1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/kernel.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <asm/unaligned.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "print-tree.h"
36 #include "ordered-data.h"
40 #include "compression.h"
42 #include "free-space-cache.h"
43 #include "inode-map.h"
49 struct btrfs_iget_args {
50 struct btrfs_key *location;
51 struct btrfs_root *root;
54 struct btrfs_dio_data {
56 u64 unsubmitted_oe_range_start;
57 u64 unsubmitted_oe_range_end;
61 static const struct inode_operations btrfs_dir_inode_operations;
62 static const struct inode_operations btrfs_symlink_inode_operations;
63 static const struct inode_operations btrfs_dir_ro_inode_operations;
64 static const struct inode_operations btrfs_special_inode_operations;
65 static const struct inode_operations btrfs_file_inode_operations;
66 static const struct address_space_operations btrfs_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static const struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73 struct kmem_cache *btrfs_free_space_cachep;
76 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
77 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
78 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
79 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
80 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
81 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
82 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
83 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
86 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
87 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
88 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
89 static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, u64 delalloc_end,
92 int *page_started, unsigned long *nr_written,
93 int unlock, struct btrfs_dedupe_hash *hash);
94 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
95 u64 orig_start, u64 block_start,
96 u64 block_len, u64 orig_block_len,
97 u64 ram_bytes, int compress_type,
100 static void __endio_write_update_ordered(struct inode *inode,
101 const u64 offset, const u64 bytes,
102 const bool uptodate);
105 * Cleanup all submitted ordered extents in specified range to handle errors
106 * from the fill_dellaloc() callback.
108 * NOTE: caller must ensure that when an error happens, it can not call
109 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
110 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
111 * to be released, which we want to happen only when finishing the ordered
112 * extent (btrfs_finish_ordered_io()). Also note that the caller of
113 * btrfs_run_delalloc_range already does proper cleanup for the first page of
114 * the range, that is, it invokes the callback writepage_end_io_hook() for the
115 * range of the first page.
117 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
121 unsigned long index = offset >> PAGE_SHIFT;
122 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
125 while (index <= end_index) {
126 page = find_get_page(inode->i_mapping, index);
130 ClearPagePrivate2(page);
133 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
134 bytes - PAGE_SIZE, false);
137 static int btrfs_dirty_inode(struct inode *inode);
139 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
140 void btrfs_test_inode_set_ops(struct inode *inode)
142 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
146 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
147 struct inode *inode, struct inode *dir,
148 const struct qstr *qstr)
152 err = btrfs_init_acl(trans, inode, dir);
154 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
159 * this does all the hard work for inserting an inline extent into
160 * the btree. The caller should have done a btrfs_drop_extents so that
161 * no overlapping inline items exist in the btree
163 static int insert_inline_extent(struct btrfs_trans_handle *trans,
164 struct btrfs_path *path, int extent_inserted,
165 struct btrfs_root *root, struct inode *inode,
166 u64 start, size_t size, size_t compressed_size,
168 struct page **compressed_pages)
170 struct extent_buffer *leaf;
171 struct page *page = NULL;
174 struct btrfs_file_extent_item *ei;
176 size_t cur_size = size;
177 unsigned long offset;
179 if (compressed_size && compressed_pages)
180 cur_size = compressed_size;
182 inode_add_bytes(inode, size);
184 if (!extent_inserted) {
185 struct btrfs_key key;
188 key.objectid = btrfs_ino(BTRFS_I(inode));
190 key.type = BTRFS_EXTENT_DATA_KEY;
192 datasize = btrfs_file_extent_calc_inline_size(cur_size);
193 path->leave_spinning = 1;
194 ret = btrfs_insert_empty_item(trans, root, path, &key,
199 leaf = path->nodes[0];
200 ei = btrfs_item_ptr(leaf, path->slots[0],
201 struct btrfs_file_extent_item);
202 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
203 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
204 btrfs_set_file_extent_encryption(leaf, ei, 0);
205 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
206 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
207 ptr = btrfs_file_extent_inline_start(ei);
209 if (compress_type != BTRFS_COMPRESS_NONE) {
212 while (compressed_size > 0) {
213 cpage = compressed_pages[i];
214 cur_size = min_t(unsigned long, compressed_size,
217 kaddr = kmap_atomic(cpage);
218 write_extent_buffer(leaf, kaddr, ptr, cur_size);
219 kunmap_atomic(kaddr);
223 compressed_size -= cur_size;
225 btrfs_set_file_extent_compression(leaf, ei,
228 page = find_get_page(inode->i_mapping,
229 start >> PAGE_SHIFT);
230 btrfs_set_file_extent_compression(leaf, ei, 0);
231 kaddr = kmap_atomic(page);
232 offset = start & (PAGE_SIZE - 1);
233 write_extent_buffer(leaf, kaddr + offset, ptr, size);
234 kunmap_atomic(kaddr);
237 btrfs_mark_buffer_dirty(leaf);
238 btrfs_release_path(path);
241 * we're an inline extent, so nobody can
242 * extend the file past i_size without locking
243 * a page we already have locked.
245 * We must do any isize and inode updates
246 * before we unlock the pages. Otherwise we
247 * could end up racing with unlink.
249 BTRFS_I(inode)->disk_i_size = inode->i_size;
250 ret = btrfs_update_inode(trans, root, inode);
258 * conditionally insert an inline extent into the file. This
259 * does the checks required to make sure the data is small enough
260 * to fit as an inline extent.
262 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
263 u64 end, size_t compressed_size,
265 struct page **compressed_pages)
267 struct btrfs_root *root = BTRFS_I(inode)->root;
268 struct btrfs_fs_info *fs_info = root->fs_info;
269 struct btrfs_trans_handle *trans;
270 u64 isize = i_size_read(inode);
271 u64 actual_end = min(end + 1, isize);
272 u64 inline_len = actual_end - start;
273 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
274 u64 data_len = inline_len;
276 struct btrfs_path *path;
277 int extent_inserted = 0;
278 u32 extent_item_size;
281 data_len = compressed_size;
284 actual_end > fs_info->sectorsize ||
285 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
287 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
289 data_len > fs_info->max_inline) {
293 path = btrfs_alloc_path();
297 trans = btrfs_join_transaction(root);
299 btrfs_free_path(path);
300 return PTR_ERR(trans);
302 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
304 if (compressed_size && compressed_pages)
305 extent_item_size = btrfs_file_extent_calc_inline_size(
308 extent_item_size = btrfs_file_extent_calc_inline_size(
311 ret = __btrfs_drop_extents(trans, root, inode, path,
312 start, aligned_end, NULL,
313 1, 1, extent_item_size, &extent_inserted);
315 btrfs_abort_transaction(trans, ret);
319 if (isize > actual_end)
320 inline_len = min_t(u64, isize, actual_end);
321 ret = insert_inline_extent(trans, path, extent_inserted,
323 inline_len, compressed_size,
324 compress_type, compressed_pages);
325 if (ret && ret != -ENOSPC) {
326 btrfs_abort_transaction(trans, ret);
328 } else if (ret == -ENOSPC) {
333 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
334 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
337 * Don't forget to free the reserved space, as for inlined extent
338 * it won't count as data extent, free them directly here.
339 * And at reserve time, it's always aligned to page size, so
340 * just free one page here.
342 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
343 btrfs_free_path(path);
344 btrfs_end_transaction(trans);
348 struct async_extent {
353 unsigned long nr_pages;
355 struct list_head list;
360 struct btrfs_root *root;
361 struct page *locked_page;
364 unsigned int write_flags;
365 struct list_head extents;
366 struct btrfs_work work;
369 static noinline int add_async_extent(struct async_cow *cow,
370 u64 start, u64 ram_size,
373 unsigned long nr_pages,
376 struct async_extent *async_extent;
378 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
379 BUG_ON(!async_extent); /* -ENOMEM */
380 async_extent->start = start;
381 async_extent->ram_size = ram_size;
382 async_extent->compressed_size = compressed_size;
383 async_extent->pages = pages;
384 async_extent->nr_pages = nr_pages;
385 async_extent->compress_type = compress_type;
386 list_add_tail(&async_extent->list, &cow->extents);
390 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
392 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
395 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
398 if (BTRFS_I(inode)->defrag_compress)
400 /* bad compression ratios */
401 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
403 if (btrfs_test_opt(fs_info, COMPRESS) ||
404 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
405 BTRFS_I(inode)->prop_compress)
406 return btrfs_compress_heuristic(inode, start, end);
410 static inline void inode_should_defrag(struct btrfs_inode *inode,
411 u64 start, u64 end, u64 num_bytes, u64 small_write)
413 /* If this is a small write inside eof, kick off a defrag */
414 if (num_bytes < small_write &&
415 (start > 0 || end + 1 < inode->disk_i_size))
416 btrfs_add_inode_defrag(NULL, inode);
420 * we create compressed extents in two phases. The first
421 * phase compresses a range of pages that have already been
422 * locked (both pages and state bits are locked).
424 * This is done inside an ordered work queue, and the compression
425 * is spread across many cpus. The actual IO submission is step
426 * two, and the ordered work queue takes care of making sure that
427 * happens in the same order things were put onto the queue by
428 * writepages and friends.
430 * If this code finds it can't get good compression, it puts an
431 * entry onto the work queue to write the uncompressed bytes. This
432 * makes sure that both compressed inodes and uncompressed inodes
433 * are written in the same order that the flusher thread sent them
436 static noinline void compress_file_range(struct inode *inode,
437 struct page *locked_page,
439 struct async_cow *async_cow,
442 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
443 u64 blocksize = fs_info->sectorsize;
445 u64 isize = i_size_read(inode);
447 struct page **pages = NULL;
448 unsigned long nr_pages;
449 unsigned long total_compressed = 0;
450 unsigned long total_in = 0;
453 int compress_type = fs_info->compress_type;
456 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
459 actual_end = min_t(u64, isize, end + 1);
462 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
463 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
464 nr_pages = min_t(unsigned long, nr_pages,
465 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
468 * we don't want to send crud past the end of i_size through
469 * compression, that's just a waste of CPU time. So, if the
470 * end of the file is before the start of our current
471 * requested range of bytes, we bail out to the uncompressed
472 * cleanup code that can deal with all of this.
474 * It isn't really the fastest way to fix things, but this is a
475 * very uncommon corner.
477 if (actual_end <= start)
478 goto cleanup_and_bail_uncompressed;
480 total_compressed = actual_end - start;
483 * skip compression for a small file range(<=blocksize) that
484 * isn't an inline extent, since it doesn't save disk space at all.
486 if (total_compressed <= blocksize &&
487 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
488 goto cleanup_and_bail_uncompressed;
490 total_compressed = min_t(unsigned long, total_compressed,
491 BTRFS_MAX_UNCOMPRESSED);
496 * we do compression for mount -o compress and when the
497 * inode has not been flagged as nocompress. This flag can
498 * change at any time if we discover bad compression ratios.
500 if (inode_need_compress(inode, start, end)) {
502 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
504 /* just bail out to the uncompressed code */
509 if (BTRFS_I(inode)->defrag_compress)
510 compress_type = BTRFS_I(inode)->defrag_compress;
511 else if (BTRFS_I(inode)->prop_compress)
512 compress_type = BTRFS_I(inode)->prop_compress;
515 * we need to call clear_page_dirty_for_io on each
516 * page in the range. Otherwise applications with the file
517 * mmap'd can wander in and change the page contents while
518 * we are compressing them.
520 * If the compression fails for any reason, we set the pages
521 * dirty again later on.
523 * Note that the remaining part is redirtied, the start pointer
524 * has moved, the end is the original one.
527 extent_range_clear_dirty_for_io(inode, start, end);
531 /* Compression level is applied here and only here */
532 ret = btrfs_compress_pages(
533 compress_type | (fs_info->compress_level << 4),
534 inode->i_mapping, start,
541 unsigned long offset = total_compressed &
543 struct page *page = pages[nr_pages - 1];
546 /* zero the tail end of the last page, we might be
547 * sending it down to disk
550 kaddr = kmap_atomic(page);
551 memset(kaddr + offset, 0,
553 kunmap_atomic(kaddr);
560 /* lets try to make an inline extent */
561 if (ret || total_in < actual_end) {
562 /* we didn't compress the entire range, try
563 * to make an uncompressed inline extent.
565 ret = cow_file_range_inline(inode, start, end, 0,
566 BTRFS_COMPRESS_NONE, NULL);
568 /* try making a compressed inline extent */
569 ret = cow_file_range_inline(inode, start, end,
571 compress_type, pages);
574 unsigned long clear_flags = EXTENT_DELALLOC |
575 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
576 EXTENT_DO_ACCOUNTING;
577 unsigned long page_error_op;
579 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
582 * inline extent creation worked or returned error,
583 * we don't need to create any more async work items.
584 * Unlock and free up our temp pages.
586 * We use DO_ACCOUNTING here because we need the
587 * delalloc_release_metadata to be done _after_ we drop
588 * our outstanding extent for clearing delalloc for this
591 extent_clear_unlock_delalloc(inode, start, end, end,
604 * we aren't doing an inline extent round the compressed size
605 * up to a block size boundary so the allocator does sane
608 total_compressed = ALIGN(total_compressed, blocksize);
611 * one last check to make sure the compression is really a
612 * win, compare the page count read with the blocks on disk,
613 * compression must free at least one sector size
615 total_in = ALIGN(total_in, PAGE_SIZE);
616 if (total_compressed + blocksize <= total_in) {
620 * The async work queues will take care of doing actual
621 * allocation on disk for these compressed pages, and
622 * will submit them to the elevator.
624 add_async_extent(async_cow, start, total_in,
625 total_compressed, pages, nr_pages,
628 if (start + total_in < end) {
639 * the compression code ran but failed to make things smaller,
640 * free any pages it allocated and our page pointer array
642 for (i = 0; i < nr_pages; i++) {
643 WARN_ON(pages[i]->mapping);
648 total_compressed = 0;
651 /* flag the file so we don't compress in the future */
652 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
653 !(BTRFS_I(inode)->prop_compress)) {
654 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
657 cleanup_and_bail_uncompressed:
659 * No compression, but we still need to write the pages in the file
660 * we've been given so far. redirty the locked page if it corresponds
661 * to our extent and set things up for the async work queue to run
662 * cow_file_range to do the normal delalloc dance.
664 if (page_offset(locked_page) >= start &&
665 page_offset(locked_page) <= end)
666 __set_page_dirty_nobuffers(locked_page);
667 /* unlocked later on in the async handlers */
670 extent_range_redirty_for_io(inode, start, end);
671 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
672 BTRFS_COMPRESS_NONE);
678 for (i = 0; i < nr_pages; i++) {
679 WARN_ON(pages[i]->mapping);
685 static void free_async_extent_pages(struct async_extent *async_extent)
689 if (!async_extent->pages)
692 for (i = 0; i < async_extent->nr_pages; i++) {
693 WARN_ON(async_extent->pages[i]->mapping);
694 put_page(async_extent->pages[i]);
696 kfree(async_extent->pages);
697 async_extent->nr_pages = 0;
698 async_extent->pages = NULL;
702 * phase two of compressed writeback. This is the ordered portion
703 * of the code, which only gets called in the order the work was
704 * queued. We walk all the async extents created by compress_file_range
705 * and send them down to the disk.
707 static noinline void submit_compressed_extents(struct inode *inode,
708 struct async_cow *async_cow)
710 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
711 struct async_extent *async_extent;
713 struct btrfs_key ins;
714 struct extent_map *em;
715 struct btrfs_root *root = BTRFS_I(inode)->root;
716 struct extent_io_tree *io_tree;
720 while (!list_empty(&async_cow->extents)) {
721 async_extent = list_entry(async_cow->extents.next,
722 struct async_extent, list);
723 list_del(&async_extent->list);
725 io_tree = &BTRFS_I(inode)->io_tree;
728 /* did the compression code fall back to uncompressed IO? */
729 if (!async_extent->pages) {
730 int page_started = 0;
731 unsigned long nr_written = 0;
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start +
735 async_extent->ram_size - 1);
737 /* allocate blocks */
738 ret = cow_file_range(inode, async_cow->locked_page,
740 async_extent->start +
741 async_extent->ram_size - 1,
742 async_extent->start +
743 async_extent->ram_size - 1,
744 &page_started, &nr_written, 0,
750 * if page_started, cow_file_range inserted an
751 * inline extent and took care of all the unlocking
752 * and IO for us. Otherwise, we need to submit
753 * all those pages down to the drive.
755 if (!page_started && !ret)
756 extent_write_locked_range(inode,
758 async_extent->start +
759 async_extent->ram_size - 1,
762 unlock_page(async_cow->locked_page);
768 lock_extent(io_tree, async_extent->start,
769 async_extent->start + async_extent->ram_size - 1);
771 ret = btrfs_reserve_extent(root, async_extent->ram_size,
772 async_extent->compressed_size,
773 async_extent->compressed_size,
774 0, alloc_hint, &ins, 1, 1);
776 free_async_extent_pages(async_extent);
778 if (ret == -ENOSPC) {
779 unlock_extent(io_tree, async_extent->start,
780 async_extent->start +
781 async_extent->ram_size - 1);
784 * we need to redirty the pages if we decide to
785 * fallback to uncompressed IO, otherwise we
786 * will not submit these pages down to lower
789 extent_range_redirty_for_io(inode,
791 async_extent->start +
792 async_extent->ram_size - 1);
799 * here we're doing allocation and writeback of the
802 em = create_io_em(inode, async_extent->start,
803 async_extent->ram_size, /* len */
804 async_extent->start, /* orig_start */
805 ins.objectid, /* block_start */
806 ins.offset, /* block_len */
807 ins.offset, /* orig_block_len */
808 async_extent->ram_size, /* ram_bytes */
809 async_extent->compress_type,
810 BTRFS_ORDERED_COMPRESSED);
812 /* ret value is not necessary due to void function */
813 goto out_free_reserve;
816 ret = btrfs_add_ordered_extent_compress(inode,
819 async_extent->ram_size,
821 BTRFS_ORDERED_COMPRESSED,
822 async_extent->compress_type);
824 btrfs_drop_extent_cache(BTRFS_I(inode),
826 async_extent->start +
827 async_extent->ram_size - 1, 0);
828 goto out_free_reserve;
830 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
833 * clear dirty, set writeback and unlock the pages.
835 extent_clear_unlock_delalloc(inode, async_extent->start,
836 async_extent->start +
837 async_extent->ram_size - 1,
838 async_extent->start +
839 async_extent->ram_size - 1,
840 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
841 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
843 if (btrfs_submit_compressed_write(inode,
845 async_extent->ram_size,
847 ins.offset, async_extent->pages,
848 async_extent->nr_pages,
849 async_cow->write_flags)) {
850 struct page *p = async_extent->pages[0];
851 const u64 start = async_extent->start;
852 const u64 end = start + async_extent->ram_size - 1;
854 p->mapping = inode->i_mapping;
855 btrfs_writepage_endio_finish_ordered(p, start, end,
859 extent_clear_unlock_delalloc(inode, start, end, end,
863 free_async_extent_pages(async_extent);
865 alloc_hint = ins.objectid + ins.offset;
871 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
872 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
874 extent_clear_unlock_delalloc(inode, async_extent->start,
875 async_extent->start +
876 async_extent->ram_size - 1,
877 async_extent->start +
878 async_extent->ram_size - 1,
879 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
880 EXTENT_DELALLOC_NEW |
881 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
882 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
883 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
885 free_async_extent_pages(async_extent);
890 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
893 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
894 struct extent_map *em;
897 read_lock(&em_tree->lock);
898 em = search_extent_mapping(em_tree, start, num_bytes);
901 * if block start isn't an actual block number then find the
902 * first block in this inode and use that as a hint. If that
903 * block is also bogus then just don't worry about it.
905 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
907 em = search_extent_mapping(em_tree, 0, 0);
908 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
909 alloc_hint = em->block_start;
913 alloc_hint = em->block_start;
917 read_unlock(&em_tree->lock);
923 * when extent_io.c finds a delayed allocation range in the file,
924 * the call backs end up in this code. The basic idea is to
925 * allocate extents on disk for the range, and create ordered data structs
926 * in ram to track those extents.
928 * locked_page is the page that writepage had locked already. We use
929 * it to make sure we don't do extra locks or unlocks.
931 * *page_started is set to one if we unlock locked_page and do everything
932 * required to start IO on it. It may be clean and already done with
935 static noinline int cow_file_range(struct inode *inode,
936 struct page *locked_page,
937 u64 start, u64 end, u64 delalloc_end,
938 int *page_started, unsigned long *nr_written,
939 int unlock, struct btrfs_dedupe_hash *hash)
941 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
942 struct btrfs_root *root = BTRFS_I(inode)->root;
945 unsigned long ram_size;
946 u64 cur_alloc_size = 0;
947 u64 blocksize = fs_info->sectorsize;
948 struct btrfs_key ins;
949 struct extent_map *em;
951 unsigned long page_ops;
952 bool extent_reserved = false;
955 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
961 num_bytes = ALIGN(end - start + 1, blocksize);
962 num_bytes = max(blocksize, num_bytes);
963 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
965 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
968 /* lets try to make an inline extent */
969 ret = cow_file_range_inline(inode, start, end, 0,
970 BTRFS_COMPRESS_NONE, NULL);
973 * We use DO_ACCOUNTING here because we need the
974 * delalloc_release_metadata to be run _after_ we drop
975 * our outstanding extent for clearing delalloc for this
978 extent_clear_unlock_delalloc(inode, start, end,
980 EXTENT_LOCKED | EXTENT_DELALLOC |
981 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
982 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
983 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
985 *nr_written = *nr_written +
986 (end - start + PAGE_SIZE) / PAGE_SIZE;
989 } else if (ret < 0) {
994 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
995 btrfs_drop_extent_cache(BTRFS_I(inode), start,
996 start + num_bytes - 1, 0);
998 while (num_bytes > 0) {
999 cur_alloc_size = num_bytes;
1000 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1001 fs_info->sectorsize, 0, alloc_hint,
1005 cur_alloc_size = ins.offset;
1006 extent_reserved = true;
1008 ram_size = ins.offset;
1009 em = create_io_em(inode, start, ins.offset, /* len */
1010 start, /* orig_start */
1011 ins.objectid, /* block_start */
1012 ins.offset, /* block_len */
1013 ins.offset, /* orig_block_len */
1014 ram_size, /* ram_bytes */
1015 BTRFS_COMPRESS_NONE, /* compress_type */
1016 BTRFS_ORDERED_REGULAR /* type */);
1021 free_extent_map(em);
1023 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1024 ram_size, cur_alloc_size, 0);
1026 goto out_drop_extent_cache;
1028 if (root->root_key.objectid ==
1029 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1030 ret = btrfs_reloc_clone_csums(inode, start,
1033 * Only drop cache here, and process as normal.
1035 * We must not allow extent_clear_unlock_delalloc()
1036 * at out_unlock label to free meta of this ordered
1037 * extent, as its meta should be freed by
1038 * btrfs_finish_ordered_io().
1040 * So we must continue until @start is increased to
1041 * skip current ordered extent.
1044 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1045 start + ram_size - 1, 0);
1048 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1050 /* we're not doing compressed IO, don't unlock the first
1051 * page (which the caller expects to stay locked), don't
1052 * clear any dirty bits and don't set any writeback bits
1054 * Do set the Private2 bit so we know this page was properly
1055 * setup for writepage
1057 page_ops = unlock ? PAGE_UNLOCK : 0;
1058 page_ops |= PAGE_SET_PRIVATE2;
1060 extent_clear_unlock_delalloc(inode, start,
1061 start + ram_size - 1,
1062 delalloc_end, locked_page,
1063 EXTENT_LOCKED | EXTENT_DELALLOC,
1065 if (num_bytes < cur_alloc_size)
1068 num_bytes -= cur_alloc_size;
1069 alloc_hint = ins.objectid + ins.offset;
1070 start += cur_alloc_size;
1071 extent_reserved = false;
1074 * btrfs_reloc_clone_csums() error, since start is increased
1075 * extent_clear_unlock_delalloc() at out_unlock label won't
1076 * free metadata of current ordered extent, we're OK to exit.
1084 out_drop_extent_cache:
1085 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1087 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1088 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1090 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1091 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1092 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1095 * If we reserved an extent for our delalloc range (or a subrange) and
1096 * failed to create the respective ordered extent, then it means that
1097 * when we reserved the extent we decremented the extent's size from
1098 * the data space_info's bytes_may_use counter and incremented the
1099 * space_info's bytes_reserved counter by the same amount. We must make
1100 * sure extent_clear_unlock_delalloc() does not try to decrement again
1101 * the data space_info's bytes_may_use counter, therefore we do not pass
1102 * it the flag EXTENT_CLEAR_DATA_RESV.
1104 if (extent_reserved) {
1105 extent_clear_unlock_delalloc(inode, start,
1106 start + cur_alloc_size,
1107 start + cur_alloc_size,
1111 start += cur_alloc_size;
1115 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1117 clear_bits | EXTENT_CLEAR_DATA_RESV,
1123 * work queue call back to started compression on a file and pages
1125 static noinline void async_cow_start(struct btrfs_work *work)
1127 struct async_cow *async_cow;
1129 async_cow = container_of(work, struct async_cow, work);
1131 compress_file_range(async_cow->inode, async_cow->locked_page,
1132 async_cow->start, async_cow->end, async_cow,
1134 if (num_added == 0) {
1135 btrfs_add_delayed_iput(async_cow->inode);
1136 async_cow->inode = NULL;
1141 * work queue call back to submit previously compressed pages
1143 static noinline void async_cow_submit(struct btrfs_work *work)
1145 struct btrfs_fs_info *fs_info;
1146 struct async_cow *async_cow;
1147 struct btrfs_root *root;
1148 unsigned long nr_pages;
1150 async_cow = container_of(work, struct async_cow, work);
1152 root = async_cow->root;
1153 fs_info = root->fs_info;
1154 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1157 /* atomic_sub_return implies a barrier */
1158 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1160 cond_wake_up_nomb(&fs_info->async_submit_wait);
1162 if (async_cow->inode)
1163 submit_compressed_extents(async_cow->inode, async_cow);
1166 static noinline void async_cow_free(struct btrfs_work *work)
1168 struct async_cow *async_cow;
1169 async_cow = container_of(work, struct async_cow, work);
1170 if (async_cow->inode)
1171 btrfs_add_delayed_iput(async_cow->inode);
1175 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1176 u64 start, u64 end, int *page_started,
1177 unsigned long *nr_written,
1178 unsigned int write_flags)
1180 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1181 struct async_cow *async_cow;
1182 struct btrfs_root *root = BTRFS_I(inode)->root;
1183 unsigned long nr_pages;
1186 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1188 while (start < end) {
1189 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1190 BUG_ON(!async_cow); /* -ENOMEM */
1191 async_cow->inode = igrab(inode);
1192 async_cow->root = root;
1193 async_cow->locked_page = locked_page;
1194 async_cow->start = start;
1195 async_cow->write_flags = write_flags;
1197 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1198 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1201 cur_end = min(end, start + SZ_512K - 1);
1203 async_cow->end = cur_end;
1204 INIT_LIST_HEAD(&async_cow->extents);
1206 btrfs_init_work(&async_cow->work,
1207 btrfs_delalloc_helper,
1208 async_cow_start, async_cow_submit,
1211 nr_pages = (cur_end - start + PAGE_SIZE) >>
1213 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1215 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1217 *nr_written += nr_pages;
1218 start = cur_end + 1;
1224 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1225 u64 bytenr, u64 num_bytes)
1228 struct btrfs_ordered_sum *sums;
1231 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1232 bytenr + num_bytes - 1, &list, 0);
1233 if (ret == 0 && list_empty(&list))
1236 while (!list_empty(&list)) {
1237 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1238 list_del(&sums->list);
1247 * when nowcow writeback call back. This checks for snapshots or COW copies
1248 * of the extents that exist in the file, and COWs the file as required.
1250 * If no cow copies or snapshots exist, we write directly to the existing
1253 static noinline int run_delalloc_nocow(struct inode *inode,
1254 struct page *locked_page,
1255 u64 start, u64 end, int *page_started, int force,
1256 unsigned long *nr_written)
1258 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1259 struct btrfs_root *root = BTRFS_I(inode)->root;
1260 struct extent_buffer *leaf;
1261 struct btrfs_path *path;
1262 struct btrfs_file_extent_item *fi;
1263 struct btrfs_key found_key;
1264 struct extent_map *em;
1279 u64 ino = btrfs_ino(BTRFS_I(inode));
1281 path = btrfs_alloc_path();
1283 extent_clear_unlock_delalloc(inode, start, end, end,
1285 EXTENT_LOCKED | EXTENT_DELALLOC |
1286 EXTENT_DO_ACCOUNTING |
1287 EXTENT_DEFRAG, PAGE_UNLOCK |
1289 PAGE_SET_WRITEBACK |
1290 PAGE_END_WRITEBACK);
1294 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1296 cow_start = (u64)-1;
1299 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1303 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1304 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &found_key,
1306 path->slots[0] - 1);
1307 if (found_key.objectid == ino &&
1308 found_key.type == BTRFS_EXTENT_DATA_KEY)
1313 leaf = path->nodes[0];
1314 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1315 ret = btrfs_next_leaf(root, path);
1317 if (cow_start != (u64)-1)
1318 cur_offset = cow_start;
1323 leaf = path->nodes[0];
1329 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1331 if (found_key.objectid > ino)
1333 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1334 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1338 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1339 found_key.offset > end)
1342 if (found_key.offset > cur_offset) {
1343 extent_end = found_key.offset;
1348 fi = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_file_extent_item);
1350 extent_type = btrfs_file_extent_type(leaf, fi);
1352 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1353 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1354 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1355 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1356 extent_offset = btrfs_file_extent_offset(leaf, fi);
1357 extent_end = found_key.offset +
1358 btrfs_file_extent_num_bytes(leaf, fi);
1360 btrfs_file_extent_disk_num_bytes(leaf, fi);
1361 if (extent_end <= start) {
1365 if (disk_bytenr == 0)
1367 if (btrfs_file_extent_compression(leaf, fi) ||
1368 btrfs_file_extent_encryption(leaf, fi) ||
1369 btrfs_file_extent_other_encoding(leaf, fi))
1372 * Do the same check as in btrfs_cross_ref_exist but
1373 * without the unnecessary search.
1375 if (btrfs_file_extent_generation(leaf, fi) <=
1376 btrfs_root_last_snapshot(&root->root_item))
1378 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1380 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1382 ret = btrfs_cross_ref_exist(root, ino,
1384 extent_offset, disk_bytenr);
1387 * ret could be -EIO if the above fails to read
1391 if (cow_start != (u64)-1)
1392 cur_offset = cow_start;
1396 WARN_ON_ONCE(nolock);
1399 disk_bytenr += extent_offset;
1400 disk_bytenr += cur_offset - found_key.offset;
1401 num_bytes = min(end + 1, extent_end) - cur_offset;
1403 * if there are pending snapshots for this root,
1404 * we fall into common COW way.
1406 if (!nolock && atomic_read(&root->snapshot_force_cow))
1409 * force cow if csum exists in the range.
1410 * this ensure that csum for a given extent are
1411 * either valid or do not exist.
1413 ret = csum_exist_in_range(fs_info, disk_bytenr,
1417 * ret could be -EIO if the above fails to read
1421 if (cow_start != (u64)-1)
1422 cur_offset = cow_start;
1425 WARN_ON_ONCE(nolock);
1428 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1431 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1432 extent_end = found_key.offset +
1433 btrfs_file_extent_ram_bytes(leaf, fi);
1434 extent_end = ALIGN(extent_end,
1435 fs_info->sectorsize);
1440 if (extent_end <= start) {
1443 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1447 if (cow_start == (u64)-1)
1448 cow_start = cur_offset;
1449 cur_offset = extent_end;
1450 if (cur_offset > end)
1456 btrfs_release_path(path);
1457 if (cow_start != (u64)-1) {
1458 ret = cow_file_range(inode, locked_page,
1459 cow_start, found_key.offset - 1,
1460 end, page_started, nr_written, 1,
1464 btrfs_dec_nocow_writers(fs_info,
1468 cow_start = (u64)-1;
1471 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1472 u64 orig_start = found_key.offset - extent_offset;
1474 em = create_io_em(inode, cur_offset, num_bytes,
1476 disk_bytenr, /* block_start */
1477 num_bytes, /* block_len */
1478 disk_num_bytes, /* orig_block_len */
1479 ram_bytes, BTRFS_COMPRESS_NONE,
1480 BTRFS_ORDERED_PREALLOC);
1483 btrfs_dec_nocow_writers(fs_info,
1488 free_extent_map(em);
1491 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1492 type = BTRFS_ORDERED_PREALLOC;
1494 type = BTRFS_ORDERED_NOCOW;
1497 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1498 num_bytes, num_bytes, type);
1500 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1501 BUG_ON(ret); /* -ENOMEM */
1503 if (root->root_key.objectid ==
1504 BTRFS_DATA_RELOC_TREE_OBJECTID)
1506 * Error handled later, as we must prevent
1507 * extent_clear_unlock_delalloc() in error handler
1508 * from freeing metadata of created ordered extent.
1510 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1513 extent_clear_unlock_delalloc(inode, cur_offset,
1514 cur_offset + num_bytes - 1, end,
1515 locked_page, EXTENT_LOCKED |
1517 EXTENT_CLEAR_DATA_RESV,
1518 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1520 cur_offset = extent_end;
1523 * btrfs_reloc_clone_csums() error, now we're OK to call error
1524 * handler, as metadata for created ordered extent will only
1525 * be freed by btrfs_finish_ordered_io().
1529 if (cur_offset > end)
1532 btrfs_release_path(path);
1534 if (cur_offset <= end && cow_start == (u64)-1)
1535 cow_start = cur_offset;
1537 if (cow_start != (u64)-1) {
1539 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1540 page_started, nr_written, 1, NULL);
1546 if (ret && cur_offset < end)
1547 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1548 locked_page, EXTENT_LOCKED |
1549 EXTENT_DELALLOC | EXTENT_DEFRAG |
1550 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1552 PAGE_SET_WRITEBACK |
1553 PAGE_END_WRITEBACK);
1554 btrfs_free_path(path);
1558 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1561 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1562 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1566 * @defrag_bytes is a hint value, no spinlock held here,
1567 * if is not zero, it means the file is defragging.
1568 * Force cow if given extent needs to be defragged.
1570 if (BTRFS_I(inode)->defrag_bytes &&
1571 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1572 EXTENT_DEFRAG, 0, NULL))
1579 * Function to process delayed allocation (create CoW) for ranges which are
1580 * being touched for the first time.
1582 int btrfs_run_delalloc_range(void *private_data, struct page *locked_page,
1583 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1584 struct writeback_control *wbc)
1586 struct inode *inode = private_data;
1588 int force_cow = need_force_cow(inode, start, end);
1589 unsigned int write_flags = wbc_to_write_flags(wbc);
1591 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1592 ret = run_delalloc_nocow(inode, locked_page, start, end,
1593 page_started, 1, nr_written);
1594 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1595 ret = run_delalloc_nocow(inode, locked_page, start, end,
1596 page_started, 0, nr_written);
1597 } else if (!inode_need_compress(inode, start, end)) {
1598 ret = cow_file_range(inode, locked_page, start, end, end,
1599 page_started, nr_written, 1, NULL);
1601 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1602 &BTRFS_I(inode)->runtime_flags);
1603 ret = cow_file_range_async(inode, locked_page, start, end,
1604 page_started, nr_written,
1608 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1612 static void btrfs_split_extent_hook(void *private_data,
1613 struct extent_state *orig, u64 split)
1615 struct inode *inode = private_data;
1618 /* not delalloc, ignore it */
1619 if (!(orig->state & EXTENT_DELALLOC))
1622 size = orig->end - orig->start + 1;
1623 if (size > BTRFS_MAX_EXTENT_SIZE) {
1628 * See the explanation in btrfs_merge_extent_hook, the same
1629 * applies here, just in reverse.
1631 new_size = orig->end - split + 1;
1632 num_extents = count_max_extents(new_size);
1633 new_size = split - orig->start;
1634 num_extents += count_max_extents(new_size);
1635 if (count_max_extents(size) >= num_extents)
1639 spin_lock(&BTRFS_I(inode)->lock);
1640 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1641 spin_unlock(&BTRFS_I(inode)->lock);
1645 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1646 * extents so we can keep track of new extents that are just merged onto old
1647 * extents, such as when we are doing sequential writes, so we can properly
1648 * account for the metadata space we'll need.
1650 static void btrfs_merge_extent_hook(void *private_data,
1651 struct extent_state *new,
1652 struct extent_state *other)
1654 struct inode *inode = private_data;
1655 u64 new_size, old_size;
1658 /* not delalloc, ignore it */
1659 if (!(other->state & EXTENT_DELALLOC))
1662 if (new->start > other->start)
1663 new_size = new->end - other->start + 1;
1665 new_size = other->end - new->start + 1;
1667 /* we're not bigger than the max, unreserve the space and go */
1668 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1669 spin_lock(&BTRFS_I(inode)->lock);
1670 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1671 spin_unlock(&BTRFS_I(inode)->lock);
1676 * We have to add up either side to figure out how many extents were
1677 * accounted for before we merged into one big extent. If the number of
1678 * extents we accounted for is <= the amount we need for the new range
1679 * then we can return, otherwise drop. Think of it like this
1683 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1684 * need 2 outstanding extents, on one side we have 1 and the other side
1685 * we have 1 so they are == and we can return. But in this case
1687 * [MAX_SIZE+4k][MAX_SIZE+4k]
1689 * Each range on their own accounts for 2 extents, but merged together
1690 * they are only 3 extents worth of accounting, so we need to drop in
1693 old_size = other->end - other->start + 1;
1694 num_extents = count_max_extents(old_size);
1695 old_size = new->end - new->start + 1;
1696 num_extents += count_max_extents(old_size);
1697 if (count_max_extents(new_size) >= num_extents)
1700 spin_lock(&BTRFS_I(inode)->lock);
1701 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1702 spin_unlock(&BTRFS_I(inode)->lock);
1705 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1706 struct inode *inode)
1708 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1710 spin_lock(&root->delalloc_lock);
1711 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1712 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1713 &root->delalloc_inodes);
1714 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1715 &BTRFS_I(inode)->runtime_flags);
1716 root->nr_delalloc_inodes++;
1717 if (root->nr_delalloc_inodes == 1) {
1718 spin_lock(&fs_info->delalloc_root_lock);
1719 BUG_ON(!list_empty(&root->delalloc_root));
1720 list_add_tail(&root->delalloc_root,
1721 &fs_info->delalloc_roots);
1722 spin_unlock(&fs_info->delalloc_root_lock);
1725 spin_unlock(&root->delalloc_lock);
1729 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1730 struct btrfs_inode *inode)
1732 struct btrfs_fs_info *fs_info = root->fs_info;
1734 if (!list_empty(&inode->delalloc_inodes)) {
1735 list_del_init(&inode->delalloc_inodes);
1736 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737 &inode->runtime_flags);
1738 root->nr_delalloc_inodes--;
1739 if (!root->nr_delalloc_inodes) {
1740 ASSERT(list_empty(&root->delalloc_inodes));
1741 spin_lock(&fs_info->delalloc_root_lock);
1742 BUG_ON(list_empty(&root->delalloc_root));
1743 list_del_init(&root->delalloc_root);
1744 spin_unlock(&fs_info->delalloc_root_lock);
1749 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1750 struct btrfs_inode *inode)
1752 spin_lock(&root->delalloc_lock);
1753 __btrfs_del_delalloc_inode(root, inode);
1754 spin_unlock(&root->delalloc_lock);
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.
1762 static void btrfs_set_bit_hook(void *private_data,
1763 struct extent_state *state, unsigned *bits)
1765 struct inode *inode = private_data;
1767 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1769 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
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
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));
1782 spin_lock(&BTRFS_I(inode)->lock);
1783 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1784 spin_unlock(&BTRFS_I(inode)->lock);
1786 /* For sanity tests */
1787 if (btrfs_is_testing(fs_info))
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);
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 -
1807 spin_unlock(&BTRFS_I(inode)->lock);
1812 * extent_io.c clear_bit_hook, see set_bit_hook for why
1814 static void btrfs_clear_bit_hook(void *private_data,
1815 struct extent_state *state,
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);
1823 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1824 spin_lock(&inode->lock);
1825 inode->defrag_bytes -= len;
1826 spin_unlock(&inode->lock);
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
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);
1838 spin_lock(&inode->lock);
1839 btrfs_mod_outstanding_extents(inode, -num_extents);
1840 spin_unlock(&inode->lock);
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
1847 if (*bits & EXTENT_CLEAR_META_RESV &&
1848 root != fs_info->tree_root)
1849 btrfs_delalloc_release_metadata(inode, len, false);
1851 /* For sanity tests. */
1852 if (btrfs_is_testing(fs_info))
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(
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);
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);
1883 * Merge bio hook, this must check the chunk tree to make sure we don't create
1884 * bios that span stripes or chunks
1886 * return 1 if page cannot be merged to bio
1887 * return 0 if page can be merged to bio
1888 * return error otherwise
1890 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1891 size_t size, struct bio *bio,
1892 unsigned long bio_flags)
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;
1901 if (bio_flags & EXTENT_BIO_COMPRESSED)
1904 length = bio->bi_iter.bi_size;
1905 map_length = length;
1906 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1910 if (map_length < length + size)
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.
1920 * At IO completion time the cums attached on the ordered extent record
1921 * are inserted into the btree
1923 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1926 struct inode *inode = private_data;
1927 blk_status_t ret = 0;
1929 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1930 BUG_ON(ret); /* -ENOMEM */
1935 * in order to insert checksums into the metadata in large chunks,
1936 * we wait until bio submission time. All the pages in the bio are
1937 * checksummed and sums are attached onto the ordered extent record.
1939 * At IO completion time the cums attached on the ordered extent record
1940 * are inserted into the btree
1942 blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1945 struct inode *inode = private_data;
1946 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1949 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1951 bio->bi_status = ret;
1958 * extent_io.c submission hook. This does the right thing for csum calculation
1959 * on write, or reading the csums from the tree before a read.
1961 * Rules about async/sync submit,
1962 * a) read: sync submit
1964 * b) write without checksum: sync submit
1966 * c) write with checksum:
1967 * c-1) if bio is issued by fsync: sync submit
1968 * (sync_writers != 0)
1970 * c-2) if root is reloc root: sync submit
1971 * (only in case of buffered IO)
1973 * c-3) otherwise: async submit
1975 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1976 int mirror_num, unsigned long bio_flags,
1979 struct inode *inode = private_data;
1980 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1981 struct btrfs_root *root = BTRFS_I(inode)->root;
1982 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1983 blk_status_t ret = 0;
1985 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1987 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1989 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1990 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1992 if (bio_op(bio) != REQ_OP_WRITE) {
1993 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1997 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1998 ret = btrfs_submit_compressed_read(inode, bio,
2002 } else if (!skip_sum) {
2003 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2008 } else if (async && !skip_sum) {
2009 /* csum items have already been cloned */
2010 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2012 /* we're doing a write, do the async checksumming */
2013 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2015 btrfs_submit_bio_start);
2017 } else if (!skip_sum) {
2018 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2024 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2028 bio->bi_status = ret;
2035 * given a list of ordered sums record them in the inode. This happens
2036 * at IO completion time based on sums calculated at bio submission time.
2038 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2039 struct inode *inode, struct list_head *list)
2041 struct btrfs_ordered_sum *sum;
2044 list_for_each_entry(sum, list, list) {
2045 trans->adding_csums = true;
2046 ret = btrfs_csum_file_blocks(trans,
2047 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2048 trans->adding_csums = false;
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)
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);
2064 /* see btrfs_writepage_start_hook for details on why this is required */
2065 struct btrfs_writepage_fixup {
2067 struct btrfs_work work;
2070 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
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;
2077 struct inode *inode;
2082 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2086 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2087 ClearPageChecked(page);
2091 inode = page->mapping->host;
2092 page_start = page_offset(page);
2093 page_end = page_offset(page) + PAGE_SIZE - 1;
2095 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2098 /* already ordered? We're done */
2099 if (PagePrivate2(page))
2102 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2105 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2106 page_end, &cached_state);
2108 btrfs_start_ordered_extent(inode, ordered, 1);
2109 btrfs_put_ordered_extent(ordered);
2113 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2116 mapping_set_error(page->mapping, ret);
2117 end_extent_writepage(page, ret, page_start, page_end);
2118 ClearPageChecked(page);
2122 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2125 mapping_set_error(page->mapping, ret);
2126 end_extent_writepage(page, ret, page_start, page_end);
2127 ClearPageChecked(page);
2131 ClearPageChecked(page);
2132 set_page_dirty(page);
2133 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2135 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2141 extent_changeset_free(data_reserved);
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.
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.
2155 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
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;
2161 /* this page is properly in the ordered list */
2162 if (TestClearPagePrivate2(page))
2165 if (PageChecked(page))
2168 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2172 SetPageChecked(page);
2174 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2175 btrfs_writepage_fixup_worker, NULL, NULL);
2177 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
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)
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;
2194 int extent_inserted = 0;
2197 path = btrfs_alloc_path();
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.
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
2210 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2211 file_pos + num_bytes, NULL, 0,
2212 1, sizeof(*fi), &extent_inserted);
2216 if (!extent_inserted) {
2217 ins.objectid = btrfs_ino(BTRFS_I(inode));
2218 ins.offset = file_pos;
2219 ins.type = BTRFS_EXTENT_DATA_KEY;
2221 path->leave_spinning = 1;
2222 ret = btrfs_insert_empty_item(trans, root, path, &ins,
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);
2241 btrfs_mark_buffer_dirty(leaf);
2242 btrfs_release_path(path);
2244 inode_add_bytes(inode, num_bytes);
2246 ins.objectid = disk_bytenr;
2247 ins.offset = disk_num_bytes;
2248 ins.type = BTRFS_EXTENT_ITEM_KEY;
2251 * Release the reserved range from inode dirty range map, as it is
2252 * already moved into delayed_ref_head
2254 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2258 ret = btrfs_alloc_reserved_file_extent(trans, root,
2259 btrfs_ino(BTRFS_I(inode)),
2260 file_pos, qg_released, &ins);
2262 btrfs_free_path(path);
2267 /* snapshot-aware defrag */
2268 struct sa_defrag_extent_backref {
2269 struct rb_node node;
2270 struct old_sa_defrag_extent *old;
2279 struct old_sa_defrag_extent {
2280 struct list_head list;
2281 struct new_sa_defrag_extent *new;
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;
2302 static int backref_comp(struct sa_defrag_extent_backref *b1,
2303 struct sa_defrag_extent_backref *b2)
2305 if (b1->root_id < b2->root_id)
2307 else if (b1->root_id > b2->root_id)
2310 if (b1->inum < b2->inum)
2312 else if (b1->inum > b2->inum)
2315 if (b1->file_pos < b2->file_pos)
2317 else if (b1->file_pos > b2->file_pos)
2321 * [------------------------------] ===> (a range of space)
2322 * |<--->| |<---->| =============> (fs/file tree A)
2323 * |<---------------------------->| ===> (fs/file tree B)
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.
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).
2335 static void backref_insert(struct rb_root *root,
2336 struct sa_defrag_extent_backref *backref)
2338 struct rb_node **p = &root->rb_node;
2339 struct rb_node *parent = NULL;
2340 struct sa_defrag_extent_backref *entry;
2345 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2347 ret = backref_comp(backref, entry);
2351 p = &(*p)->rb_right;
2354 rb_link_node(&backref->node, parent, p);
2355 rb_insert_color(&backref->node, root);
2359 * Note the backref might has changed, and in this case we just return 0.
2361 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
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);
2379 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2380 inum == btrfs_ino(BTRFS_I(inode)))
2383 key.objectid = root_id;
2384 key.type = BTRFS_ROOT_ITEM_KEY;
2385 key.offset = (u64)-1;
2387 root = btrfs_read_fs_root_no_name(fs_info, &key);
2389 if (PTR_ERR(root) == -ENOENT)
2392 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2393 inum, offset, root_id);
2394 return PTR_ERR(root);
2397 key.objectid = inum;
2398 key.type = BTRFS_EXTENT_DATA_KEY;
2399 if (offset > (u64)-1 << 32)
2402 key.offset = offset;
2404 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2405 if (WARN_ON(ret < 0))
2412 leaf = path->nodes[0];
2413 slot = path->slots[0];
2415 if (slot >= btrfs_header_nritems(leaf)) {
2416 ret = btrfs_next_leaf(root, path);
2419 } else if (ret > 0) {
2428 btrfs_item_key_to_cpu(leaf, &key, slot);
2430 if (key.objectid > inum)
2433 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2436 extent = btrfs_item_ptr(leaf, slot,
2437 struct btrfs_file_extent_item);
2439 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2443 * 'offset' refers to the exact key.offset,
2444 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2445 * (key.offset - extent_offset).
2447 if (key.offset != offset)
2450 extent_offset = btrfs_file_extent_offset(leaf, extent);
2451 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2453 if (extent_offset >= old->extent_offset + old->offset +
2454 old->len || extent_offset + num_bytes <=
2455 old->extent_offset + old->offset)
2460 backref = kmalloc(sizeof(*backref), GFP_NOFS);
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);
2473 backref_insert(&new->root, backref);
2476 btrfs_release_path(path);
2481 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2482 struct new_sa_defrag_extent *new)
2484 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2485 struct old_sa_defrag_extent *old, *tmp;
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,
2495 if (ret < 0 && ret != -ENOENT)
2498 /* no backref to be processed for this extent */
2500 list_del(&old->list);
2505 if (list_empty(&new->head))
2511 static int relink_is_mergable(struct extent_buffer *leaf,
2512 struct btrfs_file_extent_item *fi,
2513 struct new_sa_defrag_extent *new)
2515 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2518 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2521 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2524 if (btrfs_file_extent_encryption(leaf, fi) ||
2525 btrfs_file_extent_other_encoding(leaf, fi))
2532 * Note the backref might has changed, and in this case we just return 0.
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)
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;
2558 if (prev && prev->root_id == backref->root_id &&
2559 prev->inum == backref->inum &&
2560 prev->file_pos + prev->num_bytes == backref->file_pos)
2563 /* step 1: get root */
2564 key.objectid = backref->root_id;
2565 key.type = BTRFS_ROOT_ITEM_KEY;
2566 key.offset = (u64)-1;
2568 index = srcu_read_lock(&fs_info->subvol_srcu);
2570 root = btrfs_read_fs_root_no_name(fs_info, &key);
2572 srcu_read_unlock(&fs_info->subvol_srcu, index);
2573 if (PTR_ERR(root) == -ENOENT)
2575 return PTR_ERR(root);
2578 if (btrfs_root_readonly(root)) {
2579 srcu_read_unlock(&fs_info->subvol_srcu, index);
2583 /* step 2: get inode */
2584 key.objectid = backref->inum;
2585 key.type = BTRFS_INODE_ITEM_KEY;
2588 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2589 if (IS_ERR(inode)) {
2590 srcu_read_unlock(&fs_info->subvol_srcu, index);
2594 srcu_read_unlock(&fs_info->subvol_srcu, index);
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,
2602 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2604 btrfs_put_ordered_extent(ordered);
2608 trans = btrfs_join_transaction(root);
2609 if (IS_ERR(trans)) {
2610 ret = PTR_ERR(trans);
2614 key.objectid = backref->inum;
2615 key.type = BTRFS_EXTENT_DATA_KEY;
2616 key.offset = backref->file_pos;
2618 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2621 } else if (ret > 0) {
2626 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2627 struct btrfs_file_extent_item);
2629 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2630 backref->generation)
2633 btrfs_release_path(path);
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;
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);
2644 ret = btrfs_drop_extents(trans, root, inode, start,
2649 key.objectid = btrfs_ino(BTRFS_I(inode));
2650 key.type = BTRFS_EXTENT_DATA_KEY;
2653 path->leave_spinning = 1;
2655 struct btrfs_file_extent_item *fi;
2657 struct btrfs_key found_key;
2659 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2664 leaf = path->nodes[0];
2665 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
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);
2671 if (extent_len + found_key.offset == start &&
2672 relink_is_mergable(leaf, fi, new)) {
2673 btrfs_set_file_extent_num_bytes(leaf, fi,
2675 btrfs_mark_buffer_dirty(leaf);
2676 inode_add_bytes(inode, len);
2682 btrfs_release_path(path);
2687 ret = btrfs_insert_empty_item(trans, root, path, &key,
2690 btrfs_abort_transaction(trans, ret);
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);
2708 btrfs_mark_buffer_dirty(leaf);
2709 inode_add_bytes(inode, len);
2710 btrfs_release_path(path);
2712 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2714 backref->root_id, backref->inum,
2715 new->file_pos); /* start - extent_offset */
2717 btrfs_abort_transaction(trans, ret);
2723 btrfs_release_path(path);
2724 path->leave_spinning = 0;
2725 btrfs_end_transaction(trans);
2727 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2733 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2735 struct old_sa_defrag_extent *old, *tmp;
2740 list_for_each_entry_safe(old, tmp, &new->head, list) {
2746 static void relink_file_extents(struct new_sa_defrag_extent *new)
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 rb_node *node;
2755 path = btrfs_alloc_path();
2759 if (!record_extent_backrefs(path, new)) {
2760 btrfs_free_path(path);
2763 btrfs_release_path(path);
2766 node = rb_first(&new->root);
2769 rb_erase(node, &new->root);
2771 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2773 ret = relink_extent_backref(path, prev, backref);
2786 btrfs_free_path(path);
2788 free_sa_defrag_extent(new);
2790 atomic_dec(&fs_info->defrag_running);
2791 wake_up(&fs_info->transaction_wait);
2794 static struct new_sa_defrag_extent *
2795 record_old_file_extents(struct inode *inode,
2796 struct btrfs_ordered_extent *ordered)
2798 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2799 struct btrfs_root *root = BTRFS_I(inode)->root;
2800 struct btrfs_path *path;
2801 struct btrfs_key key;
2802 struct old_sa_defrag_extent *old;
2803 struct new_sa_defrag_extent *new;
2806 new = kmalloc(sizeof(*new), GFP_NOFS);
2811 new->file_pos = ordered->file_offset;
2812 new->len = ordered->len;
2813 new->bytenr = ordered->start;
2814 new->disk_len = ordered->disk_len;
2815 new->compress_type = ordered->compress_type;
2816 new->root = RB_ROOT;
2817 INIT_LIST_HEAD(&new->head);
2819 path = btrfs_alloc_path();
2823 key.objectid = btrfs_ino(BTRFS_I(inode));
2824 key.type = BTRFS_EXTENT_DATA_KEY;
2825 key.offset = new->file_pos;
2827 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2830 if (ret > 0 && path->slots[0] > 0)
2833 /* find out all the old extents for the file range */
2835 struct btrfs_file_extent_item *extent;
2836 struct extent_buffer *l;
2845 slot = path->slots[0];
2847 if (slot >= btrfs_header_nritems(l)) {
2848 ret = btrfs_next_leaf(root, path);
2856 btrfs_item_key_to_cpu(l, &key, slot);
2858 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2860 if (key.type != BTRFS_EXTENT_DATA_KEY)
2862 if (key.offset >= new->file_pos + new->len)
2865 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2867 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2868 if (key.offset + num_bytes < new->file_pos)
2871 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2875 extent_offset = btrfs_file_extent_offset(l, extent);
2877 old = kmalloc(sizeof(*old), GFP_NOFS);
2881 offset = max(new->file_pos, key.offset);
2882 end = min(new->file_pos + new->len, key.offset + num_bytes);
2884 old->bytenr = disk_bytenr;
2885 old->extent_offset = extent_offset;
2886 old->offset = offset - key.offset;
2887 old->len = end - offset;
2890 list_add_tail(&old->list, &new->head);
2896 btrfs_free_path(path);
2897 atomic_inc(&fs_info->defrag_running);
2902 btrfs_free_path(path);
2904 free_sa_defrag_extent(new);
2908 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2911 struct btrfs_block_group_cache *cache;
2913 cache = btrfs_lookup_block_group(fs_info, start);
2916 spin_lock(&cache->lock);
2917 cache->delalloc_bytes -= len;
2918 spin_unlock(&cache->lock);
2920 btrfs_put_block_group(cache);
2923 /* as ordered data IO finishes, this gets called so we can finish
2924 * an ordered extent if the range of bytes in the file it covers are
2927 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2929 struct inode *inode = ordered_extent->inode;
2930 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2931 struct btrfs_root *root = BTRFS_I(inode)->root;
2932 struct btrfs_trans_handle *trans = NULL;
2933 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2934 struct extent_state *cached_state = NULL;
2935 struct new_sa_defrag_extent *new = NULL;
2936 int compress_type = 0;
2938 u64 logical_len = ordered_extent->len;
2940 bool truncated = false;
2941 bool range_locked = false;
2942 bool clear_new_delalloc_bytes = false;
2943 bool clear_reserved_extent = true;
2945 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2946 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2947 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2948 clear_new_delalloc_bytes = true;
2950 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2952 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2957 btrfs_free_io_failure_record(BTRFS_I(inode),
2958 ordered_extent->file_offset,
2959 ordered_extent->file_offset +
2960 ordered_extent->len - 1);
2962 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2964 logical_len = ordered_extent->truncated_len;
2965 /* Truncated the entire extent, don't bother adding */
2970 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2971 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2974 * For mwrite(mmap + memset to write) case, we still reserve
2975 * space for NOCOW range.
2976 * As NOCOW won't cause a new delayed ref, just free the space
2978 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2979 ordered_extent->len);
2980 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2982 trans = btrfs_join_transaction_nolock(root);
2984 trans = btrfs_join_transaction(root);
2985 if (IS_ERR(trans)) {
2986 ret = PTR_ERR(trans);
2990 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2991 ret = btrfs_update_inode_fallback(trans, root, inode);
2992 if (ret) /* -ENOMEM or corruption */
2993 btrfs_abort_transaction(trans, ret);
2997 range_locked = true;
2998 lock_extent_bits(io_tree, ordered_extent->file_offset,
2999 ordered_extent->file_offset + ordered_extent->len - 1,
3002 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3003 ordered_extent->file_offset + ordered_extent->len - 1,
3004 EXTENT_DEFRAG, 0, cached_state);
3006 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3007 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3008 /* the inode is shared */
3009 new = record_old_file_extents(inode, ordered_extent);
3011 clear_extent_bit(io_tree, ordered_extent->file_offset,
3012 ordered_extent->file_offset + ordered_extent->len - 1,
3013 EXTENT_DEFRAG, 0, 0, &cached_state);
3017 trans = btrfs_join_transaction_nolock(root);
3019 trans = btrfs_join_transaction(root);
3020 if (IS_ERR(trans)) {
3021 ret = PTR_ERR(trans);
3026 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3028 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3029 compress_type = ordered_extent->compress_type;
3030 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3031 BUG_ON(compress_type);
3032 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3033 ordered_extent->len);
3034 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3035 ordered_extent->file_offset,
3036 ordered_extent->file_offset +
3039 BUG_ON(root == fs_info->tree_root);
3040 ret = insert_reserved_file_extent(trans, inode,
3041 ordered_extent->file_offset,
3042 ordered_extent->start,
3043 ordered_extent->disk_len,
3044 logical_len, logical_len,
3045 compress_type, 0, 0,
3046 BTRFS_FILE_EXTENT_REG);
3048 clear_reserved_extent = false;
3049 btrfs_release_delalloc_bytes(fs_info,
3050 ordered_extent->start,
3051 ordered_extent->disk_len);
3054 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3055 ordered_extent->file_offset, ordered_extent->len,
3058 btrfs_abort_transaction(trans, ret);
3062 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3064 btrfs_abort_transaction(trans, ret);
3068 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3069 ret = btrfs_update_inode_fallback(trans, root, inode);
3070 if (ret) { /* -ENOMEM or corruption */
3071 btrfs_abort_transaction(trans, ret);
3076 if (range_locked || clear_new_delalloc_bytes) {
3077 unsigned int clear_bits = 0;
3080 clear_bits |= EXTENT_LOCKED;
3081 if (clear_new_delalloc_bytes)
3082 clear_bits |= EXTENT_DELALLOC_NEW;
3083 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3084 ordered_extent->file_offset,
3085 ordered_extent->file_offset +
3086 ordered_extent->len - 1,
3088 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3093 btrfs_end_transaction(trans);
3095 if (ret || truncated) {
3099 start = ordered_extent->file_offset + logical_len;
3101 start = ordered_extent->file_offset;
3102 end = ordered_extent->file_offset + ordered_extent->len - 1;
3103 clear_extent_uptodate(io_tree, start, end, NULL);
3105 /* Drop the cache for the part of the extent we didn't write. */
3106 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3109 * If the ordered extent had an IOERR or something else went
3110 * wrong we need to return the space for this ordered extent
3111 * back to the allocator. We only free the extent in the
3112 * truncated case if we didn't write out the extent at all.
3114 * If we made it past insert_reserved_file_extent before we
3115 * errored out then we don't need to do this as the accounting
3116 * has already been done.
3118 if ((ret || !logical_len) &&
3119 clear_reserved_extent &&
3120 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3121 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3122 btrfs_free_reserved_extent(fs_info,
3123 ordered_extent->start,
3124 ordered_extent->disk_len, 1);
3129 * This needs to be done to make sure anybody waiting knows we are done
3130 * updating everything for this ordered extent.
3132 btrfs_remove_ordered_extent(inode, ordered_extent);
3134 /* for snapshot-aware defrag */
3137 free_sa_defrag_extent(new);
3138 atomic_dec(&fs_info->defrag_running);
3140 relink_file_extents(new);
3145 btrfs_put_ordered_extent(ordered_extent);
3146 /* once for the tree */
3147 btrfs_put_ordered_extent(ordered_extent);
3149 /* Try to release some metadata so we don't get an OOM but don't wait */
3150 btrfs_btree_balance_dirty_nodelay(fs_info);
3155 static void finish_ordered_fn(struct btrfs_work *work)
3157 struct btrfs_ordered_extent *ordered_extent;
3158 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3159 btrfs_finish_ordered_io(ordered_extent);
3162 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start, u64 end,
3163 struct extent_state *state, int uptodate)
3165 struct inode *inode = page->mapping->host;
3166 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3167 struct btrfs_ordered_extent *ordered_extent = NULL;
3168 struct btrfs_workqueue *wq;
3169 btrfs_work_func_t func;
3171 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3173 ClearPagePrivate2(page);
3174 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3175 end - start + 1, uptodate))
3178 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3179 wq = fs_info->endio_freespace_worker;
3180 func = btrfs_freespace_write_helper;
3182 wq = fs_info->endio_write_workers;
3183 func = btrfs_endio_write_helper;
3186 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3188 btrfs_queue_work(wq, &ordered_extent->work);
3191 static int __readpage_endio_check(struct inode *inode,
3192 struct btrfs_io_bio *io_bio,
3193 int icsum, struct page *page,
3194 int pgoff, u64 start, size_t len)
3200 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3202 kaddr = kmap_atomic(page);
3203 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3204 btrfs_csum_final(csum, (u8 *)&csum);
3205 if (csum != csum_expected)
3208 kunmap_atomic(kaddr);
3211 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3212 io_bio->mirror_num);
3213 memset(kaddr + pgoff, 1, len);
3214 flush_dcache_page(page);
3215 kunmap_atomic(kaddr);
3220 * when reads are done, we need to check csums to verify the data is correct
3221 * if there's a match, we allow the bio to finish. If not, the code in
3222 * extent_io.c will try to find good copies for us.
3224 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3225 u64 phy_offset, struct page *page,
3226 u64 start, u64 end, int mirror)
3228 size_t offset = start - page_offset(page);
3229 struct inode *inode = page->mapping->host;
3230 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3231 struct btrfs_root *root = BTRFS_I(inode)->root;
3233 if (PageChecked(page)) {
3234 ClearPageChecked(page);
3238 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3241 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3242 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3243 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3247 phy_offset >>= inode->i_sb->s_blocksize_bits;
3248 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3249 start, (size_t)(end - start + 1));
3253 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3255 * @inode: The inode we want to perform iput on
3257 * This function uses the generic vfs_inode::i_count to track whether we should
3258 * just decrement it (in case it's > 1) or if this is the last iput then link
3259 * the inode to the delayed iput machinery. Delayed iputs are processed at
3260 * transaction commit time/superblock commit/cleaner kthread.
3262 void btrfs_add_delayed_iput(struct inode *inode)
3264 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3265 struct btrfs_inode *binode = BTRFS_I(inode);
3267 if (atomic_add_unless(&inode->i_count, -1, 1))
3270 spin_lock(&fs_info->delayed_iput_lock);
3271 ASSERT(list_empty(&binode->delayed_iput));
3272 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3273 spin_unlock(&fs_info->delayed_iput_lock);
3276 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3279 spin_lock(&fs_info->delayed_iput_lock);
3280 while (!list_empty(&fs_info->delayed_iputs)) {
3281 struct btrfs_inode *inode;
3283 inode = list_first_entry(&fs_info->delayed_iputs,
3284 struct btrfs_inode, delayed_iput);
3285 list_del_init(&inode->delayed_iput);
3286 spin_unlock(&fs_info->delayed_iput_lock);
3287 iput(&inode->vfs_inode);
3288 spin_lock(&fs_info->delayed_iput_lock);
3290 spin_unlock(&fs_info->delayed_iput_lock);
3294 * This creates an orphan entry for the given inode in case something goes wrong
3295 * in the middle of an unlink.
3297 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3298 struct btrfs_inode *inode)
3302 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3303 if (ret && ret != -EEXIST) {
3304 btrfs_abort_transaction(trans, ret);
3312 * We have done the delete so we can go ahead and remove the orphan item for
3313 * this particular inode.
3315 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3316 struct btrfs_inode *inode)
3318 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3322 * this cleans up any orphans that may be left on the list from the last use
3325 int btrfs_orphan_cleanup(struct btrfs_root *root)
3327 struct btrfs_fs_info *fs_info = root->fs_info;
3328 struct btrfs_path *path;
3329 struct extent_buffer *leaf;
3330 struct btrfs_key key, found_key;
3331 struct btrfs_trans_handle *trans;
3332 struct inode *inode;
3333 u64 last_objectid = 0;
3334 int ret = 0, nr_unlink = 0;
3336 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3339 path = btrfs_alloc_path();
3344 path->reada = READA_BACK;
3346 key.objectid = BTRFS_ORPHAN_OBJECTID;
3347 key.type = BTRFS_ORPHAN_ITEM_KEY;
3348 key.offset = (u64)-1;
3351 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3356 * if ret == 0 means we found what we were searching for, which
3357 * is weird, but possible, so only screw with path if we didn't
3358 * find the key and see if we have stuff that matches
3362 if (path->slots[0] == 0)
3367 /* pull out the item */
3368 leaf = path->nodes[0];
3369 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3371 /* make sure the item matches what we want */
3372 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3374 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3377 /* release the path since we're done with it */
3378 btrfs_release_path(path);
3381 * this is where we are basically btrfs_lookup, without the
3382 * crossing root thing. we store the inode number in the
3383 * offset of the orphan item.
3386 if (found_key.offset == last_objectid) {
3388 "Error removing orphan entry, stopping orphan cleanup");
3393 last_objectid = found_key.offset;
3395 found_key.objectid = found_key.offset;
3396 found_key.type = BTRFS_INODE_ITEM_KEY;
3397 found_key.offset = 0;
3398 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3399 ret = PTR_ERR_OR_ZERO(inode);
3400 if (ret && ret != -ENOENT)
3403 if (ret == -ENOENT && root == fs_info->tree_root) {
3404 struct btrfs_root *dead_root;
3405 struct btrfs_fs_info *fs_info = root->fs_info;
3406 int is_dead_root = 0;
3409 * this is an orphan in the tree root. Currently these
3410 * could come from 2 sources:
3411 * a) a snapshot deletion in progress
3412 * b) a free space cache inode
3413 * We need to distinguish those two, as the snapshot
3414 * orphan must not get deleted.
3415 * find_dead_roots already ran before us, so if this
3416 * is a snapshot deletion, we should find the root
3417 * in the dead_roots list
3419 spin_lock(&fs_info->trans_lock);
3420 list_for_each_entry(dead_root, &fs_info->dead_roots,
3422 if (dead_root->root_key.objectid ==
3423 found_key.objectid) {
3428 spin_unlock(&fs_info->trans_lock);
3430 /* prevent this orphan from being found again */
3431 key.offset = found_key.objectid - 1;
3438 * If we have an inode with links, there are a couple of
3439 * possibilities. Old kernels (before v3.12) used to create an
3440 * orphan item for truncate indicating that there were possibly
3441 * extent items past i_size that needed to be deleted. In v3.12,
3442 * truncate was changed to update i_size in sync with the extent
3443 * items, but the (useless) orphan item was still created. Since
3444 * v4.18, we don't create the orphan item for truncate at all.
3446 * So, this item could mean that we need to do a truncate, but
3447 * only if this filesystem was last used on a pre-v3.12 kernel
3448 * and was not cleanly unmounted. The odds of that are quite
3449 * slim, and it's a pain to do the truncate now, so just delete
3452 * It's also possible that this orphan item was supposed to be
3453 * deleted but wasn't. The inode number may have been reused,
3454 * but either way, we can delete the orphan item.
3456 if (ret == -ENOENT || inode->i_nlink) {
3459 trans = btrfs_start_transaction(root, 1);
3460 if (IS_ERR(trans)) {
3461 ret = PTR_ERR(trans);
3464 btrfs_debug(fs_info, "auto deleting %Lu",
3465 found_key.objectid);
3466 ret = btrfs_del_orphan_item(trans, root,
3467 found_key.objectid);
3468 btrfs_end_transaction(trans);
3476 /* this will do delete_inode and everything for us */
3479 /* release the path since we're done with it */
3480 btrfs_release_path(path);
3482 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3484 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3485 trans = btrfs_join_transaction(root);
3487 btrfs_end_transaction(trans);
3491 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3495 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3496 btrfs_free_path(path);
3501 * very simple check to peek ahead in the leaf looking for xattrs. If we
3502 * don't find any xattrs, we know there can't be any acls.
3504 * slot is the slot the inode is in, objectid is the objectid of the inode
3506 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3507 int slot, u64 objectid,
3508 int *first_xattr_slot)
3510 u32 nritems = btrfs_header_nritems(leaf);
3511 struct btrfs_key found_key;
3512 static u64 xattr_access = 0;
3513 static u64 xattr_default = 0;
3516 if (!xattr_access) {
3517 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3518 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3519 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3520 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3524 *first_xattr_slot = -1;
3525 while (slot < nritems) {
3526 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3528 /* we found a different objectid, there must not be acls */
3529 if (found_key.objectid != objectid)
3532 /* we found an xattr, assume we've got an acl */
3533 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3534 if (*first_xattr_slot == -1)
3535 *first_xattr_slot = slot;
3536 if (found_key.offset == xattr_access ||
3537 found_key.offset == xattr_default)
3542 * we found a key greater than an xattr key, there can't
3543 * be any acls later on
3545 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3552 * it goes inode, inode backrefs, xattrs, extents,
3553 * so if there are a ton of hard links to an inode there can
3554 * be a lot of backrefs. Don't waste time searching too hard,
3555 * this is just an optimization
3560 /* we hit the end of the leaf before we found an xattr or
3561 * something larger than an xattr. We have to assume the inode
3564 if (*first_xattr_slot == -1)
3565 *first_xattr_slot = slot;
3570 * read an inode from the btree into the in-memory inode
3572 static int btrfs_read_locked_inode(struct inode *inode,
3573 struct btrfs_path *in_path)
3575 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3576 struct btrfs_path *path = in_path;
3577 struct extent_buffer *leaf;
3578 struct btrfs_inode_item *inode_item;
3579 struct btrfs_root *root = BTRFS_I(inode)->root;
3580 struct btrfs_key location;
3585 bool filled = false;
3586 int first_xattr_slot;
3588 ret = btrfs_fill_inode(inode, &rdev);
3593 path = btrfs_alloc_path();
3598 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3600 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3602 if (path != in_path)
3603 btrfs_free_path(path);
3607 leaf = path->nodes[0];
3612 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3613 struct btrfs_inode_item);
3614 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3615 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3616 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3617 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3618 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3620 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3621 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3623 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3624 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3626 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3627 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3629 BTRFS_I(inode)->i_otime.tv_sec =
3630 btrfs_timespec_sec(leaf, &inode_item->otime);
3631 BTRFS_I(inode)->i_otime.tv_nsec =
3632 btrfs_timespec_nsec(leaf, &inode_item->otime);
3634 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3635 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3636 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3638 inode_set_iversion_queried(inode,
3639 btrfs_inode_sequence(leaf, inode_item));
3640 inode->i_generation = BTRFS_I(inode)->generation;
3642 rdev = btrfs_inode_rdev(leaf, inode_item);
3644 BTRFS_I(inode)->index_cnt = (u64)-1;
3645 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3649 * If we were modified in the current generation and evicted from memory
3650 * and then re-read we need to do a full sync since we don't have any
3651 * idea about which extents were modified before we were evicted from
3654 * This is required for both inode re-read from disk and delayed inode
3655 * in delayed_nodes_tree.
3657 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3658 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3659 &BTRFS_I(inode)->runtime_flags);
3662 * We don't persist the id of the transaction where an unlink operation
3663 * against the inode was last made. So here we assume the inode might
3664 * have been evicted, and therefore the exact value of last_unlink_trans
3665 * lost, and set it to last_trans to avoid metadata inconsistencies
3666 * between the inode and its parent if the inode is fsync'ed and the log
3667 * replayed. For example, in the scenario:
3670 * ln mydir/foo mydir/bar
3673 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3674 * xfs_io -c fsync mydir/foo
3676 * mount fs, triggers fsync log replay
3678 * We must make sure that when we fsync our inode foo we also log its
3679 * parent inode, otherwise after log replay the parent still has the
3680 * dentry with the "bar" name but our inode foo has a link count of 1
3681 * and doesn't have an inode ref with the name "bar" anymore.
3683 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3684 * but it guarantees correctness at the expense of occasional full
3685 * transaction commits on fsync if our inode is a directory, or if our
3686 * inode is not a directory, logging its parent unnecessarily.
3688 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3691 if (inode->i_nlink != 1 ||
3692 path->slots[0] >= btrfs_header_nritems(leaf))
3695 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3696 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3699 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3700 if (location.type == BTRFS_INODE_REF_KEY) {
3701 struct btrfs_inode_ref *ref;
3703 ref = (struct btrfs_inode_ref *)ptr;
3704 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3705 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3706 struct btrfs_inode_extref *extref;
3708 extref = (struct btrfs_inode_extref *)ptr;
3709 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3714 * try to precache a NULL acl entry for files that don't have
3715 * any xattrs or acls
3717 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3718 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3719 if (first_xattr_slot != -1) {
3720 path->slots[0] = first_xattr_slot;
3721 ret = btrfs_load_inode_props(inode, path);
3724 "error loading props for ino %llu (root %llu): %d",
3725 btrfs_ino(BTRFS_I(inode)),
3726 root->root_key.objectid, ret);
3728 if (path != in_path)
3729 btrfs_free_path(path);
3732 cache_no_acl(inode);
3734 switch (inode->i_mode & S_IFMT) {
3736 inode->i_mapping->a_ops = &btrfs_aops;
3737 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3738 inode->i_fop = &btrfs_file_operations;
3739 inode->i_op = &btrfs_file_inode_operations;
3742 inode->i_fop = &btrfs_dir_file_operations;
3743 inode->i_op = &btrfs_dir_inode_operations;
3746 inode->i_op = &btrfs_symlink_inode_operations;
3747 inode_nohighmem(inode);
3748 inode->i_mapping->a_ops = &btrfs_aops;
3751 inode->i_op = &btrfs_special_inode_operations;
3752 init_special_inode(inode, inode->i_mode, rdev);
3756 btrfs_sync_inode_flags_to_i_flags(inode);
3761 * given a leaf and an inode, copy the inode fields into the leaf
3763 static void fill_inode_item(struct btrfs_trans_handle *trans,
3764 struct extent_buffer *leaf,
3765 struct btrfs_inode_item *item,
3766 struct inode *inode)
3768 struct btrfs_map_token token;
3770 btrfs_init_map_token(&token);
3772 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3773 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3774 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3776 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3777 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3779 btrfs_set_token_timespec_sec(leaf, &item->atime,
3780 inode->i_atime.tv_sec, &token);
3781 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3782 inode->i_atime.tv_nsec, &token);
3784 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3785 inode->i_mtime.tv_sec, &token);
3786 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3787 inode->i_mtime.tv_nsec, &token);
3789 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3790 inode->i_ctime.tv_sec, &token);
3791 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3792 inode->i_ctime.tv_nsec, &token);
3794 btrfs_set_token_timespec_sec(leaf, &item->otime,
3795 BTRFS_I(inode)->i_otime.tv_sec, &token);
3796 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3797 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3799 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3801 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3803 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3805 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3806 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3807 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3808 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3812 * copy everything in the in-memory inode into the btree.
3814 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3815 struct btrfs_root *root, struct inode *inode)
3817 struct btrfs_inode_item *inode_item;
3818 struct btrfs_path *path;
3819 struct extent_buffer *leaf;
3822 path = btrfs_alloc_path();
3826 path->leave_spinning = 1;
3827 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3835 leaf = path->nodes[0];
3836 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3837 struct btrfs_inode_item);
3839 fill_inode_item(trans, leaf, inode_item, inode);
3840 btrfs_mark_buffer_dirty(leaf);
3841 btrfs_set_inode_last_trans(trans, inode);
3844 btrfs_free_path(path);
3849 * copy everything in the in-memory inode into the btree.
3851 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3852 struct btrfs_root *root, struct inode *inode)
3854 struct btrfs_fs_info *fs_info = root->fs_info;
3858 * If the inode is a free space inode, we can deadlock during commit
3859 * if we put it into the delayed code.
3861 * The data relocation inode should also be directly updated
3864 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3865 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3866 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3867 btrfs_update_root_times(trans, root);
3869 ret = btrfs_delayed_update_inode(trans, root, inode);
3871 btrfs_set_inode_last_trans(trans, inode);
3875 return btrfs_update_inode_item(trans, root, inode);
3878 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3879 struct btrfs_root *root,
3880 struct inode *inode)
3884 ret = btrfs_update_inode(trans, root, inode);
3886 return btrfs_update_inode_item(trans, root, inode);
3891 * unlink helper that gets used here in inode.c and in the tree logging
3892 * recovery code. It remove a link in a directory with a given name, and
3893 * also drops the back refs in the inode to the directory
3895 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3896 struct btrfs_root *root,
3897 struct btrfs_inode *dir,
3898 struct btrfs_inode *inode,
3899 const char *name, int name_len)
3901 struct btrfs_fs_info *fs_info = root->fs_info;
3902 struct btrfs_path *path;
3904 struct extent_buffer *leaf;
3905 struct btrfs_dir_item *di;
3906 struct btrfs_key key;
3908 u64 ino = btrfs_ino(inode);
3909 u64 dir_ino = btrfs_ino(dir);
3911 path = btrfs_alloc_path();
3917 path->leave_spinning = 1;
3918 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3919 name, name_len, -1);
3920 if (IS_ERR_OR_NULL(di)) {
3921 ret = di ? PTR_ERR(di) : -ENOENT;
3924 leaf = path->nodes[0];
3925 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3926 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3929 btrfs_release_path(path);
3932 * If we don't have dir index, we have to get it by looking up
3933 * the inode ref, since we get the inode ref, remove it directly,
3934 * it is unnecessary to do delayed deletion.
3936 * But if we have dir index, needn't search inode ref to get it.
3937 * Since the inode ref is close to the inode item, it is better
3938 * that we delay to delete it, and just do this deletion when
3939 * we update the inode item.
3941 if (inode->dir_index) {
3942 ret = btrfs_delayed_delete_inode_ref(inode);
3944 index = inode->dir_index;
3949 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3953 "failed to delete reference to %.*s, inode %llu parent %llu",
3954 name_len, name, ino, dir_ino);
3955 btrfs_abort_transaction(trans, ret);
3959 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3961 btrfs_abort_transaction(trans, ret);
3965 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3967 if (ret != 0 && ret != -ENOENT) {
3968 btrfs_abort_transaction(trans, ret);
3972 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3977 btrfs_abort_transaction(trans, ret);
3979 btrfs_free_path(path);
3983 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3984 inode_inc_iversion(&inode->vfs_inode);
3985 inode_inc_iversion(&dir->vfs_inode);
3986 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3987 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3988 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3993 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3994 struct btrfs_root *root,
3995 struct btrfs_inode *dir, struct btrfs_inode *inode,
3996 const char *name, int name_len)
3999 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4001 drop_nlink(&inode->vfs_inode);
4002 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4008 * helper to start transaction for unlink and rmdir.
4010 * unlink and rmdir are special in btrfs, they do not always free space, so
4011 * if we cannot make our reservations the normal way try and see if there is
4012 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4013 * allow the unlink to occur.
4015 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4017 struct btrfs_root *root = BTRFS_I(dir)->root;
4020 * 1 for the possible orphan item
4021 * 1 for the dir item
4022 * 1 for the dir index
4023 * 1 for the inode ref
4026 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4029 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4031 struct btrfs_root *root = BTRFS_I(dir)->root;
4032 struct btrfs_trans_handle *trans;
4033 struct inode *inode = d_inode(dentry);
4036 trans = __unlink_start_trans(dir);
4038 return PTR_ERR(trans);
4040 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4043 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4044 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4045 dentry->d_name.len);
4049 if (inode->i_nlink == 0) {
4050 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4056 btrfs_end_transaction(trans);
4057 btrfs_btree_balance_dirty(root->fs_info);
4061 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4062 struct inode *dir, u64 objectid,
4063 const char *name, int name_len)
4065 struct btrfs_root *root = BTRFS_I(dir)->root;
4066 struct btrfs_path *path;
4067 struct extent_buffer *leaf;
4068 struct btrfs_dir_item *di;
4069 struct btrfs_key key;
4072 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4074 path = btrfs_alloc_path();
4078 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4079 name, name_len, -1);
4080 if (IS_ERR_OR_NULL(di)) {
4081 ret = di ? PTR_ERR(di) : -ENOENT;
4085 leaf = path->nodes[0];
4086 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4087 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4088 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4090 btrfs_abort_transaction(trans, ret);
4093 btrfs_release_path(path);
4095 ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid,
4096 dir_ino, &index, name, name_len);
4098 if (ret != -ENOENT) {
4099 btrfs_abort_transaction(trans, ret);
4102 di = btrfs_search_dir_index_item(root, path, dir_ino,
4104 if (IS_ERR_OR_NULL(di)) {
4109 btrfs_abort_transaction(trans, ret);
4113 leaf = path->nodes[0];
4114 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4117 btrfs_release_path(path);
4119 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4121 btrfs_abort_transaction(trans, ret);
4125 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4126 inode_inc_iversion(dir);
4127 dir->i_mtime = dir->i_ctime = current_time(dir);
4128 ret = btrfs_update_inode_fallback(trans, root, dir);
4130 btrfs_abort_transaction(trans, ret);
4132 btrfs_free_path(path);
4137 * Helper to check if the subvolume references other subvolumes or if it's
4140 static noinline int may_destroy_subvol(struct btrfs_root *root)
4142 struct btrfs_fs_info *fs_info = root->fs_info;
4143 struct btrfs_path *path;
4144 struct btrfs_dir_item *di;
4145 struct btrfs_key key;
4149 path = btrfs_alloc_path();
4153 /* Make sure this root isn't set as the default subvol */
4154 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4155 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4156 dir_id, "default", 7, 0);
4157 if (di && !IS_ERR(di)) {
4158 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4159 if (key.objectid == root->root_key.objectid) {
4162 "deleting default subvolume %llu is not allowed",
4166 btrfs_release_path(path);
4169 key.objectid = root->root_key.objectid;
4170 key.type = BTRFS_ROOT_REF_KEY;
4171 key.offset = (u64)-1;
4173 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4179 if (path->slots[0] > 0) {
4181 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4182 if (key.objectid == root->root_key.objectid &&
4183 key.type == BTRFS_ROOT_REF_KEY)
4187 btrfs_free_path(path);
4191 /* Delete all dentries for inodes belonging to the root */
4192 static void btrfs_prune_dentries(struct btrfs_root *root)
4194 struct btrfs_fs_info *fs_info = root->fs_info;
4195 struct rb_node *node;
4196 struct rb_node *prev;
4197 struct btrfs_inode *entry;
4198 struct inode *inode;
4201 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4202 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4204 spin_lock(&root->inode_lock);
4206 node = root->inode_tree.rb_node;
4210 entry = rb_entry(node, struct btrfs_inode, rb_node);
4212 if (objectid < btrfs_ino(entry))
4213 node = node->rb_left;
4214 else if (objectid > btrfs_ino(entry))
4215 node = node->rb_right;
4221 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4222 if (objectid <= btrfs_ino(entry)) {
4226 prev = rb_next(prev);
4230 entry = rb_entry(node, struct btrfs_inode, rb_node);
4231 objectid = btrfs_ino(entry) + 1;
4232 inode = igrab(&entry->vfs_inode);
4234 spin_unlock(&root->inode_lock);
4235 if (atomic_read(&inode->i_count) > 1)
4236 d_prune_aliases(inode);
4238 * btrfs_drop_inode will have it removed from the inode
4239 * cache when its usage count hits zero.
4243 spin_lock(&root->inode_lock);
4247 if (cond_resched_lock(&root->inode_lock))
4250 node = rb_next(node);
4252 spin_unlock(&root->inode_lock);
4255 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4257 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4258 struct btrfs_root *root = BTRFS_I(dir)->root;
4259 struct inode *inode = d_inode(dentry);
4260 struct btrfs_root *dest = BTRFS_I(inode)->root;
4261 struct btrfs_trans_handle *trans;
4262 struct btrfs_block_rsv block_rsv;
4268 * Don't allow to delete a subvolume with send in progress. This is
4269 * inside the inode lock so the error handling that has to drop the bit
4270 * again is not run concurrently.
4272 spin_lock(&dest->root_item_lock);
4273 if (dest->send_in_progress) {
4274 spin_unlock(&dest->root_item_lock);
4276 "attempt to delete subvolume %llu during send",
4277 dest->root_key.objectid);
4280 root_flags = btrfs_root_flags(&dest->root_item);
4281 btrfs_set_root_flags(&dest->root_item,
4282 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4283 spin_unlock(&dest->root_item_lock);
4285 down_write(&fs_info->subvol_sem);
4287 err = may_destroy_subvol(dest);
4291 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4293 * One for dir inode,
4294 * two for dir entries,
4295 * two for root ref/backref.
4297 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4301 trans = btrfs_start_transaction(root, 0);
4302 if (IS_ERR(trans)) {
4303 err = PTR_ERR(trans);
4306 trans->block_rsv = &block_rsv;
4307 trans->bytes_reserved = block_rsv.size;
4309 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4311 ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid,
4312 dentry->d_name.name, dentry->d_name.len);
4315 btrfs_abort_transaction(trans, ret);
4319 btrfs_record_root_in_trans(trans, dest);
4321 memset(&dest->root_item.drop_progress, 0,
4322 sizeof(dest->root_item.drop_progress));
4323 dest->root_item.drop_level = 0;
4324 btrfs_set_root_refs(&dest->root_item, 0);
4326 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4327 ret = btrfs_insert_orphan_item(trans,
4329 dest->root_key.objectid);
4331 btrfs_abort_transaction(trans, ret);
4337 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4338 BTRFS_UUID_KEY_SUBVOL,
4339 dest->root_key.objectid);
4340 if (ret && ret != -ENOENT) {
4341 btrfs_abort_transaction(trans, ret);
4345 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4346 ret = btrfs_uuid_tree_remove(trans,
4347 dest->root_item.received_uuid,
4348 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4349 dest->root_key.objectid);
4350 if (ret && ret != -ENOENT) {
4351 btrfs_abort_transaction(trans, ret);
4358 trans->block_rsv = NULL;
4359 trans->bytes_reserved = 0;
4360 ret = btrfs_end_transaction(trans);
4363 inode->i_flags |= S_DEAD;
4365 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4367 up_write(&fs_info->subvol_sem);
4369 spin_lock(&dest->root_item_lock);
4370 root_flags = btrfs_root_flags(&dest->root_item);
4371 btrfs_set_root_flags(&dest->root_item,
4372 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4373 spin_unlock(&dest->root_item_lock);
4375 d_invalidate(dentry);
4376 btrfs_prune_dentries(dest);
4377 ASSERT(dest->send_in_progress == 0);
4380 if (dest->ino_cache_inode) {
4381 iput(dest->ino_cache_inode);
4382 dest->ino_cache_inode = NULL;
4389 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4391 struct inode *inode = d_inode(dentry);
4393 struct btrfs_root *root = BTRFS_I(dir)->root;
4394 struct btrfs_trans_handle *trans;
4395 u64 last_unlink_trans;
4397 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4399 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4400 return btrfs_delete_subvolume(dir, dentry);
4402 trans = __unlink_start_trans(dir);
4404 return PTR_ERR(trans);
4406 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4407 err = btrfs_unlink_subvol(trans, dir,
4408 BTRFS_I(inode)->location.objectid,
4409 dentry->d_name.name,
4410 dentry->d_name.len);
4414 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4418 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4420 /* now the directory is empty */
4421 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4422 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4423 dentry->d_name.len);
4425 btrfs_i_size_write(BTRFS_I(inode), 0);
4427 * Propagate the last_unlink_trans value of the deleted dir to
4428 * its parent directory. This is to prevent an unrecoverable
4429 * log tree in the case we do something like this:
4431 * 2) create snapshot under dir foo
4432 * 3) delete the snapshot
4435 * 6) fsync foo or some file inside foo
4437 if (last_unlink_trans >= trans->transid)
4438 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4441 btrfs_end_transaction(trans);
4442 btrfs_btree_balance_dirty(root->fs_info);
4447 static int truncate_space_check(struct btrfs_trans_handle *trans,
4448 struct btrfs_root *root,
4451 struct btrfs_fs_info *fs_info = root->fs_info;
4455 * This is only used to apply pressure to the enospc system, we don't
4456 * intend to use this reservation at all.
4458 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4459 bytes_deleted *= fs_info->nodesize;
4460 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4461 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4463 trace_btrfs_space_reservation(fs_info, "transaction",
4466 trans->bytes_reserved += bytes_deleted;
4473 * Return this if we need to call truncate_block for the last bit of the
4476 #define NEED_TRUNCATE_BLOCK 1
4479 * this can truncate away extent items, csum items and directory items.
4480 * It starts at a high offset and removes keys until it can't find
4481 * any higher than new_size
4483 * csum items that cross the new i_size are truncated to the new size
4486 * min_type is the minimum key type to truncate down to. If set to 0, this
4487 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4489 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4490 struct btrfs_root *root,
4491 struct inode *inode,
4492 u64 new_size, u32 min_type)
4494 struct btrfs_fs_info *fs_info = root->fs_info;
4495 struct btrfs_path *path;
4496 struct extent_buffer *leaf;
4497 struct btrfs_file_extent_item *fi;
4498 struct btrfs_key key;
4499 struct btrfs_key found_key;
4500 u64 extent_start = 0;
4501 u64 extent_num_bytes = 0;
4502 u64 extent_offset = 0;
4504 u64 last_size = new_size;
4505 u32 found_type = (u8)-1;
4508 int pending_del_nr = 0;
4509 int pending_del_slot = 0;
4510 int extent_type = -1;
4512 u64 ino = btrfs_ino(BTRFS_I(inode));
4513 u64 bytes_deleted = 0;
4514 bool be_nice = false;
4515 bool should_throttle = false;
4516 bool should_end = false;
4518 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4521 * for non-free space inodes and ref cows, we want to back off from
4524 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4525 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4528 path = btrfs_alloc_path();
4531 path->reada = READA_BACK;
4534 * We want to drop from the next block forward in case this new size is
4535 * not block aligned since we will be keeping the last block of the
4536 * extent just the way it is.
4538 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4539 root == fs_info->tree_root)
4540 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4541 fs_info->sectorsize),
4545 * This function is also used to drop the items in the log tree before
4546 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4547 * it is used to drop the loged items. So we shouldn't kill the delayed
4550 if (min_type == 0 && root == BTRFS_I(inode)->root)
4551 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4554 key.offset = (u64)-1;
4559 * with a 16K leaf size and 128MB extents, you can actually queue
4560 * up a huge file in a single leaf. Most of the time that
4561 * bytes_deleted is > 0, it will be huge by the time we get here
4563 if (be_nice && bytes_deleted > SZ_32M &&
4564 btrfs_should_end_transaction(trans)) {
4569 path->leave_spinning = 1;
4570 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4576 /* there are no items in the tree for us to truncate, we're
4579 if (path->slots[0] == 0)
4586 leaf = path->nodes[0];
4587 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4588 found_type = found_key.type;
4590 if (found_key.objectid != ino)
4593 if (found_type < min_type)
4596 item_end = found_key.offset;
4597 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4598 fi = btrfs_item_ptr(leaf, path->slots[0],
4599 struct btrfs_file_extent_item);
4600 extent_type = btrfs_file_extent_type(leaf, fi);
4601 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4603 btrfs_file_extent_num_bytes(leaf, fi);
4605 trace_btrfs_truncate_show_fi_regular(
4606 BTRFS_I(inode), leaf, fi,
4608 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4609 item_end += btrfs_file_extent_ram_bytes(leaf,
4612 trace_btrfs_truncate_show_fi_inline(
4613 BTRFS_I(inode), leaf, fi, path->slots[0],
4618 if (found_type > min_type) {
4621 if (item_end < new_size)
4623 if (found_key.offset >= new_size)
4629 /* FIXME, shrink the extent if the ref count is only 1 */
4630 if (found_type != BTRFS_EXTENT_DATA_KEY)
4633 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4635 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4637 u64 orig_num_bytes =
4638 btrfs_file_extent_num_bytes(leaf, fi);
4639 extent_num_bytes = ALIGN(new_size -
4641 fs_info->sectorsize);
4642 btrfs_set_file_extent_num_bytes(leaf, fi,
4644 num_dec = (orig_num_bytes -
4646 if (test_bit(BTRFS_ROOT_REF_COWS,
4649 inode_sub_bytes(inode, num_dec);
4650 btrfs_mark_buffer_dirty(leaf);
4653 btrfs_file_extent_disk_num_bytes(leaf,
4655 extent_offset = found_key.offset -
4656 btrfs_file_extent_offset(leaf, fi);
4658 /* FIXME blocksize != 4096 */
4659 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4660 if (extent_start != 0) {
4662 if (test_bit(BTRFS_ROOT_REF_COWS,
4664 inode_sub_bytes(inode, num_dec);
4667 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4669 * we can't truncate inline items that have had
4673 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4674 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4675 btrfs_file_extent_compression(leaf, fi) == 0) {
4676 u32 size = (u32)(new_size - found_key.offset);
4678 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4679 size = btrfs_file_extent_calc_inline_size(size);
4680 btrfs_truncate_item(root->fs_info, path, size, 1);
4681 } else if (!del_item) {
4683 * We have to bail so the last_size is set to
4684 * just before this extent.
4686 ret = NEED_TRUNCATE_BLOCK;
4690 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4691 inode_sub_bytes(inode, item_end + 1 - new_size);
4695 last_size = found_key.offset;
4697 last_size = new_size;
4699 if (!pending_del_nr) {
4700 /* no pending yet, add ourselves */
4701 pending_del_slot = path->slots[0];
4703 } else if (pending_del_nr &&
4704 path->slots[0] + 1 == pending_del_slot) {
4705 /* hop on the pending chunk */
4707 pending_del_slot = path->slots[0];
4714 should_throttle = false;
4717 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4718 root == fs_info->tree_root)) {
4719 btrfs_set_path_blocking(path);
4720 bytes_deleted += extent_num_bytes;
4721 ret = btrfs_free_extent(trans, root, extent_start,
4722 extent_num_bytes, 0,
4723 btrfs_header_owner(leaf),
4724 ino, extent_offset);
4726 btrfs_abort_transaction(trans, ret);
4729 if (btrfs_should_throttle_delayed_refs(trans))
4730 btrfs_async_run_delayed_refs(fs_info,
4731 trans->delayed_ref_updates * 2,
4734 if (truncate_space_check(trans, root,
4735 extent_num_bytes)) {
4738 if (btrfs_should_throttle_delayed_refs(trans))
4739 should_throttle = true;
4743 if (found_type == BTRFS_INODE_ITEM_KEY)
4746 if (path->slots[0] == 0 ||
4747 path->slots[0] != pending_del_slot ||
4748 should_throttle || should_end) {
4749 if (pending_del_nr) {
4750 ret = btrfs_del_items(trans, root, path,
4754 btrfs_abort_transaction(trans, ret);
4759 btrfs_release_path(path);
4760 if (should_throttle) {
4761 unsigned long updates = trans->delayed_ref_updates;
4763 trans->delayed_ref_updates = 0;
4764 ret = btrfs_run_delayed_refs(trans,
4771 * if we failed to refill our space rsv, bail out
4772 * and let the transaction restart
4784 if (ret >= 0 && pending_del_nr) {
4787 err = btrfs_del_items(trans, root, path, pending_del_slot,
4790 btrfs_abort_transaction(trans, err);
4794 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4795 ASSERT(last_size >= new_size);
4796 if (!ret && last_size > new_size)
4797 last_size = new_size;
4798 btrfs_ordered_update_i_size(inode, last_size, NULL);
4801 btrfs_free_path(path);
4803 if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4804 unsigned long updates = trans->delayed_ref_updates;
4808 trans->delayed_ref_updates = 0;
4809 err = btrfs_run_delayed_refs(trans, updates * 2);
4818 * btrfs_truncate_block - read, zero a chunk and write a block
4819 * @inode - inode that we're zeroing
4820 * @from - the offset to start zeroing
4821 * @len - the length to zero, 0 to zero the entire range respective to the
4823 * @front - zero up to the offset instead of from the offset on
4825 * This will find the block for the "from" offset and cow the block and zero the
4826 * part we want to zero. This is used with truncate and hole punching.
4828 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4831 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4832 struct address_space *mapping = inode->i_mapping;
4833 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4834 struct btrfs_ordered_extent *ordered;
4835 struct extent_state *cached_state = NULL;
4836 struct extent_changeset *data_reserved = NULL;
4838 u32 blocksize = fs_info->sectorsize;
4839 pgoff_t index = from >> PAGE_SHIFT;
4840 unsigned offset = from & (blocksize - 1);
4842 gfp_t mask = btrfs_alloc_write_mask(mapping);
4847 if (IS_ALIGNED(offset, blocksize) &&
4848 (!len || IS_ALIGNED(len, blocksize)))
4851 block_start = round_down(from, blocksize);
4852 block_end = block_start + blocksize - 1;
4854 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4855 block_start, blocksize);
4860 page = find_or_create_page(mapping, index, mask);
4862 btrfs_delalloc_release_space(inode, data_reserved,
4863 block_start, blocksize, true);
4864 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4869 if (!PageUptodate(page)) {
4870 ret = btrfs_readpage(NULL, page);
4872 if (page->mapping != mapping) {
4877 if (!PageUptodate(page)) {
4882 wait_on_page_writeback(page);
4884 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4885 set_page_extent_mapped(page);
4887 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4889 unlock_extent_cached(io_tree, block_start, block_end,
4893 btrfs_start_ordered_extent(inode, ordered, 1);
4894 btrfs_put_ordered_extent(ordered);
4898 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4899 EXTENT_DIRTY | EXTENT_DELALLOC |
4900 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4901 0, 0, &cached_state);
4903 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4906 unlock_extent_cached(io_tree, block_start, block_end,
4911 if (offset != blocksize) {
4913 len = blocksize - offset;
4916 memset(kaddr + (block_start - page_offset(page)),
4919 memset(kaddr + (block_start - page_offset(page)) + offset,
4921 flush_dcache_page(page);
4924 ClearPageChecked(page);
4925 set_page_dirty(page);
4926 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4930 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4932 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4936 extent_changeset_free(data_reserved);
4940 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4941 u64 offset, u64 len)
4943 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4944 struct btrfs_trans_handle *trans;
4948 * Still need to make sure the inode looks like it's been updated so
4949 * that any holes get logged if we fsync.
4951 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4952 BTRFS_I(inode)->last_trans = fs_info->generation;
4953 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4954 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4959 * 1 - for the one we're dropping
4960 * 1 - for the one we're adding
4961 * 1 - for updating the inode.
4963 trans = btrfs_start_transaction(root, 3);
4965 return PTR_ERR(trans);
4967 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4969 btrfs_abort_transaction(trans, ret);
4970 btrfs_end_transaction(trans);
4974 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4975 offset, 0, 0, len, 0, len, 0, 0, 0);
4977 btrfs_abort_transaction(trans, ret);
4979 btrfs_update_inode(trans, root, inode);
4980 btrfs_end_transaction(trans);
4985 * This function puts in dummy file extents for the area we're creating a hole
4986 * for. So if we are truncating this file to a larger size we need to insert
4987 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4988 * the range between oldsize and size
4990 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4992 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4993 struct btrfs_root *root = BTRFS_I(inode)->root;
4994 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4995 struct extent_map *em = NULL;
4996 struct extent_state *cached_state = NULL;
4997 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4998 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4999 u64 block_end = ALIGN(size, fs_info->sectorsize);
5006 * If our size started in the middle of a block we need to zero out the
5007 * rest of the block before we expand the i_size, otherwise we could
5008 * expose stale data.
5010 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5014 if (size <= hole_start)
5018 struct btrfs_ordered_extent *ordered;
5020 lock_extent_bits(io_tree, hole_start, block_end - 1,
5022 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5023 block_end - hole_start);
5026 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5028 btrfs_start_ordered_extent(inode, ordered, 1);
5029 btrfs_put_ordered_extent(ordered);
5032 cur_offset = hole_start;
5034 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5035 block_end - cur_offset, 0);
5041 last_byte = min(extent_map_end(em), block_end);
5042 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5043 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5044 struct extent_map *hole_em;
5045 hole_size = last_byte - cur_offset;
5047 err = maybe_insert_hole(root, inode, cur_offset,
5051 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5052 cur_offset + hole_size - 1, 0);
5053 hole_em = alloc_extent_map();
5055 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5056 &BTRFS_I(inode)->runtime_flags);
5059 hole_em->start = cur_offset;
5060 hole_em->len = hole_size;
5061 hole_em->orig_start = cur_offset;
5063 hole_em->block_start = EXTENT_MAP_HOLE;
5064 hole_em->block_len = 0;
5065 hole_em->orig_block_len = 0;
5066 hole_em->ram_bytes = hole_size;
5067 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5068 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5069 hole_em->generation = fs_info->generation;
5072 write_lock(&em_tree->lock);
5073 err = add_extent_mapping(em_tree, hole_em, 1);
5074 write_unlock(&em_tree->lock);
5077 btrfs_drop_extent_cache(BTRFS_I(inode),
5082 free_extent_map(hole_em);
5085 free_extent_map(em);
5087 cur_offset = last_byte;
5088 if (cur_offset >= block_end)
5091 free_extent_map(em);
5092 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5096 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5098 struct btrfs_root *root = BTRFS_I(inode)->root;
5099 struct btrfs_trans_handle *trans;
5100 loff_t oldsize = i_size_read(inode);
5101 loff_t newsize = attr->ia_size;
5102 int mask = attr->ia_valid;
5106 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5107 * special case where we need to update the times despite not having
5108 * these flags set. For all other operations the VFS set these flags
5109 * explicitly if it wants a timestamp update.
5111 if (newsize != oldsize) {
5112 inode_inc_iversion(inode);
5113 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5114 inode->i_ctime = inode->i_mtime =
5115 current_time(inode);
5118 if (newsize > oldsize) {
5120 * Don't do an expanding truncate while snapshotting is ongoing.
5121 * This is to ensure the snapshot captures a fully consistent
5122 * state of this file - if the snapshot captures this expanding
5123 * truncation, it must capture all writes that happened before
5126 btrfs_wait_for_snapshot_creation(root);
5127 ret = btrfs_cont_expand(inode, oldsize, newsize);
5129 btrfs_end_write_no_snapshotting(root);
5133 trans = btrfs_start_transaction(root, 1);
5134 if (IS_ERR(trans)) {
5135 btrfs_end_write_no_snapshotting(root);
5136 return PTR_ERR(trans);
5139 i_size_write(inode, newsize);
5140 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5141 pagecache_isize_extended(inode, oldsize, newsize);
5142 ret = btrfs_update_inode(trans, root, inode);
5143 btrfs_end_write_no_snapshotting(root);
5144 btrfs_end_transaction(trans);
5148 * We're truncating a file that used to have good data down to
5149 * zero. Make sure it gets into the ordered flush list so that
5150 * any new writes get down to disk quickly.
5153 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5154 &BTRFS_I(inode)->runtime_flags);
5156 truncate_setsize(inode, newsize);
5158 /* Disable nonlocked read DIO to avoid the end less truncate */
5159 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5160 inode_dio_wait(inode);
5161 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5163 ret = btrfs_truncate(inode, newsize == oldsize);
5164 if (ret && inode->i_nlink) {
5168 * Truncate failed, so fix up the in-memory size. We
5169 * adjusted disk_i_size down as we removed extents, so
5170 * wait for disk_i_size to be stable and then update the
5171 * in-memory size to match.
5173 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5176 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5183 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5185 struct inode *inode = d_inode(dentry);
5186 struct btrfs_root *root = BTRFS_I(inode)->root;
5189 if (btrfs_root_readonly(root))
5192 err = setattr_prepare(dentry, attr);
5196 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5197 err = btrfs_setsize(inode, attr);
5202 if (attr->ia_valid) {
5203 setattr_copy(inode, attr);
5204 inode_inc_iversion(inode);
5205 err = btrfs_dirty_inode(inode);
5207 if (!err && attr->ia_valid & ATTR_MODE)
5208 err = posix_acl_chmod(inode, inode->i_mode);
5215 * While truncating the inode pages during eviction, we get the VFS calling
5216 * btrfs_invalidatepage() against each page of the inode. This is slow because
5217 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5218 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5219 * extent_state structures over and over, wasting lots of time.
5221 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5222 * those expensive operations on a per page basis and do only the ordered io
5223 * finishing, while we release here the extent_map and extent_state structures,
5224 * without the excessive merging and splitting.
5226 static void evict_inode_truncate_pages(struct inode *inode)
5228 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5229 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5230 struct rb_node *node;
5232 ASSERT(inode->i_state & I_FREEING);
5233 truncate_inode_pages_final(&inode->i_data);
5235 write_lock(&map_tree->lock);
5236 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5237 struct extent_map *em;
5239 node = rb_first_cached(&map_tree->map);
5240 em = rb_entry(node, struct extent_map, rb_node);
5241 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5242 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5243 remove_extent_mapping(map_tree, em);
5244 free_extent_map(em);
5245 if (need_resched()) {
5246 write_unlock(&map_tree->lock);
5248 write_lock(&map_tree->lock);
5251 write_unlock(&map_tree->lock);
5254 * Keep looping until we have no more ranges in the io tree.
5255 * We can have ongoing bios started by readpages (called from readahead)
5256 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5257 * still in progress (unlocked the pages in the bio but did not yet
5258 * unlocked the ranges in the io tree). Therefore this means some
5259 * ranges can still be locked and eviction started because before
5260 * submitting those bios, which are executed by a separate task (work
5261 * queue kthread), inode references (inode->i_count) were not taken
5262 * (which would be dropped in the end io callback of each bio).
5263 * Therefore here we effectively end up waiting for those bios and
5264 * anyone else holding locked ranges without having bumped the inode's
5265 * reference count - if we don't do it, when they access the inode's
5266 * io_tree to unlock a range it may be too late, leading to an
5267 * use-after-free issue.
5269 spin_lock(&io_tree->lock);
5270 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5271 struct extent_state *state;
5272 struct extent_state *cached_state = NULL;
5275 unsigned state_flags;
5277 node = rb_first(&io_tree->state);
5278 state = rb_entry(node, struct extent_state, rb_node);
5279 start = state->start;
5281 state_flags = state->state;
5282 spin_unlock(&io_tree->lock);
5284 lock_extent_bits(io_tree, start, end, &cached_state);
5287 * If still has DELALLOC flag, the extent didn't reach disk,
5288 * and its reserved space won't be freed by delayed_ref.
5289 * So we need to free its reserved space here.
5290 * (Refer to comment in btrfs_invalidatepage, case 2)
5292 * Note, end is the bytenr of last byte, so we need + 1 here.
5294 if (state_flags & EXTENT_DELALLOC)
5295 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5297 clear_extent_bit(io_tree, start, end,
5298 EXTENT_LOCKED | EXTENT_DIRTY |
5299 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5300 EXTENT_DEFRAG, 1, 1, &cached_state);
5303 spin_lock(&io_tree->lock);
5305 spin_unlock(&io_tree->lock);
5308 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5309 struct btrfs_block_rsv *rsv)
5311 struct btrfs_fs_info *fs_info = root->fs_info;
5312 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5316 struct btrfs_trans_handle *trans;
5319 ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
5320 BTRFS_RESERVE_FLUSH_LIMIT);
5322 if (ret && ++failures > 2) {
5324 "could not allocate space for a delete; will truncate on mount");
5325 return ERR_PTR(-ENOSPC);
5328 trans = btrfs_join_transaction(root);
5329 if (IS_ERR(trans) || !ret)
5333 * Try to steal from the global reserve if there is space for
5336 if (!btrfs_check_space_for_delayed_refs(trans) &&
5337 !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, false))
5340 /* If not, commit and try again. */
5341 ret = btrfs_commit_transaction(trans);
5343 return ERR_PTR(ret);
5347 void btrfs_evict_inode(struct inode *inode)
5349 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5350 struct btrfs_trans_handle *trans;
5351 struct btrfs_root *root = BTRFS_I(inode)->root;
5352 struct btrfs_block_rsv *rsv;
5355 trace_btrfs_inode_evict(inode);
5362 evict_inode_truncate_pages(inode);
5364 if (inode->i_nlink &&
5365 ((btrfs_root_refs(&root->root_item) != 0 &&
5366 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5367 btrfs_is_free_space_inode(BTRFS_I(inode))))
5370 if (is_bad_inode(inode))
5373 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5375 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5378 if (inode->i_nlink > 0) {
5379 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5380 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5384 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5388 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5391 rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5394 btrfs_i_size_write(BTRFS_I(inode), 0);
5397 trans = evict_refill_and_join(root, rsv);
5401 trans->block_rsv = rsv;
5403 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5404 trans->block_rsv = &fs_info->trans_block_rsv;
5405 btrfs_end_transaction(trans);
5406 btrfs_btree_balance_dirty(fs_info);
5407 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5414 * Errors here aren't a big deal, it just means we leave orphan items in
5415 * the tree. They will be cleaned up on the next mount. If the inode
5416 * number gets reused, cleanup deletes the orphan item without doing
5417 * anything, and unlink reuses the existing orphan item.
5419 * If it turns out that we are dropping too many of these, we might want
5420 * to add a mechanism for retrying these after a commit.
5422 trans = evict_refill_and_join(root, rsv);
5423 if (!IS_ERR(trans)) {
5424 trans->block_rsv = rsv;
5425 btrfs_orphan_del(trans, BTRFS_I(inode));
5426 trans->block_rsv = &fs_info->trans_block_rsv;
5427 btrfs_end_transaction(trans);
5430 if (!(root == fs_info->tree_root ||
5431 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5432 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5435 btrfs_free_block_rsv(fs_info, rsv);
5438 * If we didn't successfully delete, the orphan item will still be in
5439 * the tree and we'll retry on the next mount. Again, we might also want
5440 * to retry these periodically in the future.
5442 btrfs_remove_delayed_node(BTRFS_I(inode));
5447 * this returns the key found in the dir entry in the location pointer.
5448 * If no dir entries were found, returns -ENOENT.
5449 * If found a corrupted location in dir entry, returns -EUCLEAN.
5451 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5452 struct btrfs_key *location)
5454 const char *name = dentry->d_name.name;
5455 int namelen = dentry->d_name.len;
5456 struct btrfs_dir_item *di;
5457 struct btrfs_path *path;
5458 struct btrfs_root *root = BTRFS_I(dir)->root;
5461 path = btrfs_alloc_path();
5465 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5467 if (IS_ERR_OR_NULL(di)) {
5468 ret = di ? PTR_ERR(di) : -ENOENT;
5472 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5473 if (location->type != BTRFS_INODE_ITEM_KEY &&
5474 location->type != BTRFS_ROOT_ITEM_KEY) {
5476 btrfs_warn(root->fs_info,
5477 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5478 __func__, name, btrfs_ino(BTRFS_I(dir)),
5479 location->objectid, location->type, location->offset);
5482 btrfs_free_path(path);
5487 * when we hit a tree root in a directory, the btrfs part of the inode
5488 * needs to be changed to reflect the root directory of the tree root. This
5489 * is kind of like crossing a mount point.
5491 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5493 struct dentry *dentry,
5494 struct btrfs_key *location,
5495 struct btrfs_root **sub_root)
5497 struct btrfs_path *path;
5498 struct btrfs_root *new_root;
5499 struct btrfs_root_ref *ref;
5500 struct extent_buffer *leaf;
5501 struct btrfs_key key;
5505 path = btrfs_alloc_path();
5512 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5513 key.type = BTRFS_ROOT_REF_KEY;
5514 key.offset = location->objectid;
5516 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5523 leaf = path->nodes[0];
5524 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5525 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5526 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5529 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5530 (unsigned long)(ref + 1),
5531 dentry->d_name.len);
5535 btrfs_release_path(path);
5537 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5538 if (IS_ERR(new_root)) {
5539 err = PTR_ERR(new_root);
5543 *sub_root = new_root;
5544 location->objectid = btrfs_root_dirid(&new_root->root_item);
5545 location->type = BTRFS_INODE_ITEM_KEY;
5546 location->offset = 0;
5549 btrfs_free_path(path);
5553 static void inode_tree_add(struct inode *inode)
5555 struct btrfs_root *root = BTRFS_I(inode)->root;
5556 struct btrfs_inode *entry;
5558 struct rb_node *parent;
5559 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5560 u64 ino = btrfs_ino(BTRFS_I(inode));
5562 if (inode_unhashed(inode))
5565 spin_lock(&root->inode_lock);
5566 p = &root->inode_tree.rb_node;
5569 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5571 if (ino < btrfs_ino(entry))
5572 p = &parent->rb_left;
5573 else if (ino > btrfs_ino(entry))
5574 p = &parent->rb_right;
5576 WARN_ON(!(entry->vfs_inode.i_state &
5577 (I_WILL_FREE | I_FREEING)));
5578 rb_replace_node(parent, new, &root->inode_tree);
5579 RB_CLEAR_NODE(parent);
5580 spin_unlock(&root->inode_lock);
5584 rb_link_node(new, parent, p);
5585 rb_insert_color(new, &root->inode_tree);
5586 spin_unlock(&root->inode_lock);
5589 static void inode_tree_del(struct inode *inode)
5591 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5592 struct btrfs_root *root = BTRFS_I(inode)->root;
5595 spin_lock(&root->inode_lock);
5596 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5597 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5598 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5599 empty = RB_EMPTY_ROOT(&root->inode_tree);
5601 spin_unlock(&root->inode_lock);
5603 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5604 synchronize_srcu(&fs_info->subvol_srcu);
5605 spin_lock(&root->inode_lock);
5606 empty = RB_EMPTY_ROOT(&root->inode_tree);
5607 spin_unlock(&root->inode_lock);
5609 btrfs_add_dead_root(root);
5614 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5616 struct btrfs_iget_args *args = p;
5617 inode->i_ino = args->location->objectid;
5618 memcpy(&BTRFS_I(inode)->location, args->location,
5619 sizeof(*args->location));
5620 BTRFS_I(inode)->root = args->root;
5624 static int btrfs_find_actor(struct inode *inode, void *opaque)
5626 struct btrfs_iget_args *args = opaque;
5627 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5628 args->root == BTRFS_I(inode)->root;
5631 static struct inode *btrfs_iget_locked(struct super_block *s,
5632 struct btrfs_key *location,
5633 struct btrfs_root *root)
5635 struct inode *inode;
5636 struct btrfs_iget_args args;
5637 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5639 args.location = location;
5642 inode = iget5_locked(s, hashval, btrfs_find_actor,
5643 btrfs_init_locked_inode,
5648 /* Get an inode object given its location and corresponding root.
5649 * Returns in *is_new if the inode was read from disk
5651 struct inode *btrfs_iget_path(struct super_block *s, struct btrfs_key *location,
5652 struct btrfs_root *root, int *new,
5653 struct btrfs_path *path)
5655 struct inode *inode;
5657 inode = btrfs_iget_locked(s, location, root);
5659 return ERR_PTR(-ENOMEM);
5661 if (inode->i_state & I_NEW) {
5664 ret = btrfs_read_locked_inode(inode, path);
5666 inode_tree_add(inode);
5667 unlock_new_inode(inode);
5673 * ret > 0 can come from btrfs_search_slot called by
5674 * btrfs_read_locked_inode, this means the inode item
5679 inode = ERR_PTR(ret);
5686 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5687 struct btrfs_root *root, int *new)
5689 return btrfs_iget_path(s, location, root, new, NULL);
5692 static struct inode *new_simple_dir(struct super_block *s,
5693 struct btrfs_key *key,
5694 struct btrfs_root *root)
5696 struct inode *inode = new_inode(s);
5699 return ERR_PTR(-ENOMEM);
5701 BTRFS_I(inode)->root = root;
5702 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5703 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5705 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5706 inode->i_op = &btrfs_dir_ro_inode_operations;
5707 inode->i_opflags &= ~IOP_XATTR;
5708 inode->i_fop = &simple_dir_operations;
5709 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5710 inode->i_mtime = current_time(inode);
5711 inode->i_atime = inode->i_mtime;
5712 inode->i_ctime = inode->i_mtime;
5713 BTRFS_I(inode)->i_otime = inode->i_mtime;
5718 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5720 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5721 struct inode *inode;
5722 struct btrfs_root *root = BTRFS_I(dir)->root;
5723 struct btrfs_root *sub_root = root;
5724 struct btrfs_key location;
5728 if (dentry->d_name.len > BTRFS_NAME_LEN)
5729 return ERR_PTR(-ENAMETOOLONG);
5731 ret = btrfs_inode_by_name(dir, dentry, &location);
5733 return ERR_PTR(ret);
5735 if (location.type == BTRFS_INODE_ITEM_KEY) {
5736 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5740 index = srcu_read_lock(&fs_info->subvol_srcu);
5741 ret = fixup_tree_root_location(fs_info, dir, dentry,
5742 &location, &sub_root);
5745 inode = ERR_PTR(ret);
5747 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5749 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5751 srcu_read_unlock(&fs_info->subvol_srcu, index);
5753 if (!IS_ERR(inode) && root != sub_root) {
5754 down_read(&fs_info->cleanup_work_sem);
5755 if (!sb_rdonly(inode->i_sb))
5756 ret = btrfs_orphan_cleanup(sub_root);
5757 up_read(&fs_info->cleanup_work_sem);
5760 inode = ERR_PTR(ret);
5767 static int btrfs_dentry_delete(const struct dentry *dentry)
5769 struct btrfs_root *root;
5770 struct inode *inode = d_inode(dentry);
5772 if (!inode && !IS_ROOT(dentry))
5773 inode = d_inode(dentry->d_parent);
5776 root = BTRFS_I(inode)->root;
5777 if (btrfs_root_refs(&root->root_item) == 0)
5780 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5786 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5789 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5791 if (inode == ERR_PTR(-ENOENT))
5793 return d_splice_alias(inode, dentry);
5796 unsigned char btrfs_filetype_table[] = {
5797 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5801 * All this infrastructure exists because dir_emit can fault, and we are holding
5802 * the tree lock when doing readdir. For now just allocate a buffer and copy
5803 * our information into that, and then dir_emit from the buffer. This is
5804 * similar to what NFS does, only we don't keep the buffer around in pagecache
5805 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5806 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5809 static int btrfs_opendir(struct inode *inode, struct file *file)
5811 struct btrfs_file_private *private;
5813 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5816 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5817 if (!private->filldir_buf) {
5821 file->private_data = private;
5832 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5835 struct dir_entry *entry = addr;
5836 char *name = (char *)(entry + 1);
5838 ctx->pos = get_unaligned(&entry->offset);
5839 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5840 get_unaligned(&entry->ino),
5841 get_unaligned(&entry->type)))
5843 addr += sizeof(struct dir_entry) +
5844 get_unaligned(&entry->name_len);
5850 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5852 struct inode *inode = file_inode(file);
5853 struct btrfs_root *root = BTRFS_I(inode)->root;
5854 struct btrfs_file_private *private = file->private_data;
5855 struct btrfs_dir_item *di;
5856 struct btrfs_key key;
5857 struct btrfs_key found_key;
5858 struct btrfs_path *path;
5860 struct list_head ins_list;
5861 struct list_head del_list;
5863 struct extent_buffer *leaf;
5870 struct btrfs_key location;
5872 if (!dir_emit_dots(file, ctx))
5875 path = btrfs_alloc_path();
5879 addr = private->filldir_buf;
5880 path->reada = READA_FORWARD;
5882 INIT_LIST_HEAD(&ins_list);
5883 INIT_LIST_HEAD(&del_list);
5884 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5887 key.type = BTRFS_DIR_INDEX_KEY;
5888 key.offset = ctx->pos;
5889 key.objectid = btrfs_ino(BTRFS_I(inode));
5891 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5896 struct dir_entry *entry;
5898 leaf = path->nodes[0];
5899 slot = path->slots[0];
5900 if (slot >= btrfs_header_nritems(leaf)) {
5901 ret = btrfs_next_leaf(root, path);
5909 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5911 if (found_key.objectid != key.objectid)
5913 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5915 if (found_key.offset < ctx->pos)
5917 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5919 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5920 name_len = btrfs_dir_name_len(leaf, di);
5921 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5923 btrfs_release_path(path);
5924 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5927 addr = private->filldir_buf;
5934 put_unaligned(name_len, &entry->name_len);
5935 name_ptr = (char *)(entry + 1);
5936 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5938 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5940 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5941 put_unaligned(location.objectid, &entry->ino);
5942 put_unaligned(found_key.offset, &entry->offset);
5944 addr += sizeof(struct dir_entry) + name_len;
5945 total_len += sizeof(struct dir_entry) + name_len;
5949 btrfs_release_path(path);
5951 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5955 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5960 * Stop new entries from being returned after we return the last
5963 * New directory entries are assigned a strictly increasing
5964 * offset. This means that new entries created during readdir
5965 * are *guaranteed* to be seen in the future by that readdir.
5966 * This has broken buggy programs which operate on names as
5967 * they're returned by readdir. Until we re-use freed offsets
5968 * we have this hack to stop new entries from being returned
5969 * under the assumption that they'll never reach this huge
5972 * This is being careful not to overflow 32bit loff_t unless the
5973 * last entry requires it because doing so has broken 32bit apps
5976 if (ctx->pos >= INT_MAX)
5977 ctx->pos = LLONG_MAX;
5984 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5985 btrfs_free_path(path);
5990 * This is somewhat expensive, updating the tree every time the
5991 * inode changes. But, it is most likely to find the inode in cache.
5992 * FIXME, needs more benchmarking...there are no reasons other than performance
5993 * to keep or drop this code.
5995 static int btrfs_dirty_inode(struct inode *inode)
5997 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5998 struct btrfs_root *root = BTRFS_I(inode)->root;
5999 struct btrfs_trans_handle *trans;
6002 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6005 trans = btrfs_join_transaction(root);
6007 return PTR_ERR(trans);
6009 ret = btrfs_update_inode(trans, root, inode);
6010 if (ret && ret == -ENOSPC) {
6011 /* whoops, lets try again with the full transaction */
6012 btrfs_end_transaction(trans);
6013 trans = btrfs_start_transaction(root, 1);
6015 return PTR_ERR(trans);
6017 ret = btrfs_update_inode(trans, root, inode);
6019 btrfs_end_transaction(trans);
6020 if (BTRFS_I(inode)->delayed_node)
6021 btrfs_balance_delayed_items(fs_info);
6027 * This is a copy of file_update_time. We need this so we can return error on
6028 * ENOSPC for updating the inode in the case of file write and mmap writes.
6030 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6033 struct btrfs_root *root = BTRFS_I(inode)->root;
6034 bool dirty = flags & ~S_VERSION;
6036 if (btrfs_root_readonly(root))
6039 if (flags & S_VERSION)
6040 dirty |= inode_maybe_inc_iversion(inode, dirty);
6041 if (flags & S_CTIME)
6042 inode->i_ctime = *now;
6043 if (flags & S_MTIME)
6044 inode->i_mtime = *now;
6045 if (flags & S_ATIME)
6046 inode->i_atime = *now;
6047 return dirty ? btrfs_dirty_inode(inode) : 0;
6051 * find the highest existing sequence number in a directory
6052 * and then set the in-memory index_cnt variable to reflect
6053 * free sequence numbers
6055 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6057 struct btrfs_root *root = inode->root;
6058 struct btrfs_key key, found_key;
6059 struct btrfs_path *path;
6060 struct extent_buffer *leaf;
6063 key.objectid = btrfs_ino(inode);
6064 key.type = BTRFS_DIR_INDEX_KEY;
6065 key.offset = (u64)-1;
6067 path = btrfs_alloc_path();
6071 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6074 /* FIXME: we should be able to handle this */
6080 * MAGIC NUMBER EXPLANATION:
6081 * since we search a directory based on f_pos we have to start at 2
6082 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6083 * else has to start at 2
6085 if (path->slots[0] == 0) {
6086 inode->index_cnt = 2;
6092 leaf = path->nodes[0];
6093 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6095 if (found_key.objectid != btrfs_ino(inode) ||
6096 found_key.type != BTRFS_DIR_INDEX_KEY) {
6097 inode->index_cnt = 2;
6101 inode->index_cnt = found_key.offset + 1;
6103 btrfs_free_path(path);
6108 * helper to find a free sequence number in a given directory. This current
6109 * code is very simple, later versions will do smarter things in the btree
6111 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6115 if (dir->index_cnt == (u64)-1) {
6116 ret = btrfs_inode_delayed_dir_index_count(dir);
6118 ret = btrfs_set_inode_index_count(dir);
6124 *index = dir->index_cnt;
6130 static int btrfs_insert_inode_locked(struct inode *inode)
6132 struct btrfs_iget_args args;
6133 args.location = &BTRFS_I(inode)->location;
6134 args.root = BTRFS_I(inode)->root;
6136 return insert_inode_locked4(inode,
6137 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6138 btrfs_find_actor, &args);
6142 * Inherit flags from the parent inode.
6144 * Currently only the compression flags and the cow flags are inherited.
6146 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6153 flags = BTRFS_I(dir)->flags;
6155 if (flags & BTRFS_INODE_NOCOMPRESS) {
6156 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6157 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6158 } else if (flags & BTRFS_INODE_COMPRESS) {
6159 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6160 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6163 if (flags & BTRFS_INODE_NODATACOW) {
6164 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6165 if (S_ISREG(inode->i_mode))
6166 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6169 btrfs_sync_inode_flags_to_i_flags(inode);
6172 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6173 struct btrfs_root *root,
6175 const char *name, int name_len,
6176 u64 ref_objectid, u64 objectid,
6177 umode_t mode, u64 *index)
6179 struct btrfs_fs_info *fs_info = root->fs_info;
6180 struct inode *inode;
6181 struct btrfs_inode_item *inode_item;
6182 struct btrfs_key *location;
6183 struct btrfs_path *path;
6184 struct btrfs_inode_ref *ref;
6185 struct btrfs_key key[2];
6187 int nitems = name ? 2 : 1;
6191 path = btrfs_alloc_path();
6193 return ERR_PTR(-ENOMEM);
6195 inode = new_inode(fs_info->sb);
6197 btrfs_free_path(path);
6198 return ERR_PTR(-ENOMEM);
6202 * O_TMPFILE, set link count to 0, so that after this point,
6203 * we fill in an inode item with the correct link count.
6206 set_nlink(inode, 0);
6209 * we have to initialize this early, so we can reclaim the inode
6210 * number if we fail afterwards in this function.
6212 inode->i_ino = objectid;
6215 trace_btrfs_inode_request(dir);
6217 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6219 btrfs_free_path(path);
6221 return ERR_PTR(ret);
6227 * index_cnt is ignored for everything but a dir,
6228 * btrfs_set_inode_index_count has an explanation for the magic
6231 BTRFS_I(inode)->index_cnt = 2;
6232 BTRFS_I(inode)->dir_index = *index;
6233 BTRFS_I(inode)->root = root;
6234 BTRFS_I(inode)->generation = trans->transid;
6235 inode->i_generation = BTRFS_I(inode)->generation;
6238 * We could have gotten an inode number from somebody who was fsynced
6239 * and then removed in this same transaction, so let's just set full
6240 * sync since it will be a full sync anyway and this will blow away the
6241 * old info in the log.
6243 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6245 key[0].objectid = objectid;
6246 key[0].type = BTRFS_INODE_ITEM_KEY;
6249 sizes[0] = sizeof(struct btrfs_inode_item);
6253 * Start new inodes with an inode_ref. This is slightly more
6254 * efficient for small numbers of hard links since they will
6255 * be packed into one item. Extended refs will kick in if we
6256 * add more hard links than can fit in the ref item.
6258 key[1].objectid = objectid;
6259 key[1].type = BTRFS_INODE_REF_KEY;
6260 key[1].offset = ref_objectid;
6262 sizes[1] = name_len + sizeof(*ref);
6265 location = &BTRFS_I(inode)->location;
6266 location->objectid = objectid;
6267 location->offset = 0;
6268 location->type = BTRFS_INODE_ITEM_KEY;
6270 ret = btrfs_insert_inode_locked(inode);
6276 path->leave_spinning = 1;
6277 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6281 inode_init_owner(inode, dir, mode);
6282 inode_set_bytes(inode, 0);
6284 inode->i_mtime = current_time(inode);
6285 inode->i_atime = inode->i_mtime;
6286 inode->i_ctime = inode->i_mtime;
6287 BTRFS_I(inode)->i_otime = inode->i_mtime;
6289 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6290 struct btrfs_inode_item);
6291 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6292 sizeof(*inode_item));
6293 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6296 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6297 struct btrfs_inode_ref);
6298 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6299 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6300 ptr = (unsigned long)(ref + 1);
6301 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6304 btrfs_mark_buffer_dirty(path->nodes[0]);
6305 btrfs_free_path(path);
6307 btrfs_inherit_iflags(inode, dir);
6309 if (S_ISREG(mode)) {
6310 if (btrfs_test_opt(fs_info, NODATASUM))
6311 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6312 if (btrfs_test_opt(fs_info, NODATACOW))
6313 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6314 BTRFS_INODE_NODATASUM;
6317 inode_tree_add(inode);
6319 trace_btrfs_inode_new(inode);
6320 btrfs_set_inode_last_trans(trans, inode);
6322 btrfs_update_root_times(trans, root);
6324 ret = btrfs_inode_inherit_props(trans, inode, dir);
6327 "error inheriting props for ino %llu (root %llu): %d",
6328 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6333 discard_new_inode(inode);
6336 BTRFS_I(dir)->index_cnt--;
6337 btrfs_free_path(path);
6338 return ERR_PTR(ret);
6341 static inline u8 btrfs_inode_type(struct inode *inode)
6343 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6347 * utility function to add 'inode' into 'parent_inode' with
6348 * a give name and a given sequence number.
6349 * if 'add_backref' is true, also insert a backref from the
6350 * inode to the parent directory.
6352 int btrfs_add_link(struct btrfs_trans_handle *trans,
6353 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6354 const char *name, int name_len, int add_backref, u64 index)
6357 struct btrfs_key key;
6358 struct btrfs_root *root = parent_inode->root;
6359 u64 ino = btrfs_ino(inode);
6360 u64 parent_ino = btrfs_ino(parent_inode);
6362 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6363 memcpy(&key, &inode->root->root_key, sizeof(key));
6366 key.type = BTRFS_INODE_ITEM_KEY;
6370 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6371 ret = btrfs_add_root_ref(trans, key.objectid,
6372 root->root_key.objectid, parent_ino,
6373 index, name, name_len);
6374 } else if (add_backref) {
6375 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6379 /* Nothing to clean up yet */
6383 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6384 btrfs_inode_type(&inode->vfs_inode), index);
6385 if (ret == -EEXIST || ret == -EOVERFLOW)
6388 btrfs_abort_transaction(trans, ret);
6392 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6394 inode_inc_iversion(&parent_inode->vfs_inode);
6395 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6396 current_time(&parent_inode->vfs_inode);
6397 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6399 btrfs_abort_transaction(trans, ret);
6403 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6406 err = btrfs_del_root_ref(trans, key.objectid,
6407 root->root_key.objectid, parent_ino,
6408 &local_index, name, name_len);
6410 } else if (add_backref) {
6414 err = btrfs_del_inode_ref(trans, root, name, name_len,
6415 ino, parent_ino, &local_index);
6420 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6421 struct btrfs_inode *dir, struct dentry *dentry,
6422 struct btrfs_inode *inode, int backref, u64 index)
6424 int err = btrfs_add_link(trans, dir, inode,
6425 dentry->d_name.name, dentry->d_name.len,
6432 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6433 umode_t mode, dev_t rdev)
6435 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6436 struct btrfs_trans_handle *trans;
6437 struct btrfs_root *root = BTRFS_I(dir)->root;
6438 struct inode *inode = NULL;
6444 * 2 for inode item and ref
6446 * 1 for xattr if selinux is on
6448 trans = btrfs_start_transaction(root, 5);
6450 return PTR_ERR(trans);
6452 err = btrfs_find_free_ino(root, &objectid);
6456 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6457 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6459 if (IS_ERR(inode)) {
6460 err = PTR_ERR(inode);
6466 * If the active LSM wants to access the inode during
6467 * d_instantiate it needs these. Smack checks to see
6468 * if the filesystem supports xattrs by looking at the
6471 inode->i_op = &btrfs_special_inode_operations;
6472 init_special_inode(inode, inode->i_mode, rdev);
6474 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6478 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6483 btrfs_update_inode(trans, root, inode);
6484 d_instantiate_new(dentry, inode);
6487 btrfs_end_transaction(trans);
6488 btrfs_btree_balance_dirty(fs_info);
6490 inode_dec_link_count(inode);
6491 discard_new_inode(inode);
6496 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6497 umode_t mode, bool excl)
6499 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6500 struct btrfs_trans_handle *trans;
6501 struct btrfs_root *root = BTRFS_I(dir)->root;
6502 struct inode *inode = NULL;
6508 * 2 for inode item and ref
6510 * 1 for xattr if selinux is on
6512 trans = btrfs_start_transaction(root, 5);
6514 return PTR_ERR(trans);
6516 err = btrfs_find_free_ino(root, &objectid);
6520 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6521 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6523 if (IS_ERR(inode)) {
6524 err = PTR_ERR(inode);
6529 * If the active LSM wants to access the inode during
6530 * d_instantiate it needs these. Smack checks to see
6531 * if the filesystem supports xattrs by looking at the
6534 inode->i_fop = &btrfs_file_operations;
6535 inode->i_op = &btrfs_file_inode_operations;
6536 inode->i_mapping->a_ops = &btrfs_aops;
6538 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6542 err = btrfs_update_inode(trans, root, inode);
6546 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6551 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6552 d_instantiate_new(dentry, inode);
6555 btrfs_end_transaction(trans);
6557 inode_dec_link_count(inode);
6558 discard_new_inode(inode);
6560 btrfs_btree_balance_dirty(fs_info);
6564 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6565 struct dentry *dentry)
6567 struct btrfs_trans_handle *trans = NULL;
6568 struct btrfs_root *root = BTRFS_I(dir)->root;
6569 struct inode *inode = d_inode(old_dentry);
6570 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6575 /* do not allow sys_link's with other subvols of the same device */
6576 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6579 if (inode->i_nlink >= BTRFS_LINK_MAX)
6582 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6587 * 2 items for inode and inode ref
6588 * 2 items for dir items
6589 * 1 item for parent inode
6590 * 1 item for orphan item deletion if O_TMPFILE
6592 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6593 if (IS_ERR(trans)) {
6594 err = PTR_ERR(trans);
6599 /* There are several dir indexes for this inode, clear the cache. */
6600 BTRFS_I(inode)->dir_index = 0ULL;
6602 inode_inc_iversion(inode);
6603 inode->i_ctime = current_time(inode);
6605 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6607 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6613 struct dentry *parent = dentry->d_parent;
6616 err = btrfs_update_inode(trans, root, inode);
6619 if (inode->i_nlink == 1) {
6621 * If new hard link count is 1, it's a file created
6622 * with open(2) O_TMPFILE flag.
6624 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6628 d_instantiate(dentry, inode);
6629 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6631 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6632 err = btrfs_commit_transaction(trans);
6639 btrfs_end_transaction(trans);
6641 inode_dec_link_count(inode);
6644 btrfs_btree_balance_dirty(fs_info);
6648 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6650 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6651 struct inode *inode = NULL;
6652 struct btrfs_trans_handle *trans;
6653 struct btrfs_root *root = BTRFS_I(dir)->root;
6655 int drop_on_err = 0;
6660 * 2 items for inode and ref
6661 * 2 items for dir items
6662 * 1 for xattr if selinux is on
6664 trans = btrfs_start_transaction(root, 5);
6666 return PTR_ERR(trans);
6668 err = btrfs_find_free_ino(root, &objectid);
6672 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6673 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6674 S_IFDIR | mode, &index);
6675 if (IS_ERR(inode)) {
6676 err = PTR_ERR(inode);
6682 /* these must be set before we unlock the inode */
6683 inode->i_op = &btrfs_dir_inode_operations;
6684 inode->i_fop = &btrfs_dir_file_operations;
6686 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6690 btrfs_i_size_write(BTRFS_I(inode), 0);
6691 err = btrfs_update_inode(trans, root, inode);
6695 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6696 dentry->d_name.name,
6697 dentry->d_name.len, 0, index);
6701 d_instantiate_new(dentry, inode);
6705 btrfs_end_transaction(trans);
6707 inode_dec_link_count(inode);
6708 discard_new_inode(inode);
6710 btrfs_btree_balance_dirty(fs_info);
6714 static noinline int uncompress_inline(struct btrfs_path *path,
6716 size_t pg_offset, u64 extent_offset,
6717 struct btrfs_file_extent_item *item)
6720 struct extent_buffer *leaf = path->nodes[0];
6723 unsigned long inline_size;
6727 WARN_ON(pg_offset != 0);
6728 compress_type = btrfs_file_extent_compression(leaf, item);
6729 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6730 inline_size = btrfs_file_extent_inline_item_len(leaf,
6731 btrfs_item_nr(path->slots[0]));
6732 tmp = kmalloc(inline_size, GFP_NOFS);
6735 ptr = btrfs_file_extent_inline_start(item);
6737 read_extent_buffer(leaf, tmp, ptr, inline_size);
6739 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6740 ret = btrfs_decompress(compress_type, tmp, page,
6741 extent_offset, inline_size, max_size);
6744 * decompression code contains a memset to fill in any space between the end
6745 * of the uncompressed data and the end of max_size in case the decompressed
6746 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6747 * the end of an inline extent and the beginning of the next block, so we
6748 * cover that region here.
6751 if (max_size + pg_offset < PAGE_SIZE) {
6752 char *map = kmap(page);
6753 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6761 * a bit scary, this does extent mapping from logical file offset to the disk.
6762 * the ugly parts come from merging extents from the disk with the in-ram
6763 * representation. This gets more complex because of the data=ordered code,
6764 * where the in-ram extents might be locked pending data=ordered completion.
6766 * This also copies inline extents directly into the page.
6768 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6770 size_t pg_offset, u64 start, u64 len,
6773 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6776 u64 extent_start = 0;
6778 u64 objectid = btrfs_ino(inode);
6780 struct btrfs_path *path = NULL;
6781 struct btrfs_root *root = inode->root;
6782 struct btrfs_file_extent_item *item;
6783 struct extent_buffer *leaf;
6784 struct btrfs_key found_key;
6785 struct extent_map *em = NULL;
6786 struct extent_map_tree *em_tree = &inode->extent_tree;
6787 struct extent_io_tree *io_tree = &inode->io_tree;
6788 const bool new_inline = !page || create;
6790 read_lock(&em_tree->lock);
6791 em = lookup_extent_mapping(em_tree, start, len);
6793 em->bdev = fs_info->fs_devices->latest_bdev;
6794 read_unlock(&em_tree->lock);
6797 if (em->start > start || em->start + em->len <= start)
6798 free_extent_map(em);
6799 else if (em->block_start == EXTENT_MAP_INLINE && page)
6800 free_extent_map(em);
6804 em = alloc_extent_map();
6809 em->bdev = fs_info->fs_devices->latest_bdev;
6810 em->start = EXTENT_MAP_HOLE;
6811 em->orig_start = EXTENT_MAP_HOLE;
6813 em->block_len = (u64)-1;
6815 path = btrfs_alloc_path();
6821 /* Chances are we'll be called again, so go ahead and do readahead */
6822 path->reada = READA_FORWARD;
6825 * Unless we're going to uncompress the inline extent, no sleep would
6828 path->leave_spinning = 1;
6830 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6837 if (path->slots[0] == 0)
6842 leaf = path->nodes[0];
6843 item = btrfs_item_ptr(leaf, path->slots[0],
6844 struct btrfs_file_extent_item);
6845 /* are we inside the extent that was found? */
6846 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6847 found_type = found_key.type;
6848 if (found_key.objectid != objectid ||
6849 found_type != BTRFS_EXTENT_DATA_KEY) {
6851 * If we backup past the first extent we want to move forward
6852 * and see if there is an extent in front of us, otherwise we'll
6853 * say there is a hole for our whole search range which can
6860 found_type = btrfs_file_extent_type(leaf, item);
6861 extent_start = found_key.offset;
6862 if (found_type == BTRFS_FILE_EXTENT_REG ||
6863 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6864 extent_end = extent_start +
6865 btrfs_file_extent_num_bytes(leaf, item);
6867 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6869 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6872 size = btrfs_file_extent_ram_bytes(leaf, item);
6873 extent_end = ALIGN(extent_start + size,
6874 fs_info->sectorsize);
6876 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6881 if (start >= extent_end) {
6883 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6884 ret = btrfs_next_leaf(root, path);
6891 leaf = path->nodes[0];
6893 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6894 if (found_key.objectid != objectid ||
6895 found_key.type != BTRFS_EXTENT_DATA_KEY)
6897 if (start + len <= found_key.offset)
6899 if (start > found_key.offset)
6902 em->orig_start = start;
6903 em->len = found_key.offset - start;
6907 btrfs_extent_item_to_extent_map(inode, path, item,
6910 if (found_type == BTRFS_FILE_EXTENT_REG ||
6911 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6913 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6917 size_t extent_offset;
6923 size = btrfs_file_extent_ram_bytes(leaf, item);
6924 extent_offset = page_offset(page) + pg_offset - extent_start;
6925 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6926 size - extent_offset);
6927 em->start = extent_start + extent_offset;
6928 em->len = ALIGN(copy_size, fs_info->sectorsize);
6929 em->orig_block_len = em->len;
6930 em->orig_start = em->start;
6931 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6933 btrfs_set_path_blocking(path);
6934 if (!PageUptodate(page)) {
6935 if (btrfs_file_extent_compression(leaf, item) !=
6936 BTRFS_COMPRESS_NONE) {
6937 ret = uncompress_inline(path, page, pg_offset,
6938 extent_offset, item);
6945 read_extent_buffer(leaf, map + pg_offset, ptr,
6947 if (pg_offset + copy_size < PAGE_SIZE) {
6948 memset(map + pg_offset + copy_size, 0,
6949 PAGE_SIZE - pg_offset -
6954 flush_dcache_page(page);
6956 set_extent_uptodate(io_tree, em->start,
6957 extent_map_end(em) - 1, NULL, GFP_NOFS);
6962 em->orig_start = start;
6965 em->block_start = EXTENT_MAP_HOLE;
6967 btrfs_release_path(path);
6968 if (em->start > start || extent_map_end(em) <= start) {
6970 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6971 em->start, em->len, start, len);
6977 write_lock(&em_tree->lock);
6978 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6979 write_unlock(&em_tree->lock);
6981 btrfs_free_path(path);
6983 trace_btrfs_get_extent(root, inode, em);
6986 free_extent_map(em);
6987 return ERR_PTR(err);
6989 BUG_ON(!em); /* Error is always set */
6993 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6995 size_t pg_offset, u64 start, u64 len,
6998 struct extent_map *em;
6999 struct extent_map *hole_em = NULL;
7000 u64 range_start = start;
7006 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7010 * If our em maps to:
7012 * - a pre-alloc extent,
7013 * there might actually be delalloc bytes behind it.
7015 if (em->block_start != EXTENT_MAP_HOLE &&
7016 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7021 /* check to see if we've wrapped (len == -1 or similar) */
7030 /* ok, we didn't find anything, lets look for delalloc */
7031 found = count_range_bits(&inode->io_tree, &range_start,
7032 end, len, EXTENT_DELALLOC, 1);
7033 found_end = range_start + found;
7034 if (found_end < range_start)
7035 found_end = (u64)-1;
7038 * we didn't find anything useful, return
7039 * the original results from get_extent()
7041 if (range_start > end || found_end <= start) {
7047 /* adjust the range_start to make sure it doesn't
7048 * go backwards from the start they passed in
7050 range_start = max(start, range_start);
7051 found = found_end - range_start;
7054 u64 hole_start = start;
7057 em = alloc_extent_map();
7063 * when btrfs_get_extent can't find anything it
7064 * returns one huge hole
7066 * make sure what it found really fits our range, and
7067 * adjust to make sure it is based on the start from
7071 u64 calc_end = extent_map_end(hole_em);
7073 if (calc_end <= start || (hole_em->start > end)) {
7074 free_extent_map(hole_em);
7077 hole_start = max(hole_em->start, start);
7078 hole_len = calc_end - hole_start;
7082 if (hole_em && range_start > hole_start) {
7083 /* our hole starts before our delalloc, so we
7084 * have to return just the parts of the hole
7085 * that go until the delalloc starts
7087 em->len = min(hole_len,
7088 range_start - hole_start);
7089 em->start = hole_start;
7090 em->orig_start = hole_start;
7092 * don't adjust block start at all,
7093 * it is fixed at EXTENT_MAP_HOLE
7095 em->block_start = hole_em->block_start;
7096 em->block_len = hole_len;
7097 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7098 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7100 em->start = range_start;
7102 em->orig_start = range_start;
7103 em->block_start = EXTENT_MAP_DELALLOC;
7104 em->block_len = found;
7111 free_extent_map(hole_em);
7113 free_extent_map(em);
7114 return ERR_PTR(err);
7119 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7122 const u64 orig_start,
7123 const u64 block_start,
7124 const u64 block_len,
7125 const u64 orig_block_len,
7126 const u64 ram_bytes,
7129 struct extent_map *em = NULL;
7132 if (type != BTRFS_ORDERED_NOCOW) {
7133 em = create_io_em(inode, start, len, orig_start,
7134 block_start, block_len, orig_block_len,
7136 BTRFS_COMPRESS_NONE, /* compress_type */
7141 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7142 len, block_len, type);
7145 free_extent_map(em);
7146 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7147 start + len - 1, 0);
7156 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7159 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7160 struct btrfs_root *root = BTRFS_I(inode)->root;
7161 struct extent_map *em;
7162 struct btrfs_key ins;
7166 alloc_hint = get_extent_allocation_hint(inode, start, len);
7167 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7168 0, alloc_hint, &ins, 1, 1);
7170 return ERR_PTR(ret);
7172 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7173 ins.objectid, ins.offset, ins.offset,
7174 ins.offset, BTRFS_ORDERED_REGULAR);
7175 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7177 btrfs_free_reserved_extent(fs_info, ins.objectid,
7184 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7185 * block must be cow'd
7187 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7188 u64 *orig_start, u64 *orig_block_len,
7191 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7192 struct btrfs_path *path;
7194 struct extent_buffer *leaf;
7195 struct btrfs_root *root = BTRFS_I(inode)->root;
7196 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7197 struct btrfs_file_extent_item *fi;
7198 struct btrfs_key key;
7205 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7207 path = btrfs_alloc_path();
7211 ret = btrfs_lookup_file_extent(NULL, root, path,
7212 btrfs_ino(BTRFS_I(inode)), offset, 0);
7216 slot = path->slots[0];
7219 /* can't find the item, must cow */
7226 leaf = path->nodes[0];
7227 btrfs_item_key_to_cpu(leaf, &key, slot);
7228 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7229 key.type != BTRFS_EXTENT_DATA_KEY) {
7230 /* not our file or wrong item type, must cow */
7234 if (key.offset > offset) {
7235 /* Wrong offset, must cow */
7239 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7240 found_type = btrfs_file_extent_type(leaf, fi);
7241 if (found_type != BTRFS_FILE_EXTENT_REG &&
7242 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7243 /* not a regular extent, must cow */
7247 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7250 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7251 if (extent_end <= offset)
7254 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7255 if (disk_bytenr == 0)
7258 if (btrfs_file_extent_compression(leaf, fi) ||
7259 btrfs_file_extent_encryption(leaf, fi) ||
7260 btrfs_file_extent_other_encoding(leaf, fi))
7264 * Do the same check as in btrfs_cross_ref_exist but without the
7265 * unnecessary search.
7267 if (btrfs_file_extent_generation(leaf, fi) <=
7268 btrfs_root_last_snapshot(&root->root_item))
7271 backref_offset = btrfs_file_extent_offset(leaf, fi);
7274 *orig_start = key.offset - backref_offset;
7275 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7276 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7279 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7282 num_bytes = min(offset + *len, extent_end) - offset;
7283 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7286 range_end = round_up(offset + num_bytes,
7287 root->fs_info->sectorsize) - 1;
7288 ret = test_range_bit(io_tree, offset, range_end,
7289 EXTENT_DELALLOC, 0, NULL);
7296 btrfs_release_path(path);
7299 * look for other files referencing this extent, if we
7300 * find any we must cow
7303 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7304 key.offset - backref_offset, disk_bytenr);
7311 * adjust disk_bytenr and num_bytes to cover just the bytes
7312 * in this extent we are about to write. If there
7313 * are any csums in that range we have to cow in order
7314 * to keep the csums correct
7316 disk_bytenr += backref_offset;
7317 disk_bytenr += offset - key.offset;
7318 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7321 * all of the above have passed, it is safe to overwrite this extent
7327 btrfs_free_path(path);
7331 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7332 struct extent_state **cached_state, int writing)
7334 struct btrfs_ordered_extent *ordered;
7338 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7341 * We're concerned with the entire range that we're going to be
7342 * doing DIO to, so we need to make sure there's no ordered
7343 * extents in this range.
7345 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7346 lockend - lockstart + 1);
7349 * We need to make sure there are no buffered pages in this
7350 * range either, we could have raced between the invalidate in
7351 * generic_file_direct_write and locking the extent. The
7352 * invalidate needs to happen so that reads after a write do not
7356 (!writing || !filemap_range_has_page(inode->i_mapping,
7357 lockstart, lockend)))
7360 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7365 * If we are doing a DIO read and the ordered extent we
7366 * found is for a buffered write, we can not wait for it
7367 * to complete and retry, because if we do so we can
7368 * deadlock with concurrent buffered writes on page
7369 * locks. This happens only if our DIO read covers more
7370 * than one extent map, if at this point has already
7371 * created an ordered extent for a previous extent map
7372 * and locked its range in the inode's io tree, and a
7373 * concurrent write against that previous extent map's
7374 * range and this range started (we unlock the ranges
7375 * in the io tree only when the bios complete and
7376 * buffered writes always lock pages before attempting
7377 * to lock range in the io tree).
7380 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7381 btrfs_start_ordered_extent(inode, ordered, 1);
7384 btrfs_put_ordered_extent(ordered);
7387 * We could trigger writeback for this range (and wait
7388 * for it to complete) and then invalidate the pages for
7389 * this range (through invalidate_inode_pages2_range()),
7390 * but that can lead us to a deadlock with a concurrent
7391 * call to readpages() (a buffered read or a defrag call
7392 * triggered a readahead) on a page lock due to an
7393 * ordered dio extent we created before but did not have
7394 * yet a corresponding bio submitted (whence it can not
7395 * complete), which makes readpages() wait for that
7396 * ordered extent to complete while holding a lock on
7411 /* The callers of this must take lock_extent() */
7412 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7413 u64 orig_start, u64 block_start,
7414 u64 block_len, u64 orig_block_len,
7415 u64 ram_bytes, int compress_type,
7418 struct extent_map_tree *em_tree;
7419 struct extent_map *em;
7420 struct btrfs_root *root = BTRFS_I(inode)->root;
7423 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7424 type == BTRFS_ORDERED_COMPRESSED ||
7425 type == BTRFS_ORDERED_NOCOW ||
7426 type == BTRFS_ORDERED_REGULAR);
7428 em_tree = &BTRFS_I(inode)->extent_tree;
7429 em = alloc_extent_map();
7431 return ERR_PTR(-ENOMEM);
7434 em->orig_start = orig_start;
7436 em->block_len = block_len;
7437 em->block_start = block_start;
7438 em->bdev = root->fs_info->fs_devices->latest_bdev;
7439 em->orig_block_len = orig_block_len;
7440 em->ram_bytes = ram_bytes;
7441 em->generation = -1;
7442 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7443 if (type == BTRFS_ORDERED_PREALLOC) {
7444 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7445 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7446 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7447 em->compress_type = compress_type;
7451 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7452 em->start + em->len - 1, 0);
7453 write_lock(&em_tree->lock);
7454 ret = add_extent_mapping(em_tree, em, 1);
7455 write_unlock(&em_tree->lock);
7457 * The caller has taken lock_extent(), who could race with us
7460 } while (ret == -EEXIST);
7463 free_extent_map(em);
7464 return ERR_PTR(ret);
7467 /* em got 2 refs now, callers needs to do free_extent_map once. */
7472 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7473 struct buffer_head *bh_result,
7474 struct inode *inode,
7477 if (em->block_start == EXTENT_MAP_HOLE ||
7478 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7481 len = min(len, em->len - (start - em->start));
7483 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7485 bh_result->b_size = len;
7486 bh_result->b_bdev = em->bdev;
7487 set_buffer_mapped(bh_result);
7492 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7493 struct buffer_head *bh_result,
7494 struct inode *inode,
7495 struct btrfs_dio_data *dio_data,
7498 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7499 struct extent_map *em = *map;
7503 * We don't allocate a new extent in the following cases
7505 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7507 * 2) The extent is marked as PREALLOC. We're good to go here and can
7508 * just use the extent.
7511 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7512 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7513 em->block_start != EXTENT_MAP_HOLE)) {
7515 u64 block_start, orig_start, orig_block_len, ram_bytes;
7517 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7518 type = BTRFS_ORDERED_PREALLOC;
7520 type = BTRFS_ORDERED_NOCOW;
7521 len = min(len, em->len - (start - em->start));
7522 block_start = em->block_start + (start - em->start);
7524 if (can_nocow_extent(inode, start, &len, &orig_start,
7525 &orig_block_len, &ram_bytes) == 1 &&
7526 btrfs_inc_nocow_writers(fs_info, block_start)) {
7527 struct extent_map *em2;
7529 em2 = btrfs_create_dio_extent(inode, start, len,
7530 orig_start, block_start,
7531 len, orig_block_len,
7533 btrfs_dec_nocow_writers(fs_info, block_start);
7534 if (type == BTRFS_ORDERED_PREALLOC) {
7535 free_extent_map(em);
7539 if (em2 && IS_ERR(em2)) {
7544 * For inode marked NODATACOW or extent marked PREALLOC,
7545 * use the existing or preallocated extent, so does not
7546 * need to adjust btrfs_space_info's bytes_may_use.
7548 btrfs_free_reserved_data_space_noquota(inode, start,
7554 /* this will cow the extent */
7555 len = bh_result->b_size;
7556 free_extent_map(em);
7557 *map = em = btrfs_new_extent_direct(inode, start, len);
7563 len = min(len, em->len - (start - em->start));
7566 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7568 bh_result->b_size = len;
7569 bh_result->b_bdev = em->bdev;
7570 set_buffer_mapped(bh_result);
7572 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7573 set_buffer_new(bh_result);
7576 * Need to update the i_size under the extent lock so buffered
7577 * readers will get the updated i_size when we unlock.
7579 if (!dio_data->overwrite && start + len > i_size_read(inode))
7580 i_size_write(inode, start + len);
7582 WARN_ON(dio_data->reserve < len);
7583 dio_data->reserve -= len;
7584 dio_data->unsubmitted_oe_range_end = start + len;
7585 current->journal_info = dio_data;
7590 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7591 struct buffer_head *bh_result, int create)
7593 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7594 struct extent_map *em;
7595 struct extent_state *cached_state = NULL;
7596 struct btrfs_dio_data *dio_data = NULL;
7597 u64 start = iblock << inode->i_blkbits;
7598 u64 lockstart, lockend;
7599 u64 len = bh_result->b_size;
7600 int unlock_bits = EXTENT_LOCKED;
7604 unlock_bits |= EXTENT_DIRTY;
7606 len = min_t(u64, len, fs_info->sectorsize);
7609 lockend = start + len - 1;
7611 if (current->journal_info) {
7613 * Need to pull our outstanding extents and set journal_info to NULL so
7614 * that anything that needs to check if there's a transaction doesn't get
7617 dio_data = current->journal_info;
7618 current->journal_info = NULL;
7622 * If this errors out it's because we couldn't invalidate pagecache for
7623 * this range and we need to fallback to buffered.
7625 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7631 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7638 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7639 * io. INLINE is special, and we could probably kludge it in here, but
7640 * it's still buffered so for safety lets just fall back to the generic
7643 * For COMPRESSED we _have_ to read the entire extent in so we can
7644 * decompress it, so there will be buffering required no matter what we
7645 * do, so go ahead and fallback to buffered.
7647 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7648 * to buffered IO. Don't blame me, this is the price we pay for using
7651 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7652 em->block_start == EXTENT_MAP_INLINE) {
7653 free_extent_map(em);
7659 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7660 dio_data, start, len);
7664 /* clear and unlock the entire range */
7665 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7666 unlock_bits, 1, 0, &cached_state);
7668 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7670 /* Can be negative only if we read from a hole */
7673 free_extent_map(em);
7677 * We need to unlock only the end area that we aren't using.
7678 * The rest is going to be unlocked by the endio routine.
7680 lockstart = start + bh_result->b_size;
7681 if (lockstart < lockend) {
7682 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7683 lockend, unlock_bits, 1, 0,
7686 free_extent_state(cached_state);
7690 free_extent_map(em);
7695 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7696 unlock_bits, 1, 0, &cached_state);
7699 current->journal_info = dio_data;
7703 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7707 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7710 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7712 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7716 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7721 static int btrfs_check_dio_repairable(struct inode *inode,
7722 struct bio *failed_bio,
7723 struct io_failure_record *failrec,
7726 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7729 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7730 if (num_copies == 1) {
7732 * we only have a single copy of the data, so don't bother with
7733 * all the retry and error correction code that follows. no
7734 * matter what the error is, it is very likely to persist.
7736 btrfs_debug(fs_info,
7737 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7738 num_copies, failrec->this_mirror, failed_mirror);
7742 failrec->failed_mirror = failed_mirror;
7743 failrec->this_mirror++;
7744 if (failrec->this_mirror == failed_mirror)
7745 failrec->this_mirror++;
7747 if (failrec->this_mirror > num_copies) {
7748 btrfs_debug(fs_info,
7749 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7750 num_copies, failrec->this_mirror, failed_mirror);
7757 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7758 struct page *page, unsigned int pgoff,
7759 u64 start, u64 end, int failed_mirror,
7760 bio_end_io_t *repair_endio, void *repair_arg)
7762 struct io_failure_record *failrec;
7763 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7764 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7767 unsigned int read_mode = 0;
7770 blk_status_t status;
7771 struct bio_vec bvec;
7773 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7775 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7777 return errno_to_blk_status(ret);
7779 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7782 free_io_failure(failure_tree, io_tree, failrec);
7783 return BLK_STS_IOERR;
7786 segs = bio_segments(failed_bio);
7787 bio_get_first_bvec(failed_bio, &bvec);
7789 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7790 read_mode |= REQ_FAILFAST_DEV;
7792 isector = start - btrfs_io_bio(failed_bio)->logical;
7793 isector >>= inode->i_sb->s_blocksize_bits;
7794 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7795 pgoff, isector, repair_endio, repair_arg);
7796 bio->bi_opf = REQ_OP_READ | read_mode;
7798 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7799 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7800 read_mode, failrec->this_mirror, failrec->in_validation);
7802 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7804 free_io_failure(failure_tree, io_tree, failrec);
7811 struct btrfs_retry_complete {
7812 struct completion done;
7813 struct inode *inode;
7818 static void btrfs_retry_endio_nocsum(struct bio *bio)
7820 struct btrfs_retry_complete *done = bio->bi_private;
7821 struct inode *inode = done->inode;
7822 struct bio_vec *bvec;
7823 struct extent_io_tree *io_tree, *failure_tree;
7829 ASSERT(bio->bi_vcnt == 1);
7830 io_tree = &BTRFS_I(inode)->io_tree;
7831 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7832 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7835 ASSERT(!bio_flagged(bio, BIO_CLONED));
7836 bio_for_each_segment_all(bvec, bio, i)
7837 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7838 io_tree, done->start, bvec->bv_page,
7839 btrfs_ino(BTRFS_I(inode)), 0);
7841 complete(&done->done);
7845 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7846 struct btrfs_io_bio *io_bio)
7848 struct btrfs_fs_info *fs_info;
7849 struct bio_vec bvec;
7850 struct bvec_iter iter;
7851 struct btrfs_retry_complete done;
7857 blk_status_t err = BLK_STS_OK;
7859 fs_info = BTRFS_I(inode)->root->fs_info;
7860 sectorsize = fs_info->sectorsize;
7862 start = io_bio->logical;
7864 io_bio->bio.bi_iter = io_bio->iter;
7866 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7867 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7868 pgoff = bvec.bv_offset;
7870 next_block_or_try_again:
7873 init_completion(&done.done);
7875 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7876 pgoff, start, start + sectorsize - 1,
7878 btrfs_retry_endio_nocsum, &done);
7884 wait_for_completion_io(&done.done);
7886 if (!done.uptodate) {
7887 /* We might have another mirror, so try again */
7888 goto next_block_or_try_again;
7892 start += sectorsize;
7896 pgoff += sectorsize;
7897 ASSERT(pgoff < PAGE_SIZE);
7898 goto next_block_or_try_again;
7905 static void btrfs_retry_endio(struct bio *bio)
7907 struct btrfs_retry_complete *done = bio->bi_private;
7908 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7909 struct extent_io_tree *io_tree, *failure_tree;
7910 struct inode *inode = done->inode;
7911 struct bio_vec *bvec;
7921 ASSERT(bio->bi_vcnt == 1);
7922 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7924 io_tree = &BTRFS_I(inode)->io_tree;
7925 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7927 ASSERT(!bio_flagged(bio, BIO_CLONED));
7928 bio_for_each_segment_all(bvec, bio, i) {
7929 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7930 bvec->bv_offset, done->start,
7933 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7934 failure_tree, io_tree, done->start,
7936 btrfs_ino(BTRFS_I(inode)),
7942 done->uptodate = uptodate;
7944 complete(&done->done);
7948 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7949 struct btrfs_io_bio *io_bio, blk_status_t err)
7951 struct btrfs_fs_info *fs_info;
7952 struct bio_vec bvec;
7953 struct bvec_iter iter;
7954 struct btrfs_retry_complete done;
7961 bool uptodate = (err == 0);
7963 blk_status_t status;
7965 fs_info = BTRFS_I(inode)->root->fs_info;
7966 sectorsize = fs_info->sectorsize;
7969 start = io_bio->logical;
7971 io_bio->bio.bi_iter = io_bio->iter;
7973 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7974 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7976 pgoff = bvec.bv_offset;
7979 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7980 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7981 bvec.bv_page, pgoff, start, sectorsize);
7988 init_completion(&done.done);
7990 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7991 pgoff, start, start + sectorsize - 1,
7992 io_bio->mirror_num, btrfs_retry_endio,
7999 wait_for_completion_io(&done.done);
8001 if (!done.uptodate) {
8002 /* We might have another mirror, so try again */
8006 offset += sectorsize;
8007 start += sectorsize;
8013 pgoff += sectorsize;
8014 ASSERT(pgoff < PAGE_SIZE);
8022 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8023 struct btrfs_io_bio *io_bio, blk_status_t err)
8025 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8029 return __btrfs_correct_data_nocsum(inode, io_bio);
8033 return __btrfs_subio_endio_read(inode, io_bio, err);
8037 static void btrfs_endio_direct_read(struct bio *bio)
8039 struct btrfs_dio_private *dip = bio->bi_private;
8040 struct inode *inode = dip->inode;
8041 struct bio *dio_bio;
8042 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8043 blk_status_t err = bio->bi_status;
8045 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8046 err = btrfs_subio_endio_read(inode, io_bio, err);
8048 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8049 dip->logical_offset + dip->bytes - 1);
8050 dio_bio = dip->dio_bio;
8054 dio_bio->bi_status = err;
8055 dio_end_io(dio_bio);
8058 io_bio->end_io(io_bio, blk_status_to_errno(err));
8062 static void __endio_write_update_ordered(struct inode *inode,
8063 const u64 offset, const u64 bytes,
8064 const bool uptodate)
8066 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8067 struct btrfs_ordered_extent *ordered = NULL;
8068 struct btrfs_workqueue *wq;
8069 btrfs_work_func_t func;
8070 u64 ordered_offset = offset;
8071 u64 ordered_bytes = bytes;
8074 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8075 wq = fs_info->endio_freespace_worker;
8076 func = btrfs_freespace_write_helper;
8078 wq = fs_info->endio_write_workers;
8079 func = btrfs_endio_write_helper;
8082 while (ordered_offset < offset + bytes) {
8083 last_offset = ordered_offset;
8084 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8088 btrfs_init_work(&ordered->work, func,
8091 btrfs_queue_work(wq, &ordered->work);
8094 * If btrfs_dec_test_ordered_pending does not find any ordered
8095 * extent in the range, we can exit.
8097 if (ordered_offset == last_offset)
8100 * Our bio might span multiple ordered extents. In this case
8101 * we keep goin until we have accounted the whole dio.
8103 if (ordered_offset < offset + bytes) {
8104 ordered_bytes = offset + bytes - ordered_offset;
8110 static void btrfs_endio_direct_write(struct bio *bio)
8112 struct btrfs_dio_private *dip = bio->bi_private;
8113 struct bio *dio_bio = dip->dio_bio;
8115 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8116 dip->bytes, !bio->bi_status);
8120 dio_bio->bi_status = bio->bi_status;
8121 dio_end_io(dio_bio);
8125 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8126 struct bio *bio, u64 offset)
8128 struct inode *inode = private_data;
8130 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8131 BUG_ON(ret); /* -ENOMEM */
8135 static void btrfs_end_dio_bio(struct bio *bio)
8137 struct btrfs_dio_private *dip = bio->bi_private;
8138 blk_status_t err = bio->bi_status;
8141 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8142 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8143 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8145 (unsigned long long)bio->bi_iter.bi_sector,
8146 bio->bi_iter.bi_size, err);
8148 if (dip->subio_endio)
8149 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8153 * We want to perceive the errors flag being set before
8154 * decrementing the reference count. We don't need a barrier
8155 * since atomic operations with a return value are fully
8156 * ordered as per atomic_t.txt
8161 /* if there are more bios still pending for this dio, just exit */
8162 if (!atomic_dec_and_test(&dip->pending_bios))
8166 bio_io_error(dip->orig_bio);
8168 dip->dio_bio->bi_status = BLK_STS_OK;
8169 bio_endio(dip->orig_bio);
8175 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8176 struct btrfs_dio_private *dip,
8180 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8181 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8185 * We load all the csum data we need when we submit
8186 * the first bio to reduce the csum tree search and
8189 if (dip->logical_offset == file_offset) {
8190 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8196 if (bio == dip->orig_bio)
8199 file_offset -= dip->logical_offset;
8200 file_offset >>= inode->i_sb->s_blocksize_bits;
8201 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8206 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8207 struct inode *inode, u64 file_offset, int async_submit)
8209 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8210 struct btrfs_dio_private *dip = bio->bi_private;
8211 bool write = bio_op(bio) == REQ_OP_WRITE;
8214 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8216 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8219 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8224 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8227 if (write && async_submit) {
8228 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8230 btrfs_submit_bio_start_direct_io);
8234 * If we aren't doing async submit, calculate the csum of the
8237 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8241 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8247 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8252 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8254 struct inode *inode = dip->inode;
8255 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8257 struct bio *orig_bio = dip->orig_bio;
8258 u64 start_sector = orig_bio->bi_iter.bi_sector;
8259 u64 file_offset = dip->logical_offset;
8261 int async_submit = 0;
8263 int clone_offset = 0;
8266 blk_status_t status;
8268 map_length = orig_bio->bi_iter.bi_size;
8269 submit_len = map_length;
8270 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8271 &map_length, NULL, 0);
8275 if (map_length >= submit_len) {
8277 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8281 /* async crcs make it difficult to collect full stripe writes. */
8282 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8288 ASSERT(map_length <= INT_MAX);
8289 atomic_inc(&dip->pending_bios);
8291 clone_len = min_t(int, submit_len, map_length);
8294 * This will never fail as it's passing GPF_NOFS and
8295 * the allocation is backed by btrfs_bioset.
8297 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8299 bio->bi_private = dip;
8300 bio->bi_end_io = btrfs_end_dio_bio;
8301 btrfs_io_bio(bio)->logical = file_offset;
8303 ASSERT(submit_len >= clone_len);
8304 submit_len -= clone_len;
8305 if (submit_len == 0)
8309 * Increase the count before we submit the bio so we know
8310 * the end IO handler won't happen before we increase the
8311 * count. Otherwise, the dip might get freed before we're
8312 * done setting it up.
8314 atomic_inc(&dip->pending_bios);
8316 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8320 atomic_dec(&dip->pending_bios);
8324 clone_offset += clone_len;
8325 start_sector += clone_len >> 9;
8326 file_offset += clone_len;
8328 map_length = submit_len;
8329 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8330 start_sector << 9, &map_length, NULL, 0);
8333 } while (submit_len > 0);
8336 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8344 * Before atomic variable goto zero, we must make sure dip->errors is
8345 * perceived to be set. This ordering is ensured by the fact that an
8346 * atomic operations with a return value are fully ordered as per
8349 if (atomic_dec_and_test(&dip->pending_bios))
8350 bio_io_error(dip->orig_bio);
8352 /* bio_end_io() will handle error, so we needn't return it */
8356 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8359 struct btrfs_dio_private *dip = NULL;
8360 struct bio *bio = NULL;
8361 struct btrfs_io_bio *io_bio;
8362 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8365 bio = btrfs_bio_clone(dio_bio);
8367 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8373 dip->private = dio_bio->bi_private;
8375 dip->logical_offset = file_offset;
8376 dip->bytes = dio_bio->bi_iter.bi_size;
8377 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8378 bio->bi_private = dip;
8379 dip->orig_bio = bio;
8380 dip->dio_bio = dio_bio;
8381 atomic_set(&dip->pending_bios, 0);
8382 io_bio = btrfs_io_bio(bio);
8383 io_bio->logical = file_offset;
8386 bio->bi_end_io = btrfs_endio_direct_write;
8388 bio->bi_end_io = btrfs_endio_direct_read;
8389 dip->subio_endio = btrfs_subio_endio_read;
8393 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8394 * even if we fail to submit a bio, because in such case we do the
8395 * corresponding error handling below and it must not be done a second
8396 * time by btrfs_direct_IO().
8399 struct btrfs_dio_data *dio_data = current->journal_info;
8401 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8403 dio_data->unsubmitted_oe_range_start =
8404 dio_data->unsubmitted_oe_range_end;
8407 ret = btrfs_submit_direct_hook(dip);
8412 io_bio->end_io(io_bio, ret);
8416 * If we arrived here it means either we failed to submit the dip
8417 * or we either failed to clone the dio_bio or failed to allocate the
8418 * dip. If we cloned the dio_bio and allocated the dip, we can just
8419 * call bio_endio against our io_bio so that we get proper resource
8420 * cleanup if we fail to submit the dip, otherwise, we must do the
8421 * same as btrfs_endio_direct_[write|read] because we can't call these
8422 * callbacks - they require an allocated dip and a clone of dio_bio.
8427 * The end io callbacks free our dip, do the final put on bio
8428 * and all the cleanup and final put for dio_bio (through
8435 __endio_write_update_ordered(inode,
8437 dio_bio->bi_iter.bi_size,
8440 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8441 file_offset + dio_bio->bi_iter.bi_size - 1);
8443 dio_bio->bi_status = BLK_STS_IOERR;
8445 * Releases and cleans up our dio_bio, no need to bio_put()
8446 * nor bio_endio()/bio_io_error() against dio_bio.
8448 dio_end_io(dio_bio);
8455 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8456 const struct iov_iter *iter, loff_t offset)
8460 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8461 ssize_t retval = -EINVAL;
8463 if (offset & blocksize_mask)
8466 if (iov_iter_alignment(iter) & blocksize_mask)
8469 /* If this is a write we don't need to check anymore */
8470 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8473 * Check to make sure we don't have duplicate iov_base's in this
8474 * iovec, if so return EINVAL, otherwise we'll get csum errors
8475 * when reading back.
8477 for (seg = 0; seg < iter->nr_segs; seg++) {
8478 for (i = seg + 1; i < iter->nr_segs; i++) {
8479 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8488 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8490 struct file *file = iocb->ki_filp;
8491 struct inode *inode = file->f_mapping->host;
8492 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8493 struct btrfs_dio_data dio_data = { 0 };
8494 struct extent_changeset *data_reserved = NULL;
8495 loff_t offset = iocb->ki_pos;
8499 bool relock = false;
8502 if (check_direct_IO(fs_info, iter, offset))
8505 inode_dio_begin(inode);
8508 * The generic stuff only does filemap_write_and_wait_range, which
8509 * isn't enough if we've written compressed pages to this area, so
8510 * we need to flush the dirty pages again to make absolutely sure
8511 * that any outstanding dirty pages are on disk.
8513 count = iov_iter_count(iter);
8514 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8515 &BTRFS_I(inode)->runtime_flags))
8516 filemap_fdatawrite_range(inode->i_mapping, offset,
8517 offset + count - 1);
8519 if (iov_iter_rw(iter) == WRITE) {
8521 * If the write DIO is beyond the EOF, we need update
8522 * the isize, but it is protected by i_mutex. So we can
8523 * not unlock the i_mutex at this case.
8525 if (offset + count <= inode->i_size) {
8526 dio_data.overwrite = 1;
8527 inode_unlock(inode);
8529 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8533 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8539 * We need to know how many extents we reserved so that we can
8540 * do the accounting properly if we go over the number we
8541 * originally calculated. Abuse current->journal_info for this.
8543 dio_data.reserve = round_up(count,
8544 fs_info->sectorsize);
8545 dio_data.unsubmitted_oe_range_start = (u64)offset;
8546 dio_data.unsubmitted_oe_range_end = (u64)offset;
8547 current->journal_info = &dio_data;
8548 down_read(&BTRFS_I(inode)->dio_sem);
8549 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8550 &BTRFS_I(inode)->runtime_flags)) {
8551 inode_dio_end(inode);
8552 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8556 ret = __blockdev_direct_IO(iocb, inode,
8557 fs_info->fs_devices->latest_bdev,
8558 iter, btrfs_get_blocks_direct, NULL,
8559 btrfs_submit_direct, flags);
8560 if (iov_iter_rw(iter) == WRITE) {
8561 up_read(&BTRFS_I(inode)->dio_sem);
8562 current->journal_info = NULL;
8563 if (ret < 0 && ret != -EIOCBQUEUED) {
8564 if (dio_data.reserve)
8565 btrfs_delalloc_release_space(inode, data_reserved,
8566 offset, dio_data.reserve, true);
8568 * On error we might have left some ordered extents
8569 * without submitting corresponding bios for them, so
8570 * cleanup them up to avoid other tasks getting them
8571 * and waiting for them to complete forever.
8573 if (dio_data.unsubmitted_oe_range_start <
8574 dio_data.unsubmitted_oe_range_end)
8575 __endio_write_update_ordered(inode,
8576 dio_data.unsubmitted_oe_range_start,
8577 dio_data.unsubmitted_oe_range_end -
8578 dio_data.unsubmitted_oe_range_start,
8580 } else if (ret >= 0 && (size_t)ret < count)
8581 btrfs_delalloc_release_space(inode, data_reserved,
8582 offset, count - (size_t)ret, true);
8583 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8587 inode_dio_end(inode);
8591 extent_changeset_free(data_reserved);
8595 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8597 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8598 __u64 start, __u64 len)
8602 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8606 return extent_fiemap(inode, fieinfo, start, len);
8609 int btrfs_readpage(struct file *file, struct page *page)
8611 struct extent_io_tree *tree;
8612 tree = &BTRFS_I(page->mapping->host)->io_tree;
8613 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8616 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8618 struct inode *inode = page->mapping->host;
8621 if (current->flags & PF_MEMALLOC) {
8622 redirty_page_for_writepage(wbc, page);
8628 * If we are under memory pressure we will call this directly from the
8629 * VM, we need to make sure we have the inode referenced for the ordered
8630 * extent. If not just return like we didn't do anything.
8632 if (!igrab(inode)) {
8633 redirty_page_for_writepage(wbc, page);
8634 return AOP_WRITEPAGE_ACTIVATE;
8636 ret = extent_write_full_page(page, wbc);
8637 btrfs_add_delayed_iput(inode);
8641 static int btrfs_writepages(struct address_space *mapping,
8642 struct writeback_control *wbc)
8644 return extent_writepages(mapping, wbc);
8648 btrfs_readpages(struct file *file, struct address_space *mapping,
8649 struct list_head *pages, unsigned nr_pages)
8651 return extent_readpages(mapping, pages, nr_pages);
8654 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8656 int ret = try_release_extent_mapping(page, gfp_flags);
8658 ClearPagePrivate(page);
8659 set_page_private(page, 0);
8665 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8667 if (PageWriteback(page) || PageDirty(page))
8669 return __btrfs_releasepage(page, gfp_flags);
8672 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8673 unsigned int length)
8675 struct inode *inode = page->mapping->host;
8676 struct extent_io_tree *tree;
8677 struct btrfs_ordered_extent *ordered;
8678 struct extent_state *cached_state = NULL;
8679 u64 page_start = page_offset(page);
8680 u64 page_end = page_start + PAGE_SIZE - 1;
8683 int inode_evicting = inode->i_state & I_FREEING;
8686 * we have the page locked, so new writeback can't start,
8687 * and the dirty bit won't be cleared while we are here.
8689 * Wait for IO on this page so that we can safely clear
8690 * the PagePrivate2 bit and do ordered accounting
8692 wait_on_page_writeback(page);
8694 tree = &BTRFS_I(inode)->io_tree;
8696 btrfs_releasepage(page, GFP_NOFS);
8700 if (!inode_evicting)
8701 lock_extent_bits(tree, page_start, page_end, &cached_state);
8704 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8705 page_end - start + 1);
8707 end = min(page_end, ordered->file_offset + ordered->len - 1);
8709 * IO on this page will never be started, so we need
8710 * to account for any ordered extents now
8712 if (!inode_evicting)
8713 clear_extent_bit(tree, start, end,
8714 EXTENT_DIRTY | EXTENT_DELALLOC |
8715 EXTENT_DELALLOC_NEW |
8716 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8717 EXTENT_DEFRAG, 1, 0, &cached_state);
8719 * whoever cleared the private bit is responsible
8720 * for the finish_ordered_io
8722 if (TestClearPagePrivate2(page)) {
8723 struct btrfs_ordered_inode_tree *tree;
8726 tree = &BTRFS_I(inode)->ordered_tree;
8728 spin_lock_irq(&tree->lock);
8729 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8730 new_len = start - ordered->file_offset;
8731 if (new_len < ordered->truncated_len)
8732 ordered->truncated_len = new_len;
8733 spin_unlock_irq(&tree->lock);
8735 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8737 end - start + 1, 1))
8738 btrfs_finish_ordered_io(ordered);
8740 btrfs_put_ordered_extent(ordered);
8741 if (!inode_evicting) {
8742 cached_state = NULL;
8743 lock_extent_bits(tree, start, end,
8748 if (start < page_end)
8753 * Qgroup reserved space handler
8754 * Page here will be either
8755 * 1) Already written to disk
8756 * In this case, its reserved space is released from data rsv map
8757 * and will be freed by delayed_ref handler finally.
8758 * So even we call qgroup_free_data(), it won't decrease reserved
8760 * 2) Not written to disk
8761 * This means the reserved space should be freed here. However,
8762 * if a truncate invalidates the page (by clearing PageDirty)
8763 * and the page is accounted for while allocating extent
8764 * in btrfs_check_data_free_space() we let delayed_ref to
8765 * free the entire extent.
8767 if (PageDirty(page))
8768 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8769 if (!inode_evicting) {
8770 clear_extent_bit(tree, page_start, page_end,
8771 EXTENT_LOCKED | EXTENT_DIRTY |
8772 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8773 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8776 __btrfs_releasepage(page, GFP_NOFS);
8779 ClearPageChecked(page);
8780 if (PagePrivate(page)) {
8781 ClearPagePrivate(page);
8782 set_page_private(page, 0);
8788 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8789 * called from a page fault handler when a page is first dirtied. Hence we must
8790 * be careful to check for EOF conditions here. We set the page up correctly
8791 * for a written page which means we get ENOSPC checking when writing into
8792 * holes and correct delalloc and unwritten extent mapping on filesystems that
8793 * support these features.
8795 * We are not allowed to take the i_mutex here so we have to play games to
8796 * protect against truncate races as the page could now be beyond EOF. Because
8797 * truncate_setsize() writes the inode size before removing pages, once we have
8798 * the page lock we can determine safely if the page is beyond EOF. If it is not
8799 * beyond EOF, then the page is guaranteed safe against truncation until we
8802 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8804 struct page *page = vmf->page;
8805 struct inode *inode = file_inode(vmf->vma->vm_file);
8806 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8807 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8808 struct btrfs_ordered_extent *ordered;
8809 struct extent_state *cached_state = NULL;
8810 struct extent_changeset *data_reserved = NULL;
8812 unsigned long zero_start;
8822 reserved_space = PAGE_SIZE;
8824 sb_start_pagefault(inode->i_sb);
8825 page_start = page_offset(page);
8826 page_end = page_start + PAGE_SIZE - 1;
8830 * Reserving delalloc space after obtaining the page lock can lead to
8831 * deadlock. For example, if a dirty page is locked by this function
8832 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8833 * dirty page write out, then the btrfs_writepage() function could
8834 * end up waiting indefinitely to get a lock on the page currently
8835 * being processed by btrfs_page_mkwrite() function.
8837 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8840 ret2 = file_update_time(vmf->vma->vm_file);
8844 ret = vmf_error(ret2);
8850 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8853 size = i_size_read(inode);
8855 if ((page->mapping != inode->i_mapping) ||
8856 (page_start >= size)) {
8857 /* page got truncated out from underneath us */
8860 wait_on_page_writeback(page);
8862 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8863 set_page_extent_mapped(page);
8866 * we can't set the delalloc bits if there are pending ordered
8867 * extents. Drop our locks and wait for them to finish
8869 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8872 unlock_extent_cached(io_tree, page_start, page_end,
8875 btrfs_start_ordered_extent(inode, ordered, 1);
8876 btrfs_put_ordered_extent(ordered);
8880 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8881 reserved_space = round_up(size - page_start,
8882 fs_info->sectorsize);
8883 if (reserved_space < PAGE_SIZE) {
8884 end = page_start + reserved_space - 1;
8885 btrfs_delalloc_release_space(inode, data_reserved,
8886 page_start, PAGE_SIZE - reserved_space,
8892 * page_mkwrite gets called when the page is firstly dirtied after it's
8893 * faulted in, but write(2) could also dirty a page and set delalloc
8894 * bits, thus in this case for space account reason, we still need to
8895 * clear any delalloc bits within this page range since we have to
8896 * reserve data&meta space before lock_page() (see above comments).
8898 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8899 EXTENT_DIRTY | EXTENT_DELALLOC |
8900 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8901 0, 0, &cached_state);
8903 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8906 unlock_extent_cached(io_tree, page_start, page_end,
8908 ret = VM_FAULT_SIGBUS;
8913 /* page is wholly or partially inside EOF */
8914 if (page_start + PAGE_SIZE > size)
8915 zero_start = size & ~PAGE_MASK;
8917 zero_start = PAGE_SIZE;
8919 if (zero_start != PAGE_SIZE) {
8921 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8922 flush_dcache_page(page);
8925 ClearPageChecked(page);
8926 set_page_dirty(page);
8927 SetPageUptodate(page);
8929 BTRFS_I(inode)->last_trans = fs_info->generation;
8930 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8931 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8933 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8936 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8937 sb_end_pagefault(inode->i_sb);
8938 extent_changeset_free(data_reserved);
8939 return VM_FAULT_LOCKED;
8945 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8946 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8947 reserved_space, (ret != 0));
8949 sb_end_pagefault(inode->i_sb);
8950 extent_changeset_free(data_reserved);
8954 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8956 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8957 struct btrfs_root *root = BTRFS_I(inode)->root;
8958 struct btrfs_block_rsv *rsv;
8960 struct btrfs_trans_handle *trans;
8961 u64 mask = fs_info->sectorsize - 1;
8962 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8964 if (!skip_writeback) {
8965 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8972 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8973 * things going on here:
8975 * 1) We need to reserve space to update our inode.
8977 * 2) We need to have something to cache all the space that is going to
8978 * be free'd up by the truncate operation, but also have some slack
8979 * space reserved in case it uses space during the truncate (thank you
8980 * very much snapshotting).
8982 * And we need these to be separate. The fact is we can use a lot of
8983 * space doing the truncate, and we have no earthly idea how much space
8984 * we will use, so we need the truncate reservation to be separate so it
8985 * doesn't end up using space reserved for updating the inode. We also
8986 * need to be able to stop the transaction and start a new one, which
8987 * means we need to be able to update the inode several times, and we
8988 * have no idea of knowing how many times that will be, so we can't just
8989 * reserve 1 item for the entirety of the operation, so that has to be
8990 * done separately as well.
8992 * So that leaves us with
8994 * 1) rsv - for the truncate reservation, which we will steal from the
8995 * transaction reservation.
8996 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8997 * updating the inode.
8999 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9002 rsv->size = min_size;
9006 * 1 for the truncate slack space
9007 * 1 for updating the inode.
9009 trans = btrfs_start_transaction(root, 2);
9010 if (IS_ERR(trans)) {
9011 ret = PTR_ERR(trans);
9015 /* Migrate the slack space for the truncate to our reserve */
9016 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9021 * So if we truncate and then write and fsync we normally would just
9022 * write the extents that changed, which is a problem if we need to
9023 * first truncate that entire inode. So set this flag so we write out
9024 * all of the extents in the inode to the sync log so we're completely
9027 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9028 trans->block_rsv = rsv;
9031 ret = btrfs_truncate_inode_items(trans, root, inode,
9033 BTRFS_EXTENT_DATA_KEY);
9034 trans->block_rsv = &fs_info->trans_block_rsv;
9035 if (ret != -ENOSPC && ret != -EAGAIN)
9038 ret = btrfs_update_inode(trans, root, inode);
9042 btrfs_end_transaction(trans);
9043 btrfs_btree_balance_dirty(fs_info);
9045 trans = btrfs_start_transaction(root, 2);
9046 if (IS_ERR(trans)) {
9047 ret = PTR_ERR(trans);
9052 btrfs_block_rsv_release(fs_info, rsv, -1);
9053 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9054 rsv, min_size, false);
9055 BUG_ON(ret); /* shouldn't happen */
9056 trans->block_rsv = rsv;
9060 * We can't call btrfs_truncate_block inside a trans handle as we could
9061 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9062 * we've truncated everything except the last little bit, and can do
9063 * btrfs_truncate_block and then update the disk_i_size.
9065 if (ret == NEED_TRUNCATE_BLOCK) {
9066 btrfs_end_transaction(trans);
9067 btrfs_btree_balance_dirty(fs_info);
9069 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9072 trans = btrfs_start_transaction(root, 1);
9073 if (IS_ERR(trans)) {
9074 ret = PTR_ERR(trans);
9077 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9083 trans->block_rsv = &fs_info->trans_block_rsv;
9084 ret2 = btrfs_update_inode(trans, root, inode);
9088 ret2 = btrfs_end_transaction(trans);
9091 btrfs_btree_balance_dirty(fs_info);
9094 btrfs_free_block_rsv(fs_info, rsv);
9100 * create a new subvolume directory/inode (helper for the ioctl).
9102 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9103 struct btrfs_root *new_root,
9104 struct btrfs_root *parent_root,
9107 struct inode *inode;
9111 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9112 new_dirid, new_dirid,
9113 S_IFDIR | (~current_umask() & S_IRWXUGO),
9116 return PTR_ERR(inode);
9117 inode->i_op = &btrfs_dir_inode_operations;
9118 inode->i_fop = &btrfs_dir_file_operations;
9120 set_nlink(inode, 1);
9121 btrfs_i_size_write(BTRFS_I(inode), 0);
9122 unlock_new_inode(inode);
9124 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9126 btrfs_err(new_root->fs_info,
9127 "error inheriting subvolume %llu properties: %d",
9128 new_root->root_key.objectid, err);
9130 err = btrfs_update_inode(trans, new_root, inode);
9136 struct inode *btrfs_alloc_inode(struct super_block *sb)
9138 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9139 struct btrfs_inode *ei;
9140 struct inode *inode;
9142 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9149 ei->last_sub_trans = 0;
9150 ei->logged_trans = 0;
9151 ei->delalloc_bytes = 0;
9152 ei->new_delalloc_bytes = 0;
9153 ei->defrag_bytes = 0;
9154 ei->disk_i_size = 0;
9157 ei->index_cnt = (u64)-1;
9159 ei->last_unlink_trans = 0;
9160 ei->last_log_commit = 0;
9162 spin_lock_init(&ei->lock);
9163 ei->outstanding_extents = 0;
9164 if (sb->s_magic != BTRFS_TEST_MAGIC)
9165 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9166 BTRFS_BLOCK_RSV_DELALLOC);
9167 ei->runtime_flags = 0;
9168 ei->prop_compress = BTRFS_COMPRESS_NONE;
9169 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9171 ei->delayed_node = NULL;
9173 ei->i_otime.tv_sec = 0;
9174 ei->i_otime.tv_nsec = 0;
9176 inode = &ei->vfs_inode;
9177 extent_map_tree_init(&ei->extent_tree);
9178 extent_io_tree_init(&ei->io_tree, inode);
9179 extent_io_tree_init(&ei->io_failure_tree, inode);
9180 ei->io_tree.track_uptodate = 1;
9181 ei->io_failure_tree.track_uptodate = 1;
9182 atomic_set(&ei->sync_writers, 0);
9183 mutex_init(&ei->log_mutex);
9184 mutex_init(&ei->delalloc_mutex);
9185 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9186 INIT_LIST_HEAD(&ei->delalloc_inodes);
9187 INIT_LIST_HEAD(&ei->delayed_iput);
9188 RB_CLEAR_NODE(&ei->rb_node);
9189 init_rwsem(&ei->dio_sem);
9194 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9195 void btrfs_test_destroy_inode(struct inode *inode)
9197 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9198 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9202 static void btrfs_i_callback(struct rcu_head *head)
9204 struct inode *inode = container_of(head, struct inode, i_rcu);
9205 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9208 void btrfs_destroy_inode(struct inode *inode)
9210 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9211 struct btrfs_ordered_extent *ordered;
9212 struct btrfs_root *root = BTRFS_I(inode)->root;
9214 WARN_ON(!hlist_empty(&inode->i_dentry));
9215 WARN_ON(inode->i_data.nrpages);
9216 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9217 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9218 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9219 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9220 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9221 WARN_ON(BTRFS_I(inode)->csum_bytes);
9222 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9225 * This can happen where we create an inode, but somebody else also
9226 * created the same inode and we need to destroy the one we already
9233 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9238 "found ordered extent %llu %llu on inode cleanup",
9239 ordered->file_offset, ordered->len);
9240 btrfs_remove_ordered_extent(inode, ordered);
9241 btrfs_put_ordered_extent(ordered);
9242 btrfs_put_ordered_extent(ordered);
9245 btrfs_qgroup_check_reserved_leak(inode);
9246 inode_tree_del(inode);
9247 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9249 call_rcu(&inode->i_rcu, btrfs_i_callback);
9252 int btrfs_drop_inode(struct inode *inode)
9254 struct btrfs_root *root = BTRFS_I(inode)->root;
9259 /* the snap/subvol tree is on deleting */
9260 if (btrfs_root_refs(&root->root_item) == 0)
9263 return generic_drop_inode(inode);
9266 static void init_once(void *foo)
9268 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9270 inode_init_once(&ei->vfs_inode);
9273 void __cold btrfs_destroy_cachep(void)
9276 * Make sure all delayed rcu free inodes are flushed before we
9280 kmem_cache_destroy(btrfs_inode_cachep);
9281 kmem_cache_destroy(btrfs_trans_handle_cachep);
9282 kmem_cache_destroy(btrfs_path_cachep);
9283 kmem_cache_destroy(btrfs_free_space_cachep);
9286 int __init btrfs_init_cachep(void)
9288 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9289 sizeof(struct btrfs_inode), 0,
9290 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9292 if (!btrfs_inode_cachep)
9295 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9296 sizeof(struct btrfs_trans_handle), 0,
9297 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9298 if (!btrfs_trans_handle_cachep)
9301 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9302 sizeof(struct btrfs_path), 0,
9303 SLAB_MEM_SPREAD, NULL);
9304 if (!btrfs_path_cachep)
9307 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9308 sizeof(struct btrfs_free_space), 0,
9309 SLAB_MEM_SPREAD, NULL);
9310 if (!btrfs_free_space_cachep)
9315 btrfs_destroy_cachep();
9319 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9320 u32 request_mask, unsigned int flags)
9323 struct inode *inode = d_inode(path->dentry);
9324 u32 blocksize = inode->i_sb->s_blocksize;
9325 u32 bi_flags = BTRFS_I(inode)->flags;
9327 stat->result_mask |= STATX_BTIME;
9328 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9329 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9330 if (bi_flags & BTRFS_INODE_APPEND)
9331 stat->attributes |= STATX_ATTR_APPEND;
9332 if (bi_flags & BTRFS_INODE_COMPRESS)
9333 stat->attributes |= STATX_ATTR_COMPRESSED;
9334 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9335 stat->attributes |= STATX_ATTR_IMMUTABLE;
9336 if (bi_flags & BTRFS_INODE_NODUMP)
9337 stat->attributes |= STATX_ATTR_NODUMP;
9339 stat->attributes_mask |= (STATX_ATTR_APPEND |
9340 STATX_ATTR_COMPRESSED |
9341 STATX_ATTR_IMMUTABLE |
9344 generic_fillattr(inode, stat);
9345 stat->dev = BTRFS_I(inode)->root->anon_dev;
9347 spin_lock(&BTRFS_I(inode)->lock);
9348 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9349 spin_unlock(&BTRFS_I(inode)->lock);
9350 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9351 ALIGN(delalloc_bytes, blocksize)) >> 9;
9355 static int btrfs_rename_exchange(struct inode *old_dir,
9356 struct dentry *old_dentry,
9357 struct inode *new_dir,
9358 struct dentry *new_dentry)
9360 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9361 struct btrfs_trans_handle *trans;
9362 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9363 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9364 struct inode *new_inode = new_dentry->d_inode;
9365 struct inode *old_inode = old_dentry->d_inode;
9366 struct timespec64 ctime = current_time(old_inode);
9367 struct dentry *parent;
9368 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9369 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9374 bool root_log_pinned = false;
9375 bool dest_log_pinned = false;
9376 struct btrfs_log_ctx ctx_root;
9377 struct btrfs_log_ctx ctx_dest;
9378 bool sync_log_root = false;
9379 bool sync_log_dest = false;
9380 bool commit_transaction = false;
9382 /* we only allow rename subvolume link between subvolumes */
9383 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9386 btrfs_init_log_ctx(&ctx_root, old_inode);
9387 btrfs_init_log_ctx(&ctx_dest, new_inode);
9389 /* close the race window with snapshot create/destroy ioctl */
9390 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9391 down_read(&fs_info->subvol_sem);
9392 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9393 down_read(&fs_info->subvol_sem);
9396 * We want to reserve the absolute worst case amount of items. So if
9397 * both inodes are subvols and we need to unlink them then that would
9398 * require 4 item modifications, but if they are both normal inodes it
9399 * would require 5 item modifications, so we'll assume their normal
9400 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9401 * should cover the worst case number of items we'll modify.
9403 trans = btrfs_start_transaction(root, 12);
9404 if (IS_ERR(trans)) {
9405 ret = PTR_ERR(trans);
9410 * We need to find a free sequence number both in the source and
9411 * in the destination directory for the exchange.
9413 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9416 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9420 BTRFS_I(old_inode)->dir_index = 0ULL;
9421 BTRFS_I(new_inode)->dir_index = 0ULL;
9423 /* Reference for the source. */
9424 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9425 /* force full log commit if subvolume involved. */
9426 btrfs_set_log_full_commit(fs_info, trans);
9428 btrfs_pin_log_trans(root);
9429 root_log_pinned = true;
9430 ret = btrfs_insert_inode_ref(trans, dest,
9431 new_dentry->d_name.name,
9432 new_dentry->d_name.len,
9434 btrfs_ino(BTRFS_I(new_dir)),
9440 /* And now for the dest. */
9441 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9442 /* force full log commit if subvolume involved. */
9443 btrfs_set_log_full_commit(fs_info, trans);
9445 btrfs_pin_log_trans(dest);
9446 dest_log_pinned = true;
9447 ret = btrfs_insert_inode_ref(trans, root,
9448 old_dentry->d_name.name,
9449 old_dentry->d_name.len,
9451 btrfs_ino(BTRFS_I(old_dir)),
9457 /* Update inode version and ctime/mtime. */
9458 inode_inc_iversion(old_dir);
9459 inode_inc_iversion(new_dir);
9460 inode_inc_iversion(old_inode);
9461 inode_inc_iversion(new_inode);
9462 old_dir->i_ctime = old_dir->i_mtime = ctime;
9463 new_dir->i_ctime = new_dir->i_mtime = ctime;
9464 old_inode->i_ctime = ctime;
9465 new_inode->i_ctime = ctime;
9467 if (old_dentry->d_parent != new_dentry->d_parent) {
9468 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9469 BTRFS_I(old_inode), 1);
9470 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9471 BTRFS_I(new_inode), 1);
9474 /* src is a subvolume */
9475 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9476 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9477 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9478 old_dentry->d_name.name,
9479 old_dentry->d_name.len);
9480 } else { /* src is an inode */
9481 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9482 BTRFS_I(old_dentry->d_inode),
9483 old_dentry->d_name.name,
9484 old_dentry->d_name.len);
9486 ret = btrfs_update_inode(trans, root, old_inode);
9489 btrfs_abort_transaction(trans, ret);
9493 /* dest is a subvolume */
9494 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9495 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9496 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9497 new_dentry->d_name.name,
9498 new_dentry->d_name.len);
9499 } else { /* dest is an inode */
9500 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9501 BTRFS_I(new_dentry->d_inode),
9502 new_dentry->d_name.name,
9503 new_dentry->d_name.len);
9505 ret = btrfs_update_inode(trans, dest, new_inode);
9508 btrfs_abort_transaction(trans, ret);
9512 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9513 new_dentry->d_name.name,
9514 new_dentry->d_name.len, 0, old_idx);
9516 btrfs_abort_transaction(trans, ret);
9520 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9521 old_dentry->d_name.name,
9522 old_dentry->d_name.len, 0, new_idx);
9524 btrfs_abort_transaction(trans, ret);
9528 if (old_inode->i_nlink == 1)
9529 BTRFS_I(old_inode)->dir_index = old_idx;
9530 if (new_inode->i_nlink == 1)
9531 BTRFS_I(new_inode)->dir_index = new_idx;
9533 if (root_log_pinned) {
9534 parent = new_dentry->d_parent;
9535 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9536 BTRFS_I(old_dir), parent,
9538 if (ret == BTRFS_NEED_LOG_SYNC)
9539 sync_log_root = true;
9540 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9541 commit_transaction = true;
9543 btrfs_end_log_trans(root);
9544 root_log_pinned = false;
9546 if (dest_log_pinned) {
9547 if (!commit_transaction) {
9548 parent = old_dentry->d_parent;
9549 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9550 BTRFS_I(new_dir), parent,
9552 if (ret == BTRFS_NEED_LOG_SYNC)
9553 sync_log_dest = true;
9554 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9555 commit_transaction = true;
9558 btrfs_end_log_trans(dest);
9559 dest_log_pinned = false;
9563 * If we have pinned a log and an error happened, we unpin tasks
9564 * trying to sync the log and force them to fallback to a transaction
9565 * commit if the log currently contains any of the inodes involved in
9566 * this rename operation (to ensure we do not persist a log with an
9567 * inconsistent state for any of these inodes or leading to any
9568 * inconsistencies when replayed). If the transaction was aborted, the
9569 * abortion reason is propagated to userspace when attempting to commit
9570 * the transaction. If the log does not contain any of these inodes, we
9571 * allow the tasks to sync it.
9573 if (ret && (root_log_pinned || dest_log_pinned)) {
9574 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9575 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9576 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9578 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9579 btrfs_set_log_full_commit(fs_info, trans);
9581 if (root_log_pinned) {
9582 btrfs_end_log_trans(root);
9583 root_log_pinned = false;
9585 if (dest_log_pinned) {
9586 btrfs_end_log_trans(dest);
9587 dest_log_pinned = false;
9590 if (!ret && sync_log_root && !commit_transaction) {
9591 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9594 commit_transaction = true;
9596 if (!ret && sync_log_dest && !commit_transaction) {
9597 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9600 commit_transaction = true;
9602 if (commit_transaction) {
9603 ret = btrfs_commit_transaction(trans);
9607 ret2 = btrfs_end_transaction(trans);
9608 ret = ret ? ret : ret2;
9611 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9612 up_read(&fs_info->subvol_sem);
9613 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9614 up_read(&fs_info->subvol_sem);
9619 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9620 struct btrfs_root *root,
9622 struct dentry *dentry)
9625 struct inode *inode;
9629 ret = btrfs_find_free_ino(root, &objectid);
9633 inode = btrfs_new_inode(trans, root, dir,
9634 dentry->d_name.name,
9636 btrfs_ino(BTRFS_I(dir)),
9638 S_IFCHR | WHITEOUT_MODE,
9641 if (IS_ERR(inode)) {
9642 ret = PTR_ERR(inode);
9646 inode->i_op = &btrfs_special_inode_operations;
9647 init_special_inode(inode, inode->i_mode,
9650 ret = btrfs_init_inode_security(trans, inode, dir,
9655 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9656 BTRFS_I(inode), 0, index);
9660 ret = btrfs_update_inode(trans, root, inode);
9662 unlock_new_inode(inode);
9664 inode_dec_link_count(inode);
9670 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9671 struct inode *new_dir, struct dentry *new_dentry,
9674 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9675 struct btrfs_trans_handle *trans;
9676 unsigned int trans_num_items;
9677 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9678 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9679 struct inode *new_inode = d_inode(new_dentry);
9680 struct inode *old_inode = d_inode(old_dentry);
9684 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9685 bool log_pinned = false;
9686 struct btrfs_log_ctx ctx;
9687 bool sync_log = false;
9688 bool commit_transaction = false;
9690 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9693 /* we only allow rename subvolume link between subvolumes */
9694 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9697 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9698 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9701 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9702 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9706 /* check for collisions, even if the name isn't there */
9707 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9708 new_dentry->d_name.name,
9709 new_dentry->d_name.len);
9712 if (ret == -EEXIST) {
9714 * eexist without a new_inode */
9715 if (WARN_ON(!new_inode)) {
9719 /* maybe -EOVERFLOW */
9726 * we're using rename to replace one file with another. Start IO on it
9727 * now so we don't add too much work to the end of the transaction
9729 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9730 filemap_flush(old_inode->i_mapping);
9732 /* close the racy window with snapshot create/destroy ioctl */
9733 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9734 down_read(&fs_info->subvol_sem);
9736 * We want to reserve the absolute worst case amount of items. So if
9737 * both inodes are subvols and we need to unlink them then that would
9738 * require 4 item modifications, but if they are both normal inodes it
9739 * would require 5 item modifications, so we'll assume they are normal
9740 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9741 * should cover the worst case number of items we'll modify.
9742 * If our rename has the whiteout flag, we need more 5 units for the
9743 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9744 * when selinux is enabled).
9746 trans_num_items = 11;
9747 if (flags & RENAME_WHITEOUT)
9748 trans_num_items += 5;
9749 trans = btrfs_start_transaction(root, trans_num_items);
9750 if (IS_ERR(trans)) {
9751 ret = PTR_ERR(trans);
9756 btrfs_record_root_in_trans(trans, dest);
9758 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9762 BTRFS_I(old_inode)->dir_index = 0ULL;
9763 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9764 /* force full log commit if subvolume involved. */
9765 btrfs_set_log_full_commit(fs_info, trans);
9767 btrfs_pin_log_trans(root);
9769 ret = btrfs_insert_inode_ref(trans, dest,
9770 new_dentry->d_name.name,
9771 new_dentry->d_name.len,
9773 btrfs_ino(BTRFS_I(new_dir)), index);
9778 inode_inc_iversion(old_dir);
9779 inode_inc_iversion(new_dir);
9780 inode_inc_iversion(old_inode);
9781 old_dir->i_ctime = old_dir->i_mtime =
9782 new_dir->i_ctime = new_dir->i_mtime =
9783 old_inode->i_ctime = current_time(old_dir);
9785 if (old_dentry->d_parent != new_dentry->d_parent)
9786 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9787 BTRFS_I(old_inode), 1);
9789 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9790 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9791 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9792 old_dentry->d_name.name,
9793 old_dentry->d_name.len);
9795 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9796 BTRFS_I(d_inode(old_dentry)),
9797 old_dentry->d_name.name,
9798 old_dentry->d_name.len);
9800 ret = btrfs_update_inode(trans, root, old_inode);
9803 btrfs_abort_transaction(trans, ret);
9808 inode_inc_iversion(new_inode);
9809 new_inode->i_ctime = current_time(new_inode);
9810 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9811 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9812 root_objectid = BTRFS_I(new_inode)->location.objectid;
9813 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9814 new_dentry->d_name.name,
9815 new_dentry->d_name.len);
9816 BUG_ON(new_inode->i_nlink == 0);
9818 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9819 BTRFS_I(d_inode(new_dentry)),
9820 new_dentry->d_name.name,
9821 new_dentry->d_name.len);
9823 if (!ret && new_inode->i_nlink == 0)
9824 ret = btrfs_orphan_add(trans,
9825 BTRFS_I(d_inode(new_dentry)));
9827 btrfs_abort_transaction(trans, ret);
9832 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9833 new_dentry->d_name.name,
9834 new_dentry->d_name.len, 0, index);
9836 btrfs_abort_transaction(trans, ret);
9840 if (old_inode->i_nlink == 1)
9841 BTRFS_I(old_inode)->dir_index = index;
9844 struct dentry *parent = new_dentry->d_parent;
9846 btrfs_init_log_ctx(&ctx, old_inode);
9847 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9848 BTRFS_I(old_dir), parent,
9850 if (ret == BTRFS_NEED_LOG_SYNC)
9852 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9853 commit_transaction = true;
9855 btrfs_end_log_trans(root);
9859 if (flags & RENAME_WHITEOUT) {
9860 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9864 btrfs_abort_transaction(trans, ret);
9870 * If we have pinned the log and an error happened, we unpin tasks
9871 * trying to sync the log and force them to fallback to a transaction
9872 * commit if the log currently contains any of the inodes involved in
9873 * this rename operation (to ensure we do not persist a log with an
9874 * inconsistent state for any of these inodes or leading to any
9875 * inconsistencies when replayed). If the transaction was aborted, the
9876 * abortion reason is propagated to userspace when attempting to commit
9877 * the transaction. If the log does not contain any of these inodes, we
9878 * allow the tasks to sync it.
9880 if (ret && log_pinned) {
9881 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9882 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9883 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9885 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9886 btrfs_set_log_full_commit(fs_info, trans);
9888 btrfs_end_log_trans(root);
9891 if (!ret && sync_log) {
9892 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9894 commit_transaction = true;
9896 if (commit_transaction) {
9897 ret = btrfs_commit_transaction(trans);
9901 ret2 = btrfs_end_transaction(trans);
9902 ret = ret ? ret : ret2;
9905 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9906 up_read(&fs_info->subvol_sem);
9911 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9912 struct inode *new_dir, struct dentry *new_dentry,
9915 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9918 if (flags & RENAME_EXCHANGE)
9919 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9922 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9925 struct btrfs_delalloc_work {
9926 struct inode *inode;
9927 struct completion completion;
9928 struct list_head list;
9929 struct btrfs_work work;
9932 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9934 struct btrfs_delalloc_work *delalloc_work;
9935 struct inode *inode;
9937 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9939 inode = delalloc_work->inode;
9940 filemap_flush(inode->i_mapping);
9941 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9942 &BTRFS_I(inode)->runtime_flags))
9943 filemap_flush(inode->i_mapping);
9946 complete(&delalloc_work->completion);
9949 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9951 struct btrfs_delalloc_work *work;
9953 work = kmalloc(sizeof(*work), GFP_NOFS);
9957 init_completion(&work->completion);
9958 INIT_LIST_HEAD(&work->list);
9959 work->inode = inode;
9960 WARN_ON_ONCE(!inode);
9961 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9962 btrfs_run_delalloc_work, NULL, NULL);
9968 * some fairly slow code that needs optimization. This walks the list
9969 * of all the inodes with pending delalloc and forces them to disk.
9971 static int start_delalloc_inodes(struct btrfs_root *root, int nr)
9973 struct btrfs_inode *binode;
9974 struct inode *inode;
9975 struct btrfs_delalloc_work *work, *next;
9976 struct list_head works;
9977 struct list_head splice;
9980 INIT_LIST_HEAD(&works);
9981 INIT_LIST_HEAD(&splice);
9983 mutex_lock(&root->delalloc_mutex);
9984 spin_lock(&root->delalloc_lock);
9985 list_splice_init(&root->delalloc_inodes, &splice);
9986 while (!list_empty(&splice)) {
9987 binode = list_entry(splice.next, struct btrfs_inode,
9990 list_move_tail(&binode->delalloc_inodes,
9991 &root->delalloc_inodes);
9992 inode = igrab(&binode->vfs_inode);
9994 cond_resched_lock(&root->delalloc_lock);
9997 spin_unlock(&root->delalloc_lock);
9999 work = btrfs_alloc_delalloc_work(inode);
10005 list_add_tail(&work->list, &works);
10006 btrfs_queue_work(root->fs_info->flush_workers,
10009 if (nr != -1 && ret >= nr)
10012 spin_lock(&root->delalloc_lock);
10014 spin_unlock(&root->delalloc_lock);
10017 list_for_each_entry_safe(work, next, &works, list) {
10018 list_del_init(&work->list);
10019 wait_for_completion(&work->completion);
10023 if (!list_empty(&splice)) {
10024 spin_lock(&root->delalloc_lock);
10025 list_splice_tail(&splice, &root->delalloc_inodes);
10026 spin_unlock(&root->delalloc_lock);
10028 mutex_unlock(&root->delalloc_mutex);
10032 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10034 struct btrfs_fs_info *fs_info = root->fs_info;
10037 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10040 ret = start_delalloc_inodes(root, -1);
10046 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10048 struct btrfs_root *root;
10049 struct list_head splice;
10052 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10055 INIT_LIST_HEAD(&splice);
10057 mutex_lock(&fs_info->delalloc_root_mutex);
10058 spin_lock(&fs_info->delalloc_root_lock);
10059 list_splice_init(&fs_info->delalloc_roots, &splice);
10060 while (!list_empty(&splice) && nr) {
10061 root = list_first_entry(&splice, struct btrfs_root,
10063 root = btrfs_grab_fs_root(root);
10065 list_move_tail(&root->delalloc_root,
10066 &fs_info->delalloc_roots);
10067 spin_unlock(&fs_info->delalloc_root_lock);
10069 ret = start_delalloc_inodes(root, nr);
10070 btrfs_put_fs_root(root);
10078 spin_lock(&fs_info->delalloc_root_lock);
10080 spin_unlock(&fs_info->delalloc_root_lock);
10084 if (!list_empty(&splice)) {
10085 spin_lock(&fs_info->delalloc_root_lock);
10086 list_splice_tail(&splice, &fs_info->delalloc_roots);
10087 spin_unlock(&fs_info->delalloc_root_lock);
10089 mutex_unlock(&fs_info->delalloc_root_mutex);
10093 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10094 const char *symname)
10096 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10097 struct btrfs_trans_handle *trans;
10098 struct btrfs_root *root = BTRFS_I(dir)->root;
10099 struct btrfs_path *path;
10100 struct btrfs_key key;
10101 struct inode *inode = NULL;
10108 struct btrfs_file_extent_item *ei;
10109 struct extent_buffer *leaf;
10111 name_len = strlen(symname);
10112 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10113 return -ENAMETOOLONG;
10116 * 2 items for inode item and ref
10117 * 2 items for dir items
10118 * 1 item for updating parent inode item
10119 * 1 item for the inline extent item
10120 * 1 item for xattr if selinux is on
10122 trans = btrfs_start_transaction(root, 7);
10124 return PTR_ERR(trans);
10126 err = btrfs_find_free_ino(root, &objectid);
10130 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10131 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10132 objectid, S_IFLNK|S_IRWXUGO, &index);
10133 if (IS_ERR(inode)) {
10134 err = PTR_ERR(inode);
10140 * If the active LSM wants to access the inode during
10141 * d_instantiate it needs these. Smack checks to see
10142 * if the filesystem supports xattrs by looking at the
10145 inode->i_fop = &btrfs_file_operations;
10146 inode->i_op = &btrfs_file_inode_operations;
10147 inode->i_mapping->a_ops = &btrfs_aops;
10148 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10150 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10154 path = btrfs_alloc_path();
10159 key.objectid = btrfs_ino(BTRFS_I(inode));
10161 key.type = BTRFS_EXTENT_DATA_KEY;
10162 datasize = btrfs_file_extent_calc_inline_size(name_len);
10163 err = btrfs_insert_empty_item(trans, root, path, &key,
10166 btrfs_free_path(path);
10169 leaf = path->nodes[0];
10170 ei = btrfs_item_ptr(leaf, path->slots[0],
10171 struct btrfs_file_extent_item);
10172 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10173 btrfs_set_file_extent_type(leaf, ei,
10174 BTRFS_FILE_EXTENT_INLINE);
10175 btrfs_set_file_extent_encryption(leaf, ei, 0);
10176 btrfs_set_file_extent_compression(leaf, ei, 0);
10177 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10178 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10180 ptr = btrfs_file_extent_inline_start(ei);
10181 write_extent_buffer(leaf, symname, ptr, name_len);
10182 btrfs_mark_buffer_dirty(leaf);
10183 btrfs_free_path(path);
10185 inode->i_op = &btrfs_symlink_inode_operations;
10186 inode_nohighmem(inode);
10187 inode->i_mapping->a_ops = &btrfs_aops;
10188 inode_set_bytes(inode, name_len);
10189 btrfs_i_size_write(BTRFS_I(inode), name_len);
10190 err = btrfs_update_inode(trans, root, inode);
10192 * Last step, add directory indexes for our symlink inode. This is the
10193 * last step to avoid extra cleanup of these indexes if an error happens
10197 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10198 BTRFS_I(inode), 0, index);
10202 d_instantiate_new(dentry, inode);
10205 btrfs_end_transaction(trans);
10206 if (err && inode) {
10207 inode_dec_link_count(inode);
10208 discard_new_inode(inode);
10210 btrfs_btree_balance_dirty(fs_info);
10214 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10215 u64 start, u64 num_bytes, u64 min_size,
10216 loff_t actual_len, u64 *alloc_hint,
10217 struct btrfs_trans_handle *trans)
10219 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10220 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10221 struct extent_map *em;
10222 struct btrfs_root *root = BTRFS_I(inode)->root;
10223 struct btrfs_key ins;
10224 u64 cur_offset = start;
10227 u64 last_alloc = (u64)-1;
10229 bool own_trans = true;
10230 u64 end = start + num_bytes - 1;
10234 while (num_bytes > 0) {
10236 trans = btrfs_start_transaction(root, 3);
10237 if (IS_ERR(trans)) {
10238 ret = PTR_ERR(trans);
10243 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10244 cur_bytes = max(cur_bytes, min_size);
10246 * If we are severely fragmented we could end up with really
10247 * small allocations, so if the allocator is returning small
10248 * chunks lets make its job easier by only searching for those
10251 cur_bytes = min(cur_bytes, last_alloc);
10252 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10253 min_size, 0, *alloc_hint, &ins, 1, 0);
10256 btrfs_end_transaction(trans);
10259 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10261 last_alloc = ins.offset;
10262 ret = insert_reserved_file_extent(trans, inode,
10263 cur_offset, ins.objectid,
10264 ins.offset, ins.offset,
10265 ins.offset, 0, 0, 0,
10266 BTRFS_FILE_EXTENT_PREALLOC);
10268 btrfs_free_reserved_extent(fs_info, ins.objectid,
10270 btrfs_abort_transaction(trans, ret);
10272 btrfs_end_transaction(trans);
10276 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10277 cur_offset + ins.offset -1, 0);
10279 em = alloc_extent_map();
10281 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10282 &BTRFS_I(inode)->runtime_flags);
10286 em->start = cur_offset;
10287 em->orig_start = cur_offset;
10288 em->len = ins.offset;
10289 em->block_start = ins.objectid;
10290 em->block_len = ins.offset;
10291 em->orig_block_len = ins.offset;
10292 em->ram_bytes = ins.offset;
10293 em->bdev = fs_info->fs_devices->latest_bdev;
10294 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10295 em->generation = trans->transid;
10298 write_lock(&em_tree->lock);
10299 ret = add_extent_mapping(em_tree, em, 1);
10300 write_unlock(&em_tree->lock);
10301 if (ret != -EEXIST)
10303 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10304 cur_offset + ins.offset - 1,
10307 free_extent_map(em);
10309 num_bytes -= ins.offset;
10310 cur_offset += ins.offset;
10311 *alloc_hint = ins.objectid + ins.offset;
10313 inode_inc_iversion(inode);
10314 inode->i_ctime = current_time(inode);
10315 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10316 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10317 (actual_len > inode->i_size) &&
10318 (cur_offset > inode->i_size)) {
10319 if (cur_offset > actual_len)
10320 i_size = actual_len;
10322 i_size = cur_offset;
10323 i_size_write(inode, i_size);
10324 btrfs_ordered_update_i_size(inode, i_size, NULL);
10327 ret = btrfs_update_inode(trans, root, inode);
10330 btrfs_abort_transaction(trans, ret);
10332 btrfs_end_transaction(trans);
10337 btrfs_end_transaction(trans);
10339 if (cur_offset < end)
10340 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10341 end - cur_offset + 1);
10345 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10346 u64 start, u64 num_bytes, u64 min_size,
10347 loff_t actual_len, u64 *alloc_hint)
10349 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10350 min_size, actual_len, alloc_hint,
10354 int btrfs_prealloc_file_range_trans(struct inode *inode,
10355 struct btrfs_trans_handle *trans, int mode,
10356 u64 start, u64 num_bytes, u64 min_size,
10357 loff_t actual_len, u64 *alloc_hint)
10359 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10360 min_size, actual_len, alloc_hint, trans);
10363 static int btrfs_set_page_dirty(struct page *page)
10365 return __set_page_dirty_nobuffers(page);
10368 static int btrfs_permission(struct inode *inode, int mask)
10370 struct btrfs_root *root = BTRFS_I(inode)->root;
10371 umode_t mode = inode->i_mode;
10373 if (mask & MAY_WRITE &&
10374 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10375 if (btrfs_root_readonly(root))
10377 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10380 return generic_permission(inode, mask);
10383 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10385 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10386 struct btrfs_trans_handle *trans;
10387 struct btrfs_root *root = BTRFS_I(dir)->root;
10388 struct inode *inode = NULL;
10394 * 5 units required for adding orphan entry
10396 trans = btrfs_start_transaction(root, 5);
10398 return PTR_ERR(trans);
10400 ret = btrfs_find_free_ino(root, &objectid);
10404 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10405 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10406 if (IS_ERR(inode)) {
10407 ret = PTR_ERR(inode);
10412 inode->i_fop = &btrfs_file_operations;
10413 inode->i_op = &btrfs_file_inode_operations;
10415 inode->i_mapping->a_ops = &btrfs_aops;
10416 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10418 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10422 ret = btrfs_update_inode(trans, root, inode);
10425 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10430 * We set number of links to 0 in btrfs_new_inode(), and here we set
10431 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10434 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10436 set_nlink(inode, 1);
10437 d_tmpfile(dentry, inode);
10438 unlock_new_inode(inode);
10439 mark_inode_dirty(inode);
10441 btrfs_end_transaction(trans);
10443 discard_new_inode(inode);
10444 btrfs_btree_balance_dirty(fs_info);
10448 __attribute__((const))
10449 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10454 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10455 u64 start, u64 end)
10457 struct inode *inode = private_data;
10460 isize = i_size_read(inode);
10461 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10462 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10463 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10464 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10468 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10470 struct inode *inode = tree->private_data;
10471 unsigned long index = start >> PAGE_SHIFT;
10472 unsigned long end_index = end >> PAGE_SHIFT;
10475 while (index <= end_index) {
10476 page = find_get_page(inode->i_mapping, index);
10477 ASSERT(page); /* Pages should be in the extent_io_tree */
10478 set_page_writeback(page);
10484 static const struct inode_operations btrfs_dir_inode_operations = {
10485 .getattr = btrfs_getattr,
10486 .lookup = btrfs_lookup,
10487 .create = btrfs_create,
10488 .unlink = btrfs_unlink,
10489 .link = btrfs_link,
10490 .mkdir = btrfs_mkdir,
10491 .rmdir = btrfs_rmdir,
10492 .rename = btrfs_rename2,
10493 .symlink = btrfs_symlink,
10494 .setattr = btrfs_setattr,
10495 .mknod = btrfs_mknod,
10496 .listxattr = btrfs_listxattr,
10497 .permission = btrfs_permission,
10498 .get_acl = btrfs_get_acl,
10499 .set_acl = btrfs_set_acl,
10500 .update_time = btrfs_update_time,
10501 .tmpfile = btrfs_tmpfile,
10503 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10504 .lookup = btrfs_lookup,
10505 .permission = btrfs_permission,
10506 .update_time = btrfs_update_time,
10509 static const struct file_operations btrfs_dir_file_operations = {
10510 .llseek = generic_file_llseek,
10511 .read = generic_read_dir,
10512 .iterate_shared = btrfs_real_readdir,
10513 .open = btrfs_opendir,
10514 .unlocked_ioctl = btrfs_ioctl,
10515 #ifdef CONFIG_COMPAT
10516 .compat_ioctl = btrfs_compat_ioctl,
10518 .release = btrfs_release_file,
10519 .fsync = btrfs_sync_file,
10522 static const struct extent_io_ops btrfs_extent_io_ops = {
10523 /* mandatory callbacks */
10524 .submit_bio_hook = btrfs_submit_bio_hook,
10525 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10526 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10528 /* optional callbacks */
10529 .set_bit_hook = btrfs_set_bit_hook,
10530 .clear_bit_hook = btrfs_clear_bit_hook,
10531 .merge_extent_hook = btrfs_merge_extent_hook,
10532 .split_extent_hook = btrfs_split_extent_hook,
10533 .check_extent_io_range = btrfs_check_extent_io_range,
10537 * btrfs doesn't support the bmap operation because swapfiles
10538 * use bmap to make a mapping of extents in the file. They assume
10539 * these extents won't change over the life of the file and they
10540 * use the bmap result to do IO directly to the drive.
10542 * the btrfs bmap call would return logical addresses that aren't
10543 * suitable for IO and they also will change frequently as COW
10544 * operations happen. So, swapfile + btrfs == corruption.
10546 * For now we're avoiding this by dropping bmap.
10548 static const struct address_space_operations btrfs_aops = {
10549 .readpage = btrfs_readpage,
10550 .writepage = btrfs_writepage,
10551 .writepages = btrfs_writepages,
10552 .readpages = btrfs_readpages,
10553 .direct_IO = btrfs_direct_IO,
10554 .invalidatepage = btrfs_invalidatepage,
10555 .releasepage = btrfs_releasepage,
10556 .set_page_dirty = btrfs_set_page_dirty,
10557 .error_remove_page = generic_error_remove_page,
10560 static const struct inode_operations btrfs_file_inode_operations = {
10561 .getattr = btrfs_getattr,
10562 .setattr = btrfs_setattr,
10563 .listxattr = btrfs_listxattr,
10564 .permission = btrfs_permission,
10565 .fiemap = btrfs_fiemap,
10566 .get_acl = btrfs_get_acl,
10567 .set_acl = btrfs_set_acl,
10568 .update_time = btrfs_update_time,
10570 static const struct inode_operations btrfs_special_inode_operations = {
10571 .getattr = btrfs_getattr,
10572 .setattr = btrfs_setattr,
10573 .permission = btrfs_permission,
10574 .listxattr = btrfs_listxattr,
10575 .get_acl = btrfs_get_acl,
10576 .set_acl = btrfs_set_acl,
10577 .update_time = btrfs_update_time,
10579 static const struct inode_operations btrfs_symlink_inode_operations = {
10580 .get_link = page_get_link,
10581 .getattr = btrfs_getattr,
10582 .setattr = btrfs_setattr,
10583 .permission = btrfs_permission,
10584 .listxattr = btrfs_listxattr,
10585 .update_time = btrfs_update_time,
10588 const struct dentry_operations btrfs_dentry_operations = {
10589 .d_delete = btrfs_dentry_delete,