fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/compat.h>
  34 #include <linux/bit_spinlock.h>
  35 #include <linux/xattr.h>
  36 #include <linux/posix_acl.h>
  37 #include <linux/falloc.h>
  38 #include <linux/slab.h>
  39 #include <linux/ratelimit.h>
  40 #include <linux/mount.h>
  41 #include <linux/btrfs.h>
  42 #include <linux/blkdev.h>
  43 #include <linux/posix_acl_xattr.h>
  44 #include <linux/uio.h>
  45 #include <linux/magic.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62 #include "qgroup.h"
  63 #include "dedupe.h"
  64
  65 struct btrfs_iget_args {
  66         struct btrfs_key *location;
  67         struct btrfs_root *root;
  68 };
  69
  70 struct btrfs_dio_data {
  71         u64 reserve;
  72         u64 unsubmitted_oe_range_start;
  73         u64 unsubmitted_oe_range_end;
  74         int overwrite;
  75 };
  76
  77 static const struct inode_operations btrfs_dir_inode_operations;
  78 static const struct inode_operations btrfs_symlink_inode_operations;
  79 static const struct inode_operations btrfs_dir_ro_inode_operations;
  80 static const struct inode_operations btrfs_special_inode_operations;
  81 static const struct inode_operations btrfs_file_inode_operations;
  82 static const struct address_space_operations btrfs_aops;
  83 static const struct address_space_operations btrfs_symlink_aops;
  84 static const struct file_operations btrfs_dir_file_operations;
  85 static const struct extent_io_ops btrfs_extent_io_ops;
  86
  87 static struct kmem_cache *btrfs_inode_cachep;
  88 struct kmem_cache *btrfs_trans_handle_cachep;
  89 struct kmem_cache *btrfs_path_cachep;
  90 struct kmem_cache *btrfs_free_space_cachep;
  91
  92 #define S_SHIFT 12
  93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  94         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  95         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  96         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  97         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  98         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  99         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
 100         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
 101 };
 102
 103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
 104 static int btrfs_truncate(struct inode *inode);
 105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
 106 static noinline int cow_file_range(struct inode *inode,
 107                                    struct page *locked_page,
 108                                    u64 start, u64 end, u64 delalloc_end,
 109                                    int *page_started, unsigned long *nr_written,
 110                                    int unlock, struct btrfs_dedupe_hash *hash);
 111 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
 112                                        u64 orig_start, u64 block_start,
 113                                        u64 block_len, u64 orig_block_len,
 114                                        u64 ram_bytes, int compress_type,
 115                                        int type);
 116
 117 static void __endio_write_update_ordered(struct inode *inode,
 118                                          const u64 offset, const u64 bytes,
 119                                          const bool uptodate);
 120
 121 /*
 122  * Cleanup all submitted ordered extents in specified range to handle errors
 123  * from the fill_dellaloc() callback.
 124  *
 125  * NOTE: caller must ensure that when an error happens, it can not call
 126  * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
 127  * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
 128  * to be released, which we want to happen only when finishing the ordered
 129  * extent (btrfs_finish_ordered_io()). Also note that the caller of the
 130  * fill_delalloc() callback already does proper cleanup for the first page of
 131  * the range, that is, it invokes the callback writepage_end_io_hook() for the
 132  * range of the first page.
 133  */
 134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
 135                                                  const u64 offset,
 136                                                  const u64 bytes)
 137 {
 138         unsigned long index = offset >> PAGE_SHIFT;
 139         unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
 140         struct page *page;
 141
 142         while (index <= end_index) {
 143                 page = find_get_page(inode->i_mapping, index);
 144                 index++;
 145                 if (!page)
 146                         continue;
 147                 ClearPagePrivate2(page);
 148                 put_page(page);
 149         }
 150         return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
 151                                             bytes - PAGE_SIZE, false);
 152 }
 153
 154 static int btrfs_dirty_inode(struct inode *inode);
 155
 156 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 157 void btrfs_test_inode_set_ops(struct inode *inode)
 158 {
 159         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 160 }
 161 #endif
 162
 163 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 164                                      struct inode *inode,  struct inode *dir,
 165                                      const struct qstr *qstr)
 166 {
 167         int err;
 168
 169         err = btrfs_init_acl(trans, inode, dir);
 170         if (!err)
 171                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 172         return err;
 173 }
 174
 175 /*
 176  * this does all the hard work for inserting an inline extent into
 177  * the btree.  The caller should have done a btrfs_drop_extents so that
 178  * no overlapping inline items exist in the btree
 179  */
 180 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 181                                 struct btrfs_path *path, int extent_inserted,
 182                                 struct btrfs_root *root, struct inode *inode,
 183                                 u64 start, size_t size, size_t compressed_size,
 184                                 int compress_type,
 185                                 struct page **compressed_pages)
 186 {
 187         struct extent_buffer *leaf;
 188         struct page *page = NULL;
 189         char *kaddr;
 190         unsigned long ptr;
 191         struct btrfs_file_extent_item *ei;
 192         int ret;
 193         size_t cur_size = size;
 194         unsigned long offset;
 195
 196         if (compressed_size && compressed_pages)
 197                 cur_size = compressed_size;
 198
 199         inode_add_bytes(inode, size);
 200
 201         if (!extent_inserted) {
 202                 struct btrfs_key key;
 203                 size_t datasize;
 204
 205                 key.objectid = btrfs_ino(BTRFS_I(inode));
 206                 key.offset = start;
 207                 key.type = BTRFS_EXTENT_DATA_KEY;
 208
 209                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 210                 path->leave_spinning = 1;
 211                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 212                                               datasize);
 213                 if (ret)
 214                         goto fail;
 215         }
 216         leaf = path->nodes[0];
 217         ei = btrfs_item_ptr(leaf, path->slots[0],
 218                             struct btrfs_file_extent_item);
 219         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 220         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 221         btrfs_set_file_extent_encryption(leaf, ei, 0);
 222         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 223         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 224         ptr = btrfs_file_extent_inline_start(ei);
 225
 226         if (compress_type != BTRFS_COMPRESS_NONE) {
 227                 struct page *cpage;
 228                 int i = 0;
 229                 while (compressed_size > 0) {
 230                         cpage = compressed_pages[i];
 231                         cur_size = min_t(unsigned long, compressed_size,
 232                                        PAGE_SIZE);
 233
 234                         kaddr = kmap_atomic(cpage);
 235                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 236                         kunmap_atomic(kaddr);
 237
 238                         i++;
 239                         ptr += cur_size;
 240                         compressed_size -= cur_size;
 241                 }
 242                 btrfs_set_file_extent_compression(leaf, ei,
 243                                                   compress_type);
 244         } else {
 245                 page = find_get_page(inode->i_mapping,
 246                                      start >> PAGE_SHIFT);
 247                 btrfs_set_file_extent_compression(leaf, ei, 0);
 248                 kaddr = kmap_atomic(page);
 249                 offset = start & (PAGE_SIZE - 1);
 250                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 251                 kunmap_atomic(kaddr);
 252                 put_page(page);
 253         }
 254         btrfs_mark_buffer_dirty(leaf);
 255         btrfs_release_path(path);
 256
 257         /*
 258          * we're an inline extent, so nobody can
 259          * extend the file past i_size without locking
 260          * a page we already have locked.
 261          *
 262          * We must do any isize and inode updates
 263          * before we unlock the pages.  Otherwise we
 264          * could end up racing with unlink.
 265          */
 266         BTRFS_I(inode)->disk_i_size = inode->i_size;
 267         ret = btrfs_update_inode(trans, root, inode);
 268
 269 fail:
 270         return ret;
 271 }
 272
 273
 274 /*
 275  * conditionally insert an inline extent into the file.  This
 276  * does the checks required to make sure the data is small enough
 277  * to fit as an inline extent.
 278  */
 279 static noinline int cow_file_range_inline(struct btrfs_root *root,
 280                                           struct inode *inode, u64 start,
 281                                           u64 end, size_t compressed_size,
 282                                           int compress_type,
 283                                           struct page **compressed_pages)
 284 {
 285         struct btrfs_fs_info *fs_info = root->fs_info;
 286         struct btrfs_trans_handle *trans;
 287         u64 isize = i_size_read(inode);
 288         u64 actual_end = min(end + 1, isize);
 289         u64 inline_len = actual_end - start;
 290         u64 aligned_end = ALIGN(end, fs_info->sectorsize);
 291         u64 data_len = inline_len;
 292         int ret;
 293         struct btrfs_path *path;
 294         int extent_inserted = 0;
 295         u32 extent_item_size;
 296
 297         if (compressed_size)
 298                 data_len = compressed_size;
 299
 300         if (start > 0 ||
 301             actual_end > fs_info->sectorsize ||
 302             data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
 303             (!compressed_size &&
 304             (actual_end & (fs_info->sectorsize - 1)) == 0) ||
 305             end + 1 < isize ||
 306             data_len > fs_info->max_inline) {
 307                 return 1;
 308         }
 309
 310         path = btrfs_alloc_path();
 311         if (!path)
 312                 return -ENOMEM;
 313
 314         trans = btrfs_join_transaction(root);
 315         if (IS_ERR(trans)) {
 316                 btrfs_free_path(path);
 317                 return PTR_ERR(trans);
 318         }
 319         trans->block_rsv = &BTRFS_I(inode)->block_rsv;
 320
 321         if (compressed_size && compressed_pages)
 322                 extent_item_size = btrfs_file_extent_calc_inline_size(
 323                    compressed_size);
 324         else
 325                 extent_item_size = btrfs_file_extent_calc_inline_size(
 326                     inline_len);
 327
 328         ret = __btrfs_drop_extents(trans, root, inode, path,
 329                                    start, aligned_end, NULL,
 330                                    1, 1, extent_item_size, &extent_inserted);
 331         if (ret) {
 332                 btrfs_abort_transaction(trans, ret);
 333                 goto out;
 334         }
 335
 336         if (isize > actual_end)
 337                 inline_len = min_t(u64, isize, actual_end);
 338         ret = insert_inline_extent(trans, path, extent_inserted,
 339                                    root, inode, start,
 340                                    inline_len, compressed_size,
 341                                    compress_type, compressed_pages);
 342         if (ret && ret != -ENOSPC) {
 343                 btrfs_abort_transaction(trans, ret);
 344                 goto out;
 345         } else if (ret == -ENOSPC) {
 346                 ret = 1;
 347                 goto out;
 348         }
 349
 350         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 351         btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
 352 out:
 353         /*
 354          * Don't forget to free the reserved space, as for inlined extent
 355          * it won't count as data extent, free them directly here.
 356          * And at reserve time, it's always aligned to page size, so
 357          * just free one page here.
 358          */
 359         btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
 360         btrfs_free_path(path);
 361         btrfs_end_transaction(trans);
 362         return ret;
 363 }
 364
 365 struct async_extent {
 366         u64 start;
 367         u64 ram_size;
 368         u64 compressed_size;
 369         struct page **pages;
 370         unsigned long nr_pages;
 371         int compress_type;
 372         struct list_head list;
 373 };
 374
 375 struct async_cow {
 376         struct inode *inode;
 377         struct btrfs_root *root;
 378         struct page *locked_page;
 379         u64 start;
 380         u64 end;
 381         unsigned int write_flags;
 382         struct list_head extents;
 383         struct btrfs_work work;
 384 };
 385
 386 static noinline int add_async_extent(struct async_cow *cow,
 387                                      u64 start, u64 ram_size,
 388                                      u64 compressed_size,
 389                                      struct page **pages,
 390                                      unsigned long nr_pages,
 391                                      int compress_type)
 392 {
 393         struct async_extent *async_extent;
 394
 395         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 396         BUG_ON(!async_extent); /* -ENOMEM */
 397         async_extent->start = start;
 398         async_extent->ram_size = ram_size;
 399         async_extent->compressed_size = compressed_size;
 400         async_extent->pages = pages;
 401         async_extent->nr_pages = nr_pages;
 402         async_extent->compress_type = compress_type;
 403         list_add_tail(&async_extent->list, &cow->extents);
 404         return 0;
 405 }
 406
 407 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
 408 {
 409         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 410
 411         /* force compress */
 412         if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
 413                 return 1;
 414         /* defrag ioctl */
 415         if (BTRFS_I(inode)->defrag_compress)
 416                 return 1;
 417         /* bad compression ratios */
 418         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 419                 return 0;
 420         if (btrfs_test_opt(fs_info, COMPRESS) ||
 421             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 422             BTRFS_I(inode)->prop_compress)
 423                 return btrfs_compress_heuristic(inode, start, end);
 424         return 0;
 425 }
 426
 427 static inline void inode_should_defrag(struct btrfs_inode *inode,
 428                 u64 start, u64 end, u64 num_bytes, u64 small_write)
 429 {
 430         /* If this is a small write inside eof, kick off a defrag */
 431         if (num_bytes < small_write &&
 432             (start > 0 || end + 1 < inode->disk_i_size))
 433                 btrfs_add_inode_defrag(NULL, inode);
 434 }
 435
 436 /*
 437  * we create compressed extents in two phases.  The first
 438  * phase compresses a range of pages that have already been
 439  * locked (both pages and state bits are locked).
 440  *
 441  * This is done inside an ordered work queue, and the compression
 442  * is spread across many cpus.  The actual IO submission is step
 443  * two, and the ordered work queue takes care of making sure that
 444  * happens in the same order things were put onto the queue by
 445  * writepages and friends.
 446  *
 447  * If this code finds it can't get good compression, it puts an
 448  * entry onto the work queue to write the uncompressed bytes.  This
 449  * makes sure that both compressed inodes and uncompressed inodes
 450  * are written in the same order that the flusher thread sent them
 451  * down.
 452  */
 453 static noinline void compress_file_range(struct inode *inode,
 454                                         struct page *locked_page,
 455                                         u64 start, u64 end,
 456                                         struct async_cow *async_cow,
 457                                         int *num_added)
 458 {
 459         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 460         struct btrfs_root *root = BTRFS_I(inode)->root;
 461         u64 blocksize = fs_info->sectorsize;
 462         u64 actual_end;
 463         u64 isize = i_size_read(inode);
 464         int ret = 0;
 465         struct page **pages = NULL;
 466         unsigned long nr_pages;
 467         unsigned long total_compressed = 0;
 468         unsigned long total_in = 0;
 469         int i;
 470         int will_compress;
 471         int compress_type = fs_info->compress_type;
 472         int redirty = 0;
 473
 474         inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
 475                         SZ_16K);
 476
 477         actual_end = min_t(u64, isize, end + 1);
 478 again:
 479         will_compress = 0;
 480         nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
 481         BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
 482         nr_pages = min_t(unsigned long, nr_pages,
 483                         BTRFS_MAX_COMPRESSED / PAGE_SIZE);
 484
 485         /*
 486          * we don't want to send crud past the end of i_size through
 487          * compression, that's just a waste of CPU time.  So, if the
 488          * end of the file is before the start of our current
 489          * requested range of bytes, we bail out to the uncompressed
 490          * cleanup code that can deal with all of this.
 491          *
 492          * It isn't really the fastest way to fix things, but this is a
 493          * very uncommon corner.
 494          */
 495         if (actual_end <= start)
 496                 goto cleanup_and_bail_uncompressed;
 497
 498         total_compressed = actual_end - start;
 499
 500         /*
 501          * skip compression for a small file range(<=blocksize) that
 502          * isn't an inline extent, since it doesn't save disk space at all.
 503          */
 504         if (total_compressed <= blocksize &&
 505            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 506                 goto cleanup_and_bail_uncompressed;
 507
 508         total_compressed = min_t(unsigned long, total_compressed,
 509                         BTRFS_MAX_UNCOMPRESSED);
 510         total_in = 0;
 511         ret = 0;
 512
 513         /*
 514          * we do compression for mount -o compress and when the
 515          * inode has not been flagged as nocompress.  This flag can
 516          * change at any time if we discover bad compression ratios.
 517          */
 518         if (inode_need_compress(inode, start, end)) {
 519                 WARN_ON(pages);
 520                 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
 521                 if (!pages) {
 522                         /* just bail out to the uncompressed code */
 523                         goto cont;
 524                 }
 525
 526                 if (BTRFS_I(inode)->defrag_compress)
 527                         compress_type = BTRFS_I(inode)->defrag_compress;
 528                 else if (BTRFS_I(inode)->prop_compress)
 529                         compress_type = BTRFS_I(inode)->prop_compress;
 530
 531                 /*
 532                  * we need to call clear_page_dirty_for_io on each
 533                  * page in the range.  Otherwise applications with the file
 534                  * mmap'd can wander in and change the page contents while
 535                  * we are compressing them.
 536                  *
 537                  * If the compression fails for any reason, we set the pages
 538                  * dirty again later on.
 539                  *
 540                  * Note that the remaining part is redirtied, the start pointer
 541                  * has moved, the end is the original one.
 542                  */
 543                 if (!redirty) {
 544                         extent_range_clear_dirty_for_io(inode, start, end);
 545                         redirty = 1;
 546                 }
 547
 548                 /* Compression level is applied here and only here */
 549                 ret = btrfs_compress_pages(
 550                         compress_type | (fs_info->compress_level << 4),
 551                                            inode->i_mapping, start,
 552                                            pages,
 553                                            &nr_pages,
 554                                            &total_in,
 555                                            &total_compressed);
 556
 557                 if (!ret) {
 558                         unsigned long offset = total_compressed &
 559                                 (PAGE_SIZE - 1);
 560                         struct page *page = pages[nr_pages - 1];
 561                         char *kaddr;
 562
 563                         /* zero the tail end of the last page, we might be
 564                          * sending it down to disk
 565                          */
 566                         if (offset) {
 567                                 kaddr = kmap_atomic(page);
 568                                 memset(kaddr + offset, 0,
 569                                        PAGE_SIZE - offset);
 570                                 kunmap_atomic(kaddr);
 571                         }
 572                         will_compress = 1;
 573                 }
 574         }
 575 cont:
 576         if (start == 0) {
 577                 /* lets try to make an inline extent */
 578                 if (ret || total_in < actual_end) {
 579                         /* we didn't compress the entire range, try
 580                          * to make an uncompressed inline extent.
 581                          */
 582                         ret = cow_file_range_inline(root, inode, start, end,
 583                                             0, BTRFS_COMPRESS_NONE, NULL);
 584                 } else {
 585                         /* try making a compressed inline extent */
 586                         ret = cow_file_range_inline(root, inode, start, end,
 587                                                     total_compressed,
 588                                                     compress_type, pages);
 589                 }
 590                 if (ret <= 0) {
 591                         unsigned long clear_flags = EXTENT_DELALLOC |
 592                                 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
 593                                 EXTENT_DO_ACCOUNTING;
 594                         unsigned long page_error_op;
 595
 596                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 597
 598                         /*
 599                          * inline extent creation worked or returned error,
 600                          * we don't need to create any more async work items.
 601                          * Unlock and free up our temp pages.
 602                          *
 603                          * We use DO_ACCOUNTING here because we need the
 604                          * delalloc_release_metadata to be done _after_ we drop
 605                          * our outstanding extent for clearing delalloc for this
 606                          * range.
 607                          */
 608                         extent_clear_unlock_delalloc(inode, start, end, end,
 609                                                      NULL, clear_flags,
 610                                                      PAGE_UNLOCK |
 611                                                      PAGE_CLEAR_DIRTY |
 612                                                      PAGE_SET_WRITEBACK |
 613                                                      page_error_op |
 614                                                      PAGE_END_WRITEBACK);
 615                         goto free_pages_out;
 616                 }
 617         }
 618
 619         if (will_compress) {
 620                 /*
 621                  * we aren't doing an inline extent round the compressed size
 622                  * up to a block size boundary so the allocator does sane
 623                  * things
 624                  */
 625                 total_compressed = ALIGN(total_compressed, blocksize);
 626
 627                 /*
 628                  * one last check to make sure the compression is really a
 629                  * win, compare the page count read with the blocks on disk,
 630                  * compression must free at least one sector size
 631                  */
 632                 total_in = ALIGN(total_in, PAGE_SIZE);
 633                 if (total_compressed + blocksize <= total_in) {
 634                         *num_added += 1;
 635
 636                         /*
 637                          * The async work queues will take care of doing actual
 638                          * allocation on disk for these compressed pages, and
 639                          * will submit them to the elevator.
 640                          */
 641                         add_async_extent(async_cow, start, total_in,
 642                                         total_compressed, pages, nr_pages,
 643                                         compress_type);
 644
 645                         if (start + total_in < end) {
 646                                 start += total_in;
 647                                 pages = NULL;
 648                                 cond_resched();
 649                                 goto again;
 650                         }
 651                         return;
 652                 }
 653         }
 654         if (pages) {
 655                 /*
 656                  * the compression code ran but failed to make things smaller,
 657                  * free any pages it allocated and our page pointer array
 658                  */
 659                 for (i = 0; i < nr_pages; i++) {
 660                         WARN_ON(pages[i]->mapping);
 661                         put_page(pages[i]);
 662                 }
 663                 kfree(pages);
 664                 pages = NULL;
 665                 total_compressed = 0;
 666                 nr_pages = 0;
 667
 668                 /* flag the file so we don't compress in the future */
 669                 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
 670                     !(BTRFS_I(inode)->prop_compress)) {
 671                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 672                 }
 673         }
 674 cleanup_and_bail_uncompressed:
 675         /*
 676          * No compression, but we still need to write the pages in the file
 677          * we've been given so far.  redirty the locked page if it corresponds
 678          * to our extent and set things up for the async work queue to run
 679          * cow_file_range to do the normal delalloc dance.
 680          */
 681         if (page_offset(locked_page) >= start &&
 682             page_offset(locked_page) <= end)
 683                 __set_page_dirty_nobuffers(locked_page);
 684                 /* unlocked later on in the async handlers */
 685
 686         if (redirty)
 687                 extent_range_redirty_for_io(inode, start, end);
 688         add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
 689                          BTRFS_COMPRESS_NONE);
 690         *num_added += 1;
 691
 692         return;
 693
 694 free_pages_out:
 695         for (i = 0; i < nr_pages; i++) {
 696                 WARN_ON(pages[i]->mapping);
 697                 put_page(pages[i]);
 698         }
 699         kfree(pages);
 700 }
 701
 702 static void free_async_extent_pages(struct async_extent *async_extent)
 703 {
 704         int i;
 705
 706         if (!async_extent->pages)
 707                 return;
 708
 709         for (i = 0; i < async_extent->nr_pages; i++) {
 710                 WARN_ON(async_extent->pages[i]->mapping);
 711                 put_page(async_extent->pages[i]);
 712         }
 713         kfree(async_extent->pages);
 714         async_extent->nr_pages = 0;
 715         async_extent->pages = NULL;
 716 }
 717
 718 /*
 719  * phase two of compressed writeback.  This is the ordered portion
 720  * of the code, which only gets called in the order the work was
 721  * queued.  We walk all the async extents created by compress_file_range
 722  * and send them down to the disk.
 723  */
 724 static noinline void submit_compressed_extents(struct inode *inode,
 725                                               struct async_cow *async_cow)
 726 {
 727         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 728         struct async_extent *async_extent;
 729         u64 alloc_hint = 0;
 730         struct btrfs_key ins;
 731         struct extent_map *em;
 732         struct btrfs_root *root = BTRFS_I(inode)->root;
 733         struct extent_io_tree *io_tree;
 734         int ret = 0;
 735
 736 again:
 737         while (!list_empty(&async_cow->extents)) {
 738                 async_extent = list_entry(async_cow->extents.next,
 739                                           struct async_extent, list);
 740                 list_del(&async_extent->list);
 741
 742                 io_tree = &BTRFS_I(inode)->io_tree;
 743
 744 retry:
 745                 /* did the compression code fall back to uncompressed IO? */
 746                 if (!async_extent->pages) {
 747                         int page_started = 0;
 748                         unsigned long nr_written = 0;
 749
 750                         lock_extent(io_tree, async_extent->start,
 751                                          async_extent->start +
 752                                          async_extent->ram_size - 1);
 753
 754                         /* allocate blocks */
 755                         ret = cow_file_range(inode, async_cow->locked_page,
 756                                              async_extent->start,
 757                                              async_extent->start +
 758                                              async_extent->ram_size - 1,
 759                                              async_extent->start +
 760                                              async_extent->ram_size - 1,
 761                                              &page_started, &nr_written, 0,
 762                                              NULL);
 763
 764                         /* JDM XXX */
 765
 766                         /*
 767                          * if page_started, cow_file_range inserted an
 768                          * inline extent and took care of all the unlocking
 769                          * and IO for us.  Otherwise, we need to submit
 770                          * all those pages down to the drive.
 771                          */
 772                         if (!page_started && !ret)
 773                                 extent_write_locked_range(inode,
 774                                                   async_extent->start,
 775                                                   async_extent->start +
 776                                                   async_extent->ram_size - 1,
 777                                                   WB_SYNC_ALL);
 778                         else if (ret)
 779                                 unlock_page(async_cow->locked_page);
 780                         kfree(async_extent);
 781                         cond_resched();
 782                         continue;
 783                 }
 784
 785                 lock_extent(io_tree, async_extent->start,
 786                             async_extent->start + async_extent->ram_size - 1);
 787
 788                 ret = btrfs_reserve_extent(root, async_extent->ram_size,
 789                                            async_extent->compressed_size,
 790                                            async_extent->compressed_size,
 791                                            0, alloc_hint, &ins, 1, 1);
 792                 if (ret) {
 793                         free_async_extent_pages(async_extent);
 794
 795                         if (ret == -ENOSPC) {
 796                                 unlock_extent(io_tree, async_extent->start,
 797                                               async_extent->start +
 798                                               async_extent->ram_size - 1);
 799
 800                                 /*
 801                                  * we need to redirty the pages if we decide to
 802                                  * fallback to uncompressed IO, otherwise we
 803                                  * will not submit these pages down to lower
 804                                  * layers.
 805                                  */
 806                                 extent_range_redirty_for_io(inode,
 807                                                 async_extent->start,
 808                                                 async_extent->start +
 809                                                 async_extent->ram_size - 1);
 810
 811                                 goto retry;
 812                         }
 813                         goto out_free;
 814                 }
 815                 /*
 816                  * here we're doing allocation and writeback of the
 817                  * compressed pages
 818                  */
 819                 em = create_io_em(inode, async_extent->start,
 820                                   async_extent->ram_size, /* len */
 821                                   async_extent->start, /* orig_start */
 822                                   ins.objectid, /* block_start */
 823                                   ins.offset, /* block_len */
 824                                   ins.offset, /* orig_block_len */
 825                                   async_extent->ram_size, /* ram_bytes */
 826                                   async_extent->compress_type,
 827                                   BTRFS_ORDERED_COMPRESSED);
 828                 if (IS_ERR(em))
 829                         /* ret value is not necessary due to void function */
 830                         goto out_free_reserve;
 831                 free_extent_map(em);
 832
 833                 ret = btrfs_add_ordered_extent_compress(inode,
 834                                                 async_extent->start,
 835                                                 ins.objectid,
 836                                                 async_extent->ram_size,
 837                                                 ins.offset,
 838                                                 BTRFS_ORDERED_COMPRESSED,
 839                                                 async_extent->compress_type);
 840                 if (ret) {
 841                         btrfs_drop_extent_cache(BTRFS_I(inode),
 842                                                 async_extent->start,
 843                                                 async_extent->start +
 844                                                 async_extent->ram_size - 1, 0);
 845                         goto out_free_reserve;
 846                 }
 847                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 848
 849                 /*
 850                  * clear dirty, set writeback and unlock the pages.
 851                  */
 852                 extent_clear_unlock_delalloc(inode, async_extent->start,
 853                                 async_extent->start +
 854                                 async_extent->ram_size - 1,
 855                                 async_extent->start +
 856                                 async_extent->ram_size - 1,
 857                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 858                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 859                                 PAGE_SET_WRITEBACK);
 860                 if (btrfs_submit_compressed_write(inode,
 861                                     async_extent->start,
 862                                     async_extent->ram_size,
 863                                     ins.objectid,
 864                                     ins.offset, async_extent->pages,
 865                                     async_extent->nr_pages,
 866                                     async_cow->write_flags)) {
 867                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
 868                         struct page *p = async_extent->pages[0];
 869                         const u64 start = async_extent->start;
 870                         const u64 end = start + async_extent->ram_size - 1;
 871
 872                         p->mapping = inode->i_mapping;
 873                         tree->ops->writepage_end_io_hook(p, start, end,
 874                                                          NULL, 0);
 875                         p->mapping = NULL;
 876                         extent_clear_unlock_delalloc(inode, start, end, end,
 877                                                      NULL, 0,
 878                                                      PAGE_END_WRITEBACK |
 879                                                      PAGE_SET_ERROR);
 880                         free_async_extent_pages(async_extent);
 881                 }
 882                 alloc_hint = ins.objectid + ins.offset;
 883                 kfree(async_extent);
 884                 cond_resched();
 885         }
 886         return;
 887 out_free_reserve:
 888         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
 889         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
 890 out_free:
 891         extent_clear_unlock_delalloc(inode, async_extent->start,
 892                                      async_extent->start +
 893                                      async_extent->ram_size - 1,
 894                                      async_extent->start +
 895                                      async_extent->ram_size - 1,
 896                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 897                                      EXTENT_DELALLOC_NEW |
 898                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 899                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 900                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 901                                      PAGE_SET_ERROR);
 902         free_async_extent_pages(async_extent);
 903         kfree(async_extent);
 904         goto again;
 905 }
 906
 907 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 908                                       u64 num_bytes)
 909 {
 910         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 911         struct extent_map *em;
 912         u64 alloc_hint = 0;
 913
 914         read_lock(&em_tree->lock);
 915         em = search_extent_mapping(em_tree, start, num_bytes);
 916         if (em) {
 917                 /*
 918                  * if block start isn't an actual block number then find the
 919                  * first block in this inode and use that as a hint.  If that
 920                  * block is also bogus then just don't worry about it.
 921                  */
 922                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 923                         free_extent_map(em);
 924                         em = search_extent_mapping(em_tree, 0, 0);
 925                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 926                                 alloc_hint = em->block_start;
 927                         if (em)
 928                                 free_extent_map(em);
 929                 } else {
 930                         alloc_hint = em->block_start;
 931                         free_extent_map(em);
 932                 }
 933         }
 934         read_unlock(&em_tree->lock);
 935
 936         return alloc_hint;
 937 }
 938
 939 /*
 940  * when extent_io.c finds a delayed allocation range in the file,
 941  * the call backs end up in this code.  The basic idea is to
 942  * allocate extents on disk for the range, and create ordered data structs
 943  * in ram to track those extents.
 944  *
 945  * locked_page is the page that writepage had locked already.  We use
 946  * it to make sure we don't do extra locks or unlocks.
 947  *
 948  * *page_started is set to one if we unlock locked_page and do everything
 949  * required to start IO on it.  It may be clean and already done with
 950  * IO when we return.
 951  */
 952 static noinline int cow_file_range(struct inode *inode,
 953                                    struct page *locked_page,
 954                                    u64 start, u64 end, u64 delalloc_end,
 955                                    int *page_started, unsigned long *nr_written,
 956                                    int unlock, struct btrfs_dedupe_hash *hash)
 957 {
 958         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 959         struct btrfs_root *root = BTRFS_I(inode)->root;
 960         u64 alloc_hint = 0;
 961         u64 num_bytes;
 962         unsigned long ram_size;
 963         u64 disk_num_bytes;
 964         u64 cur_alloc_size = 0;
 965         u64 blocksize = fs_info->sectorsize;
 966         struct btrfs_key ins;
 967         struct extent_map *em;
 968         unsigned clear_bits;
 969         unsigned long page_ops;
 970         bool extent_reserved = false;
 971         int ret = 0;
 972
 973         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 974                 WARN_ON_ONCE(1);
 975                 ret = -EINVAL;
 976                 goto out_unlock;
 977         }
 978
 979         num_bytes = ALIGN(end - start + 1, blocksize);
 980         num_bytes = max(blocksize,  num_bytes);
 981         disk_num_bytes = num_bytes;
 982
 983         inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
 984
 985         if (start == 0) {
 986                 /* lets try to make an inline extent */
 987                 ret = cow_file_range_inline(root, inode, start, end, 0,
 988                                         BTRFS_COMPRESS_NONE, NULL);
 989                 if (ret == 0) {
 990                         /*
 991                          * We use DO_ACCOUNTING here because we need the
 992                          * delalloc_release_metadata to be run _after_ we drop
 993                          * our outstanding extent for clearing delalloc for this
 994                          * range.
 995                          */
 996                         extent_clear_unlock_delalloc(inode, start, end,
 997                                      delalloc_end, NULL,
 998                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 999                                      EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1000                                      EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1001                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1002                                      PAGE_END_WRITEBACK);
1003                         *nr_written = *nr_written +
1004                              (end - start + PAGE_SIZE) / PAGE_SIZE;
1005                         *page_started = 1;
1006                         goto out;
1007                 } else if (ret < 0) {
1008                         goto out_unlock;
1009                 }
1010         }
1011
1012         BUG_ON(disk_num_bytes >
1013                btrfs_super_total_bytes(fs_info->super_copy));
1014
1015         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1016         btrfs_drop_extent_cache(BTRFS_I(inode), start,
1017                         start + num_bytes - 1, 0);
1018
1019         while (disk_num_bytes > 0) {
1020                 cur_alloc_size = disk_num_bytes;
1021                 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1022                                            fs_info->sectorsize, 0, alloc_hint,
1023                                            &ins, 1, 1);
1024                 if (ret < 0)
1025                         goto out_unlock;
1026                 cur_alloc_size = ins.offset;
1027                 extent_reserved = true;
1028
1029                 ram_size = ins.offset;
1030                 em = create_io_em(inode, start, ins.offset, /* len */
1031                                   start, /* orig_start */
1032                                   ins.objectid, /* block_start */
1033                                   ins.offset, /* block_len */
1034                                   ins.offset, /* orig_block_len */
1035                                   ram_size, /* ram_bytes */
1036                                   BTRFS_COMPRESS_NONE, /* compress_type */
1037                                   BTRFS_ORDERED_REGULAR /* type */);
1038                 if (IS_ERR(em))
1039                         goto out_reserve;
1040                 free_extent_map(em);
1041
1042                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1043                                                ram_size, cur_alloc_size, 0);
1044                 if (ret)
1045                         goto out_drop_extent_cache;
1046
1047                 if (root->root_key.objectid ==
1048                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1049                         ret = btrfs_reloc_clone_csums(inode, start,
1050                                                       cur_alloc_size);
1051                         /*
1052                          * Only drop cache here, and process as normal.
1053                          *
1054                          * We must not allow extent_clear_unlock_delalloc()
1055                          * at out_unlock label to free meta of this ordered
1056                          * extent, as its meta should be freed by
1057                          * btrfs_finish_ordered_io().
1058                          *
1059                          * So we must continue until @start is increased to
1060                          * skip current ordered extent.
1061                          */
1062                         if (ret)
1063                                 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1064                                                 start + ram_size - 1, 0);
1065                 }
1066
1067                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1068
1069                 /* we're not doing compressed IO, don't unlock the first
1070                  * page (which the caller expects to stay locked), don't
1071                  * clear any dirty bits and don't set any writeback bits
1072                  *
1073                  * Do set the Private2 bit so we know this page was properly
1074                  * setup for writepage
1075                  */
1076                 page_ops = unlock ? PAGE_UNLOCK : 0;
1077                 page_ops |= PAGE_SET_PRIVATE2;
1078
1079                 extent_clear_unlock_delalloc(inode, start,
1080                                              start + ram_size - 1,
1081                                              delalloc_end, locked_page,
1082                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1083                                              page_ops);
1084                 if (disk_num_bytes < cur_alloc_size)
1085                         disk_num_bytes = 0;
1086                 else
1087                         disk_num_bytes -= cur_alloc_size;
1088                 num_bytes -= cur_alloc_size;
1089                 alloc_hint = ins.objectid + ins.offset;
1090                 start += cur_alloc_size;
1091                 extent_reserved = false;
1092
1093                 /*
1094                  * btrfs_reloc_clone_csums() error, since start is increased
1095                  * extent_clear_unlock_delalloc() at out_unlock label won't
1096                  * free metadata of current ordered extent, we're OK to exit.
1097                  */
1098                 if (ret)
1099                         goto out_unlock;
1100         }
1101 out:
1102         return ret;
1103
1104 out_drop_extent_cache:
1105         btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1106 out_reserve:
1107         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1108         btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1109 out_unlock:
1110         clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1111                 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1112         page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1113                 PAGE_END_WRITEBACK;
1114         /*
1115          * If we reserved an extent for our delalloc range (or a subrange) and
1116          * failed to create the respective ordered extent, then it means that
1117          * when we reserved the extent we decremented the extent's size from
1118          * the data space_info's bytes_may_use counter and incremented the
1119          * space_info's bytes_reserved counter by the same amount. We must make
1120          * sure extent_clear_unlock_delalloc() does not try to decrement again
1121          * the data space_info's bytes_may_use counter, therefore we do not pass
1122          * it the flag EXTENT_CLEAR_DATA_RESV.
1123          */
1124         if (extent_reserved) {
1125                 extent_clear_unlock_delalloc(inode, start,
1126                                              start + cur_alloc_size,
1127                                              start + cur_alloc_size,
1128                                              locked_page,
1129                                              clear_bits,
1130                                              page_ops);
1131                 start += cur_alloc_size;
1132                 if (start >= end)
1133                         goto out;
1134         }
1135         extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1136                                      locked_page,
1137                                      clear_bits | EXTENT_CLEAR_DATA_RESV,
1138                                      page_ops);
1139         goto out;
1140 }
1141
1142 /*
1143  * work queue call back to started compression on a file and pages
1144  */
1145 static noinline void async_cow_start(struct btrfs_work *work)
1146 {
1147         struct async_cow *async_cow;
1148         int num_added = 0;
1149         async_cow = container_of(work, struct async_cow, work);
1150
1151         compress_file_range(async_cow->inode, async_cow->locked_page,
1152                             async_cow->start, async_cow->end, async_cow,
1153                             &num_added);
1154         if (num_added == 0) {
1155                 btrfs_add_delayed_iput(async_cow->inode);
1156                 async_cow->inode = NULL;
1157         }
1158 }
1159
1160 /*
1161  * work queue call back to submit previously compressed pages
1162  */
1163 static noinline void async_cow_submit(struct btrfs_work *work)
1164 {
1165         struct btrfs_fs_info *fs_info;
1166         struct async_cow *async_cow;
1167         struct btrfs_root *root;
1168         unsigned long nr_pages;
1169
1170         async_cow = container_of(work, struct async_cow, work);
1171
1172         root = async_cow->root;
1173         fs_info = root->fs_info;
1174         nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1175                 PAGE_SHIFT;
1176
1177         /*
1178          * atomic_sub_return implies a barrier for waitqueue_active
1179          */
1180         if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1181             5 * SZ_1M &&
1182             waitqueue_active(&fs_info->async_submit_wait))
1183                 wake_up(&fs_info->async_submit_wait);
1184
1185         if (async_cow->inode)
1186                 submit_compressed_extents(async_cow->inode, async_cow);
1187 }
1188
1189 static noinline void async_cow_free(struct btrfs_work *work)
1190 {
1191         struct async_cow *async_cow;
1192         async_cow = container_of(work, struct async_cow, work);
1193         if (async_cow->inode)
1194                 btrfs_add_delayed_iput(async_cow->inode);
1195         kfree(async_cow);
1196 }
1197
1198 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1199                                 u64 start, u64 end, int *page_started,
1200                                 unsigned long *nr_written,
1201                                 unsigned int write_flags)
1202 {
1203         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1204         struct async_cow *async_cow;
1205         struct btrfs_root *root = BTRFS_I(inode)->root;
1206         unsigned long nr_pages;
1207         u64 cur_end;
1208
1209         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1210                          1, 0, NULL);
1211         while (start < end) {
1212                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1213                 BUG_ON(!async_cow); /* -ENOMEM */
1214                 async_cow->inode = igrab(inode);
1215                 async_cow->root = root;
1216                 async_cow->locked_page = locked_page;
1217                 async_cow->start = start;
1218                 async_cow->write_flags = write_flags;
1219
1220                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1221                     !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1222                         cur_end = end;
1223                 else
1224                         cur_end = min(end, start + SZ_512K - 1);
1225
1226                 async_cow->end = cur_end;
1227                 INIT_LIST_HEAD(&async_cow->extents);
1228
1229                 btrfs_init_work(&async_cow->work,
1230                                 btrfs_delalloc_helper,
1231                                 async_cow_start, async_cow_submit,
1232                                 async_cow_free);
1233
1234                 nr_pages = (cur_end - start + PAGE_SIZE) >>
1235                         PAGE_SHIFT;
1236                 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1237
1238                 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1239
1240                 *nr_written += nr_pages;
1241                 start = cur_end + 1;
1242         }
1243         *page_started = 1;
1244         return 0;
1245 }
1246
1247 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1248                                         u64 bytenr, u64 num_bytes)
1249 {
1250         int ret;
1251         struct btrfs_ordered_sum *sums;
1252         LIST_HEAD(list);
1253
1254         ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1255                                        bytenr + num_bytes - 1, &list, 0);
1256         if (ret == 0 && list_empty(&list))
1257                 return 0;
1258
1259         while (!list_empty(&list)) {
1260                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1261                 list_del(&sums->list);
1262                 kfree(sums);
1263         }
1264         return 1;
1265 }
1266
1267 /*
1268  * when nowcow writeback call back.  This checks for snapshots or COW copies
1269  * of the extents that exist in the file, and COWs the file as required.
1270  *
1271  * If no cow copies or snapshots exist, we write directly to the existing
1272  * blocks on disk
1273  */
1274 static noinline int run_delalloc_nocow(struct inode *inode,
1275                                        struct page *locked_page,
1276                               u64 start, u64 end, int *page_started, int force,
1277                               unsigned long *nr_written)
1278 {
1279         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1280         struct btrfs_root *root = BTRFS_I(inode)->root;
1281         struct extent_buffer *leaf;
1282         struct btrfs_path *path;
1283         struct btrfs_file_extent_item *fi;
1284         struct btrfs_key found_key;
1285         struct extent_map *em;
1286         u64 cow_start;
1287         u64 cur_offset;
1288         u64 extent_end;
1289         u64 extent_offset;
1290         u64 disk_bytenr;
1291         u64 num_bytes;
1292         u64 disk_num_bytes;
1293         u64 ram_bytes;
1294         int extent_type;
1295         int ret, err;
1296         int type;
1297         int nocow;
1298         int check_prev = 1;
1299         bool nolock;
1300         u64 ino = btrfs_ino(BTRFS_I(inode));
1301
1302         path = btrfs_alloc_path();
1303         if (!path) {
1304                 extent_clear_unlock_delalloc(inode, start, end, end,
1305                                              locked_page,
1306                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1307                                              EXTENT_DO_ACCOUNTING |
1308                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1309                                              PAGE_CLEAR_DIRTY |
1310                                              PAGE_SET_WRITEBACK |
1311                                              PAGE_END_WRITEBACK);
1312                 return -ENOMEM;
1313         }
1314
1315         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1316
1317         cow_start = (u64)-1;
1318         cur_offset = start;
1319         while (1) {
1320                 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1321                                                cur_offset, 0);
1322                 if (ret < 0)
1323                         goto error;
1324                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1325                         leaf = path->nodes[0];
1326                         btrfs_item_key_to_cpu(leaf, &found_key,
1327                                               path->slots[0] - 1);
1328                         if (found_key.objectid == ino &&
1329                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1330                                 path->slots[0]--;
1331                 }
1332                 check_prev = 0;
1333 next_slot:
1334                 leaf = path->nodes[0];
1335                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1336                         ret = btrfs_next_leaf(root, path);
1337                         if (ret < 0) {
1338                                 if (cow_start != (u64)-1)
1339                                         cur_offset = cow_start;
1340                                 goto error;
1341                         }
1342                         if (ret > 0)
1343                                 break;
1344                         leaf = path->nodes[0];
1345                 }
1346
1347                 nocow = 0;
1348                 disk_bytenr = 0;
1349                 num_bytes = 0;
1350                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1351
1352                 if (found_key.objectid > ino)
1353                         break;
1354                 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1355                     found_key.type < BTRFS_EXTENT_DATA_KEY) {
1356                         path->slots[0]++;
1357                         goto next_slot;
1358                 }
1359                 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1360                     found_key.offset > end)
1361                         break;
1362
1363                 if (found_key.offset > cur_offset) {
1364                         extent_end = found_key.offset;
1365                         extent_type = 0;
1366                         goto out_check;
1367                 }
1368
1369                 fi = btrfs_item_ptr(leaf, path->slots[0],
1370                                     struct btrfs_file_extent_item);
1371                 extent_type = btrfs_file_extent_type(leaf, fi);
1372
1373                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1374                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1375                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1376                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1377                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1378                         extent_end = found_key.offset +
1379                                 btrfs_file_extent_num_bytes(leaf, fi);
1380                         disk_num_bytes =
1381                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1382                         if (extent_end <= start) {
1383                                 path->slots[0]++;
1384                                 goto next_slot;
1385                         }
1386                         if (disk_bytenr == 0)
1387                                 goto out_check;
1388                         if (btrfs_file_extent_compression(leaf, fi) ||
1389                             btrfs_file_extent_encryption(leaf, fi) ||
1390                             btrfs_file_extent_other_encoding(leaf, fi))
1391                                 goto out_check;
1392                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1393                                 goto out_check;
1394                         if (btrfs_extent_readonly(fs_info, disk_bytenr))
1395                                 goto out_check;
1396                         if (btrfs_cross_ref_exist(root, ino,
1397                                                   found_key.offset -
1398                                                   extent_offset, disk_bytenr))
1399                                 goto out_check;
1400                         disk_bytenr += extent_offset;
1401                         disk_bytenr += cur_offset - found_key.offset;
1402                         num_bytes = min(end + 1, extent_end) - cur_offset;
1403                         /*
1404                          * if there are pending snapshots for this root,
1405                          * we fall into common COW way.
1406                          */
1407                         if (!nolock) {
1408                                 err = btrfs_start_write_no_snapshotting(root);
1409                                 if (!err)
1410                                         goto out_check;
1411                         }
1412                         /*
1413                          * force cow if csum exists in the range.
1414                          * this ensure that csum for a given extent are
1415                          * either valid or do not exist.
1416                          */
1417                         if (csum_exist_in_range(fs_info, disk_bytenr,
1418                                                 num_bytes)) {
1419                                 if (!nolock)
1420                                         btrfs_end_write_no_snapshotting(root);
1421                                 goto out_check;
1422                         }
1423                         if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1424                                 if (!nolock)
1425                                         btrfs_end_write_no_snapshotting(root);
1426                                 goto out_check;
1427                         }
1428                         nocow = 1;
1429                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1430                         extent_end = found_key.offset +
1431                                 btrfs_file_extent_inline_len(leaf,
1432                                                      path->slots[0], fi);
1433                         extent_end = ALIGN(extent_end,
1434                                            fs_info->sectorsize);
1435                 } else {
1436                         BUG_ON(1);
1437                 }
1438 out_check:
1439                 if (extent_end <= start) {
1440                         path->slots[0]++;
1441                         if (!nolock && nocow)
1442                                 btrfs_end_write_no_snapshotting(root);
1443                         if (nocow)
1444                                 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1445                         goto next_slot;
1446                 }
1447                 if (!nocow) {
1448                         if (cow_start == (u64)-1)
1449                                 cow_start = cur_offset;
1450                         cur_offset = extent_end;
1451                         if (cur_offset > end)
1452                                 break;
1453                         path->slots[0]++;
1454                         goto next_slot;
1455                 }
1456
1457                 btrfs_release_path(path);
1458                 if (cow_start != (u64)-1) {
1459                         ret = cow_file_range(inode, locked_page,
1460                                              cow_start, found_key.offset - 1,
1461                                              end, page_started, nr_written, 1,
1462                                              NULL);
1463                         if (ret) {
1464                                 if (!nolock && nocow)
1465                                         btrfs_end_write_no_snapshotting(root);
1466                                 if (nocow)
1467                                         btrfs_dec_nocow_writers(fs_info,
1468                                                                 disk_bytenr);
1469                                 goto error;
1470                         }
1471                         cow_start = (u64)-1;
1472                 }
1473
1474                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1475                         u64 orig_start = found_key.offset - extent_offset;
1476
1477                         em = create_io_em(inode, cur_offset, num_bytes,
1478                                           orig_start,
1479                                           disk_bytenr, /* block_start */
1480                                           num_bytes, /* block_len */
1481                                           disk_num_bytes, /* orig_block_len */
1482                                           ram_bytes, BTRFS_COMPRESS_NONE,
1483                                           BTRFS_ORDERED_PREALLOC);
1484                         if (IS_ERR(em)) {
1485                                 if (!nolock && nocow)
1486                                         btrfs_end_write_no_snapshotting(root);
1487                                 if (nocow)
1488                                         btrfs_dec_nocow_writers(fs_info,
1489                                                                 disk_bytenr);
1490                                 ret = PTR_ERR(em);
1491                                 goto error;
1492                         }
1493                         free_extent_map(em);
1494                 }
1495
1496                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1497                         type = BTRFS_ORDERED_PREALLOC;
1498                 } else {
1499                         type = BTRFS_ORDERED_NOCOW;
1500                 }
1501
1502                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1503                                                num_bytes, num_bytes, type);
1504                 if (nocow)
1505                         btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1506                 BUG_ON(ret); /* -ENOMEM */
1507
1508                 if (root->root_key.objectid ==
1509                     BTRFS_DATA_RELOC_TREE_OBJECTID)
1510                         /*
1511                          * Error handled later, as we must prevent
1512                          * extent_clear_unlock_delalloc() in error handler
1513                          * from freeing metadata of created ordered extent.
1514                          */
1515                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1516                                                       num_bytes);
1517
1518                 extent_clear_unlock_delalloc(inode, cur_offset,
1519                                              cur_offset + num_bytes - 1, end,
1520                                              locked_page, EXTENT_LOCKED |
1521                                              EXTENT_DELALLOC |
1522                                              EXTENT_CLEAR_DATA_RESV,
1523                                              PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1524
1525                 if (!nolock && nocow)
1526                         btrfs_end_write_no_snapshotting(root);
1527                 cur_offset = extent_end;
1528
1529                 /*
1530                  * btrfs_reloc_clone_csums() error, now we're OK to call error
1531                  * handler, as metadata for created ordered extent will only
1532                  * be freed by btrfs_finish_ordered_io().
1533                  */
1534                 if (ret)
1535                         goto error;
1536                 if (cur_offset > end)
1537                         break;
1538         }
1539         btrfs_release_path(path);
1540
1541         if (cur_offset <= end && cow_start == (u64)-1) {
1542                 cow_start = cur_offset;
1543                 cur_offset = end;
1544         }
1545
1546         if (cow_start != (u64)-1) {
1547                 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1548                                      page_started, nr_written, 1, NULL);
1549                 if (ret)
1550                         goto error;
1551         }
1552
1553 error:
1554         if (ret && cur_offset < end)
1555                 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1556                                              locked_page, EXTENT_LOCKED |
1557                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1558                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1559                                              PAGE_CLEAR_DIRTY |
1560                                              PAGE_SET_WRITEBACK |
1561                                              PAGE_END_WRITEBACK);
1562         btrfs_free_path(path);
1563         return ret;
1564 }
1565
1566 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1567 {
1568
1569         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1570             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1571                 return 0;
1572
1573         /*
1574          * @defrag_bytes is a hint value, no spinlock held here,
1575          * if is not zero, it means the file is defragging.
1576          * Force cow if given extent needs to be defragged.
1577          */
1578         if (BTRFS_I(inode)->defrag_bytes &&
1579             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1580                            EXTENT_DEFRAG, 0, NULL))
1581                 return 1;
1582
1583         return 0;
1584 }
1585
1586 /*
1587  * extent_io.c call back to do delayed allocation processing
1588  */
1589 static int run_delalloc_range(void *private_data, struct page *locked_page,
1590                               u64 start, u64 end, int *page_started,
1591                               unsigned long *nr_written,
1592                               struct writeback_control *wbc)
1593 {
1594         struct inode *inode = private_data;
1595         int ret;
1596         int force_cow = need_force_cow(inode, start, end);
1597         unsigned int write_flags = wbc_to_write_flags(wbc);
1598
1599         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1600                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1601                                          page_started, 1, nr_written);
1602         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1603                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1604                                          page_started, 0, nr_written);
1605         } else if (!inode_need_compress(inode, start, end)) {
1606                 ret = cow_file_range(inode, locked_page, start, end, end,
1607                                       page_started, nr_written, 1, NULL);
1608         } else {
1609                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1610                         &BTRFS_I(inode)->runtime_flags);
1611                 ret = cow_file_range_async(inode, locked_page, start, end,
1612                                            page_started, nr_written,
1613                                            write_flags);
1614         }
1615         if (ret)
1616                 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1617         return ret;
1618 }
1619
1620 static void btrfs_split_extent_hook(void *private_data,
1621                                     struct extent_state *orig, u64 split)
1622 {
1623         struct inode *inode = private_data;
1624         u64 size;
1625
1626         /* not delalloc, ignore it */
1627         if (!(orig->state & EXTENT_DELALLOC))
1628                 return;
1629
1630         size = orig->end - orig->start + 1;
1631         if (size > BTRFS_MAX_EXTENT_SIZE) {
1632                 u32 num_extents;
1633                 u64 new_size;
1634
1635                 /*
1636                  * See the explanation in btrfs_merge_extent_hook, the same
1637                  * applies here, just in reverse.
1638                  */
1639                 new_size = orig->end - split + 1;
1640                 num_extents = count_max_extents(new_size);
1641                 new_size = split - orig->start;
1642                 num_extents += count_max_extents(new_size);
1643                 if (count_max_extents(size) >= num_extents)
1644                         return;
1645         }
1646
1647         spin_lock(&BTRFS_I(inode)->lock);
1648         btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1649         spin_unlock(&BTRFS_I(inode)->lock);
1650 }
1651
1652 /*
1653  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1654  * extents so we can keep track of new extents that are just merged onto old
1655  * extents, such as when we are doing sequential writes, so we can properly
1656  * account for the metadata space we'll need.
1657  */
1658 static void btrfs_merge_extent_hook(void *private_data,
1659                                     struct extent_state *new,
1660                                     struct extent_state *other)
1661 {
1662         struct inode *inode = private_data;
1663         u64 new_size, old_size;
1664         u32 num_extents;
1665
1666         /* not delalloc, ignore it */
1667         if (!(other->state & EXTENT_DELALLOC))
1668                 return;
1669
1670         if (new->start > other->start)
1671                 new_size = new->end - other->start + 1;
1672         else
1673                 new_size = other->end - new->start + 1;
1674
1675         /* we're not bigger than the max, unreserve the space and go */
1676         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1677                 spin_lock(&BTRFS_I(inode)->lock);
1678                 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1679                 spin_unlock(&BTRFS_I(inode)->lock);
1680                 return;
1681         }
1682
1683         /*
1684          * We have to add up either side to figure out how many extents were
1685          * accounted for before we merged into one big extent.  If the number of
1686          * extents we accounted for is <= the amount we need for the new range
1687          * then we can return, otherwise drop.  Think of it like this
1688          *
1689          * [ 4k][MAX_SIZE]
1690          *
1691          * So we've grown the extent by a MAX_SIZE extent, this would mean we
1692          * need 2 outstanding extents, on one side we have 1 and the other side
1693          * we have 1 so they are == and we can return.  But in this case
1694          *
1695          * [MAX_SIZE+4k][MAX_SIZE+4k]
1696          *
1697          * Each range on their own accounts for 2 extents, but merged together
1698          * they are only 3 extents worth of accounting, so we need to drop in
1699          * this case.
1700          */
1701         old_size = other->end - other->start + 1;
1702         num_extents = count_max_extents(old_size);
1703         old_size = new->end - new->start + 1;
1704         num_extents += count_max_extents(old_size);
1705         if (count_max_extents(new_size) >= num_extents)
1706                 return;
1707
1708         spin_lock(&BTRFS_I(inode)->lock);
1709         btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1710         spin_unlock(&BTRFS_I(inode)->lock);
1711 }
1712
1713 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1714                                       struct inode *inode)
1715 {
1716         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1717
1718         spin_lock(&root->delalloc_lock);
1719         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1720                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1721                               &root->delalloc_inodes);
1722                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1723                         &BTRFS_I(inode)->runtime_flags);
1724                 root->nr_delalloc_inodes++;
1725                 if (root->nr_delalloc_inodes == 1) {
1726                         spin_lock(&fs_info->delalloc_root_lock);
1727                         BUG_ON(!list_empty(&root->delalloc_root));
1728                         list_add_tail(&root->delalloc_root,
1729                                       &fs_info->delalloc_roots);
1730                         spin_unlock(&fs_info->delalloc_root_lock);
1731                 }
1732         }
1733         spin_unlock(&root->delalloc_lock);
1734 }
1735
1736 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1737                                      struct btrfs_inode *inode)
1738 {
1739         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1740
1741         spin_lock(&root->delalloc_lock);
1742         if (!list_empty(&inode->delalloc_inodes)) {
1743                 list_del_init(&inode->delalloc_inodes);
1744                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1745                           &inode->runtime_flags);
1746                 root->nr_delalloc_inodes--;
1747                 if (!root->nr_delalloc_inodes) {
1748                         spin_lock(&fs_info->delalloc_root_lock);
1749                         BUG_ON(list_empty(&root->delalloc_root));
1750                         list_del_init(&root->delalloc_root);
1751                         spin_unlock(&fs_info->delalloc_root_lock);
1752                 }
1753         }
1754         spin_unlock(&root->delalloc_lock);
1755 }
1756
1757 /*
1758  * extent_io.c set_bit_hook, used to track delayed allocation
1759  * bytes in this file, and to maintain the list of inodes that
1760  * have pending delalloc work to be done.
1761  */
1762 static void btrfs_set_bit_hook(void *private_data,
1763                                struct extent_state *state, unsigned *bits)
1764 {
1765         struct inode *inode = private_data;
1766
1767         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1768
1769         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1770                 WARN_ON(1);
1771         /*
1772          * set_bit and clear bit hooks normally require _irqsave/restore
1773          * but in this case, we are only testing for the DELALLOC
1774          * bit, which is only set or cleared with irqs on
1775          */
1776         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1777                 struct btrfs_root *root = BTRFS_I(inode)->root;
1778                 u64 len = state->end + 1 - state->start;
1779                 u32 num_extents = count_max_extents(len);
1780                 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1781
1782                 spin_lock(&BTRFS_I(inode)->lock);
1783                 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1784                 spin_unlock(&BTRFS_I(inode)->lock);
1785
1786                 /* For sanity tests */
1787                 if (btrfs_is_testing(fs_info))
1788                         return;
1789
1790                 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1791                                          fs_info->delalloc_batch);
1792                 spin_lock(&BTRFS_I(inode)->lock);
1793                 BTRFS_I(inode)->delalloc_bytes += len;
1794                 if (*bits & EXTENT_DEFRAG)
1795                         BTRFS_I(inode)->defrag_bytes += len;
1796                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1797                                          &BTRFS_I(inode)->runtime_flags))
1798                         btrfs_add_delalloc_inodes(root, inode);
1799                 spin_unlock(&BTRFS_I(inode)->lock);
1800         }
1801
1802         if (!(state->state & EXTENT_DELALLOC_NEW) &&
1803             (*bits & EXTENT_DELALLOC_NEW)) {
1804                 spin_lock(&BTRFS_I(inode)->lock);
1805                 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1806                         state->start;
1807                 spin_unlock(&BTRFS_I(inode)->lock);
1808         }
1809 }
1810
1811 /*
1812  * extent_io.c clear_bit_hook, see set_bit_hook for why
1813  */
1814 static void btrfs_clear_bit_hook(void *private_data,
1815                                  struct extent_state *state,
1816                                  unsigned *bits)
1817 {
1818         struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1819         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1820         u64 len = state->end + 1 - state->start;
1821         u32 num_extents = count_max_extents(len);
1822
1823         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1824                 spin_lock(&inode->lock);
1825                 inode->defrag_bytes -= len;
1826                 spin_unlock(&inode->lock);
1827         }
1828
1829         /*
1830          * set_bit and clear bit hooks normally require _irqsave/restore
1831          * but in this case, we are only testing for the DELALLOC
1832          * bit, which is only set or cleared with irqs on
1833          */
1834         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1835                 struct btrfs_root *root = inode->root;
1836                 bool do_list = !btrfs_is_free_space_inode(inode);
1837
1838                 spin_lock(&inode->lock);
1839                 btrfs_mod_outstanding_extents(inode, -num_extents);
1840                 spin_unlock(&inode->lock);
1841
1842                 /*
1843                  * We don't reserve metadata space for space cache inodes so we
1844                  * don't need to call dellalloc_release_metadata if there is an
1845                  * error.
1846                  */
1847                 if (*bits & EXTENT_CLEAR_META_RESV &&
1848                     root != fs_info->tree_root)
1849                         btrfs_delalloc_release_metadata(inode, len);
1850
1851                 /* For sanity tests. */
1852                 if (btrfs_is_testing(fs_info))
1853                         return;
1854
1855                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1856                     do_list && !(state->state & EXTENT_NORESERVE) &&
1857                     (*bits & EXTENT_CLEAR_DATA_RESV))
1858                         btrfs_free_reserved_data_space_noquota(
1859                                         &inode->vfs_inode,
1860                                         state->start, len);
1861
1862                 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1863                                          fs_info->delalloc_batch);
1864                 spin_lock(&inode->lock);
1865                 inode->delalloc_bytes -= len;
1866                 if (do_list && inode->delalloc_bytes == 0 &&
1867                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1868                                         &inode->runtime_flags))
1869                         btrfs_del_delalloc_inode(root, inode);
1870                 spin_unlock(&inode->lock);
1871         }
1872
1873         if ((state->state & EXTENT_DELALLOC_NEW) &&
1874             (*bits & EXTENT_DELALLOC_NEW)) {
1875                 spin_lock(&inode->lock);
1876                 ASSERT(inode->new_delalloc_bytes >= len);
1877                 inode->new_delalloc_bytes -= len;
1878                 spin_unlock(&inode->lock);
1879         }
1880 }
1881
1882 /*
1883  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1884  * we don't create bios that span stripes or chunks
1885  *
1886  * return 1 if page cannot be merged to bio
1887  * return 0 if page can be merged to bio
1888  * return error otherwise
1889  */
1890 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1891                          size_t size, struct bio *bio,
1892                          unsigned long bio_flags)
1893 {
1894         struct inode *inode = page->mapping->host;
1895         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1896         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1897         u64 length = 0;
1898         u64 map_length;
1899         int ret;
1900
1901         if (bio_flags & EXTENT_BIO_COMPRESSED)
1902                 return 0;
1903
1904         length = bio->bi_iter.bi_size;
1905         map_length = length;
1906         ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1907                               NULL, 0);
1908         if (ret < 0)
1909                 return ret;
1910         if (map_length < length + size)
1911                 return 1;
1912         return 0;
1913 }
1914
1915 /*
1916  * in order to insert checksums into the metadata in large chunks,
1917  * we wait until bio submission time.   All the pages in the bio are
1918  * checksummed and sums are attached onto the ordered extent record.
1919  *
1920  * At IO completion time the cums attached on the ordered extent record
1921  * are inserted into the btree
1922  */
1923 static blk_status_t __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1924                                     int mirror_num, unsigned long bio_flags,
1925                                     u64 bio_offset)
1926 {
1927         struct inode *inode = private_data;
1928         blk_status_t ret = 0;
1929
1930         ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1931         BUG_ON(ret); /* -ENOMEM */
1932         return 0;
1933 }
1934
1935 /*
1936  * in order to insert checksums into the metadata in large chunks,
1937  * we wait until bio submission time.   All the pages in the bio are
1938  * checksummed and sums are attached onto the ordered extent record.
1939  *
1940  * At IO completion time the cums attached on the ordered extent record
1941  * are inserted into the btree
1942  */
1943 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1944                           int mirror_num, unsigned long bio_flags,
1945                           u64 bio_offset)
1946 {
1947         struct inode *inode = private_data;
1948         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1949         blk_status_t ret;
1950
1951         ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1952         if (ret) {
1953                 bio->bi_status = ret;
1954                 bio_endio(bio);
1955         }
1956         return ret;
1957 }
1958
1959 /*
1960  * extent_io.c submission hook. This does the right thing for csum calculation
1961  * on write, or reading the csums from the tree before a read.
1962  *
1963  * Rules about async/sync submit,
1964  * a) read:                             sync submit
1965  *
1966  * b) write without checksum:           sync submit
1967  *
1968  * c) write with checksum:
1969  *    c-1) if bio is issued by fsync:   sync submit
1970  *         (sync_writers != 0)
1971  *
1972  *    c-2) if root is reloc root:       sync submit
1973  *         (only in case of buffered IO)
1974  *
1975  *    c-3) otherwise:                   async submit
1976  */
1977 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1978                                  int mirror_num, unsigned long bio_flags,
1979                                  u64 bio_offset)
1980 {
1981         struct inode *inode = private_data;
1982         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1983         struct btrfs_root *root = BTRFS_I(inode)->root;
1984         enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1985         blk_status_t ret = 0;
1986         int skip_sum;
1987         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1988
1989         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1990
1991         if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1992                 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1993
1994         if (bio_op(bio) != REQ_OP_WRITE) {
1995                 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1996                 if (ret)
1997                         goto out;
1998
1999                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2000                         ret = btrfs_submit_compressed_read(inode, bio,
2001                                                            mirror_num,
2002                                                            bio_flags);
2003                         goto out;
2004                 } else if (!skip_sum) {
2005                         ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2006                         if (ret)
2007                                 goto out;
2008                 }
2009                 goto mapit;
2010         } else if (async && !skip_sum) {
2011                 /* csum items have already been cloned */
2012                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2013                         goto mapit;
2014                 /* we're doing a write, do the async checksumming */
2015                 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2016                                           bio_offset, inode,
2017                                           __btrfs_submit_bio_start,
2018                                           __btrfs_submit_bio_done);
2019                 goto out;
2020         } else if (!skip_sum) {
2021                 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2022                 if (ret)
2023                         goto out;
2024         }
2025
2026 mapit:
2027         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2028
2029 out:
2030         if (ret) {
2031                 bio->bi_status = ret;
2032                 bio_endio(bio);
2033         }
2034         return ret;
2035 }
2036
2037 /*
2038  * given a list of ordered sums record them in the inode.  This happens
2039  * at IO completion time based on sums calculated at bio submission time.
2040  */
2041 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2042                              struct inode *inode, struct list_head *list)
2043 {
2044         struct btrfs_ordered_sum *sum;
2045
2046         list_for_each_entry(sum, list, list) {
2047                 trans->adding_csums = true;
2048                 btrfs_csum_file_blocks(trans,
2049                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
2050                 trans->adding_csums = false;
2051         }
2052         return 0;
2053 }
2054
2055 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2056                               unsigned int extra_bits,
2057                               struct extent_state **cached_state, int dedupe)
2058 {
2059         WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2060         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2061                                    extra_bits, cached_state);
2062 }
2063
2064 /* see btrfs_writepage_start_hook for details on why this is required */
2065 struct btrfs_writepage_fixup {
2066         struct page *page;
2067         struct btrfs_work work;
2068 };
2069
2070 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2071 {
2072         struct btrfs_writepage_fixup *fixup;
2073         struct btrfs_ordered_extent *ordered;
2074         struct extent_state *cached_state = NULL;
2075         struct extent_changeset *data_reserved = NULL;
2076         struct page *page;
2077         struct inode *inode;
2078         u64 page_start;
2079         u64 page_end;
2080         int ret;
2081
2082         fixup = container_of(work, struct btrfs_writepage_fixup, work);
2083         page = fixup->page;
2084 again:
2085         lock_page(page);
2086         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2087                 ClearPageChecked(page);
2088                 goto out_page;
2089         }
2090
2091         inode = page->mapping->host;
2092         page_start = page_offset(page);
2093         page_end = page_offset(page) + PAGE_SIZE - 1;
2094
2095         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2096                          &cached_state);
2097
2098         /* already ordered? We're done */
2099         if (PagePrivate2(page))
2100                 goto out;
2101
2102         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2103                                         PAGE_SIZE);
2104         if (ordered) {
2105                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2106                                      page_end, &cached_state);
2107                 unlock_page(page);
2108                 btrfs_start_ordered_extent(inode, ordered, 1);
2109                 btrfs_put_ordered_extent(ordered);
2110                 goto again;
2111         }
2112
2113         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2114                                            PAGE_SIZE);
2115         if (ret) {
2116                 mapping_set_error(page->mapping, ret);
2117                 end_extent_writepage(page, ret, page_start, page_end);
2118                 ClearPageChecked(page);
2119                 goto out;
2120          }
2121
2122         ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2123                                         &cached_state, 0);
2124         if (ret) {
2125                 mapping_set_error(page->mapping, ret);
2126                 end_extent_writepage(page, ret, page_start, page_end);
2127                 ClearPageChecked(page);
2128                 goto out;
2129         }
2130
2131         ClearPageChecked(page);
2132         set_page_dirty(page);
2133         btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2134 out:
2135         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2136                              &cached_state);
2137 out_page:
2138         unlock_page(page);
2139         put_page(page);
2140         kfree(fixup);
2141         extent_changeset_free(data_reserved);
2142 }
2143
2144 /*
2145  * There are a few paths in the higher layers of the kernel that directly
2146  * set the page dirty bit without asking the filesystem if it is a
2147  * good idea.  This causes problems because we want to make sure COW
2148  * properly happens and the data=ordered rules are followed.
2149  *
2150  * In our case any range that doesn't have the ORDERED bit set
2151  * hasn't been properly setup for IO.  We kick off an async process
2152  * to fix it up.  The async helper will wait for ordered extents, set
2153  * the delalloc bit and make it safe to write the page.
2154  */
2155 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2156 {
2157         struct inode *inode = page->mapping->host;
2158         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2159         struct btrfs_writepage_fixup *fixup;
2160
2161         /* this page is properly in the ordered list */
2162         if (TestClearPagePrivate2(page))
2163                 return 0;
2164
2165         if (PageChecked(page))
2166                 return -EAGAIN;
2167
2168         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2169         if (!fixup)
2170                 return -EAGAIN;
2171
2172         SetPageChecked(page);
2173         get_page(page);
2174         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2175                         btrfs_writepage_fixup_worker, NULL, NULL);
2176         fixup->page = page;
2177         btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2178         return -EBUSY;
2179 }
2180
2181 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2182                                        struct inode *inode, u64 file_pos,
2183                                        u64 disk_bytenr, u64 disk_num_bytes,
2184                                        u64 num_bytes, u64 ram_bytes,
2185                                        u8 compression, u8 encryption,
2186                                        u16 other_encoding, int extent_type)
2187 {
2188         struct btrfs_root *root = BTRFS_I(inode)->root;
2189         struct btrfs_file_extent_item *fi;
2190         struct btrfs_path *path;
2191         struct extent_buffer *leaf;
2192         struct btrfs_key ins;
2193         u64 qg_released;
2194         int extent_inserted = 0;
2195         int ret;
2196
2197         path = btrfs_alloc_path();
2198         if (!path)
2199                 return -ENOMEM;
2200
2201         /*
2202          * we may be replacing one extent in the tree with another.
2203          * The new extent is pinned in the extent map, and we don't want
2204          * to drop it from the cache until it is completely in the btree.
2205          *
2206          * So, tell btrfs_drop_extents to leave this extent in the cache.
2207          * the caller is expected to unpin it and allow it to be merged
2208          * with the others.
2209          */
2210         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2211                                    file_pos + num_bytes, NULL, 0,
2212                                    1, sizeof(*fi), &extent_inserted);
2213         if (ret)
2214                 goto out;
2215
2216         if (!extent_inserted) {
2217                 ins.objectid = btrfs_ino(BTRFS_I(inode));
2218                 ins.offset = file_pos;
2219                 ins.type = BTRFS_EXTENT_DATA_KEY;
2220
2221                 path->leave_spinning = 1;
2222                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2223                                               sizeof(*fi));
2224                 if (ret)
2225                         goto out;
2226         }
2227         leaf = path->nodes[0];
2228         fi = btrfs_item_ptr(leaf, path->slots[0],
2229                             struct btrfs_file_extent_item);
2230         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2231         btrfs_set_file_extent_type(leaf, fi, extent_type);
2232         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2233         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2234         btrfs_set_file_extent_offset(leaf, fi, 0);
2235         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2236         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2237         btrfs_set_file_extent_compression(leaf, fi, compression);
2238         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2239         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2240
2241         btrfs_mark_buffer_dirty(leaf);
2242         btrfs_release_path(path);
2243
2244         inode_add_bytes(inode, num_bytes);
2245
2246         ins.objectid = disk_bytenr;
2247         ins.offset = disk_num_bytes;
2248         ins.type = BTRFS_EXTENT_ITEM_KEY;
2249
2250         /*
2251          * Release the reserved range from inode dirty range map, as it is
2252          * already moved into delayed_ref_head
2253          */
2254         ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2255         if (ret < 0)
2256                 goto out;
2257         qg_released = ret;
2258         ret = btrfs_alloc_reserved_file_extent(trans, root,
2259                                                btrfs_ino(BTRFS_I(inode)),
2260                                                file_pos, qg_released, &ins);
2261 out:
2262         btrfs_free_path(path);
2263
2264         return ret;
2265 }
2266
2267 /* snapshot-aware defrag */
2268 struct sa_defrag_extent_backref {
2269         struct rb_node node;
2270         struct old_sa_defrag_extent *old;
2271         u64 root_id;
2272         u64 inum;
2273         u64 file_pos;
2274         u64 extent_offset;
2275         u64 num_bytes;
2276         u64 generation;
2277 };
2278
2279 struct old_sa_defrag_extent {
2280         struct list_head list;
2281         struct new_sa_defrag_extent *new;
2282
2283         u64 extent_offset;
2284         u64 bytenr;
2285         u64 offset;
2286         u64 len;
2287         int count;
2288 };
2289
2290 struct new_sa_defrag_extent {
2291         struct rb_root root;
2292         struct list_head head;
2293         struct btrfs_path *path;
2294         struct inode *inode;
2295         u64 file_pos;
2296         u64 len;
2297         u64 bytenr;
2298         u64 disk_len;
2299         u8 compress_type;
2300 };
2301
2302 static int backref_comp(struct sa_defrag_extent_backref *b1,
2303                         struct sa_defrag_extent_backref *b2)
2304 {
2305         if (b1->root_id < b2->root_id)
2306                 return -1;
2307         else if (b1->root_id > b2->root_id)
2308                 return 1;
2309
2310         if (b1->inum < b2->inum)
2311                 return -1;
2312         else if (b1->inum > b2->inum)
2313                 return 1;
2314
2315         if (b1->file_pos < b2->file_pos)
2316                 return -1;
2317         else if (b1->file_pos > b2->file_pos)
2318                 return 1;
2319
2320         /*
2321          * [------------------------------] ===> (a range of space)
2322          *     |<--->|   |<---->| =============> (fs/file tree A)
2323          * |<---------------------------->| ===> (fs/file tree B)
2324          *
2325          * A range of space can refer to two file extents in one tree while
2326          * refer to only one file extent in another tree.
2327          *
2328          * So we may process a disk offset more than one time(two extents in A)
2329          * and locate at the same extent(one extent in B), then insert two same
2330          * backrefs(both refer to the extent in B).
2331          */
2332         return 0;
2333 }
2334
2335 static void backref_insert(struct rb_root *root,
2336                            struct sa_defrag_extent_backref *backref)
2337 {
2338         struct rb_node **p = &root->rb_node;
2339         struct rb_node *parent = NULL;
2340         struct sa_defrag_extent_backref *entry;
2341         int ret;
2342
2343         while (*p) {
2344                 parent = *p;
2345                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2346
2347                 ret = backref_comp(backref, entry);
2348                 if (ret < 0)
2349                         p = &(*p)->rb_left;
2350                 else
2351                         p = &(*p)->rb_right;
2352         }
2353
2354         rb_link_node(&backref->node, parent, p);
2355         rb_insert_color(&backref->node, root);
2356 }
2357
2358 /*
2359  * Note the backref might has changed, and in this case we just return 0.
2360  */
2361 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2362                                        void *ctx)
2363 {
2364         struct btrfs_file_extent_item *extent;
2365         struct old_sa_defrag_extent *old = ctx;
2366         struct new_sa_defrag_extent *new = old->new;
2367         struct btrfs_path *path = new->path;
2368         struct btrfs_key key;
2369         struct btrfs_root *root;
2370         struct sa_defrag_extent_backref *backref;
2371         struct extent_buffer *leaf;
2372         struct inode *inode = new->inode;
2373         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2374         int slot;
2375         int ret;
2376         u64 extent_offset;
2377         u64 num_bytes;
2378
2379         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2380             inum == btrfs_ino(BTRFS_I(inode)))
2381                 return 0;
2382
2383         key.objectid = root_id;
2384         key.type = BTRFS_ROOT_ITEM_KEY;
2385         key.offset = (u64)-1;
2386
2387         root = btrfs_read_fs_root_no_name(fs_info, &key);
2388         if (IS_ERR(root)) {
2389                 if (PTR_ERR(root) == -ENOENT)
2390                         return 0;
2391                 WARN_ON(1);
2392                 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2393                          inum, offset, root_id);
2394                 return PTR_ERR(root);
2395         }
2396
2397         key.objectid = inum;
2398         key.type = BTRFS_EXTENT_DATA_KEY;
2399         if (offset > (u64)-1 << 32)
2400                 key.offset = 0;
2401         else
2402                 key.offset = offset;
2403
2404         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2405         if (WARN_ON(ret < 0))
2406                 return ret;
2407         ret = 0;
2408
2409         while (1) {
2410                 cond_resched();
2411
2412                 leaf = path->nodes[0];
2413                 slot = path->slots[0];
2414
2415                 if (slot >= btrfs_header_nritems(leaf)) {
2416                         ret = btrfs_next_leaf(root, path);
2417                         if (ret < 0) {
2418                                 goto out;
2419                         } else if (ret > 0) {
2420                                 ret = 0;
2421                                 goto out;
2422                         }
2423                         continue;
2424                 }
2425
2426                 path->slots[0]++;
2427
2428                 btrfs_item_key_to_cpu(leaf, &key, slot);
2429
2430                 if (key.objectid > inum)
2431                         goto out;
2432
2433                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2434                         continue;
2435
2436                 extent = btrfs_item_ptr(leaf, slot,
2437                                         struct btrfs_file_extent_item);
2438
2439                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2440                         continue;
2441
2442                 /*
2443                  * 'offset' refers to the exact key.offset,
2444                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2445                  * (key.offset - extent_offset).
2446                  */
2447                 if (key.offset != offset)
2448                         continue;
2449
2450                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2451                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2452
2453                 if (extent_offset >= old->extent_offset + old->offset +
2454                     old->len || extent_offset + num_bytes <=
2455                     old->extent_offset + old->offset)
2456                         continue;
2457                 break;
2458         }
2459
2460         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2461         if (!backref) {
2462                 ret = -ENOENT;
2463                 goto out;
2464         }
2465
2466         backref->root_id = root_id;
2467         backref->inum = inum;
2468         backref->file_pos = offset;
2469         backref->num_bytes = num_bytes;
2470         backref->extent_offset = extent_offset;
2471         backref->generation = btrfs_file_extent_generation(leaf, extent);
2472         backref->old = old;
2473         backref_insert(&new->root, backref);
2474         old->count++;
2475 out:
2476         btrfs_release_path(path);
2477         WARN_ON(ret);
2478         return ret;
2479 }
2480
2481 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2482                                    struct new_sa_defrag_extent *new)
2483 {
2484         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2485         struct old_sa_defrag_extent *old, *tmp;
2486         int ret;
2487
2488         new->path = path;
2489
2490         list_for_each_entry_safe(old, tmp, &new->head, list) {
2491                 ret = iterate_inodes_from_logical(old->bytenr +
2492                                                   old->extent_offset, fs_info,
2493                                                   path, record_one_backref,
2494                                                   old, false);
2495                 if (ret < 0 && ret != -ENOENT)
2496                         return false;
2497
2498                 /* no backref to be processed for this extent */
2499                 if (!old->count) {
2500                         list_del(&old->list);
2501                         kfree(old);
2502                 }
2503         }
2504
2505         if (list_empty(&new->head))
2506                 return false;
2507
2508         return true;
2509 }
2510
2511 static int relink_is_mergable(struct extent_buffer *leaf,
2512                               struct btrfs_file_extent_item *fi,
2513                               struct new_sa_defrag_extent *new)
2514 {
2515         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2516                 return 0;
2517
2518         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2519                 return 0;
2520
2521         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2522                 return 0;
2523
2524         if (btrfs_file_extent_encryption(leaf, fi) ||
2525             btrfs_file_extent_other_encoding(leaf, fi))
2526                 return 0;
2527
2528         return 1;
2529 }
2530
2531 /*
2532  * Note the backref might has changed, and in this case we just return 0.
2533  */
2534 static noinline int relink_extent_backref(struct btrfs_path *path,
2535                                  struct sa_defrag_extent_backref *prev,
2536                                  struct sa_defrag_extent_backref *backref)
2537 {
2538         struct btrfs_file_extent_item *extent;
2539         struct btrfs_file_extent_item *item;
2540         struct btrfs_ordered_extent *ordered;
2541         struct btrfs_trans_handle *trans;
2542         struct btrfs_root *root;
2543         struct btrfs_key key;
2544         struct extent_buffer *leaf;
2545         struct old_sa_defrag_extent *old = backref->old;
2546         struct new_sa_defrag_extent *new = old->new;
2547         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2548         struct inode *inode;
2549         struct extent_state *cached = NULL;
2550         int ret = 0;
2551         u64 start;
2552         u64 len;
2553         u64 lock_start;
2554         u64 lock_end;
2555         bool merge = false;
2556         int index;
2557
2558         if (prev && prev->root_id == backref->root_id &&
2559             prev->inum == backref->inum &&
2560             prev->file_pos + prev->num_bytes == backref->file_pos)
2561                 merge = true;
2562
2563         /* step 1: get root */
2564         key.objectid = backref->root_id;
2565         key.type = BTRFS_ROOT_ITEM_KEY;
2566         key.offset = (u64)-1;
2567
2568         index = srcu_read_lock(&fs_info->subvol_srcu);
2569
2570         root = btrfs_read_fs_root_no_name(fs_info, &key);
2571         if (IS_ERR(root)) {
2572                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2573                 if (PTR_ERR(root) == -ENOENT)
2574                         return 0;
2575                 return PTR_ERR(root);
2576         }
2577
2578         if (btrfs_root_readonly(root)) {
2579                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2580                 return 0;
2581         }
2582
2583         /* step 2: get inode */
2584         key.objectid = backref->inum;
2585         key.type = BTRFS_INODE_ITEM_KEY;
2586         key.offset = 0;
2587
2588         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2589         if (IS_ERR(inode)) {
2590                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2591                 return 0;
2592         }
2593
2594         srcu_read_unlock(&fs_info->subvol_srcu, index);
2595
2596         /* step 3: relink backref */
2597         lock_start = backref->file_pos;
2598         lock_end = backref->file_pos + backref->num_bytes - 1;
2599         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2600                          &cached);
2601
2602         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2603         if (ordered) {
2604                 btrfs_put_ordered_extent(ordered);
2605                 goto out_unlock;
2606         }
2607
2608         trans = btrfs_join_transaction(root);
2609         if (IS_ERR(trans)) {
2610                 ret = PTR_ERR(trans);
2611                 goto out_unlock;
2612         }
2613
2614         key.objectid = backref->inum;
2615         key.type = BTRFS_EXTENT_DATA_KEY;
2616         key.offset = backref->file_pos;
2617
2618         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2619         if (ret < 0) {
2620                 goto out_free_path;
2621         } else if (ret > 0) {
2622                 ret = 0;
2623                 goto out_free_path;
2624         }
2625
2626         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2627                                 struct btrfs_file_extent_item);
2628
2629         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2630             backref->generation)
2631                 goto out_free_path;
2632
2633         btrfs_release_path(path);
2634
2635         start = backref->file_pos;
2636         if (backref->extent_offset < old->extent_offset + old->offset)
2637                 start += old->extent_offset + old->offset -
2638                          backref->extent_offset;
2639
2640         len = min(backref->extent_offset + backref->num_bytes,
2641                   old->extent_offset + old->offset + old->len);
2642         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2643
2644         ret = btrfs_drop_extents(trans, root, inode, start,
2645                                  start + len, 1);
2646         if (ret)
2647                 goto out_free_path;
2648 again:
2649         key.objectid = btrfs_ino(BTRFS_I(inode));
2650         key.type = BTRFS_EXTENT_DATA_KEY;
2651         key.offset = start;
2652
2653         path->leave_spinning = 1;
2654         if (merge) {
2655                 struct btrfs_file_extent_item *fi;
2656                 u64 extent_len;
2657                 struct btrfs_key found_key;
2658
2659                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2660                 if (ret < 0)
2661                         goto out_free_path;
2662
2663                 path->slots[0]--;
2664                 leaf = path->nodes[0];
2665                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2666
2667                 fi = btrfs_item_ptr(leaf, path->slots[0],
2668                                     struct btrfs_file_extent_item);
2669                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2670
2671                 if (extent_len + found_key.offset == start &&
2672                     relink_is_mergable(leaf, fi, new)) {
2673                         btrfs_set_file_extent_num_bytes(leaf, fi,
2674                                                         extent_len + len);
2675                         btrfs_mark_buffer_dirty(leaf);
2676                         inode_add_bytes(inode, len);
2677
2678                         ret = 1;
2679                         goto out_free_path;
2680                 } else {
2681                         merge = false;
2682                         btrfs_release_path(path);
2683                         goto again;
2684                 }
2685         }
2686
2687         ret = btrfs_insert_empty_item(trans, root, path, &key,
2688                                         sizeof(*extent));
2689         if (ret) {
2690                 btrfs_abort_transaction(trans, ret);
2691                 goto out_free_path;
2692         }
2693
2694         leaf = path->nodes[0];
2695         item = btrfs_item_ptr(leaf, path->slots[0],
2696                                 struct btrfs_file_extent_item);
2697         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2698         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2699         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2700         btrfs_set_file_extent_num_bytes(leaf, item, len);
2701         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2702         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2703         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2704         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2705         btrfs_set_file_extent_encryption(leaf, item, 0);
2706         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2707
2708         btrfs_mark_buffer_dirty(leaf);
2709         inode_add_bytes(inode, len);
2710         btrfs_release_path(path);
2711
2712         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2713                         new->disk_len, 0,
2714                         backref->root_id, backref->inum,
2715                         new->file_pos); /* start - extent_offset */
2716         if (ret) {
2717                 btrfs_abort_transaction(trans, ret);
2718                 goto out_free_path;
2719         }
2720
2721         ret = 1;
2722 out_free_path:
2723         btrfs_release_path(path);
2724         path->leave_spinning = 0;
2725         btrfs_end_transaction(trans);
2726 out_unlock:
2727         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2728                              &cached);
2729         iput(inode);
2730         return ret;
2731 }
2732
2733 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2734 {
2735         struct old_sa_defrag_extent *old, *tmp;
2736
2737         if (!new)
2738                 return;
2739
2740         list_for_each_entry_safe(old, tmp, &new->head, list) {
2741                 kfree(old);
2742         }
2743         kfree(new);
2744 }
2745
2746 static void relink_file_extents(struct new_sa_defrag_extent *new)
2747 {
2748         struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2749         struct btrfs_path *path;
2750         struct sa_defrag_extent_backref *backref;
2751         struct sa_defrag_extent_backref *prev = NULL;
2752         struct inode *inode;
2753         struct btrfs_root *root;
2754         struct rb_node *node;
2755         int ret;
2756
2757         inode = new->inode;
2758         root = BTRFS_I(inode)->root;
2759
2760         path = btrfs_alloc_path();
2761         if (!path)
2762                 return;
2763
2764         if (!record_extent_backrefs(path, new)) {
2765                 btrfs_free_path(path);
2766                 goto out;
2767         }
2768         btrfs_release_path(path);
2769
2770         while (1) {
2771                 node = rb_first(&new->root);
2772                 if (!node)
2773                         break;
2774                 rb_erase(node, &new->root);
2775
2776                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2777
2778                 ret = relink_extent_backref(path, prev, backref);
2779                 WARN_ON(ret < 0);
2780
2781                 kfree(prev);
2782
2783                 if (ret == 1)
2784                         prev = backref;
2785                 else
2786                         prev = NULL;
2787                 cond_resched();
2788         }
2789         kfree(prev);
2790
2791         btrfs_free_path(path);
2792 out:
2793         free_sa_defrag_extent(new);
2794
2795         atomic_dec(&fs_info->defrag_running);
2796         wake_up(&fs_info->transaction_wait);
2797 }
2798
2799 static struct new_sa_defrag_extent *
2800 record_old_file_extents(struct inode *inode,
2801                         struct btrfs_ordered_extent *ordered)
2802 {
2803         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2804         struct btrfs_root *root = BTRFS_I(inode)->root;
2805         struct btrfs_path *path;
2806         struct btrfs_key key;
2807         struct old_sa_defrag_extent *old;
2808         struct new_sa_defrag_extent *new;
2809         int ret;
2810
2811         new = kmalloc(sizeof(*new), GFP_NOFS);
2812         if (!new)
2813                 return NULL;
2814
2815         new->inode = inode;
2816         new->file_pos = ordered->file_offset;
2817         new->len = ordered->len;
2818         new->bytenr = ordered->start;
2819         new->disk_len = ordered->disk_len;
2820         new->compress_type = ordered->compress_type;
2821         new->root = RB_ROOT;
2822         INIT_LIST_HEAD(&new->head);
2823
2824         path = btrfs_alloc_path();
2825         if (!path)
2826                 goto out_kfree;
2827
2828         key.objectid = btrfs_ino(BTRFS_I(inode));
2829         key.type = BTRFS_EXTENT_DATA_KEY;
2830         key.offset = new->file_pos;
2831
2832         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2833         if (ret < 0)
2834                 goto out_free_path;
2835         if (ret > 0 && path->slots[0] > 0)
2836                 path->slots[0]--;
2837
2838         /* find out all the old extents for the file range */
2839         while (1) {
2840                 struct btrfs_file_extent_item *extent;
2841                 struct extent_buffer *l;
2842                 int slot;
2843                 u64 num_bytes;
2844                 u64 offset;
2845                 u64 end;
2846                 u64 disk_bytenr;
2847                 u64 extent_offset;
2848
2849                 l = path->nodes[0];
2850                 slot = path->slots[0];
2851
2852                 if (slot >= btrfs_header_nritems(l)) {
2853                         ret = btrfs_next_leaf(root, path);
2854                         if (ret < 0)
2855                                 goto out_free_path;
2856                         else if (ret > 0)
2857                                 break;
2858                         continue;
2859                 }
2860
2861                 btrfs_item_key_to_cpu(l, &key, slot);
2862
2863                 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2864                         break;
2865                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2866                         break;
2867                 if (key.offset >= new->file_pos + new->len)
2868                         break;
2869
2870                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2871
2872                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2873                 if (key.offset + num_bytes < new->file_pos)
2874                         goto next;
2875
2876                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2877                 if (!disk_bytenr)
2878                         goto next;
2879
2880                 extent_offset = btrfs_file_extent_offset(l, extent);
2881
2882                 old = kmalloc(sizeof(*old), GFP_NOFS);
2883                 if (!old)
2884                         goto out_free_path;
2885
2886                 offset = max(new->file_pos, key.offset);
2887                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2888
2889                 old->bytenr = disk_bytenr;
2890                 old->extent_offset = extent_offset;
2891                 old->offset = offset - key.offset;
2892                 old->len = end - offset;
2893                 old->new = new;
2894                 old->count = 0;
2895                 list_add_tail(&old->list, &new->head);
2896 next:
2897                 path->slots[0]++;
2898                 cond_resched();
2899         }
2900
2901         btrfs_free_path(path);
2902         atomic_inc(&fs_info->defrag_running);
2903
2904         return new;
2905
2906 out_free_path:
2907         btrfs_free_path(path);
2908 out_kfree:
2909         free_sa_defrag_extent(new);
2910         return NULL;
2911 }
2912
2913 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2914                                          u64 start, u64 len)
2915 {
2916         struct btrfs_block_group_cache *cache;
2917
2918         cache = btrfs_lookup_block_group(fs_info, start);
2919         ASSERT(cache);
2920
2921         spin_lock(&cache->lock);
2922         cache->delalloc_bytes -= len;
2923         spin_unlock(&cache->lock);
2924
2925         btrfs_put_block_group(cache);
2926 }
2927
2928 /* as ordered data IO finishes, this gets called so we can finish
2929  * an ordered extent if the range of bytes in the file it covers are
2930  * fully written.
2931  */
2932 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2933 {
2934         struct inode *inode = ordered_extent->inode;
2935         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2936         struct btrfs_root *root = BTRFS_I(inode)->root;
2937         struct btrfs_trans_handle *trans = NULL;
2938         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2939         struct extent_state *cached_state = NULL;
2940         struct new_sa_defrag_extent *new = NULL;
2941         int compress_type = 0;
2942         int ret = 0;
2943         u64 logical_len = ordered_extent->len;
2944         bool nolock;
2945         bool truncated = false;
2946         bool range_locked = false;
2947         bool clear_new_delalloc_bytes = false;
2948
2949         if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2950             !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2951             !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2952                 clear_new_delalloc_bytes = true;
2953
2954         nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2955
2956         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2957                 ret = -EIO;
2958                 goto out;
2959         }
2960
2961         btrfs_free_io_failure_record(BTRFS_I(inode),
2962                         ordered_extent->file_offset,
2963                         ordered_extent->file_offset +
2964                         ordered_extent->len - 1);
2965
2966         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2967                 truncated = true;
2968                 logical_len = ordered_extent->truncated_len;
2969                 /* Truncated the entire extent, don't bother adding */
2970                 if (!logical_len)
2971                         goto out;
2972         }
2973
2974         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2975                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2976
2977                 /*
2978                  * For mwrite(mmap + memset to write) case, we still reserve
2979                  * space for NOCOW range.
2980                  * As NOCOW won't cause a new delayed ref, just free the space
2981                  */
2982                 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2983                                        ordered_extent->len);
2984                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2985                 if (nolock)
2986                         trans = btrfs_join_transaction_nolock(root);
2987                 else
2988                         trans = btrfs_join_transaction(root);
2989                 if (IS_ERR(trans)) {
2990                         ret = PTR_ERR(trans);
2991                         trans = NULL;
2992                         goto out;
2993                 }
2994                 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2995                 ret = btrfs_update_inode_fallback(trans, root, inode);
2996                 if (ret) /* -ENOMEM or corruption */
2997                         btrfs_abort_transaction(trans, ret);
2998                 goto out;
2999         }
3000
3001         range_locked = true;
3002         lock_extent_bits(io_tree, ordered_extent->file_offset,
3003                          ordered_extent->file_offset + ordered_extent->len - 1,
3004                          &cached_state);
3005
3006         ret = test_range_bit(io_tree, ordered_extent->file_offset,
3007                         ordered_extent->file_offset + ordered_extent->len - 1,
3008                         EXTENT_DEFRAG, 0, cached_state);
3009         if (ret) {
3010                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3011                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3012                         /* the inode is shared */
3013                         new = record_old_file_extents(inode, ordered_extent);
3014
3015                 clear_extent_bit(io_tree, ordered_extent->file_offset,
3016                         ordered_extent->file_offset + ordered_extent->len - 1,
3017                         EXTENT_DEFRAG, 0, 0, &cached_state);
3018         }
3019
3020         if (nolock)
3021                 trans = btrfs_join_transaction_nolock(root);
3022         else
3023                 trans = btrfs_join_transaction(root);
3024         if (IS_ERR(trans)) {
3025                 ret = PTR_ERR(trans);
3026                 trans = NULL;
3027                 goto out;
3028         }
3029
3030         trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3031
3032         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3033                 compress_type = ordered_extent->compress_type;
3034         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3035                 BUG_ON(compress_type);
3036                 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3037                                        ordered_extent->len);
3038                 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3039                                                 ordered_extent->file_offset,
3040                                                 ordered_extent->file_offset +
3041                                                 logical_len);
3042         } else {
3043                 BUG_ON(root == fs_info->tree_root);
3044                 ret = insert_reserved_file_extent(trans, inode,
3045                                                 ordered_extent->file_offset,
3046                                                 ordered_extent->start,
3047                                                 ordered_extent->disk_len,
3048                                                 logical_len, logical_len,
3049                                                 compress_type, 0, 0,
3050                                                 BTRFS_FILE_EXTENT_REG);
3051                 if (!ret)
3052                         btrfs_release_delalloc_bytes(fs_info,
3053                                                      ordered_extent->start,
3054                                                      ordered_extent->disk_len);
3055         }
3056         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3057                            ordered_extent->file_offset, ordered_extent->len,
3058                            trans->transid);
3059         if (ret < 0) {
3060                 btrfs_abort_transaction(trans, ret);
3061                 goto out;
3062         }
3063
3064         add_pending_csums(trans, inode, &ordered_extent->list);
3065
3066         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3067         ret = btrfs_update_inode_fallback(trans, root, inode);
3068         if (ret) { /* -ENOMEM or corruption */
3069                 btrfs_abort_transaction(trans, ret);
3070                 goto out;
3071         }
3072         ret = 0;
3073 out:
3074         if (range_locked || clear_new_delalloc_bytes) {
3075                 unsigned int clear_bits = 0;
3076
3077                 if (range_locked)
3078                         clear_bits |= EXTENT_LOCKED;
3079                 if (clear_new_delalloc_bytes)
3080                         clear_bits |= EXTENT_DELALLOC_NEW;
3081                 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3082                                  ordered_extent->file_offset,
3083                                  ordered_extent->file_offset +
3084                                  ordered_extent->len - 1,
3085                                  clear_bits,
3086                                  (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3087                                  0, &cached_state);
3088         }
3089
3090         if (trans)
3091                 btrfs_end_transaction(trans);
3092
3093         if (ret || truncated) {
3094                 u64 start, end;
3095
3096                 if (truncated)
3097                         start = ordered_extent->file_offset + logical_len;
3098                 else
3099                         start = ordered_extent->file_offset;
3100                 end = ordered_extent->file_offset + ordered_extent->len - 1;
3101                 clear_extent_uptodate(io_tree, start, end, NULL);
3102
3103                 /* Drop the cache for the part of the extent we didn't write. */
3104                 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3105
3106                 /*
3107                  * If the ordered extent had an IOERR or something else went
3108                  * wrong we need to return the space for this ordered extent
3109                  * back to the allocator.  We only free the extent in the
3110                  * truncated case if we didn't write out the extent at all.
3111                  */
3112                 if ((ret || !logical_len) &&
3113                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3114                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3115                         btrfs_free_reserved_extent(fs_info,
3116                                                    ordered_extent->start,
3117                                                    ordered_extent->disk_len, 1);
3118         }
3119
3120
3121         /*
3122          * This needs to be done to make sure anybody waiting knows we are done
3123          * updating everything for this ordered extent.
3124          */
3125         btrfs_remove_ordered_extent(inode, ordered_extent);
3126
3127         /* for snapshot-aware defrag */
3128         if (new) {
3129                 if (ret) {
3130                         free_sa_defrag_extent(new);
3131                         atomic_dec(&fs_info->defrag_running);
3132                 } else {
3133                         relink_file_extents(new);
3134                 }
3135         }
3136
3137         /* once for us */
3138         btrfs_put_ordered_extent(ordered_extent);
3139         /* once for the tree */
3140         btrfs_put_ordered_extent(ordered_extent);
3141
3142         return ret;
3143 }
3144
3145 static void finish_ordered_fn(struct btrfs_work *work)
3146 {
3147         struct btrfs_ordered_extent *ordered_extent;
3148         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3149         btrfs_finish_ordered_io(ordered_extent);
3150 }
3151
3152 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3153                                 struct extent_state *state, int uptodate)
3154 {
3155         struct inode *inode = page->mapping->host;
3156         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3157         struct btrfs_ordered_extent *ordered_extent = NULL;
3158         struct btrfs_workqueue *wq;
3159         btrfs_work_func_t func;
3160
3161         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3162
3163         ClearPagePrivate2(page);
3164         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3165                                             end - start + 1, uptodate))
3166                 return;
3167
3168         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3169                 wq = fs_info->endio_freespace_worker;
3170                 func = btrfs_freespace_write_helper;
3171         } else {
3172                 wq = fs_info->endio_write_workers;
3173                 func = btrfs_endio_write_helper;
3174         }
3175
3176         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3177                         NULL);
3178         btrfs_queue_work(wq, &ordered_extent->work);
3179 }
3180
3181 static int __readpage_endio_check(struct inode *inode,
3182                                   struct btrfs_io_bio *io_bio,
3183                                   int icsum, struct page *page,
3184                                   int pgoff, u64 start, size_t len)
3185 {
3186         char *kaddr;
3187         u32 csum_expected;
3188         u32 csum = ~(u32)0;
3189
3190         csum_expected = *(((u32 *)io_bio->csum) + icsum);
3191
3192         kaddr = kmap_atomic(page);
3193         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3194         btrfs_csum_final(csum, (u8 *)&csum);
3195         if (csum != csum_expected)
3196                 goto zeroit;
3197
3198         kunmap_atomic(kaddr);
3199         return 0;
3200 zeroit:
3201         btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3202                                     io_bio->mirror_num);
3203         memset(kaddr + pgoff, 1, len);
3204         flush_dcache_page(page);
3205         kunmap_atomic(kaddr);
3206         return -EIO;
3207 }
3208
3209 /*
3210  * when reads are done, we need to check csums to verify the data is correct
3211  * if there's a match, we allow the bio to finish.  If not, the code in
3212  * extent_io.c will try to find good copies for us.
3213  */
3214 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3215                                       u64 phy_offset, struct page *page,
3216                                       u64 start, u64 end, int mirror)
3217 {
3218         size_t offset = start - page_offset(page);
3219         struct inode *inode = page->mapping->host;
3220         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3221         struct btrfs_root *root = BTRFS_I(inode)->root;
3222
3223         if (PageChecked(page)) {
3224                 ClearPageChecked(page);
3225                 return 0;
3226         }
3227
3228         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3229                 return 0;
3230
3231         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3232             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3233                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3234                 return 0;
3235         }
3236
3237         phy_offset >>= inode->i_sb->s_blocksize_bits;
3238         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3239                                       start, (size_t)(end - start + 1));
3240 }
3241
3242 void btrfs_add_delayed_iput(struct inode *inode)
3243 {
3244         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3245         struct btrfs_inode *binode = BTRFS_I(inode);
3246
3247         if (atomic_add_unless(&inode->i_count, -1, 1))
3248                 return;
3249
3250         spin_lock(&fs_info->delayed_iput_lock);
3251         if (binode->delayed_iput_count == 0) {
3252                 ASSERT(list_empty(&binode->delayed_iput));
3253                 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3254         } else {
3255                 binode->delayed_iput_count++;
3256         }
3257         spin_unlock(&fs_info->delayed_iput_lock);
3258 }
3259
3260 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3261 {
3262
3263         spin_lock(&fs_info->delayed_iput_lock);
3264         while (!list_empty(&fs_info->delayed_iputs)) {
3265                 struct btrfs_inode *inode;
3266
3267                 inode = list_first_entry(&fs_info->delayed_iputs,
3268                                 struct btrfs_inode, delayed_iput);
3269                 if (inode->delayed_iput_count) {
3270                         inode->delayed_iput_count--;
3271                         list_move_tail(&inode->delayed_iput,
3272                                         &fs_info->delayed_iputs);
3273                 } else {
3274                         list_del_init(&inode->delayed_iput);
3275                 }
3276                 spin_unlock(&fs_info->delayed_iput_lock);
3277                 iput(&inode->vfs_inode);
3278                 spin_lock(&fs_info->delayed_iput_lock);
3279         }
3280         spin_unlock(&fs_info->delayed_iput_lock);
3281 }
3282
3283 /*
3284  * This is called in transaction commit time. If there are no orphan
3285  * files in the subvolume, it removes orphan item and frees block_rsv
3286  * structure.
3287  */
3288 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3289                               struct btrfs_root *root)
3290 {
3291         struct btrfs_fs_info *fs_info = root->fs_info;
3292         struct btrfs_block_rsv *block_rsv;
3293         int ret;
3294
3295         if (atomic_read(&root->orphan_inodes) ||
3296             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3297                 return;
3298
3299         spin_lock(&root->orphan_lock);
3300         if (atomic_read(&root->orphan_inodes)) {
3301                 spin_unlock(&root->orphan_lock);
3302                 return;
3303         }
3304
3305         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3306                 spin_unlock(&root->orphan_lock);
3307                 return;
3308         }
3309
3310         block_rsv = root->orphan_block_rsv;
3311         root->orphan_block_rsv = NULL;
3312         spin_unlock(&root->orphan_lock);
3313
3314         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3315             btrfs_root_refs(&root->root_item) > 0) {
3316                 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3317                                             root->root_key.objectid);
3318                 if (ret)
3319                         btrfs_abort_transaction(trans, ret);
3320                 else
3321                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3322                                   &root->state);
3323         }
3324
3325         if (block_rsv) {
3326                 WARN_ON(block_rsv->size > 0);
3327                 btrfs_free_block_rsv(fs_info, block_rsv);
3328         }
3329 }
3330
3331 /*
3332  * This creates an orphan entry for the given inode in case something goes
3333  * wrong in the middle of an unlink/truncate.
3334  *
3335  * NOTE: caller of this function should reserve 5 units of metadata for
3336  *       this function.
3337  */
3338 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3339                 struct btrfs_inode *inode)
3340 {
3341         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3342         struct btrfs_root *root = inode->root;
3343         struct btrfs_block_rsv *block_rsv = NULL;
3344         int reserve = 0;
3345         int insert = 0;
3346         int ret;
3347
3348         if (!root->orphan_block_rsv) {
3349                 block_rsv = btrfs_alloc_block_rsv(fs_info,
3350                                                   BTRFS_BLOCK_RSV_TEMP);
3351                 if (!block_rsv)
3352                         return -ENOMEM;
3353         }
3354
3355         spin_lock(&root->orphan_lock);
3356         if (!root->orphan_block_rsv) {
3357                 root->orphan_block_rsv = block_rsv;
3358         } else if (block_rsv) {
3359                 btrfs_free_block_rsv(fs_info, block_rsv);
3360                 block_rsv = NULL;
3361         }
3362
3363         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3364                               &inode->runtime_flags)) {
3365 #if 0
3366                 /*
3367                  * For proper ENOSPC handling, we should do orphan
3368                  * cleanup when mounting. But this introduces backward
3369                  * compatibility issue.
3370                  */
3371                 if (!xchg(&root->orphan_item_inserted, 1))
3372                         insert = 2;
3373                 else
3374                         insert = 1;
3375 #endif
3376                 insert = 1;
3377                 atomic_inc(&root->orphan_inodes);
3378         }
3379
3380         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3381                               &inode->runtime_flags))
3382                 reserve = 1;
3383         spin_unlock(&root->orphan_lock);
3384
3385         /* grab metadata reservation from transaction handle */
3386         if (reserve) {
3387                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3388                 ASSERT(!ret);
3389                 if (ret) {
3390                         /*
3391                          * dec doesn't need spin_lock as ->orphan_block_rsv
3392                          * would be released only if ->orphan_inodes is
3393                          * zero.
3394                          */
3395                         atomic_dec(&root->orphan_inodes);
3396                         clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3397                                   &inode->runtime_flags);
3398                         if (insert)
3399                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3400                                           &inode->runtime_flags);
3401                         return ret;
3402                 }
3403         }
3404
3405         /* insert an orphan item to track this unlinked/truncated file */
3406         if (insert >= 1) {
3407                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3408                 if (ret) {
3409                         if (reserve) {
3410                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3411                                           &inode->runtime_flags);
3412                                 btrfs_orphan_release_metadata(inode);
3413                         }
3414                         /*
3415                          * btrfs_orphan_commit_root may race with us and set
3416                          * ->orphan_block_rsv to zero, in order to avoid that,
3417                          * decrease ->orphan_inodes after everything is done.
3418                          */
3419                         atomic_dec(&root->orphan_inodes);
3420                         if (ret != -EEXIST) {
3421                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3422                                           &inode->runtime_flags);
3423                                 btrfs_abort_transaction(trans, ret);
3424                                 return ret;
3425                         }
3426                 }
3427                 ret = 0;
3428         }
3429
3430         /* insert an orphan item to track subvolume contains orphan files */
3431         if (insert >= 2) {
3432                 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3433                                                root->root_key.objectid);
3434                 if (ret && ret != -EEXIST) {
3435                         btrfs_abort_transaction(trans, ret);
3436                         return ret;
3437                 }
3438         }
3439         return 0;
3440 }
3441
3442 /*
3443  * We have done the truncate/delete so we can go ahead and remove the orphan
3444  * item for this particular inode.
3445  */
3446 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3447                             struct btrfs_inode *inode)
3448 {
3449         struct btrfs_root *root = inode->root;
3450         int delete_item = 0;
3451         int ret = 0;
3452
3453         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3454                                &inode->runtime_flags))
3455                 delete_item = 1;
3456
3457         if (delete_item && trans)
3458                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3459
3460         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3461                                &inode->runtime_flags))
3462                 btrfs_orphan_release_metadata(inode);
3463
3464         /*
3465          * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3466          * to zero, in order to avoid that, decrease ->orphan_inodes after
3467          * everything is done.
3468          */
3469         if (delete_item)
3470                 atomic_dec(&root->orphan_inodes);
3471
3472         return ret;
3473 }
3474
3475 /*
3476  * this cleans up any orphans that may be left on the list from the last use
3477  * of this root.
3478  */
3479 int btrfs_orphan_cleanup(struct btrfs_root *root)
3480 {
3481         struct btrfs_fs_info *fs_info = root->fs_info;
3482         struct btrfs_path *path;
3483         struct extent_buffer *leaf;
3484         struct btrfs_key key, found_key;
3485         struct btrfs_trans_handle *trans;
3486         struct inode *inode;
3487         u64 last_objectid = 0;
3488         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3489
3490         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3491                 return 0;
3492
3493         path = btrfs_alloc_path();
3494         if (!path) {
3495                 ret = -ENOMEM;
3496                 goto out;
3497         }
3498         path->reada = READA_BACK;
3499
3500         key.objectid = BTRFS_ORPHAN_OBJECTID;
3501         key.type = BTRFS_ORPHAN_ITEM_KEY;
3502         key.offset = (u64)-1;
3503
3504         while (1) {
3505                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3506                 if (ret < 0)
3507                         goto out;
3508
3509                 /*
3510                  * if ret == 0 means we found what we were searching for, which
3511                  * is weird, but possible, so only screw with path if we didn't
3512                  * find the key and see if we have stuff that matches
3513                  */
3514                 if (ret > 0) {
3515                         ret = 0;
3516                         if (path->slots[0] == 0)
3517                                 break;
3518                         path->slots[0]--;
3519                 }
3520
3521                 /* pull out the item */
3522                 leaf = path->nodes[0];
3523                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3524
3525                 /* make sure the item matches what we want */
3526                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3527                         break;
3528                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3529                         break;
3530
3531                 /* release the path since we're done with it */
3532                 btrfs_release_path(path);
3533
3534                 /*
3535                  * this is where we are basically btrfs_lookup, without the
3536                  * crossing root thing.  we store the inode number in the
3537                  * offset of the orphan item.
3538                  */
3539
3540                 if (found_key.offset == last_objectid) {
3541                         btrfs_err(fs_info,
3542                                   "Error removing orphan entry, stopping orphan cleanup");
3543                         ret = -EINVAL;
3544                         goto out;
3545                 }
3546
3547                 last_objectid = found_key.offset;
3548
3549                 found_key.objectid = found_key.offset;
3550                 found_key.type = BTRFS_INODE_ITEM_KEY;
3551                 found_key.offset = 0;
3552                 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3553                 ret = PTR_ERR_OR_ZERO(inode);
3554                 if (ret && ret != -ENOENT)
3555                         goto out;
3556
3557                 if (ret == -ENOENT && root == fs_info->tree_root) {
3558                         struct btrfs_root *dead_root;
3559                         struct btrfs_fs_info *fs_info = root->fs_info;
3560                         int is_dead_root = 0;
3561
3562                         /*
3563                          * this is an orphan in the tree root. Currently these
3564                          * could come from 2 sources:
3565                          *  a) a snapshot deletion in progress
3566                          *  b) a free space cache inode
3567                          * We need to distinguish those two, as the snapshot
3568                          * orphan must not get deleted.
3569                          * find_dead_roots already ran before us, so if this
3570                          * is a snapshot deletion, we should find the root
3571                          * in the dead_roots list
3572                          */
3573                         spin_lock(&fs_info->trans_lock);
3574                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3575                                             root_list) {
3576                                 if (dead_root->root_key.objectid ==
3577                                     found_key.objectid) {
3578                                         is_dead_root = 1;
3579                                         break;
3580                                 }
3581                         }
3582                         spin_unlock(&fs_info->trans_lock);
3583                         if (is_dead_root) {
3584                                 /* prevent this orphan from being found again */
3585                                 key.offset = found_key.objectid - 1;
3586                                 continue;
3587                         }
3588                 }
3589                 /*
3590                  * Inode is already gone but the orphan item is still there,
3591                  * kill the orphan item.
3592                  */
3593                 if (ret == -ENOENT) {
3594                         trans = btrfs_start_transaction(root, 1);
3595                         if (IS_ERR(trans)) {
3596                                 ret = PTR_ERR(trans);
3597                                 goto out;
3598                         }
3599                         btrfs_debug(fs_info, "auto deleting %Lu",
3600                                     found_key.objectid);
3601                         ret = btrfs_del_orphan_item(trans, root,
3602                                                     found_key.objectid);
3603                         btrfs_end_transaction(trans);
3604                         if (ret)
3605                                 goto out;
3606                         continue;
3607                 }
3608
3609                 /*
3610                  * add this inode to the orphan list so btrfs_orphan_del does
3611                  * the proper thing when we hit it
3612                  */
3613                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3614                         &BTRFS_I(inode)->runtime_flags);
3615                 atomic_inc(&root->orphan_inodes);
3616
3617                 /* if we have links, this was a truncate, lets do that */
3618                 if (inode->i_nlink) {
3619                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3620                                 iput(inode);
3621                                 continue;
3622                         }
3623                         nr_truncate++;
3624
3625                         /* 1 for the orphan item deletion. */
3626                         trans = btrfs_start_transaction(root, 1);
3627                         if (IS_ERR(trans)) {
3628                                 iput(inode);
3629                                 ret = PTR_ERR(trans);
3630                                 goto out;
3631                         }
3632                         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3633                         btrfs_end_transaction(trans);
3634                         if (ret) {
3635                                 iput(inode);
3636                                 goto out;
3637                         }
3638
3639                         ret = btrfs_truncate(inode);
3640                         if (ret)
3641                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
3642                 } else {
3643                         nr_unlink++;
3644                 }
3645
3646                 /* this will do delete_inode and everything for us */
3647                 iput(inode);
3648                 if (ret)
3649                         goto out;
3650         }
3651         /* release the path since we're done with it */
3652         btrfs_release_path(path);
3653
3654         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3655
3656         if (root->orphan_block_rsv)
3657                 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3658                                         (u64)-1);
3659
3660         if (root->orphan_block_rsv ||
3661             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3662                 trans = btrfs_join_transaction(root);
3663                 if (!IS_ERR(trans))
3664                         btrfs_end_transaction(trans);
3665         }
3666
3667         if (nr_unlink)
3668                 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3669         if (nr_truncate)
3670                 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3671
3672 out:
3673         if (ret)
3674                 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3675         btrfs_free_path(path);
3676         return ret;
3677 }
3678
3679 /*
3680  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3681  * don't find any xattrs, we know there can't be any acls.
3682  *
3683  * slot is the slot the inode is in, objectid is the objectid of the inode
3684  */
3685 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3686                                           int slot, u64 objectid,
3687                                           int *first_xattr_slot)
3688 {
3689         u32 nritems = btrfs_header_nritems(leaf);
3690         struct btrfs_key found_key;
3691         static u64 xattr_access = 0;
3692         static u64 xattr_default = 0;
3693         int scanned = 0;
3694
3695         if (!xattr_access) {
3696                 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3697                                         strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3698                 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3699                                         strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3700         }
3701
3702         slot++;
3703         *first_xattr_slot = -1;
3704         while (slot < nritems) {
3705                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3706
3707                 /* we found a different objectid, there must not be acls */
3708                 if (found_key.objectid != objectid)
3709                         return 0;
3710
3711                 /* we found an xattr, assume we've got an acl */
3712                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3713                         if (*first_xattr_slot == -1)
3714                                 *first_xattr_slot = slot;
3715                         if (found_key.offset == xattr_access ||
3716                             found_key.offset == xattr_default)
3717                                 return 1;
3718                 }
3719
3720                 /*
3721                  * we found a key greater than an xattr key, there can't
3722                  * be any acls later on
3723                  */
3724                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3725                         return 0;
3726
3727                 slot++;
3728                 scanned++;
3729
3730                 /*
3731                  * it goes inode, inode backrefs, xattrs, extents,
3732                  * so if there are a ton of hard links to an inode there can
3733                  * be a lot of backrefs.  Don't waste time searching too hard,
3734                  * this is just an optimization
3735                  */
3736                 if (scanned >= 8)
3737                         break;
3738         }
3739         /* we hit the end of the leaf before we found an xattr or
3740          * something larger than an xattr.  We have to assume the inode
3741          * has acls
3742          */
3743         if (*first_xattr_slot == -1)
3744                 *first_xattr_slot = slot;
3745         return 1;
3746 }
3747
3748 /*
3749  * read an inode from the btree into the in-memory inode
3750  */
3751 static int btrfs_read_locked_inode(struct inode *inode)
3752 {
3753         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3754         struct btrfs_path *path;
3755         struct extent_buffer *leaf;
3756         struct btrfs_inode_item *inode_item;
3757         struct btrfs_root *root = BTRFS_I(inode)->root;
3758         struct btrfs_key location;
3759         unsigned long ptr;
3760         int maybe_acls;
3761         u32 rdev;
3762         int ret;
3763         bool filled = false;
3764         int first_xattr_slot;
3765
3766         ret = btrfs_fill_inode(inode, &rdev);
3767         if (!ret)
3768                 filled = true;
3769
3770         path = btrfs_alloc_path();
3771         if (!path) {
3772                 ret = -ENOMEM;
3773                 goto make_bad;
3774         }
3775
3776         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3777
3778         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3779         if (ret) {
3780                 if (ret > 0)
3781                         ret = -ENOENT;
3782                 goto make_bad;
3783         }
3784
3785         leaf = path->nodes[0];
3786
3787         if (filled)
3788                 goto cache_index;
3789
3790         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3791                                     struct btrfs_inode_item);
3792         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3793         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3794         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3795         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3796         btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3797
3798         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3799         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3800
3801         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3802         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3803
3804         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3805         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3806
3807         BTRFS_I(inode)->i_otime.tv_sec =
3808                 btrfs_timespec_sec(leaf, &inode_item->otime);
3809         BTRFS_I(inode)->i_otime.tv_nsec =
3810                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3811
3812         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3813         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3814         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3815
3816         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3817         inode->i_generation = BTRFS_I(inode)->generation;
3818         inode->i_rdev = 0;
3819         rdev = btrfs_inode_rdev(leaf, inode_item);
3820
3821         BTRFS_I(inode)->index_cnt = (u64)-1;
3822         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3823
3824 cache_index:
3825         /*
3826          * If we were modified in the current generation and evicted from memory
3827          * and then re-read we need to do a full sync since we don't have any
3828          * idea about which extents were modified before we were evicted from
3829          * cache.
3830          *
3831          * This is required for both inode re-read from disk and delayed inode
3832          * in delayed_nodes_tree.
3833          */
3834         if (BTRFS_I(inode)->last_trans == fs_info->generation)
3835                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3836                         &BTRFS_I(inode)->runtime_flags);
3837
3838         /*
3839          * We don't persist the id of the transaction where an unlink operation
3840          * against the inode was last made. So here we assume the inode might
3841          * have been evicted, and therefore the exact value of last_unlink_trans
3842          * lost, and set it to last_trans to avoid metadata inconsistencies
3843          * between the inode and its parent if the inode is fsync'ed and the log
3844          * replayed. For example, in the scenario:
3845          *
3846          * touch mydir/foo
3847          * ln mydir/foo mydir/bar
3848          * sync
3849          * unlink mydir/bar
3850          * echo 2 > /proc/sys/vm/drop_caches   # evicts inode
3851          * xfs_io -c fsync mydir/foo
3852          * <power failure>
3853          * mount fs, triggers fsync log replay
3854          *
3855          * We must make sure that when we fsync our inode foo we also log its
3856          * parent inode, otherwise after log replay the parent still has the
3857          * dentry with the "bar" name but our inode foo has a link count of 1
3858          * and doesn't have an inode ref with the name "bar" anymore.
3859          *
3860          * Setting last_unlink_trans to last_trans is a pessimistic approach,
3861          * but it guarantees correctness at the expense of occasional full
3862          * transaction commits on fsync if our inode is a directory, or if our
3863          * inode is not a directory, logging its parent unnecessarily.
3864          */
3865         BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3866
3867         path->slots[0]++;
3868         if (inode->i_nlink != 1 ||
3869             path->slots[0] >= btrfs_header_nritems(leaf))
3870                 goto cache_acl;
3871
3872         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3873         if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3874                 goto cache_acl;
3875
3876         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3877         if (location.type == BTRFS_INODE_REF_KEY) {
3878                 struct btrfs_inode_ref *ref;
3879
3880                 ref = (struct btrfs_inode_ref *)ptr;
3881                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3882         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3883                 struct btrfs_inode_extref *extref;
3884
3885                 extref = (struct btrfs_inode_extref *)ptr;
3886                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3887                                                                      extref);
3888         }
3889 cache_acl:
3890         /*
3891          * try to precache a NULL acl entry for files that don't have
3892          * any xattrs or acls
3893          */
3894         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3895                         btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3896         if (first_xattr_slot != -1) {
3897                 path->slots[0] = first_xattr_slot;
3898                 ret = btrfs_load_inode_props(inode, path);
3899                 if (ret)
3900                         btrfs_err(fs_info,
3901                                   "error loading props for ino %llu (root %llu): %d",
3902                                   btrfs_ino(BTRFS_I(inode)),
3903                                   root->root_key.objectid, ret);
3904         }
3905         btrfs_free_path(path);
3906
3907         if (!maybe_acls)
3908                 cache_no_acl(inode);
3909
3910         switch (inode->i_mode & S_IFMT) {
3911         case S_IFREG:
3912                 inode->i_mapping->a_ops = &btrfs_aops;
3913                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3914                 inode->i_fop = &btrfs_file_operations;
3915                 inode->i_op = &btrfs_file_inode_operations;
3916                 break;
3917         case S_IFDIR:
3918                 inode->i_fop = &btrfs_dir_file_operations;
3919                 inode->i_op = &btrfs_dir_inode_operations;
3920                 break;
3921         case S_IFLNK:
3922                 inode->i_op = &btrfs_symlink_inode_operations;
3923                 inode_nohighmem(inode);
3924                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3925                 break;
3926         default:
3927                 inode->i_op = &btrfs_special_inode_operations;
3928                 init_special_inode(inode, inode->i_mode, rdev);
3929                 break;
3930         }
3931
3932         btrfs_update_iflags(inode);
3933         return 0;
3934
3935 make_bad:
3936         btrfs_free_path(path);
3937         make_bad_inode(inode);
3938         return ret;
3939 }
3940
3941 /*
3942  * given a leaf and an inode, copy the inode fields into the leaf
3943  */
3944 static void fill_inode_item(struct btrfs_trans_handle *trans,
3945                             struct extent_buffer *leaf,
3946                             struct btrfs_inode_item *item,
3947                             struct inode *inode)
3948 {
3949         struct btrfs_map_token token;
3950
3951         btrfs_init_map_token(&token);
3952
3953         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3954         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3955         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3956                                    &token);
3957         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3958         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3959
3960         btrfs_set_token_timespec_sec(leaf, &item->atime,
3961                                      inode->i_atime.tv_sec, &token);
3962         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3963                                       inode->i_atime.tv_nsec, &token);
3964
3965         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3966                                      inode->i_mtime.tv_sec, &token);
3967         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3968                                       inode->i_mtime.tv_nsec, &token);
3969
3970         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3971                                      inode->i_ctime.tv_sec, &token);
3972         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3973                                       inode->i_ctime.tv_nsec, &token);
3974
3975         btrfs_set_token_timespec_sec(leaf, &item->otime,
3976                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3977         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3978                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3979
3980         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3981                                      &token);
3982         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3983                                          &token);
3984         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3985         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3986         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3987         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3988         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3989 }
3990
3991 /*
3992  * copy everything in the in-memory inode into the btree.
3993  */
3994 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3995                                 struct btrfs_root *root, struct inode *inode)
3996 {
3997         struct btrfs_inode_item *inode_item;
3998         struct btrfs_path *path;
3999         struct extent_buffer *leaf;
4000         int ret;
4001
4002         path = btrfs_alloc_path();
4003         if (!path)
4004                 return -ENOMEM;
4005
4006         path->leave_spinning = 1;
4007         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
4008                                  1);
4009         if (ret) {
4010                 if (ret > 0)
4011                         ret = -ENOENT;
4012                 goto failed;
4013         }
4014
4015         leaf = path->nodes[0];
4016         inode_item = btrfs_item_ptr(leaf, path->slots[0],
4017                                     struct btrfs_inode_item);
4018
4019         fill_inode_item(trans, leaf, inode_item, inode);
4020         btrfs_mark_buffer_dirty(leaf);
4021         btrfs_set_inode_last_trans(trans, inode);
4022         ret = 0;
4023 failed:
4024         btrfs_free_path(path);
4025         return ret;
4026 }
4027
4028 /*
4029  * copy everything in the in-memory inode into the btree.
4030  */
4031 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4032                                 struct btrfs_root *root, struct inode *inode)
4033 {
4034         struct btrfs_fs_info *fs_info = root->fs_info;
4035         int ret;
4036
4037         /*
4038          * If the inode is a free space inode, we can deadlock during commit
4039          * if we put it into the delayed code.
4040          *
4041          * The data relocation inode should also be directly updated
4042          * without delay
4043          */
4044         if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4045             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4046             && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4047                 btrfs_update_root_times(trans, root);
4048
4049                 ret = btrfs_delayed_update_inode(trans, root, inode);
4050                 if (!ret)
4051                         btrfs_set_inode_last_trans(trans, inode);
4052                 return ret;
4053         }
4054
4055         return btrfs_update_inode_item(trans, root, inode);
4056 }
4057
4058 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4059                                          struct btrfs_root *root,
4060                                          struct inode *inode)
4061 {
4062         int ret;
4063
4064         ret = btrfs_update_inode(trans, root, inode);
4065         if (ret == -ENOSPC)
4066                 return btrfs_update_inode_item(trans, root, inode);
4067         return ret;
4068 }
4069
4070 /*
4071  * unlink helper that gets used here in inode.c and in the tree logging
4072  * recovery code.  It remove a link in a directory with a given name, and
4073  * also drops the back refs in the inode to the directory
4074  */
4075 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4076                                 struct btrfs_root *root,
4077                                 struct btrfs_inode *dir,
4078                                 struct btrfs_inode *inode,
4079                                 const char *name, int name_len)
4080 {
4081         struct btrfs_fs_info *fs_info = root->fs_info;
4082         struct btrfs_path *path;
4083         int ret = 0;
4084         struct extent_buffer *leaf;
4085         struct btrfs_dir_item *di;
4086         struct btrfs_key key;
4087         u64 index;
4088         u64 ino = btrfs_ino(inode);
4089         u64 dir_ino = btrfs_ino(dir);
4090
4091         path = btrfs_alloc_path();
4092         if (!path) {
4093                 ret = -ENOMEM;
4094                 goto out;
4095         }
4096
4097         path->leave_spinning = 1;
4098         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4099                                     name, name_len, -1);
4100         if (IS_ERR(di)) {
4101                 ret = PTR_ERR(di);
4102                 goto err;
4103         }
4104         if (!di) {
4105                 ret = -ENOENT;
4106                 goto err;
4107         }
4108         leaf = path->nodes[0];
4109         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4110         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4111         if (ret)
4112                 goto err;
4113         btrfs_release_path(path);
4114
4115         /*
4116          * If we don't have dir index, we have to get it by looking up
4117          * the inode ref, since we get the inode ref, remove it directly,
4118          * it is unnecessary to do delayed deletion.
4119          *
4120          * But if we have dir index, needn't search inode ref to get it.
4121          * Since the inode ref is close to the inode item, it is better
4122          * that we delay to delete it, and just do this deletion when
4123          * we update the inode item.
4124          */
4125         if (inode->dir_index) {
4126                 ret = btrfs_delayed_delete_inode_ref(inode);
4127                 if (!ret) {
4128                         index = inode->dir_index;
4129                         goto skip_backref;
4130                 }
4131         }
4132
4133         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4134                                   dir_ino, &index);
4135         if (ret) {
4136                 btrfs_info(fs_info,
4137                         "failed to delete reference to %.*s, inode %llu parent %llu",
4138                         name_len, name, ino, dir_ino);
4139                 btrfs_abort_transaction(trans, ret);
4140                 goto err;
4141         }
4142 skip_backref:
4143         ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4144         if (ret) {
4145                 btrfs_abort_transaction(trans, ret);
4146                 goto err;
4147         }
4148
4149         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4150                         dir_ino);
4151         if (ret != 0 && ret != -ENOENT) {
4152                 btrfs_abort_transaction(trans, ret);
4153                 goto err;
4154         }
4155
4156         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4157                         index);
4158         if (ret == -ENOENT)
4159                 ret = 0;
4160         else if (ret)
4161                 btrfs_abort_transaction(trans, ret);
4162 err:
4163         btrfs_free_path(path);
4164         if (ret)
4165                 goto out;
4166
4167         btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4168         inode_inc_iversion(&inode->vfs_inode);
4169         inode_inc_iversion(&dir->vfs_inode);
4170         inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4171                 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4172         ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4173 out:
4174         return ret;
4175 }
4176
4177 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4178                        struct btrfs_root *root,
4179                        struct btrfs_inode *dir, struct btrfs_inode *inode,
4180                        const char *name, int name_len)
4181 {
4182         int ret;
4183         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4184         if (!ret) {
4185                 drop_nlink(&inode->vfs_inode);
4186                 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4187         }
4188         return ret;
4189 }
4190
4191 /*
4192  * helper to start transaction for unlink and rmdir.
4193  *
4194  * unlink and rmdir are special in btrfs, they do not always free space, so
4195  * if we cannot make our reservations the normal way try and see if there is
4196  * plenty of slack room in the global reserve to migrate, otherwise we cannot
4197  * allow the unlink to occur.
4198  */
4199 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4200 {
4201         struct btrfs_root *root = BTRFS_I(dir)->root;
4202
4203         /*
4204          * 1 for the possible orphan item
4205          * 1 for the dir item
4206          * 1 for the dir index
4207          * 1 for the inode ref
4208          * 1 for the inode
4209          */
4210         return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4211 }
4212
4213 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4214 {
4215         struct btrfs_root *root = BTRFS_I(dir)->root;
4216         struct btrfs_trans_handle *trans;
4217         struct inode *inode = d_inode(dentry);
4218         int ret;
4219
4220         trans = __unlink_start_trans(dir);
4221         if (IS_ERR(trans))
4222                 return PTR_ERR(trans);
4223
4224         btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4225                         0);
4226
4227         ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4228                         BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4229                         dentry->d_name.len);
4230         if (ret)
4231                 goto out;
4232
4233         if (inode->i_nlink == 0) {
4234                 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4235                 if (ret)
4236                         goto out;
4237         }
4238
4239 out:
4240         btrfs_end_transaction(trans);
4241         btrfs_btree_balance_dirty(root->fs_info);
4242         return ret;
4243 }
4244
4245 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4246                         struct btrfs_root *root,
4247                         struct inode *dir, u64 objectid,
4248                         const char *name, int name_len)
4249 {
4250         struct btrfs_fs_info *fs_info = root->fs_info;
4251         struct btrfs_path *path;
4252         struct extent_buffer *leaf;
4253         struct btrfs_dir_item *di;
4254         struct btrfs_key key;
4255         u64 index;
4256         int ret;
4257         u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4258
4259         path = btrfs_alloc_path();
4260         if (!path)
4261                 return -ENOMEM;
4262
4263         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4264                                    name, name_len, -1);
4265         if (IS_ERR_OR_NULL(di)) {
4266                 if (!di)
4267                         ret = -ENOENT;
4268                 else
4269                         ret = PTR_ERR(di);
4270                 goto out;
4271         }
4272
4273         leaf = path->nodes[0];
4274         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4275         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4276         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4277         if (ret) {
4278                 btrfs_abort_transaction(trans, ret);
4279                 goto out;
4280         }
4281         btrfs_release_path(path);
4282
4283         ret = btrfs_del_root_ref(trans, fs_info, objectid,
4284                                  root->root_key.objectid, dir_ino,
4285                                  &index, name, name_len);
4286         if (ret < 0) {
4287                 if (ret != -ENOENT) {
4288                         btrfs_abort_transaction(trans, ret);
4289                         goto out;
4290                 }
4291                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4292                                                  name, name_len);
4293                 if (IS_ERR_OR_NULL(di)) {
4294                         if (!di)
4295                                 ret = -ENOENT;
4296                         else
4297                                 ret = PTR_ERR(di);
4298                         btrfs_abort_transaction(trans, ret);
4299                         goto out;
4300                 }
4301
4302                 leaf = path->nodes[0];
4303                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4304                 btrfs_release_path(path);
4305                 index = key.offset;
4306         }
4307         btrfs_release_path(path);
4308
4309         ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4310         if (ret) {
4311                 btrfs_abort_transaction(trans, ret);
4312                 goto out;
4313         }
4314
4315         btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4316         inode_inc_iversion(dir);
4317         dir->i_mtime = dir->i_ctime = current_time(dir);
4318         ret = btrfs_update_inode_fallback(trans, root, dir);
4319         if (ret)
4320                 btrfs_abort_transaction(trans, ret);
4321 out:
4322         btrfs_free_path(path);
4323         return ret;
4324 }
4325
4326 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4327 {
4328         struct inode *inode = d_inode(dentry);
4329         int err = 0;
4330         struct btrfs_root *root = BTRFS_I(dir)->root;
4331         struct btrfs_trans_handle *trans;
4332         u64 last_unlink_trans;
4333
4334         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4335                 return -ENOTEMPTY;
4336         if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4337                 return -EPERM;
4338
4339         trans = __unlink_start_trans(dir);
4340         if (IS_ERR(trans))
4341                 return PTR_ERR(trans);
4342
4343         if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4344                 err = btrfs_unlink_subvol(trans, root, dir,
4345                                           BTRFS_I(inode)->location.objectid,
4346                                           dentry->d_name.name,
4347                                           dentry->d_name.len);
4348                 goto out;
4349         }
4350
4351         err = btrfs_orphan_add(trans, BTRFS_I(inode));
4352         if (err)
4353                 goto out;
4354
4355         last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4356
4357         /* now the directory is empty */
4358         err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4359                         BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4360                         dentry->d_name.len);
4361         if (!err) {
4362                 btrfs_i_size_write(BTRFS_I(inode), 0);
4363                 /*
4364                  * Propagate the last_unlink_trans value of the deleted dir to
4365                  * its parent directory. This is to prevent an unrecoverable
4366                  * log tree in the case we do something like this:
4367                  * 1) create dir foo
4368                  * 2) create snapshot under dir foo
4369                  * 3) delete the snapshot
4370                  * 4) rmdir foo
4371                  * 5) mkdir foo
4372                  * 6) fsync foo or some file inside foo
4373                  */
4374                 if (last_unlink_trans >= trans->transid)
4375                         BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4376         }
4377 out:
4378         btrfs_end_transaction(trans);
4379         btrfs_btree_balance_dirty(root->fs_info);
4380
4381         return err;
4382 }
4383
4384 static int truncate_space_check(struct btrfs_trans_handle *trans,
4385                                 struct btrfs_root *root,
4386                                 u64 bytes_deleted)
4387 {
4388         struct btrfs_fs_info *fs_info = root->fs_info;
4389         int ret;
4390
4391         /*
4392          * This is only used to apply pressure to the enospc system, we don't
4393          * intend to use this reservation at all.
4394          */
4395         bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4396         bytes_deleted *= fs_info->nodesize;
4397         ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4398                                   bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4399         if (!ret) {
4400                 trace_btrfs_space_reservation(fs_info, "transaction",
4401                                               trans->transid,
4402                                               bytes_deleted, 1);
4403                 trans->bytes_reserved += bytes_deleted;
4404         }
4405         return ret;
4406
4407 }
4408
4409 /*
4410  * Return this if we need to call truncate_block for the last bit of the
4411  * truncate.
4412  */
4413 #define NEED_TRUNCATE_BLOCK 1
4414
4415 /*
4416  * this can truncate away extent items, csum items and directory items.
4417  * It starts at a high offset and removes keys until it can't find
4418  * any higher than new_size
4419  *
4420  * csum items that cross the new i_size are truncated to the new size
4421  * as well.
4422  *
4423  * min_type is the minimum key type to truncate down to.  If set to 0, this
4424  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4425  */
4426 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4427                                struct btrfs_root *root,
4428                                struct inode *inode,
4429                                u64 new_size, u32 min_type)
4430 {
4431         struct btrfs_fs_info *fs_info = root->fs_info;
4432         struct btrfs_path *path;
4433         struct extent_buffer *leaf;
4434         struct btrfs_file_extent_item *fi;
4435         struct btrfs_key key;
4436         struct btrfs_key found_key;
4437         u64 extent_start = 0;
4438         u64 extent_num_bytes = 0;
4439         u64 extent_offset = 0;
4440         u64 item_end = 0;
4441         u64 last_size = new_size;
4442         u32 found_type = (u8)-1;
4443         int found_extent;
4444         int del_item;
4445         int pending_del_nr = 0;
4446         int pending_del_slot = 0;
4447         int extent_type = -1;
4448         int ret;
4449         int err = 0;
4450         u64 ino = btrfs_ino(BTRFS_I(inode));
4451         u64 bytes_deleted = 0;
4452         bool be_nice = false;
4453         bool should_throttle = false;
4454         bool should_end = false;
4455
4456         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4457
4458         /*
4459          * for non-free space inodes and ref cows, we want to back off from
4460          * time to time
4461          */
4462         if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4463             test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4464                 be_nice = true;
4465
4466         path = btrfs_alloc_path();
4467         if (!path)
4468                 return -ENOMEM;
4469         path->reada = READA_BACK;
4470
4471         /*
4472          * We want to drop from the next block forward in case this new size is
4473          * not block aligned since we will be keeping the last block of the
4474          * extent just the way it is.
4475          */
4476         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4477             root == fs_info->tree_root)
4478                 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4479                                         fs_info->sectorsize),
4480                                         (u64)-1, 0);
4481
4482         /*
4483          * This function is also used to drop the items in the log tree before
4484          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4485          * it is used to drop the loged items. So we shouldn't kill the delayed
4486          * items.
4487          */
4488         if (min_type == 0 && root == BTRFS_I(inode)->root)
4489                 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4490
4491         key.objectid = ino;
4492         key.offset = (u64)-1;
4493         key.type = (u8)-1;
4494
4495 search_again:
4496         /*
4497          * with a 16K leaf size and 128MB extents, you can actually queue
4498          * up a huge file in a single leaf.  Most of the time that
4499          * bytes_deleted is > 0, it will be huge by the time we get here
4500          */
4501         if (be_nice && bytes_deleted > SZ_32M) {
4502                 if (btrfs_should_end_transaction(trans)) {
4503                         err = -EAGAIN;
4504                         goto error;
4505                 }
4506         }
4507
4508
4509         path->leave_spinning = 1;
4510         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4511         if (ret < 0) {
4512                 err = ret;
4513                 goto out;
4514         }
4515
4516         if (ret > 0) {
4517                 /* there are no items in the tree for us to truncate, we're
4518                  * done
4519                  */
4520                 if (path->slots[0] == 0)
4521                         goto out;
4522                 path->slots[0]--;
4523         }
4524
4525         while (1) {
4526                 fi = NULL;
4527                 leaf = path->nodes[0];
4528                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4529                 found_type = found_key.type;
4530
4531                 if (found_key.objectid != ino)
4532                         break;
4533
4534                 if (found_type < min_type)
4535                         break;
4536
4537                 item_end = found_key.offset;
4538                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4539                         fi = btrfs_item_ptr(leaf, path->slots[0],
4540                                             struct btrfs_file_extent_item);
4541                         extent_type = btrfs_file_extent_type(leaf, fi);
4542                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4543                                 item_end +=
4544                                     btrfs_file_extent_num_bytes(leaf, fi);
4545
4546                                 trace_btrfs_truncate_show_fi_regular(
4547                                         BTRFS_I(inode), leaf, fi,
4548                                         found_key.offset);
4549                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4550                                 item_end += btrfs_file_extent_inline_len(leaf,
4551                                                          path->slots[0], fi);
4552
4553                                 trace_btrfs_truncate_show_fi_inline(
4554                                         BTRFS_I(inode), leaf, fi, path->slots[0],
4555                                         found_key.offset);
4556                         }
4557                         item_end--;
4558                 }
4559                 if (found_type > min_type) {
4560                         del_item = 1;
4561                 } else {
4562                         if (item_end < new_size)
4563                                 break;
4564                         if (found_key.offset >= new_size)
4565                                 del_item = 1;
4566                         else
4567                                 del_item = 0;
4568                 }
4569                 found_extent = 0;
4570                 /* FIXME, shrink the extent if the ref count is only 1 */
4571                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4572                         goto delete;
4573
4574                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4575                         u64 num_dec;
4576                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4577                         if (!del_item) {
4578                                 u64 orig_num_bytes =
4579                                         btrfs_file_extent_num_bytes(leaf, fi);
4580                                 extent_num_bytes = ALIGN(new_size -
4581                                                 found_key.offset,
4582                                                 fs_info->sectorsize);
4583                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4584                                                          extent_num_bytes);
4585                                 num_dec = (orig_num_bytes -
4586                                            extent_num_bytes);
4587                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4588                                              &root->state) &&
4589                                     extent_start != 0)
4590                                         inode_sub_bytes(inode, num_dec);
4591                                 btrfs_mark_buffer_dirty(leaf);
4592                         } else {
4593                                 extent_num_bytes =
4594                                         btrfs_file_extent_disk_num_bytes(leaf,
4595                                                                          fi);
4596                                 extent_offset = found_key.offset -
4597                                         btrfs_file_extent_offset(leaf, fi);
4598
4599                                 /* FIXME blocksize != 4096 */
4600                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4601                                 if (extent_start != 0) {
4602                                         found_extent = 1;
4603                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4604                                                      &root->state))
4605                                                 inode_sub_bytes(inode, num_dec);
4606                                 }
4607                         }
4608                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4609                         /*
4610                          * we can't truncate inline items that have had
4611                          * special encodings
4612                          */
4613                         if (!del_item &&
4614                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4615                             btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4616                             btrfs_file_extent_compression(leaf, fi) == 0) {
4617                                 u32 size = (u32)(new_size - found_key.offset);
4618
4619                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4620                                 size = btrfs_file_extent_calc_inline_size(size);
4621                                 btrfs_truncate_item(root->fs_info, path, size, 1);
4622                         } else if (!del_item) {
4623                                 /*
4624                                  * We have to bail so the last_size is set to
4625                                  * just before this extent.
4626                                  */
4627                                 err = NEED_TRUNCATE_BLOCK;
4628                                 break;
4629                         }
4630
4631                         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4632                                 inode_sub_bytes(inode, item_end + 1 - new_size);
4633                 }
4634 delete:
4635                 if (del_item)
4636                         last_size = found_key.offset;
4637                 else
4638                         last_size = new_size;
4639                 if (del_item) {
4640                         if (!pending_del_nr) {
4641                                 /* no pending yet, add ourselves */
4642                                 pending_del_slot = path->slots[0];
4643                                 pending_del_nr = 1;
4644                         } else if (pending_del_nr &&
4645                                    path->slots[0] + 1 == pending_del_slot) {
4646                                 /* hop on the pending chunk */
4647                                 pending_del_nr++;
4648                                 pending_del_slot = path->slots[0];
4649                         } else {
4650                                 BUG();
4651                         }
4652                 } else {
4653                         break;
4654                 }
4655                 should_throttle = false;
4656
4657                 if (found_extent &&
4658                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4659                      root == fs_info->tree_root)) {
4660                         btrfs_set_path_blocking(path);
4661                         bytes_deleted += extent_num_bytes;
4662                         ret = btrfs_free_extent(trans, root, extent_start,
4663                                                 extent_num_bytes, 0,
4664                                                 btrfs_header_owner(leaf),
4665                                                 ino, extent_offset);
4666                         BUG_ON(ret);
4667                         if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4668                                 btrfs_async_run_delayed_refs(fs_info,
4669                                         trans->delayed_ref_updates * 2,
4670                                         trans->transid, 0);
4671                         if (be_nice) {
4672                                 if (truncate_space_check(trans, root,
4673                                                          extent_num_bytes)) {
4674                                         should_end = true;
4675                                 }
4676                                 if (btrfs_should_throttle_delayed_refs(trans,
4677                                                                        fs_info))
4678                                         should_throttle = true;
4679                         }
4680                 }
4681
4682                 if (found_type == BTRFS_INODE_ITEM_KEY)
4683                         break;
4684
4685                 if (path->slots[0] == 0 ||
4686                     path->slots[0] != pending_del_slot ||
4687                     should_throttle || should_end) {
4688                         if (pending_del_nr) {
4689                                 ret = btrfs_del_items(trans, root, path,
4690                                                 pending_del_slot,
4691                                                 pending_del_nr);
4692                                 if (ret) {
4693                                         btrfs_abort_transaction(trans, ret);
4694                                         goto error;
4695                                 }
4696                                 pending_del_nr = 0;
4697                         }
4698                         btrfs_release_path(path);
4699                         if (should_throttle) {
4700                                 unsigned long updates = trans->delayed_ref_updates;
4701                                 if (updates) {
4702                                         trans->delayed_ref_updates = 0;
4703                                         ret = btrfs_run_delayed_refs(trans,
4704                                                                    fs_info,
4705                                                                    updates * 2);
4706                                         if (ret && !err)
4707                                                 err = ret;
4708                                 }
4709                         }
4710                         /*
4711                          * if we failed to refill our space rsv, bail out
4712                          * and let the transaction restart
4713                          */
4714                         if (should_end) {
4715                                 err = -EAGAIN;
4716                                 goto error;
4717                         }
4718                         goto search_again;
4719                 } else {
4720                         path->slots[0]--;
4721                 }
4722         }
4723 out:
4724         if (pending_del_nr) {
4725                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4726                                       pending_del_nr);
4727                 if (ret)
4728                         btrfs_abort_transaction(trans, ret);
4729         }
4730 error:
4731         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4732                 ASSERT(last_size >= new_size);
4733                 if (!err && last_size > new_size)
4734                         last_size = new_size;
4735                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4736         }
4737
4738         btrfs_free_path(path);
4739
4740         if (be_nice && bytes_deleted > SZ_32M) {
4741                 unsigned long updates = trans->delayed_ref_updates;
4742                 if (updates) {
4743                         trans->delayed_ref_updates = 0;
4744                         ret = btrfs_run_delayed_refs(trans, fs_info,
4745                                                      updates * 2);
4746                         if (ret && !err)
4747                                 err = ret;
4748                 }
4749         }
4750         return err;
4751 }
4752
4753 /*
4754  * btrfs_truncate_block - read, zero a chunk and write a block
4755  * @inode - inode that we're zeroing
4756  * @from - the offset to start zeroing
4757  * @len - the length to zero, 0 to zero the entire range respective to the
4758  *      offset
4759  * @front - zero up to the offset instead of from the offset on
4760  *
4761  * This will find the block for the "from" offset and cow the block and zero the
4762  * part we want to zero.  This is used with truncate and hole punching.
4763  */
4764 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4765                         int front)
4766 {
4767         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4768         struct address_space *mapping = inode->i_mapping;
4769         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4770         struct btrfs_ordered_extent *ordered;
4771         struct extent_state *cached_state = NULL;
4772         struct extent_changeset *data_reserved = NULL;
4773         char *kaddr;
4774         u32 blocksize = fs_info->sectorsize;
4775         pgoff_t index = from >> PAGE_SHIFT;
4776         unsigned offset = from & (blocksize - 1);
4777         struct page *page;
4778         gfp_t mask = btrfs_alloc_write_mask(mapping);
4779         int ret = 0;
4780         u64 block_start;
4781         u64 block_end;
4782
4783         if (IS_ALIGNED(offset, blocksize) &&
4784             (!len || IS_ALIGNED(len, blocksize)))
4785                 goto out;
4786
4787         block_start = round_down(from, blocksize);
4788         block_end = block_start + blocksize - 1;
4789
4790         ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4791                                            block_start, blocksize);
4792         if (ret)
4793                 goto out;
4794
4795 again:
4796         page = find_or_create_page(mapping, index, mask);
4797         if (!page) {
4798                 btrfs_delalloc_release_space(inode, data_reserved,
4799                                              block_start, blocksize);
4800                 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4801                 ret = -ENOMEM;
4802                 goto out;
4803         }
4804
4805         if (!PageUptodate(page)) {
4806                 ret = btrfs_readpage(NULL, page);
4807                 lock_page(page);
4808                 if (page->mapping != mapping) {
4809                         unlock_page(page);
4810                         put_page(page);
4811                         goto again;
4812                 }
4813                 if (!PageUptodate(page)) {
4814                         ret = -EIO;
4815                         goto out_unlock;
4816                 }
4817         }
4818         wait_on_page_writeback(page);
4819
4820         lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4821         set_page_extent_mapped(page);
4822
4823         ordered = btrfs_lookup_ordered_extent(inode, block_start);
4824         if (ordered) {
4825                 unlock_extent_cached(io_tree, block_start, block_end,
4826                                      &cached_state);
4827                 unlock_page(page);
4828                 put_page(page);
4829                 btrfs_start_ordered_extent(inode, ordered, 1);
4830                 btrfs_put_ordered_extent(ordered);
4831                 goto again;
4832         }
4833
4834         clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4835                           EXTENT_DIRTY | EXTENT_DELALLOC |
4836                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4837                           0, 0, &cached_state);
4838
4839         ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4840                                         &cached_state, 0);
4841         if (ret) {
4842                 unlock_extent_cached(io_tree, block_start, block_end,
4843                                      &cached_state);
4844                 goto out_unlock;
4845         }
4846
4847         if (offset != blocksize) {
4848                 if (!len)
4849                         len = blocksize - offset;
4850                 kaddr = kmap(page);
4851                 if (front)
4852                         memset(kaddr + (block_start - page_offset(page)),
4853                                 0, offset);
4854                 else
4855                         memset(kaddr + (block_start - page_offset(page)) +  offset,
4856                                 0, len);
4857                 flush_dcache_page(page);
4858                 kunmap(page);
4859         }
4860         ClearPageChecked(page);
4861         set_page_dirty(page);
4862         unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4863
4864 out_unlock:
4865         if (ret)
4866                 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4867                                              blocksize);
4868         btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4869         unlock_page(page);
4870         put_page(page);
4871 out:
4872         extent_changeset_free(data_reserved);
4873         return ret;
4874 }
4875
4876 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4877                              u64 offset, u64 len)
4878 {
4879         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4880         struct btrfs_trans_handle *trans;
4881         int ret;
4882
4883         /*
4884          * Still need to make sure the inode looks like it's been updated so
4885          * that any holes get logged if we fsync.
4886          */
4887         if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4888                 BTRFS_I(inode)->last_trans = fs_info->generation;
4889                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4890                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4891                 return 0;
4892         }
4893
4894         /*
4895          * 1 - for the one we're dropping
4896          * 1 - for the one we're adding
4897          * 1 - for updating the inode.
4898          */
4899         trans = btrfs_start_transaction(root, 3);
4900         if (IS_ERR(trans))
4901                 return PTR_ERR(trans);
4902
4903         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4904         if (ret) {
4905                 btrfs_abort_transaction(trans, ret);
4906                 btrfs_end_transaction(trans);
4907                 return ret;
4908         }
4909
4910         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4911                         offset, 0, 0, len, 0, len, 0, 0, 0);
4912         if (ret)
4913                 btrfs_abort_transaction(trans, ret);
4914         else
4915                 btrfs_update_inode(trans, root, inode);
4916         btrfs_end_transaction(trans);
4917         return ret;
4918 }
4919
4920 /*
4921  * This function puts in dummy file extents for the area we're creating a hole
4922  * for.  So if we are truncating this file to a larger size we need to insert
4923  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4924  * the range between oldsize and size
4925  */
4926 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4927 {
4928         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4929         struct btrfs_root *root = BTRFS_I(inode)->root;
4930         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4931         struct extent_map *em = NULL;
4932         struct extent_state *cached_state = NULL;
4933         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4934         u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4935         u64 block_end = ALIGN(size, fs_info->sectorsize);
4936         u64 last_byte;
4937         u64 cur_offset;
4938         u64 hole_size;
4939         int err = 0;
4940
4941         /*
4942          * If our size started in the middle of a block we need to zero out the
4943          * rest of the block before we expand the i_size, otherwise we could
4944          * expose stale data.
4945          */
4946         err = btrfs_truncate_block(inode, oldsize, 0, 0);
4947         if (err)
4948                 return err;
4949
4950         if (size <= hole_start)
4951                 return 0;
4952
4953         while (1) {
4954                 struct btrfs_ordered_extent *ordered;
4955
4956                 lock_extent_bits(io_tree, hole_start, block_end - 1,
4957                                  &cached_state);
4958                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4959                                                      block_end - hole_start);
4960                 if (!ordered)
4961                         break;
4962                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4963                                      &cached_state);
4964                 btrfs_start_ordered_extent(inode, ordered, 1);
4965                 btrfs_put_ordered_extent(ordered);
4966         }
4967
4968         cur_offset = hole_start;
4969         while (1) {
4970                 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4971                                 block_end - cur_offset, 0);
4972                 if (IS_ERR(em)) {
4973                         err = PTR_ERR(em);
4974                         em = NULL;
4975                         break;
4976                 }
4977                 last_byte = min(extent_map_end(em), block_end);
4978                 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4979                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4980                         struct extent_map *hole_em;
4981                         hole_size = last_byte - cur_offset;
4982
4983                         err = maybe_insert_hole(root, inode, cur_offset,
4984                                                 hole_size);
4985                         if (err)
4986                                 break;
4987                         btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4988                                                 cur_offset + hole_size - 1, 0);
4989                         hole_em = alloc_extent_map();
4990                         if (!hole_em) {
4991                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4992                                         &BTRFS_I(inode)->runtime_flags);
4993                                 goto next;
4994                         }
4995                         hole_em->start = cur_offset;
4996                         hole_em->len = hole_size;
4997                         hole_em->orig_start = cur_offset;
4998
4999                         hole_em->block_start = EXTENT_MAP_HOLE;
5000                         hole_em->block_len = 0;
5001                         hole_em->orig_block_len = 0;
5002                         hole_em->ram_bytes = hole_size;
5003                         hole_em->bdev = fs_info->fs_devices->latest_bdev;
5004                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
5005                         hole_em->generation = fs_info->generation;
5006
5007                         while (1) {
5008                                 write_lock(&em_tree->lock);
5009                                 err = add_extent_mapping(em_tree, hole_em, 1);
5010                                 write_unlock(&em_tree->lock);
5011                                 if (err != -EEXIST)
5012                                         break;
5013                                 btrfs_drop_extent_cache(BTRFS_I(inode),
5014                                                         cur_offset,
5015                                                         cur_offset +
5016                                                         hole_size - 1, 0);
5017                         }
5018                         free_extent_map(hole_em);
5019                 }
5020 next:
5021                 free_extent_map(em);
5022                 em = NULL;
5023                 cur_offset = last_byte;
5024                 if (cur_offset >= block_end)
5025                         break;
5026         }
5027         free_extent_map(em);
5028         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5029         return err;
5030 }
5031
5032 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5033 {
5034         struct btrfs_root *root = BTRFS_I(inode)->root;
5035         struct btrfs_trans_handle *trans;
5036         loff_t oldsize = i_size_read(inode);
5037         loff_t newsize = attr->ia_size;
5038         int mask = attr->ia_valid;
5039         int ret;
5040
5041         /*
5042          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5043          * special case where we need to update the times despite not having
5044          * these flags set.  For all other operations the VFS set these flags
5045          * explicitly if it wants a timestamp update.
5046          */
5047         if (newsize != oldsize) {
5048                 inode_inc_iversion(inode);
5049                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5050                         inode->i_ctime = inode->i_mtime =
5051                                 current_time(inode);
5052         }
5053
5054         if (newsize > oldsize) {
5055                 /*
5056                  * Don't do an expanding truncate while snapshotting is ongoing.
5057                  * This is to ensure the snapshot captures a fully consistent
5058                  * state of this file - if the snapshot captures this expanding
5059                  * truncation, it must capture all writes that happened before
5060                  * this truncation.
5061                  */
5062                 btrfs_wait_for_snapshot_creation(root);
5063                 ret = btrfs_cont_expand(inode, oldsize, newsize);
5064                 if (ret) {
5065                         btrfs_end_write_no_snapshotting(root);
5066                         return ret;
5067                 }
5068
5069                 trans = btrfs_start_transaction(root, 1);
5070                 if (IS_ERR(trans)) {
5071                         btrfs_end_write_no_snapshotting(root);
5072                         return PTR_ERR(trans);
5073                 }
5074
5075                 i_size_write(inode, newsize);
5076                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5077                 pagecache_isize_extended(inode, oldsize, newsize);
5078                 ret = btrfs_update_inode(trans, root, inode);
5079                 btrfs_end_write_no_snapshotting(root);
5080                 btrfs_end_transaction(trans);
5081         } else {
5082
5083                 /*
5084                  * We're truncating a file that used to have good data down to
5085                  * zero. Make sure it gets into the ordered flush list so that
5086                  * any new writes get down to disk quickly.
5087                  */
5088                 if (newsize == 0)
5089                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5090                                 &BTRFS_I(inode)->runtime_flags);
5091
5092                 /*
5093                  * 1 for the orphan item we're going to add
5094                  * 1 for the orphan item deletion.
5095                  */
5096                 trans = btrfs_start_transaction(root, 2);
5097                 if (IS_ERR(trans))
5098                         return PTR_ERR(trans);
5099
5100                 /*
5101                  * We need to do this in case we fail at _any_ point during the
5102                  * actual truncate.  Once we do the truncate_setsize we could
5103                  * invalidate pages which forces any outstanding ordered io to
5104                  * be instantly completed which will give us extents that need
5105                  * to be truncated.  If we fail to get an orphan inode down we
5106                  * could have left over extents that were never meant to live,
5107                  * so we need to guarantee from this point on that everything
5108                  * will be consistent.
5109                  */
5110                 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5111                 btrfs_end_transaction(trans);
5112                 if (ret)
5113                         return ret;
5114
5115                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5116                 truncate_setsize(inode, newsize);
5117
5118                 /* Disable nonlocked read DIO to avoid the end less truncate */
5119                 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5120                 inode_dio_wait(inode);
5121                 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5122
5123                 ret = btrfs_truncate(inode);
5124                 if (ret && inode->i_nlink) {
5125                         int err;
5126
5127                         /* To get a stable disk_i_size */
5128                         err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5129                         if (err) {
5130                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5131                                 return err;
5132                         }
5133
5134                         /*
5135                          * failed to truncate, disk_i_size is only adjusted down
5136                          * as we remove extents, so it should represent the true
5137                          * size of the inode, so reset the in memory size and
5138                          * delete our orphan entry.
5139                          */
5140                         trans = btrfs_join_transaction(root);
5141                         if (IS_ERR(trans)) {
5142                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5143                                 return ret;
5144                         }
5145                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5146                         err = btrfs_orphan_del(trans, BTRFS_I(inode));
5147                         if (err)
5148                                 btrfs_abort_transaction(trans, err);
5149                         btrfs_end_transaction(trans);
5150                 }
5151         }
5152
5153         return ret;
5154 }
5155
5156 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5157 {
5158         struct inode *inode = d_inode(dentry);
5159         struct btrfs_root *root = BTRFS_I(inode)->root;
5160         int err;
5161
5162         if (btrfs_root_readonly(root))
5163                 return -EROFS;
5164
5165         err = setattr_prepare(dentry, attr);
5166         if (err)
5167                 return err;
5168
5169         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5170                 err = btrfs_setsize(inode, attr);
5171                 if (err)
5172                         return err;
5173         }
5174
5175         if (attr->ia_valid) {
5176                 setattr_copy(inode, attr);
5177                 inode_inc_iversion(inode);
5178                 err = btrfs_dirty_inode(inode);
5179
5180                 if (!err && attr->ia_valid & ATTR_MODE)
5181                         err = posix_acl_chmod(inode, inode->i_mode);
5182         }
5183
5184         return err;
5185 }
5186
5187 /*
5188  * While truncating the inode pages during eviction, we get the VFS calling
5189  * btrfs_invalidatepage() against each page of the inode. This is slow because
5190  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5191  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5192  * extent_state structures over and over, wasting lots of time.
5193  *
5194  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5195  * those expensive operations on a per page basis and do only the ordered io
5196  * finishing, while we release here the extent_map and extent_state structures,
5197  * without the excessive merging and splitting.
5198  */
5199 static void evict_inode_truncate_pages(struct inode *inode)
5200 {
5201         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5202         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5203         struct rb_node *node;
5204
5205         ASSERT(inode->i_state & I_FREEING);
5206         truncate_inode_pages_final(&inode->i_data);
5207
5208         write_lock(&map_tree->lock);
5209         while (!RB_EMPTY_ROOT(&map_tree->map)) {
5210                 struct extent_map *em;
5211
5212                 node = rb_first(&map_tree->map);
5213                 em = rb_entry(node, struct extent_map, rb_node);
5214                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5215                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5216                 remove_extent_mapping(map_tree, em);
5217                 free_extent_map(em);
5218                 if (need_resched()) {
5219                         write_unlock(&map_tree->lock);
5220                         cond_resched();
5221                         write_lock(&map_tree->lock);
5222                 }
5223         }
5224         write_unlock(&map_tree->lock);
5225
5226         /*
5227          * Keep looping until we have no more ranges in the io tree.
5228          * We can have ongoing bios started by readpages (called from readahead)
5229          * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5230          * still in progress (unlocked the pages in the bio but did not yet
5231          * unlocked the ranges in the io tree). Therefore this means some
5232          * ranges can still be locked and eviction started because before
5233          * submitting those bios, which are executed by a separate task (work
5234          * queue kthread), inode references (inode->i_count) were not taken
5235          * (which would be dropped in the end io callback of each bio).
5236          * Therefore here we effectively end up waiting for those bios and
5237          * anyone else holding locked ranges without having bumped the inode's
5238          * reference count - if we don't do it, when they access the inode's
5239          * io_tree to unlock a range it may be too late, leading to an
5240          * use-after-free issue.
5241          */
5242         spin_lock(&io_tree->lock);
5243         while (!RB_EMPTY_ROOT(&io_tree->state)) {
5244                 struct extent_state *state;
5245                 struct extent_state *cached_state = NULL;
5246                 u64 start;
5247                 u64 end;
5248
5249                 node = rb_first(&io_tree->state);
5250                 state = rb_entry(node, struct extent_state, rb_node);
5251                 start = state->start;
5252                 end = state->end;
5253                 spin_unlock(&io_tree->lock);
5254
5255                 lock_extent_bits(io_tree, start, end, &cached_state);
5256
5257                 /*
5258                  * If still has DELALLOC flag, the extent didn't reach disk,
5259                  * and its reserved space won't be freed by delayed_ref.
5260                  * So we need to free its reserved space here.
5261                  * (Refer to comment in btrfs_invalidatepage, case 2)
5262                  *
5263                  * Note, end is the bytenr of last byte, so we need + 1 here.
5264                  */
5265                 if (state->state & EXTENT_DELALLOC)
5266                         btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5267
5268                 clear_extent_bit(io_tree, start, end,
5269                                  EXTENT_LOCKED | EXTENT_DIRTY |
5270                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5271                                  EXTENT_DEFRAG, 1, 1, &cached_state);
5272
5273                 cond_resched();
5274                 spin_lock(&io_tree->lock);
5275         }
5276         spin_unlock(&io_tree->lock);
5277 }
5278
5279 void btrfs_evict_inode(struct inode *inode)
5280 {
5281         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5282         struct btrfs_trans_handle *trans;
5283         struct btrfs_root *root = BTRFS_I(inode)->root;
5284         struct btrfs_block_rsv *rsv, *global_rsv;
5285         int steal_from_global = 0;
5286         u64 min_size;
5287         int ret;
5288
5289         trace_btrfs_inode_evict(inode);
5290
5291         if (!root) {
5292                 clear_inode(inode);
5293                 return;
5294         }
5295
5296         min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5297
5298         evict_inode_truncate_pages(inode);
5299
5300         if (inode->i_nlink &&
5301             ((btrfs_root_refs(&root->root_item) != 0 &&
5302               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5303              btrfs_is_free_space_inode(BTRFS_I(inode))))
5304                 goto no_delete;
5305
5306         if (is_bad_inode(inode)) {
5307                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5308                 goto no_delete;
5309         }
5310         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5311         if (!special_file(inode->i_mode))
5312                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5313
5314         btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5315
5316         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5317                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5318                                  &BTRFS_I(inode)->runtime_flags));
5319                 goto no_delete;
5320         }
5321
5322         if (inode->i_nlink > 0) {
5323                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5324                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5325                 goto no_delete;
5326         }
5327
5328         ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5329         if (ret) {
5330                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5331                 goto no_delete;
5332         }
5333
5334         rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5335         if (!rsv) {
5336                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5337                 goto no_delete;
5338         }
5339         rsv->size = min_size;
5340         rsv->failfast = 1;
5341         global_rsv = &fs_info->global_block_rsv;
5342
5343         btrfs_i_size_write(BTRFS_I(inode), 0);
5344
5345         /*
5346          * This is a bit simpler than btrfs_truncate since we've already
5347          * reserved our space for our orphan item in the unlink, so we just
5348          * need to reserve some slack space in case we add bytes and update
5349          * inode item when doing the truncate.
5350          */
5351         while (1) {
5352                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5353                                              BTRFS_RESERVE_FLUSH_LIMIT);
5354
5355                 /*
5356                  * Try and steal from the global reserve since we will
5357                  * likely not use this space anyway, we want to try as
5358                  * hard as possible to get this to work.
5359                  */
5360                 if (ret)
5361                         steal_from_global++;
5362                 else
5363                         steal_from_global = 0;
5364                 ret = 0;
5365
5366                 /*
5367                  * steal_from_global == 0: we reserved stuff, hooray!
5368                  * steal_from_global == 1: we didn't reserve stuff, boo!
5369                  * steal_from_global == 2: we've committed, still not a lot of
5370                  * room but maybe we'll have room in the global reserve this
5371                  * time.
5372                  * steal_from_global == 3: abandon all hope!
5373                  */
5374                 if (steal_from_global > 2) {
5375                         btrfs_warn(fs_info,
5376                                    "Could not get space for a delete, will truncate on mount %d",
5377                                    ret);
5378                         btrfs_orphan_del(NULL, BTRFS_I(inode));
5379                         btrfs_free_block_rsv(fs_info, rsv);
5380                         goto no_delete;
5381                 }
5382
5383                 trans = btrfs_join_transaction(root);
5384                 if (IS_ERR(trans)) {
5385                         btrfs_orphan_del(NULL, BTRFS_I(inode));
5386                         btrfs_free_block_rsv(fs_info, rsv);
5387                         goto no_delete;
5388                 }
5389
5390                 /*
5391                  * We can't just steal from the global reserve, we need to make
5392                  * sure there is room to do it, if not we need to commit and try
5393                  * again.
5394                  */
5395                 if (steal_from_global) {
5396                         if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5397                                 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5398                                                               min_size, 0);
5399                         else
5400                                 ret = -ENOSPC;
5401                 }
5402
5403                 /*
5404                  * Couldn't steal from the global reserve, we have too much
5405                  * pending stuff built up, commit the transaction and try it
5406                  * again.
5407                  */
5408                 if (ret) {
5409                         ret = btrfs_commit_transaction(trans);
5410                         if (ret) {
5411                                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5412                                 btrfs_free_block_rsv(fs_info, rsv);
5413                                 goto no_delete;
5414                         }
5415                         continue;
5416                 } else {
5417                         steal_from_global = 0;
5418                 }
5419
5420                 trans->block_rsv = rsv;
5421
5422                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5423                 if (ret != -ENOSPC && ret != -EAGAIN)
5424                         break;
5425
5426                 trans->block_rsv = &fs_info->trans_block_rsv;
5427                 btrfs_end_transaction(trans);
5428                 trans = NULL;
5429                 btrfs_btree_balance_dirty(fs_info);
5430         }
5431
5432         btrfs_free_block_rsv(fs_info, rsv);
5433
5434         /*
5435          * Errors here aren't a big deal, it just means we leave orphan items
5436          * in the tree.  They will be cleaned up on the next mount.
5437          */
5438         if (ret == 0) {
5439                 trans->block_rsv = root->orphan_block_rsv;
5440                 btrfs_orphan_del(trans, BTRFS_I(inode));
5441         } else {
5442                 btrfs_orphan_del(NULL, BTRFS_I(inode));
5443         }
5444
5445         trans->block_rsv = &fs_info->trans_block_rsv;
5446         if (!(root == fs_info->tree_root ||
5447               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5448                 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5449
5450         btrfs_end_transaction(trans);
5451         btrfs_btree_balance_dirty(fs_info);
5452 no_delete:
5453         btrfs_remove_delayed_node(BTRFS_I(inode));
5454         clear_inode(inode);
5455 }
5456
5457 /*
5458  * this returns the key found in the dir entry in the location pointer.
5459  * If no dir entries were found, location->objectid is 0.
5460  */
5461 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5462                                struct btrfs_key *location)
5463 {
5464         const char *name = dentry->d_name.name;
5465         int namelen = dentry->d_name.len;
5466         struct btrfs_dir_item *di;
5467         struct btrfs_path *path;
5468         struct btrfs_root *root = BTRFS_I(dir)->root;
5469         int ret = 0;
5470
5471         path = btrfs_alloc_path();
5472         if (!path)
5473                 return -ENOMEM;
5474
5475         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5476                         name, namelen, 0);
5477         if (IS_ERR(di))
5478                 ret = PTR_ERR(di);
5479
5480         if (IS_ERR_OR_NULL(di))
5481                 goto out_err;
5482
5483         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5484         if (location->type != BTRFS_INODE_ITEM_KEY &&
5485             location->type != BTRFS_ROOT_ITEM_KEY) {
5486                 btrfs_warn(root->fs_info,
5487 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5488                            __func__, name, btrfs_ino(BTRFS_I(dir)),
5489                            location->objectid, location->type, location->offset);
5490                 goto out_err;
5491         }
5492 out:
5493         btrfs_free_path(path);
5494         return ret;
5495 out_err:
5496         location->objectid = 0;
5497         goto out;
5498 }
5499
5500 /*
5501  * when we hit a tree root in a directory, the btrfs part of the inode
5502  * needs to be changed to reflect the root directory of the tree root.  This
5503  * is kind of like crossing a mount point.
5504  */
5505 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5506                                     struct inode *dir,
5507                                     struct dentry *dentry,
5508                                     struct btrfs_key *location,
5509                                     struct btrfs_root **sub_root)
5510 {
5511         struct btrfs_path *path;
5512         struct btrfs_root *new_root;
5513         struct btrfs_root_ref *ref;
5514         struct extent_buffer *leaf;
5515         struct btrfs_key key;
5516         int ret;
5517         int err = 0;
5518
5519         path = btrfs_alloc_path();
5520         if (!path) {
5521                 err = -ENOMEM;
5522                 goto out;
5523         }
5524
5525         err = -ENOENT;
5526         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5527         key.type = BTRFS_ROOT_REF_KEY;
5528         key.offset = location->objectid;
5529
5530         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5531         if (ret) {
5532                 if (ret < 0)
5533                         err = ret;
5534                 goto out;
5535         }
5536
5537         leaf = path->nodes[0];
5538         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5539         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5540             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5541                 goto out;
5542
5543         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5544                                    (unsigned long)(ref + 1),
5545                                    dentry->d_name.len);
5546         if (ret)
5547                 goto out;
5548
5549         btrfs_release_path(path);
5550
5551         new_root = btrfs_read_fs_root_no_name(fs_info, location);
5552         if (IS_ERR(new_root)) {
5553                 err = PTR_ERR(new_root);
5554                 goto out;
5555         }
5556
5557         *sub_root = new_root;
5558         location->objectid = btrfs_root_dirid(&new_root->root_item);
5559         location->type = BTRFS_INODE_ITEM_KEY;
5560         location->offset = 0;
5561         err = 0;
5562 out:
5563         btrfs_free_path(path);
5564         return err;
5565 }
5566
5567 static void inode_tree_add(struct inode *inode)
5568 {
5569         struct btrfs_root *root = BTRFS_I(inode)->root;
5570         struct btrfs_inode *entry;
5571         struct rb_node **p;
5572         struct rb_node *parent;
5573         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5574         u64 ino = btrfs_ino(BTRFS_I(inode));
5575
5576         if (inode_unhashed(inode))
5577                 return;
5578         parent = NULL;
5579         spin_lock(&root->inode_lock);
5580         p = &root->inode_tree.rb_node;
5581         while (*p) {
5582                 parent = *p;
5583                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5584
5585                 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5586                         p = &parent->rb_left;
5587                 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5588                         p = &parent->rb_right;
5589                 else {
5590                         WARN_ON(!(entry->vfs_inode.i_state &
5591                                   (I_WILL_FREE | I_FREEING)));
5592                         rb_replace_node(parent, new, &root->inode_tree);
5593                         RB_CLEAR_NODE(parent);
5594                         spin_unlock(&root->inode_lock);
5595                         return;
5596                 }
5597         }
5598         rb_link_node(new, parent, p);
5599         rb_insert_color(new, &root->inode_tree);
5600         spin_unlock(&root->inode_lock);
5601 }
5602
5603 static void inode_tree_del(struct inode *inode)
5604 {
5605         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5606         struct btrfs_root *root = BTRFS_I(inode)->root;
5607         int empty = 0;
5608
5609         spin_lock(&root->inode_lock);
5610         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5611                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5612                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5613                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5614         }
5615         spin_unlock(&root->inode_lock);
5616
5617         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5618                 synchronize_srcu(&fs_info->subvol_srcu);
5619                 spin_lock(&root->inode_lock);
5620                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5621                 spin_unlock(&root->inode_lock);
5622                 if (empty)
5623                         btrfs_add_dead_root(root);
5624         }
5625 }
5626
5627 void btrfs_invalidate_inodes(struct btrfs_root *root)
5628 {
5629         struct btrfs_fs_info *fs_info = root->fs_info;
5630         struct rb_node *node;
5631         struct rb_node *prev;
5632         struct btrfs_inode *entry;
5633         struct inode *inode;
5634         u64 objectid = 0;
5635
5636         if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5637                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5638
5639         spin_lock(&root->inode_lock);
5640 again:
5641         node = root->inode_tree.rb_node;
5642         prev = NULL;
5643         while (node) {
5644                 prev = node;
5645                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5646
5647                 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5648                         node = node->rb_left;
5649                 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5650                         node = node->rb_right;
5651                 else
5652                         break;
5653         }
5654         if (!node) {
5655                 while (prev) {
5656                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5657                         if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5658                                 node = prev;
5659                                 break;
5660                         }
5661                         prev = rb_next(prev);
5662                 }
5663         }
5664         while (node) {
5665                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5666                 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5667                 inode = igrab(&entry->vfs_inode);
5668                 if (inode) {
5669                         spin_unlock(&root->inode_lock);
5670                         if (atomic_read(&inode->i_count) > 1)
5671                                 d_prune_aliases(inode);
5672                         /*
5673                          * btrfs_drop_inode will have it removed from
5674                          * the inode cache when its usage count
5675                          * hits zero.
5676                          */
5677                         iput(inode);
5678                         cond_resched();
5679                         spin_lock(&root->inode_lock);
5680                         goto again;
5681                 }
5682
5683                 if (cond_resched_lock(&root->inode_lock))
5684                         goto again;
5685
5686                 node = rb_next(node);
5687         }
5688         spin_unlock(&root->inode_lock);
5689 }
5690
5691 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5692 {
5693         struct btrfs_iget_args *args = p;
5694         inode->i_ino = args->location->objectid;
5695         memcpy(&BTRFS_I(inode)->location, args->location,
5696                sizeof(*args->location));
5697         BTRFS_I(inode)->root = args->root;
5698         return 0;
5699 }
5700
5701 static int btrfs_find_actor(struct inode *inode, void *opaque)
5702 {
5703         struct btrfs_iget_args *args = opaque;
5704         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5705                 args->root == BTRFS_I(inode)->root;
5706 }
5707
5708 static struct inode *btrfs_iget_locked(struct super_block *s,
5709                                        struct btrfs_key *location,
5710                                        struct btrfs_root *root)
5711 {
5712         struct inode *inode;
5713         struct btrfs_iget_args args;
5714         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5715
5716         args.location = location;
5717         args.root = root;
5718
5719         inode = iget5_locked(s, hashval, btrfs_find_actor,
5720                              btrfs_init_locked_inode,
5721                              (void *)&args);
5722         return inode;
5723 }
5724
5725 /* Get an inode object given its location and corresponding root.
5726  * Returns in *is_new if the inode was read from disk
5727  */
5728 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5729                          struct btrfs_root *root, int *new)
5730 {
5731         struct inode *inode;
5732
5733         inode = btrfs_iget_locked(s, location, root);
5734         if (!inode)
5735                 return ERR_PTR(-ENOMEM);
5736
5737         if (inode->i_state & I_NEW) {
5738                 int ret;
5739
5740                 ret = btrfs_read_locked_inode(inode);
5741                 if (!is_bad_inode(inode)) {
5742                         inode_tree_add(inode);
5743                         unlock_new_inode(inode);
5744                         if (new)
5745                                 *new = 1;
5746                 } else {
5747                         unlock_new_inode(inode);
5748                         iput(inode);
5749                         ASSERT(ret < 0);
5750                         inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5751                 }
5752         }
5753
5754         return inode;
5755 }
5756
5757 static struct inode *new_simple_dir(struct super_block *s,
5758                                     struct btrfs_key *key,
5759                                     struct btrfs_root *root)
5760 {
5761         struct inode *inode = new_inode(s);
5762
5763         if (!inode)
5764                 return ERR_PTR(-ENOMEM);
5765
5766         BTRFS_I(inode)->root = root;
5767         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5768         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5769
5770         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5771         inode->i_op = &btrfs_dir_ro_inode_operations;
5772         inode->i_opflags &= ~IOP_XATTR;
5773         inode->i_fop = &simple_dir_operations;
5774         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5775         inode->i_mtime = current_time(inode);
5776         inode->i_atime = inode->i_mtime;
5777         inode->i_ctime = inode->i_mtime;
5778         BTRFS_I(inode)->i_otime = inode->i_mtime;
5779
5780         return inode;
5781 }
5782
5783 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5784 {
5785         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5786         struct inode *inode;
5787         struct btrfs_root *root = BTRFS_I(dir)->root;
5788         struct btrfs_root *sub_root = root;
5789         struct btrfs_key location;
5790         int index;
5791         int ret = 0;
5792
5793         if (dentry->d_name.len > BTRFS_NAME_LEN)
5794                 return ERR_PTR(-ENAMETOOLONG);
5795
5796         ret = btrfs_inode_by_name(dir, dentry, &location);
5797         if (ret < 0)
5798                 return ERR_PTR(ret);
5799
5800         if (location.objectid == 0)
5801                 return ERR_PTR(-ENOENT);
5802
5803         if (location.type == BTRFS_INODE_ITEM_KEY) {
5804                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5805                 return inode;
5806         }
5807
5808         index = srcu_read_lock(&fs_info->subvol_srcu);
5809         ret = fixup_tree_root_location(fs_info, dir, dentry,
5810                                        &location, &sub_root);
5811         if (ret < 0) {
5812                 if (ret != -ENOENT)
5813                         inode = ERR_PTR(ret);
5814                 else
5815                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5816         } else {
5817                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5818         }
5819         srcu_read_unlock(&fs_info->subvol_srcu, index);
5820
5821         if (!IS_ERR(inode) && root != sub_root) {
5822                 down_read(&fs_info->cleanup_work_sem);
5823                 if (!sb_rdonly(inode->i_sb))
5824                         ret = btrfs_orphan_cleanup(sub_root);
5825                 up_read(&fs_info->cleanup_work_sem);
5826                 if (ret) {
5827                         iput(inode);
5828                         inode = ERR_PTR(ret);
5829                 }
5830         }
5831
5832         return inode;
5833 }
5834
5835 static int btrfs_dentry_delete(const struct dentry *dentry)
5836 {
5837         struct btrfs_root *root;
5838         struct inode *inode = d_inode(dentry);
5839
5840         if (!inode && !IS_ROOT(dentry))
5841                 inode = d_inode(dentry->d_parent);
5842
5843         if (inode) {
5844                 root = BTRFS_I(inode)->root;
5845                 if (btrfs_root_refs(&root->root_item) == 0)
5846                         return 1;
5847
5848                 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5849                         return 1;
5850         }
5851         return 0;
5852 }
5853
5854 static void btrfs_dentry_release(struct dentry *dentry)
5855 {
5856         kfree(dentry->d_fsdata);
5857 }
5858
5859 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5860                                    unsigned int flags)
5861 {
5862         struct inode *inode;
5863
5864         inode = btrfs_lookup_dentry(dir, dentry);
5865         if (IS_ERR(inode)) {
5866                 if (PTR_ERR(inode) == -ENOENT)
5867                         inode = NULL;
5868                 else
5869                         return ERR_CAST(inode);
5870         }
5871
5872         return d_splice_alias(inode, dentry);
5873 }
5874
5875 unsigned char btrfs_filetype_table[] = {
5876         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5877 };
5878
5879 /*
5880  * All this infrastructure exists because dir_emit can fault, and we are holding
5881  * the tree lock when doing readdir.  For now just allocate a buffer and copy
5882  * our information into that, and then dir_emit from the buffer.  This is
5883  * similar to what NFS does, only we don't keep the buffer around in pagecache
5884  * because I'm afraid I'll mess that up.  Long term we need to make filldir do
5885  * copy_to_user_inatomic so we don't have to worry about page faulting under the
5886  * tree lock.
5887  */
5888 static int btrfs_opendir(struct inode *inode, struct file *file)
5889 {
5890         struct btrfs_file_private *private;
5891
5892         private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5893         if (!private)
5894                 return -ENOMEM;
5895         private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5896         if (!private->filldir_buf) {
5897                 kfree(private);
5898                 return -ENOMEM;
5899         }
5900         file->private_data = private;
5901         return 0;
5902 }
5903
5904 struct dir_entry {
5905         u64 ino;
5906         u64 offset;
5907         unsigned type;
5908         int name_len;
5909 };
5910
5911 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5912 {
5913         while (entries--) {
5914                 struct dir_entry *entry = addr;
5915                 char *name = (char *)(entry + 1);
5916
5917                 ctx->pos = entry->offset;
5918                 if (!dir_emit(ctx, name, entry->name_len, entry->ino,
5919                               entry->type))
5920                         return 1;
5921                 addr += sizeof(struct dir_entry) + entry->name_len;
5922                 ctx->pos++;
5923         }
5924         return 0;
5925 }
5926
5927 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5928 {
5929         struct inode *inode = file_inode(file);
5930         struct btrfs_root *root = BTRFS_I(inode)->root;
5931         struct btrfs_file_private *private = file->private_data;
5932         struct btrfs_dir_item *di;
5933         struct btrfs_key key;
5934         struct btrfs_key found_key;
5935         struct btrfs_path *path;
5936         void *addr;
5937         struct list_head ins_list;
5938         struct list_head del_list;
5939         int ret;
5940         struct extent_buffer *leaf;
5941         int slot;
5942         char *name_ptr;
5943         int name_len;
5944         int entries = 0;
5945         int total_len = 0;
5946         bool put = false;
5947         struct btrfs_key location;
5948
5949         if (!dir_emit_dots(file, ctx))
5950                 return 0;
5951
5952         path = btrfs_alloc_path();
5953         if (!path)
5954                 return -ENOMEM;
5955
5956         addr = private->filldir_buf;
5957         path->reada = READA_FORWARD;
5958
5959         INIT_LIST_HEAD(&ins_list);
5960         INIT_LIST_HEAD(&del_list);
5961         put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5962
5963 again:
5964         key.type = BTRFS_DIR_INDEX_KEY;
5965         key.offset = ctx->pos;
5966         key.objectid = btrfs_ino(BTRFS_I(inode));
5967
5968         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5969         if (ret < 0)
5970                 goto err;
5971
5972         while (1) {
5973                 struct dir_entry *entry;
5974
5975                 leaf = path->nodes[0];
5976                 slot = path->slots[0];
5977                 if (slot >= btrfs_header_nritems(leaf)) {
5978                         ret = btrfs_next_leaf(root, path);
5979                         if (ret < 0)
5980                                 goto err;
5981                         else if (ret > 0)
5982                                 break;
5983                         continue;
5984                 }
5985
5986                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5987
5988                 if (found_key.objectid != key.objectid)
5989                         break;
5990                 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5991                         break;
5992                 if (found_key.offset < ctx->pos)
5993                         goto next;
5994                 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5995                         goto next;
5996                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5997                 name_len = btrfs_dir_name_len(leaf, di);
5998                 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5999                     PAGE_SIZE) {
6000                         btrfs_release_path(path);
6001                         ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6002                         if (ret)
6003                                 goto nopos;
6004                         addr = private->filldir_buf;
6005                         entries = 0;
6006                         total_len = 0;
6007                         goto again;
6008                 }
6009
6010                 entry = addr;
6011                 entry->name_len = name_len;
6012                 name_ptr = (char *)(entry + 1);
6013                 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6014                                    name_len);
6015                 entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
6016                 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6017                 entry->ino = location.objectid;
6018                 entry->offset = found_key.offset;
6019                 entries++;
6020                 addr += sizeof(struct dir_entry) + name_len;
6021                 total_len += sizeof(struct dir_entry) + name_len;
6022 next:
6023                 path->slots[0]++;
6024         }
6025         btrfs_release_path(path);
6026
6027         ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6028         if (ret)
6029                 goto nopos;
6030
6031         ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6032         if (ret)
6033                 goto nopos;
6034
6035         /*
6036          * Stop new entries from being returned after we return the last
6037          * entry.
6038          *
6039          * New directory entries are assigned a strictly increasing
6040          * offset.  This means that new entries created during readdir
6041          * are *guaranteed* to be seen in the future by that readdir.
6042          * This has broken buggy programs which operate on names as
6043          * they're returned by readdir.  Until we re-use freed offsets
6044          * we have this hack to stop new entries from being returned
6045          * under the assumption that they'll never reach this huge
6046          * offset.
6047          *
6048          * This is being careful not to overflow 32bit loff_t unless the
6049          * last entry requires it because doing so has broken 32bit apps
6050          * in the past.
6051          */
6052         if (ctx->pos >= INT_MAX)
6053                 ctx->pos = LLONG_MAX;
6054         else
6055                 ctx->pos = INT_MAX;
6056 nopos:
6057         ret = 0;
6058 err:
6059         if (put)
6060                 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6061         btrfs_free_path(path);
6062         return ret;
6063 }
6064
6065 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6066 {
6067         struct btrfs_root *root = BTRFS_I(inode)->root;
6068         struct btrfs_trans_handle *trans;
6069         int ret = 0;
6070         bool nolock = false;
6071
6072         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6073                 return 0;
6074
6075         if (btrfs_fs_closing(root->fs_info) &&
6076                         btrfs_is_free_space_inode(BTRFS_I(inode)))
6077                 nolock = true;
6078
6079         if (wbc->sync_mode == WB_SYNC_ALL) {
6080                 if (nolock)
6081                         trans = btrfs_join_transaction_nolock(root);
6082                 else
6083                         trans = btrfs_join_transaction(root);
6084                 if (IS_ERR(trans))
6085                         return PTR_ERR(trans);
6086                 ret = btrfs_commit_transaction(trans);
6087         }
6088         return ret;
6089 }
6090
6091 /*
6092  * This is somewhat expensive, updating the tree every time the
6093  * inode changes.  But, it is most likely to find the inode in cache.
6094  * FIXME, needs more benchmarking...there are no reasons other than performance
6095  * to keep or drop this code.
6096  */
6097 static int btrfs_dirty_inode(struct inode *inode)
6098 {
6099         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6100         struct btrfs_root *root = BTRFS_I(inode)->root;
6101         struct btrfs_trans_handle *trans;
6102         int ret;
6103
6104         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6105                 return 0;
6106
6107         trans = btrfs_join_transaction(root);
6108         if (IS_ERR(trans))
6109                 return PTR_ERR(trans);
6110
6111         ret = btrfs_update_inode(trans, root, inode);
6112         if (ret && ret == -ENOSPC) {
6113                 /* whoops, lets try again with the full transaction */
6114                 btrfs_end_transaction(trans);
6115                 trans = btrfs_start_transaction(root, 1);
6116                 if (IS_ERR(trans))
6117                         return PTR_ERR(trans);
6118
6119                 ret = btrfs_update_inode(trans, root, inode);
6120         }
6121         btrfs_end_transaction(trans);
6122         if (BTRFS_I(inode)->delayed_node)
6123                 btrfs_balance_delayed_items(fs_info);
6124
6125         return ret;
6126 }
6127
6128 /*
6129  * This is a copy of file_update_time.  We need this so we can return error on
6130  * ENOSPC for updating the inode in the case of file write and mmap writes.
6131  */
6132 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6133                              int flags)
6134 {
6135         struct btrfs_root *root = BTRFS_I(inode)->root;
6136
6137         if (btrfs_root_readonly(root))
6138                 return -EROFS;
6139
6140         if (flags & S_VERSION)
6141                 inode_inc_iversion(inode);
6142         if (flags & S_CTIME)
6143                 inode->i_ctime = *now;
6144         if (flags & S_MTIME)
6145                 inode->i_mtime = *now;
6146         if (flags & S_ATIME)
6147                 inode->i_atime = *now;
6148         return btrfs_dirty_inode(inode);
6149 }
6150
6151 /*
6152  * find the highest existing sequence number in a directory
6153  * and then set the in-memory index_cnt variable to reflect
6154  * free sequence numbers
6155  */
6156 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6157 {
6158         struct btrfs_root *root = inode->root;
6159         struct btrfs_key key, found_key;
6160         struct btrfs_path *path;
6161         struct extent_buffer *leaf;
6162         int ret;
6163
6164         key.objectid = btrfs_ino(inode);
6165         key.type = BTRFS_DIR_INDEX_KEY;
6166         key.offset = (u64)-1;
6167
6168         path = btrfs_alloc_path();
6169         if (!path)
6170                 return -ENOMEM;
6171
6172         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6173         if (ret < 0)
6174                 goto out;
6175         /* FIXME: we should be able to handle this */
6176         if (ret == 0)
6177                 goto out;
6178         ret = 0;
6179
6180         /*
6181          * MAGIC NUMBER EXPLANATION:
6182          * since we search a directory based on f_pos we have to start at 2
6183          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6184          * else has to start at 2
6185          */
6186         if (path->slots[0] == 0) {
6187                 inode->index_cnt = 2;
6188                 goto out;
6189         }
6190
6191         path->slots[0]--;
6192
6193         leaf = path->nodes[0];
6194         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6195
6196         if (found_key.objectid != btrfs_ino(inode) ||
6197             found_key.type != BTRFS_DIR_INDEX_KEY) {
6198                 inode->index_cnt = 2;
6199                 goto out;
6200         }
6201
6202         inode->index_cnt = found_key.offset + 1;
6203 out:
6204         btrfs_free_path(path);
6205         return ret;
6206 }
6207
6208 /*
6209  * helper to find a free sequence number in a given directory.  This current
6210  * code is very simple, later versions will do smarter things in the btree
6211  */
6212 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6213 {
6214         int ret = 0;
6215
6216         if (dir->index_cnt == (u64)-1) {
6217                 ret = btrfs_inode_delayed_dir_index_count(dir);
6218                 if (ret) {
6219                         ret = btrfs_set_inode_index_count(dir);
6220                         if (ret)
6221                                 return ret;
6222                 }
6223         }
6224
6225         *index = dir->index_cnt;
6226         dir->index_cnt++;
6227
6228         return ret;
6229 }
6230
6231 static int btrfs_insert_inode_locked(struct inode *inode)
6232 {
6233         struct btrfs_iget_args args;
6234         args.location = &BTRFS_I(inode)->location;
6235         args.root = BTRFS_I(inode)->root;
6236
6237         return insert_inode_locked4(inode,
6238                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6239                    btrfs_find_actor, &args);
6240 }
6241
6242 /*
6243  * Inherit flags from the parent inode.
6244  *
6245  * Currently only the compression flags and the cow flags are inherited.
6246  */
6247 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6248 {
6249         unsigned int flags;
6250
6251         if (!dir)
6252                 return;
6253
6254         flags = BTRFS_I(dir)->flags;
6255
6256         if (flags & BTRFS_INODE_NOCOMPRESS) {
6257                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6258                 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6259         } else if (flags & BTRFS_INODE_COMPRESS) {
6260                 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6261                 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6262         }
6263
6264         if (flags & BTRFS_INODE_NODATACOW) {
6265                 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6266                 if (S_ISREG(inode->i_mode))
6267                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6268         }
6269
6270         btrfs_update_iflags(inode);
6271 }
6272
6273 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6274                                      struct btrfs_root *root,
6275                                      struct inode *dir,
6276                                      const char *name, int name_len,
6277                                      u64 ref_objectid, u64 objectid,
6278                                      umode_t mode, u64 *index)
6279 {
6280         struct btrfs_fs_info *fs_info = root->fs_info;
6281         struct inode *inode;
6282         struct btrfs_inode_item *inode_item;
6283         struct btrfs_key *location;
6284         struct btrfs_path *path;
6285         struct btrfs_inode_ref *ref;
6286         struct btrfs_key key[2];
6287         u32 sizes[2];
6288         int nitems = name ? 2 : 1;
6289         unsigned long ptr;
6290         int ret;
6291
6292         path = btrfs_alloc_path();
6293         if (!path)
6294                 return ERR_PTR(-ENOMEM);
6295
6296         inode = new_inode(fs_info->sb);
6297         if (!inode) {
6298                 btrfs_free_path(path);
6299                 return ERR_PTR(-ENOMEM);
6300         }
6301
6302         /*
6303          * O_TMPFILE, set link count to 0, so that after this point,
6304          * we fill in an inode item with the correct link count.
6305          */
6306         if (!name)
6307                 set_nlink(inode, 0);
6308
6309         /*
6310          * we have to initialize this early, so we can reclaim the inode
6311          * number if we fail afterwards in this function.
6312          */
6313         inode->i_ino = objectid;
6314
6315         if (dir && name) {
6316                 trace_btrfs_inode_request(dir);
6317
6318                 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6319                 if (ret) {
6320                         btrfs_free_path(path);
6321                         iput(inode);
6322                         return ERR_PTR(ret);
6323                 }
6324         } else if (dir) {
6325                 *index = 0;
6326         }
6327         /*
6328          * index_cnt is ignored for everything but a dir,
6329          * btrfs_set_inode_index_count has an explanation for the magic
6330          * number
6331          */
6332         BTRFS_I(inode)->index_cnt = 2;
6333         BTRFS_I(inode)->dir_index = *index;
6334         BTRFS_I(inode)->root = root;
6335         BTRFS_I(inode)->generation = trans->transid;
6336         inode->i_generation = BTRFS_I(inode)->generation;
6337
6338         /*
6339          * We could have gotten an inode number from somebody who was fsynced
6340          * and then removed in this same transaction, so let's just set full
6341          * sync since it will be a full sync anyway and this will blow away the
6342          * old info in the log.
6343          */
6344         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6345
6346         key[0].objectid = objectid;
6347         key[0].type = BTRFS_INODE_ITEM_KEY;
6348         key[0].offset = 0;
6349
6350         sizes[0] = sizeof(struct btrfs_inode_item);
6351
6352         if (name) {
6353                 /*
6354                  * Start new inodes with an inode_ref. This is slightly more
6355                  * efficient for small numbers of hard links since they will
6356                  * be packed into one item. Extended refs will kick in if we
6357                  * add more hard links than can fit in the ref item.
6358                  */
6359                 key[1].objectid = objectid;
6360                 key[1].type = BTRFS_INODE_REF_KEY;
6361                 key[1].offset = ref_objectid;
6362
6363                 sizes[1] = name_len + sizeof(*ref);
6364         }
6365
6366         location = &BTRFS_I(inode)->location;
6367         location->objectid = objectid;
6368         location->offset = 0;
6369         location->type = BTRFS_INODE_ITEM_KEY;
6370
6371         ret = btrfs_insert_inode_locked(inode);
6372         if (ret < 0)
6373                 goto fail;
6374
6375         path->leave_spinning = 1;
6376         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6377         if (ret != 0)
6378                 goto fail_unlock;
6379
6380         inode_init_owner(inode, dir, mode);
6381         inode_set_bytes(inode, 0);
6382
6383         inode->i_mtime = current_time(inode);
6384         inode->i_atime = inode->i_mtime;
6385         inode->i_ctime = inode->i_mtime;
6386         BTRFS_I(inode)->i_otime = inode->i_mtime;
6387
6388         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6389                                   struct btrfs_inode_item);
6390         memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6391                              sizeof(*inode_item));
6392         fill_inode_item(trans, path->nodes[0], inode_item, inode);
6393
6394         if (name) {
6395                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6396                                      struct btrfs_inode_ref);
6397                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6398                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6399                 ptr = (unsigned long)(ref + 1);
6400                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6401         }
6402
6403         btrfs_mark_buffer_dirty(path->nodes[0]);
6404         btrfs_free_path(path);
6405
6406         btrfs_inherit_iflags(inode, dir);
6407
6408         if (S_ISREG(mode)) {
6409                 if (btrfs_test_opt(fs_info, NODATASUM))
6410                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6411                 if (btrfs_test_opt(fs_info, NODATACOW))
6412                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6413                                 BTRFS_INODE_NODATASUM;
6414         }
6415
6416         inode_tree_add(inode);
6417
6418         trace_btrfs_inode_new(inode);
6419         btrfs_set_inode_last_trans(trans, inode);
6420
6421         btrfs_update_root_times(trans, root);
6422
6423         ret = btrfs_inode_inherit_props(trans, inode, dir);
6424         if (ret)
6425                 btrfs_err(fs_info,
6426                           "error inheriting props for ino %llu (root %llu): %d",
6427                         btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6428
6429         return inode;
6430
6431 fail_unlock:
6432         unlock_new_inode(inode);
6433 fail:
6434         if (dir && name)
6435                 BTRFS_I(dir)->index_cnt--;
6436         btrfs_free_path(path);
6437         iput(inode);
6438         return ERR_PTR(ret);
6439 }
6440
6441 static inline u8 btrfs_inode_type(struct inode *inode)
6442 {
6443         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6444 }
6445
6446 /*
6447  * utility function to add 'inode' into 'parent_inode' with
6448  * a give name and a given sequence number.
6449  * if 'add_backref' is true, also insert a backref from the
6450  * inode to the parent directory.
6451  */
6452 int btrfs_add_link(struct btrfs_trans_handle *trans,
6453                    struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6454                    const char *name, int name_len, int add_backref, u64 index)
6455 {
6456         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6457         int ret = 0;
6458         struct btrfs_key key;
6459         struct btrfs_root *root = parent_inode->root;
6460         u64 ino = btrfs_ino(inode);
6461         u64 parent_ino = btrfs_ino(parent_inode);
6462
6463         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6464                 memcpy(&key, &inode->root->root_key, sizeof(key));
6465         } else {
6466                 key.objectid = ino;
6467                 key.type = BTRFS_INODE_ITEM_KEY;
6468                 key.offset = 0;
6469         }
6470
6471         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6472                 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6473                                          root->root_key.objectid, parent_ino,
6474                                          index, name, name_len);
6475         } else if (add_backref) {
6476                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6477                                              parent_ino, index);
6478         }
6479
6480         /* Nothing to clean up yet */
6481         if (ret)
6482                 return ret;
6483
6484         ret = btrfs_insert_dir_item(trans, root, name, name_len,
6485                                     parent_inode, &key,
6486                                     btrfs_inode_type(&inode->vfs_inode), index);
6487         if (ret == -EEXIST || ret == -EOVERFLOW)
6488                 goto fail_dir_item;
6489         else if (ret) {
6490                 btrfs_abort_transaction(trans, ret);
6491                 return ret;
6492         }
6493
6494         btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6495                            name_len * 2);
6496         inode_inc_iversion(&parent_inode->vfs_inode);
6497         parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6498                 current_time(&parent_inode->vfs_inode);
6499         ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6500         if (ret)
6501                 btrfs_abort_transaction(trans, ret);
6502         return ret;
6503
6504 fail_dir_item:
6505         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6506                 u64 local_index;
6507                 int err;
6508                 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6509                                          root->root_key.objectid, parent_ino,
6510                                          &local_index, name, name_len);
6511
6512         } else if (add_backref) {
6513                 u64 local_index;
6514                 int err;
6515
6516                 err = btrfs_del_inode_ref(trans, root, name, name_len,
6517                                           ino, parent_ino, &local_index);
6518         }
6519         return ret;
6520 }
6521
6522 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6523                             struct btrfs_inode *dir, struct dentry *dentry,
6524                             struct btrfs_inode *inode, int backref, u64 index)
6525 {
6526         int err = btrfs_add_link(trans, dir, inode,
6527                                  dentry->d_name.name, dentry->d_name.len,
6528                                  backref, index);
6529         if (err > 0)
6530                 err = -EEXIST;
6531         return err;
6532 }
6533
6534 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6535                         umode_t mode, dev_t rdev)
6536 {
6537         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6538         struct btrfs_trans_handle *trans;
6539         struct btrfs_root *root = BTRFS_I(dir)->root;
6540         struct inode *inode = NULL;
6541         int err;
6542         int drop_inode = 0;
6543         u64 objectid;
6544         u64 index = 0;
6545
6546         /*
6547          * 2 for inode item and ref
6548          * 2 for dir items
6549          * 1 for xattr if selinux is on
6550          */
6551         trans = btrfs_start_transaction(root, 5);
6552         if (IS_ERR(trans))
6553                 return PTR_ERR(trans);
6554
6555         err = btrfs_find_free_ino(root, &objectid);
6556         if (err)
6557                 goto out_unlock;
6558
6559         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6560                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6561                         mode, &index);
6562         if (IS_ERR(inode)) {
6563                 err = PTR_ERR(inode);
6564                 goto out_unlock;
6565         }
6566
6567         /*
6568         * If the active LSM wants to access the inode during
6569         * d_instantiate it needs these. Smack checks to see
6570         * if the filesystem supports xattrs by looking at the
6571         * ops vector.
6572         */
6573         inode->i_op = &btrfs_special_inode_operations;
6574         init_special_inode(inode, inode->i_mode, rdev);
6575
6576         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6577         if (err)
6578                 goto out_unlock_inode;
6579
6580         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6581                         0, index);
6582         if (err) {
6583                 goto out_unlock_inode;
6584         } else {
6585                 btrfs_update_inode(trans, root, inode);
6586                 unlock_new_inode(inode);
6587                 d_instantiate(dentry, inode);
6588         }
6589
6590 out_unlock:
6591         btrfs_end_transaction(trans);
6592         btrfs_btree_balance_dirty(fs_info);
6593         if (drop_inode) {
6594                 inode_dec_link_count(inode);
6595                 iput(inode);
6596         }
6597         return err;
6598
6599 out_unlock_inode:
6600         drop_inode = 1;
6601         unlock_new_inode(inode);
6602         goto out_unlock;
6603
6604 }
6605
6606 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6607                         umode_t mode, bool excl)
6608 {
6609         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6610         struct btrfs_trans_handle *trans;
6611         struct btrfs_root *root = BTRFS_I(dir)->root;
6612         struct inode *inode = NULL;
6613         int drop_inode_on_err = 0;
6614         int err;
6615         u64 objectid;
6616         u64 index = 0;
6617
6618         /*
6619          * 2 for inode item and ref
6620          * 2 for dir items
6621          * 1 for xattr if selinux is on
6622          */
6623         trans = btrfs_start_transaction(root, 5);
6624         if (IS_ERR(trans))
6625                 return PTR_ERR(trans);
6626
6627         err = btrfs_find_free_ino(root, &objectid);
6628         if (err)
6629                 goto out_unlock;
6630
6631         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6632                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6633                         mode, &index);
6634         if (IS_ERR(inode)) {
6635                 err = PTR_ERR(inode);
6636                 goto out_unlock;
6637         }
6638         drop_inode_on_err = 1;
6639         /*
6640         * If the active LSM wants to access the inode during
6641         * d_instantiate it needs these. Smack checks to see
6642         * if the filesystem supports xattrs by looking at the
6643         * ops vector.
6644         */
6645         inode->i_fop = &btrfs_file_operations;
6646         inode->i_op = &btrfs_file_inode_operations;
6647         inode->i_mapping->a_ops = &btrfs_aops;
6648
6649         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6650         if (err)
6651                 goto out_unlock_inode;
6652
6653         err = btrfs_update_inode(trans, root, inode);
6654         if (err)
6655                 goto out_unlock_inode;
6656
6657         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6658                         0, index);
6659         if (err)
6660                 goto out_unlock_inode;
6661
6662         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6663         unlock_new_inode(inode);
6664         d_instantiate(dentry, inode);
6665
6666 out_unlock:
6667         btrfs_end_transaction(trans);
6668         if (err && drop_inode_on_err) {
6669                 inode_dec_link_count(inode);
6670                 iput(inode);
6671         }
6672         btrfs_btree_balance_dirty(fs_info);
6673         return err;
6674
6675 out_unlock_inode:
6676         unlock_new_inode(inode);
6677         goto out_unlock;
6678
6679 }
6680
6681 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6682                       struct dentry *dentry)
6683 {
6684         struct btrfs_trans_handle *trans = NULL;
6685         struct btrfs_root *root = BTRFS_I(dir)->root;
6686         struct inode *inode = d_inode(old_dentry);
6687         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6688         u64 index;
6689         int err;
6690         int drop_inode = 0;
6691
6692         /* do not allow sys_link's with other subvols of the same device */
6693         if (root->objectid != BTRFS_I(inode)->root->objectid)
6694                 return -EXDEV;
6695
6696         if (inode->i_nlink >= BTRFS_LINK_MAX)
6697                 return -EMLINK;
6698
6699         err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6700         if (err)
6701                 goto fail;
6702
6703         /*
6704          * 2 items for inode and inode ref
6705          * 2 items for dir items
6706          * 1 item for parent inode
6707          */
6708         trans = btrfs_start_transaction(root, 5);
6709         if (IS_ERR(trans)) {
6710                 err = PTR_ERR(trans);
6711                 trans = NULL;
6712                 goto fail;
6713         }
6714
6715         /* There are several dir indexes for this inode, clear the cache. */
6716         BTRFS_I(inode)->dir_index = 0ULL;
6717         inc_nlink(inode);
6718         inode_inc_iversion(inode);
6719         inode->i_ctime = current_time(inode);
6720         ihold(inode);
6721         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6722
6723         err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6724                         1, index);
6725
6726         if (err) {
6727                 drop_inode = 1;
6728         } else {
6729                 struct dentry *parent = dentry->d_parent;
6730                 err = btrfs_update_inode(trans, root, inode);
6731                 if (err)
6732                         goto fail;
6733                 if (inode->i_nlink == 1) {
6734                         /*
6735                          * If new hard link count is 1, it's a file created
6736                          * with open(2) O_TMPFILE flag.
6737                          */
6738                         err = btrfs_orphan_del(trans, BTRFS_I(inode));
6739                         if (err)
6740                                 goto fail;
6741                 }
6742                 d_instantiate(dentry, inode);
6743                 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6744         }
6745
6746 fail:
6747         if (trans)
6748                 btrfs_end_transaction(trans);
6749         if (drop_inode) {
6750                 inode_dec_link_count(inode);
6751                 iput(inode);
6752         }
6753         btrfs_btree_balance_dirty(fs_info);
6754         return err;
6755 }
6756
6757 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6758 {
6759         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6760         struct inode *inode = NULL;
6761         struct btrfs_trans_handle *trans;
6762         struct btrfs_root *root = BTRFS_I(dir)->root;
6763         int err = 0;
6764         int drop_on_err = 0;
6765         u64 objectid = 0;
6766         u64 index = 0;
6767
6768         /*
6769          * 2 items for inode and ref
6770          * 2 items for dir items
6771          * 1 for xattr if selinux is on
6772          */
6773         trans = btrfs_start_transaction(root, 5);
6774         if (IS_ERR(trans))
6775                 return PTR_ERR(trans);
6776
6777         err = btrfs_find_free_ino(root, &objectid);
6778         if (err)
6779                 goto out_fail;
6780
6781         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6782                         dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6783                         S_IFDIR | mode, &index);
6784         if (IS_ERR(inode)) {
6785                 err = PTR_ERR(inode);
6786                 goto out_fail;
6787         }
6788
6789         drop_on_err = 1;
6790         /* these must be set before we unlock the inode */
6791         inode->i_op = &btrfs_dir_inode_operations;
6792         inode->i_fop = &btrfs_dir_file_operations;
6793
6794         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6795         if (err)
6796                 goto out_fail_inode;
6797
6798         btrfs_i_size_write(BTRFS_I(inode), 0);
6799         err = btrfs_update_inode(trans, root, inode);
6800         if (err)
6801                 goto out_fail_inode;
6802
6803         err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6804                         dentry->d_name.name,
6805                         dentry->d_name.len, 0, index);
6806         if (err)
6807                 goto out_fail_inode;
6808
6809         d_instantiate(dentry, inode);
6810         /*
6811          * mkdir is special.  We're unlocking after we call d_instantiate
6812          * to avoid a race with nfsd calling d_instantiate.
6813          */
6814         unlock_new_inode(inode);
6815         drop_on_err = 0;
6816
6817 out_fail:
6818         btrfs_end_transaction(trans);
6819         if (drop_on_err) {
6820                 inode_dec_link_count(inode);
6821                 iput(inode);
6822         }
6823         btrfs_btree_balance_dirty(fs_info);
6824         return err;
6825
6826 out_fail_inode:
6827         unlock_new_inode(inode);
6828         goto out_fail;
6829 }
6830
6831 static noinline int uncompress_inline(struct btrfs_path *path,
6832                                       struct page *page,
6833                                       size_t pg_offset, u64 extent_offset,
6834                                       struct btrfs_file_extent_item *item)
6835 {
6836         int ret;
6837         struct extent_buffer *leaf = path->nodes[0];
6838         char *tmp;
6839         size_t max_size;
6840         unsigned long inline_size;
6841         unsigned long ptr;
6842         int compress_type;
6843
6844         WARN_ON(pg_offset != 0);
6845         compress_type = btrfs_file_extent_compression(leaf, item);
6846         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6847         inline_size = btrfs_file_extent_inline_item_len(leaf,
6848                                         btrfs_item_nr(path->slots[0]));
6849         tmp = kmalloc(inline_size, GFP_NOFS);
6850         if (!tmp)
6851                 return -ENOMEM;
6852         ptr = btrfs_file_extent_inline_start(item);
6853
6854         read_extent_buffer(leaf, tmp, ptr, inline_size);
6855
6856         max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6857         ret = btrfs_decompress(compress_type, tmp, page,
6858                                extent_offset, inline_size, max_size);
6859
6860         /*
6861          * decompression code contains a memset to fill in any space between the end
6862          * of the uncompressed data and the end of max_size in case the decompressed
6863          * data ends up shorter than ram_bytes.  That doesn't cover the hole between
6864          * the end of an inline extent and the beginning of the next block, so we
6865          * cover that region here.
6866          */
6867
6868         if (max_size + pg_offset < PAGE_SIZE) {
6869                 char *map = kmap(page);
6870                 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6871                 kunmap(page);
6872         }
6873         kfree(tmp);
6874         return ret;
6875 }
6876
6877 /*
6878  * a bit scary, this does extent mapping from logical file offset to the disk.
6879  * the ugly parts come from merging extents from the disk with the in-ram
6880  * representation.  This gets more complex because of the data=ordered code,
6881  * where the in-ram extents might be locked pending data=ordered completion.
6882  *
6883  * This also copies inline extents directly into the page.
6884  */
6885 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6886                 struct page *page,
6887             size_t pg_offset, u64 start, u64 len,
6888                 int create)
6889 {
6890         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6891         int ret;
6892         int err = 0;
6893         u64 extent_start = 0;
6894         u64 extent_end = 0;
6895         u64 objectid = btrfs_ino(inode);
6896         u32 found_type;
6897         struct btrfs_path *path = NULL;
6898         struct btrfs_root *root = inode->root;
6899         struct btrfs_file_extent_item *item;
6900         struct extent_buffer *leaf;
6901         struct btrfs_key found_key;
6902         struct extent_map *em = NULL;
6903         struct extent_map_tree *em_tree = &inode->extent_tree;
6904         struct extent_io_tree *io_tree = &inode->io_tree;
6905         const bool new_inline = !page || create;
6906
6907         read_lock(&em_tree->lock);
6908         em = lookup_extent_mapping(em_tree, start, len);
6909         if (em)
6910                 em->bdev = fs_info->fs_devices->latest_bdev;
6911         read_unlock(&em_tree->lock);
6912
6913         if (em) {
6914                 if (em->start > start || em->start + em->len <= start)
6915                         free_extent_map(em);
6916                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6917                         free_extent_map(em);
6918                 else
6919                         goto out;
6920         }
6921         em = alloc_extent_map();
6922         if (!em) {
6923                 err = -ENOMEM;
6924                 goto out;
6925         }
6926         em->bdev = fs_info->fs_devices->latest_bdev;
6927         em->start = EXTENT_MAP_HOLE;
6928         em->orig_start = EXTENT_MAP_HOLE;
6929         em->len = (u64)-1;
6930         em->block_len = (u64)-1;
6931
6932         if (!path) {
6933                 path = btrfs_alloc_path();
6934                 if (!path) {
6935                         err = -ENOMEM;
6936                         goto out;
6937                 }
6938                 /*
6939                  * Chances are we'll be called again, so go ahead and do
6940                  * readahead
6941                  */
6942                 path->reada = READA_FORWARD;
6943         }
6944
6945         ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6946         if (ret < 0) {
6947                 err = ret;
6948                 goto out;
6949         }
6950
6951         if (ret != 0) {
6952                 if (path->slots[0] == 0)
6953                         goto not_found;
6954                 path->slots[0]--;
6955         }
6956
6957         leaf = path->nodes[0];
6958         item = btrfs_item_ptr(leaf, path->slots[0],
6959                               struct btrfs_file_extent_item);
6960         /* are we inside the extent that was found? */
6961         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6962         found_type = found_key.type;
6963         if (found_key.objectid != objectid ||
6964             found_type != BTRFS_EXTENT_DATA_KEY) {
6965                 /*
6966                  * If we backup past the first extent we want to move forward
6967                  * and see if there is an extent in front of us, otherwise we'll
6968                  * say there is a hole for our whole search range which can
6969                  * cause problems.
6970                  */
6971                 extent_end = start;
6972                 goto next;
6973         }
6974
6975         found_type = btrfs_file_extent_type(leaf, item);
6976         extent_start = found_key.offset;
6977         if (found_type == BTRFS_FILE_EXTENT_REG ||
6978             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6979                 extent_end = extent_start +
6980                        btrfs_file_extent_num_bytes(leaf, item);
6981
6982                 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6983                                                        extent_start);
6984         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6985                 size_t size;
6986                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6987                 extent_end = ALIGN(extent_start + size,
6988                                    fs_info->sectorsize);
6989
6990                 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6991                                                       path->slots[0],
6992                                                       extent_start);
6993         }
6994 next:
6995         if (start >= extent_end) {
6996                 path->slots[0]++;
6997                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6998                         ret = btrfs_next_leaf(root, path);
6999                         if (ret < 0) {
7000                                 err = ret;
7001                                 goto out;
7002                         }
7003                         if (ret > 0)
7004                                 goto not_found;
7005                         leaf = path->nodes[0];
7006                 }
7007                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7008                 if (found_key.objectid != objectid ||
7009                     found_key.type != BTRFS_EXTENT_DATA_KEY)
7010                         goto not_found;
7011                 if (start + len <= found_key.offset)
7012                         goto not_found;
7013                 if (start > found_key.offset)
7014                         goto next;
7015                 em->start = start;
7016                 em->orig_start = start;
7017                 em->len = found_key.offset - start;
7018                 goto not_found_em;
7019         }
7020
7021         btrfs_extent_item_to_extent_map(inode, path, item,
7022                         new_inline, em);
7023
7024         if (found_type == BTRFS_FILE_EXTENT_REG ||
7025             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7026                 goto insert;
7027         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7028                 unsigned long ptr;
7029                 char *map;
7030                 size_t size;
7031                 size_t extent_offset;
7032                 size_t copy_size;
7033
7034                 if (new_inline)
7035                         goto out;
7036
7037                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7038                 extent_offset = page_offset(page) + pg_offset - extent_start;
7039                 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7040                                   size - extent_offset);
7041                 em->start = extent_start + extent_offset;
7042                 em->len = ALIGN(copy_size, fs_info->sectorsize);
7043                 em->orig_block_len = em->len;
7044                 em->orig_start = em->start;
7045                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7046                 if (!PageUptodate(page)) {
7047                         if (btrfs_file_extent_compression(leaf, item) !=
7048                             BTRFS_COMPRESS_NONE) {
7049                                 ret = uncompress_inline(path, page, pg_offset,
7050                                                         extent_offset, item);
7051                                 if (ret) {
7052                                         err = ret;
7053                                         goto out;
7054                                 }
7055                         } else {
7056                                 map = kmap(page);
7057                                 read_extent_buffer(leaf, map + pg_offset, ptr,
7058                                                    copy_size);
7059                                 if (pg_offset + copy_size < PAGE_SIZE) {
7060                                         memset(map + pg_offset + copy_size, 0,
7061                                                PAGE_SIZE - pg_offset -
7062                                                copy_size);
7063                                 }
7064                                 kunmap(page);
7065                         }
7066                         flush_dcache_page(page);
7067                 }
7068                 set_extent_uptodate(io_tree, em->start,
7069                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
7070                 goto insert;
7071         }
7072 not_found:
7073         em->start = start;
7074         em->orig_start = start;
7075         em->len = len;
7076 not_found_em:
7077         em->block_start = EXTENT_MAP_HOLE;
7078 insert:
7079         btrfs_release_path(path);
7080         if (em->start > start || extent_map_end(em) <= start) {
7081                 btrfs_err(fs_info,
7082                           "bad extent! em: [%llu %llu] passed [%llu %llu]",
7083                           em->start, em->len, start, len);
7084                 err = -EIO;
7085                 goto out;
7086         }
7087
7088         err = 0;
7089         write_lock(&em_tree->lock);
7090         err = btrfs_add_extent_mapping(em_tree, &em, start, len);
7091         write_unlock(&em_tree->lock);
7092 out:
7093
7094         trace_btrfs_get_extent(root, inode, em);
7095
7096         btrfs_free_path(path);
7097         if (err) {
7098                 free_extent_map(em);
7099                 return ERR_PTR(err);
7100         }
7101         BUG_ON(!em); /* Error is always set */
7102         return em;
7103 }
7104
7105 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7106                 struct page *page,
7107                 size_t pg_offset, u64 start, u64 len,
7108                 int create)
7109 {
7110         struct extent_map *em;
7111         struct extent_map *hole_em = NULL;
7112         u64 range_start = start;
7113         u64 end;
7114         u64 found;
7115         u64 found_end;
7116         int err = 0;
7117
7118         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7119         if (IS_ERR(em))
7120                 return em;
7121         /*
7122          * If our em maps to:
7123          * - a hole or
7124          * - a pre-alloc extent,
7125          * there might actually be delalloc bytes behind it.
7126          */
7127         if (em->block_start != EXTENT_MAP_HOLE &&
7128             !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7129                 return em;
7130         else
7131                 hole_em = em;
7132
7133         /* check to see if we've wrapped (len == -1 or similar) */
7134         end = start + len;
7135         if (end < start)
7136                 end = (u64)-1;
7137         else
7138                 end -= 1;
7139
7140         em = NULL;
7141
7142         /* ok, we didn't find anything, lets look for delalloc */
7143         found = count_range_bits(&inode->io_tree, &range_start,
7144                                  end, len, EXTENT_DELALLOC, 1);
7145         found_end = range_start + found;
7146         if (found_end < range_start)
7147                 found_end = (u64)-1;
7148
7149         /*
7150          * we didn't find anything useful, return
7151          * the original results from get_extent()
7152          */
7153         if (range_start > end || found_end <= start) {
7154                 em = hole_em;
7155                 hole_em = NULL;
7156                 goto out;
7157         }
7158
7159         /* adjust the range_start to make sure it doesn't
7160          * go backwards from the start they passed in
7161          */
7162         range_start = max(start, range_start);
7163         found = found_end - range_start;
7164
7165         if (found > 0) {
7166                 u64 hole_start = start;
7167                 u64 hole_len = len;
7168
7169                 em = alloc_extent_map();
7170                 if (!em) {
7171                         err = -ENOMEM;
7172                         goto out;
7173                 }
7174                 /*
7175                  * when btrfs_get_extent can't find anything it
7176                  * returns one huge hole
7177                  *
7178                  * make sure what it found really fits our range, and
7179                  * adjust to make sure it is based on the start from
7180                  * the caller
7181                  */
7182                 if (hole_em) {
7183                         u64 calc_end = extent_map_end(hole_em);
7184
7185                         if (calc_end <= start || (hole_em->start > end)) {
7186                                 free_extent_map(hole_em);
7187                                 hole_em = NULL;
7188                         } else {
7189                                 hole_start = max(hole_em->start, start);
7190                                 hole_len = calc_end - hole_start;
7191                         }
7192                 }
7193                 em->bdev = NULL;
7194                 if (hole_em && range_start > hole_start) {
7195                         /* our hole starts before our delalloc, so we
7196                          * have to return just the parts of the hole
7197                          * that go until  the delalloc starts
7198                          */
7199                         em->len = min(hole_len,
7200                                       range_start - hole_start);
7201                         em->start = hole_start;
7202                         em->orig_start = hole_start;
7203                         /*
7204                          * don't adjust block start at all,
7205                          * it is fixed at EXTENT_MAP_HOLE
7206                          */
7207                         em->block_start = hole_em->block_start;
7208                         em->block_len = hole_len;
7209                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7210                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7211                 } else {
7212                         em->start = range_start;
7213                         em->len = found;
7214                         em->orig_start = range_start;
7215                         em->block_start = EXTENT_MAP_DELALLOC;
7216                         em->block_len = found;
7217                 }
7218         } else {
7219                 return hole_em;
7220         }
7221 out:
7222
7223         free_extent_map(hole_em);
7224         if (err) {
7225                 free_extent_map(em);
7226                 return ERR_PTR(err);
7227         }
7228         return em;
7229 }
7230
7231 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7232                                                   const u64 start,
7233                                                   const u64 len,
7234                                                   const u64 orig_start,
7235                                                   const u64 block_start,
7236                                                   const u64 block_len,
7237                                                   const u64 orig_block_len,
7238                                                   const u64 ram_bytes,
7239                                                   const int type)
7240 {
7241         struct extent_map *em = NULL;
7242         int ret;
7243
7244         if (type != BTRFS_ORDERED_NOCOW) {
7245                 em = create_io_em(inode, start, len, orig_start,
7246                                   block_start, block_len, orig_block_len,
7247                                   ram_bytes,
7248                                   BTRFS_COMPRESS_NONE, /* compress_type */
7249                                   type);
7250                 if (IS_ERR(em))
7251                         goto out;
7252         }
7253         ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7254                                            len, block_len, type);
7255         if (ret) {
7256                 if (em) {
7257                         free_extent_map(em);
7258                         btrfs_drop_extent_cache(BTRFS_I(inode), start,
7259                                                 start + len - 1, 0);
7260                 }
7261                 em = ERR_PTR(ret);
7262         }
7263  out:
7264
7265         return em;
7266 }
7267
7268 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7269                                                   u64 start, u64 len)
7270 {
7271         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7272         struct btrfs_root *root = BTRFS_I(inode)->root;
7273         struct extent_map *em;
7274         struct btrfs_key ins;
7275         u64 alloc_hint;
7276         int ret;
7277
7278         alloc_hint = get_extent_allocation_hint(inode, start, len);
7279         ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7280                                    0, alloc_hint, &ins, 1, 1);
7281         if (ret)
7282                 return ERR_PTR(ret);
7283
7284         em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7285                                      ins.objectid, ins.offset, ins.offset,
7286                                      ins.offset, BTRFS_ORDERED_REGULAR);
7287         btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7288         if (IS_ERR(em))
7289                 btrfs_free_reserved_extent(fs_info, ins.objectid,
7290                                            ins.offset, 1);
7291
7292         return em;
7293 }
7294
7295 /*
7296  * returns 1 when the nocow is safe, < 1 on error, 0 if the
7297  * block must be cow'd
7298  */
7299 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7300                               u64 *orig_start, u64 *orig_block_len,
7301                               u64 *ram_bytes)
7302 {
7303         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7304         struct btrfs_path *path;
7305         int ret;
7306         struct extent_buffer *leaf;
7307         struct btrfs_root *root = BTRFS_I(inode)->root;
7308         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7309         struct btrfs_file_extent_item *fi;
7310         struct btrfs_key key;
7311         u64 disk_bytenr;
7312         u64 backref_offset;
7313         u64 extent_end;
7314         u64 num_bytes;
7315         int slot;
7316         int found_type;
7317         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7318
7319         path = btrfs_alloc_path();
7320         if (!path)
7321                 return -ENOMEM;
7322
7323         ret = btrfs_lookup_file_extent(NULL, root, path,
7324                         btrfs_ino(BTRFS_I(inode)), offset, 0);
7325         if (ret < 0)
7326                 goto out;
7327
7328         slot = path->slots[0];
7329         if (ret == 1) {
7330                 if (slot == 0) {
7331                         /* can't find the item, must cow */
7332                         ret = 0;
7333                         goto out;
7334                 }
7335                 slot--;
7336         }
7337         ret = 0;
7338         leaf = path->nodes[0];
7339         btrfs_item_key_to_cpu(leaf, &key, slot);
7340         if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7341             key.type != BTRFS_EXTENT_DATA_KEY) {
7342                 /* not our file or wrong item type, must cow */
7343                 goto out;
7344         }
7345
7346         if (key.offset > offset) {
7347                 /* Wrong offset, must cow */
7348                 goto out;
7349         }
7350
7351         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7352         found_type = btrfs_file_extent_type(leaf, fi);
7353         if (found_type != BTRFS_FILE_EXTENT_REG &&
7354             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7355                 /* not a regular extent, must cow */
7356                 goto out;
7357         }
7358
7359         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7360                 goto out;
7361
7362         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7363         if (extent_end <= offset)
7364                 goto out;
7365
7366         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7367         if (disk_bytenr == 0)
7368                 goto out;
7369
7370         if (btrfs_file_extent_compression(leaf, fi) ||
7371             btrfs_file_extent_encryption(leaf, fi) ||
7372             btrfs_file_extent_other_encoding(leaf, fi))
7373                 goto out;
7374
7375         backref_offset = btrfs_file_extent_offset(leaf, fi);
7376
7377         if (orig_start) {
7378                 *orig_start = key.offset - backref_offset;
7379                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7380                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7381         }
7382
7383         if (btrfs_extent_readonly(fs_info, disk_bytenr))
7384                 goto out;
7385
7386         num_bytes = min(offset + *len, extent_end) - offset;
7387         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7388                 u64 range_end;
7389
7390                 range_end = round_up(offset + num_bytes,
7391                                      root->fs_info->sectorsize) - 1;
7392                 ret = test_range_bit(io_tree, offset, range_end,
7393                                      EXTENT_DELALLOC, 0, NULL);
7394                 if (ret) {
7395                         ret = -EAGAIN;
7396                         goto out;
7397                 }
7398         }
7399
7400         btrfs_release_path(path);
7401
7402         /*
7403          * look for other files referencing this extent, if we
7404          * find any we must cow
7405          */
7406
7407         ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7408                                     key.offset - backref_offset, disk_bytenr);
7409         if (ret) {
7410                 ret = 0;
7411                 goto out;
7412         }
7413
7414         /*
7415          * adjust disk_bytenr and num_bytes to cover just the bytes
7416          * in this extent we are about to write.  If there
7417          * are any csums in that range we have to cow in order
7418          * to keep the csums correct
7419          */
7420         disk_bytenr += backref_offset;
7421         disk_bytenr += offset - key.offset;
7422         if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7423                 goto out;
7424         /*
7425          * all of the above have passed, it is safe to overwrite this extent
7426          * without cow
7427          */
7428         *len = num_bytes;
7429         ret = 1;
7430 out:
7431         btrfs_free_path(path);
7432         return ret;
7433 }
7434
7435 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7436 {
7437         struct radix_tree_root *root = &inode->i_mapping->page_tree;
7438         bool found = false;
7439         void **pagep = NULL;
7440         struct page *page = NULL;
7441         unsigned long start_idx;
7442         unsigned long end_idx;
7443
7444         start_idx = start >> PAGE_SHIFT;
7445
7446         /*
7447          * end is the last byte in the last page.  end == start is legal
7448          */
7449         end_idx = end >> PAGE_SHIFT;
7450
7451         rcu_read_lock();
7452
7453         /* Most of the code in this while loop is lifted from
7454          * find_get_page.  It's been modified to begin searching from a
7455          * page and return just the first page found in that range.  If the
7456          * found idx is less than or equal to the end idx then we know that
7457          * a page exists.  If no pages are found or if those pages are
7458          * outside of the range then we're fine (yay!) */
7459         while (page == NULL &&
7460                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7461                 page = radix_tree_deref_slot(pagep);
7462                 if (unlikely(!page))
7463                         break;
7464
7465                 if (radix_tree_exception(page)) {
7466                         if (radix_tree_deref_retry(page)) {
7467                                 page = NULL;
7468                                 continue;
7469                         }
7470                         /*
7471                          * Otherwise, shmem/tmpfs must be storing a swap entry
7472                          * here as an exceptional entry: so return it without
7473                          * attempting to raise page count.
7474                          */
7475                         page = NULL;
7476                         break; /* TODO: Is this relevant for this use case? */
7477                 }
7478
7479                 if (!page_cache_get_speculative(page)) {
7480                         page = NULL;
7481                         continue;
7482                 }
7483
7484                 /*
7485                  * Has the page moved?
7486                  * This is part of the lockless pagecache protocol. See
7487                  * include/linux/pagemap.h for details.
7488                  */
7489                 if (unlikely(page != *pagep)) {
7490                         put_page(page);
7491                         page = NULL;
7492                 }
7493         }
7494
7495         if (page) {
7496                 if (page->index <= end_idx)
7497                         found = true;
7498                 put_page(page);
7499         }
7500
7501         rcu_read_unlock();
7502         return found;
7503 }
7504
7505 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7506                               struct extent_state **cached_state, int writing)
7507 {
7508         struct btrfs_ordered_extent *ordered;
7509         int ret = 0;
7510
7511         while (1) {
7512                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7513                                  cached_state);
7514                 /*
7515                  * We're concerned with the entire range that we're going to be
7516                  * doing DIO to, so we need to make sure there's no ordered
7517                  * extents in this range.
7518                  */
7519                 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7520                                                      lockend - lockstart + 1);
7521
7522                 /*
7523                  * We need to make sure there are no buffered pages in this
7524                  * range either, we could have raced between the invalidate in
7525                  * generic_file_direct_write and locking the extent.  The
7526                  * invalidate needs to happen so that reads after a write do not
7527                  * get stale data.
7528                  */
7529                 if (!ordered &&
7530                     (!writing ||
7531                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7532                         break;
7533
7534                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7535                                      cached_state);
7536
7537                 if (ordered) {
7538                         /*
7539                          * If we are doing a DIO read and the ordered extent we
7540                          * found is for a buffered write, we can not wait for it
7541                          * to complete and retry, because if we do so we can
7542                          * deadlock with concurrent buffered writes on page
7543                          * locks. This happens only if our DIO read covers more
7544                          * than one extent map, if at this point has already
7545                          * created an ordered extent for a previous extent map
7546                          * and locked its range in the inode's io tree, and a
7547                          * concurrent write against that previous extent map's
7548                          * range and this range started (we unlock the ranges
7549                          * in the io tree only when the bios complete and
7550                          * buffered writes always lock pages before attempting
7551                          * to lock range in the io tree).
7552                          */
7553                         if (writing ||
7554                             test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7555                                 btrfs_start_ordered_extent(inode, ordered, 1);
7556                         else
7557                                 ret = -ENOTBLK;
7558                         btrfs_put_ordered_extent(ordered);
7559                 } else {
7560                         /*
7561                          * We could trigger writeback for this range (and wait
7562                          * for it to complete) and then invalidate the pages for
7563                          * this range (through invalidate_inode_pages2_range()),
7564                          * but that can lead us to a deadlock with a concurrent
7565                          * call to readpages() (a buffered read or a defrag call
7566                          * triggered a readahead) on a page lock due to an
7567                          * ordered dio extent we created before but did not have
7568                          * yet a corresponding bio submitted (whence it can not
7569                          * complete), which makes readpages() wait for that
7570                          * ordered extent to complete while holding a lock on
7571                          * that page.
7572                          */
7573                         ret = -ENOTBLK;
7574                 }
7575
7576                 if (ret)
7577                         break;
7578
7579                 cond_resched();
7580         }
7581
7582         return ret;
7583 }
7584
7585 /* The callers of this must take lock_extent() */
7586 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7587                                        u64 orig_start, u64 block_start,
7588                                        u64 block_len, u64 orig_block_len,
7589                                        u64 ram_bytes, int compress_type,
7590                                        int type)
7591 {
7592         struct extent_map_tree *em_tree;
7593         struct extent_map *em;
7594         struct btrfs_root *root = BTRFS_I(inode)->root;
7595         int ret;
7596
7597         ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7598                type == BTRFS_ORDERED_COMPRESSED ||
7599                type == BTRFS_ORDERED_NOCOW ||
7600                type == BTRFS_ORDERED_REGULAR);
7601
7602         em_tree = &BTRFS_I(inode)->extent_tree;
7603         em = alloc_extent_map();
7604         if (!em)
7605                 return ERR_PTR(-ENOMEM);
7606
7607         em->start = start;
7608         em->orig_start = orig_start;
7609         em->len = len;
7610         em->block_len = block_len;
7611         em->block_start = block_start;
7612         em->bdev = root->fs_info->fs_devices->latest_bdev;
7613         em->orig_block_len = orig_block_len;
7614         em->ram_bytes = ram_bytes;
7615         em->generation = -1;
7616         set_bit(EXTENT_FLAG_PINNED, &em->flags);
7617         if (type == BTRFS_ORDERED_PREALLOC) {
7618                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7619         } else if (type == BTRFS_ORDERED_COMPRESSED) {
7620                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7621                 em->compress_type = compress_type;
7622         }
7623
7624         do {
7625                 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7626                                 em->start + em->len - 1, 0);
7627                 write_lock(&em_tree->lock);
7628                 ret = add_extent_mapping(em_tree, em, 1);
7629                 write_unlock(&em_tree->lock);
7630                 /*
7631                  * The caller has taken lock_extent(), who could race with us
7632                  * to add em?
7633                  */
7634         } while (ret == -EEXIST);
7635
7636         if (ret) {
7637                 free_extent_map(em);
7638                 return ERR_PTR(ret);
7639         }
7640
7641         /* em got 2 refs now, callers needs to do free_extent_map once. */
7642         return em;
7643 }
7644
7645 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7646                                    struct buffer_head *bh_result, int create)
7647 {
7648         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7649         struct extent_map *em;
7650         struct extent_state *cached_state = NULL;
7651         struct btrfs_dio_data *dio_data = NULL;
7652         u64 start = iblock << inode->i_blkbits;
7653         u64 lockstart, lockend;
7654         u64 len = bh_result->b_size;
7655         int unlock_bits = EXTENT_LOCKED;
7656         int ret = 0;
7657
7658         if (create)
7659                 unlock_bits |= EXTENT_DIRTY;
7660         else
7661                 len = min_t(u64, len, fs_info->sectorsize);
7662
7663         lockstart = start;
7664         lockend = start + len - 1;
7665
7666         if (current->journal_info) {
7667                 /*
7668                  * Need to pull our outstanding extents and set journal_info to NULL so
7669                  * that anything that needs to check if there's a transaction doesn't get
7670                  * confused.
7671                  */
7672                 dio_data = current->journal_info;
7673                 current->journal_info = NULL;
7674         }
7675
7676         /*
7677          * If this errors out it's because we couldn't invalidate pagecache for
7678          * this range and we need to fallback to buffered.
7679          */
7680         if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7681                                create)) {
7682                 ret = -ENOTBLK;
7683                 goto err;
7684         }
7685
7686         em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7687         if (IS_ERR(em)) {
7688                 ret = PTR_ERR(em);
7689                 goto unlock_err;
7690         }
7691
7692         /*
7693          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7694          * io.  INLINE is special, and we could probably kludge it in here, but
7695          * it's still buffered so for safety lets just fall back to the generic
7696          * buffered path.
7697          *
7698          * For COMPRESSED we _have_ to read the entire extent in so we can
7699          * decompress it, so there will be buffering required no matter what we
7700          * do, so go ahead and fallback to buffered.
7701          *
7702          * We return -ENOTBLK because that's what makes DIO go ahead and go back
7703          * to buffered IO.  Don't blame me, this is the price we pay for using
7704          * the generic code.
7705          */
7706         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7707             em->block_start == EXTENT_MAP_INLINE) {
7708                 free_extent_map(em);
7709                 ret = -ENOTBLK;
7710                 goto unlock_err;
7711         }
7712
7713         /* Just a good old fashioned hole, return */
7714         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7715                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7716                 free_extent_map(em);
7717                 goto unlock_err;
7718         }
7719
7720         /*
7721          * We don't allocate a new extent in the following cases
7722          *
7723          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7724          * existing extent.
7725          * 2) The extent is marked as PREALLOC.  We're good to go here and can
7726          * just use the extent.
7727          *
7728          */
7729         if (!create) {
7730                 len = min(len, em->len - (start - em->start));
7731                 lockstart = start + len;
7732                 goto unlock;
7733         }
7734
7735         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7736             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7737              em->block_start != EXTENT_MAP_HOLE)) {
7738                 int type;
7739                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7740
7741                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7742                         type = BTRFS_ORDERED_PREALLOC;
7743                 else
7744                         type = BTRFS_ORDERED_NOCOW;
7745                 len = min(len, em->len - (start - em->start));
7746                 block_start = em->block_start + (start - em->start);
7747
7748                 if (can_nocow_extent(inode, start, &len, &orig_start,
7749                                      &orig_block_len, &ram_bytes) == 1 &&
7750                     btrfs_inc_nocow_writers(fs_info, block_start)) {
7751                         struct extent_map *em2;
7752
7753                         em2 = btrfs_create_dio_extent(inode, start, len,
7754                                                       orig_start, block_start,
7755                                                       len, orig_block_len,
7756                                                       ram_bytes, type);
7757                         btrfs_dec_nocow_writers(fs_info, block_start);
7758                         if (type == BTRFS_ORDERED_PREALLOC) {
7759                                 free_extent_map(em);
7760                                 em = em2;
7761                         }
7762                         if (em2 && IS_ERR(em2)) {
7763                                 ret = PTR_ERR(em2);
7764                                 goto unlock_err;
7765                         }
7766                         /*
7767                          * For inode marked NODATACOW or extent marked PREALLOC,
7768                          * use the existing or preallocated extent, so does not
7769                          * need to adjust btrfs_space_info's bytes_may_use.
7770                          */
7771                         btrfs_free_reserved_data_space_noquota(inode,
7772                                         start, len);
7773                         goto unlock;
7774                 }
7775         }
7776
7777         /*
7778          * this will cow the extent, reset the len in case we changed
7779          * it above
7780          */
7781         len = bh_result->b_size;
7782         free_extent_map(em);
7783         em = btrfs_new_extent_direct(inode, start, len);
7784         if (IS_ERR(em)) {
7785                 ret = PTR_ERR(em);
7786                 goto unlock_err;
7787         }
7788         len = min(len, em->len - (start - em->start));
7789 unlock:
7790         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7791                 inode->i_blkbits;
7792         bh_result->b_size = len;
7793         bh_result->b_bdev = em->bdev;
7794         set_buffer_mapped(bh_result);
7795         if (create) {
7796                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7797                         set_buffer_new(bh_result);
7798
7799                 /*
7800                  * Need to update the i_size under the extent lock so buffered
7801                  * readers will get the updated i_size when we unlock.
7802                  */
7803                 if (!dio_data->overwrite && start + len > i_size_read(inode))
7804                         i_size_write(inode, start + len);
7805
7806                 WARN_ON(dio_data->reserve < len);
7807                 dio_data->reserve -= len;
7808                 dio_data->unsubmitted_oe_range_end = start + len;
7809                 current->journal_info = dio_data;
7810         }
7811
7812         /*
7813          * In the case of write we need to clear and unlock the entire range,
7814          * in the case of read we need to unlock only the end area that we
7815          * aren't using if there is any left over space.
7816          */
7817         if (lockstart < lockend) {
7818                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7819                                  lockend, unlock_bits, 1, 0,
7820                                  &cached_state);
7821         } else {
7822                 free_extent_state(cached_state);
7823         }
7824
7825         free_extent_map(em);
7826
7827         return 0;
7828
7829 unlock_err:
7830         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7831                          unlock_bits, 1, 0, &cached_state);
7832 err:
7833         if (dio_data)
7834                 current->journal_info = dio_data;
7835         return ret;
7836 }
7837
7838 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7839                                                  struct bio *bio,
7840                                                  int mirror_num)
7841 {
7842         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7843         blk_status_t ret;
7844
7845         BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7846
7847         ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7848         if (ret)
7849                 return ret;
7850
7851         ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7852
7853         return ret;
7854 }
7855
7856 static int btrfs_check_dio_repairable(struct inode *inode,
7857                                       struct bio *failed_bio,
7858                                       struct io_failure_record *failrec,
7859                                       int failed_mirror)
7860 {
7861         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7862         int num_copies;
7863
7864         num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7865         if (num_copies == 1) {
7866                 /*
7867                  * we only have a single copy of the data, so don't bother with
7868                  * all the retry and error correction code that follows. no
7869                  * matter what the error is, it is very likely to persist.
7870                  */
7871                 btrfs_debug(fs_info,
7872                         "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7873                         num_copies, failrec->this_mirror, failed_mirror);
7874                 return 0;
7875         }
7876
7877         failrec->failed_mirror = failed_mirror;
7878         failrec->this_mirror++;
7879         if (failrec->this_mirror == failed_mirror)
7880                 failrec->this_mirror++;
7881
7882         if (failrec->this_mirror > num_copies) {
7883                 btrfs_debug(fs_info,
7884                         "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7885                         num_copies, failrec->this_mirror, failed_mirror);
7886                 return 0;
7887         }
7888
7889         return 1;
7890 }
7891
7892 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7893                                    struct page *page, unsigned int pgoff,
7894                                    u64 start, u64 end, int failed_mirror,
7895                                    bio_end_io_t *repair_endio, void *repair_arg)
7896 {
7897         struct io_failure_record *failrec;
7898         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7899         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7900         struct bio *bio;
7901         int isector;
7902         unsigned int read_mode = 0;
7903         int segs;
7904         int ret;
7905         blk_status_t status;
7906
7907         BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7908
7909         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7910         if (ret)
7911                 return errno_to_blk_status(ret);
7912
7913         ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7914                                          failed_mirror);
7915         if (!ret) {
7916                 free_io_failure(failure_tree, io_tree, failrec);
7917                 return BLK_STS_IOERR;
7918         }
7919
7920         segs = bio_segments(failed_bio);
7921         if (segs > 1 ||
7922             (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
7923                 read_mode |= REQ_FAILFAST_DEV;
7924
7925         isector = start - btrfs_io_bio(failed_bio)->logical;
7926         isector >>= inode->i_sb->s_blocksize_bits;
7927         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7928                                 pgoff, isector, repair_endio, repair_arg);
7929         bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7930
7931         btrfs_debug(BTRFS_I(inode)->root->fs_info,
7932                     "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7933                     read_mode, failrec->this_mirror, failrec->in_validation);
7934
7935         status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7936         if (status) {
7937                 free_io_failure(failure_tree, io_tree, failrec);
7938                 bio_put(bio);
7939         }
7940
7941         return status;
7942 }
7943
7944 struct btrfs_retry_complete {
7945         struct completion done;
7946         struct inode *inode;
7947         u64 start;
7948         int uptodate;
7949 };
7950
7951 static void btrfs_retry_endio_nocsum(struct bio *bio)
7952 {
7953         struct btrfs_retry_complete *done = bio->bi_private;
7954         struct inode *inode = done->inode;
7955         struct bio_vec *bvec;
7956         struct extent_io_tree *io_tree, *failure_tree;
7957         int i;
7958
7959         if (bio->bi_status)
7960                 goto end;
7961
7962         ASSERT(bio->bi_vcnt == 1);
7963         io_tree = &BTRFS_I(inode)->io_tree;
7964         failure_tree = &BTRFS_I(inode)->io_failure_tree;
7965         ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
7966
7967         done->uptodate = 1;
7968         ASSERT(!bio_flagged(bio, BIO_CLONED));
7969         bio_for_each_segment_all(bvec, bio, i)
7970                 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7971                                  io_tree, done->start, bvec->bv_page,
7972                                  btrfs_ino(BTRFS_I(inode)), 0);
7973 end:
7974         complete(&done->done);
7975         bio_put(bio);
7976 }
7977
7978 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7979                                                 struct btrfs_io_bio *io_bio)
7980 {
7981         struct btrfs_fs_info *fs_info;
7982         struct bio_vec bvec;
7983         struct bvec_iter iter;
7984         struct btrfs_retry_complete done;
7985         u64 start;
7986         unsigned int pgoff;
7987         u32 sectorsize;
7988         int nr_sectors;
7989         blk_status_t ret;
7990         blk_status_t err = BLK_STS_OK;
7991
7992         fs_info = BTRFS_I(inode)->root->fs_info;
7993         sectorsize = fs_info->sectorsize;
7994
7995         start = io_bio->logical;
7996         done.inode = inode;
7997         io_bio->bio.bi_iter = io_bio->iter;
7998
7999         bio_for_each_segment(bvec, &io_bio->bio, iter) {
8000                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8001                 pgoff = bvec.bv_offset;
8002
8003 next_block_or_try_again:
8004                 done.uptodate = 0;
8005                 done.start = start;
8006                 init_completion(&done.done);
8007
8008                 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8009                                 pgoff, start, start + sectorsize - 1,
8010                                 io_bio->mirror_num,
8011                                 btrfs_retry_endio_nocsum, &done);
8012                 if (ret) {
8013                         err = ret;
8014                         goto next;
8015                 }
8016
8017                 wait_for_completion_io(&done.done);
8018
8019                 if (!done.uptodate) {
8020                         /* We might have another mirror, so try again */
8021                         goto next_block_or_try_again;
8022                 }
8023
8024 next:
8025                 start += sectorsize;
8026
8027                 nr_sectors--;
8028                 if (nr_sectors) {
8029                         pgoff += sectorsize;
8030                         ASSERT(pgoff < PAGE_SIZE);
8031                         goto next_block_or_try_again;
8032                 }
8033         }
8034
8035         return err;
8036 }
8037
8038 static void btrfs_retry_endio(struct bio *bio)
8039 {
8040         struct btrfs_retry_complete *done = bio->bi_private;
8041         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8042         struct extent_io_tree *io_tree, *failure_tree;
8043         struct inode *inode = done->inode;
8044         struct bio_vec *bvec;
8045         int uptodate;
8046         int ret;
8047         int i;
8048
8049         if (bio->bi_status)
8050                 goto end;
8051
8052         uptodate = 1;
8053
8054         ASSERT(bio->bi_vcnt == 1);
8055         ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8056
8057         io_tree = &BTRFS_I(inode)->io_tree;
8058         failure_tree = &BTRFS_I(inode)->io_failure_tree;
8059
8060         ASSERT(!bio_flagged(bio, BIO_CLONED));
8061         bio_for_each_segment_all(bvec, bio, i) {
8062                 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8063                                              bvec->bv_offset, done->start,
8064                                              bvec->bv_len);
8065                 if (!ret)
8066                         clean_io_failure(BTRFS_I(inode)->root->fs_info,
8067                                          failure_tree, io_tree, done->start,
8068                                          bvec->bv_page,
8069                                          btrfs_ino(BTRFS_I(inode)),
8070                                          bvec->bv_offset);
8071                 else
8072                         uptodate = 0;
8073         }
8074
8075         done->uptodate = uptodate;
8076 end:
8077         complete(&done->done);
8078         bio_put(bio);
8079 }
8080
8081 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8082                 struct btrfs_io_bio *io_bio, blk_status_t err)
8083 {
8084         struct btrfs_fs_info *fs_info;
8085         struct bio_vec bvec;
8086         struct bvec_iter iter;
8087         struct btrfs_retry_complete done;
8088         u64 start;
8089         u64 offset = 0;
8090         u32 sectorsize;
8091         int nr_sectors;
8092         unsigned int pgoff;
8093         int csum_pos;
8094         bool uptodate = (err == 0);
8095         int ret;
8096         blk_status_t status;
8097
8098         fs_info = BTRFS_I(inode)->root->fs_info;
8099         sectorsize = fs_info->sectorsize;
8100
8101         err = BLK_STS_OK;
8102         start = io_bio->logical;
8103         done.inode = inode;
8104         io_bio->bio.bi_iter = io_bio->iter;
8105
8106         bio_for_each_segment(bvec, &io_bio->bio, iter) {
8107                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8108
8109                 pgoff = bvec.bv_offset;
8110 next_block:
8111                 if (uptodate) {
8112                         csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8113                         ret = __readpage_endio_check(inode, io_bio, csum_pos,
8114                                         bvec.bv_page, pgoff, start, sectorsize);
8115                         if (likely(!ret))
8116                                 goto next;
8117                 }
8118 try_again:
8119                 done.uptodate = 0;
8120                 done.start = start;
8121                 init_completion(&done.done);
8122
8123                 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8124                                         pgoff, start, start + sectorsize - 1,
8125                                         io_bio->mirror_num, btrfs_retry_endio,
8126                                         &done);
8127                 if (status) {
8128                         err = status;
8129                         goto next;
8130                 }
8131
8132                 wait_for_completion_io(&done.done);
8133
8134                 if (!done.uptodate) {
8135                         /* We might have another mirror, so try again */
8136                         goto try_again;
8137                 }
8138 next:
8139                 offset += sectorsize;
8140                 start += sectorsize;
8141
8142                 ASSERT(nr_sectors);
8143
8144                 nr_sectors--;
8145                 if (nr_sectors) {
8146                         pgoff += sectorsize;
8147                         ASSERT(pgoff < PAGE_SIZE);
8148                         goto next_block;
8149                 }
8150         }
8151
8152         return err;
8153 }
8154
8155 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8156                 struct btrfs_io_bio *io_bio, blk_status_t err)
8157 {
8158         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8159
8160         if (skip_csum) {
8161                 if (unlikely(err))
8162                         return __btrfs_correct_data_nocsum(inode, io_bio);
8163                 else
8164                         return BLK_STS_OK;
8165         } else {
8166                 return __btrfs_subio_endio_read(inode, io_bio, err);
8167         }
8168 }
8169
8170 static void btrfs_endio_direct_read(struct bio *bio)
8171 {
8172         struct btrfs_dio_private *dip = bio->bi_private;
8173         struct inode *inode = dip->inode;
8174         struct bio *dio_bio;
8175         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8176         blk_status_t err = bio->bi_status;
8177
8178         if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8179                 err = btrfs_subio_endio_read(inode, io_bio, err);
8180
8181         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8182                       dip->logical_offset + dip->bytes - 1);
8183         dio_bio = dip->dio_bio;
8184
8185         kfree(dip);
8186
8187         dio_bio->bi_status = err;
8188         dio_end_io(dio_bio);
8189
8190         if (io_bio->end_io)
8191                 io_bio->end_io(io_bio, blk_status_to_errno(err));
8192         bio_put(bio);
8193 }
8194
8195 static void __endio_write_update_ordered(struct inode *inode,
8196                                          const u64 offset, const u64 bytes,
8197                                          const bool uptodate)
8198 {
8199         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8200         struct btrfs_ordered_extent *ordered = NULL;
8201         struct btrfs_workqueue *wq;
8202         btrfs_work_func_t func;
8203         u64 ordered_offset = offset;
8204         u64 ordered_bytes = bytes;
8205         u64 last_offset;
8206         int ret;
8207
8208         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8209                 wq = fs_info->endio_freespace_worker;
8210                 func = btrfs_freespace_write_helper;
8211         } else {
8212                 wq = fs_info->endio_write_workers;
8213                 func = btrfs_endio_write_helper;
8214         }
8215
8216 again:
8217         last_offset = ordered_offset;
8218         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8219                                                    &ordered_offset,
8220                                                    ordered_bytes,
8221                                                    uptodate);
8222         if (!ret)
8223                 goto out_test;
8224
8225         btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8226         btrfs_queue_work(wq, &ordered->work);
8227 out_test:
8228         /*
8229          * If btrfs_dec_test_ordered_pending does not find any ordered extent
8230          * in the range, we can exit.
8231          */
8232         if (ordered_offset == last_offset)
8233                 return;
8234         /*
8235          * our bio might span multiple ordered extents.  If we haven't
8236          * completed the accounting for the whole dio, go back and try again
8237          */
8238         if (ordered_offset < offset + bytes) {
8239                 ordered_bytes = offset + bytes - ordered_offset;
8240                 ordered = NULL;
8241                 goto again;
8242         }
8243 }
8244
8245 static void btrfs_endio_direct_write(struct bio *bio)
8246 {
8247         struct btrfs_dio_private *dip = bio->bi_private;
8248         struct bio *dio_bio = dip->dio_bio;
8249
8250         __endio_write_update_ordered(dip->inode, dip->logical_offset,
8251                                      dip->bytes, !bio->bi_status);
8252
8253         kfree(dip);
8254
8255         dio_bio->bi_status = bio->bi_status;
8256         dio_end_io(dio_bio);
8257         bio_put(bio);
8258 }
8259
8260 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8261                                     struct bio *bio, int mirror_num,
8262                                     unsigned long bio_flags, u64 offset)
8263 {
8264         struct inode *inode = private_data;
8265         blk_status_t ret;
8266         ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8267         BUG_ON(ret); /* -ENOMEM */
8268         return 0;
8269 }
8270
8271 static void btrfs_end_dio_bio(struct bio *bio)
8272 {
8273         struct btrfs_dio_private *dip = bio->bi_private;
8274         blk_status_t err = bio->bi_status;
8275
8276         if (err)
8277                 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8278                            "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8279                            btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8280                            bio->bi_opf,
8281                            (unsigned long long)bio->bi_iter.bi_sector,
8282                            bio->bi_iter.bi_size, err);
8283
8284         if (dip->subio_endio)
8285                 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8286
8287         if (err) {
8288                 dip->errors = 1;
8289
8290                 /*
8291                  * before atomic variable goto zero, we must make sure
8292                  * dip->errors is perceived to be set.
8293                  */
8294                 smp_mb__before_atomic();
8295         }
8296
8297         /* if there are more bios still pending for this dio, just exit */
8298         if (!atomic_dec_and_test(&dip->pending_bios))
8299                 goto out;
8300
8301         if (dip->errors) {
8302                 bio_io_error(dip->orig_bio);
8303         } else {
8304                 dip->dio_bio->bi_status = BLK_STS_OK;
8305                 bio_endio(dip->orig_bio);
8306         }
8307 out:
8308         bio_put(bio);
8309 }
8310
8311 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8312                                                  struct btrfs_dio_private *dip,
8313                                                  struct bio *bio,
8314                                                  u64 file_offset)
8315 {
8316         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8317         struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8318         blk_status_t ret;
8319
8320         /*
8321          * We load all the csum data we need when we submit
8322          * the first bio to reduce the csum tree search and
8323          * contention.
8324          */
8325         if (dip->logical_offset == file_offset) {
8326                 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8327                                                 file_offset);
8328                 if (ret)
8329                         return ret;
8330         }
8331
8332         if (bio == dip->orig_bio)
8333                 return 0;
8334
8335         file_offset -= dip->logical_offset;
8336         file_offset >>= inode->i_sb->s_blocksize_bits;
8337         io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8338
8339         return 0;
8340 }
8341
8342 static inline blk_status_t
8343 __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
8344                        int async_submit)
8345 {
8346         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8347         struct btrfs_dio_private *dip = bio->bi_private;
8348         bool write = bio_op(bio) == REQ_OP_WRITE;
8349         blk_status_t ret;
8350
8351         /* Check btrfs_submit_bio_hook() for rules about async submit. */
8352         if (async_submit)
8353                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8354
8355         if (!write) {
8356                 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8357                 if (ret)
8358                         goto err;
8359         }
8360
8361         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8362                 goto map;
8363
8364         if (write && async_submit) {
8365                 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8366                                           file_offset, inode,
8367                                           __btrfs_submit_bio_start_direct_io,
8368                                           __btrfs_submit_bio_done);
8369                 goto err;
8370         } else if (write) {
8371                 /*
8372                  * If we aren't doing async submit, calculate the csum of the
8373                  * bio now.
8374                  */
8375                 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8376                 if (ret)
8377                         goto err;
8378         } else {
8379                 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8380                                                      file_offset);
8381                 if (ret)
8382                         goto err;
8383         }
8384 map:
8385         ret = btrfs_map_bio(fs_info, bio, 0, 0);
8386 err:
8387         return ret;
8388 }
8389
8390 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8391 {
8392         struct inode *inode = dip->inode;
8393         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8394         struct bio *bio;
8395         struct bio *orig_bio = dip->orig_bio;
8396         u64 start_sector = orig_bio->bi_iter.bi_sector;
8397         u64 file_offset = dip->logical_offset;
8398         u64 map_length;
8399         int async_submit = 0;
8400         u64 submit_len;
8401         int clone_offset = 0;
8402         int clone_len;
8403         int ret;
8404         blk_status_t status;
8405
8406         map_length = orig_bio->bi_iter.bi_size;
8407         submit_len = map_length;
8408         ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8409                               &map_length, NULL, 0);
8410         if (ret)
8411                 return -EIO;
8412
8413         if (map_length >= submit_len) {
8414                 bio = orig_bio;
8415                 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8416                 goto submit;
8417         }
8418
8419         /* async crcs make it difficult to collect full stripe writes. */
8420         if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8421                 async_submit = 0;
8422         else
8423                 async_submit = 1;
8424
8425         /* bio split */
8426         ASSERT(map_length <= INT_MAX);
8427         atomic_inc(&dip->pending_bios);
8428         do {
8429                 clone_len = min_t(int, submit_len, map_length);
8430
8431                 /*
8432                  * This will never fail as it's passing GPF_NOFS and
8433                  * the allocation is backed by btrfs_bioset.
8434                  */
8435                 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8436                                               clone_len);
8437                 bio->bi_private = dip;
8438                 bio->bi_end_io = btrfs_end_dio_bio;
8439                 btrfs_io_bio(bio)->logical = file_offset;
8440
8441                 ASSERT(submit_len >= clone_len);
8442                 submit_len -= clone_len;
8443                 if (submit_len == 0)
8444                         break;
8445
8446                 /*
8447                  * Increase the count before we submit the bio so we know
8448                  * the end IO handler won't happen before we increase the
8449                  * count. Otherwise, the dip might get freed before we're
8450                  * done setting it up.
8451                  */
8452                 atomic_inc(&dip->pending_bios);
8453
8454                 status = __btrfs_submit_dio_bio(bio, inode, file_offset,
8455                                                 async_submit);
8456                 if (status) {
8457                         bio_put(bio);
8458                         atomic_dec(&dip->pending_bios);
8459                         goto out_err;
8460                 }
8461
8462                 clone_offset += clone_len;
8463                 start_sector += clone_len >> 9;
8464                 file_offset += clone_len;
8465
8466                 map_length = submit_len;
8467                 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8468                                       start_sector << 9, &map_length, NULL, 0);
8469                 if (ret)
8470                         goto out_err;
8471         } while (submit_len > 0);
8472
8473 submit:
8474         status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8475         if (!status)
8476                 return 0;
8477
8478         bio_put(bio);
8479 out_err:
8480         dip->errors = 1;
8481         /*
8482          * before atomic variable goto zero, we must
8483          * make sure dip->errors is perceived to be set.
8484          */
8485         smp_mb__before_atomic();
8486         if (atomic_dec_and_test(&dip->pending_bios))
8487                 bio_io_error(dip->orig_bio);
8488
8489         /* bio_end_io() will handle error, so we needn't return it */
8490         return 0;
8491 }
8492
8493 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8494                                 loff_t file_offset)
8495 {
8496         struct btrfs_dio_private *dip = NULL;
8497         struct bio *bio = NULL;
8498         struct btrfs_io_bio *io_bio;
8499         bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8500         int ret = 0;
8501
8502         bio = btrfs_bio_clone(dio_bio);
8503
8504         dip = kzalloc(sizeof(*dip), GFP_NOFS);
8505         if (!dip) {
8506                 ret = -ENOMEM;
8507                 goto free_ordered;
8508         }
8509
8510         dip->private = dio_bio->bi_private;
8511         dip->inode = inode;
8512         dip->logical_offset = file_offset;
8513         dip->bytes = dio_bio->bi_iter.bi_size;
8514         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8515         bio->bi_private = dip;
8516         dip->orig_bio = bio;
8517         dip->dio_bio = dio_bio;
8518         atomic_set(&dip->pending_bios, 0);
8519         io_bio = btrfs_io_bio(bio);
8520         io_bio->logical = file_offset;
8521
8522         if (write) {
8523                 bio->bi_end_io = btrfs_endio_direct_write;
8524         } else {
8525                 bio->bi_end_io = btrfs_endio_direct_read;
8526                 dip->subio_endio = btrfs_subio_endio_read;
8527         }
8528
8529         /*
8530          * Reset the range for unsubmitted ordered extents (to a 0 length range)
8531          * even if we fail to submit a bio, because in such case we do the
8532          * corresponding error handling below and it must not be done a second
8533          * time by btrfs_direct_IO().
8534          */
8535         if (write) {
8536                 struct btrfs_dio_data *dio_data = current->journal_info;
8537
8538                 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8539                         dip->bytes;
8540                 dio_data->unsubmitted_oe_range_start =
8541                         dio_data->unsubmitted_oe_range_end;
8542         }
8543
8544         ret = btrfs_submit_direct_hook(dip);
8545         if (!ret)
8546                 return;
8547
8548         if (io_bio->end_io)
8549                 io_bio->end_io(io_bio, ret);
8550
8551 free_ordered:
8552         /*
8553          * If we arrived here it means either we failed to submit the dip
8554          * or we either failed to clone the dio_bio or failed to allocate the
8555          * dip. If we cloned the dio_bio and allocated the dip, we can just
8556          * call bio_endio against our io_bio so that we get proper resource
8557          * cleanup if we fail to submit the dip, otherwise, we must do the
8558          * same as btrfs_endio_direct_[write|read] because we can't call these
8559          * callbacks - they require an allocated dip and a clone of dio_bio.
8560          */
8561         if (bio && dip) {
8562                 bio_io_error(bio);
8563                 /*
8564                  * The end io callbacks free our dip, do the final put on bio
8565                  * and all the cleanup and final put for dio_bio (through
8566                  * dio_end_io()).
8567                  */
8568                 dip = NULL;
8569                 bio = NULL;
8570         } else {
8571                 if (write)
8572                         __endio_write_update_ordered(inode,
8573                                                 file_offset,
8574                                                 dio_bio->bi_iter.bi_size,
8575                                                 false);
8576                 else
8577                         unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8578                               file_offset + dio_bio->bi_iter.bi_size - 1);
8579
8580                 dio_bio->bi_status = BLK_STS_IOERR;
8581                 /*
8582                  * Releases and cleans up our dio_bio, no need to bio_put()
8583                  * nor bio_endio()/bio_io_error() against dio_bio.
8584                  */
8585                 dio_end_io(dio_bio);
8586         }
8587         if (bio)
8588                 bio_put(bio);
8589         kfree(dip);
8590 }
8591
8592 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8593                                const struct iov_iter *iter, loff_t offset)
8594 {
8595         int seg;
8596         int i;
8597         unsigned int blocksize_mask = fs_info->sectorsize - 1;
8598         ssize_t retval = -EINVAL;
8599
8600         if (offset & blocksize_mask)
8601                 goto out;
8602
8603         if (iov_iter_alignment(iter) & blocksize_mask)
8604                 goto out;
8605
8606         /* If this is a write we don't need to check anymore */
8607         if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8608                 return 0;
8609         /*
8610          * Check to make sure we don't have duplicate iov_base's in this
8611          * iovec, if so return EINVAL, otherwise we'll get csum errors
8612          * when reading back.
8613          */
8614         for (seg = 0; seg < iter->nr_segs; seg++) {
8615                 for (i = seg + 1; i < iter->nr_segs; i++) {
8616                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8617                                 goto out;
8618                 }
8619         }
8620         retval = 0;
8621 out:
8622         return retval;
8623 }
8624
8625 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8626 {
8627         struct file *file = iocb->ki_filp;
8628         struct inode *inode = file->f_mapping->host;
8629         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8630         struct btrfs_dio_data dio_data = { 0 };
8631         struct extent_changeset *data_reserved = NULL;
8632         loff_t offset = iocb->ki_pos;
8633         size_t count = 0;
8634         int flags = 0;
8635         bool wakeup = true;
8636         bool relock = false;
8637         ssize_t ret;
8638
8639         if (check_direct_IO(fs_info, iter, offset))
8640                 return 0;
8641
8642         inode_dio_begin(inode);
8643
8644         /*
8645          * The generic stuff only does filemap_write_and_wait_range, which
8646          * isn't enough if we've written compressed pages to this area, so
8647          * we need to flush the dirty pages again to make absolutely sure
8648          * that any outstanding dirty pages are on disk.
8649          */
8650         count = iov_iter_count(iter);
8651         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8652                      &BTRFS_I(inode)->runtime_flags))
8653                 filemap_fdatawrite_range(inode->i_mapping, offset,
8654                                          offset + count - 1);
8655
8656         if (iov_iter_rw(iter) == WRITE) {
8657                 /*
8658                  * If the write DIO is beyond the EOF, we need update
8659                  * the isize, but it is protected by i_mutex. So we can
8660                  * not unlock the i_mutex at this case.
8661                  */
8662                 if (offset + count <= inode->i_size) {
8663                         dio_data.overwrite = 1;
8664                         inode_unlock(inode);
8665                         relock = true;
8666                 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8667                         ret = -EAGAIN;
8668                         goto out;
8669                 }
8670                 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8671                                                    offset, count);
8672                 if (ret)
8673                         goto out;
8674
8675                 /*
8676                  * We need to know how many extents we reserved so that we can
8677                  * do the accounting properly if we go over the number we
8678                  * originally calculated.  Abuse current->journal_info for this.
8679                  */
8680                 dio_data.reserve = round_up(count,
8681                                             fs_info->sectorsize);
8682                 dio_data.unsubmitted_oe_range_start = (u64)offset;
8683                 dio_data.unsubmitted_oe_range_end = (u64)offset;
8684                 current->journal_info = &dio_data;
8685                 down_read(&BTRFS_I(inode)->dio_sem);
8686         } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8687                                      &BTRFS_I(inode)->runtime_flags)) {
8688                 inode_dio_end(inode);
8689                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8690                 wakeup = false;
8691         }
8692
8693         ret = __blockdev_direct_IO(iocb, inode,
8694                                    fs_info->fs_devices->latest_bdev,
8695                                    iter, btrfs_get_blocks_direct, NULL,
8696                                    btrfs_submit_direct, flags);
8697         if (iov_iter_rw(iter) == WRITE) {
8698                 up_read(&BTRFS_I(inode)->dio_sem);
8699                 current->journal_info = NULL;
8700                 if (ret < 0 && ret != -EIOCBQUEUED) {
8701                         if (dio_data.reserve)
8702                                 btrfs_delalloc_release_space(inode, data_reserved,
8703                                         offset, dio_data.reserve);
8704                         /*
8705                          * On error we might have left some ordered extents
8706                          * without submitting corresponding bios for them, so
8707                          * cleanup them up to avoid other tasks getting them
8708                          * and waiting for them to complete forever.
8709                          */
8710                         if (dio_data.unsubmitted_oe_range_start <
8711                             dio_data.unsubmitted_oe_range_end)
8712                                 __endio_write_update_ordered(inode,
8713                                         dio_data.unsubmitted_oe_range_start,
8714                                         dio_data.unsubmitted_oe_range_end -
8715                                         dio_data.unsubmitted_oe_range_start,
8716                                         false);
8717                 } else if (ret >= 0 && (size_t)ret < count)
8718                         btrfs_delalloc_release_space(inode, data_reserved,
8719                                         offset, count - (size_t)ret);
8720                 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
8721         }
8722 out:
8723         if (wakeup)
8724                 inode_dio_end(inode);
8725         if (relock)
8726                 inode_lock(inode);
8727
8728         extent_changeset_free(data_reserved);
8729         return ret;
8730 }
8731
8732 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
8733
8734 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8735                 __u64 start, __u64 len)
8736 {
8737         int     ret;
8738
8739         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8740         if (ret)
8741                 return ret;
8742
8743         return extent_fiemap(inode, fieinfo, start, len);
8744 }
8745
8746 int btrfs_readpage(struct file *file, struct page *page)
8747 {
8748         struct extent_io_tree *tree;
8749         tree = &BTRFS_I(page->mapping->host)->io_tree;
8750         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8751 }
8752
8753 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8754 {
8755         struct inode *inode = page->mapping->host;
8756         int ret;
8757
8758         if (current->flags & PF_MEMALLOC) {
8759                 redirty_page_for_writepage(wbc, page);
8760                 unlock_page(page);
8761                 return 0;
8762         }
8763
8764         /*
8765          * If we are under memory pressure we will call this directly from the
8766          * VM, we need to make sure we have the inode referenced for the ordered
8767          * extent.  If not just return like we didn't do anything.
8768          */
8769         if (!igrab(inode)) {
8770                 redirty_page_for_writepage(wbc, page);
8771                 return AOP_WRITEPAGE_ACTIVATE;
8772         }
8773         ret = extent_write_full_page(page, wbc);
8774         btrfs_add_delayed_iput(inode);
8775         return ret;
8776 }
8777
8778 static int btrfs_writepages(struct address_space *mapping,
8779                             struct writeback_control *wbc)
8780 {
8781         struct extent_io_tree *tree;
8782
8783         tree = &BTRFS_I(mapping->host)->io_tree;
8784         return extent_writepages(tree, mapping, wbc);
8785 }
8786
8787 static int
8788 btrfs_readpages(struct file *file, struct address_space *mapping,
8789                 struct list_head *pages, unsigned nr_pages)
8790 {
8791         struct extent_io_tree *tree;
8792         tree = &BTRFS_I(mapping->host)->io_tree;
8793         return extent_readpages(tree, mapping, pages, nr_pages);
8794 }
8795 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8796 {
8797         struct extent_io_tree *tree;
8798         struct extent_map_tree *map;
8799         int ret;
8800
8801         tree = &BTRFS_I(page->mapping->host)->io_tree;
8802         map = &BTRFS_I(page->mapping->host)->extent_tree;
8803         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8804         if (ret == 1) {
8805                 ClearPagePrivate(page);
8806                 set_page_private(page, 0);
8807                 put_page(page);
8808         }
8809         return ret;
8810 }
8811
8812 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8813 {
8814         if (PageWriteback(page) || PageDirty(page))
8815                 return 0;
8816         return __btrfs_releasepage(page, gfp_flags);
8817 }
8818
8819 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8820                                  unsigned int length)
8821 {
8822         struct inode *inode = page->mapping->host;
8823         struct extent_io_tree *tree;
8824         struct btrfs_ordered_extent *ordered;
8825         struct extent_state *cached_state = NULL;
8826         u64 page_start = page_offset(page);
8827         u64 page_end = page_start + PAGE_SIZE - 1;
8828         u64 start;
8829         u64 end;
8830         int inode_evicting = inode->i_state & I_FREEING;
8831
8832         /*
8833          * we have the page locked, so new writeback can't start,
8834          * and the dirty bit won't be cleared while we are here.
8835          *
8836          * Wait for IO on this page so that we can safely clear
8837          * the PagePrivate2 bit and do ordered accounting
8838          */
8839         wait_on_page_writeback(page);
8840
8841         tree = &BTRFS_I(inode)->io_tree;
8842         if (offset) {
8843                 btrfs_releasepage(page, GFP_NOFS);
8844                 return;
8845         }
8846
8847         if (!inode_evicting)
8848                 lock_extent_bits(tree, page_start, page_end, &cached_state);
8849 again:
8850         start = page_start;
8851         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8852                                         page_end - start + 1);
8853         if (ordered) {
8854                 end = min(page_end, ordered->file_offset + ordered->len - 1);
8855                 /*
8856                  * IO on this page will never be started, so we need
8857                  * to account for any ordered extents now
8858                  */
8859                 if (!inode_evicting)
8860                         clear_extent_bit(tree, start, end,
8861                                          EXTENT_DIRTY | EXTENT_DELALLOC |
8862                                          EXTENT_DELALLOC_NEW |
8863                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8864                                          EXTENT_DEFRAG, 1, 0, &cached_state);
8865                 /*
8866                  * whoever cleared the private bit is responsible
8867                  * for the finish_ordered_io
8868                  */
8869                 if (TestClearPagePrivate2(page)) {
8870                         struct btrfs_ordered_inode_tree *tree;
8871                         u64 new_len;
8872
8873                         tree = &BTRFS_I(inode)->ordered_tree;
8874
8875                         spin_lock_irq(&tree->lock);
8876                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8877                         new_len = start - ordered->file_offset;
8878                         if (new_len < ordered->truncated_len)
8879                                 ordered->truncated_len = new_len;
8880                         spin_unlock_irq(&tree->lock);
8881
8882                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
8883                                                            start,
8884                                                            end - start + 1, 1))
8885                                 btrfs_finish_ordered_io(ordered);
8886                 }
8887                 btrfs_put_ordered_extent(ordered);
8888                 if (!inode_evicting) {
8889                         cached_state = NULL;
8890                         lock_extent_bits(tree, start, end,
8891                                          &cached_state);
8892                 }
8893
8894                 start = end + 1;
8895                 if (start < page_end)
8896                         goto again;
8897         }
8898
8899         /*
8900          * Qgroup reserved space handler
8901          * Page here will be either
8902          * 1) Already written to disk
8903          *    In this case, its reserved space is released from data rsv map
8904          *    and will be freed by delayed_ref handler finally.
8905          *    So even we call qgroup_free_data(), it won't decrease reserved
8906          *    space.
8907          * 2) Not written to disk
8908          *    This means the reserved space should be freed here. However,
8909          *    if a truncate invalidates the page (by clearing PageDirty)
8910          *    and the page is accounted for while allocating extent
8911          *    in btrfs_check_data_free_space() we let delayed_ref to
8912          *    free the entire extent.
8913          */
8914         if (PageDirty(page))
8915                 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8916         if (!inode_evicting) {
8917                 clear_extent_bit(tree, page_start, page_end,
8918                                  EXTENT_LOCKED | EXTENT_DIRTY |
8919                                  EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8920                                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8921                                  &cached_state);
8922
8923                 __btrfs_releasepage(page, GFP_NOFS);
8924         }
8925
8926         ClearPageChecked(page);
8927         if (PagePrivate(page)) {
8928                 ClearPagePrivate(page);
8929                 set_page_private(page, 0);
8930                 put_page(page);
8931         }
8932 }
8933
8934 /*
8935  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8936  * called from a page fault handler when a page is first dirtied. Hence we must
8937  * be careful to check for EOF conditions here. We set the page up correctly
8938  * for a written page which means we get ENOSPC checking when writing into
8939  * holes and correct delalloc and unwritten extent mapping on filesystems that
8940  * support these features.
8941  *
8942  * We are not allowed to take the i_mutex here so we have to play games to
8943  * protect against truncate races as the page could now be beyond EOF.  Because
8944  * vmtruncate() writes the inode size before removing pages, once we have the
8945  * page lock we can determine safely if the page is beyond EOF. If it is not
8946  * beyond EOF, then the page is guaranteed safe against truncation until we
8947  * unlock the page.
8948  */
8949 int btrfs_page_mkwrite(struct vm_fault *vmf)
8950 {
8951         struct page *page = vmf->page;
8952         struct inode *inode = file_inode(vmf->vma->vm_file);
8953         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8954         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8955         struct btrfs_ordered_extent *ordered;
8956         struct extent_state *cached_state = NULL;
8957         struct extent_changeset *data_reserved = NULL;
8958         char *kaddr;
8959         unsigned long zero_start;
8960         loff_t size;
8961         int ret;
8962         int reserved = 0;
8963         u64 reserved_space;
8964         u64 page_start;
8965         u64 page_end;
8966         u64 end;
8967
8968         reserved_space = PAGE_SIZE;
8969
8970         sb_start_pagefault(inode->i_sb);
8971         page_start = page_offset(page);
8972         page_end = page_start + PAGE_SIZE - 1;
8973         end = page_end;
8974
8975         /*
8976          * Reserving delalloc space after obtaining the page lock can lead to
8977          * deadlock. For example, if a dirty page is locked by this function
8978          * and the call to btrfs_delalloc_reserve_space() ends up triggering
8979          * dirty page write out, then the btrfs_writepage() function could
8980          * end up waiting indefinitely to get a lock on the page currently
8981          * being processed by btrfs_page_mkwrite() function.
8982          */
8983         ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8984                                            reserved_space);
8985         if (!ret) {
8986                 ret = file_update_time(vmf->vma->vm_file);
8987                 reserved = 1;
8988         }
8989         if (ret) {
8990                 if (ret == -ENOMEM)
8991                         ret = VM_FAULT_OOM;
8992                 else /* -ENOSPC, -EIO, etc */
8993                         ret = VM_FAULT_SIGBUS;
8994                 if (reserved)
8995                         goto out;
8996                 goto out_noreserve;
8997         }
8998
8999         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9000 again:
9001         lock_page(page);
9002         size = i_size_read(inode);
9003
9004         if ((page->mapping != inode->i_mapping) ||
9005             (page_start >= size)) {
9006                 /* page got truncated out from underneath us */
9007                 goto out_unlock;
9008         }
9009         wait_on_page_writeback(page);
9010
9011         lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9012         set_page_extent_mapped(page);
9013
9014         /*
9015          * we can't set the delalloc bits if there are pending ordered
9016          * extents.  Drop our locks and wait for them to finish
9017          */
9018         ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9019                         PAGE_SIZE);
9020         if (ordered) {
9021                 unlock_extent_cached(io_tree, page_start, page_end,
9022                                      &cached_state);
9023                 unlock_page(page);
9024                 btrfs_start_ordered_extent(inode, ordered, 1);
9025                 btrfs_put_ordered_extent(ordered);
9026                 goto again;
9027         }
9028
9029         if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9030                 reserved_space = round_up(size - page_start,
9031                                           fs_info->sectorsize);
9032                 if (reserved_space < PAGE_SIZE) {
9033                         end = page_start + reserved_space - 1;
9034                         btrfs_delalloc_release_space(inode, data_reserved,
9035                                         page_start, PAGE_SIZE - reserved_space);
9036                 }
9037         }
9038
9039         /*
9040          * page_mkwrite gets called when the page is firstly dirtied after it's
9041          * faulted in, but write(2) could also dirty a page and set delalloc
9042          * bits, thus in this case for space account reason, we still need to
9043          * clear any delalloc bits within this page range since we have to
9044          * reserve data&meta space before lock_page() (see above comments).
9045          */
9046         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9047                           EXTENT_DIRTY | EXTENT_DELALLOC |
9048                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9049                           0, 0, &cached_state);
9050
9051         ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
9052                                         &cached_state, 0);
9053         if (ret) {
9054                 unlock_extent_cached(io_tree, page_start, page_end,
9055                                      &cached_state);
9056                 ret = VM_FAULT_SIGBUS;
9057                 goto out_unlock;
9058         }
9059         ret = 0;
9060
9061         /* page is wholly or partially inside EOF */
9062         if (page_start + PAGE_SIZE > size)
9063                 zero_start = size & ~PAGE_MASK;
9064         else
9065                 zero_start = PAGE_SIZE;
9066
9067         if (zero_start != PAGE_SIZE) {
9068                 kaddr = kmap(page);
9069                 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9070                 flush_dcache_page(page);
9071                 kunmap(page);
9072         }
9073         ClearPageChecked(page);
9074         set_page_dirty(page);
9075         SetPageUptodate(page);
9076
9077         BTRFS_I(inode)->last_trans = fs_info->generation;
9078         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9079         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9080
9081         unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
9082
9083 out_unlock:
9084         if (!ret) {
9085                 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9086                 sb_end_pagefault(inode->i_sb);
9087                 extent_changeset_free(data_reserved);
9088                 return VM_FAULT_LOCKED;
9089         }
9090         unlock_page(page);
9091 out:
9092         btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9093         btrfs_delalloc_release_space(inode, data_reserved, page_start,
9094                                      reserved_space);
9095 out_noreserve:
9096         sb_end_pagefault(inode->i_sb);
9097         extent_changeset_free(data_reserved);
9098         return ret;
9099 }
9100
9101 static int btrfs_truncate(struct inode *inode)
9102 {
9103         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9104         struct btrfs_root *root = BTRFS_I(inode)->root;
9105         struct btrfs_block_rsv *rsv;
9106         int ret = 0;
9107         int err = 0;
9108         struct btrfs_trans_handle *trans;
9109         u64 mask = fs_info->sectorsize - 1;
9110         u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9111
9112         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9113                                        (u64)-1);
9114         if (ret)
9115                 return ret;
9116
9117         /*
9118          * Yes ladies and gentlemen, this is indeed ugly.  The fact is we have
9119          * 3 things going on here
9120          *
9121          * 1) We need to reserve space for our orphan item and the space to
9122          * delete our orphan item.  Lord knows we don't want to have a dangling
9123          * orphan item because we didn't reserve space to remove it.
9124          *
9125          * 2) We need to reserve space to update our inode.
9126          *
9127          * 3) We need to have something to cache all the space that is going to
9128          * be free'd up by the truncate operation, but also have some slack
9129          * space reserved in case it uses space during the truncate (thank you
9130          * very much snapshotting).
9131          *
9132          * And we need these to all be separate.  The fact is we can use a lot of
9133          * space doing the truncate, and we have no earthly idea how much space
9134          * we will use, so we need the truncate reservation to be separate so it
9135          * doesn't end up using space reserved for updating the inode or
9136          * removing the orphan item.  We also need to be able to stop the
9137          * transaction and start a new one, which means we need to be able to
9138          * update the inode several times, and we have no idea of knowing how
9139          * many times that will be, so we can't just reserve 1 item for the
9140          * entirety of the operation, so that has to be done separately as well.
9141          * Then there is the orphan item, which does indeed need to be held on
9142          * to for the whole operation, and we need nobody to touch this reserved
9143          * space except the orphan code.
9144          *
9145          * So that leaves us with
9146          *
9147          * 1) root->orphan_block_rsv - for the orphan deletion.
9148          * 2) rsv - for the truncate reservation, which we will steal from the
9149          * transaction reservation.
9150          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9151          * updating the inode.
9152          */
9153         rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9154         if (!rsv)
9155                 return -ENOMEM;
9156         rsv->size = min_size;
9157         rsv->failfast = 1;
9158
9159         /*
9160          * 1 for the truncate slack space
9161          * 1 for updating the inode.
9162          */
9163         trans = btrfs_start_transaction(root, 2);
9164         if (IS_ERR(trans)) {
9165                 err = PTR_ERR(trans);
9166                 goto out;
9167         }
9168
9169         /* Migrate the slack space for the truncate to our reserve */
9170         ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9171                                       min_size, 0);
9172         BUG_ON(ret);
9173
9174         /*
9175          * So if we truncate and then write and fsync we normally would just
9176          * write the extents that changed, which is a problem if we need to
9177          * first truncate that entire inode.  So set this flag so we write out
9178          * all of the extents in the inode to the sync log so we're completely
9179          * safe.
9180          */
9181         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9182         trans->block_rsv = rsv;
9183
9184         while (1) {
9185                 ret = btrfs_truncate_inode_items(trans, root, inode,
9186                                                  inode->i_size,
9187                                                  BTRFS_EXTENT_DATA_KEY);
9188                 trans->block_rsv = &fs_info->trans_block_rsv;
9189                 if (ret != -ENOSPC && ret != -EAGAIN) {
9190                         err = ret;
9191                         break;
9192                 }
9193
9194                 ret = btrfs_update_inode(trans, root, inode);
9195                 if (ret) {
9196                         err = ret;
9197                         break;
9198                 }
9199
9200                 btrfs_end_transaction(trans);
9201                 btrfs_btree_balance_dirty(fs_info);
9202
9203                 trans = btrfs_start_transaction(root, 2);
9204                 if (IS_ERR(trans)) {
9205                         ret = err = PTR_ERR(trans);
9206                         trans = NULL;
9207                         break;
9208                 }
9209
9210                 btrfs_block_rsv_release(fs_info, rsv, -1);
9211                 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9212                                               rsv, min_size, 0);
9213                 BUG_ON(ret);    /* shouldn't happen */
9214                 trans->block_rsv = rsv;
9215         }
9216
9217         /*
9218          * We can't call btrfs_truncate_block inside a trans handle as we could
9219          * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9220          * we've truncated everything except the last little bit, and can do
9221          * btrfs_truncate_block and then update the disk_i_size.
9222          */
9223         if (ret == NEED_TRUNCATE_BLOCK) {
9224                 btrfs_end_transaction(trans);
9225                 btrfs_btree_balance_dirty(fs_info);
9226
9227                 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9228                 if (ret)
9229                         goto out;
9230                 trans = btrfs_start_transaction(root, 1);
9231                 if (IS_ERR(trans)) {
9232                         ret = PTR_ERR(trans);
9233                         goto out;
9234                 }
9235                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9236         }
9237
9238         if (ret == 0 && inode->i_nlink > 0) {
9239                 trans->block_rsv = root->orphan_block_rsv;
9240                 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9241                 if (ret)
9242                         err = ret;
9243         }
9244
9245         if (trans) {
9246                 trans->block_rsv = &fs_info->trans_block_rsv;
9247                 ret = btrfs_update_inode(trans, root, inode);
9248                 if (ret && !err)
9249                         err = ret;
9250
9251                 ret = btrfs_end_transaction(trans);
9252                 btrfs_btree_balance_dirty(fs_info);
9253         }
9254 out:
9255         btrfs_free_block_rsv(fs_info, rsv);
9256
9257         if (ret && !err)
9258                 err = ret;
9259
9260         return err;
9261 }
9262
9263 /*
9264  * create a new subvolume directory/inode (helper for the ioctl).
9265  */
9266 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9267                              struct btrfs_root *new_root,
9268                              struct btrfs_root *parent_root,
9269                              u64 new_dirid)
9270 {
9271         struct inode *inode;
9272         int err;
9273         u64 index = 0;
9274
9275         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9276                                 new_dirid, new_dirid,
9277                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
9278                                 &index);
9279         if (IS_ERR(inode))
9280                 return PTR_ERR(inode);
9281         inode->i_op = &btrfs_dir_inode_operations;
9282         inode->i_fop = &btrfs_dir_file_operations;
9283
9284         set_nlink(inode, 1);
9285         btrfs_i_size_write(BTRFS_I(inode), 0);
9286         unlock_new_inode(inode);
9287
9288         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9289         if (err)
9290                 btrfs_err(new_root->fs_info,
9291                           "error inheriting subvolume %llu properties: %d",
9292                           new_root->root_key.objectid, err);
9293
9294         err = btrfs_update_inode(trans, new_root, inode);
9295
9296         iput(inode);
9297         return err;
9298 }
9299
9300 struct inode *btrfs_alloc_inode(struct super_block *sb)
9301 {
9302         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9303         struct btrfs_inode *ei;
9304         struct inode *inode;
9305
9306         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9307         if (!ei)
9308                 return NULL;
9309
9310         ei->root = NULL;
9311         ei->generation = 0;
9312         ei->last_trans = 0;
9313         ei->last_sub_trans = 0;
9314         ei->logged_trans = 0;
9315         ei->delalloc_bytes = 0;
9316         ei->new_delalloc_bytes = 0;
9317         ei->defrag_bytes = 0;
9318         ei->disk_i_size = 0;
9319         ei->flags = 0;
9320         ei->csum_bytes = 0;
9321         ei->index_cnt = (u64)-1;
9322         ei->dir_index = 0;
9323         ei->last_unlink_trans = 0;
9324         ei->last_log_commit = 0;
9325         ei->delayed_iput_count = 0;
9326
9327         spin_lock_init(&ei->lock);
9328         ei->outstanding_extents = 0;
9329         if (sb->s_magic != BTRFS_TEST_MAGIC)
9330                 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9331                                               BTRFS_BLOCK_RSV_DELALLOC);
9332         ei->runtime_flags = 0;
9333         ei->prop_compress = BTRFS_COMPRESS_NONE;
9334         ei->defrag_compress = BTRFS_COMPRESS_NONE;
9335
9336         ei->delayed_node = NULL;
9337
9338         ei->i_otime.tv_sec = 0;
9339         ei->i_otime.tv_nsec = 0;
9340
9341         inode = &ei->vfs_inode;
9342         extent_map_tree_init(&ei->extent_tree);
9343         extent_io_tree_init(&ei->io_tree, inode);
9344         extent_io_tree_init(&ei->io_failure_tree, inode);
9345         ei->io_tree.track_uptodate = 1;
9346         ei->io_failure_tree.track_uptodate = 1;
9347         atomic_set(&ei->sync_writers, 0);
9348         mutex_init(&ei->log_mutex);
9349         mutex_init(&ei->delalloc_mutex);
9350         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9351         INIT_LIST_HEAD(&ei->delalloc_inodes);
9352         INIT_LIST_HEAD(&ei->delayed_iput);
9353         RB_CLEAR_NODE(&ei->rb_node);
9354         init_rwsem(&ei->dio_sem);
9355
9356         return inode;
9357 }
9358
9359 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9360 void btrfs_test_destroy_inode(struct inode *inode)
9361 {
9362         btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9363         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9364 }
9365 #endif
9366
9367 static void btrfs_i_callback(struct rcu_head *head)
9368 {
9369         struct inode *inode = container_of(head, struct inode, i_rcu);
9370         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9371 }
9372
9373 void btrfs_destroy_inode(struct inode *inode)
9374 {
9375         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9376         struct btrfs_ordered_extent *ordered;
9377         struct btrfs_root *root = BTRFS_I(inode)->root;
9378
9379         WARN_ON(!hlist_empty(&inode->i_dentry));
9380         WARN_ON(inode->i_data.nrpages);
9381         WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9382         WARN_ON(BTRFS_I(inode)->block_rsv.size);
9383         WARN_ON(BTRFS_I(inode)->outstanding_extents);
9384         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9385         WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9386         WARN_ON(BTRFS_I(inode)->csum_bytes);
9387         WARN_ON(BTRFS_I(inode)->defrag_bytes);
9388
9389         /*
9390          * This can happen where we create an inode, but somebody else also
9391          * created the same inode and we need to destroy the one we already
9392          * created.
9393          */
9394         if (!root)
9395                 goto free;
9396
9397         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9398                      &BTRFS_I(inode)->runtime_flags)) {
9399                 btrfs_info(fs_info, "inode %llu still on the orphan list",
9400                            btrfs_ino(BTRFS_I(inode)));
9401                 atomic_dec(&root->orphan_inodes);
9402         }
9403
9404         while (1) {
9405                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9406                 if (!ordered)
9407                         break;
9408                 else {
9409                         btrfs_err(fs_info,
9410                                   "found ordered extent %llu %llu on inode cleanup",
9411                                   ordered->file_offset, ordered->len);
9412                         btrfs_remove_ordered_extent(inode, ordered);
9413                         btrfs_put_ordered_extent(ordered);
9414                         btrfs_put_ordered_extent(ordered);
9415                 }
9416         }
9417         btrfs_qgroup_check_reserved_leak(inode);
9418         inode_tree_del(inode);
9419         btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9420 free:
9421         call_rcu(&inode->i_rcu, btrfs_i_callback);
9422 }
9423
9424 int btrfs_drop_inode(struct inode *inode)
9425 {
9426         struct btrfs_root *root = BTRFS_I(inode)->root;
9427
9428         if (root == NULL)
9429                 return 1;
9430
9431         /* the snap/subvol tree is on deleting */
9432         if (btrfs_root_refs(&root->root_item) == 0)
9433                 return 1;
9434         else
9435                 return generic_drop_inode(inode);
9436 }
9437
9438 static void init_once(void *foo)
9439 {
9440         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9441
9442         inode_init_once(&ei->vfs_inode);
9443 }
9444
9445 void btrfs_destroy_cachep(void)
9446 {
9447         /*
9448          * Make sure all delayed rcu free inodes are flushed before we
9449          * destroy cache.
9450          */
9451         rcu_barrier();
9452         kmem_cache_destroy(btrfs_inode_cachep);
9453         kmem_cache_destroy(btrfs_trans_handle_cachep);
9454         kmem_cache_destroy(btrfs_path_cachep);
9455         kmem_cache_destroy(btrfs_free_space_cachep);
9456 }
9457
9458 int __init btrfs_init_cachep(void)
9459 {
9460         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9461                         sizeof(struct btrfs_inode), 0,
9462                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9463                         init_once);
9464         if (!btrfs_inode_cachep)
9465                 goto fail;
9466
9467         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9468                         sizeof(struct btrfs_trans_handle), 0,
9469                         SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9470         if (!btrfs_trans_handle_cachep)
9471                 goto fail;
9472
9473         btrfs_path_cachep = kmem_cache_create("btrfs_path",
9474                         sizeof(struct btrfs_path), 0,
9475                         SLAB_MEM_SPREAD, NULL);
9476         if (!btrfs_path_cachep)
9477                 goto fail;
9478
9479         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9480                         sizeof(struct btrfs_free_space), 0,
9481                         SLAB_MEM_SPREAD, NULL);
9482         if (!btrfs_free_space_cachep)
9483                 goto fail;
9484
9485         return 0;
9486 fail:
9487         btrfs_destroy_cachep();
9488         return -ENOMEM;
9489 }
9490
9491 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9492                          u32 request_mask, unsigned int flags)
9493 {
9494         u64 delalloc_bytes;
9495         struct inode *inode = d_inode(path->dentry);
9496         u32 blocksize = inode->i_sb->s_blocksize;
9497         u32 bi_flags = BTRFS_I(inode)->flags;
9498
9499         stat->result_mask |= STATX_BTIME;
9500         stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9501         stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9502         if (bi_flags & BTRFS_INODE_APPEND)
9503                 stat->attributes |= STATX_ATTR_APPEND;
9504         if (bi_flags & BTRFS_INODE_COMPRESS)
9505                 stat->attributes |= STATX_ATTR_COMPRESSED;
9506         if (bi_flags & BTRFS_INODE_IMMUTABLE)
9507                 stat->attributes |= STATX_ATTR_IMMUTABLE;
9508         if (bi_flags & BTRFS_INODE_NODUMP)
9509                 stat->attributes |= STATX_ATTR_NODUMP;
9510
9511         stat->attributes_mask |= (STATX_ATTR_APPEND |
9512                                   STATX_ATTR_COMPRESSED |
9513                                   STATX_ATTR_IMMUTABLE |
9514                                   STATX_ATTR_NODUMP);
9515
9516         generic_fillattr(inode, stat);
9517         stat->dev = BTRFS_I(inode)->root->anon_dev;
9518
9519         spin_lock(&BTRFS_I(inode)->lock);
9520         delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9521         spin_unlock(&BTRFS_I(inode)->lock);
9522         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9523                         ALIGN(delalloc_bytes, blocksize)) >> 9;
9524         return 0;
9525 }
9526
9527 static int btrfs_rename_exchange(struct inode *old_dir,
9528                               struct dentry *old_dentry,
9529                               struct inode *new_dir,
9530                               struct dentry *new_dentry)
9531 {
9532         struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9533         struct btrfs_trans_handle *trans;
9534         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9535         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9536         struct inode *new_inode = new_dentry->d_inode;
9537         struct inode *old_inode = old_dentry->d_inode;
9538         struct timespec ctime = current_time(old_inode);
9539         struct dentry *parent;
9540         u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9541         u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9542         u64 old_idx = 0;
9543         u64 new_idx = 0;
9544         u64 root_objectid;
9545         int ret;
9546         bool root_log_pinned = false;
9547         bool dest_log_pinned = false;
9548
9549         /* we only allow rename subvolume link between subvolumes */
9550         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9551                 return -EXDEV;
9552
9553         /* close the race window with snapshot create/destroy ioctl */
9554         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9555                 down_read(&fs_info->subvol_sem);
9556         if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9557                 down_read(&fs_info->subvol_sem);
9558
9559         /*
9560          * We want to reserve the absolute worst case amount of items.  So if
9561          * both inodes are subvols and we need to unlink them then that would
9562          * require 4 item modifications, but if they are both normal inodes it
9563          * would require 5 item modifications, so we'll assume their normal
9564          * inodes.  So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9565          * should cover the worst case number of items we'll modify.
9566          */
9567         trans = btrfs_start_transaction(root, 12);
9568         if (IS_ERR(trans)) {
9569                 ret = PTR_ERR(trans);
9570                 goto out_notrans;
9571         }
9572
9573         /*
9574          * We need to find a free sequence number both in the source and
9575          * in the destination directory for the exchange.
9576          */
9577         ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9578         if (ret)
9579                 goto out_fail;
9580         ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9581         if (ret)
9582                 goto out_fail;
9583
9584         BTRFS_I(old_inode)->dir_index = 0ULL;
9585         BTRFS_I(new_inode)->dir_index = 0ULL;
9586
9587         /* Reference for the source. */
9588         if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9589                 /* force full log commit if subvolume involved. */
9590                 btrfs_set_log_full_commit(fs_info, trans);
9591         } else {
9592                 btrfs_pin_log_trans(root);
9593                 root_log_pinned = true;
9594                 ret = btrfs_insert_inode_ref(trans, dest,
9595                                              new_dentry->d_name.name,
9596                                              new_dentry->d_name.len,
9597                                              old_ino,
9598                                              btrfs_ino(BTRFS_I(new_dir)),
9599                                              old_idx);
9600                 if (ret)
9601                         goto out_fail;
9602         }
9603
9604         /* And now for the dest. */
9605         if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9606                 /* force full log commit if subvolume involved. */
9607                 btrfs_set_log_full_commit(fs_info, trans);
9608         } else {
9609                 btrfs_pin_log_trans(dest);
9610                 dest_log_pinned = true;
9611                 ret = btrfs_insert_inode_ref(trans, root,
9612                                              old_dentry->d_name.name,
9613                                              old_dentry->d_name.len,
9614                                              new_ino,
9615                                              btrfs_ino(BTRFS_I(old_dir)),
9616                                              new_idx);
9617                 if (ret)
9618                         goto out_fail;
9619         }
9620
9621         /* Update inode version and ctime/mtime. */
9622         inode_inc_iversion(old_dir);
9623         inode_inc_iversion(new_dir);
9624         inode_inc_iversion(old_inode);
9625         inode_inc_iversion(new_inode);
9626         old_dir->i_ctime = old_dir->i_mtime = ctime;
9627         new_dir->i_ctime = new_dir->i_mtime = ctime;
9628         old_inode->i_ctime = ctime;
9629         new_inode->i_ctime = ctime;
9630
9631         if (old_dentry->d_parent != new_dentry->d_parent) {
9632                 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9633                                 BTRFS_I(old_inode), 1);
9634                 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9635                                 BTRFS_I(new_inode), 1);
9636         }
9637
9638         /* src is a subvolume */
9639         if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9640                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9641                 ret = btrfs_unlink_subvol(trans, root, old_dir,
9642                                           root_objectid,
9643                                           old_dentry->d_name.name,
9644                                           old_dentry->d_name.len);
9645         } else { /* src is an inode */
9646                 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9647                                            BTRFS_I(old_dentry->d_inode),
9648                                            old_dentry->d_name.name,
9649                                            old_dentry->d_name.len);
9650                 if (!ret)
9651                         ret = btrfs_update_inode(trans, root, old_inode);
9652         }
9653         if (ret) {
9654                 btrfs_abort_transaction(trans, ret);
9655                 goto out_fail;
9656         }
9657
9658         /* dest is a subvolume */
9659         if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9660                 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9661                 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9662                                           root_objectid,
9663                                           new_dentry->d_name.name,
9664                                           new_dentry->d_name.len);
9665         } else { /* dest is an inode */
9666                 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9667                                            BTRFS_I(new_dentry->d_inode),
9668                                            new_dentry->d_name.name,
9669                                            new_dentry->d_name.len);
9670                 if (!ret)
9671                         ret = btrfs_update_inode(trans, dest, new_inode);
9672         }
9673         if (ret) {
9674                 btrfs_abort_transaction(trans, ret);
9675                 goto out_fail;
9676         }
9677
9678         ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9679                              new_dentry->d_name.name,
9680                              new_dentry->d_name.len, 0, old_idx);
9681         if (ret) {
9682                 btrfs_abort_transaction(trans, ret);
9683                 goto out_fail;
9684         }
9685
9686         ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9687                              old_dentry->d_name.name,
9688                              old_dentry->d_name.len, 0, new_idx);
9689         if (ret) {
9690                 btrfs_abort_transaction(trans, ret);
9691                 goto out_fail;
9692         }
9693
9694         if (old_inode->i_nlink == 1)
9695                 BTRFS_I(old_inode)->dir_index = old_idx;
9696         if (new_inode->i_nlink == 1)
9697                 BTRFS_I(new_inode)->dir_index = new_idx;
9698
9699         if (root_log_pinned) {
9700                 parent = new_dentry->d_parent;
9701                 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9702                                 parent);
9703                 btrfs_end_log_trans(root);
9704                 root_log_pinned = false;
9705         }
9706         if (dest_log_pinned) {
9707                 parent = old_dentry->d_parent;
9708                 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9709                                 parent);
9710                 btrfs_end_log_trans(dest);
9711                 dest_log_pinned = false;
9712         }
9713 out_fail:
9714         /*
9715          * If we have pinned a log and an error happened, we unpin tasks
9716          * trying to sync the log and force them to fallback to a transaction
9717          * commit if the log currently contains any of the inodes involved in
9718          * this rename operation (to ensure we do not persist a log with an
9719          * inconsistent state for any of these inodes or leading to any
9720          * inconsistencies when replayed). If the transaction was aborted, the
9721          * abortion reason is propagated to userspace when attempting to commit
9722          * the transaction. If the log does not contain any of these inodes, we
9723          * allow the tasks to sync it.
9724          */
9725         if (ret && (root_log_pinned || dest_log_pinned)) {
9726                 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9727                     btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9728                     btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9729                     (new_inode &&
9730                      btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9731                         btrfs_set_log_full_commit(fs_info, trans);
9732
9733                 if (root_log_pinned) {
9734                         btrfs_end_log_trans(root);
9735                         root_log_pinned = false;
9736                 }
9737                 if (dest_log_pinned) {
9738                         btrfs_end_log_trans(dest);
9739                         dest_log_pinned = false;
9740                 }
9741         }
9742         ret = btrfs_end_transaction(trans);
9743 out_notrans:
9744         if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9745                 up_read(&fs_info->subvol_sem);
9746         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9747                 up_read(&fs_info->subvol_sem);
9748
9749         return ret;
9750 }
9751
9752 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9753                                      struct btrfs_root *root,
9754                                      struct inode *dir,
9755                                      struct dentry *dentry)
9756 {
9757         int ret;
9758         struct inode *inode;
9759         u64 objectid;
9760         u64 index;
9761
9762         ret = btrfs_find_free_ino(root, &objectid);
9763         if (ret)
9764                 return ret;
9765
9766         inode = btrfs_new_inode(trans, root, dir,
9767                                 dentry->d_name.name,
9768                                 dentry->d_name.len,
9769                                 btrfs_ino(BTRFS_I(dir)),
9770                                 objectid,
9771                                 S_IFCHR | WHITEOUT_MODE,
9772                                 &index);
9773
9774         if (IS_ERR(inode)) {
9775                 ret = PTR_ERR(inode);
9776                 return ret;
9777         }
9778
9779         inode->i_op = &btrfs_special_inode_operations;
9780         init_special_inode(inode, inode->i_mode,
9781                 WHITEOUT_DEV);
9782
9783         ret = btrfs_init_inode_security(trans, inode, dir,
9784                                 &dentry->d_name);
9785         if (ret)
9786                 goto out;
9787
9788         ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9789                                 BTRFS_I(inode), 0, index);
9790         if (ret)
9791                 goto out;
9792
9793         ret = btrfs_update_inode(trans, root, inode);
9794 out:
9795         unlock_new_inode(inode);
9796         if (ret)
9797                 inode_dec_link_count(inode);
9798         iput(inode);
9799
9800         return ret;
9801 }
9802
9803 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9804                            struct inode *new_dir, struct dentry *new_dentry,
9805                            unsigned int flags)
9806 {
9807         struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9808         struct btrfs_trans_handle *trans;
9809         unsigned int trans_num_items;
9810         struct btrfs_root *root = BTRFS_I(old_dir)->root;
9811         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9812         struct inode *new_inode = d_inode(new_dentry);
9813         struct inode *old_inode = d_inode(old_dentry);
9814         u64 index = 0;
9815         u64 root_objectid;
9816         int ret;
9817         u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9818         bool log_pinned = false;
9819
9820         if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9821                 return -EPERM;
9822
9823         /* we only allow rename subvolume link between subvolumes */
9824         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9825                 return -EXDEV;
9826
9827         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9828             (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9829                 return -ENOTEMPTY;
9830
9831         if (S_ISDIR(old_inode->i_mode) && new_inode &&
9832             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9833                 return -ENOTEMPTY;
9834
9835
9836         /* check for collisions, even if the  name isn't there */
9837         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9838                              new_dentry->d_name.name,
9839                              new_dentry->d_name.len);
9840
9841         if (ret) {
9842                 if (ret == -EEXIST) {
9843                         /* we shouldn't get
9844                          * eexist without a new_inode */
9845                         if (WARN_ON(!new_inode)) {
9846                                 return ret;
9847                         }
9848                 } else {
9849                         /* maybe -EOVERFLOW */
9850                         return ret;
9851                 }
9852         }
9853         ret = 0;
9854
9855         /*
9856          * we're using rename to replace one file with another.  Start IO on it
9857          * now so  we don't add too much work to the end of the transaction
9858          */
9859         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9860                 filemap_flush(old_inode->i_mapping);
9861
9862         /* close the racy window with snapshot create/destroy ioctl */
9863         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9864                 down_read(&fs_info->subvol_sem);
9865         /*
9866          * We want to reserve the absolute worst case amount of items.  So if
9867          * both inodes are subvols and we need to unlink them then that would
9868          * require 4 item modifications, but if they are both normal inodes it
9869          * would require 5 item modifications, so we'll assume they are normal
9870          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9871          * should cover the worst case number of items we'll modify.
9872          * If our rename has the whiteout flag, we need more 5 units for the
9873          * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9874          * when selinux is enabled).
9875          */
9876         trans_num_items = 11;
9877         if (flags & RENAME_WHITEOUT)
9878                 trans_num_items += 5;
9879         trans = btrfs_start_transaction(root, trans_num_items);
9880         if (IS_ERR(trans)) {
9881                 ret = PTR_ERR(trans);
9882                 goto out_notrans;
9883         }
9884
9885         if (dest != root)
9886                 btrfs_record_root_in_trans(trans, dest);
9887
9888         ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9889         if (ret)
9890                 goto out_fail;
9891
9892         BTRFS_I(old_inode)->dir_index = 0ULL;
9893         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9894                 /* force full log commit if subvolume involved. */
9895                 btrfs_set_log_full_commit(fs_info, trans);
9896         } else {
9897                 btrfs_pin_log_trans(root);
9898                 log_pinned = true;
9899                 ret = btrfs_insert_inode_ref(trans, dest,
9900                                              new_dentry->d_name.name,
9901                                              new_dentry->d_name.len,
9902                                              old_ino,
9903                                              btrfs_ino(BTRFS_I(new_dir)), index);
9904                 if (ret)
9905                         goto out_fail;
9906         }
9907
9908         inode_inc_iversion(old_dir);
9909         inode_inc_iversion(new_dir);
9910         inode_inc_iversion(old_inode);
9911         old_dir->i_ctime = old_dir->i_mtime =
9912         new_dir->i_ctime = new_dir->i_mtime =
9913         old_inode->i_ctime = current_time(old_dir);
9914
9915         if (old_dentry->d_parent != new_dentry->d_parent)
9916                 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9917                                 BTRFS_I(old_inode), 1);
9918
9919         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9920                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9921                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9922                                         old_dentry->d_name.name,
9923                                         old_dentry->d_name.len);
9924         } else {
9925                 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9926                                         BTRFS_I(d_inode(old_dentry)),
9927                                         old_dentry->d_name.name,
9928                                         old_dentry->d_name.len);
9929                 if (!ret)
9930                         ret = btrfs_update_inode(trans, root, old_inode);
9931         }
9932         if (ret) {
9933                 btrfs_abort_transaction(trans, ret);
9934                 goto out_fail;
9935         }
9936
9937         if (new_inode) {
9938                 inode_inc_iversion(new_inode);
9939                 new_inode->i_ctime = current_time(new_inode);
9940                 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9941                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9942                         root_objectid = BTRFS_I(new_inode)->location.objectid;
9943                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
9944                                                 root_objectid,
9945                                                 new_dentry->d_name.name,
9946                                                 new_dentry->d_name.len);
9947                         BUG_ON(new_inode->i_nlink == 0);
9948                 } else {
9949                         ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9950                                                  BTRFS_I(d_inode(new_dentry)),
9951                                                  new_dentry->d_name.name,
9952                                                  new_dentry->d_name.len);
9953                 }
9954                 if (!ret && new_inode->i_nlink == 0)
9955                         ret = btrfs_orphan_add(trans,
9956                                         BTRFS_I(d_inode(new_dentry)));
9957                 if (ret) {
9958                         btrfs_abort_transaction(trans, ret);
9959                         goto out_fail;
9960                 }
9961         }
9962
9963         ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9964                              new_dentry->d_name.name,
9965                              new_dentry->d_name.len, 0, index);
9966         if (ret) {
9967                 btrfs_abort_transaction(trans, ret);
9968                 goto out_fail;
9969         }
9970
9971         if (old_inode->i_nlink == 1)
9972                 BTRFS_I(old_inode)->dir_index = index;
9973
9974         if (log_pinned) {
9975                 struct dentry *parent = new_dentry->d_parent;
9976
9977                 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9978                                 parent);
9979                 btrfs_end_log_trans(root);
9980                 log_pinned = false;
9981         }
9982
9983         if (flags & RENAME_WHITEOUT) {
9984                 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9985                                                 old_dentry);
9986
9987                 if (ret) {
9988                         btrfs_abort_transaction(trans, ret);
9989                         goto out_fail;
9990                 }
9991         }
9992 out_fail:
9993         /*
9994          * If we have pinned the log and an error happened, we unpin tasks
9995          * trying to sync the log and force them to fallback to a transaction
9996          * commit if the log currently contains any of the inodes involved in
9997          * this rename operation (to ensure we do not persist a log with an
9998          * inconsistent state for any of these inodes or leading to any
9999          * inconsistencies when replayed). If the transaction was aborted, the
10000          * abortion reason is propagated to userspace when attempting to commit
10001          * the transaction. If the log does not contain any of these inodes, we
10002          * allow the tasks to sync it.
10003          */
10004         if (ret && log_pinned) {
10005                 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10006                     btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10007                     btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10008                     (new_inode &&
10009                      btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10010                         btrfs_set_log_full_commit(fs_info, trans);
10011
10012                 btrfs_end_log_trans(root);
10013                 log_pinned = false;
10014         }
10015         btrfs_end_transaction(trans);
10016 out_notrans:
10017         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10018                 up_read(&fs_info->subvol_sem);
10019
10020         return ret;
10021 }
10022
10023 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10024                          struct inode *new_dir, struct dentry *new_dentry,
10025                          unsigned int flags)
10026 {
10027         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10028                 return -EINVAL;
10029
10030         if (flags & RENAME_EXCHANGE)
10031                 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10032                                           new_dentry);
10033
10034         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10035 }
10036
10037 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10038 {
10039         struct btrfs_delalloc_work *delalloc_work;
10040         struct inode *inode;
10041
10042         delalloc_work = container_of(work, struct btrfs_delalloc_work,
10043                                      work);
10044         inode = delalloc_work->inode;
10045         filemap_flush(inode->i_mapping);
10046         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10047                                 &BTRFS_I(inode)->runtime_flags))
10048                 filemap_flush(inode->i_mapping);
10049
10050         if (delalloc_work->delay_iput)
10051                 btrfs_add_delayed_iput(inode);
10052         else
10053                 iput(inode);
10054         complete(&delalloc_work->completion);
10055 }
10056
10057 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10058                                                     int delay_iput)
10059 {
10060         struct btrfs_delalloc_work *work;
10061
10062         work = kmalloc(sizeof(*work), GFP_NOFS);
10063         if (!work)
10064                 return NULL;
10065
10066         init_completion(&work->completion);
10067         INIT_LIST_HEAD(&work->list);
10068         work->inode = inode;
10069         work->delay_iput = delay_iput;
10070         WARN_ON_ONCE(!inode);
10071         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10072                         btrfs_run_delalloc_work, NULL, NULL);
10073
10074         return work;
10075 }
10076
10077 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10078 {
10079         wait_for_completion(&work->completion);
10080         kfree(work);
10081 }
10082
10083 /*
10084  * some fairly slow code that needs optimization. This walks the list
10085  * of all the inodes with pending delalloc and forces them to disk.
10086  */
10087 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10088                                    int nr)
10089 {
10090         struct btrfs_inode *binode;
10091         struct inode *inode;
10092         struct btrfs_delalloc_work *work, *next;
10093         struct list_head works;
10094         struct list_head splice;
10095         int ret = 0;
10096
10097         INIT_LIST_HEAD(&works);
10098         INIT_LIST_HEAD(&splice);
10099
10100         mutex_lock(&root->delalloc_mutex);
10101         spin_lock(&root->delalloc_lock);
10102         list_splice_init(&root->delalloc_inodes, &splice);
10103         while (!list_empty(&splice)) {
10104                 binode = list_entry(splice.next, struct btrfs_inode,
10105                                     delalloc_inodes);
10106
10107                 list_move_tail(&binode->delalloc_inodes,
10108                                &root->delalloc_inodes);
10109                 inode = igrab(&binode->vfs_inode);
10110                 if (!inode) {
10111                         cond_resched_lock(&root->delalloc_lock);
10112                         continue;
10113                 }
10114                 spin_unlock(&root->delalloc_lock);
10115
10116                 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10117                 if (!work) {
10118                         if (delay_iput)
10119                                 btrfs_add_delayed_iput(inode);
10120                         else
10121                                 iput(inode);
10122                         ret = -ENOMEM;
10123                         goto out;
10124                 }
10125                 list_add_tail(&work->list, &works);
10126                 btrfs_queue_work(root->fs_info->flush_workers,
10127                                  &work->work);
10128                 ret++;
10129                 if (nr != -1 && ret >= nr)
10130                         goto out;
10131                 cond_resched();
10132                 spin_lock(&root->delalloc_lock);
10133         }
10134         spin_unlock(&root->delalloc_lock);
10135
10136 out:
10137         list_for_each_entry_safe(work, next, &works, list) {
10138                 list_del_init(&work->list);
10139                 btrfs_wait_and_free_delalloc_work(work);
10140         }
10141
10142         if (!list_empty_careful(&splice)) {
10143                 spin_lock(&root->delalloc_lock);
10144                 list_splice_tail(&splice, &root->delalloc_inodes);
10145                 spin_unlock(&root->delalloc_lock);
10146         }
10147         mutex_unlock(&root->delalloc_mutex);
10148         return ret;
10149 }
10150
10151 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10152 {
10153         struct btrfs_fs_info *fs_info = root->fs_info;
10154         int ret;
10155
10156         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10157                 return -EROFS;
10158
10159         ret = __start_delalloc_inodes(root, delay_iput, -1);
10160         if (ret > 0)
10161                 ret = 0;
10162         return ret;
10163 }
10164
10165 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10166                                int nr)
10167 {
10168         struct btrfs_root *root;
10169         struct list_head splice;
10170         int ret;
10171
10172         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10173                 return -EROFS;
10174
10175         INIT_LIST_HEAD(&splice);
10176
10177         mutex_lock(&fs_info->delalloc_root_mutex);
10178         spin_lock(&fs_info->delalloc_root_lock);
10179         list_splice_init(&fs_info->delalloc_roots, &splice);
10180         while (!list_empty(&splice) && nr) {
10181                 root = list_first_entry(&splice, struct btrfs_root,
10182                                         delalloc_root);
10183                 root = btrfs_grab_fs_root(root);
10184                 BUG_ON(!root);
10185                 list_move_tail(&root->delalloc_root,
10186                                &fs_info->delalloc_roots);
10187                 spin_unlock(&fs_info->delalloc_root_lock);
10188
10189                 ret = __start_delalloc_inodes(root, delay_iput, nr);
10190                 btrfs_put_fs_root(root);
10191                 if (ret < 0)
10192                         goto out;
10193
10194                 if (nr != -1) {
10195                         nr -= ret;
10196                         WARN_ON(nr < 0);
10197                 }
10198                 spin_lock(&fs_info->delalloc_root_lock);
10199         }
10200         spin_unlock(&fs_info->delalloc_root_lock);
10201
10202         ret = 0;
10203 out:
10204         if (!list_empty_careful(&splice)) {
10205                 spin_lock(&fs_info->delalloc_root_lock);
10206                 list_splice_tail(&splice, &fs_info->delalloc_roots);
10207                 spin_unlock(&fs_info->delalloc_root_lock);
10208         }
10209         mutex_unlock(&fs_info->delalloc_root_mutex);
10210         return ret;
10211 }
10212
10213 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10214                          const char *symname)
10215 {
10216         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10217         struct btrfs_trans_handle *trans;
10218         struct btrfs_root *root = BTRFS_I(dir)->root;
10219         struct btrfs_path *path;
10220         struct btrfs_key key;
10221         struct inode *inode = NULL;
10222         int err;
10223         int drop_inode = 0;
10224         u64 objectid;
10225         u64 index = 0;
10226         int name_len;
10227         int datasize;
10228         unsigned long ptr;
10229         struct btrfs_file_extent_item *ei;
10230         struct extent_buffer *leaf;
10231
10232         name_len = strlen(symname);
10233         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10234                 return -ENAMETOOLONG;
10235
10236         /*
10237          * 2 items for inode item and ref
10238          * 2 items for dir items
10239          * 1 item for updating parent inode item
10240          * 1 item for the inline extent item
10241          * 1 item for xattr if selinux is on
10242          */
10243         trans = btrfs_start_transaction(root, 7);
10244         if (IS_ERR(trans))
10245                 return PTR_ERR(trans);
10246
10247         err = btrfs_find_free_ino(root, &objectid);
10248         if (err)
10249                 goto out_unlock;
10250
10251         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10252                                 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10253                                 objectid, S_IFLNK|S_IRWXUGO, &index);
10254         if (IS_ERR(inode)) {
10255                 err = PTR_ERR(inode);
10256                 goto out_unlock;
10257         }
10258
10259         /*
10260         * If the active LSM wants to access the inode during
10261         * d_instantiate it needs these. Smack checks to see
10262         * if the filesystem supports xattrs by looking at the
10263         * ops vector.
10264         */
10265         inode->i_fop = &btrfs_file_operations;
10266         inode->i_op = &btrfs_file_inode_operations;
10267         inode->i_mapping->a_ops = &btrfs_aops;
10268         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10269
10270         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10271         if (err)
10272                 goto out_unlock_inode;
10273
10274         path = btrfs_alloc_path();
10275         if (!path) {
10276                 err = -ENOMEM;
10277                 goto out_unlock_inode;
10278         }
10279         key.objectid = btrfs_ino(BTRFS_I(inode));
10280         key.offset = 0;
10281         key.type = BTRFS_EXTENT_DATA_KEY;
10282         datasize = btrfs_file_extent_calc_inline_size(name_len);
10283         err = btrfs_insert_empty_item(trans, root, path, &key,
10284                                       datasize);
10285         if (err) {
10286                 btrfs_free_path(path);
10287                 goto out_unlock_inode;
10288         }
10289         leaf = path->nodes[0];
10290         ei = btrfs_item_ptr(leaf, path->slots[0],
10291                             struct btrfs_file_extent_item);
10292         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10293         btrfs_set_file_extent_type(leaf, ei,
10294                                    BTRFS_FILE_EXTENT_INLINE);
10295         btrfs_set_file_extent_encryption(leaf, ei, 0);
10296         btrfs_set_file_extent_compression(leaf, ei, 0);
10297         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10298         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10299
10300         ptr = btrfs_file_extent_inline_start(ei);
10301         write_extent_buffer(leaf, symname, ptr, name_len);
10302         btrfs_mark_buffer_dirty(leaf);
10303         btrfs_free_path(path);
10304
10305         inode->i_op = &btrfs_symlink_inode_operations;
10306         inode_nohighmem(inode);
10307         inode->i_mapping->a_ops = &btrfs_symlink_aops;
10308         inode_set_bytes(inode, name_len);
10309         btrfs_i_size_write(BTRFS_I(inode), name_len);
10310         err = btrfs_update_inode(trans, root, inode);
10311         /*
10312          * Last step, add directory indexes for our symlink inode. This is the
10313          * last step to avoid extra cleanup of these indexes if an error happens
10314          * elsewhere above.
10315          */
10316         if (!err)
10317                 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10318                                 BTRFS_I(inode), 0, index);
10319         if (err) {
10320                 drop_inode = 1;
10321                 goto out_unlock_inode;
10322         }
10323
10324         unlock_new_inode(inode);
10325         d_instantiate(dentry, inode);
10326
10327 out_unlock:
10328         btrfs_end_transaction(trans);
10329         if (drop_inode) {
10330                 inode_dec_link_count(inode);
10331                 iput(inode);
10332         }
10333         btrfs_btree_balance_dirty(fs_info);
10334         return err;
10335
10336 out_unlock_inode:
10337         drop_inode = 1;
10338         unlock_new_inode(inode);
10339         goto out_unlock;
10340 }
10341
10342 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10343                                        u64 start, u64 num_bytes, u64 min_size,
10344                                        loff_t actual_len, u64 *alloc_hint,
10345                                        struct btrfs_trans_handle *trans)
10346 {
10347         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10348         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10349         struct extent_map *em;
10350         struct btrfs_root *root = BTRFS_I(inode)->root;
10351         struct btrfs_key ins;
10352         u64 cur_offset = start;
10353         u64 i_size;
10354         u64 cur_bytes;
10355         u64 last_alloc = (u64)-1;
10356         int ret = 0;
10357         bool own_trans = true;
10358         u64 end = start + num_bytes - 1;
10359
10360         if (trans)
10361                 own_trans = false;
10362         while (num_bytes > 0) {
10363                 if (own_trans) {
10364                         trans = btrfs_start_transaction(root, 3);
10365                         if (IS_ERR(trans)) {
10366                                 ret = PTR_ERR(trans);
10367                                 break;
10368                         }
10369                 }
10370
10371                 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10372                 cur_bytes = max(cur_bytes, min_size);
10373                 /*
10374                  * If we are severely fragmented we could end up with really
10375                  * small allocations, so if the allocator is returning small
10376                  * chunks lets make its job easier by only searching for those
10377                  * sized chunks.
10378                  */
10379                 cur_bytes = min(cur_bytes, last_alloc);
10380                 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10381                                 min_size, 0, *alloc_hint, &ins, 1, 0);
10382                 if (ret) {
10383                         if (own_trans)
10384                                 btrfs_end_transaction(trans);
10385                         break;
10386                 }
10387                 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10388
10389                 last_alloc = ins.offset;
10390                 ret = insert_reserved_file_extent(trans, inode,
10391                                                   cur_offset, ins.objectid,
10392                                                   ins.offset, ins.offset,
10393                                                   ins.offset, 0, 0, 0,
10394                                                   BTRFS_FILE_EXTENT_PREALLOC);
10395                 if (ret) {
10396                         btrfs_free_reserved_extent(fs_info, ins.objectid,
10397                                                    ins.offset, 0);
10398                         btrfs_abort_transaction(trans, ret);
10399                         if (own_trans)
10400                                 btrfs_end_transaction(trans);
10401                         break;
10402                 }
10403
10404                 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10405                                         cur_offset + ins.offset -1, 0);
10406
10407                 em = alloc_extent_map();
10408                 if (!em) {
10409                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10410                                 &BTRFS_I(inode)->runtime_flags);
10411                         goto next;
10412                 }
10413
10414                 em->start = cur_offset;
10415                 em->orig_start = cur_offset;
10416                 em->len = ins.offset;
10417                 em->block_start = ins.objectid;
10418                 em->block_len = ins.offset;
10419                 em->orig_block_len = ins.offset;
10420                 em->ram_bytes = ins.offset;
10421                 em->bdev = fs_info->fs_devices->latest_bdev;
10422                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10423                 em->generation = trans->transid;
10424
10425                 while (1) {
10426                         write_lock(&em_tree->lock);
10427                         ret = add_extent_mapping(em_tree, em, 1);
10428                         write_unlock(&em_tree->lock);
10429                         if (ret != -EEXIST)
10430                                 break;
10431                         btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10432                                                 cur_offset + ins.offset - 1,
10433                                                 0);
10434                 }
10435                 free_extent_map(em);
10436 next:
10437                 num_bytes -= ins.offset;
10438                 cur_offset += ins.offset;
10439                 *alloc_hint = ins.objectid + ins.offset;
10440
10441                 inode_inc_iversion(inode);
10442                 inode->i_ctime = current_time(inode);
10443                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10444                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10445                     (actual_len > inode->i_size) &&
10446                     (cur_offset > inode->i_size)) {
10447                         if (cur_offset > actual_len)
10448                                 i_size = actual_len;
10449                         else
10450                                 i_size = cur_offset;
10451                         i_size_write(inode, i_size);
10452                         btrfs_ordered_update_i_size(inode, i_size, NULL);
10453                 }
10454
10455                 ret = btrfs_update_inode(trans, root, inode);
10456
10457                 if (ret) {
10458                         btrfs_abort_transaction(trans, ret);
10459                         if (own_trans)
10460                                 btrfs_end_transaction(trans);
10461                         break;
10462                 }
10463
10464                 if (own_trans)
10465                         btrfs_end_transaction(trans);
10466         }
10467         if (cur_offset < end)
10468                 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10469                         end - cur_offset + 1);
10470         return ret;
10471 }
10472
10473 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10474                               u64 start, u64 num_bytes, u64 min_size,
10475                               loff_t actual_len, u64 *alloc_hint)
10476 {
10477         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10478                                            min_size, actual_len, alloc_hint,
10479                                            NULL);
10480 }
10481
10482 int btrfs_prealloc_file_range_trans(struct inode *inode,
10483                                     struct btrfs_trans_handle *trans, int mode,
10484                                     u64 start, u64 num_bytes, u64 min_size,
10485                                     loff_t actual_len, u64 *alloc_hint)
10486 {
10487         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10488                                            min_size, actual_len, alloc_hint, trans);
10489 }
10490
10491 static int btrfs_set_page_dirty(struct page *page)
10492 {
10493         return __set_page_dirty_nobuffers(page);
10494 }
10495
10496 static int btrfs_permission(struct inode *inode, int mask)
10497 {
10498         struct btrfs_root *root = BTRFS_I(inode)->root;
10499         umode_t mode = inode->i_mode;
10500
10501         if (mask & MAY_WRITE &&
10502             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10503                 if (btrfs_root_readonly(root))
10504                         return -EROFS;
10505                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10506                         return -EACCES;
10507         }
10508         return generic_permission(inode, mask);
10509 }
10510
10511 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10512 {
10513         struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10514         struct btrfs_trans_handle *trans;
10515         struct btrfs_root *root = BTRFS_I(dir)->root;
10516         struct inode *inode = NULL;
10517         u64 objectid;
10518         u64 index;
10519         int ret = 0;
10520
10521         /*
10522          * 5 units required for adding orphan entry
10523          */
10524         trans = btrfs_start_transaction(root, 5);
10525         if (IS_ERR(trans))
10526                 return PTR_ERR(trans);
10527
10528         ret = btrfs_find_free_ino(root, &objectid);
10529         if (ret)
10530                 goto out;
10531
10532         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10533                         btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10534         if (IS_ERR(inode)) {
10535                 ret = PTR_ERR(inode);
10536                 inode = NULL;
10537                 goto out;
10538         }
10539
10540         inode->i_fop = &btrfs_file_operations;
10541         inode->i_op = &btrfs_file_inode_operations;
10542
10543         inode->i_mapping->a_ops = &btrfs_aops;
10544         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10545
10546         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10547         if (ret)
10548                 goto out_inode;
10549
10550         ret = btrfs_update_inode(trans, root, inode);
10551         if (ret)
10552                 goto out_inode;
10553         ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10554         if (ret)
10555                 goto out_inode;
10556
10557         /*
10558          * We set number of links to 0 in btrfs_new_inode(), and here we set
10559          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10560          * through:
10561          *
10562          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10563          */
10564         set_nlink(inode, 1);
10565         unlock_new_inode(inode);
10566         d_tmpfile(dentry, inode);
10567         mark_inode_dirty(inode);
10568
10569 out:
10570         btrfs_end_transaction(trans);
10571         if (ret)
10572                 iput(inode);
10573         btrfs_btree_balance_dirty(fs_info);
10574         return ret;
10575
10576 out_inode:
10577         unlock_new_inode(inode);
10578         goto out;
10579
10580 }
10581
10582 __attribute__((const))
10583 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10584 {
10585         return -EAGAIN;
10586 }
10587
10588 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10589 {
10590         struct inode *inode = private_data;
10591         return btrfs_sb(inode->i_sb);
10592 }
10593
10594 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10595                                         u64 start, u64 end)
10596 {
10597         struct inode *inode = private_data;
10598         u64 isize;
10599
10600         isize = i_size_read(inode);
10601         if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10602                 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10603                     "%s: ino %llu isize %llu odd range [%llu,%llu]",
10604                         caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10605         }
10606 }
10607
10608 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10609 {
10610         struct inode *inode = private_data;
10611         unsigned long index = start >> PAGE_SHIFT;
10612         unsigned long end_index = end >> PAGE_SHIFT;
10613         struct page *page;
10614
10615         while (index <= end_index) {
10616                 page = find_get_page(inode->i_mapping, index);
10617                 ASSERT(page); /* Pages should be in the extent_io_tree */
10618                 set_page_writeback(page);
10619                 put_page(page);
10620                 index++;
10621         }
10622 }
10623
10624 static const struct inode_operations btrfs_dir_inode_operations = {
10625         .getattr        = btrfs_getattr,
10626         .lookup         = btrfs_lookup,
10627         .create         = btrfs_create,
10628         .unlink         = btrfs_unlink,
10629         .link           = btrfs_link,
10630         .mkdir          = btrfs_mkdir,
10631         .rmdir          = btrfs_rmdir,
10632         .rename         = btrfs_rename2,
10633         .symlink        = btrfs_symlink,
10634         .setattr        = btrfs_setattr,
10635         .mknod          = btrfs_mknod,
10636         .listxattr      = btrfs_listxattr,
10637         .permission     = btrfs_permission,
10638         .get_acl        = btrfs_get_acl,
10639         .set_acl        = btrfs_set_acl,
10640         .update_time    = btrfs_update_time,
10641         .tmpfile        = btrfs_tmpfile,
10642 };
10643 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10644         .lookup         = btrfs_lookup,
10645         .permission     = btrfs_permission,
10646         .update_time    = btrfs_update_time,
10647 };
10648
10649 static const struct file_operations btrfs_dir_file_operations = {
10650         .llseek         = generic_file_llseek,
10651         .read           = generic_read_dir,
10652         .iterate_shared = btrfs_real_readdir,
10653         .open           = btrfs_opendir,
10654         .unlocked_ioctl = btrfs_ioctl,
10655 #ifdef CONFIG_COMPAT
10656         .compat_ioctl   = btrfs_compat_ioctl,
10657 #endif
10658         .release        = btrfs_release_file,
10659         .fsync          = btrfs_sync_file,
10660 };
10661
10662 static const struct extent_io_ops btrfs_extent_io_ops = {
10663         /* mandatory callbacks */
10664         .submit_bio_hook = btrfs_submit_bio_hook,
10665         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10666         .merge_bio_hook = btrfs_merge_bio_hook,
10667         .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10668         .tree_fs_info = iotree_fs_info,
10669         .set_range_writeback = btrfs_set_range_writeback,
10670
10671         /* optional callbacks */
10672         .fill_delalloc = run_delalloc_range,
10673         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10674         .writepage_start_hook = btrfs_writepage_start_hook,
10675         .set_bit_hook = btrfs_set_bit_hook,
10676         .clear_bit_hook = btrfs_clear_bit_hook,
10677         .merge_extent_hook = btrfs_merge_extent_hook,
10678         .split_extent_hook = btrfs_split_extent_hook,
10679         .check_extent_io_range = btrfs_check_extent_io_range,
10680 };
10681
10682 /*
10683  * btrfs doesn't support the bmap operation because swapfiles
10684  * use bmap to make a mapping of extents in the file.  They assume
10685  * these extents won't change over the life of the file and they
10686  * use the bmap result to do IO directly to the drive.
10687  *
10688  * the btrfs bmap call would return logical addresses that aren't
10689  * suitable for IO and they also will change frequently as COW
10690  * operations happen.  So, swapfile + btrfs == corruption.
10691  *
10692  * For now we're avoiding this by dropping bmap.
10693  */
10694 static const struct address_space_operations btrfs_aops = {
10695         .readpage       = btrfs_readpage,
10696         .writepage      = btrfs_writepage,
10697         .writepages     = btrfs_writepages,
10698         .readpages      = btrfs_readpages,
10699         .direct_IO      = btrfs_direct_IO,
10700         .invalidatepage = btrfs_invalidatepage,
10701         .releasepage    = btrfs_releasepage,
10702         .set_page_dirty = btrfs_set_page_dirty,
10703         .error_remove_page = generic_error_remove_page,
10704 };
10705
10706 static const struct address_space_operations btrfs_symlink_aops = {
10707         .readpage       = btrfs_readpage,
10708         .writepage      = btrfs_writepage,
10709         .invalidatepage = btrfs_invalidatepage,
10710         .releasepage    = btrfs_releasepage,
10711 };
10712
10713 static const struct inode_operations btrfs_file_inode_operations = {
10714         .getattr        = btrfs_getattr,
10715         .setattr        = btrfs_setattr,
10716         .listxattr      = btrfs_listxattr,
10717         .permission     = btrfs_permission,
10718         .fiemap         = btrfs_fiemap,
10719         .get_acl        = btrfs_get_acl,
10720         .set_acl        = btrfs_set_acl,
10721         .update_time    = btrfs_update_time,
10722 };
10723 static const struct inode_operations btrfs_special_inode_operations = {
10724         .getattr        = btrfs_getattr,
10725         .setattr        = btrfs_setattr,
10726         .permission     = btrfs_permission,
10727         .listxattr      = btrfs_listxattr,
10728         .get_acl        = btrfs_get_acl,
10729         .set_acl        = btrfs_set_acl,
10730         .update_time    = btrfs_update_time,
10731 };
10732 static const struct inode_operations btrfs_symlink_inode_operations = {
10733         .get_link       = page_get_link,
10734         .getattr        = btrfs_getattr,
10735         .setattr        = btrfs_setattr,
10736         .permission     = btrfs_permission,
10737         .listxattr      = btrfs_listxattr,
10738         .update_time    = btrfs_update_time,
10739 };
10740
10741 const struct dentry_operations btrfs_dentry_operations = {
10742         .d_delete       = btrfs_dentry_delete,
10743         .d_release      = btrfs_dentry_release,
10744 };