2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
45 #include <linux/magic.h>
46 #include <linux/iversion.h>
49 #include "transaction.h"
50 #include "btrfs_inode.h"
51 #include "print-tree.h"
52 #include "ordered-data.h"
56 #include "compression.h"
58 #include "free-space-cache.h"
59 #include "inode-map.h"
65 struct btrfs_iget_args {
66 struct btrfs_key *location;
67 struct btrfs_root *root;
70 struct btrfs_dio_data {
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, u64 delalloc_end,
109 int *page_started, unsigned long *nr_written,
110 int unlock, struct btrfs_dedupe_hash *hash);
111 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
112 u64 orig_start, u64 block_start,
113 u64 block_len, u64 orig_block_len,
114 u64 ram_bytes, int compress_type,
117 static void __endio_write_update_ordered(struct inode *inode,
118 const u64 offset, const u64 bytes,
119 const bool uptodate);
122 * Cleanup all submitted ordered extents in specified range to handle errors
123 * from the fill_dellaloc() callback.
125 * NOTE: caller must ensure that when an error happens, it can not call
126 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
127 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
128 * to be released, which we want to happen only when finishing the ordered
129 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
130 * fill_delalloc() callback already does proper cleanup for the first page of
131 * the range, that is, it invokes the callback writepage_end_io_hook() for the
132 * range of the first page.
134 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
138 unsigned long index = offset >> PAGE_SHIFT;
139 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
142 while (index <= end_index) {
143 page = find_get_page(inode->i_mapping, index);
147 ClearPagePrivate2(page);
150 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
151 bytes - PAGE_SIZE, false);
154 static int btrfs_dirty_inode(struct inode *inode);
156 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
157 void btrfs_test_inode_set_ops(struct inode *inode)
159 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
163 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
164 struct inode *inode, struct inode *dir,
165 const struct qstr *qstr)
169 err = btrfs_init_acl(trans, inode, dir);
171 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
176 * this does all the hard work for inserting an inline extent into
177 * the btree. The caller should have done a btrfs_drop_extents so that
178 * no overlapping inline items exist in the btree
180 static int insert_inline_extent(struct btrfs_trans_handle *trans,
181 struct btrfs_path *path, int extent_inserted,
182 struct btrfs_root *root, struct inode *inode,
183 u64 start, size_t size, size_t compressed_size,
185 struct page **compressed_pages)
187 struct extent_buffer *leaf;
188 struct page *page = NULL;
191 struct btrfs_file_extent_item *ei;
193 size_t cur_size = size;
194 unsigned long offset;
196 if (compressed_size && compressed_pages)
197 cur_size = compressed_size;
199 inode_add_bytes(inode, size);
201 if (!extent_inserted) {
202 struct btrfs_key key;
205 key.objectid = btrfs_ino(BTRFS_I(inode));
207 key.type = BTRFS_EXTENT_DATA_KEY;
209 datasize = btrfs_file_extent_calc_inline_size(cur_size);
210 path->leave_spinning = 1;
211 ret = btrfs_insert_empty_item(trans, root, path, &key,
216 leaf = path->nodes[0];
217 ei = btrfs_item_ptr(leaf, path->slots[0],
218 struct btrfs_file_extent_item);
219 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
220 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
221 btrfs_set_file_extent_encryption(leaf, ei, 0);
222 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
223 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
224 ptr = btrfs_file_extent_inline_start(ei);
226 if (compress_type != BTRFS_COMPRESS_NONE) {
229 while (compressed_size > 0) {
230 cpage = compressed_pages[i];
231 cur_size = min_t(unsigned long, compressed_size,
234 kaddr = kmap_atomic(cpage);
235 write_extent_buffer(leaf, kaddr, ptr, cur_size);
236 kunmap_atomic(kaddr);
240 compressed_size -= cur_size;
242 btrfs_set_file_extent_compression(leaf, ei,
245 page = find_get_page(inode->i_mapping,
246 start >> PAGE_SHIFT);
247 btrfs_set_file_extent_compression(leaf, ei, 0);
248 kaddr = kmap_atomic(page);
249 offset = start & (PAGE_SIZE - 1);
250 write_extent_buffer(leaf, kaddr + offset, ptr, size);
251 kunmap_atomic(kaddr);
254 btrfs_mark_buffer_dirty(leaf);
255 btrfs_release_path(path);
258 * we're an inline extent, so nobody can
259 * extend the file past i_size without locking
260 * a page we already have locked.
262 * We must do any isize and inode updates
263 * before we unlock the pages. Otherwise we
264 * could end up racing with unlink.
266 BTRFS_I(inode)->disk_i_size = inode->i_size;
267 ret = btrfs_update_inode(trans, root, inode);
275 * conditionally insert an inline extent into the file. This
276 * does the checks required to make sure the data is small enough
277 * to fit as an inline extent.
279 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
280 u64 end, size_t compressed_size,
282 struct page **compressed_pages)
284 struct btrfs_root *root = BTRFS_I(inode)->root;
285 struct btrfs_fs_info *fs_info = root->fs_info;
286 struct btrfs_trans_handle *trans;
287 u64 isize = i_size_read(inode);
288 u64 actual_end = min(end + 1, isize);
289 u64 inline_len = actual_end - start;
290 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
291 u64 data_len = inline_len;
293 struct btrfs_path *path;
294 int extent_inserted = 0;
295 u32 extent_item_size;
298 data_len = compressed_size;
301 actual_end > fs_info->sectorsize ||
302 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
304 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
306 data_len > fs_info->max_inline) {
310 path = btrfs_alloc_path();
314 trans = btrfs_join_transaction(root);
316 btrfs_free_path(path);
317 return PTR_ERR(trans);
319 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
321 if (compressed_size && compressed_pages)
322 extent_item_size = btrfs_file_extent_calc_inline_size(
325 extent_item_size = btrfs_file_extent_calc_inline_size(
328 ret = __btrfs_drop_extents(trans, root, inode, path,
329 start, aligned_end, NULL,
330 1, 1, extent_item_size, &extent_inserted);
332 btrfs_abort_transaction(trans, ret);
336 if (isize > actual_end)
337 inline_len = min_t(u64, isize, actual_end);
338 ret = insert_inline_extent(trans, path, extent_inserted,
340 inline_len, compressed_size,
341 compress_type, compressed_pages);
342 if (ret && ret != -ENOSPC) {
343 btrfs_abort_transaction(trans, ret);
345 } else if (ret == -ENOSPC) {
350 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
351 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
354 * Don't forget to free the reserved space, as for inlined extent
355 * it won't count as data extent, free them directly here.
356 * And at reserve time, it's always aligned to page size, so
357 * just free one page here.
359 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
360 btrfs_free_path(path);
361 btrfs_end_transaction(trans);
365 struct async_extent {
370 unsigned long nr_pages;
372 struct list_head list;
377 struct btrfs_root *root;
378 struct page *locked_page;
381 unsigned int write_flags;
382 struct list_head extents;
383 struct btrfs_work work;
386 static noinline int add_async_extent(struct async_cow *cow,
387 u64 start, u64 ram_size,
390 unsigned long nr_pages,
393 struct async_extent *async_extent;
395 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
396 BUG_ON(!async_extent); /* -ENOMEM */
397 async_extent->start = start;
398 async_extent->ram_size = ram_size;
399 async_extent->compressed_size = compressed_size;
400 async_extent->pages = pages;
401 async_extent->nr_pages = nr_pages;
402 async_extent->compress_type = compress_type;
403 list_add_tail(&async_extent->list, &cow->extents);
407 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
409 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
412 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
415 if (BTRFS_I(inode)->defrag_compress)
417 /* bad compression ratios */
418 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
420 if (btrfs_test_opt(fs_info, COMPRESS) ||
421 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
422 BTRFS_I(inode)->prop_compress)
423 return btrfs_compress_heuristic(inode, start, end);
427 static inline void inode_should_defrag(struct btrfs_inode *inode,
428 u64 start, u64 end, u64 num_bytes, u64 small_write)
430 /* If this is a small write inside eof, kick off a defrag */
431 if (num_bytes < small_write &&
432 (start > 0 || end + 1 < inode->disk_i_size))
433 btrfs_add_inode_defrag(NULL, inode);
437 * we create compressed extents in two phases. The first
438 * phase compresses a range of pages that have already been
439 * locked (both pages and state bits are locked).
441 * This is done inside an ordered work queue, and the compression
442 * is spread across many cpus. The actual IO submission is step
443 * two, and the ordered work queue takes care of making sure that
444 * happens in the same order things were put onto the queue by
445 * writepages and friends.
447 * If this code finds it can't get good compression, it puts an
448 * entry onto the work queue to write the uncompressed bytes. This
449 * makes sure that both compressed inodes and uncompressed inodes
450 * are written in the same order that the flusher thread sent them
453 static noinline void compress_file_range(struct inode *inode,
454 struct page *locked_page,
456 struct async_cow *async_cow,
459 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
460 u64 blocksize = fs_info->sectorsize;
462 u64 isize = i_size_read(inode);
464 struct page **pages = NULL;
465 unsigned long nr_pages;
466 unsigned long total_compressed = 0;
467 unsigned long total_in = 0;
470 int compress_type = fs_info->compress_type;
473 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
476 actual_end = min_t(u64, isize, end + 1);
479 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
480 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
481 nr_pages = min_t(unsigned long, nr_pages,
482 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
485 * we don't want to send crud past the end of i_size through
486 * compression, that's just a waste of CPU time. So, if the
487 * end of the file is before the start of our current
488 * requested range of bytes, we bail out to the uncompressed
489 * cleanup code that can deal with all of this.
491 * It isn't really the fastest way to fix things, but this is a
492 * very uncommon corner.
494 if (actual_end <= start)
495 goto cleanup_and_bail_uncompressed;
497 total_compressed = actual_end - start;
500 * skip compression for a small file range(<=blocksize) that
501 * isn't an inline extent, since it doesn't save disk space at all.
503 if (total_compressed <= blocksize &&
504 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
505 goto cleanup_and_bail_uncompressed;
507 total_compressed = min_t(unsigned long, total_compressed,
508 BTRFS_MAX_UNCOMPRESSED);
513 * we do compression for mount -o compress and when the
514 * inode has not been flagged as nocompress. This flag can
515 * change at any time if we discover bad compression ratios.
517 if (inode_need_compress(inode, start, end)) {
519 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
521 /* just bail out to the uncompressed code */
525 if (BTRFS_I(inode)->defrag_compress)
526 compress_type = BTRFS_I(inode)->defrag_compress;
527 else if (BTRFS_I(inode)->prop_compress)
528 compress_type = BTRFS_I(inode)->prop_compress;
531 * we need to call clear_page_dirty_for_io on each
532 * page in the range. Otherwise applications with the file
533 * mmap'd can wander in and change the page contents while
534 * we are compressing them.
536 * If the compression fails for any reason, we set the pages
537 * dirty again later on.
539 * Note that the remaining part is redirtied, the start pointer
540 * has moved, the end is the original one.
543 extent_range_clear_dirty_for_io(inode, start, end);
547 /* Compression level is applied here and only here */
548 ret = btrfs_compress_pages(
549 compress_type | (fs_info->compress_level << 4),
550 inode->i_mapping, start,
557 unsigned long offset = total_compressed &
559 struct page *page = pages[nr_pages - 1];
562 /* zero the tail end of the last page, we might be
563 * sending it down to disk
566 kaddr = kmap_atomic(page);
567 memset(kaddr + offset, 0,
569 kunmap_atomic(kaddr);
576 /* lets try to make an inline extent */
577 if (ret || total_in < actual_end) {
578 /* we didn't compress the entire range, try
579 * to make an uncompressed inline extent.
581 ret = cow_file_range_inline(inode, start, end, 0,
582 BTRFS_COMPRESS_NONE, NULL);
584 /* try making a compressed inline extent */
585 ret = cow_file_range_inline(inode, start, end,
587 compress_type, pages);
590 unsigned long clear_flags = EXTENT_DELALLOC |
591 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
592 EXTENT_DO_ACCOUNTING;
593 unsigned long page_error_op;
595 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
598 * inline extent creation worked or returned error,
599 * we don't need to create any more async work items.
600 * Unlock and free up our temp pages.
602 * We use DO_ACCOUNTING here because we need the
603 * delalloc_release_metadata to be done _after_ we drop
604 * our outstanding extent for clearing delalloc for this
607 extent_clear_unlock_delalloc(inode, start, end, end,
620 * we aren't doing an inline extent round the compressed size
621 * up to a block size boundary so the allocator does sane
624 total_compressed = ALIGN(total_compressed, blocksize);
627 * one last check to make sure the compression is really a
628 * win, compare the page count read with the blocks on disk,
629 * compression must free at least one sector size
631 total_in = ALIGN(total_in, PAGE_SIZE);
632 if (total_compressed + blocksize <= total_in) {
636 * The async work queues will take care of doing actual
637 * allocation on disk for these compressed pages, and
638 * will submit them to the elevator.
640 add_async_extent(async_cow, start, total_in,
641 total_compressed, pages, nr_pages,
644 if (start + total_in < end) {
655 * the compression code ran but failed to make things smaller,
656 * free any pages it allocated and our page pointer array
658 for (i = 0; i < nr_pages; i++) {
659 WARN_ON(pages[i]->mapping);
664 total_compressed = 0;
667 /* flag the file so we don't compress in the future */
668 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
669 !(BTRFS_I(inode)->prop_compress)) {
670 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
673 cleanup_and_bail_uncompressed:
675 * No compression, but we still need to write the pages in the file
676 * we've been given so far. redirty the locked page if it corresponds
677 * to our extent and set things up for the async work queue to run
678 * cow_file_range to do the normal delalloc dance.
680 if (page_offset(locked_page) >= start &&
681 page_offset(locked_page) <= end)
682 __set_page_dirty_nobuffers(locked_page);
683 /* unlocked later on in the async handlers */
686 extent_range_redirty_for_io(inode, start, end);
687 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
688 BTRFS_COMPRESS_NONE);
694 for (i = 0; i < nr_pages; i++) {
695 WARN_ON(pages[i]->mapping);
701 static void free_async_extent_pages(struct async_extent *async_extent)
705 if (!async_extent->pages)
708 for (i = 0; i < async_extent->nr_pages; i++) {
709 WARN_ON(async_extent->pages[i]->mapping);
710 put_page(async_extent->pages[i]);
712 kfree(async_extent->pages);
713 async_extent->nr_pages = 0;
714 async_extent->pages = NULL;
718 * phase two of compressed writeback. This is the ordered portion
719 * of the code, which only gets called in the order the work was
720 * queued. We walk all the async extents created by compress_file_range
721 * and send them down to the disk.
723 static noinline void submit_compressed_extents(struct inode *inode,
724 struct async_cow *async_cow)
726 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
727 struct async_extent *async_extent;
729 struct btrfs_key ins;
730 struct extent_map *em;
731 struct btrfs_root *root = BTRFS_I(inode)->root;
732 struct extent_io_tree *io_tree;
736 while (!list_empty(&async_cow->extents)) {
737 async_extent = list_entry(async_cow->extents.next,
738 struct async_extent, list);
739 list_del(&async_extent->list);
741 io_tree = &BTRFS_I(inode)->io_tree;
744 /* did the compression code fall back to uncompressed IO? */
745 if (!async_extent->pages) {
746 int page_started = 0;
747 unsigned long nr_written = 0;
749 lock_extent(io_tree, async_extent->start,
750 async_extent->start +
751 async_extent->ram_size - 1);
753 /* allocate blocks */
754 ret = cow_file_range(inode, async_cow->locked_page,
756 async_extent->start +
757 async_extent->ram_size - 1,
758 async_extent->start +
759 async_extent->ram_size - 1,
760 &page_started, &nr_written, 0,
766 * if page_started, cow_file_range inserted an
767 * inline extent and took care of all the unlocking
768 * and IO for us. Otherwise, we need to submit
769 * all those pages down to the drive.
771 if (!page_started && !ret)
772 extent_write_locked_range(inode,
774 async_extent->start +
775 async_extent->ram_size - 1,
778 unlock_page(async_cow->locked_page);
784 lock_extent(io_tree, async_extent->start,
785 async_extent->start + async_extent->ram_size - 1);
787 ret = btrfs_reserve_extent(root, async_extent->ram_size,
788 async_extent->compressed_size,
789 async_extent->compressed_size,
790 0, alloc_hint, &ins, 1, 1);
792 free_async_extent_pages(async_extent);
794 if (ret == -ENOSPC) {
795 unlock_extent(io_tree, async_extent->start,
796 async_extent->start +
797 async_extent->ram_size - 1);
800 * we need to redirty the pages if we decide to
801 * fallback to uncompressed IO, otherwise we
802 * will not submit these pages down to lower
805 extent_range_redirty_for_io(inode,
807 async_extent->start +
808 async_extent->ram_size - 1);
815 * here we're doing allocation and writeback of the
818 em = create_io_em(inode, async_extent->start,
819 async_extent->ram_size, /* len */
820 async_extent->start, /* orig_start */
821 ins.objectid, /* block_start */
822 ins.offset, /* block_len */
823 ins.offset, /* orig_block_len */
824 async_extent->ram_size, /* ram_bytes */
825 async_extent->compress_type,
826 BTRFS_ORDERED_COMPRESSED);
828 /* ret value is not necessary due to void function */
829 goto out_free_reserve;
832 ret = btrfs_add_ordered_extent_compress(inode,
835 async_extent->ram_size,
837 BTRFS_ORDERED_COMPRESSED,
838 async_extent->compress_type);
840 btrfs_drop_extent_cache(BTRFS_I(inode),
842 async_extent->start +
843 async_extent->ram_size - 1, 0);
844 goto out_free_reserve;
846 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
849 * clear dirty, set writeback and unlock the pages.
851 extent_clear_unlock_delalloc(inode, async_extent->start,
852 async_extent->start +
853 async_extent->ram_size - 1,
854 async_extent->start +
855 async_extent->ram_size - 1,
856 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
857 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
859 if (btrfs_submit_compressed_write(inode,
861 async_extent->ram_size,
863 ins.offset, async_extent->pages,
864 async_extent->nr_pages,
865 async_cow->write_flags)) {
866 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
867 struct page *p = async_extent->pages[0];
868 const u64 start = async_extent->start;
869 const u64 end = start + async_extent->ram_size - 1;
871 p->mapping = inode->i_mapping;
872 tree->ops->writepage_end_io_hook(p, start, end,
875 extent_clear_unlock_delalloc(inode, start, end, end,
879 free_async_extent_pages(async_extent);
881 alloc_hint = ins.objectid + ins.offset;
887 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
888 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
890 extent_clear_unlock_delalloc(inode, async_extent->start,
891 async_extent->start +
892 async_extent->ram_size - 1,
893 async_extent->start +
894 async_extent->ram_size - 1,
895 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
896 EXTENT_DELALLOC_NEW |
897 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
898 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
899 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
901 free_async_extent_pages(async_extent);
906 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
909 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
910 struct extent_map *em;
913 read_lock(&em_tree->lock);
914 em = search_extent_mapping(em_tree, start, num_bytes);
917 * if block start isn't an actual block number then find the
918 * first block in this inode and use that as a hint. If that
919 * block is also bogus then just don't worry about it.
921 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
923 em = search_extent_mapping(em_tree, 0, 0);
924 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
925 alloc_hint = em->block_start;
929 alloc_hint = em->block_start;
933 read_unlock(&em_tree->lock);
939 * when extent_io.c finds a delayed allocation range in the file,
940 * the call backs end up in this code. The basic idea is to
941 * allocate extents on disk for the range, and create ordered data structs
942 * in ram to track those extents.
944 * locked_page is the page that writepage had locked already. We use
945 * it to make sure we don't do extra locks or unlocks.
947 * *page_started is set to one if we unlock locked_page and do everything
948 * required to start IO on it. It may be clean and already done with
951 static noinline int cow_file_range(struct inode *inode,
952 struct page *locked_page,
953 u64 start, u64 end, u64 delalloc_end,
954 int *page_started, unsigned long *nr_written,
955 int unlock, struct btrfs_dedupe_hash *hash)
957 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
958 struct btrfs_root *root = BTRFS_I(inode)->root;
961 unsigned long ram_size;
962 u64 cur_alloc_size = 0;
963 u64 blocksize = fs_info->sectorsize;
964 struct btrfs_key ins;
965 struct extent_map *em;
967 unsigned long page_ops;
968 bool extent_reserved = false;
971 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
977 num_bytes = ALIGN(end - start + 1, blocksize);
978 num_bytes = max(blocksize, num_bytes);
979 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
981 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
984 /* lets try to make an inline extent */
985 ret = cow_file_range_inline(inode, start, end, 0,
986 BTRFS_COMPRESS_NONE, NULL);
989 * We use DO_ACCOUNTING here because we need the
990 * delalloc_release_metadata to be run _after_ we drop
991 * our outstanding extent for clearing delalloc for this
994 extent_clear_unlock_delalloc(inode, start, end,
996 EXTENT_LOCKED | EXTENT_DELALLOC |
997 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
998 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
999 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1000 PAGE_END_WRITEBACK);
1001 *nr_written = *nr_written +
1002 (end - start + PAGE_SIZE) / PAGE_SIZE;
1005 } else if (ret < 0) {
1010 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1011 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1012 start + num_bytes - 1, 0);
1014 while (num_bytes > 0) {
1015 cur_alloc_size = num_bytes;
1016 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1017 fs_info->sectorsize, 0, alloc_hint,
1021 cur_alloc_size = ins.offset;
1022 extent_reserved = true;
1024 ram_size = ins.offset;
1025 em = create_io_em(inode, start, ins.offset, /* len */
1026 start, /* orig_start */
1027 ins.objectid, /* block_start */
1028 ins.offset, /* block_len */
1029 ins.offset, /* orig_block_len */
1030 ram_size, /* ram_bytes */
1031 BTRFS_COMPRESS_NONE, /* compress_type */
1032 BTRFS_ORDERED_REGULAR /* type */);
1035 free_extent_map(em);
1037 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1038 ram_size, cur_alloc_size, 0);
1040 goto out_drop_extent_cache;
1042 if (root->root_key.objectid ==
1043 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1044 ret = btrfs_reloc_clone_csums(inode, start,
1047 * Only drop cache here, and process as normal.
1049 * We must not allow extent_clear_unlock_delalloc()
1050 * at out_unlock label to free meta of this ordered
1051 * extent, as its meta should be freed by
1052 * btrfs_finish_ordered_io().
1054 * So we must continue until @start is increased to
1055 * skip current ordered extent.
1058 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1059 start + ram_size - 1, 0);
1062 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1064 /* we're not doing compressed IO, don't unlock the first
1065 * page (which the caller expects to stay locked), don't
1066 * clear any dirty bits and don't set any writeback bits
1068 * Do set the Private2 bit so we know this page was properly
1069 * setup for writepage
1071 page_ops = unlock ? PAGE_UNLOCK : 0;
1072 page_ops |= PAGE_SET_PRIVATE2;
1074 extent_clear_unlock_delalloc(inode, start,
1075 start + ram_size - 1,
1076 delalloc_end, locked_page,
1077 EXTENT_LOCKED | EXTENT_DELALLOC,
1079 if (num_bytes < cur_alloc_size)
1082 num_bytes -= cur_alloc_size;
1083 alloc_hint = ins.objectid + ins.offset;
1084 start += cur_alloc_size;
1085 extent_reserved = false;
1088 * btrfs_reloc_clone_csums() error, since start is increased
1089 * extent_clear_unlock_delalloc() at out_unlock label won't
1090 * free metadata of current ordered extent, we're OK to exit.
1098 out_drop_extent_cache:
1099 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1101 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1102 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1104 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1105 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1106 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1109 * If we reserved an extent for our delalloc range (or a subrange) and
1110 * failed to create the respective ordered extent, then it means that
1111 * when we reserved the extent we decremented the extent's size from
1112 * the data space_info's bytes_may_use counter and incremented the
1113 * space_info's bytes_reserved counter by the same amount. We must make
1114 * sure extent_clear_unlock_delalloc() does not try to decrement again
1115 * the data space_info's bytes_may_use counter, therefore we do not pass
1116 * it the flag EXTENT_CLEAR_DATA_RESV.
1118 if (extent_reserved) {
1119 extent_clear_unlock_delalloc(inode, start,
1120 start + cur_alloc_size,
1121 start + cur_alloc_size,
1125 start += cur_alloc_size;
1129 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1131 clear_bits | EXTENT_CLEAR_DATA_RESV,
1137 * work queue call back to started compression on a file and pages
1139 static noinline void async_cow_start(struct btrfs_work *work)
1141 struct async_cow *async_cow;
1143 async_cow = container_of(work, struct async_cow, work);
1145 compress_file_range(async_cow->inode, async_cow->locked_page,
1146 async_cow->start, async_cow->end, async_cow,
1148 if (num_added == 0) {
1149 btrfs_add_delayed_iput(async_cow->inode);
1150 async_cow->inode = NULL;
1155 * work queue call back to submit previously compressed pages
1157 static noinline void async_cow_submit(struct btrfs_work *work)
1159 struct btrfs_fs_info *fs_info;
1160 struct async_cow *async_cow;
1161 struct btrfs_root *root;
1162 unsigned long nr_pages;
1164 async_cow = container_of(work, struct async_cow, work);
1166 root = async_cow->root;
1167 fs_info = root->fs_info;
1168 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1172 * atomic_sub_return implies a barrier for waitqueue_active
1174 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1176 waitqueue_active(&fs_info->async_submit_wait))
1177 wake_up(&fs_info->async_submit_wait);
1179 if (async_cow->inode)
1180 submit_compressed_extents(async_cow->inode, async_cow);
1183 static noinline void async_cow_free(struct btrfs_work *work)
1185 struct async_cow *async_cow;
1186 async_cow = container_of(work, struct async_cow, work);
1187 if (async_cow->inode)
1188 btrfs_add_delayed_iput(async_cow->inode);
1192 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1193 u64 start, u64 end, int *page_started,
1194 unsigned long *nr_written,
1195 unsigned int write_flags)
1197 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1198 struct async_cow *async_cow;
1199 struct btrfs_root *root = BTRFS_I(inode)->root;
1200 unsigned long nr_pages;
1203 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1205 while (start < end) {
1206 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1207 BUG_ON(!async_cow); /* -ENOMEM */
1208 async_cow->inode = igrab(inode);
1209 async_cow->root = root;
1210 async_cow->locked_page = locked_page;
1211 async_cow->start = start;
1212 async_cow->write_flags = write_flags;
1214 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1215 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1218 cur_end = min(end, start + SZ_512K - 1);
1220 async_cow->end = cur_end;
1221 INIT_LIST_HEAD(&async_cow->extents);
1223 btrfs_init_work(&async_cow->work,
1224 btrfs_delalloc_helper,
1225 async_cow_start, async_cow_submit,
1228 nr_pages = (cur_end - start + PAGE_SIZE) >>
1230 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1232 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1234 *nr_written += nr_pages;
1235 start = cur_end + 1;
1241 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1242 u64 bytenr, u64 num_bytes)
1245 struct btrfs_ordered_sum *sums;
1248 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1249 bytenr + num_bytes - 1, &list, 0);
1250 if (ret == 0 && list_empty(&list))
1253 while (!list_empty(&list)) {
1254 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1255 list_del(&sums->list);
1264 * when nowcow writeback call back. This checks for snapshots or COW copies
1265 * of the extents that exist in the file, and COWs the file as required.
1267 * If no cow copies or snapshots exist, we write directly to the existing
1270 static noinline int run_delalloc_nocow(struct inode *inode,
1271 struct page *locked_page,
1272 u64 start, u64 end, int *page_started, int force,
1273 unsigned long *nr_written)
1275 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1276 struct btrfs_root *root = BTRFS_I(inode)->root;
1277 struct extent_buffer *leaf;
1278 struct btrfs_path *path;
1279 struct btrfs_file_extent_item *fi;
1280 struct btrfs_key found_key;
1281 struct extent_map *em;
1296 u64 ino = btrfs_ino(BTRFS_I(inode));
1298 path = btrfs_alloc_path();
1300 extent_clear_unlock_delalloc(inode, start, end, end,
1302 EXTENT_LOCKED | EXTENT_DELALLOC |
1303 EXTENT_DO_ACCOUNTING |
1304 EXTENT_DEFRAG, PAGE_UNLOCK |
1306 PAGE_SET_WRITEBACK |
1307 PAGE_END_WRITEBACK);
1311 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1313 cow_start = (u64)-1;
1316 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1320 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1321 leaf = path->nodes[0];
1322 btrfs_item_key_to_cpu(leaf, &found_key,
1323 path->slots[0] - 1);
1324 if (found_key.objectid == ino &&
1325 found_key.type == BTRFS_EXTENT_DATA_KEY)
1330 leaf = path->nodes[0];
1331 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1332 ret = btrfs_next_leaf(root, path);
1334 if (cow_start != (u64)-1)
1335 cur_offset = cow_start;
1340 leaf = path->nodes[0];
1346 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1348 if (found_key.objectid > ino)
1350 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1351 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1355 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1356 found_key.offset > end)
1359 if (found_key.offset > cur_offset) {
1360 extent_end = found_key.offset;
1365 fi = btrfs_item_ptr(leaf, path->slots[0],
1366 struct btrfs_file_extent_item);
1367 extent_type = btrfs_file_extent_type(leaf, fi);
1369 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1370 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1371 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1372 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1373 extent_offset = btrfs_file_extent_offset(leaf, fi);
1374 extent_end = found_key.offset +
1375 btrfs_file_extent_num_bytes(leaf, fi);
1377 btrfs_file_extent_disk_num_bytes(leaf, fi);
1378 if (extent_end <= start) {
1382 if (disk_bytenr == 0)
1384 if (btrfs_file_extent_compression(leaf, fi) ||
1385 btrfs_file_extent_encryption(leaf, fi) ||
1386 btrfs_file_extent_other_encoding(leaf, fi))
1388 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1390 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1392 ret = btrfs_cross_ref_exist(root, ino,
1394 extent_offset, disk_bytenr);
1397 * ret could be -EIO if the above fails to read
1401 if (cow_start != (u64)-1)
1402 cur_offset = cow_start;
1406 WARN_ON_ONCE(nolock);
1409 disk_bytenr += extent_offset;
1410 disk_bytenr += cur_offset - found_key.offset;
1411 num_bytes = min(end + 1, extent_end) - cur_offset;
1413 * if there are pending snapshots for this root,
1414 * we fall into common COW way.
1417 err = btrfs_start_write_no_snapshotting(root);
1422 * force cow if csum exists in the range.
1423 * this ensure that csum for a given extent are
1424 * either valid or do not exist.
1426 ret = csum_exist_in_range(fs_info, disk_bytenr,
1430 btrfs_end_write_no_snapshotting(root);
1433 * ret could be -EIO if the above fails to read
1437 if (cow_start != (u64)-1)
1438 cur_offset = cow_start;
1441 WARN_ON_ONCE(nolock);
1444 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1446 btrfs_end_write_no_snapshotting(root);
1450 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1451 extent_end = found_key.offset +
1452 btrfs_file_extent_inline_len(leaf,
1453 path->slots[0], fi);
1454 extent_end = ALIGN(extent_end,
1455 fs_info->sectorsize);
1460 if (extent_end <= start) {
1462 if (!nolock && nocow)
1463 btrfs_end_write_no_snapshotting(root);
1465 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1469 if (cow_start == (u64)-1)
1470 cow_start = cur_offset;
1471 cur_offset = extent_end;
1472 if (cur_offset > end)
1478 btrfs_release_path(path);
1479 if (cow_start != (u64)-1) {
1480 ret = cow_file_range(inode, locked_page,
1481 cow_start, found_key.offset - 1,
1482 end, page_started, nr_written, 1,
1485 if (!nolock && nocow)
1486 btrfs_end_write_no_snapshotting(root);
1488 btrfs_dec_nocow_writers(fs_info,
1492 cow_start = (u64)-1;
1495 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1496 u64 orig_start = found_key.offset - extent_offset;
1498 em = create_io_em(inode, cur_offset, num_bytes,
1500 disk_bytenr, /* block_start */
1501 num_bytes, /* block_len */
1502 disk_num_bytes, /* orig_block_len */
1503 ram_bytes, BTRFS_COMPRESS_NONE,
1504 BTRFS_ORDERED_PREALLOC);
1506 if (!nolock && nocow)
1507 btrfs_end_write_no_snapshotting(root);
1509 btrfs_dec_nocow_writers(fs_info,
1514 free_extent_map(em);
1517 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1518 type = BTRFS_ORDERED_PREALLOC;
1520 type = BTRFS_ORDERED_NOCOW;
1523 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1524 num_bytes, num_bytes, type);
1526 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1527 BUG_ON(ret); /* -ENOMEM */
1529 if (root->root_key.objectid ==
1530 BTRFS_DATA_RELOC_TREE_OBJECTID)
1532 * Error handled later, as we must prevent
1533 * extent_clear_unlock_delalloc() in error handler
1534 * from freeing metadata of created ordered extent.
1536 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1539 extent_clear_unlock_delalloc(inode, cur_offset,
1540 cur_offset + num_bytes - 1, end,
1541 locked_page, EXTENT_LOCKED |
1543 EXTENT_CLEAR_DATA_RESV,
1544 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1546 if (!nolock && nocow)
1547 btrfs_end_write_no_snapshotting(root);
1548 cur_offset = extent_end;
1551 * btrfs_reloc_clone_csums() error, now we're OK to call error
1552 * handler, as metadata for created ordered extent will only
1553 * be freed by btrfs_finish_ordered_io().
1557 if (cur_offset > end)
1560 btrfs_release_path(path);
1562 if (cur_offset <= end && cow_start == (u64)-1) {
1563 cow_start = cur_offset;
1567 if (cow_start != (u64)-1) {
1568 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1569 page_started, nr_written, 1, NULL);
1575 if (ret && cur_offset < end)
1576 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1577 locked_page, EXTENT_LOCKED |
1578 EXTENT_DELALLOC | EXTENT_DEFRAG |
1579 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1581 PAGE_SET_WRITEBACK |
1582 PAGE_END_WRITEBACK);
1583 btrfs_free_path(path);
1587 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1590 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1591 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1595 * @defrag_bytes is a hint value, no spinlock held here,
1596 * if is not zero, it means the file is defragging.
1597 * Force cow if given extent needs to be defragged.
1599 if (BTRFS_I(inode)->defrag_bytes &&
1600 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1601 EXTENT_DEFRAG, 0, NULL))
1608 * extent_io.c call back to do delayed allocation processing
1610 static int run_delalloc_range(void *private_data, struct page *locked_page,
1611 u64 start, u64 end, int *page_started,
1612 unsigned long *nr_written,
1613 struct writeback_control *wbc)
1615 struct inode *inode = private_data;
1617 int force_cow = need_force_cow(inode, start, end);
1618 unsigned int write_flags = wbc_to_write_flags(wbc);
1620 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1621 ret = run_delalloc_nocow(inode, locked_page, start, end,
1622 page_started, 1, nr_written);
1623 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1624 ret = run_delalloc_nocow(inode, locked_page, start, end,
1625 page_started, 0, nr_written);
1626 } else if (!inode_need_compress(inode, start, end)) {
1627 ret = cow_file_range(inode, locked_page, start, end, end,
1628 page_started, nr_written, 1, NULL);
1630 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1631 &BTRFS_I(inode)->runtime_flags);
1632 ret = cow_file_range_async(inode, locked_page, start, end,
1633 page_started, nr_written,
1637 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1641 static void btrfs_split_extent_hook(void *private_data,
1642 struct extent_state *orig, u64 split)
1644 struct inode *inode = private_data;
1647 /* not delalloc, ignore it */
1648 if (!(orig->state & EXTENT_DELALLOC))
1651 size = orig->end - orig->start + 1;
1652 if (size > BTRFS_MAX_EXTENT_SIZE) {
1657 * See the explanation in btrfs_merge_extent_hook, the same
1658 * applies here, just in reverse.
1660 new_size = orig->end - split + 1;
1661 num_extents = count_max_extents(new_size);
1662 new_size = split - orig->start;
1663 num_extents += count_max_extents(new_size);
1664 if (count_max_extents(size) >= num_extents)
1668 spin_lock(&BTRFS_I(inode)->lock);
1669 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1670 spin_unlock(&BTRFS_I(inode)->lock);
1674 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1675 * extents so we can keep track of new extents that are just merged onto old
1676 * extents, such as when we are doing sequential writes, so we can properly
1677 * account for the metadata space we'll need.
1679 static void btrfs_merge_extent_hook(void *private_data,
1680 struct extent_state *new,
1681 struct extent_state *other)
1683 struct inode *inode = private_data;
1684 u64 new_size, old_size;
1687 /* not delalloc, ignore it */
1688 if (!(other->state & EXTENT_DELALLOC))
1691 if (new->start > other->start)
1692 new_size = new->end - other->start + 1;
1694 new_size = other->end - new->start + 1;
1696 /* we're not bigger than the max, unreserve the space and go */
1697 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1698 spin_lock(&BTRFS_I(inode)->lock);
1699 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1700 spin_unlock(&BTRFS_I(inode)->lock);
1705 * We have to add up either side to figure out how many extents were
1706 * accounted for before we merged into one big extent. If the number of
1707 * extents we accounted for is <= the amount we need for the new range
1708 * then we can return, otherwise drop. Think of it like this
1712 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1713 * need 2 outstanding extents, on one side we have 1 and the other side
1714 * we have 1 so they are == and we can return. But in this case
1716 * [MAX_SIZE+4k][MAX_SIZE+4k]
1718 * Each range on their own accounts for 2 extents, but merged together
1719 * they are only 3 extents worth of accounting, so we need to drop in
1722 old_size = other->end - other->start + 1;
1723 num_extents = count_max_extents(old_size);
1724 old_size = new->end - new->start + 1;
1725 num_extents += count_max_extents(old_size);
1726 if (count_max_extents(new_size) >= num_extents)
1729 spin_lock(&BTRFS_I(inode)->lock);
1730 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1731 spin_unlock(&BTRFS_I(inode)->lock);
1734 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1735 struct inode *inode)
1737 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1739 spin_lock(&root->delalloc_lock);
1740 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1741 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1742 &root->delalloc_inodes);
1743 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1744 &BTRFS_I(inode)->runtime_flags);
1745 root->nr_delalloc_inodes++;
1746 if (root->nr_delalloc_inodes == 1) {
1747 spin_lock(&fs_info->delalloc_root_lock);
1748 BUG_ON(!list_empty(&root->delalloc_root));
1749 list_add_tail(&root->delalloc_root,
1750 &fs_info->delalloc_roots);
1751 spin_unlock(&fs_info->delalloc_root_lock);
1754 spin_unlock(&root->delalloc_lock);
1757 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1758 struct btrfs_inode *inode)
1760 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1762 spin_lock(&root->delalloc_lock);
1763 if (!list_empty(&inode->delalloc_inodes)) {
1764 list_del_init(&inode->delalloc_inodes);
1765 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1766 &inode->runtime_flags);
1767 root->nr_delalloc_inodes--;
1768 if (!root->nr_delalloc_inodes) {
1769 spin_lock(&fs_info->delalloc_root_lock);
1770 BUG_ON(list_empty(&root->delalloc_root));
1771 list_del_init(&root->delalloc_root);
1772 spin_unlock(&fs_info->delalloc_root_lock);
1775 spin_unlock(&root->delalloc_lock);
1779 * extent_io.c set_bit_hook, used to track delayed allocation
1780 * bytes in this file, and to maintain the list of inodes that
1781 * have pending delalloc work to be done.
1783 static void btrfs_set_bit_hook(void *private_data,
1784 struct extent_state *state, unsigned *bits)
1786 struct inode *inode = private_data;
1788 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1790 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1793 * set_bit and clear bit hooks normally require _irqsave/restore
1794 * but in this case, we are only testing for the DELALLOC
1795 * bit, which is only set or cleared with irqs on
1797 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1798 struct btrfs_root *root = BTRFS_I(inode)->root;
1799 u64 len = state->end + 1 - state->start;
1800 u32 num_extents = count_max_extents(len);
1801 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1803 spin_lock(&BTRFS_I(inode)->lock);
1804 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1805 spin_unlock(&BTRFS_I(inode)->lock);
1807 /* For sanity tests */
1808 if (btrfs_is_testing(fs_info))
1811 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1812 fs_info->delalloc_batch);
1813 spin_lock(&BTRFS_I(inode)->lock);
1814 BTRFS_I(inode)->delalloc_bytes += len;
1815 if (*bits & EXTENT_DEFRAG)
1816 BTRFS_I(inode)->defrag_bytes += len;
1817 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1818 &BTRFS_I(inode)->runtime_flags))
1819 btrfs_add_delalloc_inodes(root, inode);
1820 spin_unlock(&BTRFS_I(inode)->lock);
1823 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1824 (*bits & EXTENT_DELALLOC_NEW)) {
1825 spin_lock(&BTRFS_I(inode)->lock);
1826 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1828 spin_unlock(&BTRFS_I(inode)->lock);
1833 * extent_io.c clear_bit_hook, see set_bit_hook for why
1835 static void btrfs_clear_bit_hook(void *private_data,
1836 struct extent_state *state,
1839 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1840 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1841 u64 len = state->end + 1 - state->start;
1842 u32 num_extents = count_max_extents(len);
1844 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1845 spin_lock(&inode->lock);
1846 inode->defrag_bytes -= len;
1847 spin_unlock(&inode->lock);
1851 * set_bit and clear bit hooks normally require _irqsave/restore
1852 * but in this case, we are only testing for the DELALLOC
1853 * bit, which is only set or cleared with irqs on
1855 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1856 struct btrfs_root *root = inode->root;
1857 bool do_list = !btrfs_is_free_space_inode(inode);
1859 spin_lock(&inode->lock);
1860 btrfs_mod_outstanding_extents(inode, -num_extents);
1861 spin_unlock(&inode->lock);
1864 * We don't reserve metadata space for space cache inodes so we
1865 * don't need to call dellalloc_release_metadata if there is an
1868 if (*bits & EXTENT_CLEAR_META_RESV &&
1869 root != fs_info->tree_root)
1870 btrfs_delalloc_release_metadata(inode, len);
1872 /* For sanity tests. */
1873 if (btrfs_is_testing(fs_info))
1876 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1877 do_list && !(state->state & EXTENT_NORESERVE) &&
1878 (*bits & EXTENT_CLEAR_DATA_RESV))
1879 btrfs_free_reserved_data_space_noquota(
1883 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1884 fs_info->delalloc_batch);
1885 spin_lock(&inode->lock);
1886 inode->delalloc_bytes -= len;
1887 if (do_list && inode->delalloc_bytes == 0 &&
1888 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1889 &inode->runtime_flags))
1890 btrfs_del_delalloc_inode(root, inode);
1891 spin_unlock(&inode->lock);
1894 if ((state->state & EXTENT_DELALLOC_NEW) &&
1895 (*bits & EXTENT_DELALLOC_NEW)) {
1896 spin_lock(&inode->lock);
1897 ASSERT(inode->new_delalloc_bytes >= len);
1898 inode->new_delalloc_bytes -= len;
1899 spin_unlock(&inode->lock);
1904 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1905 * we don't create bios that span stripes or chunks
1907 * return 1 if page cannot be merged to bio
1908 * return 0 if page can be merged to bio
1909 * return error otherwise
1911 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1912 size_t size, struct bio *bio,
1913 unsigned long bio_flags)
1915 struct inode *inode = page->mapping->host;
1916 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1917 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1922 if (bio_flags & EXTENT_BIO_COMPRESSED)
1925 length = bio->bi_iter.bi_size;
1926 map_length = length;
1927 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1931 if (map_length < length + size)
1937 * in order to insert checksums into the metadata in large chunks,
1938 * we wait until bio submission time. All the pages in the bio are
1939 * checksummed and sums are attached onto the ordered extent record.
1941 * At IO completion time the cums attached on the ordered extent record
1942 * are inserted into the btree
1944 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1947 struct inode *inode = private_data;
1948 blk_status_t ret = 0;
1950 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1951 BUG_ON(ret); /* -ENOMEM */
1956 * in order to insert checksums into the metadata in large chunks,
1957 * we wait until bio submission time. All the pages in the bio are
1958 * checksummed and sums are attached onto the ordered extent record.
1960 * At IO completion time the cums attached on the ordered extent record
1961 * are inserted into the btree
1963 static blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1966 struct inode *inode = private_data;
1967 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1970 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1972 bio->bi_status = ret;
1979 * extent_io.c submission hook. This does the right thing for csum calculation
1980 * on write, or reading the csums from the tree before a read.
1982 * Rules about async/sync submit,
1983 * a) read: sync submit
1985 * b) write without checksum: sync submit
1987 * c) write with checksum:
1988 * c-1) if bio is issued by fsync: sync submit
1989 * (sync_writers != 0)
1991 * c-2) if root is reloc root: sync submit
1992 * (only in case of buffered IO)
1994 * c-3) otherwise: async submit
1996 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1997 int mirror_num, unsigned long bio_flags,
2000 struct inode *inode = private_data;
2001 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2002 struct btrfs_root *root = BTRFS_I(inode)->root;
2003 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2004 blk_status_t ret = 0;
2006 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2008 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2010 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2011 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2013 if (bio_op(bio) != REQ_OP_WRITE) {
2014 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2018 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2019 ret = btrfs_submit_compressed_read(inode, bio,
2023 } else if (!skip_sum) {
2024 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2029 } else if (async && !skip_sum) {
2030 /* csum items have already been cloned */
2031 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2033 /* we're doing a write, do the async checksumming */
2034 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2036 btrfs_submit_bio_start,
2037 btrfs_submit_bio_done);
2039 } else if (!skip_sum) {
2040 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2046 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2050 bio->bi_status = ret;
2057 * given a list of ordered sums record them in the inode. This happens
2058 * at IO completion time based on sums calculated at bio submission time.
2060 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2061 struct inode *inode, struct list_head *list)
2063 struct btrfs_ordered_sum *sum;
2066 list_for_each_entry(sum, list, list) {
2067 trans->adding_csums = true;
2068 ret = btrfs_csum_file_blocks(trans,
2069 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2070 trans->adding_csums = false;
2077 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2078 unsigned int extra_bits,
2079 struct extent_state **cached_state, int dedupe)
2081 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2082 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2083 extra_bits, cached_state);
2086 /* see btrfs_writepage_start_hook for details on why this is required */
2087 struct btrfs_writepage_fixup {
2089 struct btrfs_work work;
2092 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2094 struct btrfs_writepage_fixup *fixup;
2095 struct btrfs_ordered_extent *ordered;
2096 struct extent_state *cached_state = NULL;
2097 struct extent_changeset *data_reserved = NULL;
2099 struct inode *inode;
2104 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2108 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2109 ClearPageChecked(page);
2113 inode = page->mapping->host;
2114 page_start = page_offset(page);
2115 page_end = page_offset(page) + PAGE_SIZE - 1;
2117 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2120 /* already ordered? We're done */
2121 if (PagePrivate2(page))
2124 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2127 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2128 page_end, &cached_state);
2130 btrfs_start_ordered_extent(inode, ordered, 1);
2131 btrfs_put_ordered_extent(ordered);
2135 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2138 mapping_set_error(page->mapping, ret);
2139 end_extent_writepage(page, ret, page_start, page_end);
2140 ClearPageChecked(page);
2144 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2147 mapping_set_error(page->mapping, ret);
2148 end_extent_writepage(page, ret, page_start, page_end);
2149 ClearPageChecked(page);
2153 ClearPageChecked(page);
2154 set_page_dirty(page);
2155 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2157 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2163 extent_changeset_free(data_reserved);
2167 * There are a few paths in the higher layers of the kernel that directly
2168 * set the page dirty bit without asking the filesystem if it is a
2169 * good idea. This causes problems because we want to make sure COW
2170 * properly happens and the data=ordered rules are followed.
2172 * In our case any range that doesn't have the ORDERED bit set
2173 * hasn't been properly setup for IO. We kick off an async process
2174 * to fix it up. The async helper will wait for ordered extents, set
2175 * the delalloc bit and make it safe to write the page.
2177 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2179 struct inode *inode = page->mapping->host;
2180 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2181 struct btrfs_writepage_fixup *fixup;
2183 /* this page is properly in the ordered list */
2184 if (TestClearPagePrivate2(page))
2187 if (PageChecked(page))
2190 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2194 SetPageChecked(page);
2196 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2197 btrfs_writepage_fixup_worker, NULL, NULL);
2199 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2203 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2204 struct inode *inode, u64 file_pos,
2205 u64 disk_bytenr, u64 disk_num_bytes,
2206 u64 num_bytes, u64 ram_bytes,
2207 u8 compression, u8 encryption,
2208 u16 other_encoding, int extent_type)
2210 struct btrfs_root *root = BTRFS_I(inode)->root;
2211 struct btrfs_file_extent_item *fi;
2212 struct btrfs_path *path;
2213 struct extent_buffer *leaf;
2214 struct btrfs_key ins;
2216 int extent_inserted = 0;
2219 path = btrfs_alloc_path();
2224 * we may be replacing one extent in the tree with another.
2225 * The new extent is pinned in the extent map, and we don't want
2226 * to drop it from the cache until it is completely in the btree.
2228 * So, tell btrfs_drop_extents to leave this extent in the cache.
2229 * the caller is expected to unpin it and allow it to be merged
2232 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2233 file_pos + num_bytes, NULL, 0,
2234 1, sizeof(*fi), &extent_inserted);
2238 if (!extent_inserted) {
2239 ins.objectid = btrfs_ino(BTRFS_I(inode));
2240 ins.offset = file_pos;
2241 ins.type = BTRFS_EXTENT_DATA_KEY;
2243 path->leave_spinning = 1;
2244 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2249 leaf = path->nodes[0];
2250 fi = btrfs_item_ptr(leaf, path->slots[0],
2251 struct btrfs_file_extent_item);
2252 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2253 btrfs_set_file_extent_type(leaf, fi, extent_type);
2254 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2255 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2256 btrfs_set_file_extent_offset(leaf, fi, 0);
2257 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2258 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2259 btrfs_set_file_extent_compression(leaf, fi, compression);
2260 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2261 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2263 btrfs_mark_buffer_dirty(leaf);
2264 btrfs_release_path(path);
2266 inode_add_bytes(inode, num_bytes);
2268 ins.objectid = disk_bytenr;
2269 ins.offset = disk_num_bytes;
2270 ins.type = BTRFS_EXTENT_ITEM_KEY;
2273 * Release the reserved range from inode dirty range map, as it is
2274 * already moved into delayed_ref_head
2276 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2280 ret = btrfs_alloc_reserved_file_extent(trans, root,
2281 btrfs_ino(BTRFS_I(inode)),
2282 file_pos, qg_released, &ins);
2284 btrfs_free_path(path);
2289 /* snapshot-aware defrag */
2290 struct sa_defrag_extent_backref {
2291 struct rb_node node;
2292 struct old_sa_defrag_extent *old;
2301 struct old_sa_defrag_extent {
2302 struct list_head list;
2303 struct new_sa_defrag_extent *new;
2312 struct new_sa_defrag_extent {
2313 struct rb_root root;
2314 struct list_head head;
2315 struct btrfs_path *path;
2316 struct inode *inode;
2324 static int backref_comp(struct sa_defrag_extent_backref *b1,
2325 struct sa_defrag_extent_backref *b2)
2327 if (b1->root_id < b2->root_id)
2329 else if (b1->root_id > b2->root_id)
2332 if (b1->inum < b2->inum)
2334 else if (b1->inum > b2->inum)
2337 if (b1->file_pos < b2->file_pos)
2339 else if (b1->file_pos > b2->file_pos)
2343 * [------------------------------] ===> (a range of space)
2344 * |<--->| |<---->| =============> (fs/file tree A)
2345 * |<---------------------------->| ===> (fs/file tree B)
2347 * A range of space can refer to two file extents in one tree while
2348 * refer to only one file extent in another tree.
2350 * So we may process a disk offset more than one time(two extents in A)
2351 * and locate at the same extent(one extent in B), then insert two same
2352 * backrefs(both refer to the extent in B).
2357 static void backref_insert(struct rb_root *root,
2358 struct sa_defrag_extent_backref *backref)
2360 struct rb_node **p = &root->rb_node;
2361 struct rb_node *parent = NULL;
2362 struct sa_defrag_extent_backref *entry;
2367 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2369 ret = backref_comp(backref, entry);
2373 p = &(*p)->rb_right;
2376 rb_link_node(&backref->node, parent, p);
2377 rb_insert_color(&backref->node, root);
2381 * Note the backref might has changed, and in this case we just return 0.
2383 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2386 struct btrfs_file_extent_item *extent;
2387 struct old_sa_defrag_extent *old = ctx;
2388 struct new_sa_defrag_extent *new = old->new;
2389 struct btrfs_path *path = new->path;
2390 struct btrfs_key key;
2391 struct btrfs_root *root;
2392 struct sa_defrag_extent_backref *backref;
2393 struct extent_buffer *leaf;
2394 struct inode *inode = new->inode;
2395 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2401 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2402 inum == btrfs_ino(BTRFS_I(inode)))
2405 key.objectid = root_id;
2406 key.type = BTRFS_ROOT_ITEM_KEY;
2407 key.offset = (u64)-1;
2409 root = btrfs_read_fs_root_no_name(fs_info, &key);
2411 if (PTR_ERR(root) == -ENOENT)
2414 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2415 inum, offset, root_id);
2416 return PTR_ERR(root);
2419 key.objectid = inum;
2420 key.type = BTRFS_EXTENT_DATA_KEY;
2421 if (offset > (u64)-1 << 32)
2424 key.offset = offset;
2426 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2427 if (WARN_ON(ret < 0))
2434 leaf = path->nodes[0];
2435 slot = path->slots[0];
2437 if (slot >= btrfs_header_nritems(leaf)) {
2438 ret = btrfs_next_leaf(root, path);
2441 } else if (ret > 0) {
2450 btrfs_item_key_to_cpu(leaf, &key, slot);
2452 if (key.objectid > inum)
2455 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2458 extent = btrfs_item_ptr(leaf, slot,
2459 struct btrfs_file_extent_item);
2461 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2465 * 'offset' refers to the exact key.offset,
2466 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2467 * (key.offset - extent_offset).
2469 if (key.offset != offset)
2472 extent_offset = btrfs_file_extent_offset(leaf, extent);
2473 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2475 if (extent_offset >= old->extent_offset + old->offset +
2476 old->len || extent_offset + num_bytes <=
2477 old->extent_offset + old->offset)
2482 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2488 backref->root_id = root_id;
2489 backref->inum = inum;
2490 backref->file_pos = offset;
2491 backref->num_bytes = num_bytes;
2492 backref->extent_offset = extent_offset;
2493 backref->generation = btrfs_file_extent_generation(leaf, extent);
2495 backref_insert(&new->root, backref);
2498 btrfs_release_path(path);
2503 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2504 struct new_sa_defrag_extent *new)
2506 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2507 struct old_sa_defrag_extent *old, *tmp;
2512 list_for_each_entry_safe(old, tmp, &new->head, list) {
2513 ret = iterate_inodes_from_logical(old->bytenr +
2514 old->extent_offset, fs_info,
2515 path, record_one_backref,
2517 if (ret < 0 && ret != -ENOENT)
2520 /* no backref to be processed for this extent */
2522 list_del(&old->list);
2527 if (list_empty(&new->head))
2533 static int relink_is_mergable(struct extent_buffer *leaf,
2534 struct btrfs_file_extent_item *fi,
2535 struct new_sa_defrag_extent *new)
2537 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2540 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2543 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2546 if (btrfs_file_extent_encryption(leaf, fi) ||
2547 btrfs_file_extent_other_encoding(leaf, fi))
2554 * Note the backref might has changed, and in this case we just return 0.
2556 static noinline int relink_extent_backref(struct btrfs_path *path,
2557 struct sa_defrag_extent_backref *prev,
2558 struct sa_defrag_extent_backref *backref)
2560 struct btrfs_file_extent_item *extent;
2561 struct btrfs_file_extent_item *item;
2562 struct btrfs_ordered_extent *ordered;
2563 struct btrfs_trans_handle *trans;
2564 struct btrfs_root *root;
2565 struct btrfs_key key;
2566 struct extent_buffer *leaf;
2567 struct old_sa_defrag_extent *old = backref->old;
2568 struct new_sa_defrag_extent *new = old->new;
2569 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2570 struct inode *inode;
2571 struct extent_state *cached = NULL;
2580 if (prev && prev->root_id == backref->root_id &&
2581 prev->inum == backref->inum &&
2582 prev->file_pos + prev->num_bytes == backref->file_pos)
2585 /* step 1: get root */
2586 key.objectid = backref->root_id;
2587 key.type = BTRFS_ROOT_ITEM_KEY;
2588 key.offset = (u64)-1;
2590 index = srcu_read_lock(&fs_info->subvol_srcu);
2592 root = btrfs_read_fs_root_no_name(fs_info, &key);
2594 srcu_read_unlock(&fs_info->subvol_srcu, index);
2595 if (PTR_ERR(root) == -ENOENT)
2597 return PTR_ERR(root);
2600 if (btrfs_root_readonly(root)) {
2601 srcu_read_unlock(&fs_info->subvol_srcu, index);
2605 /* step 2: get inode */
2606 key.objectid = backref->inum;
2607 key.type = BTRFS_INODE_ITEM_KEY;
2610 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2611 if (IS_ERR(inode)) {
2612 srcu_read_unlock(&fs_info->subvol_srcu, index);
2616 srcu_read_unlock(&fs_info->subvol_srcu, index);
2618 /* step 3: relink backref */
2619 lock_start = backref->file_pos;
2620 lock_end = backref->file_pos + backref->num_bytes - 1;
2621 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2624 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2626 btrfs_put_ordered_extent(ordered);
2630 trans = btrfs_join_transaction(root);
2631 if (IS_ERR(trans)) {
2632 ret = PTR_ERR(trans);
2636 key.objectid = backref->inum;
2637 key.type = BTRFS_EXTENT_DATA_KEY;
2638 key.offset = backref->file_pos;
2640 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2643 } else if (ret > 0) {
2648 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2649 struct btrfs_file_extent_item);
2651 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2652 backref->generation)
2655 btrfs_release_path(path);
2657 start = backref->file_pos;
2658 if (backref->extent_offset < old->extent_offset + old->offset)
2659 start += old->extent_offset + old->offset -
2660 backref->extent_offset;
2662 len = min(backref->extent_offset + backref->num_bytes,
2663 old->extent_offset + old->offset + old->len);
2664 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2666 ret = btrfs_drop_extents(trans, root, inode, start,
2671 key.objectid = btrfs_ino(BTRFS_I(inode));
2672 key.type = BTRFS_EXTENT_DATA_KEY;
2675 path->leave_spinning = 1;
2677 struct btrfs_file_extent_item *fi;
2679 struct btrfs_key found_key;
2681 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2686 leaf = path->nodes[0];
2687 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2689 fi = btrfs_item_ptr(leaf, path->slots[0],
2690 struct btrfs_file_extent_item);
2691 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2693 if (extent_len + found_key.offset == start &&
2694 relink_is_mergable(leaf, fi, new)) {
2695 btrfs_set_file_extent_num_bytes(leaf, fi,
2697 btrfs_mark_buffer_dirty(leaf);
2698 inode_add_bytes(inode, len);
2704 btrfs_release_path(path);
2709 ret = btrfs_insert_empty_item(trans, root, path, &key,
2712 btrfs_abort_transaction(trans, ret);
2716 leaf = path->nodes[0];
2717 item = btrfs_item_ptr(leaf, path->slots[0],
2718 struct btrfs_file_extent_item);
2719 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2720 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2721 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2722 btrfs_set_file_extent_num_bytes(leaf, item, len);
2723 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2724 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2725 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2726 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2727 btrfs_set_file_extent_encryption(leaf, item, 0);
2728 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2730 btrfs_mark_buffer_dirty(leaf);
2731 inode_add_bytes(inode, len);
2732 btrfs_release_path(path);
2734 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2736 backref->root_id, backref->inum,
2737 new->file_pos); /* start - extent_offset */
2739 btrfs_abort_transaction(trans, ret);
2745 btrfs_release_path(path);
2746 path->leave_spinning = 0;
2747 btrfs_end_transaction(trans);
2749 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2755 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2757 struct old_sa_defrag_extent *old, *tmp;
2762 list_for_each_entry_safe(old, tmp, &new->head, list) {
2768 static void relink_file_extents(struct new_sa_defrag_extent *new)
2770 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2771 struct btrfs_path *path;
2772 struct sa_defrag_extent_backref *backref;
2773 struct sa_defrag_extent_backref *prev = NULL;
2774 struct inode *inode;
2775 struct rb_node *node;
2780 path = btrfs_alloc_path();
2784 if (!record_extent_backrefs(path, new)) {
2785 btrfs_free_path(path);
2788 btrfs_release_path(path);
2791 node = rb_first(&new->root);
2794 rb_erase(node, &new->root);
2796 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2798 ret = relink_extent_backref(path, prev, backref);
2811 btrfs_free_path(path);
2813 free_sa_defrag_extent(new);
2815 atomic_dec(&fs_info->defrag_running);
2816 wake_up(&fs_info->transaction_wait);
2819 static struct new_sa_defrag_extent *
2820 record_old_file_extents(struct inode *inode,
2821 struct btrfs_ordered_extent *ordered)
2823 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2824 struct btrfs_root *root = BTRFS_I(inode)->root;
2825 struct btrfs_path *path;
2826 struct btrfs_key key;
2827 struct old_sa_defrag_extent *old;
2828 struct new_sa_defrag_extent *new;
2831 new = kmalloc(sizeof(*new), GFP_NOFS);
2836 new->file_pos = ordered->file_offset;
2837 new->len = ordered->len;
2838 new->bytenr = ordered->start;
2839 new->disk_len = ordered->disk_len;
2840 new->compress_type = ordered->compress_type;
2841 new->root = RB_ROOT;
2842 INIT_LIST_HEAD(&new->head);
2844 path = btrfs_alloc_path();
2848 key.objectid = btrfs_ino(BTRFS_I(inode));
2849 key.type = BTRFS_EXTENT_DATA_KEY;
2850 key.offset = new->file_pos;
2852 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2855 if (ret > 0 && path->slots[0] > 0)
2858 /* find out all the old extents for the file range */
2860 struct btrfs_file_extent_item *extent;
2861 struct extent_buffer *l;
2870 slot = path->slots[0];
2872 if (slot >= btrfs_header_nritems(l)) {
2873 ret = btrfs_next_leaf(root, path);
2881 btrfs_item_key_to_cpu(l, &key, slot);
2883 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2885 if (key.type != BTRFS_EXTENT_DATA_KEY)
2887 if (key.offset >= new->file_pos + new->len)
2890 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2892 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2893 if (key.offset + num_bytes < new->file_pos)
2896 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2900 extent_offset = btrfs_file_extent_offset(l, extent);
2902 old = kmalloc(sizeof(*old), GFP_NOFS);
2906 offset = max(new->file_pos, key.offset);
2907 end = min(new->file_pos + new->len, key.offset + num_bytes);
2909 old->bytenr = disk_bytenr;
2910 old->extent_offset = extent_offset;
2911 old->offset = offset - key.offset;
2912 old->len = end - offset;
2915 list_add_tail(&old->list, &new->head);
2921 btrfs_free_path(path);
2922 atomic_inc(&fs_info->defrag_running);
2927 btrfs_free_path(path);
2929 free_sa_defrag_extent(new);
2933 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2936 struct btrfs_block_group_cache *cache;
2938 cache = btrfs_lookup_block_group(fs_info, start);
2941 spin_lock(&cache->lock);
2942 cache->delalloc_bytes -= len;
2943 spin_unlock(&cache->lock);
2945 btrfs_put_block_group(cache);
2948 /* as ordered data IO finishes, this gets called so we can finish
2949 * an ordered extent if the range of bytes in the file it covers are
2952 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2954 struct inode *inode = ordered_extent->inode;
2955 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2956 struct btrfs_root *root = BTRFS_I(inode)->root;
2957 struct btrfs_trans_handle *trans = NULL;
2958 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2959 struct extent_state *cached_state = NULL;
2960 struct new_sa_defrag_extent *new = NULL;
2961 int compress_type = 0;
2963 u64 logical_len = ordered_extent->len;
2965 bool truncated = false;
2966 bool range_locked = false;
2967 bool clear_new_delalloc_bytes = false;
2969 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2970 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2971 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2972 clear_new_delalloc_bytes = true;
2974 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2976 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2981 btrfs_free_io_failure_record(BTRFS_I(inode),
2982 ordered_extent->file_offset,
2983 ordered_extent->file_offset +
2984 ordered_extent->len - 1);
2986 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2988 logical_len = ordered_extent->truncated_len;
2989 /* Truncated the entire extent, don't bother adding */
2994 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2995 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2998 * For mwrite(mmap + memset to write) case, we still reserve
2999 * space for NOCOW range.
3000 * As NOCOW won't cause a new delayed ref, just free the space
3002 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3003 ordered_extent->len);
3004 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3006 trans = btrfs_join_transaction_nolock(root);
3008 trans = btrfs_join_transaction(root);
3009 if (IS_ERR(trans)) {
3010 ret = PTR_ERR(trans);
3014 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3015 ret = btrfs_update_inode_fallback(trans, root, inode);
3016 if (ret) /* -ENOMEM or corruption */
3017 btrfs_abort_transaction(trans, ret);
3021 range_locked = true;
3022 lock_extent_bits(io_tree, ordered_extent->file_offset,
3023 ordered_extent->file_offset + ordered_extent->len - 1,
3026 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3027 ordered_extent->file_offset + ordered_extent->len - 1,
3028 EXTENT_DEFRAG, 0, cached_state);
3030 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3031 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3032 /* the inode is shared */
3033 new = record_old_file_extents(inode, ordered_extent);
3035 clear_extent_bit(io_tree, ordered_extent->file_offset,
3036 ordered_extent->file_offset + ordered_extent->len - 1,
3037 EXTENT_DEFRAG, 0, 0, &cached_state);
3041 trans = btrfs_join_transaction_nolock(root);
3043 trans = btrfs_join_transaction(root);
3044 if (IS_ERR(trans)) {
3045 ret = PTR_ERR(trans);
3050 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3052 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3053 compress_type = ordered_extent->compress_type;
3054 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3055 BUG_ON(compress_type);
3056 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3057 ordered_extent->len);
3058 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3059 ordered_extent->file_offset,
3060 ordered_extent->file_offset +
3063 BUG_ON(root == fs_info->tree_root);
3064 ret = insert_reserved_file_extent(trans, inode,
3065 ordered_extent->file_offset,
3066 ordered_extent->start,
3067 ordered_extent->disk_len,
3068 logical_len, logical_len,
3069 compress_type, 0, 0,
3070 BTRFS_FILE_EXTENT_REG);
3072 btrfs_release_delalloc_bytes(fs_info,
3073 ordered_extent->start,
3074 ordered_extent->disk_len);
3076 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3077 ordered_extent->file_offset, ordered_extent->len,
3080 btrfs_abort_transaction(trans, ret);
3084 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3086 btrfs_abort_transaction(trans, ret);
3090 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3091 ret = btrfs_update_inode_fallback(trans, root, inode);
3092 if (ret) { /* -ENOMEM or corruption */
3093 btrfs_abort_transaction(trans, ret);
3098 if (range_locked || clear_new_delalloc_bytes) {
3099 unsigned int clear_bits = 0;
3102 clear_bits |= EXTENT_LOCKED;
3103 if (clear_new_delalloc_bytes)
3104 clear_bits |= EXTENT_DELALLOC_NEW;
3105 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3106 ordered_extent->file_offset,
3107 ordered_extent->file_offset +
3108 ordered_extent->len - 1,
3110 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3115 btrfs_end_transaction(trans);
3117 if (ret || truncated) {
3121 start = ordered_extent->file_offset + logical_len;
3123 start = ordered_extent->file_offset;
3124 end = ordered_extent->file_offset + ordered_extent->len - 1;
3125 clear_extent_uptodate(io_tree, start, end, NULL);
3127 /* Drop the cache for the part of the extent we didn't write. */
3128 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3131 * If the ordered extent had an IOERR or something else went
3132 * wrong we need to return the space for this ordered extent
3133 * back to the allocator. We only free the extent in the
3134 * truncated case if we didn't write out the extent at all.
3136 if ((ret || !logical_len) &&
3137 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3138 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3139 btrfs_free_reserved_extent(fs_info,
3140 ordered_extent->start,
3141 ordered_extent->disk_len, 1);
3146 * This needs to be done to make sure anybody waiting knows we are done
3147 * updating everything for this ordered extent.
3149 btrfs_remove_ordered_extent(inode, ordered_extent);
3151 /* for snapshot-aware defrag */
3154 free_sa_defrag_extent(new);
3155 atomic_dec(&fs_info->defrag_running);
3157 relink_file_extents(new);
3162 btrfs_put_ordered_extent(ordered_extent);
3163 /* once for the tree */
3164 btrfs_put_ordered_extent(ordered_extent);
3169 static void finish_ordered_fn(struct btrfs_work *work)
3171 struct btrfs_ordered_extent *ordered_extent;
3172 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3173 btrfs_finish_ordered_io(ordered_extent);
3176 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3177 struct extent_state *state, int uptodate)
3179 struct inode *inode = page->mapping->host;
3180 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3181 struct btrfs_ordered_extent *ordered_extent = NULL;
3182 struct btrfs_workqueue *wq;
3183 btrfs_work_func_t func;
3185 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3187 ClearPagePrivate2(page);
3188 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3189 end - start + 1, uptodate))
3192 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3193 wq = fs_info->endio_freespace_worker;
3194 func = btrfs_freespace_write_helper;
3196 wq = fs_info->endio_write_workers;
3197 func = btrfs_endio_write_helper;
3200 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3202 btrfs_queue_work(wq, &ordered_extent->work);
3205 static int __readpage_endio_check(struct inode *inode,
3206 struct btrfs_io_bio *io_bio,
3207 int icsum, struct page *page,
3208 int pgoff, u64 start, size_t len)
3214 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3216 kaddr = kmap_atomic(page);
3217 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3218 btrfs_csum_final(csum, (u8 *)&csum);
3219 if (csum != csum_expected)
3222 kunmap_atomic(kaddr);
3225 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3226 io_bio->mirror_num);
3227 memset(kaddr + pgoff, 1, len);
3228 flush_dcache_page(page);
3229 kunmap_atomic(kaddr);
3234 * when reads are done, we need to check csums to verify the data is correct
3235 * if there's a match, we allow the bio to finish. If not, the code in
3236 * extent_io.c will try to find good copies for us.
3238 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3239 u64 phy_offset, struct page *page,
3240 u64 start, u64 end, int mirror)
3242 size_t offset = start - page_offset(page);
3243 struct inode *inode = page->mapping->host;
3244 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3245 struct btrfs_root *root = BTRFS_I(inode)->root;
3247 if (PageChecked(page)) {
3248 ClearPageChecked(page);
3252 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3255 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3256 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3257 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3261 phy_offset >>= inode->i_sb->s_blocksize_bits;
3262 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3263 start, (size_t)(end - start + 1));
3267 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3269 * @inode: The inode we want to perform iput on
3271 * This function uses the generic vfs_inode::i_count to track whether we should
3272 * just decrement it (in case it's > 1) or if this is the last iput then link
3273 * the inode to the delayed iput machinery. Delayed iputs are processed at
3274 * transaction commit time/superblock commit/cleaner kthread.
3276 void btrfs_add_delayed_iput(struct inode *inode)
3278 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3279 struct btrfs_inode *binode = BTRFS_I(inode);
3281 if (atomic_add_unless(&inode->i_count, -1, 1))
3284 spin_lock(&fs_info->delayed_iput_lock);
3285 ASSERT(list_empty(&binode->delayed_iput));
3286 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3287 spin_unlock(&fs_info->delayed_iput_lock);
3290 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3293 spin_lock(&fs_info->delayed_iput_lock);
3294 while (!list_empty(&fs_info->delayed_iputs)) {
3295 struct btrfs_inode *inode;
3297 inode = list_first_entry(&fs_info->delayed_iputs,
3298 struct btrfs_inode, delayed_iput);
3299 list_del_init(&inode->delayed_iput);
3300 spin_unlock(&fs_info->delayed_iput_lock);
3301 iput(&inode->vfs_inode);
3302 spin_lock(&fs_info->delayed_iput_lock);
3304 spin_unlock(&fs_info->delayed_iput_lock);
3308 * This is called in transaction commit time. If there are no orphan
3309 * files in the subvolume, it removes orphan item and frees block_rsv
3312 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3313 struct btrfs_root *root)
3315 struct btrfs_fs_info *fs_info = root->fs_info;
3316 struct btrfs_block_rsv *block_rsv;
3319 if (atomic_read(&root->orphan_inodes) ||
3320 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3323 spin_lock(&root->orphan_lock);
3324 if (atomic_read(&root->orphan_inodes)) {
3325 spin_unlock(&root->orphan_lock);
3329 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3330 spin_unlock(&root->orphan_lock);
3334 block_rsv = root->orphan_block_rsv;
3335 root->orphan_block_rsv = NULL;
3336 spin_unlock(&root->orphan_lock);
3338 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3339 btrfs_root_refs(&root->root_item) > 0) {
3340 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3341 root->root_key.objectid);
3343 btrfs_abort_transaction(trans, ret);
3345 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3350 WARN_ON(block_rsv->size > 0);
3351 btrfs_free_block_rsv(fs_info, block_rsv);
3356 * This creates an orphan entry for the given inode in case something goes
3357 * wrong in the middle of an unlink/truncate.
3359 * NOTE: caller of this function should reserve 5 units of metadata for
3362 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3363 struct btrfs_inode *inode)
3365 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3366 struct btrfs_root *root = inode->root;
3367 struct btrfs_block_rsv *block_rsv = NULL;
3372 if (!root->orphan_block_rsv) {
3373 block_rsv = btrfs_alloc_block_rsv(fs_info,
3374 BTRFS_BLOCK_RSV_TEMP);
3379 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3380 &inode->runtime_flags)) {
3383 * For proper ENOSPC handling, we should do orphan
3384 * cleanup when mounting. But this introduces backward
3385 * compatibility issue.
3387 if (!xchg(&root->orphan_item_inserted, 1))
3395 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3396 &inode->runtime_flags))
3399 spin_lock(&root->orphan_lock);
3400 /* If someone has created ->orphan_block_rsv, be happy to use it. */
3401 if (!root->orphan_block_rsv) {
3402 root->orphan_block_rsv = block_rsv;
3403 } else if (block_rsv) {
3404 btrfs_free_block_rsv(fs_info, block_rsv);
3409 atomic_inc(&root->orphan_inodes);
3410 spin_unlock(&root->orphan_lock);
3412 /* grab metadata reservation from transaction handle */
3414 ret = btrfs_orphan_reserve_metadata(trans, inode);
3418 * dec doesn't need spin_lock as ->orphan_block_rsv
3419 * would be released only if ->orphan_inodes is
3422 atomic_dec(&root->orphan_inodes);
3423 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3424 &inode->runtime_flags);
3426 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3427 &inode->runtime_flags);
3432 /* insert an orphan item to track this unlinked/truncated file */
3434 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3437 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3438 &inode->runtime_flags);
3439 btrfs_orphan_release_metadata(inode);
3442 * btrfs_orphan_commit_root may race with us and set
3443 * ->orphan_block_rsv to zero, in order to avoid that,
3444 * decrease ->orphan_inodes after everything is done.
3446 atomic_dec(&root->orphan_inodes);
3447 if (ret != -EEXIST) {
3448 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3449 &inode->runtime_flags);
3450 btrfs_abort_transaction(trans, ret);
3457 /* insert an orphan item to track subvolume contains orphan files */
3459 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3460 root->root_key.objectid);
3461 if (ret && ret != -EEXIST) {
3462 btrfs_abort_transaction(trans, ret);
3470 * We have done the truncate/delete so we can go ahead and remove the orphan
3471 * item for this particular inode.
3473 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3474 struct btrfs_inode *inode)
3476 struct btrfs_root *root = inode->root;
3477 int delete_item = 0;
3480 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3481 &inode->runtime_flags))
3484 if (delete_item && trans)
3485 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3487 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3488 &inode->runtime_flags))
3489 btrfs_orphan_release_metadata(inode);
3492 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3493 * to zero, in order to avoid that, decrease ->orphan_inodes after
3494 * everything is done.
3497 atomic_dec(&root->orphan_inodes);
3503 * this cleans up any orphans that may be left on the list from the last use
3506 int btrfs_orphan_cleanup(struct btrfs_root *root)
3508 struct btrfs_fs_info *fs_info = root->fs_info;
3509 struct btrfs_path *path;
3510 struct extent_buffer *leaf;
3511 struct btrfs_key key, found_key;
3512 struct btrfs_trans_handle *trans;
3513 struct inode *inode;
3514 u64 last_objectid = 0;
3515 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3517 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3520 path = btrfs_alloc_path();
3525 path->reada = READA_BACK;
3527 key.objectid = BTRFS_ORPHAN_OBJECTID;
3528 key.type = BTRFS_ORPHAN_ITEM_KEY;
3529 key.offset = (u64)-1;
3532 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3537 * if ret == 0 means we found what we were searching for, which
3538 * is weird, but possible, so only screw with path if we didn't
3539 * find the key and see if we have stuff that matches
3543 if (path->slots[0] == 0)
3548 /* pull out the item */
3549 leaf = path->nodes[0];
3550 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3552 /* make sure the item matches what we want */
3553 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3555 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3558 /* release the path since we're done with it */
3559 btrfs_release_path(path);
3562 * this is where we are basically btrfs_lookup, without the
3563 * crossing root thing. we store the inode number in the
3564 * offset of the orphan item.
3567 if (found_key.offset == last_objectid) {
3569 "Error removing orphan entry, stopping orphan cleanup");
3574 last_objectid = found_key.offset;
3576 found_key.objectid = found_key.offset;
3577 found_key.type = BTRFS_INODE_ITEM_KEY;
3578 found_key.offset = 0;
3579 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3580 ret = PTR_ERR_OR_ZERO(inode);
3581 if (ret && ret != -ENOENT)
3584 if (ret == -ENOENT && root == fs_info->tree_root) {
3585 struct btrfs_root *dead_root;
3586 struct btrfs_fs_info *fs_info = root->fs_info;
3587 int is_dead_root = 0;
3590 * this is an orphan in the tree root. Currently these
3591 * could come from 2 sources:
3592 * a) a snapshot deletion in progress
3593 * b) a free space cache inode
3594 * We need to distinguish those two, as the snapshot
3595 * orphan must not get deleted.
3596 * find_dead_roots already ran before us, so if this
3597 * is a snapshot deletion, we should find the root
3598 * in the dead_roots list
3600 spin_lock(&fs_info->trans_lock);
3601 list_for_each_entry(dead_root, &fs_info->dead_roots,
3603 if (dead_root->root_key.objectid ==
3604 found_key.objectid) {
3609 spin_unlock(&fs_info->trans_lock);
3611 /* prevent this orphan from being found again */
3612 key.offset = found_key.objectid - 1;
3617 * Inode is already gone but the orphan item is still there,
3618 * kill the orphan item.
3620 if (ret == -ENOENT) {
3621 trans = btrfs_start_transaction(root, 1);
3622 if (IS_ERR(trans)) {
3623 ret = PTR_ERR(trans);
3626 btrfs_debug(fs_info, "auto deleting %Lu",
3627 found_key.objectid);
3628 ret = btrfs_del_orphan_item(trans, root,
3629 found_key.objectid);
3630 btrfs_end_transaction(trans);
3637 * add this inode to the orphan list so btrfs_orphan_del does
3638 * the proper thing when we hit it
3640 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3641 &BTRFS_I(inode)->runtime_flags);
3642 atomic_inc(&root->orphan_inodes);
3644 /* if we have links, this was a truncate, lets do that */
3645 if (inode->i_nlink) {
3646 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3652 /* 1 for the orphan item deletion. */
3653 trans = btrfs_start_transaction(root, 1);
3654 if (IS_ERR(trans)) {
3656 ret = PTR_ERR(trans);
3659 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3660 btrfs_end_transaction(trans);
3666 ret = btrfs_truncate(inode, false);
3668 btrfs_orphan_del(NULL, BTRFS_I(inode));
3673 /* this will do delete_inode and everything for us */
3678 /* release the path since we're done with it */
3679 btrfs_release_path(path);
3681 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3683 if (root->orphan_block_rsv)
3684 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3687 if (root->orphan_block_rsv ||
3688 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3689 trans = btrfs_join_transaction(root);
3691 btrfs_end_transaction(trans);
3695 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3697 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3701 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3702 btrfs_free_path(path);
3707 * very simple check to peek ahead in the leaf looking for xattrs. If we
3708 * don't find any xattrs, we know there can't be any acls.
3710 * slot is the slot the inode is in, objectid is the objectid of the inode
3712 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3713 int slot, u64 objectid,
3714 int *first_xattr_slot)
3716 u32 nritems = btrfs_header_nritems(leaf);
3717 struct btrfs_key found_key;
3718 static u64 xattr_access = 0;
3719 static u64 xattr_default = 0;
3722 if (!xattr_access) {
3723 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3724 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3725 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3726 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3730 *first_xattr_slot = -1;
3731 while (slot < nritems) {
3732 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3734 /* we found a different objectid, there must not be acls */
3735 if (found_key.objectid != objectid)
3738 /* we found an xattr, assume we've got an acl */
3739 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3740 if (*first_xattr_slot == -1)
3741 *first_xattr_slot = slot;
3742 if (found_key.offset == xattr_access ||
3743 found_key.offset == xattr_default)
3748 * we found a key greater than an xattr key, there can't
3749 * be any acls later on
3751 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3758 * it goes inode, inode backrefs, xattrs, extents,
3759 * so if there are a ton of hard links to an inode there can
3760 * be a lot of backrefs. Don't waste time searching too hard,
3761 * this is just an optimization
3766 /* we hit the end of the leaf before we found an xattr or
3767 * something larger than an xattr. We have to assume the inode
3770 if (*first_xattr_slot == -1)
3771 *first_xattr_slot = slot;
3776 * read an inode from the btree into the in-memory inode
3778 static int btrfs_read_locked_inode(struct inode *inode)
3780 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3781 struct btrfs_path *path;
3782 struct extent_buffer *leaf;
3783 struct btrfs_inode_item *inode_item;
3784 struct btrfs_root *root = BTRFS_I(inode)->root;
3785 struct btrfs_key location;
3790 bool filled = false;
3791 int first_xattr_slot;
3793 ret = btrfs_fill_inode(inode, &rdev);
3797 path = btrfs_alloc_path();
3803 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3805 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3812 leaf = path->nodes[0];
3817 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3818 struct btrfs_inode_item);
3819 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3820 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3821 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3822 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3823 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3825 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3826 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3828 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3829 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3831 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3832 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3834 BTRFS_I(inode)->i_otime.tv_sec =
3835 btrfs_timespec_sec(leaf, &inode_item->otime);
3836 BTRFS_I(inode)->i_otime.tv_nsec =
3837 btrfs_timespec_nsec(leaf, &inode_item->otime);
3839 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3840 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3841 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3843 inode_set_iversion_queried(inode,
3844 btrfs_inode_sequence(leaf, inode_item));
3845 inode->i_generation = BTRFS_I(inode)->generation;
3847 rdev = btrfs_inode_rdev(leaf, inode_item);
3849 BTRFS_I(inode)->index_cnt = (u64)-1;
3850 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3854 * If we were modified in the current generation and evicted from memory
3855 * and then re-read we need to do a full sync since we don't have any
3856 * idea about which extents were modified before we were evicted from
3859 * This is required for both inode re-read from disk and delayed inode
3860 * in delayed_nodes_tree.
3862 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3863 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3864 &BTRFS_I(inode)->runtime_flags);
3867 * We don't persist the id of the transaction where an unlink operation
3868 * against the inode was last made. So here we assume the inode might
3869 * have been evicted, and therefore the exact value of last_unlink_trans
3870 * lost, and set it to last_trans to avoid metadata inconsistencies
3871 * between the inode and its parent if the inode is fsync'ed and the log
3872 * replayed. For example, in the scenario:
3875 * ln mydir/foo mydir/bar
3878 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3879 * xfs_io -c fsync mydir/foo
3881 * mount fs, triggers fsync log replay
3883 * We must make sure that when we fsync our inode foo we also log its
3884 * parent inode, otherwise after log replay the parent still has the
3885 * dentry with the "bar" name but our inode foo has a link count of 1
3886 * and doesn't have an inode ref with the name "bar" anymore.
3888 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3889 * but it guarantees correctness at the expense of occasional full
3890 * transaction commits on fsync if our inode is a directory, or if our
3891 * inode is not a directory, logging its parent unnecessarily.
3893 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3896 if (inode->i_nlink != 1 ||
3897 path->slots[0] >= btrfs_header_nritems(leaf))
3900 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3901 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3904 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3905 if (location.type == BTRFS_INODE_REF_KEY) {
3906 struct btrfs_inode_ref *ref;
3908 ref = (struct btrfs_inode_ref *)ptr;
3909 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3910 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3911 struct btrfs_inode_extref *extref;
3913 extref = (struct btrfs_inode_extref *)ptr;
3914 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3919 * try to precache a NULL acl entry for files that don't have
3920 * any xattrs or acls
3922 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3923 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3924 if (first_xattr_slot != -1) {
3925 path->slots[0] = first_xattr_slot;
3926 ret = btrfs_load_inode_props(inode, path);
3929 "error loading props for ino %llu (root %llu): %d",
3930 btrfs_ino(BTRFS_I(inode)),
3931 root->root_key.objectid, ret);
3933 btrfs_free_path(path);
3936 cache_no_acl(inode);
3938 switch (inode->i_mode & S_IFMT) {
3940 inode->i_mapping->a_ops = &btrfs_aops;
3941 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3942 inode->i_fop = &btrfs_file_operations;
3943 inode->i_op = &btrfs_file_inode_operations;
3946 inode->i_fop = &btrfs_dir_file_operations;
3947 inode->i_op = &btrfs_dir_inode_operations;
3950 inode->i_op = &btrfs_symlink_inode_operations;
3951 inode_nohighmem(inode);
3952 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3955 inode->i_op = &btrfs_special_inode_operations;
3956 init_special_inode(inode, inode->i_mode, rdev);
3960 btrfs_update_iflags(inode);
3964 btrfs_free_path(path);
3965 make_bad_inode(inode);
3970 * given a leaf and an inode, copy the inode fields into the leaf
3972 static void fill_inode_item(struct btrfs_trans_handle *trans,
3973 struct extent_buffer *leaf,
3974 struct btrfs_inode_item *item,
3975 struct inode *inode)
3977 struct btrfs_map_token token;
3979 btrfs_init_map_token(&token);
3981 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3982 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3983 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3985 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3986 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3988 btrfs_set_token_timespec_sec(leaf, &item->atime,
3989 inode->i_atime.tv_sec, &token);
3990 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3991 inode->i_atime.tv_nsec, &token);
3993 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3994 inode->i_mtime.tv_sec, &token);
3995 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3996 inode->i_mtime.tv_nsec, &token);
3998 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3999 inode->i_ctime.tv_sec, &token);
4000 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
4001 inode->i_ctime.tv_nsec, &token);
4003 btrfs_set_token_timespec_sec(leaf, &item->otime,
4004 BTRFS_I(inode)->i_otime.tv_sec, &token);
4005 btrfs_set_token_timespec_nsec(leaf, &item->otime,
4006 BTRFS_I(inode)->i_otime.tv_nsec, &token);
4008 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
4010 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
4012 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
4014 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
4015 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
4016 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
4017 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
4021 * copy everything in the in-memory inode into the btree.
4023 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4024 struct btrfs_root *root, struct inode *inode)
4026 struct btrfs_inode_item *inode_item;
4027 struct btrfs_path *path;
4028 struct extent_buffer *leaf;
4031 path = btrfs_alloc_path();
4035 path->leave_spinning = 1;
4036 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
4044 leaf = path->nodes[0];
4045 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4046 struct btrfs_inode_item);
4048 fill_inode_item(trans, leaf, inode_item, inode);
4049 btrfs_mark_buffer_dirty(leaf);
4050 btrfs_set_inode_last_trans(trans, inode);
4053 btrfs_free_path(path);
4058 * copy everything in the in-memory inode into the btree.
4060 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4061 struct btrfs_root *root, struct inode *inode)
4063 struct btrfs_fs_info *fs_info = root->fs_info;
4067 * If the inode is a free space inode, we can deadlock during commit
4068 * if we put it into the delayed code.
4070 * The data relocation inode should also be directly updated
4073 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4074 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4075 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4076 btrfs_update_root_times(trans, root);
4078 ret = btrfs_delayed_update_inode(trans, root, inode);
4080 btrfs_set_inode_last_trans(trans, inode);
4084 return btrfs_update_inode_item(trans, root, inode);
4087 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4088 struct btrfs_root *root,
4089 struct inode *inode)
4093 ret = btrfs_update_inode(trans, root, inode);
4095 return btrfs_update_inode_item(trans, root, inode);
4100 * unlink helper that gets used here in inode.c and in the tree logging
4101 * recovery code. It remove a link in a directory with a given name, and
4102 * also drops the back refs in the inode to the directory
4104 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4105 struct btrfs_root *root,
4106 struct btrfs_inode *dir,
4107 struct btrfs_inode *inode,
4108 const char *name, int name_len)
4110 struct btrfs_fs_info *fs_info = root->fs_info;
4111 struct btrfs_path *path;
4113 struct extent_buffer *leaf;
4114 struct btrfs_dir_item *di;
4115 struct btrfs_key key;
4117 u64 ino = btrfs_ino(inode);
4118 u64 dir_ino = btrfs_ino(dir);
4120 path = btrfs_alloc_path();
4126 path->leave_spinning = 1;
4127 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4128 name, name_len, -1);
4137 leaf = path->nodes[0];
4138 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4139 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4142 btrfs_release_path(path);
4145 * If we don't have dir index, we have to get it by looking up
4146 * the inode ref, since we get the inode ref, remove it directly,
4147 * it is unnecessary to do delayed deletion.
4149 * But if we have dir index, needn't search inode ref to get it.
4150 * Since the inode ref is close to the inode item, it is better
4151 * that we delay to delete it, and just do this deletion when
4152 * we update the inode item.
4154 if (inode->dir_index) {
4155 ret = btrfs_delayed_delete_inode_ref(inode);
4157 index = inode->dir_index;
4162 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4166 "failed to delete reference to %.*s, inode %llu parent %llu",
4167 name_len, name, ino, dir_ino);
4168 btrfs_abort_transaction(trans, ret);
4172 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4174 btrfs_abort_transaction(trans, ret);
4178 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4180 if (ret != 0 && ret != -ENOENT) {
4181 btrfs_abort_transaction(trans, ret);
4185 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4190 btrfs_abort_transaction(trans, ret);
4192 btrfs_free_path(path);
4196 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4197 inode_inc_iversion(&inode->vfs_inode);
4198 inode_inc_iversion(&dir->vfs_inode);
4199 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4200 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4201 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4206 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4207 struct btrfs_root *root,
4208 struct btrfs_inode *dir, struct btrfs_inode *inode,
4209 const char *name, int name_len)
4212 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4214 drop_nlink(&inode->vfs_inode);
4215 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4221 * helper to start transaction for unlink and rmdir.
4223 * unlink and rmdir are special in btrfs, they do not always free space, so
4224 * if we cannot make our reservations the normal way try and see if there is
4225 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4226 * allow the unlink to occur.
4228 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4230 struct btrfs_root *root = BTRFS_I(dir)->root;
4233 * 1 for the possible orphan item
4234 * 1 for the dir item
4235 * 1 for the dir index
4236 * 1 for the inode ref
4239 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4242 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4244 struct btrfs_root *root = BTRFS_I(dir)->root;
4245 struct btrfs_trans_handle *trans;
4246 struct inode *inode = d_inode(dentry);
4249 trans = __unlink_start_trans(dir);
4251 return PTR_ERR(trans);
4253 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4256 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4257 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4258 dentry->d_name.len);
4262 if (inode->i_nlink == 0) {
4263 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4269 btrfs_end_transaction(trans);
4270 btrfs_btree_balance_dirty(root->fs_info);
4274 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4275 struct btrfs_root *root,
4276 struct inode *dir, u64 objectid,
4277 const char *name, int name_len)
4279 struct btrfs_fs_info *fs_info = root->fs_info;
4280 struct btrfs_path *path;
4281 struct extent_buffer *leaf;
4282 struct btrfs_dir_item *di;
4283 struct btrfs_key key;
4286 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4288 path = btrfs_alloc_path();
4292 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4293 name, name_len, -1);
4294 if (IS_ERR_OR_NULL(di)) {
4302 leaf = path->nodes[0];
4303 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4304 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4305 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4307 btrfs_abort_transaction(trans, ret);
4310 btrfs_release_path(path);
4312 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4313 root->root_key.objectid, dir_ino,
4314 &index, name, name_len);
4316 if (ret != -ENOENT) {
4317 btrfs_abort_transaction(trans, ret);
4320 di = btrfs_search_dir_index_item(root, path, dir_ino,
4322 if (IS_ERR_OR_NULL(di)) {
4327 btrfs_abort_transaction(trans, ret);
4331 leaf = path->nodes[0];
4332 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4333 btrfs_release_path(path);
4336 btrfs_release_path(path);
4338 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4340 btrfs_abort_transaction(trans, ret);
4344 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4345 inode_inc_iversion(dir);
4346 dir->i_mtime = dir->i_ctime = current_time(dir);
4347 ret = btrfs_update_inode_fallback(trans, root, dir);
4349 btrfs_abort_transaction(trans, ret);
4351 btrfs_free_path(path);
4355 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4357 struct inode *inode = d_inode(dentry);
4359 struct btrfs_root *root = BTRFS_I(dir)->root;
4360 struct btrfs_trans_handle *trans;
4361 u64 last_unlink_trans;
4363 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4365 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4368 trans = __unlink_start_trans(dir);
4370 return PTR_ERR(trans);
4372 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4373 err = btrfs_unlink_subvol(trans, root, dir,
4374 BTRFS_I(inode)->location.objectid,
4375 dentry->d_name.name,
4376 dentry->d_name.len);
4380 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4384 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4386 /* now the directory is empty */
4387 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4388 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4389 dentry->d_name.len);
4391 btrfs_i_size_write(BTRFS_I(inode), 0);
4393 * Propagate the last_unlink_trans value of the deleted dir to
4394 * its parent directory. This is to prevent an unrecoverable
4395 * log tree in the case we do something like this:
4397 * 2) create snapshot under dir foo
4398 * 3) delete the snapshot
4401 * 6) fsync foo or some file inside foo
4403 if (last_unlink_trans >= trans->transid)
4404 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4407 btrfs_end_transaction(trans);
4408 btrfs_btree_balance_dirty(root->fs_info);
4413 static int truncate_space_check(struct btrfs_trans_handle *trans,
4414 struct btrfs_root *root,
4417 struct btrfs_fs_info *fs_info = root->fs_info;
4421 * This is only used to apply pressure to the enospc system, we don't
4422 * intend to use this reservation at all.
4424 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4425 bytes_deleted *= fs_info->nodesize;
4426 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4427 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4429 trace_btrfs_space_reservation(fs_info, "transaction",
4432 trans->bytes_reserved += bytes_deleted;
4439 * Return this if we need to call truncate_block for the last bit of the
4442 #define NEED_TRUNCATE_BLOCK 1
4445 * this can truncate away extent items, csum items and directory items.
4446 * It starts at a high offset and removes keys until it can't find
4447 * any higher than new_size
4449 * csum items that cross the new i_size are truncated to the new size
4452 * min_type is the minimum key type to truncate down to. If set to 0, this
4453 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4455 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4456 struct btrfs_root *root,
4457 struct inode *inode,
4458 u64 new_size, u32 min_type)
4460 struct btrfs_fs_info *fs_info = root->fs_info;
4461 struct btrfs_path *path;
4462 struct extent_buffer *leaf;
4463 struct btrfs_file_extent_item *fi;
4464 struct btrfs_key key;
4465 struct btrfs_key found_key;
4466 u64 extent_start = 0;
4467 u64 extent_num_bytes = 0;
4468 u64 extent_offset = 0;
4470 u64 last_size = new_size;
4471 u32 found_type = (u8)-1;
4474 int pending_del_nr = 0;
4475 int pending_del_slot = 0;
4476 int extent_type = -1;
4479 u64 ino = btrfs_ino(BTRFS_I(inode));
4480 u64 bytes_deleted = 0;
4481 bool be_nice = false;
4482 bool should_throttle = false;
4483 bool should_end = false;
4485 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4488 * for non-free space inodes and ref cows, we want to back off from
4491 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4492 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4495 path = btrfs_alloc_path();
4498 path->reada = READA_BACK;
4501 * We want to drop from the next block forward in case this new size is
4502 * not block aligned since we will be keeping the last block of the
4503 * extent just the way it is.
4505 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4506 root == fs_info->tree_root)
4507 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4508 fs_info->sectorsize),
4512 * This function is also used to drop the items in the log tree before
4513 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4514 * it is used to drop the loged items. So we shouldn't kill the delayed
4517 if (min_type == 0 && root == BTRFS_I(inode)->root)
4518 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4521 key.offset = (u64)-1;
4526 * with a 16K leaf size and 128MB extents, you can actually queue
4527 * up a huge file in a single leaf. Most of the time that
4528 * bytes_deleted is > 0, it will be huge by the time we get here
4530 if (be_nice && bytes_deleted > SZ_32M) {
4531 if (btrfs_should_end_transaction(trans)) {
4538 path->leave_spinning = 1;
4539 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4546 /* there are no items in the tree for us to truncate, we're
4549 if (path->slots[0] == 0)
4556 leaf = path->nodes[0];
4557 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4558 found_type = found_key.type;
4560 if (found_key.objectid != ino)
4563 if (found_type < min_type)
4566 item_end = found_key.offset;
4567 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4568 fi = btrfs_item_ptr(leaf, path->slots[0],
4569 struct btrfs_file_extent_item);
4570 extent_type = btrfs_file_extent_type(leaf, fi);
4571 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4573 btrfs_file_extent_num_bytes(leaf, fi);
4575 trace_btrfs_truncate_show_fi_regular(
4576 BTRFS_I(inode), leaf, fi,
4578 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4579 item_end += btrfs_file_extent_inline_len(leaf,
4580 path->slots[0], fi);
4582 trace_btrfs_truncate_show_fi_inline(
4583 BTRFS_I(inode), leaf, fi, path->slots[0],
4588 if (found_type > min_type) {
4591 if (item_end < new_size)
4593 if (found_key.offset >= new_size)
4599 /* FIXME, shrink the extent if the ref count is only 1 */
4600 if (found_type != BTRFS_EXTENT_DATA_KEY)
4603 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4605 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4607 u64 orig_num_bytes =
4608 btrfs_file_extent_num_bytes(leaf, fi);
4609 extent_num_bytes = ALIGN(new_size -
4611 fs_info->sectorsize);
4612 btrfs_set_file_extent_num_bytes(leaf, fi,
4614 num_dec = (orig_num_bytes -
4616 if (test_bit(BTRFS_ROOT_REF_COWS,
4619 inode_sub_bytes(inode, num_dec);
4620 btrfs_mark_buffer_dirty(leaf);
4623 btrfs_file_extent_disk_num_bytes(leaf,
4625 extent_offset = found_key.offset -
4626 btrfs_file_extent_offset(leaf, fi);
4628 /* FIXME blocksize != 4096 */
4629 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4630 if (extent_start != 0) {
4632 if (test_bit(BTRFS_ROOT_REF_COWS,
4634 inode_sub_bytes(inode, num_dec);
4637 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4639 * we can't truncate inline items that have had
4643 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4644 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4645 btrfs_file_extent_compression(leaf, fi) == 0) {
4646 u32 size = (u32)(new_size - found_key.offset);
4648 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4649 size = btrfs_file_extent_calc_inline_size(size);
4650 btrfs_truncate_item(root->fs_info, path, size, 1);
4651 } else if (!del_item) {
4653 * We have to bail so the last_size is set to
4654 * just before this extent.
4656 err = NEED_TRUNCATE_BLOCK;
4660 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4661 inode_sub_bytes(inode, item_end + 1 - new_size);
4665 last_size = found_key.offset;
4667 last_size = new_size;
4669 if (!pending_del_nr) {
4670 /* no pending yet, add ourselves */
4671 pending_del_slot = path->slots[0];
4673 } else if (pending_del_nr &&
4674 path->slots[0] + 1 == pending_del_slot) {
4675 /* hop on the pending chunk */
4677 pending_del_slot = path->slots[0];
4684 should_throttle = false;
4687 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4688 root == fs_info->tree_root)) {
4689 btrfs_set_path_blocking(path);
4690 bytes_deleted += extent_num_bytes;
4691 ret = btrfs_free_extent(trans, root, extent_start,
4692 extent_num_bytes, 0,
4693 btrfs_header_owner(leaf),
4694 ino, extent_offset);
4696 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4697 btrfs_async_run_delayed_refs(fs_info,
4698 trans->delayed_ref_updates * 2,
4701 if (truncate_space_check(trans, root,
4702 extent_num_bytes)) {
4705 if (btrfs_should_throttle_delayed_refs(trans,
4707 should_throttle = true;
4711 if (found_type == BTRFS_INODE_ITEM_KEY)
4714 if (path->slots[0] == 0 ||
4715 path->slots[0] != pending_del_slot ||
4716 should_throttle || should_end) {
4717 if (pending_del_nr) {
4718 ret = btrfs_del_items(trans, root, path,
4722 btrfs_abort_transaction(trans, ret);
4727 btrfs_release_path(path);
4728 if (should_throttle) {
4729 unsigned long updates = trans->delayed_ref_updates;
4731 trans->delayed_ref_updates = 0;
4732 ret = btrfs_run_delayed_refs(trans,
4739 * if we failed to refill our space rsv, bail out
4740 * and let the transaction restart
4752 if (pending_del_nr) {
4753 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4756 btrfs_abort_transaction(trans, ret);
4759 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4760 ASSERT(last_size >= new_size);
4761 if (!err && last_size > new_size)
4762 last_size = new_size;
4763 btrfs_ordered_update_i_size(inode, last_size, NULL);
4766 btrfs_free_path(path);
4768 if (be_nice && bytes_deleted > SZ_32M) {
4769 unsigned long updates = trans->delayed_ref_updates;
4771 trans->delayed_ref_updates = 0;
4772 ret = btrfs_run_delayed_refs(trans, updates * 2);
4781 * btrfs_truncate_block - read, zero a chunk and write a block
4782 * @inode - inode that we're zeroing
4783 * @from - the offset to start zeroing
4784 * @len - the length to zero, 0 to zero the entire range respective to the
4786 * @front - zero up to the offset instead of from the offset on
4788 * This will find the block for the "from" offset and cow the block and zero the
4789 * part we want to zero. This is used with truncate and hole punching.
4791 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4794 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4795 struct address_space *mapping = inode->i_mapping;
4796 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4797 struct btrfs_ordered_extent *ordered;
4798 struct extent_state *cached_state = NULL;
4799 struct extent_changeset *data_reserved = NULL;
4801 u32 blocksize = fs_info->sectorsize;
4802 pgoff_t index = from >> PAGE_SHIFT;
4803 unsigned offset = from & (blocksize - 1);
4805 gfp_t mask = btrfs_alloc_write_mask(mapping);
4810 if (IS_ALIGNED(offset, blocksize) &&
4811 (!len || IS_ALIGNED(len, blocksize)))
4814 block_start = round_down(from, blocksize);
4815 block_end = block_start + blocksize - 1;
4817 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4818 block_start, blocksize);
4823 page = find_or_create_page(mapping, index, mask);
4825 btrfs_delalloc_release_space(inode, data_reserved,
4826 block_start, blocksize);
4827 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4832 if (!PageUptodate(page)) {
4833 ret = btrfs_readpage(NULL, page);
4835 if (page->mapping != mapping) {
4840 if (!PageUptodate(page)) {
4845 wait_on_page_writeback(page);
4847 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4848 set_page_extent_mapped(page);
4850 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4852 unlock_extent_cached(io_tree, block_start, block_end,
4856 btrfs_start_ordered_extent(inode, ordered, 1);
4857 btrfs_put_ordered_extent(ordered);
4861 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4862 EXTENT_DIRTY | EXTENT_DELALLOC |
4863 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4864 0, 0, &cached_state);
4866 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4869 unlock_extent_cached(io_tree, block_start, block_end,
4874 if (offset != blocksize) {
4876 len = blocksize - offset;
4879 memset(kaddr + (block_start - page_offset(page)),
4882 memset(kaddr + (block_start - page_offset(page)) + offset,
4884 flush_dcache_page(page);
4887 ClearPageChecked(page);
4888 set_page_dirty(page);
4889 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4893 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4895 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4899 extent_changeset_free(data_reserved);
4903 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4904 u64 offset, u64 len)
4906 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4907 struct btrfs_trans_handle *trans;
4911 * Still need to make sure the inode looks like it's been updated so
4912 * that any holes get logged if we fsync.
4914 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4915 BTRFS_I(inode)->last_trans = fs_info->generation;
4916 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4917 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4922 * 1 - for the one we're dropping
4923 * 1 - for the one we're adding
4924 * 1 - for updating the inode.
4926 trans = btrfs_start_transaction(root, 3);
4928 return PTR_ERR(trans);
4930 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4932 btrfs_abort_transaction(trans, ret);
4933 btrfs_end_transaction(trans);
4937 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4938 offset, 0, 0, len, 0, len, 0, 0, 0);
4940 btrfs_abort_transaction(trans, ret);
4942 btrfs_update_inode(trans, root, inode);
4943 btrfs_end_transaction(trans);
4948 * This function puts in dummy file extents for the area we're creating a hole
4949 * for. So if we are truncating this file to a larger size we need to insert
4950 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4951 * the range between oldsize and size
4953 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4955 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4956 struct btrfs_root *root = BTRFS_I(inode)->root;
4957 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4958 struct extent_map *em = NULL;
4959 struct extent_state *cached_state = NULL;
4960 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4961 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4962 u64 block_end = ALIGN(size, fs_info->sectorsize);
4969 * If our size started in the middle of a block we need to zero out the
4970 * rest of the block before we expand the i_size, otherwise we could
4971 * expose stale data.
4973 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4977 if (size <= hole_start)
4981 struct btrfs_ordered_extent *ordered;
4983 lock_extent_bits(io_tree, hole_start, block_end - 1,
4985 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4986 block_end - hole_start);
4989 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4991 btrfs_start_ordered_extent(inode, ordered, 1);
4992 btrfs_put_ordered_extent(ordered);
4995 cur_offset = hole_start;
4997 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4998 block_end - cur_offset, 0);
5004 last_byte = min(extent_map_end(em), block_end);
5005 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5006 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5007 struct extent_map *hole_em;
5008 hole_size = last_byte - cur_offset;
5010 err = maybe_insert_hole(root, inode, cur_offset,
5014 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5015 cur_offset + hole_size - 1, 0);
5016 hole_em = alloc_extent_map();
5018 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5019 &BTRFS_I(inode)->runtime_flags);
5022 hole_em->start = cur_offset;
5023 hole_em->len = hole_size;
5024 hole_em->orig_start = cur_offset;
5026 hole_em->block_start = EXTENT_MAP_HOLE;
5027 hole_em->block_len = 0;
5028 hole_em->orig_block_len = 0;
5029 hole_em->ram_bytes = hole_size;
5030 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5031 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5032 hole_em->generation = fs_info->generation;
5035 write_lock(&em_tree->lock);
5036 err = add_extent_mapping(em_tree, hole_em, 1);
5037 write_unlock(&em_tree->lock);
5040 btrfs_drop_extent_cache(BTRFS_I(inode),
5045 free_extent_map(hole_em);
5048 free_extent_map(em);
5050 cur_offset = last_byte;
5051 if (cur_offset >= block_end)
5054 free_extent_map(em);
5055 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5059 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5061 struct btrfs_root *root = BTRFS_I(inode)->root;
5062 struct btrfs_trans_handle *trans;
5063 loff_t oldsize = i_size_read(inode);
5064 loff_t newsize = attr->ia_size;
5065 int mask = attr->ia_valid;
5069 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5070 * special case where we need to update the times despite not having
5071 * these flags set. For all other operations the VFS set these flags
5072 * explicitly if it wants a timestamp update.
5074 if (newsize != oldsize) {
5075 inode_inc_iversion(inode);
5076 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5077 inode->i_ctime = inode->i_mtime =
5078 current_time(inode);
5081 if (newsize > oldsize) {
5083 * Don't do an expanding truncate while snapshotting is ongoing.
5084 * This is to ensure the snapshot captures a fully consistent
5085 * state of this file - if the snapshot captures this expanding
5086 * truncation, it must capture all writes that happened before
5089 btrfs_wait_for_snapshot_creation(root);
5090 ret = btrfs_cont_expand(inode, oldsize, newsize);
5092 btrfs_end_write_no_snapshotting(root);
5096 trans = btrfs_start_transaction(root, 1);
5097 if (IS_ERR(trans)) {
5098 btrfs_end_write_no_snapshotting(root);
5099 return PTR_ERR(trans);
5102 i_size_write(inode, newsize);
5103 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5104 pagecache_isize_extended(inode, oldsize, newsize);
5105 ret = btrfs_update_inode(trans, root, inode);
5106 btrfs_end_write_no_snapshotting(root);
5107 btrfs_end_transaction(trans);
5111 * We're truncating a file that used to have good data down to
5112 * zero. Make sure it gets into the ordered flush list so that
5113 * any new writes get down to disk quickly.
5116 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5117 &BTRFS_I(inode)->runtime_flags);
5120 * 1 for the orphan item we're going to add
5121 * 1 for the orphan item deletion.
5123 trans = btrfs_start_transaction(root, 2);
5125 return PTR_ERR(trans);
5128 * We need to do this in case we fail at _any_ point during the
5129 * actual truncate. Once we do the truncate_setsize we could
5130 * invalidate pages which forces any outstanding ordered io to
5131 * be instantly completed which will give us extents that need
5132 * to be truncated. If we fail to get an orphan inode down we
5133 * could have left over extents that were never meant to live,
5134 * so we need to guarantee from this point on that everything
5135 * will be consistent.
5137 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5138 btrfs_end_transaction(trans);
5142 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5143 truncate_setsize(inode, newsize);
5145 /* Disable nonlocked read DIO to avoid the end less truncate */
5146 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5147 inode_dio_wait(inode);
5148 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5150 ret = btrfs_truncate(inode, newsize == oldsize);
5151 if (ret && inode->i_nlink) {
5154 /* To get a stable disk_i_size */
5155 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5157 btrfs_orphan_del(NULL, BTRFS_I(inode));
5162 * failed to truncate, disk_i_size is only adjusted down
5163 * as we remove extents, so it should represent the true
5164 * size of the inode, so reset the in memory size and
5165 * delete our orphan entry.
5167 trans = btrfs_join_transaction(root);
5168 if (IS_ERR(trans)) {
5169 btrfs_orphan_del(NULL, BTRFS_I(inode));
5172 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5173 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5175 btrfs_abort_transaction(trans, err);
5176 btrfs_end_transaction(trans);
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)) {
5237 struct extent_map *em;
5239 node = rb_first(&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;
5276 node = rb_first(&io_tree->state);
5277 state = rb_entry(node, struct extent_state, rb_node);
5278 start = state->start;
5280 spin_unlock(&io_tree->lock);
5282 lock_extent_bits(io_tree, start, end, &cached_state);
5285 * If still has DELALLOC flag, the extent didn't reach disk,
5286 * and its reserved space won't be freed by delayed_ref.
5287 * So we need to free its reserved space here.
5288 * (Refer to comment in btrfs_invalidatepage, case 2)
5290 * Note, end is the bytenr of last byte, so we need + 1 here.
5292 if (state->state & EXTENT_DELALLOC)
5293 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5295 clear_extent_bit(io_tree, start, end,
5296 EXTENT_LOCKED | EXTENT_DIRTY |
5297 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5298 EXTENT_DEFRAG, 1, 1, &cached_state);
5301 spin_lock(&io_tree->lock);
5303 spin_unlock(&io_tree->lock);
5306 void btrfs_evict_inode(struct inode *inode)
5308 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5309 struct btrfs_trans_handle *trans;
5310 struct btrfs_root *root = BTRFS_I(inode)->root;
5311 struct btrfs_block_rsv *rsv, *global_rsv;
5312 int steal_from_global = 0;
5316 trace_btrfs_inode_evict(inode);
5323 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5325 evict_inode_truncate_pages(inode);
5327 if (inode->i_nlink &&
5328 ((btrfs_root_refs(&root->root_item) != 0 &&
5329 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5330 btrfs_is_free_space_inode(BTRFS_I(inode))))
5333 if (is_bad_inode(inode)) {
5334 btrfs_orphan_del(NULL, BTRFS_I(inode));
5337 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5338 if (!special_file(inode->i_mode))
5339 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5341 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5343 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5344 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5345 &BTRFS_I(inode)->runtime_flags));
5349 if (inode->i_nlink > 0) {
5350 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5351 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5355 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5357 btrfs_orphan_del(NULL, BTRFS_I(inode));
5361 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5363 btrfs_orphan_del(NULL, BTRFS_I(inode));
5366 rsv->size = min_size;
5368 global_rsv = &fs_info->global_block_rsv;
5370 btrfs_i_size_write(BTRFS_I(inode), 0);
5373 * This is a bit simpler than btrfs_truncate since we've already
5374 * reserved our space for our orphan item in the unlink, so we just
5375 * need to reserve some slack space in case we add bytes and update
5376 * inode item when doing the truncate.
5379 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5380 BTRFS_RESERVE_FLUSH_LIMIT);
5383 * Try and steal from the global reserve since we will
5384 * likely not use this space anyway, we want to try as
5385 * hard as possible to get this to work.
5388 steal_from_global++;
5390 steal_from_global = 0;
5394 * steal_from_global == 0: we reserved stuff, hooray!
5395 * steal_from_global == 1: we didn't reserve stuff, boo!
5396 * steal_from_global == 2: we've committed, still not a lot of
5397 * room but maybe we'll have room in the global reserve this
5399 * steal_from_global == 3: abandon all hope!
5401 if (steal_from_global > 2) {
5403 "Could not get space for a delete, will truncate on mount %d",
5405 btrfs_orphan_del(NULL, BTRFS_I(inode));
5406 btrfs_free_block_rsv(fs_info, rsv);
5410 trans = btrfs_join_transaction(root);
5411 if (IS_ERR(trans)) {
5412 btrfs_orphan_del(NULL, BTRFS_I(inode));
5413 btrfs_free_block_rsv(fs_info, rsv);
5418 * We can't just steal from the global reserve, we need to make
5419 * sure there is room to do it, if not we need to commit and try
5422 if (steal_from_global) {
5423 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5424 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5431 * Couldn't steal from the global reserve, we have too much
5432 * pending stuff built up, commit the transaction and try it
5436 ret = btrfs_commit_transaction(trans);
5438 btrfs_orphan_del(NULL, BTRFS_I(inode));
5439 btrfs_free_block_rsv(fs_info, rsv);
5444 steal_from_global = 0;
5447 trans->block_rsv = rsv;
5449 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5450 if (ret != -ENOSPC && ret != -EAGAIN)
5453 trans->block_rsv = &fs_info->trans_block_rsv;
5454 btrfs_end_transaction(trans);
5456 btrfs_btree_balance_dirty(fs_info);
5459 btrfs_free_block_rsv(fs_info, rsv);
5462 * Errors here aren't a big deal, it just means we leave orphan items
5463 * in the tree. They will be cleaned up on the next mount.
5466 trans->block_rsv = root->orphan_block_rsv;
5467 btrfs_orphan_del(trans, BTRFS_I(inode));
5469 btrfs_orphan_del(NULL, BTRFS_I(inode));
5472 trans->block_rsv = &fs_info->trans_block_rsv;
5473 if (!(root == fs_info->tree_root ||
5474 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5475 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5477 btrfs_end_transaction(trans);
5478 btrfs_btree_balance_dirty(fs_info);
5480 btrfs_remove_delayed_node(BTRFS_I(inode));
5485 * this returns the key found in the dir entry in the location pointer.
5486 * If no dir entries were found, returns -ENOENT.
5487 * If found a corrupted location in dir entry, returns -EUCLEAN.
5489 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5490 struct btrfs_key *location)
5492 const char *name = dentry->d_name.name;
5493 int namelen = dentry->d_name.len;
5494 struct btrfs_dir_item *di;
5495 struct btrfs_path *path;
5496 struct btrfs_root *root = BTRFS_I(dir)->root;
5499 path = btrfs_alloc_path();
5503 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5514 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5515 if (location->type != BTRFS_INODE_ITEM_KEY &&
5516 location->type != BTRFS_ROOT_ITEM_KEY) {
5518 btrfs_warn(root->fs_info,
5519 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5520 __func__, name, btrfs_ino(BTRFS_I(dir)),
5521 location->objectid, location->type, location->offset);
5524 btrfs_free_path(path);
5529 * when we hit a tree root in a directory, the btrfs part of the inode
5530 * needs to be changed to reflect the root directory of the tree root. This
5531 * is kind of like crossing a mount point.
5533 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5535 struct dentry *dentry,
5536 struct btrfs_key *location,
5537 struct btrfs_root **sub_root)
5539 struct btrfs_path *path;
5540 struct btrfs_root *new_root;
5541 struct btrfs_root_ref *ref;
5542 struct extent_buffer *leaf;
5543 struct btrfs_key key;
5547 path = btrfs_alloc_path();
5554 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5555 key.type = BTRFS_ROOT_REF_KEY;
5556 key.offset = location->objectid;
5558 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5565 leaf = path->nodes[0];
5566 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5567 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5568 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5571 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5572 (unsigned long)(ref + 1),
5573 dentry->d_name.len);
5577 btrfs_release_path(path);
5579 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5580 if (IS_ERR(new_root)) {
5581 err = PTR_ERR(new_root);
5585 *sub_root = new_root;
5586 location->objectid = btrfs_root_dirid(&new_root->root_item);
5587 location->type = BTRFS_INODE_ITEM_KEY;
5588 location->offset = 0;
5591 btrfs_free_path(path);
5595 static void inode_tree_add(struct inode *inode)
5597 struct btrfs_root *root = BTRFS_I(inode)->root;
5598 struct btrfs_inode *entry;
5600 struct rb_node *parent;
5601 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5602 u64 ino = btrfs_ino(BTRFS_I(inode));
5604 if (inode_unhashed(inode))
5607 spin_lock(&root->inode_lock);
5608 p = &root->inode_tree.rb_node;
5611 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5613 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5614 p = &parent->rb_left;
5615 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5616 p = &parent->rb_right;
5618 WARN_ON(!(entry->vfs_inode.i_state &
5619 (I_WILL_FREE | I_FREEING)));
5620 rb_replace_node(parent, new, &root->inode_tree);
5621 RB_CLEAR_NODE(parent);
5622 spin_unlock(&root->inode_lock);
5626 rb_link_node(new, parent, p);
5627 rb_insert_color(new, &root->inode_tree);
5628 spin_unlock(&root->inode_lock);
5631 static void inode_tree_del(struct inode *inode)
5633 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5634 struct btrfs_root *root = BTRFS_I(inode)->root;
5637 spin_lock(&root->inode_lock);
5638 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5639 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5640 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5641 empty = RB_EMPTY_ROOT(&root->inode_tree);
5643 spin_unlock(&root->inode_lock);
5645 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5646 synchronize_srcu(&fs_info->subvol_srcu);
5647 spin_lock(&root->inode_lock);
5648 empty = RB_EMPTY_ROOT(&root->inode_tree);
5649 spin_unlock(&root->inode_lock);
5651 btrfs_add_dead_root(root);
5655 void btrfs_invalidate_inodes(struct btrfs_root *root)
5657 struct btrfs_fs_info *fs_info = root->fs_info;
5658 struct rb_node *node;
5659 struct rb_node *prev;
5660 struct btrfs_inode *entry;
5661 struct inode *inode;
5664 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5665 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5667 spin_lock(&root->inode_lock);
5669 node = root->inode_tree.rb_node;
5673 entry = rb_entry(node, struct btrfs_inode, rb_node);
5675 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5676 node = node->rb_left;
5677 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5678 node = node->rb_right;
5684 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5685 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5689 prev = rb_next(prev);
5693 entry = rb_entry(node, struct btrfs_inode, rb_node);
5694 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5695 inode = igrab(&entry->vfs_inode);
5697 spin_unlock(&root->inode_lock);
5698 if (atomic_read(&inode->i_count) > 1)
5699 d_prune_aliases(inode);
5701 * btrfs_drop_inode will have it removed from
5702 * the inode cache when its usage count
5707 spin_lock(&root->inode_lock);
5711 if (cond_resched_lock(&root->inode_lock))
5714 node = rb_next(node);
5716 spin_unlock(&root->inode_lock);
5719 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5721 struct btrfs_iget_args *args = p;
5722 inode->i_ino = args->location->objectid;
5723 memcpy(&BTRFS_I(inode)->location, args->location,
5724 sizeof(*args->location));
5725 BTRFS_I(inode)->root = args->root;
5729 static int btrfs_find_actor(struct inode *inode, void *opaque)
5731 struct btrfs_iget_args *args = opaque;
5732 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5733 args->root == BTRFS_I(inode)->root;
5736 static struct inode *btrfs_iget_locked(struct super_block *s,
5737 struct btrfs_key *location,
5738 struct btrfs_root *root)
5740 struct inode *inode;
5741 struct btrfs_iget_args args;
5742 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5744 args.location = location;
5747 inode = iget5_locked(s, hashval, btrfs_find_actor,
5748 btrfs_init_locked_inode,
5753 /* Get an inode object given its location and corresponding root.
5754 * Returns in *is_new if the inode was read from disk
5756 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5757 struct btrfs_root *root, int *new)
5759 struct inode *inode;
5761 inode = btrfs_iget_locked(s, location, root);
5763 return ERR_PTR(-ENOMEM);
5765 if (inode->i_state & I_NEW) {
5768 ret = btrfs_read_locked_inode(inode);
5769 if (!is_bad_inode(inode)) {
5770 inode_tree_add(inode);
5771 unlock_new_inode(inode);
5775 unlock_new_inode(inode);
5778 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5785 static struct inode *new_simple_dir(struct super_block *s,
5786 struct btrfs_key *key,
5787 struct btrfs_root *root)
5789 struct inode *inode = new_inode(s);
5792 return ERR_PTR(-ENOMEM);
5794 BTRFS_I(inode)->root = root;
5795 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5796 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5798 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5799 inode->i_op = &btrfs_dir_ro_inode_operations;
5800 inode->i_opflags &= ~IOP_XATTR;
5801 inode->i_fop = &simple_dir_operations;
5802 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5803 inode->i_mtime = current_time(inode);
5804 inode->i_atime = inode->i_mtime;
5805 inode->i_ctime = inode->i_mtime;
5806 BTRFS_I(inode)->i_otime = inode->i_mtime;
5811 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5813 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5814 struct inode *inode;
5815 struct btrfs_root *root = BTRFS_I(dir)->root;
5816 struct btrfs_root *sub_root = root;
5817 struct btrfs_key location;
5821 if (dentry->d_name.len > BTRFS_NAME_LEN)
5822 return ERR_PTR(-ENAMETOOLONG);
5824 ret = btrfs_inode_by_name(dir, dentry, &location);
5826 return ERR_PTR(ret);
5828 if (location.type == BTRFS_INODE_ITEM_KEY) {
5829 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5833 index = srcu_read_lock(&fs_info->subvol_srcu);
5834 ret = fixup_tree_root_location(fs_info, dir, dentry,
5835 &location, &sub_root);
5838 inode = ERR_PTR(ret);
5840 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5842 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5844 srcu_read_unlock(&fs_info->subvol_srcu, index);
5846 if (!IS_ERR(inode) && root != sub_root) {
5847 down_read(&fs_info->cleanup_work_sem);
5848 if (!sb_rdonly(inode->i_sb))
5849 ret = btrfs_orphan_cleanup(sub_root);
5850 up_read(&fs_info->cleanup_work_sem);
5853 inode = ERR_PTR(ret);
5860 static int btrfs_dentry_delete(const struct dentry *dentry)
5862 struct btrfs_root *root;
5863 struct inode *inode = d_inode(dentry);
5865 if (!inode && !IS_ROOT(dentry))
5866 inode = d_inode(dentry->d_parent);
5869 root = BTRFS_I(inode)->root;
5870 if (btrfs_root_refs(&root->root_item) == 0)
5873 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5879 static void btrfs_dentry_release(struct dentry *dentry)
5881 kfree(dentry->d_fsdata);
5884 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5887 struct inode *inode;
5889 inode = btrfs_lookup_dentry(dir, dentry);
5890 if (IS_ERR(inode)) {
5891 if (PTR_ERR(inode) == -ENOENT)
5894 return ERR_CAST(inode);
5897 return d_splice_alias(inode, dentry);
5900 unsigned char btrfs_filetype_table[] = {
5901 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5905 * All this infrastructure exists because dir_emit can fault, and we are holding
5906 * the tree lock when doing readdir. For now just allocate a buffer and copy
5907 * our information into that, and then dir_emit from the buffer. This is
5908 * similar to what NFS does, only we don't keep the buffer around in pagecache
5909 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5910 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5913 static int btrfs_opendir(struct inode *inode, struct file *file)
5915 struct btrfs_file_private *private;
5917 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5920 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5921 if (!private->filldir_buf) {
5925 file->private_data = private;
5936 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5939 struct dir_entry *entry = addr;
5940 char *name = (char *)(entry + 1);
5942 ctx->pos = entry->offset;
5943 if (!dir_emit(ctx, name, entry->name_len, entry->ino,
5946 addr += sizeof(struct dir_entry) + entry->name_len;
5952 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5954 struct inode *inode = file_inode(file);
5955 struct btrfs_root *root = BTRFS_I(inode)->root;
5956 struct btrfs_file_private *private = file->private_data;
5957 struct btrfs_dir_item *di;
5958 struct btrfs_key key;
5959 struct btrfs_key found_key;
5960 struct btrfs_path *path;
5962 struct list_head ins_list;
5963 struct list_head del_list;
5965 struct extent_buffer *leaf;
5972 struct btrfs_key location;
5974 if (!dir_emit_dots(file, ctx))
5977 path = btrfs_alloc_path();
5981 addr = private->filldir_buf;
5982 path->reada = READA_FORWARD;
5984 INIT_LIST_HEAD(&ins_list);
5985 INIT_LIST_HEAD(&del_list);
5986 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5989 key.type = BTRFS_DIR_INDEX_KEY;
5990 key.offset = ctx->pos;
5991 key.objectid = btrfs_ino(BTRFS_I(inode));
5993 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5998 struct dir_entry *entry;
6000 leaf = path->nodes[0];
6001 slot = path->slots[0];
6002 if (slot >= btrfs_header_nritems(leaf)) {
6003 ret = btrfs_next_leaf(root, path);
6011 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6013 if (found_key.objectid != key.objectid)
6015 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6017 if (found_key.offset < ctx->pos)
6019 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6021 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6022 name_len = btrfs_dir_name_len(leaf, di);
6023 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6025 btrfs_release_path(path);
6026 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6029 addr = private->filldir_buf;
6036 entry->name_len = name_len;
6037 name_ptr = (char *)(entry + 1);
6038 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6040 entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
6041 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6042 entry->ino = location.objectid;
6043 entry->offset = found_key.offset;
6045 addr += sizeof(struct dir_entry) + name_len;
6046 total_len += sizeof(struct dir_entry) + name_len;
6050 btrfs_release_path(path);
6052 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6056 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6061 * Stop new entries from being returned after we return the last
6064 * New directory entries are assigned a strictly increasing
6065 * offset. This means that new entries created during readdir
6066 * are *guaranteed* to be seen in the future by that readdir.
6067 * This has broken buggy programs which operate on names as
6068 * they're returned by readdir. Until we re-use freed offsets
6069 * we have this hack to stop new entries from being returned
6070 * under the assumption that they'll never reach this huge
6073 * This is being careful not to overflow 32bit loff_t unless the
6074 * last entry requires it because doing so has broken 32bit apps
6077 if (ctx->pos >= INT_MAX)
6078 ctx->pos = LLONG_MAX;
6085 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6086 btrfs_free_path(path);
6090 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6092 struct btrfs_root *root = BTRFS_I(inode)->root;
6093 struct btrfs_trans_handle *trans;
6095 bool nolock = false;
6097 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6100 if (btrfs_fs_closing(root->fs_info) &&
6101 btrfs_is_free_space_inode(BTRFS_I(inode)))
6104 if (wbc->sync_mode == WB_SYNC_ALL) {
6106 trans = btrfs_join_transaction_nolock(root);
6108 trans = btrfs_join_transaction(root);
6110 return PTR_ERR(trans);
6111 ret = btrfs_commit_transaction(trans);
6117 * This is somewhat expensive, updating the tree every time the
6118 * inode changes. But, it is most likely to find the inode in cache.
6119 * FIXME, needs more benchmarking...there are no reasons other than performance
6120 * to keep or drop this code.
6122 static int btrfs_dirty_inode(struct inode *inode)
6124 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6125 struct btrfs_root *root = BTRFS_I(inode)->root;
6126 struct btrfs_trans_handle *trans;
6129 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6132 trans = btrfs_join_transaction(root);
6134 return PTR_ERR(trans);
6136 ret = btrfs_update_inode(trans, root, inode);
6137 if (ret && ret == -ENOSPC) {
6138 /* whoops, lets try again with the full transaction */
6139 btrfs_end_transaction(trans);
6140 trans = btrfs_start_transaction(root, 1);
6142 return PTR_ERR(trans);
6144 ret = btrfs_update_inode(trans, root, inode);
6146 btrfs_end_transaction(trans);
6147 if (BTRFS_I(inode)->delayed_node)
6148 btrfs_balance_delayed_items(fs_info);
6154 * This is a copy of file_update_time. We need this so we can return error on
6155 * ENOSPC for updating the inode in the case of file write and mmap writes.
6157 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6160 struct btrfs_root *root = BTRFS_I(inode)->root;
6161 bool dirty = flags & ~S_VERSION;
6163 if (btrfs_root_readonly(root))
6166 if (flags & S_VERSION)
6167 dirty |= inode_maybe_inc_iversion(inode, dirty);
6168 if (flags & S_CTIME)
6169 inode->i_ctime = *now;
6170 if (flags & S_MTIME)
6171 inode->i_mtime = *now;
6172 if (flags & S_ATIME)
6173 inode->i_atime = *now;
6174 return dirty ? btrfs_dirty_inode(inode) : 0;
6178 * find the highest existing sequence number in a directory
6179 * and then set the in-memory index_cnt variable to reflect
6180 * free sequence numbers
6182 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6184 struct btrfs_root *root = inode->root;
6185 struct btrfs_key key, found_key;
6186 struct btrfs_path *path;
6187 struct extent_buffer *leaf;
6190 key.objectid = btrfs_ino(inode);
6191 key.type = BTRFS_DIR_INDEX_KEY;
6192 key.offset = (u64)-1;
6194 path = btrfs_alloc_path();
6198 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6201 /* FIXME: we should be able to handle this */
6207 * MAGIC NUMBER EXPLANATION:
6208 * since we search a directory based on f_pos we have to start at 2
6209 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6210 * else has to start at 2
6212 if (path->slots[0] == 0) {
6213 inode->index_cnt = 2;
6219 leaf = path->nodes[0];
6220 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6222 if (found_key.objectid != btrfs_ino(inode) ||
6223 found_key.type != BTRFS_DIR_INDEX_KEY) {
6224 inode->index_cnt = 2;
6228 inode->index_cnt = found_key.offset + 1;
6230 btrfs_free_path(path);
6235 * helper to find a free sequence number in a given directory. This current
6236 * code is very simple, later versions will do smarter things in the btree
6238 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6242 if (dir->index_cnt == (u64)-1) {
6243 ret = btrfs_inode_delayed_dir_index_count(dir);
6245 ret = btrfs_set_inode_index_count(dir);
6251 *index = dir->index_cnt;
6257 static int btrfs_insert_inode_locked(struct inode *inode)
6259 struct btrfs_iget_args args;
6260 args.location = &BTRFS_I(inode)->location;
6261 args.root = BTRFS_I(inode)->root;
6263 return insert_inode_locked4(inode,
6264 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6265 btrfs_find_actor, &args);
6269 * Inherit flags from the parent inode.
6271 * Currently only the compression flags and the cow flags are inherited.
6273 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6280 flags = BTRFS_I(dir)->flags;
6282 if (flags & BTRFS_INODE_NOCOMPRESS) {
6283 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6284 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6285 } else if (flags & BTRFS_INODE_COMPRESS) {
6286 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6287 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6290 if (flags & BTRFS_INODE_NODATACOW) {
6291 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6292 if (S_ISREG(inode->i_mode))
6293 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6296 btrfs_update_iflags(inode);
6299 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6300 struct btrfs_root *root,
6302 const char *name, int name_len,
6303 u64 ref_objectid, u64 objectid,
6304 umode_t mode, u64 *index)
6306 struct btrfs_fs_info *fs_info = root->fs_info;
6307 struct inode *inode;
6308 struct btrfs_inode_item *inode_item;
6309 struct btrfs_key *location;
6310 struct btrfs_path *path;
6311 struct btrfs_inode_ref *ref;
6312 struct btrfs_key key[2];
6314 int nitems = name ? 2 : 1;
6318 path = btrfs_alloc_path();
6320 return ERR_PTR(-ENOMEM);
6322 inode = new_inode(fs_info->sb);
6324 btrfs_free_path(path);
6325 return ERR_PTR(-ENOMEM);
6329 * O_TMPFILE, set link count to 0, so that after this point,
6330 * we fill in an inode item with the correct link count.
6333 set_nlink(inode, 0);
6336 * we have to initialize this early, so we can reclaim the inode
6337 * number if we fail afterwards in this function.
6339 inode->i_ino = objectid;
6342 trace_btrfs_inode_request(dir);
6344 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6346 btrfs_free_path(path);
6348 return ERR_PTR(ret);
6354 * index_cnt is ignored for everything but a dir,
6355 * btrfs_set_inode_index_count has an explanation for the magic
6358 BTRFS_I(inode)->index_cnt = 2;
6359 BTRFS_I(inode)->dir_index = *index;
6360 BTRFS_I(inode)->root = root;
6361 BTRFS_I(inode)->generation = trans->transid;
6362 inode->i_generation = BTRFS_I(inode)->generation;
6365 * We could have gotten an inode number from somebody who was fsynced
6366 * and then removed in this same transaction, so let's just set full
6367 * sync since it will be a full sync anyway and this will blow away the
6368 * old info in the log.
6370 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6372 key[0].objectid = objectid;
6373 key[0].type = BTRFS_INODE_ITEM_KEY;
6376 sizes[0] = sizeof(struct btrfs_inode_item);
6380 * Start new inodes with an inode_ref. This is slightly more
6381 * efficient for small numbers of hard links since they will
6382 * be packed into one item. Extended refs will kick in if we
6383 * add more hard links than can fit in the ref item.
6385 key[1].objectid = objectid;
6386 key[1].type = BTRFS_INODE_REF_KEY;
6387 key[1].offset = ref_objectid;
6389 sizes[1] = name_len + sizeof(*ref);
6392 location = &BTRFS_I(inode)->location;
6393 location->objectid = objectid;
6394 location->offset = 0;
6395 location->type = BTRFS_INODE_ITEM_KEY;
6397 ret = btrfs_insert_inode_locked(inode);
6401 path->leave_spinning = 1;
6402 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6406 inode_init_owner(inode, dir, mode);
6407 inode_set_bytes(inode, 0);
6409 inode->i_mtime = current_time(inode);
6410 inode->i_atime = inode->i_mtime;
6411 inode->i_ctime = inode->i_mtime;
6412 BTRFS_I(inode)->i_otime = inode->i_mtime;
6414 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6415 struct btrfs_inode_item);
6416 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6417 sizeof(*inode_item));
6418 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6421 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6422 struct btrfs_inode_ref);
6423 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6424 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6425 ptr = (unsigned long)(ref + 1);
6426 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6429 btrfs_mark_buffer_dirty(path->nodes[0]);
6430 btrfs_free_path(path);
6432 btrfs_inherit_iflags(inode, dir);
6434 if (S_ISREG(mode)) {
6435 if (btrfs_test_opt(fs_info, NODATASUM))
6436 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6437 if (btrfs_test_opt(fs_info, NODATACOW))
6438 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6439 BTRFS_INODE_NODATASUM;
6442 inode_tree_add(inode);
6444 trace_btrfs_inode_new(inode);
6445 btrfs_set_inode_last_trans(trans, inode);
6447 btrfs_update_root_times(trans, root);
6449 ret = btrfs_inode_inherit_props(trans, inode, dir);
6452 "error inheriting props for ino %llu (root %llu): %d",
6453 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6458 unlock_new_inode(inode);
6461 BTRFS_I(dir)->index_cnt--;
6462 btrfs_free_path(path);
6464 return ERR_PTR(ret);
6467 static inline u8 btrfs_inode_type(struct inode *inode)
6469 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6473 * utility function to add 'inode' into 'parent_inode' with
6474 * a give name and a given sequence number.
6475 * if 'add_backref' is true, also insert a backref from the
6476 * inode to the parent directory.
6478 int btrfs_add_link(struct btrfs_trans_handle *trans,
6479 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6480 const char *name, int name_len, int add_backref, u64 index)
6482 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6484 struct btrfs_key key;
6485 struct btrfs_root *root = parent_inode->root;
6486 u64 ino = btrfs_ino(inode);
6487 u64 parent_ino = btrfs_ino(parent_inode);
6489 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6490 memcpy(&key, &inode->root->root_key, sizeof(key));
6493 key.type = BTRFS_INODE_ITEM_KEY;
6497 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6498 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6499 root->root_key.objectid, parent_ino,
6500 index, name, name_len);
6501 } else if (add_backref) {
6502 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6506 /* Nothing to clean up yet */
6510 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6512 btrfs_inode_type(&inode->vfs_inode), index);
6513 if (ret == -EEXIST || ret == -EOVERFLOW)
6516 btrfs_abort_transaction(trans, ret);
6520 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6522 inode_inc_iversion(&parent_inode->vfs_inode);
6523 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6524 current_time(&parent_inode->vfs_inode);
6525 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6527 btrfs_abort_transaction(trans, ret);
6531 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6534 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6535 root->root_key.objectid, parent_ino,
6536 &local_index, name, name_len);
6538 } else if (add_backref) {
6542 err = btrfs_del_inode_ref(trans, root, name, name_len,
6543 ino, parent_ino, &local_index);
6548 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6549 struct btrfs_inode *dir, struct dentry *dentry,
6550 struct btrfs_inode *inode, int backref, u64 index)
6552 int err = btrfs_add_link(trans, dir, inode,
6553 dentry->d_name.name, dentry->d_name.len,
6560 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6561 umode_t mode, dev_t rdev)
6563 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6564 struct btrfs_trans_handle *trans;
6565 struct btrfs_root *root = BTRFS_I(dir)->root;
6566 struct inode *inode = NULL;
6573 * 2 for inode item and ref
6575 * 1 for xattr if selinux is on
6577 trans = btrfs_start_transaction(root, 5);
6579 return PTR_ERR(trans);
6581 err = btrfs_find_free_ino(root, &objectid);
6585 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6586 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6588 if (IS_ERR(inode)) {
6589 err = PTR_ERR(inode);
6594 * If the active LSM wants to access the inode during
6595 * d_instantiate it needs these. Smack checks to see
6596 * if the filesystem supports xattrs by looking at the
6599 inode->i_op = &btrfs_special_inode_operations;
6600 init_special_inode(inode, inode->i_mode, rdev);
6602 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6604 goto out_unlock_inode;
6606 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6609 goto out_unlock_inode;
6611 btrfs_update_inode(trans, root, inode);
6612 unlock_new_inode(inode);
6613 d_instantiate(dentry, inode);
6617 btrfs_end_transaction(trans);
6618 btrfs_btree_balance_dirty(fs_info);
6620 inode_dec_link_count(inode);
6627 unlock_new_inode(inode);
6632 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6633 umode_t mode, bool excl)
6635 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6636 struct btrfs_trans_handle *trans;
6637 struct btrfs_root *root = BTRFS_I(dir)->root;
6638 struct inode *inode = NULL;
6639 int drop_inode_on_err = 0;
6645 * 2 for inode item and ref
6647 * 1 for xattr if selinux is on
6649 trans = btrfs_start_transaction(root, 5);
6651 return PTR_ERR(trans);
6653 err = btrfs_find_free_ino(root, &objectid);
6657 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6658 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6660 if (IS_ERR(inode)) {
6661 err = PTR_ERR(inode);
6664 drop_inode_on_err = 1;
6666 * If the active LSM wants to access the inode during
6667 * d_instantiate it needs these. Smack checks to see
6668 * if the filesystem supports xattrs by looking at the
6671 inode->i_fop = &btrfs_file_operations;
6672 inode->i_op = &btrfs_file_inode_operations;
6673 inode->i_mapping->a_ops = &btrfs_aops;
6675 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6677 goto out_unlock_inode;
6679 err = btrfs_update_inode(trans, root, inode);
6681 goto out_unlock_inode;
6683 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6686 goto out_unlock_inode;
6688 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6689 unlock_new_inode(inode);
6690 d_instantiate(dentry, inode);
6693 btrfs_end_transaction(trans);
6694 if (err && drop_inode_on_err) {
6695 inode_dec_link_count(inode);
6698 btrfs_btree_balance_dirty(fs_info);
6702 unlock_new_inode(inode);
6707 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6708 struct dentry *dentry)
6710 struct btrfs_trans_handle *trans = NULL;
6711 struct btrfs_root *root = BTRFS_I(dir)->root;
6712 struct inode *inode = d_inode(old_dentry);
6713 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6718 /* do not allow sys_link's with other subvols of the same device */
6719 if (root->objectid != BTRFS_I(inode)->root->objectid)
6722 if (inode->i_nlink >= BTRFS_LINK_MAX)
6725 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6730 * 2 items for inode and inode ref
6731 * 2 items for dir items
6732 * 1 item for parent inode
6734 trans = btrfs_start_transaction(root, 5);
6735 if (IS_ERR(trans)) {
6736 err = PTR_ERR(trans);
6741 /* There are several dir indexes for this inode, clear the cache. */
6742 BTRFS_I(inode)->dir_index = 0ULL;
6744 inode_inc_iversion(inode);
6745 inode->i_ctime = current_time(inode);
6747 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6749 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6755 struct dentry *parent = dentry->d_parent;
6756 err = btrfs_update_inode(trans, root, inode);
6759 if (inode->i_nlink == 1) {
6761 * If new hard link count is 1, it's a file created
6762 * with open(2) O_TMPFILE flag.
6764 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6768 d_instantiate(dentry, inode);
6769 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6774 btrfs_end_transaction(trans);
6776 inode_dec_link_count(inode);
6779 btrfs_btree_balance_dirty(fs_info);
6783 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6785 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6786 struct inode *inode = NULL;
6787 struct btrfs_trans_handle *trans;
6788 struct btrfs_root *root = BTRFS_I(dir)->root;
6790 int drop_on_err = 0;
6795 * 2 items for inode and ref
6796 * 2 items for dir items
6797 * 1 for xattr if selinux is on
6799 trans = btrfs_start_transaction(root, 5);
6801 return PTR_ERR(trans);
6803 err = btrfs_find_free_ino(root, &objectid);
6807 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6808 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6809 S_IFDIR | mode, &index);
6810 if (IS_ERR(inode)) {
6811 err = PTR_ERR(inode);
6816 /* these must be set before we unlock the inode */
6817 inode->i_op = &btrfs_dir_inode_operations;
6818 inode->i_fop = &btrfs_dir_file_operations;
6820 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6822 goto out_fail_inode;
6824 btrfs_i_size_write(BTRFS_I(inode), 0);
6825 err = btrfs_update_inode(trans, root, inode);
6827 goto out_fail_inode;
6829 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6830 dentry->d_name.name,
6831 dentry->d_name.len, 0, index);
6833 goto out_fail_inode;
6835 d_instantiate(dentry, inode);
6837 * mkdir is special. We're unlocking after we call d_instantiate
6838 * to avoid a race with nfsd calling d_instantiate.
6840 unlock_new_inode(inode);
6844 btrfs_end_transaction(trans);
6846 inode_dec_link_count(inode);
6849 btrfs_btree_balance_dirty(fs_info);
6853 unlock_new_inode(inode);
6857 static noinline int uncompress_inline(struct btrfs_path *path,
6859 size_t pg_offset, u64 extent_offset,
6860 struct btrfs_file_extent_item *item)
6863 struct extent_buffer *leaf = path->nodes[0];
6866 unsigned long inline_size;
6870 WARN_ON(pg_offset != 0);
6871 compress_type = btrfs_file_extent_compression(leaf, item);
6872 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6873 inline_size = btrfs_file_extent_inline_item_len(leaf,
6874 btrfs_item_nr(path->slots[0]));
6875 tmp = kmalloc(inline_size, GFP_NOFS);
6878 ptr = btrfs_file_extent_inline_start(item);
6880 read_extent_buffer(leaf, tmp, ptr, inline_size);
6882 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6883 ret = btrfs_decompress(compress_type, tmp, page,
6884 extent_offset, inline_size, max_size);
6887 * decompression code contains a memset to fill in any space between the end
6888 * of the uncompressed data and the end of max_size in case the decompressed
6889 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6890 * the end of an inline extent and the beginning of the next block, so we
6891 * cover that region here.
6894 if (max_size + pg_offset < PAGE_SIZE) {
6895 char *map = kmap(page);
6896 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6904 * a bit scary, this does extent mapping from logical file offset to the disk.
6905 * the ugly parts come from merging extents from the disk with the in-ram
6906 * representation. This gets more complex because of the data=ordered code,
6907 * where the in-ram extents might be locked pending data=ordered completion.
6909 * This also copies inline extents directly into the page.
6911 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6913 size_t pg_offset, u64 start, u64 len,
6916 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6919 u64 extent_start = 0;
6921 u64 objectid = btrfs_ino(inode);
6923 struct btrfs_path *path = NULL;
6924 struct btrfs_root *root = inode->root;
6925 struct btrfs_file_extent_item *item;
6926 struct extent_buffer *leaf;
6927 struct btrfs_key found_key;
6928 struct extent_map *em = NULL;
6929 struct extent_map_tree *em_tree = &inode->extent_tree;
6930 struct extent_io_tree *io_tree = &inode->io_tree;
6931 const bool new_inline = !page || create;
6933 read_lock(&em_tree->lock);
6934 em = lookup_extent_mapping(em_tree, start, len);
6936 em->bdev = fs_info->fs_devices->latest_bdev;
6937 read_unlock(&em_tree->lock);
6940 if (em->start > start || em->start + em->len <= start)
6941 free_extent_map(em);
6942 else if (em->block_start == EXTENT_MAP_INLINE && page)
6943 free_extent_map(em);
6947 em = alloc_extent_map();
6952 em->bdev = fs_info->fs_devices->latest_bdev;
6953 em->start = EXTENT_MAP_HOLE;
6954 em->orig_start = EXTENT_MAP_HOLE;
6956 em->block_len = (u64)-1;
6959 path = btrfs_alloc_path();
6965 * Chances are we'll be called again, so go ahead and do
6968 path->reada = READA_FORWARD;
6971 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6978 if (path->slots[0] == 0)
6983 leaf = path->nodes[0];
6984 item = btrfs_item_ptr(leaf, path->slots[0],
6985 struct btrfs_file_extent_item);
6986 /* are we inside the extent that was found? */
6987 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6988 found_type = found_key.type;
6989 if (found_key.objectid != objectid ||
6990 found_type != BTRFS_EXTENT_DATA_KEY) {
6992 * If we backup past the first extent we want to move forward
6993 * and see if there is an extent in front of us, otherwise we'll
6994 * say there is a hole for our whole search range which can
7001 found_type = btrfs_file_extent_type(leaf, item);
7002 extent_start = found_key.offset;
7003 if (found_type == BTRFS_FILE_EXTENT_REG ||
7004 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7005 extent_end = extent_start +
7006 btrfs_file_extent_num_bytes(leaf, item);
7008 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7010 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7012 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7013 extent_end = ALIGN(extent_start + size,
7014 fs_info->sectorsize);
7016 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7021 if (start >= extent_end) {
7023 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7024 ret = btrfs_next_leaf(root, path);
7031 leaf = path->nodes[0];
7033 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7034 if (found_key.objectid != objectid ||
7035 found_key.type != BTRFS_EXTENT_DATA_KEY)
7037 if (start + len <= found_key.offset)
7039 if (start > found_key.offset)
7042 em->orig_start = start;
7043 em->len = found_key.offset - start;
7047 btrfs_extent_item_to_extent_map(inode, path, item,
7050 if (found_type == BTRFS_FILE_EXTENT_REG ||
7051 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7053 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7057 size_t extent_offset;
7063 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7064 extent_offset = page_offset(page) + pg_offset - extent_start;
7065 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7066 size - extent_offset);
7067 em->start = extent_start + extent_offset;
7068 em->len = ALIGN(copy_size, fs_info->sectorsize);
7069 em->orig_block_len = em->len;
7070 em->orig_start = em->start;
7071 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7072 if (!PageUptodate(page)) {
7073 if (btrfs_file_extent_compression(leaf, item) !=
7074 BTRFS_COMPRESS_NONE) {
7075 ret = uncompress_inline(path, page, pg_offset,
7076 extent_offset, item);
7083 read_extent_buffer(leaf, map + pg_offset, ptr,
7085 if (pg_offset + copy_size < PAGE_SIZE) {
7086 memset(map + pg_offset + copy_size, 0,
7087 PAGE_SIZE - pg_offset -
7092 flush_dcache_page(page);
7094 set_extent_uptodate(io_tree, em->start,
7095 extent_map_end(em) - 1, NULL, GFP_NOFS);
7100 em->orig_start = start;
7103 em->block_start = EXTENT_MAP_HOLE;
7105 btrfs_release_path(path);
7106 if (em->start > start || extent_map_end(em) <= start) {
7108 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7109 em->start, em->len, start, len);
7115 write_lock(&em_tree->lock);
7116 err = btrfs_add_extent_mapping(em_tree, &em, start, len);
7117 write_unlock(&em_tree->lock);
7120 trace_btrfs_get_extent(root, inode, em);
7122 btrfs_free_path(path);
7124 free_extent_map(em);
7125 return ERR_PTR(err);
7127 BUG_ON(!em); /* Error is always set */
7131 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7133 size_t pg_offset, u64 start, u64 len,
7136 struct extent_map *em;
7137 struct extent_map *hole_em = NULL;
7138 u64 range_start = start;
7144 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7148 * If our em maps to:
7150 * - a pre-alloc extent,
7151 * there might actually be delalloc bytes behind it.
7153 if (em->block_start != EXTENT_MAP_HOLE &&
7154 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7159 /* check to see if we've wrapped (len == -1 or similar) */
7168 /* ok, we didn't find anything, lets look for delalloc */
7169 found = count_range_bits(&inode->io_tree, &range_start,
7170 end, len, EXTENT_DELALLOC, 1);
7171 found_end = range_start + found;
7172 if (found_end < range_start)
7173 found_end = (u64)-1;
7176 * we didn't find anything useful, return
7177 * the original results from get_extent()
7179 if (range_start > end || found_end <= start) {
7185 /* adjust the range_start to make sure it doesn't
7186 * go backwards from the start they passed in
7188 range_start = max(start, range_start);
7189 found = found_end - range_start;
7192 u64 hole_start = start;
7195 em = alloc_extent_map();
7201 * when btrfs_get_extent can't find anything it
7202 * returns one huge hole
7204 * make sure what it found really fits our range, and
7205 * adjust to make sure it is based on the start from
7209 u64 calc_end = extent_map_end(hole_em);
7211 if (calc_end <= start || (hole_em->start > end)) {
7212 free_extent_map(hole_em);
7215 hole_start = max(hole_em->start, start);
7216 hole_len = calc_end - hole_start;
7220 if (hole_em && range_start > hole_start) {
7221 /* our hole starts before our delalloc, so we
7222 * have to return just the parts of the hole
7223 * that go until the delalloc starts
7225 em->len = min(hole_len,
7226 range_start - hole_start);
7227 em->start = hole_start;
7228 em->orig_start = hole_start;
7230 * don't adjust block start at all,
7231 * it is fixed at EXTENT_MAP_HOLE
7233 em->block_start = hole_em->block_start;
7234 em->block_len = hole_len;
7235 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7236 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7238 em->start = range_start;
7240 em->orig_start = range_start;
7241 em->block_start = EXTENT_MAP_DELALLOC;
7242 em->block_len = found;
7249 free_extent_map(hole_em);
7251 free_extent_map(em);
7252 return ERR_PTR(err);
7257 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7260 const u64 orig_start,
7261 const u64 block_start,
7262 const u64 block_len,
7263 const u64 orig_block_len,
7264 const u64 ram_bytes,
7267 struct extent_map *em = NULL;
7270 if (type != BTRFS_ORDERED_NOCOW) {
7271 em = create_io_em(inode, start, len, orig_start,
7272 block_start, block_len, orig_block_len,
7274 BTRFS_COMPRESS_NONE, /* compress_type */
7279 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7280 len, block_len, type);
7283 free_extent_map(em);
7284 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7285 start + len - 1, 0);
7294 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7297 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7298 struct btrfs_root *root = BTRFS_I(inode)->root;
7299 struct extent_map *em;
7300 struct btrfs_key ins;
7304 alloc_hint = get_extent_allocation_hint(inode, start, len);
7305 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7306 0, alloc_hint, &ins, 1, 1);
7308 return ERR_PTR(ret);
7310 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7311 ins.objectid, ins.offset, ins.offset,
7312 ins.offset, BTRFS_ORDERED_REGULAR);
7313 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7315 btrfs_free_reserved_extent(fs_info, ins.objectid,
7322 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7323 * block must be cow'd
7325 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7326 u64 *orig_start, u64 *orig_block_len,
7329 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7330 struct btrfs_path *path;
7332 struct extent_buffer *leaf;
7333 struct btrfs_root *root = BTRFS_I(inode)->root;
7334 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7335 struct btrfs_file_extent_item *fi;
7336 struct btrfs_key key;
7343 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7345 path = btrfs_alloc_path();
7349 ret = btrfs_lookup_file_extent(NULL, root, path,
7350 btrfs_ino(BTRFS_I(inode)), offset, 0);
7354 slot = path->slots[0];
7357 /* can't find the item, must cow */
7364 leaf = path->nodes[0];
7365 btrfs_item_key_to_cpu(leaf, &key, slot);
7366 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7367 key.type != BTRFS_EXTENT_DATA_KEY) {
7368 /* not our file or wrong item type, must cow */
7372 if (key.offset > offset) {
7373 /* Wrong offset, must cow */
7377 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7378 found_type = btrfs_file_extent_type(leaf, fi);
7379 if (found_type != BTRFS_FILE_EXTENT_REG &&
7380 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7381 /* not a regular extent, must cow */
7385 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7388 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7389 if (extent_end <= offset)
7392 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7393 if (disk_bytenr == 0)
7396 if (btrfs_file_extent_compression(leaf, fi) ||
7397 btrfs_file_extent_encryption(leaf, fi) ||
7398 btrfs_file_extent_other_encoding(leaf, fi))
7401 backref_offset = btrfs_file_extent_offset(leaf, fi);
7404 *orig_start = key.offset - backref_offset;
7405 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7406 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7409 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7412 num_bytes = min(offset + *len, extent_end) - offset;
7413 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7416 range_end = round_up(offset + num_bytes,
7417 root->fs_info->sectorsize) - 1;
7418 ret = test_range_bit(io_tree, offset, range_end,
7419 EXTENT_DELALLOC, 0, NULL);
7426 btrfs_release_path(path);
7429 * look for other files referencing this extent, if we
7430 * find any we must cow
7433 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7434 key.offset - backref_offset, disk_bytenr);
7441 * adjust disk_bytenr and num_bytes to cover just the bytes
7442 * in this extent we are about to write. If there
7443 * are any csums in that range we have to cow in order
7444 * to keep the csums correct
7446 disk_bytenr += backref_offset;
7447 disk_bytenr += offset - key.offset;
7448 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7451 * all of the above have passed, it is safe to overwrite this extent
7457 btrfs_free_path(path);
7461 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7462 struct extent_state **cached_state, int writing)
7464 struct btrfs_ordered_extent *ordered;
7468 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7471 * We're concerned with the entire range that we're going to be
7472 * doing DIO to, so we need to make sure there's no ordered
7473 * extents in this range.
7475 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7476 lockend - lockstart + 1);
7479 * We need to make sure there are no buffered pages in this
7480 * range either, we could have raced between the invalidate in
7481 * generic_file_direct_write and locking the extent. The
7482 * invalidate needs to happen so that reads after a write do not
7486 (!writing || !filemap_range_has_page(inode->i_mapping,
7487 lockstart, lockend)))
7490 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7495 * If we are doing a DIO read and the ordered extent we
7496 * found is for a buffered write, we can not wait for it
7497 * to complete and retry, because if we do so we can
7498 * deadlock with concurrent buffered writes on page
7499 * locks. This happens only if our DIO read covers more
7500 * than one extent map, if at this point has already
7501 * created an ordered extent for a previous extent map
7502 * and locked its range in the inode's io tree, and a
7503 * concurrent write against that previous extent map's
7504 * range and this range started (we unlock the ranges
7505 * in the io tree only when the bios complete and
7506 * buffered writes always lock pages before attempting
7507 * to lock range in the io tree).
7510 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7511 btrfs_start_ordered_extent(inode, ordered, 1);
7514 btrfs_put_ordered_extent(ordered);
7517 * We could trigger writeback for this range (and wait
7518 * for it to complete) and then invalidate the pages for
7519 * this range (through invalidate_inode_pages2_range()),
7520 * but that can lead us to a deadlock with a concurrent
7521 * call to readpages() (a buffered read or a defrag call
7522 * triggered a readahead) on a page lock due to an
7523 * ordered dio extent we created before but did not have
7524 * yet a corresponding bio submitted (whence it can not
7525 * complete), which makes readpages() wait for that
7526 * ordered extent to complete while holding a lock on
7541 /* The callers of this must take lock_extent() */
7542 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7543 u64 orig_start, u64 block_start,
7544 u64 block_len, u64 orig_block_len,
7545 u64 ram_bytes, int compress_type,
7548 struct extent_map_tree *em_tree;
7549 struct extent_map *em;
7550 struct btrfs_root *root = BTRFS_I(inode)->root;
7553 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7554 type == BTRFS_ORDERED_COMPRESSED ||
7555 type == BTRFS_ORDERED_NOCOW ||
7556 type == BTRFS_ORDERED_REGULAR);
7558 em_tree = &BTRFS_I(inode)->extent_tree;
7559 em = alloc_extent_map();
7561 return ERR_PTR(-ENOMEM);
7564 em->orig_start = orig_start;
7566 em->block_len = block_len;
7567 em->block_start = block_start;
7568 em->bdev = root->fs_info->fs_devices->latest_bdev;
7569 em->orig_block_len = orig_block_len;
7570 em->ram_bytes = ram_bytes;
7571 em->generation = -1;
7572 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7573 if (type == BTRFS_ORDERED_PREALLOC) {
7574 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7575 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7576 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7577 em->compress_type = compress_type;
7581 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7582 em->start + em->len - 1, 0);
7583 write_lock(&em_tree->lock);
7584 ret = add_extent_mapping(em_tree, em, 1);
7585 write_unlock(&em_tree->lock);
7587 * The caller has taken lock_extent(), who could race with us
7590 } while (ret == -EEXIST);
7593 free_extent_map(em);
7594 return ERR_PTR(ret);
7597 /* em got 2 refs now, callers needs to do free_extent_map once. */
7601 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7602 struct buffer_head *bh_result, int create)
7604 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7605 struct extent_map *em;
7606 struct extent_state *cached_state = NULL;
7607 struct btrfs_dio_data *dio_data = NULL;
7608 u64 start = iblock << inode->i_blkbits;
7609 u64 lockstart, lockend;
7610 u64 len = bh_result->b_size;
7611 int unlock_bits = EXTENT_LOCKED;
7615 unlock_bits |= EXTENT_DIRTY;
7617 len = min_t(u64, len, fs_info->sectorsize);
7620 lockend = start + len - 1;
7622 if (current->journal_info) {
7624 * Need to pull our outstanding extents and set journal_info to NULL so
7625 * that anything that needs to check if there's a transaction doesn't get
7628 dio_data = current->journal_info;
7629 current->journal_info = NULL;
7633 * If this errors out it's because we couldn't invalidate pagecache for
7634 * this range and we need to fallback to buffered.
7636 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7642 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7649 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7650 * io. INLINE is special, and we could probably kludge it in here, but
7651 * it's still buffered so for safety lets just fall back to the generic
7654 * For COMPRESSED we _have_ to read the entire extent in so we can
7655 * decompress it, so there will be buffering required no matter what we
7656 * do, so go ahead and fallback to buffered.
7658 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7659 * to buffered IO. Don't blame me, this is the price we pay for using
7662 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7663 em->block_start == EXTENT_MAP_INLINE) {
7664 free_extent_map(em);
7669 /* Just a good old fashioned hole, return */
7670 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7671 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7672 free_extent_map(em);
7677 * We don't allocate a new extent in the following cases
7679 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7681 * 2) The extent is marked as PREALLOC. We're good to go here and can
7682 * just use the extent.
7686 len = min(len, em->len - (start - em->start));
7687 lockstart = start + len;
7691 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7692 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7693 em->block_start != EXTENT_MAP_HOLE)) {
7695 u64 block_start, orig_start, orig_block_len, ram_bytes;
7697 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7698 type = BTRFS_ORDERED_PREALLOC;
7700 type = BTRFS_ORDERED_NOCOW;
7701 len = min(len, em->len - (start - em->start));
7702 block_start = em->block_start + (start - em->start);
7704 if (can_nocow_extent(inode, start, &len, &orig_start,
7705 &orig_block_len, &ram_bytes) == 1 &&
7706 btrfs_inc_nocow_writers(fs_info, block_start)) {
7707 struct extent_map *em2;
7709 em2 = btrfs_create_dio_extent(inode, start, len,
7710 orig_start, block_start,
7711 len, orig_block_len,
7713 btrfs_dec_nocow_writers(fs_info, block_start);
7714 if (type == BTRFS_ORDERED_PREALLOC) {
7715 free_extent_map(em);
7718 if (em2 && IS_ERR(em2)) {
7723 * For inode marked NODATACOW or extent marked PREALLOC,
7724 * use the existing or preallocated extent, so does not
7725 * need to adjust btrfs_space_info's bytes_may_use.
7727 btrfs_free_reserved_data_space_noquota(inode,
7734 * this will cow the extent, reset the len in case we changed
7737 len = bh_result->b_size;
7738 free_extent_map(em);
7739 em = btrfs_new_extent_direct(inode, start, len);
7744 len = min(len, em->len - (start - em->start));
7746 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7748 bh_result->b_size = len;
7749 bh_result->b_bdev = em->bdev;
7750 set_buffer_mapped(bh_result);
7752 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7753 set_buffer_new(bh_result);
7756 * Need to update the i_size under the extent lock so buffered
7757 * readers will get the updated i_size when we unlock.
7759 if (!dio_data->overwrite && start + len > i_size_read(inode))
7760 i_size_write(inode, start + len);
7762 WARN_ON(dio_data->reserve < len);
7763 dio_data->reserve -= len;
7764 dio_data->unsubmitted_oe_range_end = start + len;
7765 current->journal_info = dio_data;
7769 * In the case of write we need to clear and unlock the entire range,
7770 * in the case of read we need to unlock only the end area that we
7771 * aren't using if there is any left over space.
7773 if (lockstart < lockend) {
7774 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7775 lockend, unlock_bits, 1, 0,
7778 free_extent_state(cached_state);
7781 free_extent_map(em);
7786 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7787 unlock_bits, 1, 0, &cached_state);
7790 current->journal_info = dio_data;
7794 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7798 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7801 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7803 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7807 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7812 static int btrfs_check_dio_repairable(struct inode *inode,
7813 struct bio *failed_bio,
7814 struct io_failure_record *failrec,
7817 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7820 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7821 if (num_copies == 1) {
7823 * we only have a single copy of the data, so don't bother with
7824 * all the retry and error correction code that follows. no
7825 * matter what the error is, it is very likely to persist.
7827 btrfs_debug(fs_info,
7828 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7829 num_copies, failrec->this_mirror, failed_mirror);
7833 failrec->failed_mirror = failed_mirror;
7834 failrec->this_mirror++;
7835 if (failrec->this_mirror == failed_mirror)
7836 failrec->this_mirror++;
7838 if (failrec->this_mirror > num_copies) {
7839 btrfs_debug(fs_info,
7840 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7841 num_copies, failrec->this_mirror, failed_mirror);
7848 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7849 struct page *page, unsigned int pgoff,
7850 u64 start, u64 end, int failed_mirror,
7851 bio_end_io_t *repair_endio, void *repair_arg)
7853 struct io_failure_record *failrec;
7854 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7855 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7858 unsigned int read_mode = 0;
7861 blk_status_t status;
7862 struct bio_vec bvec;
7864 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7866 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7868 return errno_to_blk_status(ret);
7870 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7873 free_io_failure(failure_tree, io_tree, failrec);
7874 return BLK_STS_IOERR;
7877 segs = bio_segments(failed_bio);
7878 bio_get_first_bvec(failed_bio, &bvec);
7880 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7881 read_mode |= REQ_FAILFAST_DEV;
7883 isector = start - btrfs_io_bio(failed_bio)->logical;
7884 isector >>= inode->i_sb->s_blocksize_bits;
7885 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7886 pgoff, isector, repair_endio, repair_arg);
7887 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7889 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7890 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7891 read_mode, failrec->this_mirror, failrec->in_validation);
7893 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7895 free_io_failure(failure_tree, io_tree, failrec);
7902 struct btrfs_retry_complete {
7903 struct completion done;
7904 struct inode *inode;
7909 static void btrfs_retry_endio_nocsum(struct bio *bio)
7911 struct btrfs_retry_complete *done = bio->bi_private;
7912 struct inode *inode = done->inode;
7913 struct bio_vec *bvec;
7914 struct extent_io_tree *io_tree, *failure_tree;
7920 ASSERT(bio->bi_vcnt == 1);
7921 io_tree = &BTRFS_I(inode)->io_tree;
7922 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7923 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7926 ASSERT(!bio_flagged(bio, BIO_CLONED));
7927 bio_for_each_segment_all(bvec, bio, i)
7928 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7929 io_tree, done->start, bvec->bv_page,
7930 btrfs_ino(BTRFS_I(inode)), 0);
7932 complete(&done->done);
7936 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7937 struct btrfs_io_bio *io_bio)
7939 struct btrfs_fs_info *fs_info;
7940 struct bio_vec bvec;
7941 struct bvec_iter iter;
7942 struct btrfs_retry_complete done;
7948 blk_status_t err = BLK_STS_OK;
7950 fs_info = BTRFS_I(inode)->root->fs_info;
7951 sectorsize = fs_info->sectorsize;
7953 start = io_bio->logical;
7955 io_bio->bio.bi_iter = io_bio->iter;
7957 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7958 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7959 pgoff = bvec.bv_offset;
7961 next_block_or_try_again:
7964 init_completion(&done.done);
7966 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7967 pgoff, start, start + sectorsize - 1,
7969 btrfs_retry_endio_nocsum, &done);
7975 wait_for_completion_io(&done.done);
7977 if (!done.uptodate) {
7978 /* We might have another mirror, so try again */
7979 goto next_block_or_try_again;
7983 start += sectorsize;
7987 pgoff += sectorsize;
7988 ASSERT(pgoff < PAGE_SIZE);
7989 goto next_block_or_try_again;
7996 static void btrfs_retry_endio(struct bio *bio)
7998 struct btrfs_retry_complete *done = bio->bi_private;
7999 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8000 struct extent_io_tree *io_tree, *failure_tree;
8001 struct inode *inode = done->inode;
8002 struct bio_vec *bvec;
8012 ASSERT(bio->bi_vcnt == 1);
8013 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
8015 io_tree = &BTRFS_I(inode)->io_tree;
8016 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8018 ASSERT(!bio_flagged(bio, BIO_CLONED));
8019 bio_for_each_segment_all(bvec, bio, i) {
8020 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8021 bvec->bv_offset, done->start,
8024 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8025 failure_tree, io_tree, done->start,
8027 btrfs_ino(BTRFS_I(inode)),
8033 done->uptodate = uptodate;
8035 complete(&done->done);
8039 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8040 struct btrfs_io_bio *io_bio, blk_status_t err)
8042 struct btrfs_fs_info *fs_info;
8043 struct bio_vec bvec;
8044 struct bvec_iter iter;
8045 struct btrfs_retry_complete done;
8052 bool uptodate = (err == 0);
8054 blk_status_t status;
8056 fs_info = BTRFS_I(inode)->root->fs_info;
8057 sectorsize = fs_info->sectorsize;
8060 start = io_bio->logical;
8062 io_bio->bio.bi_iter = io_bio->iter;
8064 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8065 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8067 pgoff = bvec.bv_offset;
8070 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8071 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8072 bvec.bv_page, pgoff, start, sectorsize);
8079 init_completion(&done.done);
8081 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8082 pgoff, start, start + sectorsize - 1,
8083 io_bio->mirror_num, btrfs_retry_endio,
8090 wait_for_completion_io(&done.done);
8092 if (!done.uptodate) {
8093 /* We might have another mirror, so try again */
8097 offset += sectorsize;
8098 start += sectorsize;
8104 pgoff += sectorsize;
8105 ASSERT(pgoff < PAGE_SIZE);
8113 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8114 struct btrfs_io_bio *io_bio, blk_status_t err)
8116 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8120 return __btrfs_correct_data_nocsum(inode, io_bio);
8124 return __btrfs_subio_endio_read(inode, io_bio, err);
8128 static void btrfs_endio_direct_read(struct bio *bio)
8130 struct btrfs_dio_private *dip = bio->bi_private;
8131 struct inode *inode = dip->inode;
8132 struct bio *dio_bio;
8133 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8134 blk_status_t err = bio->bi_status;
8136 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8137 err = btrfs_subio_endio_read(inode, io_bio, err);
8139 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8140 dip->logical_offset + dip->bytes - 1);
8141 dio_bio = dip->dio_bio;
8145 dio_bio->bi_status = err;
8146 dio_end_io(dio_bio);
8149 io_bio->end_io(io_bio, blk_status_to_errno(err));
8153 static void __endio_write_update_ordered(struct inode *inode,
8154 const u64 offset, const u64 bytes,
8155 const bool uptodate)
8157 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8158 struct btrfs_ordered_extent *ordered = NULL;
8159 struct btrfs_workqueue *wq;
8160 btrfs_work_func_t func;
8161 u64 ordered_offset = offset;
8162 u64 ordered_bytes = bytes;
8166 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8167 wq = fs_info->endio_freespace_worker;
8168 func = btrfs_freespace_write_helper;
8170 wq = fs_info->endio_write_workers;
8171 func = btrfs_endio_write_helper;
8175 last_offset = ordered_offset;
8176 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8183 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8184 btrfs_queue_work(wq, &ordered->work);
8187 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8188 * in the range, we can exit.
8190 if (ordered_offset == last_offset)
8193 * our bio might span multiple ordered extents. If we haven't
8194 * completed the accounting for the whole dio, go back and try again
8196 if (ordered_offset < offset + bytes) {
8197 ordered_bytes = offset + bytes - ordered_offset;
8203 static void btrfs_endio_direct_write(struct bio *bio)
8205 struct btrfs_dio_private *dip = bio->bi_private;
8206 struct bio *dio_bio = dip->dio_bio;
8208 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8209 dip->bytes, !bio->bi_status);
8213 dio_bio->bi_status = bio->bi_status;
8214 dio_end_io(dio_bio);
8218 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8219 struct bio *bio, u64 offset)
8221 struct inode *inode = private_data;
8223 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8224 BUG_ON(ret); /* -ENOMEM */
8228 static void btrfs_end_dio_bio(struct bio *bio)
8230 struct btrfs_dio_private *dip = bio->bi_private;
8231 blk_status_t err = bio->bi_status;
8234 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8235 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8236 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8238 (unsigned long long)bio->bi_iter.bi_sector,
8239 bio->bi_iter.bi_size, err);
8241 if (dip->subio_endio)
8242 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8246 * We want to perceive the errors flag being set before
8247 * decrementing the reference count. We don't need a barrier
8248 * since atomic operations with a return value are fully
8249 * ordered as per atomic_t.txt
8254 /* if there are more bios still pending for this dio, just exit */
8255 if (!atomic_dec_and_test(&dip->pending_bios))
8259 bio_io_error(dip->orig_bio);
8261 dip->dio_bio->bi_status = BLK_STS_OK;
8262 bio_endio(dip->orig_bio);
8268 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8269 struct btrfs_dio_private *dip,
8273 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8274 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8278 * We load all the csum data we need when we submit
8279 * the first bio to reduce the csum tree search and
8282 if (dip->logical_offset == file_offset) {
8283 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8289 if (bio == dip->orig_bio)
8292 file_offset -= dip->logical_offset;
8293 file_offset >>= inode->i_sb->s_blocksize_bits;
8294 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8299 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8300 struct inode *inode, u64 file_offset, int async_submit)
8302 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8303 struct btrfs_dio_private *dip = bio->bi_private;
8304 bool write = bio_op(bio) == REQ_OP_WRITE;
8307 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8309 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8312 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8317 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8320 if (write && async_submit) {
8321 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8323 btrfs_submit_bio_start_direct_io,
8324 btrfs_submit_bio_done);
8328 * If we aren't doing async submit, calculate the csum of the
8331 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8335 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8341 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8346 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8348 struct inode *inode = dip->inode;
8349 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8351 struct bio *orig_bio = dip->orig_bio;
8352 u64 start_sector = orig_bio->bi_iter.bi_sector;
8353 u64 file_offset = dip->logical_offset;
8355 int async_submit = 0;
8357 int clone_offset = 0;
8360 blk_status_t status;
8362 map_length = orig_bio->bi_iter.bi_size;
8363 submit_len = map_length;
8364 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8365 &map_length, NULL, 0);
8369 if (map_length >= submit_len) {
8371 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8375 /* async crcs make it difficult to collect full stripe writes. */
8376 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8382 ASSERT(map_length <= INT_MAX);
8383 atomic_inc(&dip->pending_bios);
8385 clone_len = min_t(int, submit_len, map_length);
8388 * This will never fail as it's passing GPF_NOFS and
8389 * the allocation is backed by btrfs_bioset.
8391 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8393 bio->bi_private = dip;
8394 bio->bi_end_io = btrfs_end_dio_bio;
8395 btrfs_io_bio(bio)->logical = file_offset;
8397 ASSERT(submit_len >= clone_len);
8398 submit_len -= clone_len;
8399 if (submit_len == 0)
8403 * Increase the count before we submit the bio so we know
8404 * the end IO handler won't happen before we increase the
8405 * count. Otherwise, the dip might get freed before we're
8406 * done setting it up.
8408 atomic_inc(&dip->pending_bios);
8410 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8414 atomic_dec(&dip->pending_bios);
8418 clone_offset += clone_len;
8419 start_sector += clone_len >> 9;
8420 file_offset += clone_len;
8422 map_length = submit_len;
8423 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8424 start_sector << 9, &map_length, NULL, 0);
8427 } while (submit_len > 0);
8430 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8438 * Before atomic variable goto zero, we must make sure dip->errors is
8439 * perceived to be set. This ordering is ensured by the fact that an
8440 * atomic operations with a return value are fully ordered as per
8443 if (atomic_dec_and_test(&dip->pending_bios))
8444 bio_io_error(dip->orig_bio);
8446 /* bio_end_io() will handle error, so we needn't return it */
8450 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8453 struct btrfs_dio_private *dip = NULL;
8454 struct bio *bio = NULL;
8455 struct btrfs_io_bio *io_bio;
8456 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8459 bio = btrfs_bio_clone(dio_bio);
8461 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8467 dip->private = dio_bio->bi_private;
8469 dip->logical_offset = file_offset;
8470 dip->bytes = dio_bio->bi_iter.bi_size;
8471 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8472 bio->bi_private = dip;
8473 dip->orig_bio = bio;
8474 dip->dio_bio = dio_bio;
8475 atomic_set(&dip->pending_bios, 0);
8476 io_bio = btrfs_io_bio(bio);
8477 io_bio->logical = file_offset;
8480 bio->bi_end_io = btrfs_endio_direct_write;
8482 bio->bi_end_io = btrfs_endio_direct_read;
8483 dip->subio_endio = btrfs_subio_endio_read;
8487 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8488 * even if we fail to submit a bio, because in such case we do the
8489 * corresponding error handling below and it must not be done a second
8490 * time by btrfs_direct_IO().
8493 struct btrfs_dio_data *dio_data = current->journal_info;
8495 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8497 dio_data->unsubmitted_oe_range_start =
8498 dio_data->unsubmitted_oe_range_end;
8501 ret = btrfs_submit_direct_hook(dip);
8506 io_bio->end_io(io_bio, ret);
8510 * If we arrived here it means either we failed to submit the dip
8511 * or we either failed to clone the dio_bio or failed to allocate the
8512 * dip. If we cloned the dio_bio and allocated the dip, we can just
8513 * call bio_endio against our io_bio so that we get proper resource
8514 * cleanup if we fail to submit the dip, otherwise, we must do the
8515 * same as btrfs_endio_direct_[write|read] because we can't call these
8516 * callbacks - they require an allocated dip and a clone of dio_bio.
8521 * The end io callbacks free our dip, do the final put on bio
8522 * and all the cleanup and final put for dio_bio (through
8529 __endio_write_update_ordered(inode,
8531 dio_bio->bi_iter.bi_size,
8534 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8535 file_offset + dio_bio->bi_iter.bi_size - 1);
8537 dio_bio->bi_status = BLK_STS_IOERR;
8539 * Releases and cleans up our dio_bio, no need to bio_put()
8540 * nor bio_endio()/bio_io_error() against dio_bio.
8542 dio_end_io(dio_bio);
8549 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8550 const struct iov_iter *iter, loff_t offset)
8554 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8555 ssize_t retval = -EINVAL;
8557 if (offset & blocksize_mask)
8560 if (iov_iter_alignment(iter) & blocksize_mask)
8563 /* If this is a write we don't need to check anymore */
8564 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8567 * Check to make sure we don't have duplicate iov_base's in this
8568 * iovec, if so return EINVAL, otherwise we'll get csum errors
8569 * when reading back.
8571 for (seg = 0; seg < iter->nr_segs; seg++) {
8572 for (i = seg + 1; i < iter->nr_segs; i++) {
8573 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8582 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8584 struct file *file = iocb->ki_filp;
8585 struct inode *inode = file->f_mapping->host;
8586 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8587 struct btrfs_dio_data dio_data = { 0 };
8588 struct extent_changeset *data_reserved = NULL;
8589 loff_t offset = iocb->ki_pos;
8593 bool relock = false;
8596 if (check_direct_IO(fs_info, iter, offset))
8599 inode_dio_begin(inode);
8602 * The generic stuff only does filemap_write_and_wait_range, which
8603 * isn't enough if we've written compressed pages to this area, so
8604 * we need to flush the dirty pages again to make absolutely sure
8605 * that any outstanding dirty pages are on disk.
8607 count = iov_iter_count(iter);
8608 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8609 &BTRFS_I(inode)->runtime_flags))
8610 filemap_fdatawrite_range(inode->i_mapping, offset,
8611 offset + count - 1);
8613 if (iov_iter_rw(iter) == WRITE) {
8615 * If the write DIO is beyond the EOF, we need update
8616 * the isize, but it is protected by i_mutex. So we can
8617 * not unlock the i_mutex at this case.
8619 if (offset + count <= inode->i_size) {
8620 dio_data.overwrite = 1;
8621 inode_unlock(inode);
8623 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8627 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8633 * We need to know how many extents we reserved so that we can
8634 * do the accounting properly if we go over the number we
8635 * originally calculated. Abuse current->journal_info for this.
8637 dio_data.reserve = round_up(count,
8638 fs_info->sectorsize);
8639 dio_data.unsubmitted_oe_range_start = (u64)offset;
8640 dio_data.unsubmitted_oe_range_end = (u64)offset;
8641 current->journal_info = &dio_data;
8642 down_read(&BTRFS_I(inode)->dio_sem);
8643 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8644 &BTRFS_I(inode)->runtime_flags)) {
8645 inode_dio_end(inode);
8646 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8650 ret = __blockdev_direct_IO(iocb, inode,
8651 fs_info->fs_devices->latest_bdev,
8652 iter, btrfs_get_blocks_direct, NULL,
8653 btrfs_submit_direct, flags);
8654 if (iov_iter_rw(iter) == WRITE) {
8655 up_read(&BTRFS_I(inode)->dio_sem);
8656 current->journal_info = NULL;
8657 if (ret < 0 && ret != -EIOCBQUEUED) {
8658 if (dio_data.reserve)
8659 btrfs_delalloc_release_space(inode, data_reserved,
8660 offset, dio_data.reserve);
8662 * On error we might have left some ordered extents
8663 * without submitting corresponding bios for them, so
8664 * cleanup them up to avoid other tasks getting them
8665 * and waiting for them to complete forever.
8667 if (dio_data.unsubmitted_oe_range_start <
8668 dio_data.unsubmitted_oe_range_end)
8669 __endio_write_update_ordered(inode,
8670 dio_data.unsubmitted_oe_range_start,
8671 dio_data.unsubmitted_oe_range_end -
8672 dio_data.unsubmitted_oe_range_start,
8674 } else if (ret >= 0 && (size_t)ret < count)
8675 btrfs_delalloc_release_space(inode, data_reserved,
8676 offset, count - (size_t)ret);
8677 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
8681 inode_dio_end(inode);
8685 extent_changeset_free(data_reserved);
8689 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8691 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8692 __u64 start, __u64 len)
8696 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8700 return extent_fiemap(inode, fieinfo, start, len);
8703 int btrfs_readpage(struct file *file, struct page *page)
8705 struct extent_io_tree *tree;
8706 tree = &BTRFS_I(page->mapping->host)->io_tree;
8707 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8710 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8712 struct inode *inode = page->mapping->host;
8715 if (current->flags & PF_MEMALLOC) {
8716 redirty_page_for_writepage(wbc, page);
8722 * If we are under memory pressure we will call this directly from the
8723 * VM, we need to make sure we have the inode referenced for the ordered
8724 * extent. If not just return like we didn't do anything.
8726 if (!igrab(inode)) {
8727 redirty_page_for_writepage(wbc, page);
8728 return AOP_WRITEPAGE_ACTIVATE;
8730 ret = extent_write_full_page(page, wbc);
8731 btrfs_add_delayed_iput(inode);
8735 static int btrfs_writepages(struct address_space *mapping,
8736 struct writeback_control *wbc)
8738 struct extent_io_tree *tree;
8740 tree = &BTRFS_I(mapping->host)->io_tree;
8741 return extent_writepages(tree, mapping, wbc);
8745 btrfs_readpages(struct file *file, struct address_space *mapping,
8746 struct list_head *pages, unsigned nr_pages)
8748 struct extent_io_tree *tree;
8749 tree = &BTRFS_I(mapping->host)->io_tree;
8750 return extent_readpages(tree, mapping, pages, nr_pages);
8752 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8754 struct extent_io_tree *tree;
8755 struct extent_map_tree *map;
8758 tree = &BTRFS_I(page->mapping->host)->io_tree;
8759 map = &BTRFS_I(page->mapping->host)->extent_tree;
8760 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8762 ClearPagePrivate(page);
8763 set_page_private(page, 0);
8769 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8771 if (PageWriteback(page) || PageDirty(page))
8773 return __btrfs_releasepage(page, gfp_flags);
8776 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8777 unsigned int length)
8779 struct inode *inode = page->mapping->host;
8780 struct extent_io_tree *tree;
8781 struct btrfs_ordered_extent *ordered;
8782 struct extent_state *cached_state = NULL;
8783 u64 page_start = page_offset(page);
8784 u64 page_end = page_start + PAGE_SIZE - 1;
8787 int inode_evicting = inode->i_state & I_FREEING;
8790 * we have the page locked, so new writeback can't start,
8791 * and the dirty bit won't be cleared while we are here.
8793 * Wait for IO on this page so that we can safely clear
8794 * the PagePrivate2 bit and do ordered accounting
8796 wait_on_page_writeback(page);
8798 tree = &BTRFS_I(inode)->io_tree;
8800 btrfs_releasepage(page, GFP_NOFS);
8804 if (!inode_evicting)
8805 lock_extent_bits(tree, page_start, page_end, &cached_state);
8808 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8809 page_end - start + 1);
8811 end = min(page_end, ordered->file_offset + ordered->len - 1);
8813 * IO on this page will never be started, so we need
8814 * to account for any ordered extents now
8816 if (!inode_evicting)
8817 clear_extent_bit(tree, start, end,
8818 EXTENT_DIRTY | EXTENT_DELALLOC |
8819 EXTENT_DELALLOC_NEW |
8820 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8821 EXTENT_DEFRAG, 1, 0, &cached_state);
8823 * whoever cleared the private bit is responsible
8824 * for the finish_ordered_io
8826 if (TestClearPagePrivate2(page)) {
8827 struct btrfs_ordered_inode_tree *tree;
8830 tree = &BTRFS_I(inode)->ordered_tree;
8832 spin_lock_irq(&tree->lock);
8833 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8834 new_len = start - ordered->file_offset;
8835 if (new_len < ordered->truncated_len)
8836 ordered->truncated_len = new_len;
8837 spin_unlock_irq(&tree->lock);
8839 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8841 end - start + 1, 1))
8842 btrfs_finish_ordered_io(ordered);
8844 btrfs_put_ordered_extent(ordered);
8845 if (!inode_evicting) {
8846 cached_state = NULL;
8847 lock_extent_bits(tree, start, end,
8852 if (start < page_end)
8857 * Qgroup reserved space handler
8858 * Page here will be either
8859 * 1) Already written to disk
8860 * In this case, its reserved space is released from data rsv map
8861 * and will be freed by delayed_ref handler finally.
8862 * So even we call qgroup_free_data(), it won't decrease reserved
8864 * 2) Not written to disk
8865 * This means the reserved space should be freed here. However,
8866 * if a truncate invalidates the page (by clearing PageDirty)
8867 * and the page is accounted for while allocating extent
8868 * in btrfs_check_data_free_space() we let delayed_ref to
8869 * free the entire extent.
8871 if (PageDirty(page))
8872 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8873 if (!inode_evicting) {
8874 clear_extent_bit(tree, page_start, page_end,
8875 EXTENT_LOCKED | EXTENT_DIRTY |
8876 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8877 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8880 __btrfs_releasepage(page, GFP_NOFS);
8883 ClearPageChecked(page);
8884 if (PagePrivate(page)) {
8885 ClearPagePrivate(page);
8886 set_page_private(page, 0);
8892 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8893 * called from a page fault handler when a page is first dirtied. Hence we must
8894 * be careful to check for EOF conditions here. We set the page up correctly
8895 * for a written page which means we get ENOSPC checking when writing into
8896 * holes and correct delalloc and unwritten extent mapping on filesystems that
8897 * support these features.
8899 * We are not allowed to take the i_mutex here so we have to play games to
8900 * protect against truncate races as the page could now be beyond EOF. Because
8901 * vmtruncate() writes the inode size before removing pages, once we have the
8902 * page lock we can determine safely if the page is beyond EOF. If it is not
8903 * beyond EOF, then the page is guaranteed safe against truncation until we
8906 int btrfs_page_mkwrite(struct vm_fault *vmf)
8908 struct page *page = vmf->page;
8909 struct inode *inode = file_inode(vmf->vma->vm_file);
8910 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8911 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8912 struct btrfs_ordered_extent *ordered;
8913 struct extent_state *cached_state = NULL;
8914 struct extent_changeset *data_reserved = NULL;
8916 unsigned long zero_start;
8925 reserved_space = PAGE_SIZE;
8927 sb_start_pagefault(inode->i_sb);
8928 page_start = page_offset(page);
8929 page_end = page_start + PAGE_SIZE - 1;
8933 * Reserving delalloc space after obtaining the page lock can lead to
8934 * deadlock. For example, if a dirty page is locked by this function
8935 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8936 * dirty page write out, then the btrfs_writepage() function could
8937 * end up waiting indefinitely to get a lock on the page currently
8938 * being processed by btrfs_page_mkwrite() function.
8940 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8943 ret = file_update_time(vmf->vma->vm_file);
8949 else /* -ENOSPC, -EIO, etc */
8950 ret = VM_FAULT_SIGBUS;
8956 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8959 size = i_size_read(inode);
8961 if ((page->mapping != inode->i_mapping) ||
8962 (page_start >= size)) {
8963 /* page got truncated out from underneath us */
8966 wait_on_page_writeback(page);
8968 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8969 set_page_extent_mapped(page);
8972 * we can't set the delalloc bits if there are pending ordered
8973 * extents. Drop our locks and wait for them to finish
8975 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8978 unlock_extent_cached(io_tree, page_start, page_end,
8981 btrfs_start_ordered_extent(inode, ordered, 1);
8982 btrfs_put_ordered_extent(ordered);
8986 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8987 reserved_space = round_up(size - page_start,
8988 fs_info->sectorsize);
8989 if (reserved_space < PAGE_SIZE) {
8990 end = page_start + reserved_space - 1;
8991 btrfs_delalloc_release_space(inode, data_reserved,
8992 page_start, PAGE_SIZE - reserved_space);
8997 * page_mkwrite gets called when the page is firstly dirtied after it's
8998 * faulted in, but write(2) could also dirty a page and set delalloc
8999 * bits, thus in this case for space account reason, we still need to
9000 * clear any delalloc bits within this page range since we have to
9001 * reserve data&meta space before lock_page() (see above comments).
9003 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9004 EXTENT_DIRTY | EXTENT_DELALLOC |
9005 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9006 0, 0, &cached_state);
9008 ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
9011 unlock_extent_cached(io_tree, page_start, page_end,
9013 ret = VM_FAULT_SIGBUS;
9018 /* page is wholly or partially inside EOF */
9019 if (page_start + PAGE_SIZE > size)
9020 zero_start = size & ~PAGE_MASK;
9022 zero_start = PAGE_SIZE;
9024 if (zero_start != PAGE_SIZE) {
9026 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9027 flush_dcache_page(page);
9030 ClearPageChecked(page);
9031 set_page_dirty(page);
9032 SetPageUptodate(page);
9034 BTRFS_I(inode)->last_trans = fs_info->generation;
9035 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9036 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9038 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
9042 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9043 sb_end_pagefault(inode->i_sb);
9044 extent_changeset_free(data_reserved);
9045 return VM_FAULT_LOCKED;
9049 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9050 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9053 sb_end_pagefault(inode->i_sb);
9054 extent_changeset_free(data_reserved);
9058 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
9060 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9061 struct btrfs_root *root = BTRFS_I(inode)->root;
9062 struct btrfs_block_rsv *rsv;
9065 struct btrfs_trans_handle *trans;
9066 u64 mask = fs_info->sectorsize - 1;
9067 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9069 if (!skip_writeback) {
9070 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9077 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9078 * 3 things going on here
9080 * 1) We need to reserve space for our orphan item and the space to
9081 * delete our orphan item. Lord knows we don't want to have a dangling
9082 * orphan item because we didn't reserve space to remove it.
9084 * 2) We need to reserve space to update our inode.
9086 * 3) We need to have something to cache all the space that is going to
9087 * be free'd up by the truncate operation, but also have some slack
9088 * space reserved in case it uses space during the truncate (thank you
9089 * very much snapshotting).
9091 * And we need these to all be separate. The fact is we can use a lot of
9092 * space doing the truncate, and we have no earthly idea how much space
9093 * we will use, so we need the truncate reservation to be separate so it
9094 * doesn't end up using space reserved for updating the inode or
9095 * removing the orphan item. We also need to be able to stop the
9096 * transaction and start a new one, which means we need to be able to
9097 * update the inode several times, and we have no idea of knowing how
9098 * many times that will be, so we can't just reserve 1 item for the
9099 * entirety of the operation, so that has to be done separately as well.
9100 * Then there is the orphan item, which does indeed need to be held on
9101 * to for the whole operation, and we need nobody to touch this reserved
9102 * space except the orphan code.
9104 * So that leaves us with
9106 * 1) root->orphan_block_rsv - for the orphan deletion.
9107 * 2) rsv - for the truncate reservation, which we will steal from the
9108 * transaction reservation.
9109 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9110 * updating the inode.
9112 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9115 rsv->size = min_size;
9119 * 1 for the truncate slack space
9120 * 1 for updating the inode.
9122 trans = btrfs_start_transaction(root, 2);
9123 if (IS_ERR(trans)) {
9124 err = PTR_ERR(trans);
9128 /* Migrate the slack space for the truncate to our reserve */
9129 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9134 * So if we truncate and then write and fsync we normally would just
9135 * write the extents that changed, which is a problem if we need to
9136 * first truncate that entire inode. So set this flag so we write out
9137 * all of the extents in the inode to the sync log so we're completely
9140 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9141 trans->block_rsv = rsv;
9144 ret = btrfs_truncate_inode_items(trans, root, inode,
9146 BTRFS_EXTENT_DATA_KEY);
9147 trans->block_rsv = &fs_info->trans_block_rsv;
9148 if (ret != -ENOSPC && ret != -EAGAIN) {
9153 ret = btrfs_update_inode(trans, root, inode);
9159 btrfs_end_transaction(trans);
9160 btrfs_btree_balance_dirty(fs_info);
9162 trans = btrfs_start_transaction(root, 2);
9163 if (IS_ERR(trans)) {
9164 ret = err = PTR_ERR(trans);
9169 btrfs_block_rsv_release(fs_info, rsv, -1);
9170 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9172 BUG_ON(ret); /* shouldn't happen */
9173 trans->block_rsv = rsv;
9177 * We can't call btrfs_truncate_block inside a trans handle as we could
9178 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9179 * we've truncated everything except the last little bit, and can do
9180 * btrfs_truncate_block and then update the disk_i_size.
9182 if (ret == NEED_TRUNCATE_BLOCK) {
9183 btrfs_end_transaction(trans);
9184 btrfs_btree_balance_dirty(fs_info);
9186 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9189 trans = btrfs_start_transaction(root, 1);
9190 if (IS_ERR(trans)) {
9191 ret = PTR_ERR(trans);
9194 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9197 if (ret == 0 && inode->i_nlink > 0) {
9198 trans->block_rsv = root->orphan_block_rsv;
9199 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9205 trans->block_rsv = &fs_info->trans_block_rsv;
9206 ret = btrfs_update_inode(trans, root, inode);
9210 ret = btrfs_end_transaction(trans);
9211 btrfs_btree_balance_dirty(fs_info);
9214 btrfs_free_block_rsv(fs_info, rsv);
9223 * create a new subvolume directory/inode (helper for the ioctl).
9225 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9226 struct btrfs_root *new_root,
9227 struct btrfs_root *parent_root,
9230 struct inode *inode;
9234 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9235 new_dirid, new_dirid,
9236 S_IFDIR | (~current_umask() & S_IRWXUGO),
9239 return PTR_ERR(inode);
9240 inode->i_op = &btrfs_dir_inode_operations;
9241 inode->i_fop = &btrfs_dir_file_operations;
9243 set_nlink(inode, 1);
9244 btrfs_i_size_write(BTRFS_I(inode), 0);
9245 unlock_new_inode(inode);
9247 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9249 btrfs_err(new_root->fs_info,
9250 "error inheriting subvolume %llu properties: %d",
9251 new_root->root_key.objectid, err);
9253 err = btrfs_update_inode(trans, new_root, inode);
9259 struct inode *btrfs_alloc_inode(struct super_block *sb)
9261 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9262 struct btrfs_inode *ei;
9263 struct inode *inode;
9265 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9272 ei->last_sub_trans = 0;
9273 ei->logged_trans = 0;
9274 ei->delalloc_bytes = 0;
9275 ei->new_delalloc_bytes = 0;
9276 ei->defrag_bytes = 0;
9277 ei->disk_i_size = 0;
9280 ei->index_cnt = (u64)-1;
9282 ei->last_unlink_trans = 0;
9283 ei->last_log_commit = 0;
9285 spin_lock_init(&ei->lock);
9286 ei->outstanding_extents = 0;
9287 if (sb->s_magic != BTRFS_TEST_MAGIC)
9288 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9289 BTRFS_BLOCK_RSV_DELALLOC);
9290 ei->runtime_flags = 0;
9291 ei->prop_compress = BTRFS_COMPRESS_NONE;
9292 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9294 ei->delayed_node = NULL;
9296 ei->i_otime.tv_sec = 0;
9297 ei->i_otime.tv_nsec = 0;
9299 inode = &ei->vfs_inode;
9300 extent_map_tree_init(&ei->extent_tree);
9301 extent_io_tree_init(&ei->io_tree, inode);
9302 extent_io_tree_init(&ei->io_failure_tree, inode);
9303 ei->io_tree.track_uptodate = 1;
9304 ei->io_failure_tree.track_uptodate = 1;
9305 atomic_set(&ei->sync_writers, 0);
9306 mutex_init(&ei->log_mutex);
9307 mutex_init(&ei->delalloc_mutex);
9308 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9309 INIT_LIST_HEAD(&ei->delalloc_inodes);
9310 INIT_LIST_HEAD(&ei->delayed_iput);
9311 RB_CLEAR_NODE(&ei->rb_node);
9312 init_rwsem(&ei->dio_sem);
9317 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9318 void btrfs_test_destroy_inode(struct inode *inode)
9320 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9321 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9325 static void btrfs_i_callback(struct rcu_head *head)
9327 struct inode *inode = container_of(head, struct inode, i_rcu);
9328 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9331 void btrfs_destroy_inode(struct inode *inode)
9333 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9334 struct btrfs_ordered_extent *ordered;
9335 struct btrfs_root *root = BTRFS_I(inode)->root;
9337 WARN_ON(!hlist_empty(&inode->i_dentry));
9338 WARN_ON(inode->i_data.nrpages);
9339 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9340 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9341 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9342 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9343 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9344 WARN_ON(BTRFS_I(inode)->csum_bytes);
9345 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9348 * This can happen where we create an inode, but somebody else also
9349 * created the same inode and we need to destroy the one we already
9355 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9356 &BTRFS_I(inode)->runtime_flags)) {
9357 btrfs_info(fs_info, "inode %llu still on the orphan list",
9358 btrfs_ino(BTRFS_I(inode)));
9359 atomic_dec(&root->orphan_inodes);
9363 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9368 "found ordered extent %llu %llu on inode cleanup",
9369 ordered->file_offset, ordered->len);
9370 btrfs_remove_ordered_extent(inode, ordered);
9371 btrfs_put_ordered_extent(ordered);
9372 btrfs_put_ordered_extent(ordered);
9375 btrfs_qgroup_check_reserved_leak(inode);
9376 inode_tree_del(inode);
9377 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9379 call_rcu(&inode->i_rcu, btrfs_i_callback);
9382 int btrfs_drop_inode(struct inode *inode)
9384 struct btrfs_root *root = BTRFS_I(inode)->root;
9389 /* the snap/subvol tree is on deleting */
9390 if (btrfs_root_refs(&root->root_item) == 0)
9393 return generic_drop_inode(inode);
9396 static void init_once(void *foo)
9398 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9400 inode_init_once(&ei->vfs_inode);
9403 void __cold btrfs_destroy_cachep(void)
9406 * Make sure all delayed rcu free inodes are flushed before we
9410 kmem_cache_destroy(btrfs_inode_cachep);
9411 kmem_cache_destroy(btrfs_trans_handle_cachep);
9412 kmem_cache_destroy(btrfs_path_cachep);
9413 kmem_cache_destroy(btrfs_free_space_cachep);
9416 int __init btrfs_init_cachep(void)
9418 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9419 sizeof(struct btrfs_inode), 0,
9420 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9422 if (!btrfs_inode_cachep)
9425 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9426 sizeof(struct btrfs_trans_handle), 0,
9427 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9428 if (!btrfs_trans_handle_cachep)
9431 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9432 sizeof(struct btrfs_path), 0,
9433 SLAB_MEM_SPREAD, NULL);
9434 if (!btrfs_path_cachep)
9437 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9438 sizeof(struct btrfs_free_space), 0,
9439 SLAB_MEM_SPREAD, NULL);
9440 if (!btrfs_free_space_cachep)
9445 btrfs_destroy_cachep();
9449 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9450 u32 request_mask, unsigned int flags)
9453 struct inode *inode = d_inode(path->dentry);
9454 u32 blocksize = inode->i_sb->s_blocksize;
9455 u32 bi_flags = BTRFS_I(inode)->flags;
9457 stat->result_mask |= STATX_BTIME;
9458 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9459 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9460 if (bi_flags & BTRFS_INODE_APPEND)
9461 stat->attributes |= STATX_ATTR_APPEND;
9462 if (bi_flags & BTRFS_INODE_COMPRESS)
9463 stat->attributes |= STATX_ATTR_COMPRESSED;
9464 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9465 stat->attributes |= STATX_ATTR_IMMUTABLE;
9466 if (bi_flags & BTRFS_INODE_NODUMP)
9467 stat->attributes |= STATX_ATTR_NODUMP;
9469 stat->attributes_mask |= (STATX_ATTR_APPEND |
9470 STATX_ATTR_COMPRESSED |
9471 STATX_ATTR_IMMUTABLE |
9474 generic_fillattr(inode, stat);
9475 stat->dev = BTRFS_I(inode)->root->anon_dev;
9477 spin_lock(&BTRFS_I(inode)->lock);
9478 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9479 spin_unlock(&BTRFS_I(inode)->lock);
9480 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9481 ALIGN(delalloc_bytes, blocksize)) >> 9;
9485 static int btrfs_rename_exchange(struct inode *old_dir,
9486 struct dentry *old_dentry,
9487 struct inode *new_dir,
9488 struct dentry *new_dentry)
9490 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9491 struct btrfs_trans_handle *trans;
9492 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9493 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9494 struct inode *new_inode = new_dentry->d_inode;
9495 struct inode *old_inode = old_dentry->d_inode;
9496 struct timespec ctime = current_time(old_inode);
9497 struct dentry *parent;
9498 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9499 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9504 bool root_log_pinned = false;
9505 bool dest_log_pinned = false;
9507 /* we only allow rename subvolume link between subvolumes */
9508 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9511 /* close the race window with snapshot create/destroy ioctl */
9512 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9513 down_read(&fs_info->subvol_sem);
9514 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9515 down_read(&fs_info->subvol_sem);
9518 * We want to reserve the absolute worst case amount of items. So if
9519 * both inodes are subvols and we need to unlink them then that would
9520 * require 4 item modifications, but if they are both normal inodes it
9521 * would require 5 item modifications, so we'll assume their normal
9522 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9523 * should cover the worst case number of items we'll modify.
9525 trans = btrfs_start_transaction(root, 12);
9526 if (IS_ERR(trans)) {
9527 ret = PTR_ERR(trans);
9532 * We need to find a free sequence number both in the source and
9533 * in the destination directory for the exchange.
9535 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9538 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9542 BTRFS_I(old_inode)->dir_index = 0ULL;
9543 BTRFS_I(new_inode)->dir_index = 0ULL;
9545 /* Reference for the source. */
9546 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9547 /* force full log commit if subvolume involved. */
9548 btrfs_set_log_full_commit(fs_info, trans);
9550 btrfs_pin_log_trans(root);
9551 root_log_pinned = true;
9552 ret = btrfs_insert_inode_ref(trans, dest,
9553 new_dentry->d_name.name,
9554 new_dentry->d_name.len,
9556 btrfs_ino(BTRFS_I(new_dir)),
9562 /* And now for the dest. */
9563 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9564 /* force full log commit if subvolume involved. */
9565 btrfs_set_log_full_commit(fs_info, trans);
9567 btrfs_pin_log_trans(dest);
9568 dest_log_pinned = true;
9569 ret = btrfs_insert_inode_ref(trans, root,
9570 old_dentry->d_name.name,
9571 old_dentry->d_name.len,
9573 btrfs_ino(BTRFS_I(old_dir)),
9579 /* Update inode version and ctime/mtime. */
9580 inode_inc_iversion(old_dir);
9581 inode_inc_iversion(new_dir);
9582 inode_inc_iversion(old_inode);
9583 inode_inc_iversion(new_inode);
9584 old_dir->i_ctime = old_dir->i_mtime = ctime;
9585 new_dir->i_ctime = new_dir->i_mtime = ctime;
9586 old_inode->i_ctime = ctime;
9587 new_inode->i_ctime = ctime;
9589 if (old_dentry->d_parent != new_dentry->d_parent) {
9590 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9591 BTRFS_I(old_inode), 1);
9592 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9593 BTRFS_I(new_inode), 1);
9596 /* src is a subvolume */
9597 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9598 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9599 ret = btrfs_unlink_subvol(trans, root, old_dir,
9601 old_dentry->d_name.name,
9602 old_dentry->d_name.len);
9603 } else { /* src is an inode */
9604 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9605 BTRFS_I(old_dentry->d_inode),
9606 old_dentry->d_name.name,
9607 old_dentry->d_name.len);
9609 ret = btrfs_update_inode(trans, root, old_inode);
9612 btrfs_abort_transaction(trans, ret);
9616 /* dest is a subvolume */
9617 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9618 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9619 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9621 new_dentry->d_name.name,
9622 new_dentry->d_name.len);
9623 } else { /* dest is an inode */
9624 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9625 BTRFS_I(new_dentry->d_inode),
9626 new_dentry->d_name.name,
9627 new_dentry->d_name.len);
9629 ret = btrfs_update_inode(trans, dest, new_inode);
9632 btrfs_abort_transaction(trans, ret);
9636 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9637 new_dentry->d_name.name,
9638 new_dentry->d_name.len, 0, old_idx);
9640 btrfs_abort_transaction(trans, ret);
9644 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9645 old_dentry->d_name.name,
9646 old_dentry->d_name.len, 0, new_idx);
9648 btrfs_abort_transaction(trans, ret);
9652 if (old_inode->i_nlink == 1)
9653 BTRFS_I(old_inode)->dir_index = old_idx;
9654 if (new_inode->i_nlink == 1)
9655 BTRFS_I(new_inode)->dir_index = new_idx;
9657 if (root_log_pinned) {
9658 parent = new_dentry->d_parent;
9659 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9661 btrfs_end_log_trans(root);
9662 root_log_pinned = false;
9664 if (dest_log_pinned) {
9665 parent = old_dentry->d_parent;
9666 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9668 btrfs_end_log_trans(dest);
9669 dest_log_pinned = false;
9673 * If we have pinned a log and an error happened, we unpin tasks
9674 * trying to sync the log and force them to fallback to a transaction
9675 * commit if the log currently contains any of the inodes involved in
9676 * this rename operation (to ensure we do not persist a log with an
9677 * inconsistent state for any of these inodes or leading to any
9678 * inconsistencies when replayed). If the transaction was aborted, the
9679 * abortion reason is propagated to userspace when attempting to commit
9680 * the transaction. If the log does not contain any of these inodes, we
9681 * allow the tasks to sync it.
9683 if (ret && (root_log_pinned || dest_log_pinned)) {
9684 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9685 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9686 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9688 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9689 btrfs_set_log_full_commit(fs_info, trans);
9691 if (root_log_pinned) {
9692 btrfs_end_log_trans(root);
9693 root_log_pinned = false;
9695 if (dest_log_pinned) {
9696 btrfs_end_log_trans(dest);
9697 dest_log_pinned = false;
9700 ret = btrfs_end_transaction(trans);
9702 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9703 up_read(&fs_info->subvol_sem);
9704 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9705 up_read(&fs_info->subvol_sem);
9710 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9711 struct btrfs_root *root,
9713 struct dentry *dentry)
9716 struct inode *inode;
9720 ret = btrfs_find_free_ino(root, &objectid);
9724 inode = btrfs_new_inode(trans, root, dir,
9725 dentry->d_name.name,
9727 btrfs_ino(BTRFS_I(dir)),
9729 S_IFCHR | WHITEOUT_MODE,
9732 if (IS_ERR(inode)) {
9733 ret = PTR_ERR(inode);
9737 inode->i_op = &btrfs_special_inode_operations;
9738 init_special_inode(inode, inode->i_mode,
9741 ret = btrfs_init_inode_security(trans, inode, dir,
9746 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9747 BTRFS_I(inode), 0, index);
9751 ret = btrfs_update_inode(trans, root, inode);
9753 unlock_new_inode(inode);
9755 inode_dec_link_count(inode);
9761 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9762 struct inode *new_dir, struct dentry *new_dentry,
9765 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9766 struct btrfs_trans_handle *trans;
9767 unsigned int trans_num_items;
9768 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9769 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9770 struct inode *new_inode = d_inode(new_dentry);
9771 struct inode *old_inode = d_inode(old_dentry);
9775 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9776 bool log_pinned = false;
9778 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9781 /* we only allow rename subvolume link between subvolumes */
9782 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9785 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9786 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9789 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9790 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9794 /* check for collisions, even if the name isn't there */
9795 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9796 new_dentry->d_name.name,
9797 new_dentry->d_name.len);
9800 if (ret == -EEXIST) {
9802 * eexist without a new_inode */
9803 if (WARN_ON(!new_inode)) {
9807 /* maybe -EOVERFLOW */
9814 * we're using rename to replace one file with another. Start IO on it
9815 * now so we don't add too much work to the end of the transaction
9817 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9818 filemap_flush(old_inode->i_mapping);
9820 /* close the racy window with snapshot create/destroy ioctl */
9821 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9822 down_read(&fs_info->subvol_sem);
9824 * We want to reserve the absolute worst case amount of items. So if
9825 * both inodes are subvols and we need to unlink them then that would
9826 * require 4 item modifications, but if they are both normal inodes it
9827 * would require 5 item modifications, so we'll assume they are normal
9828 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9829 * should cover the worst case number of items we'll modify.
9830 * If our rename has the whiteout flag, we need more 5 units for the
9831 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9832 * when selinux is enabled).
9834 trans_num_items = 11;
9835 if (flags & RENAME_WHITEOUT)
9836 trans_num_items += 5;
9837 trans = btrfs_start_transaction(root, trans_num_items);
9838 if (IS_ERR(trans)) {
9839 ret = PTR_ERR(trans);
9844 btrfs_record_root_in_trans(trans, dest);
9846 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9850 BTRFS_I(old_inode)->dir_index = 0ULL;
9851 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9852 /* force full log commit if subvolume involved. */
9853 btrfs_set_log_full_commit(fs_info, trans);
9855 btrfs_pin_log_trans(root);
9857 ret = btrfs_insert_inode_ref(trans, dest,
9858 new_dentry->d_name.name,
9859 new_dentry->d_name.len,
9861 btrfs_ino(BTRFS_I(new_dir)), index);
9866 inode_inc_iversion(old_dir);
9867 inode_inc_iversion(new_dir);
9868 inode_inc_iversion(old_inode);
9869 old_dir->i_ctime = old_dir->i_mtime =
9870 new_dir->i_ctime = new_dir->i_mtime =
9871 old_inode->i_ctime = current_time(old_dir);
9873 if (old_dentry->d_parent != new_dentry->d_parent)
9874 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9875 BTRFS_I(old_inode), 1);
9877 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9878 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9879 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9880 old_dentry->d_name.name,
9881 old_dentry->d_name.len);
9883 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9884 BTRFS_I(d_inode(old_dentry)),
9885 old_dentry->d_name.name,
9886 old_dentry->d_name.len);
9888 ret = btrfs_update_inode(trans, root, old_inode);
9891 btrfs_abort_transaction(trans, ret);
9896 inode_inc_iversion(new_inode);
9897 new_inode->i_ctime = current_time(new_inode);
9898 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9899 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9900 root_objectid = BTRFS_I(new_inode)->location.objectid;
9901 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9903 new_dentry->d_name.name,
9904 new_dentry->d_name.len);
9905 BUG_ON(new_inode->i_nlink == 0);
9907 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9908 BTRFS_I(d_inode(new_dentry)),
9909 new_dentry->d_name.name,
9910 new_dentry->d_name.len);
9912 if (!ret && new_inode->i_nlink == 0)
9913 ret = btrfs_orphan_add(trans,
9914 BTRFS_I(d_inode(new_dentry)));
9916 btrfs_abort_transaction(trans, ret);
9921 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9922 new_dentry->d_name.name,
9923 new_dentry->d_name.len, 0, index);
9925 btrfs_abort_transaction(trans, ret);
9929 if (old_inode->i_nlink == 1)
9930 BTRFS_I(old_inode)->dir_index = index;
9933 struct dentry *parent = new_dentry->d_parent;
9935 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9937 btrfs_end_log_trans(root);
9941 if (flags & RENAME_WHITEOUT) {
9942 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9946 btrfs_abort_transaction(trans, ret);
9952 * If we have pinned the log and an error happened, we unpin tasks
9953 * trying to sync the log and force them to fallback to a transaction
9954 * commit if the log currently contains any of the inodes involved in
9955 * this rename operation (to ensure we do not persist a log with an
9956 * inconsistent state for any of these inodes or leading to any
9957 * inconsistencies when replayed). If the transaction was aborted, the
9958 * abortion reason is propagated to userspace when attempting to commit
9959 * the transaction. If the log does not contain any of these inodes, we
9960 * allow the tasks to sync it.
9962 if (ret && log_pinned) {
9963 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9964 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9965 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9967 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9968 btrfs_set_log_full_commit(fs_info, trans);
9970 btrfs_end_log_trans(root);
9973 btrfs_end_transaction(trans);
9975 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9976 up_read(&fs_info->subvol_sem);
9981 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9982 struct inode *new_dir, struct dentry *new_dentry,
9985 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9988 if (flags & RENAME_EXCHANGE)
9989 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9992 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9995 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9997 struct btrfs_delalloc_work *delalloc_work;
9998 struct inode *inode;
10000 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10002 inode = delalloc_work->inode;
10003 filemap_flush(inode->i_mapping);
10004 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10005 &BTRFS_I(inode)->runtime_flags))
10006 filemap_flush(inode->i_mapping);
10008 if (delalloc_work->delay_iput)
10009 btrfs_add_delayed_iput(inode);
10012 complete(&delalloc_work->completion);
10015 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10018 struct btrfs_delalloc_work *work;
10020 work = kmalloc(sizeof(*work), GFP_NOFS);
10024 init_completion(&work->completion);
10025 INIT_LIST_HEAD(&work->list);
10026 work->inode = inode;
10027 work->delay_iput = delay_iput;
10028 WARN_ON_ONCE(!inode);
10029 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10030 btrfs_run_delalloc_work, NULL, NULL);
10035 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10037 wait_for_completion(&work->completion);
10042 * some fairly slow code that needs optimization. This walks the list
10043 * of all the inodes with pending delalloc and forces them to disk.
10045 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10048 struct btrfs_inode *binode;
10049 struct inode *inode;
10050 struct btrfs_delalloc_work *work, *next;
10051 struct list_head works;
10052 struct list_head splice;
10055 INIT_LIST_HEAD(&works);
10056 INIT_LIST_HEAD(&splice);
10058 mutex_lock(&root->delalloc_mutex);
10059 spin_lock(&root->delalloc_lock);
10060 list_splice_init(&root->delalloc_inodes, &splice);
10061 while (!list_empty(&splice)) {
10062 binode = list_entry(splice.next, struct btrfs_inode,
10065 list_move_tail(&binode->delalloc_inodes,
10066 &root->delalloc_inodes);
10067 inode = igrab(&binode->vfs_inode);
10069 cond_resched_lock(&root->delalloc_lock);
10072 spin_unlock(&root->delalloc_lock);
10074 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10077 btrfs_add_delayed_iput(inode);
10083 list_add_tail(&work->list, &works);
10084 btrfs_queue_work(root->fs_info->flush_workers,
10087 if (nr != -1 && ret >= nr)
10090 spin_lock(&root->delalloc_lock);
10092 spin_unlock(&root->delalloc_lock);
10095 list_for_each_entry_safe(work, next, &works, list) {
10096 list_del_init(&work->list);
10097 btrfs_wait_and_free_delalloc_work(work);
10100 if (!list_empty_careful(&splice)) {
10101 spin_lock(&root->delalloc_lock);
10102 list_splice_tail(&splice, &root->delalloc_inodes);
10103 spin_unlock(&root->delalloc_lock);
10105 mutex_unlock(&root->delalloc_mutex);
10109 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10111 struct btrfs_fs_info *fs_info = root->fs_info;
10114 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10117 ret = __start_delalloc_inodes(root, delay_iput, -1);
10123 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10126 struct btrfs_root *root;
10127 struct list_head splice;
10130 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10133 INIT_LIST_HEAD(&splice);
10135 mutex_lock(&fs_info->delalloc_root_mutex);
10136 spin_lock(&fs_info->delalloc_root_lock);
10137 list_splice_init(&fs_info->delalloc_roots, &splice);
10138 while (!list_empty(&splice) && nr) {
10139 root = list_first_entry(&splice, struct btrfs_root,
10141 root = btrfs_grab_fs_root(root);
10143 list_move_tail(&root->delalloc_root,
10144 &fs_info->delalloc_roots);
10145 spin_unlock(&fs_info->delalloc_root_lock);
10147 ret = __start_delalloc_inodes(root, delay_iput, nr);
10148 btrfs_put_fs_root(root);
10156 spin_lock(&fs_info->delalloc_root_lock);
10158 spin_unlock(&fs_info->delalloc_root_lock);
10162 if (!list_empty_careful(&splice)) {
10163 spin_lock(&fs_info->delalloc_root_lock);
10164 list_splice_tail(&splice, &fs_info->delalloc_roots);
10165 spin_unlock(&fs_info->delalloc_root_lock);
10167 mutex_unlock(&fs_info->delalloc_root_mutex);
10171 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10172 const char *symname)
10174 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10175 struct btrfs_trans_handle *trans;
10176 struct btrfs_root *root = BTRFS_I(dir)->root;
10177 struct btrfs_path *path;
10178 struct btrfs_key key;
10179 struct inode *inode = NULL;
10181 int drop_inode = 0;
10187 struct btrfs_file_extent_item *ei;
10188 struct extent_buffer *leaf;
10190 name_len = strlen(symname);
10191 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10192 return -ENAMETOOLONG;
10195 * 2 items for inode item and ref
10196 * 2 items for dir items
10197 * 1 item for updating parent inode item
10198 * 1 item for the inline extent item
10199 * 1 item for xattr if selinux is on
10201 trans = btrfs_start_transaction(root, 7);
10203 return PTR_ERR(trans);
10205 err = btrfs_find_free_ino(root, &objectid);
10209 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10210 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10211 objectid, S_IFLNK|S_IRWXUGO, &index);
10212 if (IS_ERR(inode)) {
10213 err = PTR_ERR(inode);
10218 * If the active LSM wants to access the inode during
10219 * d_instantiate it needs these. Smack checks to see
10220 * if the filesystem supports xattrs by looking at the
10223 inode->i_fop = &btrfs_file_operations;
10224 inode->i_op = &btrfs_file_inode_operations;
10225 inode->i_mapping->a_ops = &btrfs_aops;
10226 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10228 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10230 goto out_unlock_inode;
10232 path = btrfs_alloc_path();
10235 goto out_unlock_inode;
10237 key.objectid = btrfs_ino(BTRFS_I(inode));
10239 key.type = BTRFS_EXTENT_DATA_KEY;
10240 datasize = btrfs_file_extent_calc_inline_size(name_len);
10241 err = btrfs_insert_empty_item(trans, root, path, &key,
10244 btrfs_free_path(path);
10245 goto out_unlock_inode;
10247 leaf = path->nodes[0];
10248 ei = btrfs_item_ptr(leaf, path->slots[0],
10249 struct btrfs_file_extent_item);
10250 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10251 btrfs_set_file_extent_type(leaf, ei,
10252 BTRFS_FILE_EXTENT_INLINE);
10253 btrfs_set_file_extent_encryption(leaf, ei, 0);
10254 btrfs_set_file_extent_compression(leaf, ei, 0);
10255 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10256 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10258 ptr = btrfs_file_extent_inline_start(ei);
10259 write_extent_buffer(leaf, symname, ptr, name_len);
10260 btrfs_mark_buffer_dirty(leaf);
10261 btrfs_free_path(path);
10263 inode->i_op = &btrfs_symlink_inode_operations;
10264 inode_nohighmem(inode);
10265 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10266 inode_set_bytes(inode, name_len);
10267 btrfs_i_size_write(BTRFS_I(inode), name_len);
10268 err = btrfs_update_inode(trans, root, inode);
10270 * Last step, add directory indexes for our symlink inode. This is the
10271 * last step to avoid extra cleanup of these indexes if an error happens
10275 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10276 BTRFS_I(inode), 0, index);
10279 goto out_unlock_inode;
10282 unlock_new_inode(inode);
10283 d_instantiate(dentry, inode);
10286 btrfs_end_transaction(trans);
10288 inode_dec_link_count(inode);
10291 btrfs_btree_balance_dirty(fs_info);
10296 unlock_new_inode(inode);
10300 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10301 u64 start, u64 num_bytes, u64 min_size,
10302 loff_t actual_len, u64 *alloc_hint,
10303 struct btrfs_trans_handle *trans)
10305 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10306 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10307 struct extent_map *em;
10308 struct btrfs_root *root = BTRFS_I(inode)->root;
10309 struct btrfs_key ins;
10310 u64 cur_offset = start;
10313 u64 last_alloc = (u64)-1;
10315 bool own_trans = true;
10316 u64 end = start + num_bytes - 1;
10320 while (num_bytes > 0) {
10322 trans = btrfs_start_transaction(root, 3);
10323 if (IS_ERR(trans)) {
10324 ret = PTR_ERR(trans);
10329 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10330 cur_bytes = max(cur_bytes, min_size);
10332 * If we are severely fragmented we could end up with really
10333 * small allocations, so if the allocator is returning small
10334 * chunks lets make its job easier by only searching for those
10337 cur_bytes = min(cur_bytes, last_alloc);
10338 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10339 min_size, 0, *alloc_hint, &ins, 1, 0);
10342 btrfs_end_transaction(trans);
10345 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10347 last_alloc = ins.offset;
10348 ret = insert_reserved_file_extent(trans, inode,
10349 cur_offset, ins.objectid,
10350 ins.offset, ins.offset,
10351 ins.offset, 0, 0, 0,
10352 BTRFS_FILE_EXTENT_PREALLOC);
10354 btrfs_free_reserved_extent(fs_info, ins.objectid,
10356 btrfs_abort_transaction(trans, ret);
10358 btrfs_end_transaction(trans);
10362 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10363 cur_offset + ins.offset -1, 0);
10365 em = alloc_extent_map();
10367 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10368 &BTRFS_I(inode)->runtime_flags);
10372 em->start = cur_offset;
10373 em->orig_start = cur_offset;
10374 em->len = ins.offset;
10375 em->block_start = ins.objectid;
10376 em->block_len = ins.offset;
10377 em->orig_block_len = ins.offset;
10378 em->ram_bytes = ins.offset;
10379 em->bdev = fs_info->fs_devices->latest_bdev;
10380 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10381 em->generation = trans->transid;
10384 write_lock(&em_tree->lock);
10385 ret = add_extent_mapping(em_tree, em, 1);
10386 write_unlock(&em_tree->lock);
10387 if (ret != -EEXIST)
10389 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10390 cur_offset + ins.offset - 1,
10393 free_extent_map(em);
10395 num_bytes -= ins.offset;
10396 cur_offset += ins.offset;
10397 *alloc_hint = ins.objectid + ins.offset;
10399 inode_inc_iversion(inode);
10400 inode->i_ctime = current_time(inode);
10401 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10402 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10403 (actual_len > inode->i_size) &&
10404 (cur_offset > inode->i_size)) {
10405 if (cur_offset > actual_len)
10406 i_size = actual_len;
10408 i_size = cur_offset;
10409 i_size_write(inode, i_size);
10410 btrfs_ordered_update_i_size(inode, i_size, NULL);
10413 ret = btrfs_update_inode(trans, root, inode);
10416 btrfs_abort_transaction(trans, ret);
10418 btrfs_end_transaction(trans);
10423 btrfs_end_transaction(trans);
10425 if (cur_offset < end)
10426 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10427 end - cur_offset + 1);
10431 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10432 u64 start, u64 num_bytes, u64 min_size,
10433 loff_t actual_len, u64 *alloc_hint)
10435 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10436 min_size, actual_len, alloc_hint,
10440 int btrfs_prealloc_file_range_trans(struct inode *inode,
10441 struct btrfs_trans_handle *trans, int mode,
10442 u64 start, u64 num_bytes, u64 min_size,
10443 loff_t actual_len, u64 *alloc_hint)
10445 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10446 min_size, actual_len, alloc_hint, trans);
10449 static int btrfs_set_page_dirty(struct page *page)
10451 return __set_page_dirty_nobuffers(page);
10454 static int btrfs_permission(struct inode *inode, int mask)
10456 struct btrfs_root *root = BTRFS_I(inode)->root;
10457 umode_t mode = inode->i_mode;
10459 if (mask & MAY_WRITE &&
10460 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10461 if (btrfs_root_readonly(root))
10463 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10466 return generic_permission(inode, mask);
10469 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10471 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10472 struct btrfs_trans_handle *trans;
10473 struct btrfs_root *root = BTRFS_I(dir)->root;
10474 struct inode *inode = NULL;
10480 * 5 units required for adding orphan entry
10482 trans = btrfs_start_transaction(root, 5);
10484 return PTR_ERR(trans);
10486 ret = btrfs_find_free_ino(root, &objectid);
10490 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10491 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10492 if (IS_ERR(inode)) {
10493 ret = PTR_ERR(inode);
10498 inode->i_fop = &btrfs_file_operations;
10499 inode->i_op = &btrfs_file_inode_operations;
10501 inode->i_mapping->a_ops = &btrfs_aops;
10502 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10504 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10508 ret = btrfs_update_inode(trans, root, inode);
10511 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10516 * We set number of links to 0 in btrfs_new_inode(), and here we set
10517 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10520 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10522 set_nlink(inode, 1);
10523 unlock_new_inode(inode);
10524 d_tmpfile(dentry, inode);
10525 mark_inode_dirty(inode);
10528 btrfs_end_transaction(trans);
10531 btrfs_btree_balance_dirty(fs_info);
10535 unlock_new_inode(inode);
10540 __attribute__((const))
10541 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10546 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10548 struct inode *inode = private_data;
10549 return btrfs_sb(inode->i_sb);
10552 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10553 u64 start, u64 end)
10555 struct inode *inode = private_data;
10558 isize = i_size_read(inode);
10559 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10560 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10561 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10562 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10566 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10568 struct inode *inode = private_data;
10569 unsigned long index = start >> PAGE_SHIFT;
10570 unsigned long end_index = end >> PAGE_SHIFT;
10573 while (index <= end_index) {
10574 page = find_get_page(inode->i_mapping, index);
10575 ASSERT(page); /* Pages should be in the extent_io_tree */
10576 set_page_writeback(page);
10582 static const struct inode_operations btrfs_dir_inode_operations = {
10583 .getattr = btrfs_getattr,
10584 .lookup = btrfs_lookup,
10585 .create = btrfs_create,
10586 .unlink = btrfs_unlink,
10587 .link = btrfs_link,
10588 .mkdir = btrfs_mkdir,
10589 .rmdir = btrfs_rmdir,
10590 .rename = btrfs_rename2,
10591 .symlink = btrfs_symlink,
10592 .setattr = btrfs_setattr,
10593 .mknod = btrfs_mknod,
10594 .listxattr = btrfs_listxattr,
10595 .permission = btrfs_permission,
10596 .get_acl = btrfs_get_acl,
10597 .set_acl = btrfs_set_acl,
10598 .update_time = btrfs_update_time,
10599 .tmpfile = btrfs_tmpfile,
10601 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10602 .lookup = btrfs_lookup,
10603 .permission = btrfs_permission,
10604 .update_time = btrfs_update_time,
10607 static const struct file_operations btrfs_dir_file_operations = {
10608 .llseek = generic_file_llseek,
10609 .read = generic_read_dir,
10610 .iterate_shared = btrfs_real_readdir,
10611 .open = btrfs_opendir,
10612 .unlocked_ioctl = btrfs_ioctl,
10613 #ifdef CONFIG_COMPAT
10614 .compat_ioctl = btrfs_compat_ioctl,
10616 .release = btrfs_release_file,
10617 .fsync = btrfs_sync_file,
10620 static const struct extent_io_ops btrfs_extent_io_ops = {
10621 /* mandatory callbacks */
10622 .submit_bio_hook = btrfs_submit_bio_hook,
10623 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10624 .merge_bio_hook = btrfs_merge_bio_hook,
10625 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10626 .tree_fs_info = iotree_fs_info,
10627 .set_range_writeback = btrfs_set_range_writeback,
10629 /* optional callbacks */
10630 .fill_delalloc = run_delalloc_range,
10631 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10632 .writepage_start_hook = btrfs_writepage_start_hook,
10633 .set_bit_hook = btrfs_set_bit_hook,
10634 .clear_bit_hook = btrfs_clear_bit_hook,
10635 .merge_extent_hook = btrfs_merge_extent_hook,
10636 .split_extent_hook = btrfs_split_extent_hook,
10637 .check_extent_io_range = btrfs_check_extent_io_range,
10641 * btrfs doesn't support the bmap operation because swapfiles
10642 * use bmap to make a mapping of extents in the file. They assume
10643 * these extents won't change over the life of the file and they
10644 * use the bmap result to do IO directly to the drive.
10646 * the btrfs bmap call would return logical addresses that aren't
10647 * suitable for IO and they also will change frequently as COW
10648 * operations happen. So, swapfile + btrfs == corruption.
10650 * For now we're avoiding this by dropping bmap.
10652 static const struct address_space_operations btrfs_aops = {
10653 .readpage = btrfs_readpage,
10654 .writepage = btrfs_writepage,
10655 .writepages = btrfs_writepages,
10656 .readpages = btrfs_readpages,
10657 .direct_IO = btrfs_direct_IO,
10658 .invalidatepage = btrfs_invalidatepage,
10659 .releasepage = btrfs_releasepage,
10660 .set_page_dirty = btrfs_set_page_dirty,
10661 .error_remove_page = generic_error_remove_page,
10664 static const struct address_space_operations btrfs_symlink_aops = {
10665 .readpage = btrfs_readpage,
10666 .writepage = btrfs_writepage,
10667 .invalidatepage = btrfs_invalidatepage,
10668 .releasepage = btrfs_releasepage,
10671 static const struct inode_operations btrfs_file_inode_operations = {
10672 .getattr = btrfs_getattr,
10673 .setattr = btrfs_setattr,
10674 .listxattr = btrfs_listxattr,
10675 .permission = btrfs_permission,
10676 .fiemap = btrfs_fiemap,
10677 .get_acl = btrfs_get_acl,
10678 .set_acl = btrfs_set_acl,
10679 .update_time = btrfs_update_time,
10681 static const struct inode_operations btrfs_special_inode_operations = {
10682 .getattr = btrfs_getattr,
10683 .setattr = btrfs_setattr,
10684 .permission = btrfs_permission,
10685 .listxattr = btrfs_listxattr,
10686 .get_acl = btrfs_get_acl,
10687 .set_acl = btrfs_set_acl,
10688 .update_time = btrfs_update_time,
10690 static const struct inode_operations btrfs_symlink_inode_operations = {
10691 .get_link = page_get_link,
10692 .getattr = btrfs_getattr,
10693 .setattr = btrfs_setattr,
10694 .permission = btrfs_permission,
10695 .listxattr = btrfs_listxattr,
10696 .update_time = btrfs_update_time,
10699 const struct dentry_operations btrfs_dentry_operations = {
10700 .d_delete = btrfs_dentry_delete,
10701 .d_release = btrfs_dentry_release,