fs/btrfs/file-item.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2007 Oracle.  All rights reserved.
   4  */
   5
   6 #include <linux/bio.h>
   7 #include <linux/slab.h>
   8 #include <linux/pagemap.h>
   9 #include <linux/highmem.h>
  10 #include <linux/sched/mm.h>
  11 #include <crypto/hash.h>
  12 #include "ctree.h"
  13 #include "disk-io.h"
  14 #include "transaction.h"
  15 #include "volumes.h"
  16 #include "print-tree.h"
  17 #include "compression.h"
  18
  19 #define __MAX_CSUM_ITEMS(r, size) ((unsigned long)(((BTRFS_LEAF_DATA_SIZE(r) - \
  20                                    sizeof(struct btrfs_item) * 2) / \
  21                                   size) - 1))
  22
  23 #define MAX_CSUM_ITEMS(r, size) (min_t(u32, __MAX_CSUM_ITEMS(r, size), \
  24                                        PAGE_SIZE))
  25
  26 static inline u32 max_ordered_sum_bytes(struct btrfs_fs_info *fs_info,
  27                                         u16 csum_size)
  28 {
  29         u32 ncsums = (PAGE_SIZE - sizeof(struct btrfs_ordered_sum)) / csum_size;
  30
  31         return ncsums * fs_info->sectorsize;
  32 }
  33
  34 int btrfs_insert_file_extent(struct btrfs_trans_handle *trans,
  35                              struct btrfs_root *root,
  36                              u64 objectid, u64 pos,
  37                              u64 disk_offset, u64 disk_num_bytes,
  38                              u64 num_bytes, u64 offset, u64 ram_bytes,
  39                              u8 compression, u8 encryption, u16 other_encoding)
  40 {
  41         int ret = 0;
  42         struct btrfs_file_extent_item *item;
  43         struct btrfs_key file_key;
  44         struct btrfs_path *path;
  45         struct extent_buffer *leaf;
  46
  47         path = btrfs_alloc_path();
  48         if (!path)
  49                 return -ENOMEM;
  50         file_key.objectid = objectid;
  51         file_key.offset = pos;
  52         file_key.type = BTRFS_EXTENT_DATA_KEY;
  53
  54         path->leave_spinning = 1;
  55         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
  56                                       sizeof(*item));
  57         if (ret < 0)
  58                 goto out;
  59         BUG_ON(ret); /* Can't happen */
  60         leaf = path->nodes[0];
  61         item = btrfs_item_ptr(leaf, path->slots[0],
  62                               struct btrfs_file_extent_item);
  63         btrfs_set_file_extent_disk_bytenr(leaf, item, disk_offset);
  64         btrfs_set_file_extent_disk_num_bytes(leaf, item, disk_num_bytes);
  65         btrfs_set_file_extent_offset(leaf, item, offset);
  66         btrfs_set_file_extent_num_bytes(leaf, item, num_bytes);
  67         btrfs_set_file_extent_ram_bytes(leaf, item, ram_bytes);
  68         btrfs_set_file_extent_generation(leaf, item, trans->transid);
  69         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
  70         btrfs_set_file_extent_compression(leaf, item, compression);
  71         btrfs_set_file_extent_encryption(leaf, item, encryption);
  72         btrfs_set_file_extent_other_encoding(leaf, item, other_encoding);
  73
  74         btrfs_mark_buffer_dirty(leaf);
  75 out:
  76         btrfs_free_path(path);
  77         return ret;
  78 }
  79
  80 static struct btrfs_csum_item *
  81 btrfs_lookup_csum(struct btrfs_trans_handle *trans,
  82                   struct btrfs_root *root,
  83                   struct btrfs_path *path,
  84                   u64 bytenr, int cow)
  85 {
  86         struct btrfs_fs_info *fs_info = root->fs_info;
  87         int ret;
  88         struct btrfs_key file_key;
  89         struct btrfs_key found_key;
  90         struct btrfs_csum_item *item;
  91         struct extent_buffer *leaf;
  92         u64 csum_offset = 0;
  93         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
  94         int csums_in_item;
  95
  96         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
  97         file_key.offset = bytenr;
  98         file_key.type = BTRFS_EXTENT_CSUM_KEY;
  99         ret = btrfs_search_slot(trans, root, &file_key, path, 0, cow);
 100         if (ret < 0)
 101                 goto fail;
 102         leaf = path->nodes[0];
 103         if (ret > 0) {
 104                 ret = 1;
 105                 if (path->slots[0] == 0)
 106                         goto fail;
 107                 path->slots[0]--;
 108                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 109                 if (found_key.type != BTRFS_EXTENT_CSUM_KEY)
 110                         goto fail;
 111
 112                 csum_offset = (bytenr - found_key.offset) >>
 113                                 fs_info->sb->s_blocksize_bits;
 114                 csums_in_item = btrfs_item_size_nr(leaf, path->slots[0]);
 115                 csums_in_item /= csum_size;
 116
 117                 if (csum_offset == csums_in_item) {
 118                         ret = -EFBIG;
 119                         goto fail;
 120                 } else if (csum_offset > csums_in_item) {
 121                         goto fail;
 122                 }
 123         }
 124         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 125         item = (struct btrfs_csum_item *)((unsigned char *)item +
 126                                           csum_offset * csum_size);
 127         return item;
 128 fail:
 129         if (ret > 0)
 130                 ret = -ENOENT;
 131         return ERR_PTR(ret);
 132 }
 133
 134 int btrfs_lookup_file_extent(struct btrfs_trans_handle *trans,
 135                              struct btrfs_root *root,
 136                              struct btrfs_path *path, u64 objectid,
 137                              u64 offset, int mod)
 138 {
 139         int ret;
 140         struct btrfs_key file_key;
 141         int ins_len = mod < 0 ? -1 : 0;
 142         int cow = mod != 0;
 143
 144         file_key.objectid = objectid;
 145         file_key.offset = offset;
 146         file_key.type = BTRFS_EXTENT_DATA_KEY;
 147         ret = btrfs_search_slot(trans, root, &file_key, path, ins_len, cow);
 148         return ret;
 149 }
 150
 151 /**
 152  * btrfs_lookup_bio_sums - Look up checksums for a bio.
 153  * @inode: inode that the bio is for.
 154  * @bio: bio embedded in btrfs_io_bio.
 155  * @offset: Unless (u64)-1, look up checksums for this offset in the file.
 156  *          If (u64)-1, use the page offsets from the bio instead.
 157  * @dst: Buffer of size btrfs_super_csum_size() used to return checksum. If
 158  *       NULL, the checksum is returned in btrfs_io_bio(bio)->csum instead.
 159  *
 160  * Return: BLK_STS_RESOURCE if allocating memory fails, BLK_STS_OK otherwise.
 161  */
 162 blk_status_t btrfs_lookup_bio_sums(struct inode *inode, struct bio *bio,
 163                                    u64 offset, u8 *dst)
 164 {
 165         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 166         struct bio_vec bvec;
 167         struct bvec_iter iter;
 168         struct btrfs_io_bio *btrfs_bio = btrfs_io_bio(bio);
 169         struct btrfs_csum_item *item = NULL;
 170         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 171         struct btrfs_path *path;
 172         const bool page_offsets = (offset == (u64)-1);
 173         u8 *csum;
 174         u64 item_start_offset = 0;
 175         u64 item_last_offset = 0;
 176         u64 disk_bytenr;
 177         u64 page_bytes_left;
 178         u32 diff;
 179         int nblocks;
 180         int count = 0;
 181         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 182
 183         path = btrfs_alloc_path();
 184         if (!path)
 185                 return BLK_STS_RESOURCE;
 186
 187         nblocks = bio->bi_iter.bi_size >> inode->i_sb->s_blocksize_bits;
 188         if (!dst) {
 189                 if (nblocks * csum_size > BTRFS_BIO_INLINE_CSUM_SIZE) {
 190                         btrfs_bio->csum = kmalloc_array(nblocks, csum_size,
 191                                                         GFP_NOFS);
 192                         if (!btrfs_bio->csum) {
 193                                 btrfs_free_path(path);
 194                                 return BLK_STS_RESOURCE;
 195                         }
 196                 } else {
 197                         btrfs_bio->csum = btrfs_bio->csum_inline;
 198                 }
 199                 csum = btrfs_bio->csum;
 200         } else {
 201                 csum = dst;
 202         }
 203
 204         if (bio->bi_iter.bi_size > PAGE_SIZE * 8)
 205                 path->reada = READA_FORWARD;
 206
 207         /*
 208          * the free space stuff is only read when it hasn't been
 209          * updated in the current transaction.  So, we can safely
 210          * read from the commit root and sidestep a nasty deadlock
 211          * between reading the free space cache and updating the csum tree.
 212          */
 213         if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
 214                 path->search_commit_root = 1;
 215                 path->skip_locking = 1;
 216         }
 217
 218         disk_bytenr = (u64)bio->bi_iter.bi_sector << 9;
 219
 220         bio_for_each_segment(bvec, bio, iter) {
 221                 page_bytes_left = bvec.bv_len;
 222                 if (count)
 223                         goto next;
 224
 225                 if (page_offsets)
 226                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 227                 count = btrfs_find_ordered_sum(inode, offset, disk_bytenr,
 228                                                csum, nblocks);
 229                 if (count)
 230                         goto found;
 231
 232                 if (!item || disk_bytenr < item_start_offset ||
 233                     disk_bytenr >= item_last_offset) {
 234                         struct btrfs_key found_key;
 235                         u32 item_size;
 236
 237                         if (item)
 238                                 btrfs_release_path(path);
 239                         item = btrfs_lookup_csum(NULL, fs_info->csum_root,
 240                                                  path, disk_bytenr, 0);
 241                         if (IS_ERR(item)) {
 242                                 count = 1;
 243                                 memset(csum, 0, csum_size);
 244                                 if (BTRFS_I(inode)->root->root_key.objectid ==
 245                                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 246                                         set_extent_bits(io_tree, offset,
 247                                                 offset + fs_info->sectorsize - 1,
 248                                                 EXTENT_NODATASUM);
 249                                 } else {
 250                                         btrfs_info_rl(fs_info,
 251                                                    "no csum found for inode %llu start %llu",
 252                                                btrfs_ino(BTRFS_I(inode)), offset);
 253                                 }
 254                                 item = NULL;
 255                                 btrfs_release_path(path);
 256                                 goto found;
 257                         }
 258                         btrfs_item_key_to_cpu(path->nodes[0], &found_key,
 259                                               path->slots[0]);
 260
 261                         item_start_offset = found_key.offset;
 262                         item_size = btrfs_item_size_nr(path->nodes[0],
 263                                                        path->slots[0]);
 264                         item_last_offset = item_start_offset +
 265                                 (item_size / csum_size) *
 266                                 fs_info->sectorsize;
 267                         item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 268                                               struct btrfs_csum_item);
 269                 }
 270                 /*
 271                  * this byte range must be able to fit inside
 272                  * a single leaf so it will also fit inside a u32
 273                  */
 274                 diff = disk_bytenr - item_start_offset;
 275                 diff = diff / fs_info->sectorsize;
 276                 diff = diff * csum_size;
 277                 count = min_t(int, nblocks, (item_last_offset - disk_bytenr) >>
 278                                             inode->i_sb->s_blocksize_bits);
 279                 read_extent_buffer(path->nodes[0], csum,
 280                                    ((unsigned long)item) + diff,
 281                                    csum_size * count);
 282 found:
 283                 csum += count * csum_size;
 284                 nblocks -= count;
 285 next:
 286                 while (count > 0) {
 287                         count--;
 288                         disk_bytenr += fs_info->sectorsize;
 289                         offset += fs_info->sectorsize;
 290                         page_bytes_left -= fs_info->sectorsize;
 291                         if (!page_bytes_left)
 292                                 break; /* move to next bio */
 293                 }
 294         }
 295
 296         WARN_ON_ONCE(count);
 297         btrfs_free_path(path);
 298         return BLK_STS_OK;
 299 }
 300
 301 int btrfs_lookup_csums_range(struct btrfs_root *root, u64 start, u64 end,
 302                              struct list_head *list, int search_commit)
 303 {
 304         struct btrfs_fs_info *fs_info = root->fs_info;
 305         struct btrfs_key key;
 306         struct btrfs_path *path;
 307         struct extent_buffer *leaf;
 308         struct btrfs_ordered_sum *sums;
 309         struct btrfs_csum_item *item;
 310         LIST_HEAD(tmplist);
 311         unsigned long offset;
 312         int ret;
 313         size_t size;
 314         u64 csum_end;
 315         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 316
 317         ASSERT(IS_ALIGNED(start, fs_info->sectorsize) &&
 318                IS_ALIGNED(end + 1, fs_info->sectorsize));
 319
 320         path = btrfs_alloc_path();
 321         if (!path)
 322                 return -ENOMEM;
 323
 324         if (search_commit) {
 325                 path->skip_locking = 1;
 326                 path->reada = READA_FORWARD;
 327                 path->search_commit_root = 1;
 328         }
 329
 330         key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 331         key.offset = start;
 332         key.type = BTRFS_EXTENT_CSUM_KEY;
 333
 334         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 335         if (ret < 0)
 336                 goto fail;
 337         if (ret > 0 && path->slots[0] > 0) {
 338                 leaf = path->nodes[0];
 339                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 340                 if (key.objectid == BTRFS_EXTENT_CSUM_OBJECTID &&
 341                     key.type == BTRFS_EXTENT_CSUM_KEY) {
 342                         offset = (start - key.offset) >>
 343                                  fs_info->sb->s_blocksize_bits;
 344                         if (offset * csum_size <
 345                             btrfs_item_size_nr(leaf, path->slots[0] - 1))
 346                                 path->slots[0]--;
 347                 }
 348         }
 349
 350         while (start <= end) {
 351                 leaf = path->nodes[0];
 352                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 353                         ret = btrfs_next_leaf(root, path);
 354                         if (ret < 0)
 355                                 goto fail;
 356                         if (ret > 0)
 357                                 break;
 358                         leaf = path->nodes[0];
 359                 }
 360
 361                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 362                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 363                     key.type != BTRFS_EXTENT_CSUM_KEY ||
 364                     key.offset > end)
 365                         break;
 366
 367                 if (key.offset > start)
 368                         start = key.offset;
 369
 370                 size = btrfs_item_size_nr(leaf, path->slots[0]);
 371                 csum_end = key.offset + (size / csum_size) * fs_info->sectorsize;
 372                 if (csum_end <= start) {
 373                         path->slots[0]++;
 374                         continue;
 375                 }
 376
 377                 csum_end = min(csum_end, end + 1);
 378                 item = btrfs_item_ptr(path->nodes[0], path->slots[0],
 379                                       struct btrfs_csum_item);
 380                 while (start < csum_end) {
 381                         size = min_t(size_t, csum_end - start,
 382                                      max_ordered_sum_bytes(fs_info, csum_size));
 383                         sums = kzalloc(btrfs_ordered_sum_size(fs_info, size),
 384                                        GFP_NOFS);
 385                         if (!sums) {
 386                                 ret = -ENOMEM;
 387                                 goto fail;
 388                         }
 389
 390                         sums->bytenr = start;
 391                         sums->len = (int)size;
 392
 393                         offset = (start - key.offset) >>
 394                                 fs_info->sb->s_blocksize_bits;
 395                         offset *= csum_size;
 396                         size >>= fs_info->sb->s_blocksize_bits;
 397
 398                         read_extent_buffer(path->nodes[0],
 399                                            sums->sums,
 400                                            ((unsigned long)item) + offset,
 401                                            csum_size * size);
 402
 403                         start += fs_info->sectorsize * size;
 404                         list_add_tail(&sums->list, &tmplist);
 405                 }
 406                 path->slots[0]++;
 407         }
 408         ret = 0;
 409 fail:
 410         while (ret < 0 && !list_empty(&tmplist)) {
 411                 sums = list_entry(tmplist.next, struct btrfs_ordered_sum, list);
 412                 list_del(&sums->list);
 413                 kfree(sums);
 414         }
 415         list_splice_tail(&tmplist, list);
 416
 417         btrfs_free_path(path);
 418         return ret;
 419 }
 420
 421 /*
 422  * btrfs_csum_one_bio - Calculates checksums of the data contained inside a bio
 423  * @inode:       Owner of the data inside the bio
 424  * @bio:         Contains the data to be checksummed
 425  * @file_start:  offset in file this bio begins to describe
 426  * @contig:      Boolean. If true/1 means all bio vecs in this bio are
 427  *               contiguous and they begin at @file_start in the file. False/0
 428  *               means this bio can contains potentially discontigous bio vecs
 429  *               so the logical offset of each should be calculated separately.
 430  */
 431 blk_status_t btrfs_csum_one_bio(struct inode *inode, struct bio *bio,
 432                        u64 file_start, int contig)
 433 {
 434         struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 435         SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
 436         struct btrfs_ordered_sum *sums;
 437         struct btrfs_ordered_extent *ordered = NULL;
 438         char *data;
 439         struct bvec_iter iter;
 440         struct bio_vec bvec;
 441         int index;
 442         int nr_sectors;
 443         unsigned long total_bytes = 0;
 444         unsigned long this_sum_bytes = 0;
 445         int i;
 446         u64 offset;
 447         unsigned nofs_flag;
 448         const u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 449
 450         nofs_flag = memalloc_nofs_save();
 451         sums = kvzalloc(btrfs_ordered_sum_size(fs_info, bio->bi_iter.bi_size),
 452                        GFP_KERNEL);
 453         memalloc_nofs_restore(nofs_flag);
 454
 455         if (!sums)
 456                 return BLK_STS_RESOURCE;
 457
 458         sums->len = bio->bi_iter.bi_size;
 459         INIT_LIST_HEAD(&sums->list);
 460
 461         if (contig)
 462                 offset = file_start;
 463         else
 464                 offset = 0; /* shut up gcc */
 465
 466         sums->bytenr = (u64)bio->bi_iter.bi_sector << 9;
 467         index = 0;
 468
 469         shash->tfm = fs_info->csum_shash;
 470
 471         bio_for_each_segment(bvec, bio, iter) {
 472                 if (!contig)
 473                         offset = page_offset(bvec.bv_page) + bvec.bv_offset;
 474
 475                 if (!ordered) {
 476                         ordered = btrfs_lookup_ordered_extent(inode, offset);
 477                         BUG_ON(!ordered); /* Logic error */
 478                 }
 479
 480                 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info,
 481                                                  bvec.bv_len + fs_info->sectorsize
 482                                                  - 1);
 483
 484                 for (i = 0; i < nr_sectors; i++) {
 485                         if (offset >= ordered->file_offset + ordered->num_bytes ||
 486                             offset < ordered->file_offset) {
 487                                 unsigned long bytes_left;
 488
 489                                 sums->len = this_sum_bytes;
 490                                 this_sum_bytes = 0;
 491                                 btrfs_add_ordered_sum(ordered, sums);
 492                                 btrfs_put_ordered_extent(ordered);
 493
 494                                 bytes_left = bio->bi_iter.bi_size - total_bytes;
 495
 496                                 nofs_flag = memalloc_nofs_save();
 497                                 sums = kvzalloc(btrfs_ordered_sum_size(fs_info,
 498                                                       bytes_left), GFP_KERNEL);
 499                                 memalloc_nofs_restore(nofs_flag);
 500                                 BUG_ON(!sums); /* -ENOMEM */
 501                                 sums->len = bytes_left;
 502                                 ordered = btrfs_lookup_ordered_extent(inode,
 503                                                                 offset);
 504                                 ASSERT(ordered); /* Logic error */
 505                                 sums->bytenr = ((u64)bio->bi_iter.bi_sector << 9)
 506                                         + total_bytes;
 507                                 index = 0;
 508                         }
 509
 510                         crypto_shash_init(shash);
 511                         data = kmap_atomic(bvec.bv_page);
 512                         crypto_shash_update(shash, data + bvec.bv_offset
 513                                             + (i * fs_info->sectorsize),
 514                                             fs_info->sectorsize);
 515                         kunmap_atomic(data);
 516                         crypto_shash_final(shash, (char *)(sums->sums + index));
 517                         index += csum_size;
 518                         offset += fs_info->sectorsize;
 519                         this_sum_bytes += fs_info->sectorsize;
 520                         total_bytes += fs_info->sectorsize;
 521                 }
 522
 523         }
 524         this_sum_bytes = 0;
 525         btrfs_add_ordered_sum(ordered, sums);
 526         btrfs_put_ordered_extent(ordered);
 527         return 0;
 528 }
 529
 530 /*
 531  * helper function for csum removal, this expects the
 532  * key to describe the csum pointed to by the path, and it expects
 533  * the csum to overlap the range [bytenr, len]
 534  *
 535  * The csum should not be entirely contained in the range and the
 536  * range should not be entirely contained in the csum.
 537  *
 538  * This calls btrfs_truncate_item with the correct args based on the
 539  * overlap, and fixes up the key as required.
 540  */
 541 static noinline void truncate_one_csum(struct btrfs_fs_info *fs_info,
 542                                        struct btrfs_path *path,
 543                                        struct btrfs_key *key,
 544                                        u64 bytenr, u64 len)
 545 {
 546         struct extent_buffer *leaf;
 547         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 548         u64 csum_end;
 549         u64 end_byte = bytenr + len;
 550         u32 blocksize_bits = fs_info->sb->s_blocksize_bits;
 551
 552         leaf = path->nodes[0];
 553         csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 554         csum_end <<= fs_info->sb->s_blocksize_bits;
 555         csum_end += key->offset;
 556
 557         if (key->offset < bytenr && csum_end <= end_byte) {
 558                 /*
 559                  *         [ bytenr - len ]
 560                  *         [   ]
 561                  *   [csum     ]
 562                  *   A simple truncate off the end of the item
 563                  */
 564                 u32 new_size = (bytenr - key->offset) >> blocksize_bits;
 565                 new_size *= csum_size;
 566                 btrfs_truncate_item(path, new_size, 1);
 567         } else if (key->offset >= bytenr && csum_end > end_byte &&
 568                    end_byte > key->offset) {
 569                 /*
 570                  *         [ bytenr - len ]
 571                  *                 [ ]
 572                  *                 [csum     ]
 573                  * we need to truncate from the beginning of the csum
 574                  */
 575                 u32 new_size = (csum_end - end_byte) >> blocksize_bits;
 576                 new_size *= csum_size;
 577
 578                 btrfs_truncate_item(path, new_size, 0);
 579
 580                 key->offset = end_byte;
 581                 btrfs_set_item_key_safe(fs_info, path, key);
 582         } else {
 583                 BUG();
 584         }
 585 }
 586
 587 /*
 588  * deletes the csum items from the csum tree for a given
 589  * range of bytes.
 590  */
 591 int btrfs_del_csums(struct btrfs_trans_handle *trans,
 592                     struct btrfs_root *root, u64 bytenr, u64 len)
 593 {
 594         struct btrfs_fs_info *fs_info = trans->fs_info;
 595         struct btrfs_path *path;
 596         struct btrfs_key key;
 597         u64 end_byte = bytenr + len;
 598         u64 csum_end;
 599         struct extent_buffer *leaf;
 600         int ret;
 601         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 602         int blocksize_bits = fs_info->sb->s_blocksize_bits;
 603
 604         ASSERT(root == fs_info->csum_root ||
 605                root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
 606
 607         path = btrfs_alloc_path();
 608         if (!path)
 609                 return -ENOMEM;
 610
 611         while (1) {
 612                 key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 613                 key.offset = end_byte - 1;
 614                 key.type = BTRFS_EXTENT_CSUM_KEY;
 615
 616                 path->leave_spinning = 1;
 617                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 618                 if (ret > 0) {
 619                         if (path->slots[0] == 0)
 620                                 break;
 621                         path->slots[0]--;
 622                 } else if (ret < 0) {
 623                         break;
 624                 }
 625
 626                 leaf = path->nodes[0];
 627                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 628
 629                 if (key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 630                     key.type != BTRFS_EXTENT_CSUM_KEY) {
 631                         break;
 632                 }
 633
 634                 if (key.offset >= end_byte)
 635                         break;
 636
 637                 csum_end = btrfs_item_size_nr(leaf, path->slots[0]) / csum_size;
 638                 csum_end <<= blocksize_bits;
 639                 csum_end += key.offset;
 640
 641                 /* this csum ends before we start, we're done */
 642                 if (csum_end <= bytenr)
 643                         break;
 644
 645                 /* delete the entire item, it is inside our range */
 646                 if (key.offset >= bytenr && csum_end <= end_byte) {
 647                         int del_nr = 1;
 648
 649                         /*
 650                          * Check how many csum items preceding this one in this
 651                          * leaf correspond to our range and then delete them all
 652                          * at once.
 653                          */
 654                         if (key.offset > bytenr && path->slots[0] > 0) {
 655                                 int slot = path->slots[0] - 1;
 656
 657                                 while (slot >= 0) {
 658                                         struct btrfs_key pk;
 659
 660                                         btrfs_item_key_to_cpu(leaf, &pk, slot);
 661                                         if (pk.offset < bytenr ||
 662                                             pk.type != BTRFS_EXTENT_CSUM_KEY ||
 663                                             pk.objectid !=
 664                                             BTRFS_EXTENT_CSUM_OBJECTID)
 665                                                 break;
 666                                         path->slots[0] = slot;
 667                                         del_nr++;
 668                                         key.offset = pk.offset;
 669                                         slot--;
 670                                 }
 671                         }
 672                         ret = btrfs_del_items(trans, root, path,
 673                                               path->slots[0], del_nr);
 674                         if (ret)
 675                                 goto out;
 676                         if (key.offset == bytenr)
 677                                 break;
 678                 } else if (key.offset < bytenr && csum_end > end_byte) {
 679                         unsigned long offset;
 680                         unsigned long shift_len;
 681                         unsigned long item_offset;
 682                         /*
 683                          *        [ bytenr - len ]
 684                          *     [csum                ]
 685                          *
 686                          * Our bytes are in the middle of the csum,
 687                          * we need to split this item and insert a new one.
 688                          *
 689                          * But we can't drop the path because the
 690                          * csum could change, get removed, extended etc.
 691                          *
 692                          * The trick here is the max size of a csum item leaves
 693                          * enough room in the tree block for a single
 694                          * item header.  So, we split the item in place,
 695                          * adding a new header pointing to the existing
 696                          * bytes.  Then we loop around again and we have
 697                          * a nicely formed csum item that we can neatly
 698                          * truncate.
 699                          */
 700                         offset = (bytenr - key.offset) >> blocksize_bits;
 701                         offset *= csum_size;
 702
 703                         shift_len = (len >> blocksize_bits) * csum_size;
 704
 705                         item_offset = btrfs_item_ptr_offset(leaf,
 706                                                             path->slots[0]);
 707
 708                         memzero_extent_buffer(leaf, item_offset + offset,
 709                                              shift_len);
 710                         key.offset = bytenr;
 711
 712                         /*
 713                          * btrfs_split_item returns -EAGAIN when the
 714                          * item changed size or key
 715                          */
 716                         ret = btrfs_split_item(trans, root, path, &key, offset);
 717                         if (ret && ret != -EAGAIN) {
 718                                 btrfs_abort_transaction(trans, ret);
 719                                 goto out;
 720                         }
 721
 722                         key.offset = end_byte - 1;
 723                 } else {
 724                         truncate_one_csum(fs_info, path, &key, bytenr, len);
 725                         if (key.offset < bytenr)
 726                                 break;
 727                 }
 728                 btrfs_release_path(path);
 729         }
 730         ret = 0;
 731 out:
 732         btrfs_free_path(path);
 733         return ret;
 734 }
 735
 736 int btrfs_csum_file_blocks(struct btrfs_trans_handle *trans,
 737                            struct btrfs_root *root,
 738                            struct btrfs_ordered_sum *sums)
 739 {
 740         struct btrfs_fs_info *fs_info = root->fs_info;
 741         struct btrfs_key file_key;
 742         struct btrfs_key found_key;
 743         struct btrfs_path *path;
 744         struct btrfs_csum_item *item;
 745         struct btrfs_csum_item *item_end;
 746         struct extent_buffer *leaf = NULL;
 747         u64 next_offset;
 748         u64 total_bytes = 0;
 749         u64 csum_offset;
 750         u64 bytenr;
 751         u32 nritems;
 752         u32 ins_size;
 753         int index = 0;
 754         int found_next;
 755         int ret;
 756         u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
 757
 758         path = btrfs_alloc_path();
 759         if (!path)
 760                 return -ENOMEM;
 761 again:
 762         next_offset = (u64)-1;
 763         found_next = 0;
 764         bytenr = sums->bytenr + total_bytes;
 765         file_key.objectid = BTRFS_EXTENT_CSUM_OBJECTID;
 766         file_key.offset = bytenr;
 767         file_key.type = BTRFS_EXTENT_CSUM_KEY;
 768
 769         item = btrfs_lookup_csum(trans, root, path, bytenr, 1);
 770         if (!IS_ERR(item)) {
 771                 ret = 0;
 772                 leaf = path->nodes[0];
 773                 item_end = btrfs_item_ptr(leaf, path->slots[0],
 774                                           struct btrfs_csum_item);
 775                 item_end = (struct btrfs_csum_item *)((char *)item_end +
 776                            btrfs_item_size_nr(leaf, path->slots[0]));
 777                 goto found;
 778         }
 779         ret = PTR_ERR(item);
 780         if (ret != -EFBIG && ret != -ENOENT)
 781                 goto fail_unlock;
 782
 783         if (ret == -EFBIG) {
 784                 u32 item_size;
 785                 /* we found one, but it isn't big enough yet */
 786                 leaf = path->nodes[0];
 787                 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
 788                 if ((item_size / csum_size) >=
 789                     MAX_CSUM_ITEMS(fs_info, csum_size)) {
 790                         /* already at max size, make a new one */
 791                         goto insert;
 792                 }
 793         } else {
 794                 int slot = path->slots[0] + 1;
 795                 /* we didn't find a csum item, insert one */
 796                 nritems = btrfs_header_nritems(path->nodes[0]);
 797                 if (!nritems || (path->slots[0] >= nritems - 1)) {
 798                         ret = btrfs_next_leaf(root, path);
 799                         if (ret == 1)
 800                                 found_next = 1;
 801                         if (ret != 0)
 802                                 goto insert;
 803                         slot = path->slots[0];
 804                 }
 805                 btrfs_item_key_to_cpu(path->nodes[0], &found_key, slot);
 806                 if (found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 807                     found_key.type != BTRFS_EXTENT_CSUM_KEY) {
 808                         found_next = 1;
 809                         goto insert;
 810                 }
 811                 next_offset = found_key.offset;
 812                 found_next = 1;
 813                 goto insert;
 814         }
 815
 816         /*
 817          * at this point, we know the tree has an item, but it isn't big
 818          * enough yet to put our csum in.  Grow it
 819          */
 820         btrfs_release_path(path);
 821         ret = btrfs_search_slot(trans, root, &file_key, path,
 822                                 csum_size, 1);
 823         if (ret < 0)
 824                 goto fail_unlock;
 825
 826         if (ret > 0) {
 827                 if (path->slots[0] == 0)
 828                         goto insert;
 829                 path->slots[0]--;
 830         }
 831
 832         leaf = path->nodes[0];
 833         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 834         csum_offset = (bytenr - found_key.offset) >>
 835                         fs_info->sb->s_blocksize_bits;
 836
 837         if (found_key.type != BTRFS_EXTENT_CSUM_KEY ||
 838             found_key.objectid != BTRFS_EXTENT_CSUM_OBJECTID ||
 839             csum_offset >= MAX_CSUM_ITEMS(fs_info, csum_size)) {
 840                 goto insert;
 841         }
 842
 843         if (csum_offset == btrfs_item_size_nr(leaf, path->slots[0]) /
 844             csum_size) {
 845                 int extend_nr;
 846                 u64 tmp;
 847                 u32 diff;
 848                 u32 free_space;
 849
 850                 if (btrfs_leaf_free_space(leaf) <
 851                                  sizeof(struct btrfs_item) + csum_size * 2)
 852                         goto insert;
 853
 854                 free_space = btrfs_leaf_free_space(leaf) -
 855                                          sizeof(struct btrfs_item) - csum_size;
 856                 tmp = sums->len - total_bytes;
 857                 tmp >>= fs_info->sb->s_blocksize_bits;
 858                 WARN_ON(tmp < 1);
 859
 860                 extend_nr = max_t(int, 1, (int)tmp);
 861                 diff = (csum_offset + extend_nr) * csum_size;
 862                 diff = min(diff,
 863                            MAX_CSUM_ITEMS(fs_info, csum_size) * csum_size);
 864
 865                 diff = diff - btrfs_item_size_nr(leaf, path->slots[0]);
 866                 diff = min(free_space, diff);
 867                 diff /= csum_size;
 868                 diff *= csum_size;
 869
 870                 btrfs_extend_item(path, diff);
 871                 ret = 0;
 872                 goto csum;
 873         }
 874
 875 insert:
 876         btrfs_release_path(path);
 877         csum_offset = 0;
 878         if (found_next) {
 879                 u64 tmp;
 880
 881                 tmp = sums->len - total_bytes;
 882                 tmp >>= fs_info->sb->s_blocksize_bits;
 883                 tmp = min(tmp, (next_offset - file_key.offset) >>
 884                                          fs_info->sb->s_blocksize_bits);
 885
 886                 tmp = max_t(u64, 1, tmp);
 887                 tmp = min_t(u64, tmp, MAX_CSUM_ITEMS(fs_info, csum_size));
 888                 ins_size = csum_size * tmp;
 889         } else {
 890                 ins_size = csum_size;
 891         }
 892         path->leave_spinning = 1;
 893         ret = btrfs_insert_empty_item(trans, root, path, &file_key,
 894                                       ins_size);
 895         path->leave_spinning = 0;
 896         if (ret < 0)
 897                 goto fail_unlock;
 898         if (WARN_ON(ret != 0))
 899                 goto fail_unlock;
 900         leaf = path->nodes[0];
 901 csum:
 902         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_csum_item);
 903         item_end = (struct btrfs_csum_item *)((unsigned char *)item +
 904                                       btrfs_item_size_nr(leaf, path->slots[0]));
 905         item = (struct btrfs_csum_item *)((unsigned char *)item +
 906                                           csum_offset * csum_size);
 907 found:
 908         ins_size = (u32)(sums->len - total_bytes) >>
 909                    fs_info->sb->s_blocksize_bits;
 910         ins_size *= csum_size;
 911         ins_size = min_t(u32, (unsigned long)item_end - (unsigned long)item,
 912                               ins_size);
 913         write_extent_buffer(leaf, sums->sums + index, (unsigned long)item,
 914                             ins_size);
 915
 916         index += ins_size;
 917         ins_size /= csum_size;
 918         total_bytes += ins_size * fs_info->sectorsize;
 919
 920         btrfs_mark_buffer_dirty(path->nodes[0]);
 921         if (total_bytes < sums->len) {
 922                 btrfs_release_path(path);
 923                 cond_resched();
 924                 goto again;
 925         }
 926 out:
 927         btrfs_free_path(path);
 928         return ret;
 929
 930 fail_unlock:
 931         goto out;
 932 }
 933
 934 void btrfs_extent_item_to_extent_map(struct btrfs_inode *inode,
 935                                      const struct btrfs_path *path,
 936                                      struct btrfs_file_extent_item *fi,
 937                                      const bool new_inline,
 938                                      struct extent_map *em)
 939 {
 940         struct btrfs_fs_info *fs_info = inode->root->fs_info;
 941         struct btrfs_root *root = inode->root;
 942         struct extent_buffer *leaf = path->nodes[0];
 943         const int slot = path->slots[0];
 944         struct btrfs_key key;
 945         u64 extent_start, extent_end;
 946         u64 bytenr;
 947         u8 type = btrfs_file_extent_type(leaf, fi);
 948         int compress_type = btrfs_file_extent_compression(leaf, fi);
 949
 950         btrfs_item_key_to_cpu(leaf, &key, slot);
 951         extent_start = key.offset;
 952
 953         if (type == BTRFS_FILE_EXTENT_REG ||
 954             type == BTRFS_FILE_EXTENT_PREALLOC) {
 955                 extent_end = extent_start +
 956                         btrfs_file_extent_num_bytes(leaf, fi);
 957         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 958                 size_t size;
 959                 size = btrfs_file_extent_ram_bytes(leaf, fi);
 960                 extent_end = ALIGN(extent_start + size,
 961                                    fs_info->sectorsize);
 962         }
 963
 964         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
 965         if (type == BTRFS_FILE_EXTENT_REG ||
 966             type == BTRFS_FILE_EXTENT_PREALLOC) {
 967                 em->start = extent_start;
 968                 em->len = extent_end - extent_start;
 969                 em->orig_start = extent_start -
 970                         btrfs_file_extent_offset(leaf, fi);
 971                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
 972                 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 973                 if (bytenr == 0) {
 974                         em->block_start = EXTENT_MAP_HOLE;
 975                         return;
 976                 }
 977                 if (compress_type != BTRFS_COMPRESS_NONE) {
 978                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 979                         em->compress_type = compress_type;
 980                         em->block_start = bytenr;
 981                         em->block_len = em->orig_block_len;
 982                 } else {
 983                         bytenr += btrfs_file_extent_offset(leaf, fi);
 984                         em->block_start = bytenr;
 985                         em->block_len = em->len;
 986                         if (type == BTRFS_FILE_EXTENT_PREALLOC)
 987                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
 988                 }
 989         } else if (type == BTRFS_FILE_EXTENT_INLINE) {
 990                 em->block_start = EXTENT_MAP_INLINE;
 991                 em->start = extent_start;
 992                 em->len = extent_end - extent_start;
 993                 /*
 994                  * Initialize orig_start and block_len with the same values
 995                  * as in inode.c:btrfs_get_extent().
 996                  */
 997                 em->orig_start = EXTENT_MAP_HOLE;
 998                 em->block_len = (u64)-1;
 999                 if (!new_inline && compress_type != BTRFS_COMPRESS_NONE) {
1000                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
1001                         em->compress_type = compress_type;
1002                 }
1003         } else {
1004                 btrfs_err(fs_info,
1005                           "unknown file extent item type %d, inode %llu, offset %llu, "
1006                           "root %llu", type, btrfs_ino(inode), extent_start,
1007                           root->root_key.objectid);
1008         }
1009 }