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,
1945 int mirror_num, unsigned long bio_flags,
1948 struct inode *inode = private_data;
1949 blk_status_t ret = 0;
1951 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1952 BUG_ON(ret); /* -ENOMEM */
1957 * in order to insert checksums into the metadata in large chunks,
1958 * we wait until bio submission time. All the pages in the bio are
1959 * checksummed and sums are attached onto the ordered extent record.
1961 * At IO completion time the cums attached on the ordered extent record
1962 * are inserted into the btree
1964 static blk_status_t __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1965 int mirror_num, unsigned long bio_flags,
1968 struct inode *inode = private_data;
1969 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1972 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1974 bio->bi_status = ret;
1981 * extent_io.c submission hook. This does the right thing for csum calculation
1982 * on write, or reading the csums from the tree before a read.
1984 * Rules about async/sync submit,
1985 * a) read: sync submit
1987 * b) write without checksum: sync submit
1989 * c) write with checksum:
1990 * c-1) if bio is issued by fsync: sync submit
1991 * (sync_writers != 0)
1993 * c-2) if root is reloc root: sync submit
1994 * (only in case of buffered IO)
1996 * c-3) otherwise: async submit
1998 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1999 int mirror_num, unsigned long bio_flags,
2002 struct inode *inode = private_data;
2003 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2004 struct btrfs_root *root = BTRFS_I(inode)->root;
2005 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2006 blk_status_t ret = 0;
2008 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2010 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2012 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2013 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2015 if (bio_op(bio) != REQ_OP_WRITE) {
2016 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2020 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2021 ret = btrfs_submit_compressed_read(inode, bio,
2025 } else if (!skip_sum) {
2026 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2031 } else if (async && !skip_sum) {
2032 /* csum items have already been cloned */
2033 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2035 /* we're doing a write, do the async checksumming */
2036 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2038 __btrfs_submit_bio_start,
2039 __btrfs_submit_bio_done);
2041 } else if (!skip_sum) {
2042 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2048 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2052 bio->bi_status = ret;
2059 * given a list of ordered sums record them in the inode. This happens
2060 * at IO completion time based on sums calculated at bio submission time.
2062 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2063 struct inode *inode, struct list_head *list)
2065 struct btrfs_ordered_sum *sum;
2068 list_for_each_entry(sum, list, list) {
2069 trans->adding_csums = true;
2070 ret = btrfs_csum_file_blocks(trans,
2071 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2072 trans->adding_csums = false;
2079 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2080 unsigned int extra_bits,
2081 struct extent_state **cached_state, int dedupe)
2083 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2084 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2085 extra_bits, cached_state);
2088 /* see btrfs_writepage_start_hook for details on why this is required */
2089 struct btrfs_writepage_fixup {
2091 struct btrfs_work work;
2094 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2096 struct btrfs_writepage_fixup *fixup;
2097 struct btrfs_ordered_extent *ordered;
2098 struct extent_state *cached_state = NULL;
2099 struct extent_changeset *data_reserved = NULL;
2101 struct inode *inode;
2106 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2110 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2111 ClearPageChecked(page);
2115 inode = page->mapping->host;
2116 page_start = page_offset(page);
2117 page_end = page_offset(page) + PAGE_SIZE - 1;
2119 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2122 /* already ordered? We're done */
2123 if (PagePrivate2(page))
2126 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2129 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2130 page_end, &cached_state);
2132 btrfs_start_ordered_extent(inode, ordered, 1);
2133 btrfs_put_ordered_extent(ordered);
2137 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2140 mapping_set_error(page->mapping, ret);
2141 end_extent_writepage(page, ret, page_start, page_end);
2142 ClearPageChecked(page);
2146 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2149 mapping_set_error(page->mapping, ret);
2150 end_extent_writepage(page, ret, page_start, page_end);
2151 ClearPageChecked(page);
2155 ClearPageChecked(page);
2156 set_page_dirty(page);
2157 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
2159 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2165 extent_changeset_free(data_reserved);
2169 * There are a few paths in the higher layers of the kernel that directly
2170 * set the page dirty bit without asking the filesystem if it is a
2171 * good idea. This causes problems because we want to make sure COW
2172 * properly happens and the data=ordered rules are followed.
2174 * In our case any range that doesn't have the ORDERED bit set
2175 * hasn't been properly setup for IO. We kick off an async process
2176 * to fix it up. The async helper will wait for ordered extents, set
2177 * the delalloc bit and make it safe to write the page.
2179 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2181 struct inode *inode = page->mapping->host;
2182 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2183 struct btrfs_writepage_fixup *fixup;
2185 /* this page is properly in the ordered list */
2186 if (TestClearPagePrivate2(page))
2189 if (PageChecked(page))
2192 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2196 SetPageChecked(page);
2198 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2199 btrfs_writepage_fixup_worker, NULL, NULL);
2201 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2205 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2206 struct inode *inode, u64 file_pos,
2207 u64 disk_bytenr, u64 disk_num_bytes,
2208 u64 num_bytes, u64 ram_bytes,
2209 u8 compression, u8 encryption,
2210 u16 other_encoding, int extent_type)
2212 struct btrfs_root *root = BTRFS_I(inode)->root;
2213 struct btrfs_file_extent_item *fi;
2214 struct btrfs_path *path;
2215 struct extent_buffer *leaf;
2216 struct btrfs_key ins;
2218 int extent_inserted = 0;
2221 path = btrfs_alloc_path();
2226 * we may be replacing one extent in the tree with another.
2227 * The new extent is pinned in the extent map, and we don't want
2228 * to drop it from the cache until it is completely in the btree.
2230 * So, tell btrfs_drop_extents to leave this extent in the cache.
2231 * the caller is expected to unpin it and allow it to be merged
2234 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2235 file_pos + num_bytes, NULL, 0,
2236 1, sizeof(*fi), &extent_inserted);
2240 if (!extent_inserted) {
2241 ins.objectid = btrfs_ino(BTRFS_I(inode));
2242 ins.offset = file_pos;
2243 ins.type = BTRFS_EXTENT_DATA_KEY;
2245 path->leave_spinning = 1;
2246 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2251 leaf = path->nodes[0];
2252 fi = btrfs_item_ptr(leaf, path->slots[0],
2253 struct btrfs_file_extent_item);
2254 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2255 btrfs_set_file_extent_type(leaf, fi, extent_type);
2256 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2257 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2258 btrfs_set_file_extent_offset(leaf, fi, 0);
2259 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2260 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2261 btrfs_set_file_extent_compression(leaf, fi, compression);
2262 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2263 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2265 btrfs_mark_buffer_dirty(leaf);
2266 btrfs_release_path(path);
2268 inode_add_bytes(inode, num_bytes);
2270 ins.objectid = disk_bytenr;
2271 ins.offset = disk_num_bytes;
2272 ins.type = BTRFS_EXTENT_ITEM_KEY;
2275 * Release the reserved range from inode dirty range map, as it is
2276 * already moved into delayed_ref_head
2278 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2282 ret = btrfs_alloc_reserved_file_extent(trans, root,
2283 btrfs_ino(BTRFS_I(inode)),
2284 file_pos, qg_released, &ins);
2286 btrfs_free_path(path);
2291 /* snapshot-aware defrag */
2292 struct sa_defrag_extent_backref {
2293 struct rb_node node;
2294 struct old_sa_defrag_extent *old;
2303 struct old_sa_defrag_extent {
2304 struct list_head list;
2305 struct new_sa_defrag_extent *new;
2314 struct new_sa_defrag_extent {
2315 struct rb_root root;
2316 struct list_head head;
2317 struct btrfs_path *path;
2318 struct inode *inode;
2326 static int backref_comp(struct sa_defrag_extent_backref *b1,
2327 struct sa_defrag_extent_backref *b2)
2329 if (b1->root_id < b2->root_id)
2331 else if (b1->root_id > b2->root_id)
2334 if (b1->inum < b2->inum)
2336 else if (b1->inum > b2->inum)
2339 if (b1->file_pos < b2->file_pos)
2341 else if (b1->file_pos > b2->file_pos)
2345 * [------------------------------] ===> (a range of space)
2346 * |<--->| |<---->| =============> (fs/file tree A)
2347 * |<---------------------------->| ===> (fs/file tree B)
2349 * A range of space can refer to two file extents in one tree while
2350 * refer to only one file extent in another tree.
2352 * So we may process a disk offset more than one time(two extents in A)
2353 * and locate at the same extent(one extent in B), then insert two same
2354 * backrefs(both refer to the extent in B).
2359 static void backref_insert(struct rb_root *root,
2360 struct sa_defrag_extent_backref *backref)
2362 struct rb_node **p = &root->rb_node;
2363 struct rb_node *parent = NULL;
2364 struct sa_defrag_extent_backref *entry;
2369 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2371 ret = backref_comp(backref, entry);
2375 p = &(*p)->rb_right;
2378 rb_link_node(&backref->node, parent, p);
2379 rb_insert_color(&backref->node, root);
2383 * Note the backref might has changed, and in this case we just return 0.
2385 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2388 struct btrfs_file_extent_item *extent;
2389 struct old_sa_defrag_extent *old = ctx;
2390 struct new_sa_defrag_extent *new = old->new;
2391 struct btrfs_path *path = new->path;
2392 struct btrfs_key key;
2393 struct btrfs_root *root;
2394 struct sa_defrag_extent_backref *backref;
2395 struct extent_buffer *leaf;
2396 struct inode *inode = new->inode;
2397 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2403 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2404 inum == btrfs_ino(BTRFS_I(inode)))
2407 key.objectid = root_id;
2408 key.type = BTRFS_ROOT_ITEM_KEY;
2409 key.offset = (u64)-1;
2411 root = btrfs_read_fs_root_no_name(fs_info, &key);
2413 if (PTR_ERR(root) == -ENOENT)
2416 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2417 inum, offset, root_id);
2418 return PTR_ERR(root);
2421 key.objectid = inum;
2422 key.type = BTRFS_EXTENT_DATA_KEY;
2423 if (offset > (u64)-1 << 32)
2426 key.offset = offset;
2428 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2429 if (WARN_ON(ret < 0))
2436 leaf = path->nodes[0];
2437 slot = path->slots[0];
2439 if (slot >= btrfs_header_nritems(leaf)) {
2440 ret = btrfs_next_leaf(root, path);
2443 } else if (ret > 0) {
2452 btrfs_item_key_to_cpu(leaf, &key, slot);
2454 if (key.objectid > inum)
2457 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2460 extent = btrfs_item_ptr(leaf, slot,
2461 struct btrfs_file_extent_item);
2463 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2467 * 'offset' refers to the exact key.offset,
2468 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2469 * (key.offset - extent_offset).
2471 if (key.offset != offset)
2474 extent_offset = btrfs_file_extent_offset(leaf, extent);
2475 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2477 if (extent_offset >= old->extent_offset + old->offset +
2478 old->len || extent_offset + num_bytes <=
2479 old->extent_offset + old->offset)
2484 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2490 backref->root_id = root_id;
2491 backref->inum = inum;
2492 backref->file_pos = offset;
2493 backref->num_bytes = num_bytes;
2494 backref->extent_offset = extent_offset;
2495 backref->generation = btrfs_file_extent_generation(leaf, extent);
2497 backref_insert(&new->root, backref);
2500 btrfs_release_path(path);
2505 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2506 struct new_sa_defrag_extent *new)
2508 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2509 struct old_sa_defrag_extent *old, *tmp;
2514 list_for_each_entry_safe(old, tmp, &new->head, list) {
2515 ret = iterate_inodes_from_logical(old->bytenr +
2516 old->extent_offset, fs_info,
2517 path, record_one_backref,
2519 if (ret < 0 && ret != -ENOENT)
2522 /* no backref to be processed for this extent */
2524 list_del(&old->list);
2529 if (list_empty(&new->head))
2535 static int relink_is_mergable(struct extent_buffer *leaf,
2536 struct btrfs_file_extent_item *fi,
2537 struct new_sa_defrag_extent *new)
2539 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2542 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2545 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2548 if (btrfs_file_extent_encryption(leaf, fi) ||
2549 btrfs_file_extent_other_encoding(leaf, fi))
2556 * Note the backref might has changed, and in this case we just return 0.
2558 static noinline int relink_extent_backref(struct btrfs_path *path,
2559 struct sa_defrag_extent_backref *prev,
2560 struct sa_defrag_extent_backref *backref)
2562 struct btrfs_file_extent_item *extent;
2563 struct btrfs_file_extent_item *item;
2564 struct btrfs_ordered_extent *ordered;
2565 struct btrfs_trans_handle *trans;
2566 struct btrfs_root *root;
2567 struct btrfs_key key;
2568 struct extent_buffer *leaf;
2569 struct old_sa_defrag_extent *old = backref->old;
2570 struct new_sa_defrag_extent *new = old->new;
2571 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2572 struct inode *inode;
2573 struct extent_state *cached = NULL;
2582 if (prev && prev->root_id == backref->root_id &&
2583 prev->inum == backref->inum &&
2584 prev->file_pos + prev->num_bytes == backref->file_pos)
2587 /* step 1: get root */
2588 key.objectid = backref->root_id;
2589 key.type = BTRFS_ROOT_ITEM_KEY;
2590 key.offset = (u64)-1;
2592 index = srcu_read_lock(&fs_info->subvol_srcu);
2594 root = btrfs_read_fs_root_no_name(fs_info, &key);
2596 srcu_read_unlock(&fs_info->subvol_srcu, index);
2597 if (PTR_ERR(root) == -ENOENT)
2599 return PTR_ERR(root);
2602 if (btrfs_root_readonly(root)) {
2603 srcu_read_unlock(&fs_info->subvol_srcu, index);
2607 /* step 2: get inode */
2608 key.objectid = backref->inum;
2609 key.type = BTRFS_INODE_ITEM_KEY;
2612 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2613 if (IS_ERR(inode)) {
2614 srcu_read_unlock(&fs_info->subvol_srcu, index);
2618 srcu_read_unlock(&fs_info->subvol_srcu, index);
2620 /* step 3: relink backref */
2621 lock_start = backref->file_pos;
2622 lock_end = backref->file_pos + backref->num_bytes - 1;
2623 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2626 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2628 btrfs_put_ordered_extent(ordered);
2632 trans = btrfs_join_transaction(root);
2633 if (IS_ERR(trans)) {
2634 ret = PTR_ERR(trans);
2638 key.objectid = backref->inum;
2639 key.type = BTRFS_EXTENT_DATA_KEY;
2640 key.offset = backref->file_pos;
2642 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2645 } else if (ret > 0) {
2650 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2651 struct btrfs_file_extent_item);
2653 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2654 backref->generation)
2657 btrfs_release_path(path);
2659 start = backref->file_pos;
2660 if (backref->extent_offset < old->extent_offset + old->offset)
2661 start += old->extent_offset + old->offset -
2662 backref->extent_offset;
2664 len = min(backref->extent_offset + backref->num_bytes,
2665 old->extent_offset + old->offset + old->len);
2666 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2668 ret = btrfs_drop_extents(trans, root, inode, start,
2673 key.objectid = btrfs_ino(BTRFS_I(inode));
2674 key.type = BTRFS_EXTENT_DATA_KEY;
2677 path->leave_spinning = 1;
2679 struct btrfs_file_extent_item *fi;
2681 struct btrfs_key found_key;
2683 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2688 leaf = path->nodes[0];
2689 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2691 fi = btrfs_item_ptr(leaf, path->slots[0],
2692 struct btrfs_file_extent_item);
2693 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2695 if (extent_len + found_key.offset == start &&
2696 relink_is_mergable(leaf, fi, new)) {
2697 btrfs_set_file_extent_num_bytes(leaf, fi,
2699 btrfs_mark_buffer_dirty(leaf);
2700 inode_add_bytes(inode, len);
2706 btrfs_release_path(path);
2711 ret = btrfs_insert_empty_item(trans, root, path, &key,
2714 btrfs_abort_transaction(trans, ret);
2718 leaf = path->nodes[0];
2719 item = btrfs_item_ptr(leaf, path->slots[0],
2720 struct btrfs_file_extent_item);
2721 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2722 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2723 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2724 btrfs_set_file_extent_num_bytes(leaf, item, len);
2725 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2726 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2727 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2728 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2729 btrfs_set_file_extent_encryption(leaf, item, 0);
2730 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2732 btrfs_mark_buffer_dirty(leaf);
2733 inode_add_bytes(inode, len);
2734 btrfs_release_path(path);
2736 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2738 backref->root_id, backref->inum,
2739 new->file_pos); /* start - extent_offset */
2741 btrfs_abort_transaction(trans, ret);
2747 btrfs_release_path(path);
2748 path->leave_spinning = 0;
2749 btrfs_end_transaction(trans);
2751 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2757 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2759 struct old_sa_defrag_extent *old, *tmp;
2764 list_for_each_entry_safe(old, tmp, &new->head, list) {
2770 static void relink_file_extents(struct new_sa_defrag_extent *new)
2772 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2773 struct btrfs_path *path;
2774 struct sa_defrag_extent_backref *backref;
2775 struct sa_defrag_extent_backref *prev = NULL;
2776 struct inode *inode;
2777 struct btrfs_root *root;
2778 struct rb_node *node;
2782 root = BTRFS_I(inode)->root;
2784 path = btrfs_alloc_path();
2788 if (!record_extent_backrefs(path, new)) {
2789 btrfs_free_path(path);
2792 btrfs_release_path(path);
2795 node = rb_first(&new->root);
2798 rb_erase(node, &new->root);
2800 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2802 ret = relink_extent_backref(path, prev, backref);
2815 btrfs_free_path(path);
2817 free_sa_defrag_extent(new);
2819 atomic_dec(&fs_info->defrag_running);
2820 wake_up(&fs_info->transaction_wait);
2823 static struct new_sa_defrag_extent *
2824 record_old_file_extents(struct inode *inode,
2825 struct btrfs_ordered_extent *ordered)
2827 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2828 struct btrfs_root *root = BTRFS_I(inode)->root;
2829 struct btrfs_path *path;
2830 struct btrfs_key key;
2831 struct old_sa_defrag_extent *old;
2832 struct new_sa_defrag_extent *new;
2835 new = kmalloc(sizeof(*new), GFP_NOFS);
2840 new->file_pos = ordered->file_offset;
2841 new->len = ordered->len;
2842 new->bytenr = ordered->start;
2843 new->disk_len = ordered->disk_len;
2844 new->compress_type = ordered->compress_type;
2845 new->root = RB_ROOT;
2846 INIT_LIST_HEAD(&new->head);
2848 path = btrfs_alloc_path();
2852 key.objectid = btrfs_ino(BTRFS_I(inode));
2853 key.type = BTRFS_EXTENT_DATA_KEY;
2854 key.offset = new->file_pos;
2856 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2859 if (ret > 0 && path->slots[0] > 0)
2862 /* find out all the old extents for the file range */
2864 struct btrfs_file_extent_item *extent;
2865 struct extent_buffer *l;
2874 slot = path->slots[0];
2876 if (slot >= btrfs_header_nritems(l)) {
2877 ret = btrfs_next_leaf(root, path);
2885 btrfs_item_key_to_cpu(l, &key, slot);
2887 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2889 if (key.type != BTRFS_EXTENT_DATA_KEY)
2891 if (key.offset >= new->file_pos + new->len)
2894 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2896 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2897 if (key.offset + num_bytes < new->file_pos)
2900 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2904 extent_offset = btrfs_file_extent_offset(l, extent);
2906 old = kmalloc(sizeof(*old), GFP_NOFS);
2910 offset = max(new->file_pos, key.offset);
2911 end = min(new->file_pos + new->len, key.offset + num_bytes);
2913 old->bytenr = disk_bytenr;
2914 old->extent_offset = extent_offset;
2915 old->offset = offset - key.offset;
2916 old->len = end - offset;
2919 list_add_tail(&old->list, &new->head);
2925 btrfs_free_path(path);
2926 atomic_inc(&fs_info->defrag_running);
2931 btrfs_free_path(path);
2933 free_sa_defrag_extent(new);
2937 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2940 struct btrfs_block_group_cache *cache;
2942 cache = btrfs_lookup_block_group(fs_info, start);
2945 spin_lock(&cache->lock);
2946 cache->delalloc_bytes -= len;
2947 spin_unlock(&cache->lock);
2949 btrfs_put_block_group(cache);
2952 /* as ordered data IO finishes, this gets called so we can finish
2953 * an ordered extent if the range of bytes in the file it covers are
2956 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2958 struct inode *inode = ordered_extent->inode;
2959 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2960 struct btrfs_root *root = BTRFS_I(inode)->root;
2961 struct btrfs_trans_handle *trans = NULL;
2962 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2963 struct extent_state *cached_state = NULL;
2964 struct new_sa_defrag_extent *new = NULL;
2965 int compress_type = 0;
2967 u64 logical_len = ordered_extent->len;
2969 bool truncated = false;
2970 bool range_locked = false;
2971 bool clear_new_delalloc_bytes = false;
2973 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2974 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2975 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2976 clear_new_delalloc_bytes = true;
2978 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2980 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2985 btrfs_free_io_failure_record(BTRFS_I(inode),
2986 ordered_extent->file_offset,
2987 ordered_extent->file_offset +
2988 ordered_extent->len - 1);
2990 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2992 logical_len = ordered_extent->truncated_len;
2993 /* Truncated the entire extent, don't bother adding */
2998 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2999 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3002 * For mwrite(mmap + memset to write) case, we still reserve
3003 * space for NOCOW range.
3004 * As NOCOW won't cause a new delayed ref, just free the space
3006 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3007 ordered_extent->len);
3008 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3010 trans = btrfs_join_transaction_nolock(root);
3012 trans = btrfs_join_transaction(root);
3013 if (IS_ERR(trans)) {
3014 ret = PTR_ERR(trans);
3018 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3019 ret = btrfs_update_inode_fallback(trans, root, inode);
3020 if (ret) /* -ENOMEM or corruption */
3021 btrfs_abort_transaction(trans, ret);
3025 range_locked = true;
3026 lock_extent_bits(io_tree, ordered_extent->file_offset,
3027 ordered_extent->file_offset + ordered_extent->len - 1,
3030 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3031 ordered_extent->file_offset + ordered_extent->len - 1,
3032 EXTENT_DEFRAG, 0, cached_state);
3034 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3035 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3036 /* the inode is shared */
3037 new = record_old_file_extents(inode, ordered_extent);
3039 clear_extent_bit(io_tree, ordered_extent->file_offset,
3040 ordered_extent->file_offset + ordered_extent->len - 1,
3041 EXTENT_DEFRAG, 0, 0, &cached_state);
3045 trans = btrfs_join_transaction_nolock(root);
3047 trans = btrfs_join_transaction(root);
3048 if (IS_ERR(trans)) {
3049 ret = PTR_ERR(trans);
3054 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3056 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3057 compress_type = ordered_extent->compress_type;
3058 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3059 BUG_ON(compress_type);
3060 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3061 ordered_extent->len);
3062 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3063 ordered_extent->file_offset,
3064 ordered_extent->file_offset +
3067 BUG_ON(root == fs_info->tree_root);
3068 ret = insert_reserved_file_extent(trans, inode,
3069 ordered_extent->file_offset,
3070 ordered_extent->start,
3071 ordered_extent->disk_len,
3072 logical_len, logical_len,
3073 compress_type, 0, 0,
3074 BTRFS_FILE_EXTENT_REG);
3076 btrfs_release_delalloc_bytes(fs_info,
3077 ordered_extent->start,
3078 ordered_extent->disk_len);
3080 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3081 ordered_extent->file_offset, ordered_extent->len,
3084 btrfs_abort_transaction(trans, ret);
3088 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3090 btrfs_abort_transaction(trans, ret);
3094 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3095 ret = btrfs_update_inode_fallback(trans, root, inode);
3096 if (ret) { /* -ENOMEM or corruption */
3097 btrfs_abort_transaction(trans, ret);
3102 if (range_locked || clear_new_delalloc_bytes) {
3103 unsigned int clear_bits = 0;
3106 clear_bits |= EXTENT_LOCKED;
3107 if (clear_new_delalloc_bytes)
3108 clear_bits |= EXTENT_DELALLOC_NEW;
3109 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3110 ordered_extent->file_offset,
3111 ordered_extent->file_offset +
3112 ordered_extent->len - 1,
3114 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3119 btrfs_end_transaction(trans);
3121 if (ret || truncated) {
3125 start = ordered_extent->file_offset + logical_len;
3127 start = ordered_extent->file_offset;
3128 end = ordered_extent->file_offset + ordered_extent->len - 1;
3129 clear_extent_uptodate(io_tree, start, end, NULL);
3131 /* Drop the cache for the part of the extent we didn't write. */
3132 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3135 * If the ordered extent had an IOERR or something else went
3136 * wrong we need to return the space for this ordered extent
3137 * back to the allocator. We only free the extent in the
3138 * truncated case if we didn't write out the extent at all.
3140 if ((ret || !logical_len) &&
3141 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3142 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3143 btrfs_free_reserved_extent(fs_info,
3144 ordered_extent->start,
3145 ordered_extent->disk_len, 1);
3150 * This needs to be done to make sure anybody waiting knows we are done
3151 * updating everything for this ordered extent.
3153 btrfs_remove_ordered_extent(inode, ordered_extent);
3155 /* for snapshot-aware defrag */
3158 free_sa_defrag_extent(new);
3159 atomic_dec(&fs_info->defrag_running);
3161 relink_file_extents(new);
3166 btrfs_put_ordered_extent(ordered_extent);
3167 /* once for the tree */
3168 btrfs_put_ordered_extent(ordered_extent);
3173 static void finish_ordered_fn(struct btrfs_work *work)
3175 struct btrfs_ordered_extent *ordered_extent;
3176 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3177 btrfs_finish_ordered_io(ordered_extent);
3180 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3181 struct extent_state *state, int uptodate)
3183 struct inode *inode = page->mapping->host;
3184 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3185 struct btrfs_ordered_extent *ordered_extent = NULL;
3186 struct btrfs_workqueue *wq;
3187 btrfs_work_func_t func;
3189 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3191 ClearPagePrivate2(page);
3192 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3193 end - start + 1, uptodate))
3196 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3197 wq = fs_info->endio_freespace_worker;
3198 func = btrfs_freespace_write_helper;
3200 wq = fs_info->endio_write_workers;
3201 func = btrfs_endio_write_helper;
3204 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3206 btrfs_queue_work(wq, &ordered_extent->work);
3209 static int __readpage_endio_check(struct inode *inode,
3210 struct btrfs_io_bio *io_bio,
3211 int icsum, struct page *page,
3212 int pgoff, u64 start, size_t len)
3218 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3220 kaddr = kmap_atomic(page);
3221 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3222 btrfs_csum_final(csum, (u8 *)&csum);
3223 if (csum != csum_expected)
3226 kunmap_atomic(kaddr);
3229 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3230 io_bio->mirror_num);
3231 memset(kaddr + pgoff, 1, len);
3232 flush_dcache_page(page);
3233 kunmap_atomic(kaddr);
3238 * when reads are done, we need to check csums to verify the data is correct
3239 * if there's a match, we allow the bio to finish. If not, the code in
3240 * extent_io.c will try to find good copies for us.
3242 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3243 u64 phy_offset, struct page *page,
3244 u64 start, u64 end, int mirror)
3246 size_t offset = start - page_offset(page);
3247 struct inode *inode = page->mapping->host;
3248 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3249 struct btrfs_root *root = BTRFS_I(inode)->root;
3251 if (PageChecked(page)) {
3252 ClearPageChecked(page);
3256 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3259 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3260 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3261 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3265 phy_offset >>= inode->i_sb->s_blocksize_bits;
3266 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3267 start, (size_t)(end - start + 1));
3271 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3273 * @inode: The inode we want to perform iput on
3275 * This function uses the generic vfs_inode::i_count to track whether we should
3276 * just decrement it (in case it's > 1) or if this is the last iput then link
3277 * the inode to the delayed iput machinery. Delayed iputs are processed at
3278 * transaction commit time/superblock commit/cleaner kthread.
3280 void btrfs_add_delayed_iput(struct inode *inode)
3282 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3283 struct btrfs_inode *binode = BTRFS_I(inode);
3285 if (atomic_add_unless(&inode->i_count, -1, 1))
3288 spin_lock(&fs_info->delayed_iput_lock);
3289 ASSERT(list_empty(&binode->delayed_iput));
3290 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3291 spin_unlock(&fs_info->delayed_iput_lock);
3294 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3297 spin_lock(&fs_info->delayed_iput_lock);
3298 while (!list_empty(&fs_info->delayed_iputs)) {
3299 struct btrfs_inode *inode;
3301 inode = list_first_entry(&fs_info->delayed_iputs,
3302 struct btrfs_inode, delayed_iput);
3303 list_del_init(&inode->delayed_iput);
3304 spin_unlock(&fs_info->delayed_iput_lock);
3305 iput(&inode->vfs_inode);
3306 spin_lock(&fs_info->delayed_iput_lock);
3308 spin_unlock(&fs_info->delayed_iput_lock);
3312 * This is called in transaction commit time. If there are no orphan
3313 * files in the subvolume, it removes orphan item and frees block_rsv
3316 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3317 struct btrfs_root *root)
3319 struct btrfs_fs_info *fs_info = root->fs_info;
3320 struct btrfs_block_rsv *block_rsv;
3323 if (atomic_read(&root->orphan_inodes) ||
3324 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3327 spin_lock(&root->orphan_lock);
3328 if (atomic_read(&root->orphan_inodes)) {
3329 spin_unlock(&root->orphan_lock);
3333 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3334 spin_unlock(&root->orphan_lock);
3338 block_rsv = root->orphan_block_rsv;
3339 root->orphan_block_rsv = NULL;
3340 spin_unlock(&root->orphan_lock);
3342 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3343 btrfs_root_refs(&root->root_item) > 0) {
3344 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3345 root->root_key.objectid);
3347 btrfs_abort_transaction(trans, ret);
3349 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3354 WARN_ON(block_rsv->size > 0);
3355 btrfs_free_block_rsv(fs_info, block_rsv);
3360 * This creates an orphan entry for the given inode in case something goes
3361 * wrong in the middle of an unlink/truncate.
3363 * NOTE: caller of this function should reserve 5 units of metadata for
3366 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3367 struct btrfs_inode *inode)
3369 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3370 struct btrfs_root *root = inode->root;
3371 struct btrfs_block_rsv *block_rsv = NULL;
3376 if (!root->orphan_block_rsv) {
3377 block_rsv = btrfs_alloc_block_rsv(fs_info,
3378 BTRFS_BLOCK_RSV_TEMP);
3383 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3384 &inode->runtime_flags)) {
3387 * For proper ENOSPC handling, we should do orphan
3388 * cleanup when mounting. But this introduces backward
3389 * compatibility issue.
3391 if (!xchg(&root->orphan_item_inserted, 1))
3399 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3400 &inode->runtime_flags))
3403 spin_lock(&root->orphan_lock);
3404 /* If someone has created ->orphan_block_rsv, be happy to use it. */
3405 if (!root->orphan_block_rsv) {
3406 root->orphan_block_rsv = block_rsv;
3407 } else if (block_rsv) {
3408 btrfs_free_block_rsv(fs_info, block_rsv);
3413 atomic_inc(&root->orphan_inodes);
3414 spin_unlock(&root->orphan_lock);
3416 /* grab metadata reservation from transaction handle */
3418 ret = btrfs_orphan_reserve_metadata(trans, inode);
3422 * dec doesn't need spin_lock as ->orphan_block_rsv
3423 * would be released only if ->orphan_inodes is
3426 atomic_dec(&root->orphan_inodes);
3427 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3428 &inode->runtime_flags);
3430 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3431 &inode->runtime_flags);
3436 /* insert an orphan item to track this unlinked/truncated file */
3438 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3441 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3442 &inode->runtime_flags);
3443 btrfs_orphan_release_metadata(inode);
3446 * btrfs_orphan_commit_root may race with us and set
3447 * ->orphan_block_rsv to zero, in order to avoid that,
3448 * decrease ->orphan_inodes after everything is done.
3450 atomic_dec(&root->orphan_inodes);
3451 if (ret != -EEXIST) {
3452 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3453 &inode->runtime_flags);
3454 btrfs_abort_transaction(trans, ret);
3461 /* insert an orphan item to track subvolume contains orphan files */
3463 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3464 root->root_key.objectid);
3465 if (ret && ret != -EEXIST) {
3466 btrfs_abort_transaction(trans, ret);
3474 * We have done the truncate/delete so we can go ahead and remove the orphan
3475 * item for this particular inode.
3477 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3478 struct btrfs_inode *inode)
3480 struct btrfs_root *root = inode->root;
3481 int delete_item = 0;
3484 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3485 &inode->runtime_flags))
3488 if (delete_item && trans)
3489 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3491 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3492 &inode->runtime_flags))
3493 btrfs_orphan_release_metadata(inode);
3496 * btrfs_orphan_commit_root may race with us and set ->orphan_block_rsv
3497 * to zero, in order to avoid that, decrease ->orphan_inodes after
3498 * everything is done.
3501 atomic_dec(&root->orphan_inodes);
3507 * this cleans up any orphans that may be left on the list from the last use
3510 int btrfs_orphan_cleanup(struct btrfs_root *root)
3512 struct btrfs_fs_info *fs_info = root->fs_info;
3513 struct btrfs_path *path;
3514 struct extent_buffer *leaf;
3515 struct btrfs_key key, found_key;
3516 struct btrfs_trans_handle *trans;
3517 struct inode *inode;
3518 u64 last_objectid = 0;
3519 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3521 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3524 path = btrfs_alloc_path();
3529 path->reada = READA_BACK;
3531 key.objectid = BTRFS_ORPHAN_OBJECTID;
3532 key.type = BTRFS_ORPHAN_ITEM_KEY;
3533 key.offset = (u64)-1;
3536 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3541 * if ret == 0 means we found what we were searching for, which
3542 * is weird, but possible, so only screw with path if we didn't
3543 * find the key and see if we have stuff that matches
3547 if (path->slots[0] == 0)
3552 /* pull out the item */
3553 leaf = path->nodes[0];
3554 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3556 /* make sure the item matches what we want */
3557 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3559 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3562 /* release the path since we're done with it */
3563 btrfs_release_path(path);
3566 * this is where we are basically btrfs_lookup, without the
3567 * crossing root thing. we store the inode number in the
3568 * offset of the orphan item.
3571 if (found_key.offset == last_objectid) {
3573 "Error removing orphan entry, stopping orphan cleanup");
3578 last_objectid = found_key.offset;
3580 found_key.objectid = found_key.offset;
3581 found_key.type = BTRFS_INODE_ITEM_KEY;
3582 found_key.offset = 0;
3583 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3584 ret = PTR_ERR_OR_ZERO(inode);
3585 if (ret && ret != -ENOENT)
3588 if (ret == -ENOENT && root == fs_info->tree_root) {
3589 struct btrfs_root *dead_root;
3590 struct btrfs_fs_info *fs_info = root->fs_info;
3591 int is_dead_root = 0;
3594 * this is an orphan in the tree root. Currently these
3595 * could come from 2 sources:
3596 * a) a snapshot deletion in progress
3597 * b) a free space cache inode
3598 * We need to distinguish those two, as the snapshot
3599 * orphan must not get deleted.
3600 * find_dead_roots already ran before us, so if this
3601 * is a snapshot deletion, we should find the root
3602 * in the dead_roots list
3604 spin_lock(&fs_info->trans_lock);
3605 list_for_each_entry(dead_root, &fs_info->dead_roots,
3607 if (dead_root->root_key.objectid ==
3608 found_key.objectid) {
3613 spin_unlock(&fs_info->trans_lock);
3615 /* prevent this orphan from being found again */
3616 key.offset = found_key.objectid - 1;
3621 * Inode is already gone but the orphan item is still there,
3622 * kill the orphan item.
3624 if (ret == -ENOENT) {
3625 trans = btrfs_start_transaction(root, 1);
3626 if (IS_ERR(trans)) {
3627 ret = PTR_ERR(trans);
3630 btrfs_debug(fs_info, "auto deleting %Lu",
3631 found_key.objectid);
3632 ret = btrfs_del_orphan_item(trans, root,
3633 found_key.objectid);
3634 btrfs_end_transaction(trans);
3641 * add this inode to the orphan list so btrfs_orphan_del does
3642 * the proper thing when we hit it
3644 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3645 &BTRFS_I(inode)->runtime_flags);
3646 atomic_inc(&root->orphan_inodes);
3648 /* if we have links, this was a truncate, lets do that */
3649 if (inode->i_nlink) {
3650 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3656 /* 1 for the orphan item deletion. */
3657 trans = btrfs_start_transaction(root, 1);
3658 if (IS_ERR(trans)) {
3660 ret = PTR_ERR(trans);
3663 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3664 btrfs_end_transaction(trans);
3670 ret = btrfs_truncate(inode, false);
3672 btrfs_orphan_del(NULL, BTRFS_I(inode));
3677 /* this will do delete_inode and everything for us */
3682 /* release the path since we're done with it */
3683 btrfs_release_path(path);
3685 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3687 if (root->orphan_block_rsv)
3688 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3691 if (root->orphan_block_rsv ||
3692 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3693 trans = btrfs_join_transaction(root);
3695 btrfs_end_transaction(trans);
3699 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3701 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3705 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3706 btrfs_free_path(path);
3711 * very simple check to peek ahead in the leaf looking for xattrs. If we
3712 * don't find any xattrs, we know there can't be any acls.
3714 * slot is the slot the inode is in, objectid is the objectid of the inode
3716 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3717 int slot, u64 objectid,
3718 int *first_xattr_slot)
3720 u32 nritems = btrfs_header_nritems(leaf);
3721 struct btrfs_key found_key;
3722 static u64 xattr_access = 0;
3723 static u64 xattr_default = 0;
3726 if (!xattr_access) {
3727 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3728 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3729 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3730 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3734 *first_xattr_slot = -1;
3735 while (slot < nritems) {
3736 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3738 /* we found a different objectid, there must not be acls */
3739 if (found_key.objectid != objectid)
3742 /* we found an xattr, assume we've got an acl */
3743 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3744 if (*first_xattr_slot == -1)
3745 *first_xattr_slot = slot;
3746 if (found_key.offset == xattr_access ||
3747 found_key.offset == xattr_default)
3752 * we found a key greater than an xattr key, there can't
3753 * be any acls later on
3755 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3762 * it goes inode, inode backrefs, xattrs, extents,
3763 * so if there are a ton of hard links to an inode there can
3764 * be a lot of backrefs. Don't waste time searching too hard,
3765 * this is just an optimization
3770 /* we hit the end of the leaf before we found an xattr or
3771 * something larger than an xattr. We have to assume the inode
3774 if (*first_xattr_slot == -1)
3775 *first_xattr_slot = slot;
3780 * read an inode from the btree into the in-memory inode
3782 static int btrfs_read_locked_inode(struct inode *inode)
3784 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3785 struct btrfs_path *path;
3786 struct extent_buffer *leaf;
3787 struct btrfs_inode_item *inode_item;
3788 struct btrfs_root *root = BTRFS_I(inode)->root;
3789 struct btrfs_key location;
3794 bool filled = false;
3795 int first_xattr_slot;
3797 ret = btrfs_fill_inode(inode, &rdev);
3801 path = btrfs_alloc_path();
3807 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3809 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3816 leaf = path->nodes[0];
3821 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3822 struct btrfs_inode_item);
3823 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3824 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3825 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3826 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3827 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3829 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3830 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3832 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3833 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3835 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3836 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3838 BTRFS_I(inode)->i_otime.tv_sec =
3839 btrfs_timespec_sec(leaf, &inode_item->otime);
3840 BTRFS_I(inode)->i_otime.tv_nsec =
3841 btrfs_timespec_nsec(leaf, &inode_item->otime);
3843 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3844 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3845 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3847 inode_set_iversion_queried(inode,
3848 btrfs_inode_sequence(leaf, inode_item));
3849 inode->i_generation = BTRFS_I(inode)->generation;
3851 rdev = btrfs_inode_rdev(leaf, inode_item);
3853 BTRFS_I(inode)->index_cnt = (u64)-1;
3854 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3858 * If we were modified in the current generation and evicted from memory
3859 * and then re-read we need to do a full sync since we don't have any
3860 * idea about which extents were modified before we were evicted from
3863 * This is required for both inode re-read from disk and delayed inode
3864 * in delayed_nodes_tree.
3866 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3867 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3868 &BTRFS_I(inode)->runtime_flags);
3871 * We don't persist the id of the transaction where an unlink operation
3872 * against the inode was last made. So here we assume the inode might
3873 * have been evicted, and therefore the exact value of last_unlink_trans
3874 * lost, and set it to last_trans to avoid metadata inconsistencies
3875 * between the inode and its parent if the inode is fsync'ed and the log
3876 * replayed. For example, in the scenario:
3879 * ln mydir/foo mydir/bar
3882 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3883 * xfs_io -c fsync mydir/foo
3885 * mount fs, triggers fsync log replay
3887 * We must make sure that when we fsync our inode foo we also log its
3888 * parent inode, otherwise after log replay the parent still has the
3889 * dentry with the "bar" name but our inode foo has a link count of 1
3890 * and doesn't have an inode ref with the name "bar" anymore.
3892 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3893 * but it guarantees correctness at the expense of occasional full
3894 * transaction commits on fsync if our inode is a directory, or if our
3895 * inode is not a directory, logging its parent unnecessarily.
3897 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3900 if (inode->i_nlink != 1 ||
3901 path->slots[0] >= btrfs_header_nritems(leaf))
3904 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3905 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3908 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3909 if (location.type == BTRFS_INODE_REF_KEY) {
3910 struct btrfs_inode_ref *ref;
3912 ref = (struct btrfs_inode_ref *)ptr;
3913 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3914 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3915 struct btrfs_inode_extref *extref;
3917 extref = (struct btrfs_inode_extref *)ptr;
3918 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3923 * try to precache a NULL acl entry for files that don't have
3924 * any xattrs or acls
3926 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3927 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3928 if (first_xattr_slot != -1) {
3929 path->slots[0] = first_xattr_slot;
3930 ret = btrfs_load_inode_props(inode, path);
3933 "error loading props for ino %llu (root %llu): %d",
3934 btrfs_ino(BTRFS_I(inode)),
3935 root->root_key.objectid, ret);
3937 btrfs_free_path(path);
3940 cache_no_acl(inode);
3942 switch (inode->i_mode & S_IFMT) {
3944 inode->i_mapping->a_ops = &btrfs_aops;
3945 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3946 inode->i_fop = &btrfs_file_operations;
3947 inode->i_op = &btrfs_file_inode_operations;
3950 inode->i_fop = &btrfs_dir_file_operations;
3951 inode->i_op = &btrfs_dir_inode_operations;
3954 inode->i_op = &btrfs_symlink_inode_operations;
3955 inode_nohighmem(inode);
3956 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3959 inode->i_op = &btrfs_special_inode_operations;
3960 init_special_inode(inode, inode->i_mode, rdev);
3964 btrfs_update_iflags(inode);
3968 btrfs_free_path(path);
3969 make_bad_inode(inode);
3974 * given a leaf and an inode, copy the inode fields into the leaf
3976 static void fill_inode_item(struct btrfs_trans_handle *trans,
3977 struct extent_buffer *leaf,
3978 struct btrfs_inode_item *item,
3979 struct inode *inode)
3981 struct btrfs_map_token token;
3983 btrfs_init_map_token(&token);
3985 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3986 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3987 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3989 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3990 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3992 btrfs_set_token_timespec_sec(leaf, &item->atime,
3993 inode->i_atime.tv_sec, &token);
3994 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3995 inode->i_atime.tv_nsec, &token);
3997 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3998 inode->i_mtime.tv_sec, &token);
3999 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
4000 inode->i_mtime.tv_nsec, &token);
4002 btrfs_set_token_timespec_sec(leaf, &item->ctime,
4003 inode->i_ctime.tv_sec, &token);
4004 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
4005 inode->i_ctime.tv_nsec, &token);
4007 btrfs_set_token_timespec_sec(leaf, &item->otime,
4008 BTRFS_I(inode)->i_otime.tv_sec, &token);
4009 btrfs_set_token_timespec_nsec(leaf, &item->otime,
4010 BTRFS_I(inode)->i_otime.tv_nsec, &token);
4012 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
4014 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
4016 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
4018 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
4019 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
4020 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
4021 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
4025 * copy everything in the in-memory inode into the btree.
4027 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4028 struct btrfs_root *root, struct inode *inode)
4030 struct btrfs_inode_item *inode_item;
4031 struct btrfs_path *path;
4032 struct extent_buffer *leaf;
4035 path = btrfs_alloc_path();
4039 path->leave_spinning = 1;
4040 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
4048 leaf = path->nodes[0];
4049 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4050 struct btrfs_inode_item);
4052 fill_inode_item(trans, leaf, inode_item, inode);
4053 btrfs_mark_buffer_dirty(leaf);
4054 btrfs_set_inode_last_trans(trans, inode);
4057 btrfs_free_path(path);
4062 * copy everything in the in-memory inode into the btree.
4064 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4065 struct btrfs_root *root, struct inode *inode)
4067 struct btrfs_fs_info *fs_info = root->fs_info;
4071 * If the inode is a free space inode, we can deadlock during commit
4072 * if we put it into the delayed code.
4074 * The data relocation inode should also be directly updated
4077 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
4078 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
4079 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4080 btrfs_update_root_times(trans, root);
4082 ret = btrfs_delayed_update_inode(trans, root, inode);
4084 btrfs_set_inode_last_trans(trans, inode);
4088 return btrfs_update_inode_item(trans, root, inode);
4091 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4092 struct btrfs_root *root,
4093 struct inode *inode)
4097 ret = btrfs_update_inode(trans, root, inode);
4099 return btrfs_update_inode_item(trans, root, inode);
4104 * unlink helper that gets used here in inode.c and in the tree logging
4105 * recovery code. It remove a link in a directory with a given name, and
4106 * also drops the back refs in the inode to the directory
4108 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4109 struct btrfs_root *root,
4110 struct btrfs_inode *dir,
4111 struct btrfs_inode *inode,
4112 const char *name, int name_len)
4114 struct btrfs_fs_info *fs_info = root->fs_info;
4115 struct btrfs_path *path;
4117 struct extent_buffer *leaf;
4118 struct btrfs_dir_item *di;
4119 struct btrfs_key key;
4121 u64 ino = btrfs_ino(inode);
4122 u64 dir_ino = btrfs_ino(dir);
4124 path = btrfs_alloc_path();
4130 path->leave_spinning = 1;
4131 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4132 name, name_len, -1);
4141 leaf = path->nodes[0];
4142 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4143 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4146 btrfs_release_path(path);
4149 * If we don't have dir index, we have to get it by looking up
4150 * the inode ref, since we get the inode ref, remove it directly,
4151 * it is unnecessary to do delayed deletion.
4153 * But if we have dir index, needn't search inode ref to get it.
4154 * Since the inode ref is close to the inode item, it is better
4155 * that we delay to delete it, and just do this deletion when
4156 * we update the inode item.
4158 if (inode->dir_index) {
4159 ret = btrfs_delayed_delete_inode_ref(inode);
4161 index = inode->dir_index;
4166 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4170 "failed to delete reference to %.*s, inode %llu parent %llu",
4171 name_len, name, ino, dir_ino);
4172 btrfs_abort_transaction(trans, ret);
4176 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4178 btrfs_abort_transaction(trans, ret);
4182 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4184 if (ret != 0 && ret != -ENOENT) {
4185 btrfs_abort_transaction(trans, ret);
4189 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4194 btrfs_abort_transaction(trans, ret);
4196 btrfs_free_path(path);
4200 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4201 inode_inc_iversion(&inode->vfs_inode);
4202 inode_inc_iversion(&dir->vfs_inode);
4203 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4204 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4205 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4210 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4211 struct btrfs_root *root,
4212 struct btrfs_inode *dir, struct btrfs_inode *inode,
4213 const char *name, int name_len)
4216 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4218 drop_nlink(&inode->vfs_inode);
4219 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4225 * helper to start transaction for unlink and rmdir.
4227 * unlink and rmdir are special in btrfs, they do not always free space, so
4228 * if we cannot make our reservations the normal way try and see if there is
4229 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4230 * allow the unlink to occur.
4232 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4234 struct btrfs_root *root = BTRFS_I(dir)->root;
4237 * 1 for the possible orphan item
4238 * 1 for the dir item
4239 * 1 for the dir index
4240 * 1 for the inode ref
4243 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4246 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4248 struct btrfs_root *root = BTRFS_I(dir)->root;
4249 struct btrfs_trans_handle *trans;
4250 struct inode *inode = d_inode(dentry);
4253 trans = __unlink_start_trans(dir);
4255 return PTR_ERR(trans);
4257 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4260 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4261 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4262 dentry->d_name.len);
4266 if (inode->i_nlink == 0) {
4267 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4273 btrfs_end_transaction(trans);
4274 btrfs_btree_balance_dirty(root->fs_info);
4278 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4279 struct btrfs_root *root,
4280 struct inode *dir, u64 objectid,
4281 const char *name, int name_len)
4283 struct btrfs_fs_info *fs_info = root->fs_info;
4284 struct btrfs_path *path;
4285 struct extent_buffer *leaf;
4286 struct btrfs_dir_item *di;
4287 struct btrfs_key key;
4290 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4292 path = btrfs_alloc_path();
4296 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4297 name, name_len, -1);
4298 if (IS_ERR_OR_NULL(di)) {
4306 leaf = path->nodes[0];
4307 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4308 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4309 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4311 btrfs_abort_transaction(trans, ret);
4314 btrfs_release_path(path);
4316 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4317 root->root_key.objectid, dir_ino,
4318 &index, name, name_len);
4320 if (ret != -ENOENT) {
4321 btrfs_abort_transaction(trans, ret);
4324 di = btrfs_search_dir_index_item(root, path, dir_ino,
4326 if (IS_ERR_OR_NULL(di)) {
4331 btrfs_abort_transaction(trans, ret);
4335 leaf = path->nodes[0];
4336 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4337 btrfs_release_path(path);
4340 btrfs_release_path(path);
4342 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4344 btrfs_abort_transaction(trans, ret);
4348 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4349 inode_inc_iversion(dir);
4350 dir->i_mtime = dir->i_ctime = current_time(dir);
4351 ret = btrfs_update_inode_fallback(trans, root, dir);
4353 btrfs_abort_transaction(trans, ret);
4355 btrfs_free_path(path);
4359 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4361 struct inode *inode = d_inode(dentry);
4363 struct btrfs_root *root = BTRFS_I(dir)->root;
4364 struct btrfs_trans_handle *trans;
4365 u64 last_unlink_trans;
4367 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4369 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4372 trans = __unlink_start_trans(dir);
4374 return PTR_ERR(trans);
4376 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4377 err = btrfs_unlink_subvol(trans, root, dir,
4378 BTRFS_I(inode)->location.objectid,
4379 dentry->d_name.name,
4380 dentry->d_name.len);
4384 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4388 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4390 /* now the directory is empty */
4391 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4392 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4393 dentry->d_name.len);
4395 btrfs_i_size_write(BTRFS_I(inode), 0);
4397 * Propagate the last_unlink_trans value of the deleted dir to
4398 * its parent directory. This is to prevent an unrecoverable
4399 * log tree in the case we do something like this:
4401 * 2) create snapshot under dir foo
4402 * 3) delete the snapshot
4405 * 6) fsync foo or some file inside foo
4407 if (last_unlink_trans >= trans->transid)
4408 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4411 btrfs_end_transaction(trans);
4412 btrfs_btree_balance_dirty(root->fs_info);
4417 static int truncate_space_check(struct btrfs_trans_handle *trans,
4418 struct btrfs_root *root,
4421 struct btrfs_fs_info *fs_info = root->fs_info;
4425 * This is only used to apply pressure to the enospc system, we don't
4426 * intend to use this reservation at all.
4428 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4429 bytes_deleted *= fs_info->nodesize;
4430 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4431 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4433 trace_btrfs_space_reservation(fs_info, "transaction",
4436 trans->bytes_reserved += bytes_deleted;
4443 * Return this if we need to call truncate_block for the last bit of the
4446 #define NEED_TRUNCATE_BLOCK 1
4449 * this can truncate away extent items, csum items and directory items.
4450 * It starts at a high offset and removes keys until it can't find
4451 * any higher than new_size
4453 * csum items that cross the new i_size are truncated to the new size
4456 * min_type is the minimum key type to truncate down to. If set to 0, this
4457 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4459 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4460 struct btrfs_root *root,
4461 struct inode *inode,
4462 u64 new_size, u32 min_type)
4464 struct btrfs_fs_info *fs_info = root->fs_info;
4465 struct btrfs_path *path;
4466 struct extent_buffer *leaf;
4467 struct btrfs_file_extent_item *fi;
4468 struct btrfs_key key;
4469 struct btrfs_key found_key;
4470 u64 extent_start = 0;
4471 u64 extent_num_bytes = 0;
4472 u64 extent_offset = 0;
4474 u64 last_size = new_size;
4475 u32 found_type = (u8)-1;
4478 int pending_del_nr = 0;
4479 int pending_del_slot = 0;
4480 int extent_type = -1;
4483 u64 ino = btrfs_ino(BTRFS_I(inode));
4484 u64 bytes_deleted = 0;
4485 bool be_nice = false;
4486 bool should_throttle = false;
4487 bool should_end = false;
4489 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4492 * for non-free space inodes and ref cows, we want to back off from
4495 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4496 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4499 path = btrfs_alloc_path();
4502 path->reada = READA_BACK;
4505 * We want to drop from the next block forward in case this new size is
4506 * not block aligned since we will be keeping the last block of the
4507 * extent just the way it is.
4509 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4510 root == fs_info->tree_root)
4511 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4512 fs_info->sectorsize),
4516 * This function is also used to drop the items in the log tree before
4517 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4518 * it is used to drop the loged items. So we shouldn't kill the delayed
4521 if (min_type == 0 && root == BTRFS_I(inode)->root)
4522 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4525 key.offset = (u64)-1;
4530 * with a 16K leaf size and 128MB extents, you can actually queue
4531 * up a huge file in a single leaf. Most of the time that
4532 * bytes_deleted is > 0, it will be huge by the time we get here
4534 if (be_nice && bytes_deleted > SZ_32M) {
4535 if (btrfs_should_end_transaction(trans)) {
4542 path->leave_spinning = 1;
4543 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4550 /* there are no items in the tree for us to truncate, we're
4553 if (path->slots[0] == 0)
4560 leaf = path->nodes[0];
4561 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4562 found_type = found_key.type;
4564 if (found_key.objectid != ino)
4567 if (found_type < min_type)
4570 item_end = found_key.offset;
4571 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4572 fi = btrfs_item_ptr(leaf, path->slots[0],
4573 struct btrfs_file_extent_item);
4574 extent_type = btrfs_file_extent_type(leaf, fi);
4575 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4577 btrfs_file_extent_num_bytes(leaf, fi);
4579 trace_btrfs_truncate_show_fi_regular(
4580 BTRFS_I(inode), leaf, fi,
4582 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4583 item_end += btrfs_file_extent_inline_len(leaf,
4584 path->slots[0], fi);
4586 trace_btrfs_truncate_show_fi_inline(
4587 BTRFS_I(inode), leaf, fi, path->slots[0],
4592 if (found_type > min_type) {
4595 if (item_end < new_size)
4597 if (found_key.offset >= new_size)
4603 /* FIXME, shrink the extent if the ref count is only 1 */
4604 if (found_type != BTRFS_EXTENT_DATA_KEY)
4607 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4609 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4611 u64 orig_num_bytes =
4612 btrfs_file_extent_num_bytes(leaf, fi);
4613 extent_num_bytes = ALIGN(new_size -
4615 fs_info->sectorsize);
4616 btrfs_set_file_extent_num_bytes(leaf, fi,
4618 num_dec = (orig_num_bytes -
4620 if (test_bit(BTRFS_ROOT_REF_COWS,
4623 inode_sub_bytes(inode, num_dec);
4624 btrfs_mark_buffer_dirty(leaf);
4627 btrfs_file_extent_disk_num_bytes(leaf,
4629 extent_offset = found_key.offset -
4630 btrfs_file_extent_offset(leaf, fi);
4632 /* FIXME blocksize != 4096 */
4633 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4634 if (extent_start != 0) {
4636 if (test_bit(BTRFS_ROOT_REF_COWS,
4638 inode_sub_bytes(inode, num_dec);
4641 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4643 * we can't truncate inline items that have had
4647 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4648 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4649 btrfs_file_extent_compression(leaf, fi) == 0) {
4650 u32 size = (u32)(new_size - found_key.offset);
4652 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4653 size = btrfs_file_extent_calc_inline_size(size);
4654 btrfs_truncate_item(root->fs_info, path, size, 1);
4655 } else if (!del_item) {
4657 * We have to bail so the last_size is set to
4658 * just before this extent.
4660 err = NEED_TRUNCATE_BLOCK;
4664 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4665 inode_sub_bytes(inode, item_end + 1 - new_size);
4669 last_size = found_key.offset;
4671 last_size = new_size;
4673 if (!pending_del_nr) {
4674 /* no pending yet, add ourselves */
4675 pending_del_slot = path->slots[0];
4677 } else if (pending_del_nr &&
4678 path->slots[0] + 1 == pending_del_slot) {
4679 /* hop on the pending chunk */
4681 pending_del_slot = path->slots[0];
4688 should_throttle = false;
4691 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4692 root == fs_info->tree_root)) {
4693 btrfs_set_path_blocking(path);
4694 bytes_deleted += extent_num_bytes;
4695 ret = btrfs_free_extent(trans, root, extent_start,
4696 extent_num_bytes, 0,
4697 btrfs_header_owner(leaf),
4698 ino, extent_offset);
4700 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4701 btrfs_async_run_delayed_refs(fs_info,
4702 trans->delayed_ref_updates * 2,
4705 if (truncate_space_check(trans, root,
4706 extent_num_bytes)) {
4709 if (btrfs_should_throttle_delayed_refs(trans,
4711 should_throttle = true;
4715 if (found_type == BTRFS_INODE_ITEM_KEY)
4718 if (path->slots[0] == 0 ||
4719 path->slots[0] != pending_del_slot ||
4720 should_throttle || should_end) {
4721 if (pending_del_nr) {
4722 ret = btrfs_del_items(trans, root, path,
4726 btrfs_abort_transaction(trans, ret);
4731 btrfs_release_path(path);
4732 if (should_throttle) {
4733 unsigned long updates = trans->delayed_ref_updates;
4735 trans->delayed_ref_updates = 0;
4736 ret = btrfs_run_delayed_refs(trans,
4744 * if we failed to refill our space rsv, bail out
4745 * and let the transaction restart
4757 if (pending_del_nr) {
4758 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4761 btrfs_abort_transaction(trans, ret);
4764 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4765 ASSERT(last_size >= new_size);
4766 if (!err && last_size > new_size)
4767 last_size = new_size;
4768 btrfs_ordered_update_i_size(inode, last_size, NULL);
4771 btrfs_free_path(path);
4773 if (be_nice && bytes_deleted > SZ_32M) {
4774 unsigned long updates = trans->delayed_ref_updates;
4776 trans->delayed_ref_updates = 0;
4777 ret = btrfs_run_delayed_refs(trans, fs_info,
4787 * btrfs_truncate_block - read, zero a chunk and write a block
4788 * @inode - inode that we're zeroing
4789 * @from - the offset to start zeroing
4790 * @len - the length to zero, 0 to zero the entire range respective to the
4792 * @front - zero up to the offset instead of from the offset on
4794 * This will find the block for the "from" offset and cow the block and zero the
4795 * part we want to zero. This is used with truncate and hole punching.
4797 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4800 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4801 struct address_space *mapping = inode->i_mapping;
4802 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4803 struct btrfs_ordered_extent *ordered;
4804 struct extent_state *cached_state = NULL;
4805 struct extent_changeset *data_reserved = NULL;
4807 u32 blocksize = fs_info->sectorsize;
4808 pgoff_t index = from >> PAGE_SHIFT;
4809 unsigned offset = from & (blocksize - 1);
4811 gfp_t mask = btrfs_alloc_write_mask(mapping);
4816 if (IS_ALIGNED(offset, blocksize) &&
4817 (!len || IS_ALIGNED(len, blocksize)))
4820 block_start = round_down(from, blocksize);
4821 block_end = block_start + blocksize - 1;
4823 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4824 block_start, blocksize);
4829 page = find_or_create_page(mapping, index, mask);
4831 btrfs_delalloc_release_space(inode, data_reserved,
4832 block_start, blocksize);
4833 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4838 if (!PageUptodate(page)) {
4839 ret = btrfs_readpage(NULL, page);
4841 if (page->mapping != mapping) {
4846 if (!PageUptodate(page)) {
4851 wait_on_page_writeback(page);
4853 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4854 set_page_extent_mapped(page);
4856 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4858 unlock_extent_cached(io_tree, block_start, block_end,
4862 btrfs_start_ordered_extent(inode, ordered, 1);
4863 btrfs_put_ordered_extent(ordered);
4867 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4868 EXTENT_DIRTY | EXTENT_DELALLOC |
4869 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4870 0, 0, &cached_state);
4872 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4875 unlock_extent_cached(io_tree, block_start, block_end,
4880 if (offset != blocksize) {
4882 len = blocksize - offset;
4885 memset(kaddr + (block_start - page_offset(page)),
4888 memset(kaddr + (block_start - page_offset(page)) + offset,
4890 flush_dcache_page(page);
4893 ClearPageChecked(page);
4894 set_page_dirty(page);
4895 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4899 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4901 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4905 extent_changeset_free(data_reserved);
4909 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4910 u64 offset, u64 len)
4912 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4913 struct btrfs_trans_handle *trans;
4917 * Still need to make sure the inode looks like it's been updated so
4918 * that any holes get logged if we fsync.
4920 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4921 BTRFS_I(inode)->last_trans = fs_info->generation;
4922 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4923 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4928 * 1 - for the one we're dropping
4929 * 1 - for the one we're adding
4930 * 1 - for updating the inode.
4932 trans = btrfs_start_transaction(root, 3);
4934 return PTR_ERR(trans);
4936 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4938 btrfs_abort_transaction(trans, ret);
4939 btrfs_end_transaction(trans);
4943 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4944 offset, 0, 0, len, 0, len, 0, 0, 0);
4946 btrfs_abort_transaction(trans, ret);
4948 btrfs_update_inode(trans, root, inode);
4949 btrfs_end_transaction(trans);
4954 * This function puts in dummy file extents for the area we're creating a hole
4955 * for. So if we are truncating this file to a larger size we need to insert
4956 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4957 * the range between oldsize and size
4959 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4961 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4962 struct btrfs_root *root = BTRFS_I(inode)->root;
4963 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4964 struct extent_map *em = NULL;
4965 struct extent_state *cached_state = NULL;
4966 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4967 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4968 u64 block_end = ALIGN(size, fs_info->sectorsize);
4975 * If our size started in the middle of a block we need to zero out the
4976 * rest of the block before we expand the i_size, otherwise we could
4977 * expose stale data.
4979 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4983 if (size <= hole_start)
4987 struct btrfs_ordered_extent *ordered;
4989 lock_extent_bits(io_tree, hole_start, block_end - 1,
4991 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4992 block_end - hole_start);
4995 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4997 btrfs_start_ordered_extent(inode, ordered, 1);
4998 btrfs_put_ordered_extent(ordered);
5001 cur_offset = hole_start;
5003 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5004 block_end - cur_offset, 0);
5010 last_byte = min(extent_map_end(em), block_end);
5011 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5012 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5013 struct extent_map *hole_em;
5014 hole_size = last_byte - cur_offset;
5016 err = maybe_insert_hole(root, inode, cur_offset,
5020 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5021 cur_offset + hole_size - 1, 0);
5022 hole_em = alloc_extent_map();
5024 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5025 &BTRFS_I(inode)->runtime_flags);
5028 hole_em->start = cur_offset;
5029 hole_em->len = hole_size;
5030 hole_em->orig_start = cur_offset;
5032 hole_em->block_start = EXTENT_MAP_HOLE;
5033 hole_em->block_len = 0;
5034 hole_em->orig_block_len = 0;
5035 hole_em->ram_bytes = hole_size;
5036 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5037 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5038 hole_em->generation = fs_info->generation;
5041 write_lock(&em_tree->lock);
5042 err = add_extent_mapping(em_tree, hole_em, 1);
5043 write_unlock(&em_tree->lock);
5046 btrfs_drop_extent_cache(BTRFS_I(inode),
5051 free_extent_map(hole_em);
5054 free_extent_map(em);
5056 cur_offset = last_byte;
5057 if (cur_offset >= block_end)
5060 free_extent_map(em);
5061 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5065 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5067 struct btrfs_root *root = BTRFS_I(inode)->root;
5068 struct btrfs_trans_handle *trans;
5069 loff_t oldsize = i_size_read(inode);
5070 loff_t newsize = attr->ia_size;
5071 int mask = attr->ia_valid;
5075 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5076 * special case where we need to update the times despite not having
5077 * these flags set. For all other operations the VFS set these flags
5078 * explicitly if it wants a timestamp update.
5080 if (newsize != oldsize) {
5081 inode_inc_iversion(inode);
5082 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5083 inode->i_ctime = inode->i_mtime =
5084 current_time(inode);
5087 if (newsize > oldsize) {
5089 * Don't do an expanding truncate while snapshotting is ongoing.
5090 * This is to ensure the snapshot captures a fully consistent
5091 * state of this file - if the snapshot captures this expanding
5092 * truncation, it must capture all writes that happened before
5095 btrfs_wait_for_snapshot_creation(root);
5096 ret = btrfs_cont_expand(inode, oldsize, newsize);
5098 btrfs_end_write_no_snapshotting(root);
5102 trans = btrfs_start_transaction(root, 1);
5103 if (IS_ERR(trans)) {
5104 btrfs_end_write_no_snapshotting(root);
5105 return PTR_ERR(trans);
5108 i_size_write(inode, newsize);
5109 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5110 pagecache_isize_extended(inode, oldsize, newsize);
5111 ret = btrfs_update_inode(trans, root, inode);
5112 btrfs_end_write_no_snapshotting(root);
5113 btrfs_end_transaction(trans);
5117 * We're truncating a file that used to have good data down to
5118 * zero. Make sure it gets into the ordered flush list so that
5119 * any new writes get down to disk quickly.
5122 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5123 &BTRFS_I(inode)->runtime_flags);
5126 * 1 for the orphan item we're going to add
5127 * 1 for the orphan item deletion.
5129 trans = btrfs_start_transaction(root, 2);
5131 return PTR_ERR(trans);
5134 * We need to do this in case we fail at _any_ point during the
5135 * actual truncate. Once we do the truncate_setsize we could
5136 * invalidate pages which forces any outstanding ordered io to
5137 * be instantly completed which will give us extents that need
5138 * to be truncated. If we fail to get an orphan inode down we
5139 * could have left over extents that were never meant to live,
5140 * so we need to guarantee from this point on that everything
5141 * will be consistent.
5143 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5144 btrfs_end_transaction(trans);
5148 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5149 truncate_setsize(inode, newsize);
5151 /* Disable nonlocked read DIO to avoid the end less truncate */
5152 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5153 inode_dio_wait(inode);
5154 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5156 ret = btrfs_truncate(inode, newsize == oldsize);
5157 if (ret && inode->i_nlink) {
5160 /* To get a stable disk_i_size */
5161 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5163 btrfs_orphan_del(NULL, BTRFS_I(inode));
5168 * failed to truncate, disk_i_size is only adjusted down
5169 * as we remove extents, so it should represent the true
5170 * size of the inode, so reset the in memory size and
5171 * delete our orphan entry.
5173 trans = btrfs_join_transaction(root);
5174 if (IS_ERR(trans)) {
5175 btrfs_orphan_del(NULL, BTRFS_I(inode));
5178 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5179 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5181 btrfs_abort_transaction(trans, err);
5182 btrfs_end_transaction(trans);
5189 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5191 struct inode *inode = d_inode(dentry);
5192 struct btrfs_root *root = BTRFS_I(inode)->root;
5195 if (btrfs_root_readonly(root))
5198 err = setattr_prepare(dentry, attr);
5202 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5203 err = btrfs_setsize(inode, attr);
5208 if (attr->ia_valid) {
5209 setattr_copy(inode, attr);
5210 inode_inc_iversion(inode);
5211 err = btrfs_dirty_inode(inode);
5213 if (!err && attr->ia_valid & ATTR_MODE)
5214 err = posix_acl_chmod(inode, inode->i_mode);
5221 * While truncating the inode pages during eviction, we get the VFS calling
5222 * btrfs_invalidatepage() against each page of the inode. This is slow because
5223 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5224 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5225 * extent_state structures over and over, wasting lots of time.
5227 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5228 * those expensive operations on a per page basis and do only the ordered io
5229 * finishing, while we release here the extent_map and extent_state structures,
5230 * without the excessive merging and splitting.
5232 static void evict_inode_truncate_pages(struct inode *inode)
5234 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5235 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5236 struct rb_node *node;
5238 ASSERT(inode->i_state & I_FREEING);
5239 truncate_inode_pages_final(&inode->i_data);
5241 write_lock(&map_tree->lock);
5242 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5243 struct extent_map *em;
5245 node = rb_first(&map_tree->map);
5246 em = rb_entry(node, struct extent_map, rb_node);
5247 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5248 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5249 remove_extent_mapping(map_tree, em);
5250 free_extent_map(em);
5251 if (need_resched()) {
5252 write_unlock(&map_tree->lock);
5254 write_lock(&map_tree->lock);
5257 write_unlock(&map_tree->lock);
5260 * Keep looping until we have no more ranges in the io tree.
5261 * We can have ongoing bios started by readpages (called from readahead)
5262 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5263 * still in progress (unlocked the pages in the bio but did not yet
5264 * unlocked the ranges in the io tree). Therefore this means some
5265 * ranges can still be locked and eviction started because before
5266 * submitting those bios, which are executed by a separate task (work
5267 * queue kthread), inode references (inode->i_count) were not taken
5268 * (which would be dropped in the end io callback of each bio).
5269 * Therefore here we effectively end up waiting for those bios and
5270 * anyone else holding locked ranges without having bumped the inode's
5271 * reference count - if we don't do it, when they access the inode's
5272 * io_tree to unlock a range it may be too late, leading to an
5273 * use-after-free issue.
5275 spin_lock(&io_tree->lock);
5276 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5277 struct extent_state *state;
5278 struct extent_state *cached_state = NULL;
5282 node = rb_first(&io_tree->state);
5283 state = rb_entry(node, struct extent_state, rb_node);
5284 start = state->start;
5286 spin_unlock(&io_tree->lock);
5288 lock_extent_bits(io_tree, start, end, &cached_state);
5291 * If still has DELALLOC flag, the extent didn't reach disk,
5292 * and its reserved space won't be freed by delayed_ref.
5293 * So we need to free its reserved space here.
5294 * (Refer to comment in btrfs_invalidatepage, case 2)
5296 * Note, end is the bytenr of last byte, so we need + 1 here.
5298 if (state->state & EXTENT_DELALLOC)
5299 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5301 clear_extent_bit(io_tree, start, end,
5302 EXTENT_LOCKED | EXTENT_DIRTY |
5303 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5304 EXTENT_DEFRAG, 1, 1, &cached_state);
5307 spin_lock(&io_tree->lock);
5309 spin_unlock(&io_tree->lock);
5312 void btrfs_evict_inode(struct inode *inode)
5314 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5315 struct btrfs_trans_handle *trans;
5316 struct btrfs_root *root = BTRFS_I(inode)->root;
5317 struct btrfs_block_rsv *rsv, *global_rsv;
5318 int steal_from_global = 0;
5322 trace_btrfs_inode_evict(inode);
5329 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5331 evict_inode_truncate_pages(inode);
5333 if (inode->i_nlink &&
5334 ((btrfs_root_refs(&root->root_item) != 0 &&
5335 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5336 btrfs_is_free_space_inode(BTRFS_I(inode))))
5339 if (is_bad_inode(inode)) {
5340 btrfs_orphan_del(NULL, BTRFS_I(inode));
5343 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5344 if (!special_file(inode->i_mode))
5345 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5347 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5349 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5350 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5351 &BTRFS_I(inode)->runtime_flags));
5355 if (inode->i_nlink > 0) {
5356 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5357 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5361 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5363 btrfs_orphan_del(NULL, BTRFS_I(inode));
5367 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5369 btrfs_orphan_del(NULL, BTRFS_I(inode));
5372 rsv->size = min_size;
5374 global_rsv = &fs_info->global_block_rsv;
5376 btrfs_i_size_write(BTRFS_I(inode), 0);
5379 * This is a bit simpler than btrfs_truncate since we've already
5380 * reserved our space for our orphan item in the unlink, so we just
5381 * need to reserve some slack space in case we add bytes and update
5382 * inode item when doing the truncate.
5385 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5386 BTRFS_RESERVE_FLUSH_LIMIT);
5389 * Try and steal from the global reserve since we will
5390 * likely not use this space anyway, we want to try as
5391 * hard as possible to get this to work.
5394 steal_from_global++;
5396 steal_from_global = 0;
5400 * steal_from_global == 0: we reserved stuff, hooray!
5401 * steal_from_global == 1: we didn't reserve stuff, boo!
5402 * steal_from_global == 2: we've committed, still not a lot of
5403 * room but maybe we'll have room in the global reserve this
5405 * steal_from_global == 3: abandon all hope!
5407 if (steal_from_global > 2) {
5409 "Could not get space for a delete, will truncate on mount %d",
5411 btrfs_orphan_del(NULL, BTRFS_I(inode));
5412 btrfs_free_block_rsv(fs_info, rsv);
5416 trans = btrfs_join_transaction(root);
5417 if (IS_ERR(trans)) {
5418 btrfs_orphan_del(NULL, BTRFS_I(inode));
5419 btrfs_free_block_rsv(fs_info, rsv);
5424 * We can't just steal from the global reserve, we need to make
5425 * sure there is room to do it, if not we need to commit and try
5428 if (steal_from_global) {
5429 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5430 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5437 * Couldn't steal from the global reserve, we have too much
5438 * pending stuff built up, commit the transaction and try it
5442 ret = btrfs_commit_transaction(trans);
5444 btrfs_orphan_del(NULL, BTRFS_I(inode));
5445 btrfs_free_block_rsv(fs_info, rsv);
5450 steal_from_global = 0;
5453 trans->block_rsv = rsv;
5455 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5456 if (ret != -ENOSPC && ret != -EAGAIN)
5459 trans->block_rsv = &fs_info->trans_block_rsv;
5460 btrfs_end_transaction(trans);
5462 btrfs_btree_balance_dirty(fs_info);
5465 btrfs_free_block_rsv(fs_info, rsv);
5468 * Errors here aren't a big deal, it just means we leave orphan items
5469 * in the tree. They will be cleaned up on the next mount.
5472 trans->block_rsv = root->orphan_block_rsv;
5473 btrfs_orphan_del(trans, BTRFS_I(inode));
5475 btrfs_orphan_del(NULL, BTRFS_I(inode));
5478 trans->block_rsv = &fs_info->trans_block_rsv;
5479 if (!(root == fs_info->tree_root ||
5480 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5481 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5483 btrfs_end_transaction(trans);
5484 btrfs_btree_balance_dirty(fs_info);
5486 btrfs_remove_delayed_node(BTRFS_I(inode));
5491 * this returns the key found in the dir entry in the location pointer.
5492 * If no dir entries were found, returns -ENOENT.
5493 * If found a corrupted location in dir entry, returns -EUCLEAN.
5495 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5496 struct btrfs_key *location)
5498 const char *name = dentry->d_name.name;
5499 int namelen = dentry->d_name.len;
5500 struct btrfs_dir_item *di;
5501 struct btrfs_path *path;
5502 struct btrfs_root *root = BTRFS_I(dir)->root;
5505 path = btrfs_alloc_path();
5509 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5520 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5521 if (location->type != BTRFS_INODE_ITEM_KEY &&
5522 location->type != BTRFS_ROOT_ITEM_KEY) {
5524 btrfs_warn(root->fs_info,
5525 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5526 __func__, name, btrfs_ino(BTRFS_I(dir)),
5527 location->objectid, location->type, location->offset);
5530 btrfs_free_path(path);
5535 * when we hit a tree root in a directory, the btrfs part of the inode
5536 * needs to be changed to reflect the root directory of the tree root. This
5537 * is kind of like crossing a mount point.
5539 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5541 struct dentry *dentry,
5542 struct btrfs_key *location,
5543 struct btrfs_root **sub_root)
5545 struct btrfs_path *path;
5546 struct btrfs_root *new_root;
5547 struct btrfs_root_ref *ref;
5548 struct extent_buffer *leaf;
5549 struct btrfs_key key;
5553 path = btrfs_alloc_path();
5560 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5561 key.type = BTRFS_ROOT_REF_KEY;
5562 key.offset = location->objectid;
5564 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5571 leaf = path->nodes[0];
5572 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5573 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5574 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5577 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5578 (unsigned long)(ref + 1),
5579 dentry->d_name.len);
5583 btrfs_release_path(path);
5585 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5586 if (IS_ERR(new_root)) {
5587 err = PTR_ERR(new_root);
5591 *sub_root = new_root;
5592 location->objectid = btrfs_root_dirid(&new_root->root_item);
5593 location->type = BTRFS_INODE_ITEM_KEY;
5594 location->offset = 0;
5597 btrfs_free_path(path);
5601 static void inode_tree_add(struct inode *inode)
5603 struct btrfs_root *root = BTRFS_I(inode)->root;
5604 struct btrfs_inode *entry;
5606 struct rb_node *parent;
5607 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5608 u64 ino = btrfs_ino(BTRFS_I(inode));
5610 if (inode_unhashed(inode))
5613 spin_lock(&root->inode_lock);
5614 p = &root->inode_tree.rb_node;
5617 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5619 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5620 p = &parent->rb_left;
5621 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5622 p = &parent->rb_right;
5624 WARN_ON(!(entry->vfs_inode.i_state &
5625 (I_WILL_FREE | I_FREEING)));
5626 rb_replace_node(parent, new, &root->inode_tree);
5627 RB_CLEAR_NODE(parent);
5628 spin_unlock(&root->inode_lock);
5632 rb_link_node(new, parent, p);
5633 rb_insert_color(new, &root->inode_tree);
5634 spin_unlock(&root->inode_lock);
5637 static void inode_tree_del(struct inode *inode)
5639 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5640 struct btrfs_root *root = BTRFS_I(inode)->root;
5643 spin_lock(&root->inode_lock);
5644 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5645 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5646 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5647 empty = RB_EMPTY_ROOT(&root->inode_tree);
5649 spin_unlock(&root->inode_lock);
5651 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5652 synchronize_srcu(&fs_info->subvol_srcu);
5653 spin_lock(&root->inode_lock);
5654 empty = RB_EMPTY_ROOT(&root->inode_tree);
5655 spin_unlock(&root->inode_lock);
5657 btrfs_add_dead_root(root);
5661 void btrfs_invalidate_inodes(struct btrfs_root *root)
5663 struct btrfs_fs_info *fs_info = root->fs_info;
5664 struct rb_node *node;
5665 struct rb_node *prev;
5666 struct btrfs_inode *entry;
5667 struct inode *inode;
5670 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5671 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5673 spin_lock(&root->inode_lock);
5675 node = root->inode_tree.rb_node;
5679 entry = rb_entry(node, struct btrfs_inode, rb_node);
5681 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5682 node = node->rb_left;
5683 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5684 node = node->rb_right;
5690 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5691 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5695 prev = rb_next(prev);
5699 entry = rb_entry(node, struct btrfs_inode, rb_node);
5700 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5701 inode = igrab(&entry->vfs_inode);
5703 spin_unlock(&root->inode_lock);
5704 if (atomic_read(&inode->i_count) > 1)
5705 d_prune_aliases(inode);
5707 * btrfs_drop_inode will have it removed from
5708 * the inode cache when its usage count
5713 spin_lock(&root->inode_lock);
5717 if (cond_resched_lock(&root->inode_lock))
5720 node = rb_next(node);
5722 spin_unlock(&root->inode_lock);
5725 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5727 struct btrfs_iget_args *args = p;
5728 inode->i_ino = args->location->objectid;
5729 memcpy(&BTRFS_I(inode)->location, args->location,
5730 sizeof(*args->location));
5731 BTRFS_I(inode)->root = args->root;
5735 static int btrfs_find_actor(struct inode *inode, void *opaque)
5737 struct btrfs_iget_args *args = opaque;
5738 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5739 args->root == BTRFS_I(inode)->root;
5742 static struct inode *btrfs_iget_locked(struct super_block *s,
5743 struct btrfs_key *location,
5744 struct btrfs_root *root)
5746 struct inode *inode;
5747 struct btrfs_iget_args args;
5748 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5750 args.location = location;
5753 inode = iget5_locked(s, hashval, btrfs_find_actor,
5754 btrfs_init_locked_inode,
5759 /* Get an inode object given its location and corresponding root.
5760 * Returns in *is_new if the inode was read from disk
5762 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5763 struct btrfs_root *root, int *new)
5765 struct inode *inode;
5767 inode = btrfs_iget_locked(s, location, root);
5769 return ERR_PTR(-ENOMEM);
5771 if (inode->i_state & I_NEW) {
5774 ret = btrfs_read_locked_inode(inode);
5775 if (!is_bad_inode(inode)) {
5776 inode_tree_add(inode);
5777 unlock_new_inode(inode);
5781 unlock_new_inode(inode);
5784 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5791 static struct inode *new_simple_dir(struct super_block *s,
5792 struct btrfs_key *key,
5793 struct btrfs_root *root)
5795 struct inode *inode = new_inode(s);
5798 return ERR_PTR(-ENOMEM);
5800 BTRFS_I(inode)->root = root;
5801 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5802 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5804 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5805 inode->i_op = &btrfs_dir_ro_inode_operations;
5806 inode->i_opflags &= ~IOP_XATTR;
5807 inode->i_fop = &simple_dir_operations;
5808 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5809 inode->i_mtime = current_time(inode);
5810 inode->i_atime = inode->i_mtime;
5811 inode->i_ctime = inode->i_mtime;
5812 BTRFS_I(inode)->i_otime = inode->i_mtime;
5817 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5819 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5820 struct inode *inode;
5821 struct btrfs_root *root = BTRFS_I(dir)->root;
5822 struct btrfs_root *sub_root = root;
5823 struct btrfs_key location;
5827 if (dentry->d_name.len > BTRFS_NAME_LEN)
5828 return ERR_PTR(-ENAMETOOLONG);
5830 ret = btrfs_inode_by_name(dir, dentry, &location);
5832 return ERR_PTR(ret);
5834 if (location.type == BTRFS_INODE_ITEM_KEY) {
5835 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5839 index = srcu_read_lock(&fs_info->subvol_srcu);
5840 ret = fixup_tree_root_location(fs_info, dir, dentry,
5841 &location, &sub_root);
5844 inode = ERR_PTR(ret);
5846 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5848 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5850 srcu_read_unlock(&fs_info->subvol_srcu, index);
5852 if (!IS_ERR(inode) && root != sub_root) {
5853 down_read(&fs_info->cleanup_work_sem);
5854 if (!sb_rdonly(inode->i_sb))
5855 ret = btrfs_orphan_cleanup(sub_root);
5856 up_read(&fs_info->cleanup_work_sem);
5859 inode = ERR_PTR(ret);
5866 static int btrfs_dentry_delete(const struct dentry *dentry)
5868 struct btrfs_root *root;
5869 struct inode *inode = d_inode(dentry);
5871 if (!inode && !IS_ROOT(dentry))
5872 inode = d_inode(dentry->d_parent);
5875 root = BTRFS_I(inode)->root;
5876 if (btrfs_root_refs(&root->root_item) == 0)
5879 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5885 static void btrfs_dentry_release(struct dentry *dentry)
5887 kfree(dentry->d_fsdata);
5890 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5893 struct inode *inode;
5895 inode = btrfs_lookup_dentry(dir, dentry);
5896 if (IS_ERR(inode)) {
5897 if (PTR_ERR(inode) == -ENOENT)
5900 return ERR_CAST(inode);
5903 return d_splice_alias(inode, dentry);
5906 unsigned char btrfs_filetype_table[] = {
5907 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5911 * All this infrastructure exists because dir_emit can fault, and we are holding
5912 * the tree lock when doing readdir. For now just allocate a buffer and copy
5913 * our information into that, and then dir_emit from the buffer. This is
5914 * similar to what NFS does, only we don't keep the buffer around in pagecache
5915 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5916 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5919 static int btrfs_opendir(struct inode *inode, struct file *file)
5921 struct btrfs_file_private *private;
5923 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5926 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5927 if (!private->filldir_buf) {
5931 file->private_data = private;
5942 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5945 struct dir_entry *entry = addr;
5946 char *name = (char *)(entry + 1);
5948 ctx->pos = entry->offset;
5949 if (!dir_emit(ctx, name, entry->name_len, entry->ino,
5952 addr += sizeof(struct dir_entry) + entry->name_len;
5958 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5960 struct inode *inode = file_inode(file);
5961 struct btrfs_root *root = BTRFS_I(inode)->root;
5962 struct btrfs_file_private *private = file->private_data;
5963 struct btrfs_dir_item *di;
5964 struct btrfs_key key;
5965 struct btrfs_key found_key;
5966 struct btrfs_path *path;
5968 struct list_head ins_list;
5969 struct list_head del_list;
5971 struct extent_buffer *leaf;
5978 struct btrfs_key location;
5980 if (!dir_emit_dots(file, ctx))
5983 path = btrfs_alloc_path();
5987 addr = private->filldir_buf;
5988 path->reada = READA_FORWARD;
5990 INIT_LIST_HEAD(&ins_list);
5991 INIT_LIST_HEAD(&del_list);
5992 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5995 key.type = BTRFS_DIR_INDEX_KEY;
5996 key.offset = ctx->pos;
5997 key.objectid = btrfs_ino(BTRFS_I(inode));
5999 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6004 struct dir_entry *entry;
6006 leaf = path->nodes[0];
6007 slot = path->slots[0];
6008 if (slot >= btrfs_header_nritems(leaf)) {
6009 ret = btrfs_next_leaf(root, path);
6017 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6019 if (found_key.objectid != key.objectid)
6021 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6023 if (found_key.offset < ctx->pos)
6025 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6027 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
6028 name_len = btrfs_dir_name_len(leaf, di);
6029 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6031 btrfs_release_path(path);
6032 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6035 addr = private->filldir_buf;
6042 entry->name_len = name_len;
6043 name_ptr = (char *)(entry + 1);
6044 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
6046 entry->type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
6047 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6048 entry->ino = location.objectid;
6049 entry->offset = found_key.offset;
6051 addr += sizeof(struct dir_entry) + name_len;
6052 total_len += sizeof(struct dir_entry) + name_len;
6056 btrfs_release_path(path);
6058 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6062 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6067 * Stop new entries from being returned after we return the last
6070 * New directory entries are assigned a strictly increasing
6071 * offset. This means that new entries created during readdir
6072 * are *guaranteed* to be seen in the future by that readdir.
6073 * This has broken buggy programs which operate on names as
6074 * they're returned by readdir. Until we re-use freed offsets
6075 * we have this hack to stop new entries from being returned
6076 * under the assumption that they'll never reach this huge
6079 * This is being careful not to overflow 32bit loff_t unless the
6080 * last entry requires it because doing so has broken 32bit apps
6083 if (ctx->pos >= INT_MAX)
6084 ctx->pos = LLONG_MAX;
6091 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6092 btrfs_free_path(path);
6096 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6098 struct btrfs_root *root = BTRFS_I(inode)->root;
6099 struct btrfs_trans_handle *trans;
6101 bool nolock = false;
6103 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6106 if (btrfs_fs_closing(root->fs_info) &&
6107 btrfs_is_free_space_inode(BTRFS_I(inode)))
6110 if (wbc->sync_mode == WB_SYNC_ALL) {
6112 trans = btrfs_join_transaction_nolock(root);
6114 trans = btrfs_join_transaction(root);
6116 return PTR_ERR(trans);
6117 ret = btrfs_commit_transaction(trans);
6123 * This is somewhat expensive, updating the tree every time the
6124 * inode changes. But, it is most likely to find the inode in cache.
6125 * FIXME, needs more benchmarking...there are no reasons other than performance
6126 * to keep or drop this code.
6128 static int btrfs_dirty_inode(struct inode *inode)
6130 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6131 struct btrfs_root *root = BTRFS_I(inode)->root;
6132 struct btrfs_trans_handle *trans;
6135 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6138 trans = btrfs_join_transaction(root);
6140 return PTR_ERR(trans);
6142 ret = btrfs_update_inode(trans, root, inode);
6143 if (ret && ret == -ENOSPC) {
6144 /* whoops, lets try again with the full transaction */
6145 btrfs_end_transaction(trans);
6146 trans = btrfs_start_transaction(root, 1);
6148 return PTR_ERR(trans);
6150 ret = btrfs_update_inode(trans, root, inode);
6152 btrfs_end_transaction(trans);
6153 if (BTRFS_I(inode)->delayed_node)
6154 btrfs_balance_delayed_items(fs_info);
6160 * This is a copy of file_update_time. We need this so we can return error on
6161 * ENOSPC for updating the inode in the case of file write and mmap writes.
6163 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6166 struct btrfs_root *root = BTRFS_I(inode)->root;
6167 bool dirty = flags & ~S_VERSION;
6169 if (btrfs_root_readonly(root))
6172 if (flags & S_VERSION)
6173 dirty |= inode_maybe_inc_iversion(inode, dirty);
6174 if (flags & S_CTIME)
6175 inode->i_ctime = *now;
6176 if (flags & S_MTIME)
6177 inode->i_mtime = *now;
6178 if (flags & S_ATIME)
6179 inode->i_atime = *now;
6180 return dirty ? btrfs_dirty_inode(inode) : 0;
6184 * find the highest existing sequence number in a directory
6185 * and then set the in-memory index_cnt variable to reflect
6186 * free sequence numbers
6188 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6190 struct btrfs_root *root = inode->root;
6191 struct btrfs_key key, found_key;
6192 struct btrfs_path *path;
6193 struct extent_buffer *leaf;
6196 key.objectid = btrfs_ino(inode);
6197 key.type = BTRFS_DIR_INDEX_KEY;
6198 key.offset = (u64)-1;
6200 path = btrfs_alloc_path();
6204 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6207 /* FIXME: we should be able to handle this */
6213 * MAGIC NUMBER EXPLANATION:
6214 * since we search a directory based on f_pos we have to start at 2
6215 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6216 * else has to start at 2
6218 if (path->slots[0] == 0) {
6219 inode->index_cnt = 2;
6225 leaf = path->nodes[0];
6226 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6228 if (found_key.objectid != btrfs_ino(inode) ||
6229 found_key.type != BTRFS_DIR_INDEX_KEY) {
6230 inode->index_cnt = 2;
6234 inode->index_cnt = found_key.offset + 1;
6236 btrfs_free_path(path);
6241 * helper to find a free sequence number in a given directory. This current
6242 * code is very simple, later versions will do smarter things in the btree
6244 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6248 if (dir->index_cnt == (u64)-1) {
6249 ret = btrfs_inode_delayed_dir_index_count(dir);
6251 ret = btrfs_set_inode_index_count(dir);
6257 *index = dir->index_cnt;
6263 static int btrfs_insert_inode_locked(struct inode *inode)
6265 struct btrfs_iget_args args;
6266 args.location = &BTRFS_I(inode)->location;
6267 args.root = BTRFS_I(inode)->root;
6269 return insert_inode_locked4(inode,
6270 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6271 btrfs_find_actor, &args);
6275 * Inherit flags from the parent inode.
6277 * Currently only the compression flags and the cow flags are inherited.
6279 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6286 flags = BTRFS_I(dir)->flags;
6288 if (flags & BTRFS_INODE_NOCOMPRESS) {
6289 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6290 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6291 } else if (flags & BTRFS_INODE_COMPRESS) {
6292 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6293 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6296 if (flags & BTRFS_INODE_NODATACOW) {
6297 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6298 if (S_ISREG(inode->i_mode))
6299 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6302 btrfs_update_iflags(inode);
6305 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6306 struct btrfs_root *root,
6308 const char *name, int name_len,
6309 u64 ref_objectid, u64 objectid,
6310 umode_t mode, u64 *index)
6312 struct btrfs_fs_info *fs_info = root->fs_info;
6313 struct inode *inode;
6314 struct btrfs_inode_item *inode_item;
6315 struct btrfs_key *location;
6316 struct btrfs_path *path;
6317 struct btrfs_inode_ref *ref;
6318 struct btrfs_key key[2];
6320 int nitems = name ? 2 : 1;
6324 path = btrfs_alloc_path();
6326 return ERR_PTR(-ENOMEM);
6328 inode = new_inode(fs_info->sb);
6330 btrfs_free_path(path);
6331 return ERR_PTR(-ENOMEM);
6335 * O_TMPFILE, set link count to 0, so that after this point,
6336 * we fill in an inode item with the correct link count.
6339 set_nlink(inode, 0);
6342 * we have to initialize this early, so we can reclaim the inode
6343 * number if we fail afterwards in this function.
6345 inode->i_ino = objectid;
6348 trace_btrfs_inode_request(dir);
6350 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6352 btrfs_free_path(path);
6354 return ERR_PTR(ret);
6360 * index_cnt is ignored for everything but a dir,
6361 * btrfs_set_inode_index_count has an explanation for the magic
6364 BTRFS_I(inode)->index_cnt = 2;
6365 BTRFS_I(inode)->dir_index = *index;
6366 BTRFS_I(inode)->root = root;
6367 BTRFS_I(inode)->generation = trans->transid;
6368 inode->i_generation = BTRFS_I(inode)->generation;
6371 * We could have gotten an inode number from somebody who was fsynced
6372 * and then removed in this same transaction, so let's just set full
6373 * sync since it will be a full sync anyway and this will blow away the
6374 * old info in the log.
6376 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6378 key[0].objectid = objectid;
6379 key[0].type = BTRFS_INODE_ITEM_KEY;
6382 sizes[0] = sizeof(struct btrfs_inode_item);
6386 * Start new inodes with an inode_ref. This is slightly more
6387 * efficient for small numbers of hard links since they will
6388 * be packed into one item. Extended refs will kick in if we
6389 * add more hard links than can fit in the ref item.
6391 key[1].objectid = objectid;
6392 key[1].type = BTRFS_INODE_REF_KEY;
6393 key[1].offset = ref_objectid;
6395 sizes[1] = name_len + sizeof(*ref);
6398 location = &BTRFS_I(inode)->location;
6399 location->objectid = objectid;
6400 location->offset = 0;
6401 location->type = BTRFS_INODE_ITEM_KEY;
6403 ret = btrfs_insert_inode_locked(inode);
6407 path->leave_spinning = 1;
6408 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6412 inode_init_owner(inode, dir, mode);
6413 inode_set_bytes(inode, 0);
6415 inode->i_mtime = current_time(inode);
6416 inode->i_atime = inode->i_mtime;
6417 inode->i_ctime = inode->i_mtime;
6418 BTRFS_I(inode)->i_otime = inode->i_mtime;
6420 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6421 struct btrfs_inode_item);
6422 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6423 sizeof(*inode_item));
6424 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6427 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6428 struct btrfs_inode_ref);
6429 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6430 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6431 ptr = (unsigned long)(ref + 1);
6432 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6435 btrfs_mark_buffer_dirty(path->nodes[0]);
6436 btrfs_free_path(path);
6438 btrfs_inherit_iflags(inode, dir);
6440 if (S_ISREG(mode)) {
6441 if (btrfs_test_opt(fs_info, NODATASUM))
6442 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6443 if (btrfs_test_opt(fs_info, NODATACOW))
6444 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6445 BTRFS_INODE_NODATASUM;
6448 inode_tree_add(inode);
6450 trace_btrfs_inode_new(inode);
6451 btrfs_set_inode_last_trans(trans, inode);
6453 btrfs_update_root_times(trans, root);
6455 ret = btrfs_inode_inherit_props(trans, inode, dir);
6458 "error inheriting props for ino %llu (root %llu): %d",
6459 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6464 unlock_new_inode(inode);
6467 BTRFS_I(dir)->index_cnt--;
6468 btrfs_free_path(path);
6470 return ERR_PTR(ret);
6473 static inline u8 btrfs_inode_type(struct inode *inode)
6475 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6479 * utility function to add 'inode' into 'parent_inode' with
6480 * a give name and a given sequence number.
6481 * if 'add_backref' is true, also insert a backref from the
6482 * inode to the parent directory.
6484 int btrfs_add_link(struct btrfs_trans_handle *trans,
6485 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6486 const char *name, int name_len, int add_backref, u64 index)
6488 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6490 struct btrfs_key key;
6491 struct btrfs_root *root = parent_inode->root;
6492 u64 ino = btrfs_ino(inode);
6493 u64 parent_ino = btrfs_ino(parent_inode);
6495 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6496 memcpy(&key, &inode->root->root_key, sizeof(key));
6499 key.type = BTRFS_INODE_ITEM_KEY;
6503 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6504 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6505 root->root_key.objectid, parent_ino,
6506 index, name, name_len);
6507 } else if (add_backref) {
6508 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6512 /* Nothing to clean up yet */
6516 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6518 btrfs_inode_type(&inode->vfs_inode), index);
6519 if (ret == -EEXIST || ret == -EOVERFLOW)
6522 btrfs_abort_transaction(trans, ret);
6526 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6528 inode_inc_iversion(&parent_inode->vfs_inode);
6529 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6530 current_time(&parent_inode->vfs_inode);
6531 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6533 btrfs_abort_transaction(trans, ret);
6537 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6540 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6541 root->root_key.objectid, parent_ino,
6542 &local_index, name, name_len);
6544 } else if (add_backref) {
6548 err = btrfs_del_inode_ref(trans, root, name, name_len,
6549 ino, parent_ino, &local_index);
6554 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6555 struct btrfs_inode *dir, struct dentry *dentry,
6556 struct btrfs_inode *inode, int backref, u64 index)
6558 int err = btrfs_add_link(trans, dir, inode,
6559 dentry->d_name.name, dentry->d_name.len,
6566 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6567 umode_t mode, dev_t rdev)
6569 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6570 struct btrfs_trans_handle *trans;
6571 struct btrfs_root *root = BTRFS_I(dir)->root;
6572 struct inode *inode = NULL;
6579 * 2 for inode item and ref
6581 * 1 for xattr if selinux is on
6583 trans = btrfs_start_transaction(root, 5);
6585 return PTR_ERR(trans);
6587 err = btrfs_find_free_ino(root, &objectid);
6591 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6592 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6594 if (IS_ERR(inode)) {
6595 err = PTR_ERR(inode);
6600 * If the active LSM wants to access the inode during
6601 * d_instantiate it needs these. Smack checks to see
6602 * if the filesystem supports xattrs by looking at the
6605 inode->i_op = &btrfs_special_inode_operations;
6606 init_special_inode(inode, inode->i_mode, rdev);
6608 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6610 goto out_unlock_inode;
6612 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6615 goto out_unlock_inode;
6617 btrfs_update_inode(trans, root, inode);
6618 unlock_new_inode(inode);
6619 d_instantiate(dentry, inode);
6623 btrfs_end_transaction(trans);
6624 btrfs_btree_balance_dirty(fs_info);
6626 inode_dec_link_count(inode);
6633 unlock_new_inode(inode);
6638 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6639 umode_t mode, bool excl)
6641 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6642 struct btrfs_trans_handle *trans;
6643 struct btrfs_root *root = BTRFS_I(dir)->root;
6644 struct inode *inode = NULL;
6645 int drop_inode_on_err = 0;
6651 * 2 for inode item and ref
6653 * 1 for xattr if selinux is on
6655 trans = btrfs_start_transaction(root, 5);
6657 return PTR_ERR(trans);
6659 err = btrfs_find_free_ino(root, &objectid);
6663 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6664 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6666 if (IS_ERR(inode)) {
6667 err = PTR_ERR(inode);
6670 drop_inode_on_err = 1;
6672 * If the active LSM wants to access the inode during
6673 * d_instantiate it needs these. Smack checks to see
6674 * if the filesystem supports xattrs by looking at the
6677 inode->i_fop = &btrfs_file_operations;
6678 inode->i_op = &btrfs_file_inode_operations;
6679 inode->i_mapping->a_ops = &btrfs_aops;
6681 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6683 goto out_unlock_inode;
6685 err = btrfs_update_inode(trans, root, inode);
6687 goto out_unlock_inode;
6689 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6692 goto out_unlock_inode;
6694 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6695 unlock_new_inode(inode);
6696 d_instantiate(dentry, inode);
6699 btrfs_end_transaction(trans);
6700 if (err && drop_inode_on_err) {
6701 inode_dec_link_count(inode);
6704 btrfs_btree_balance_dirty(fs_info);
6708 unlock_new_inode(inode);
6713 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6714 struct dentry *dentry)
6716 struct btrfs_trans_handle *trans = NULL;
6717 struct btrfs_root *root = BTRFS_I(dir)->root;
6718 struct inode *inode = d_inode(old_dentry);
6719 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6724 /* do not allow sys_link's with other subvols of the same device */
6725 if (root->objectid != BTRFS_I(inode)->root->objectid)
6728 if (inode->i_nlink >= BTRFS_LINK_MAX)
6731 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6736 * 2 items for inode and inode ref
6737 * 2 items for dir items
6738 * 1 item for parent inode
6740 trans = btrfs_start_transaction(root, 5);
6741 if (IS_ERR(trans)) {
6742 err = PTR_ERR(trans);
6747 /* There are several dir indexes for this inode, clear the cache. */
6748 BTRFS_I(inode)->dir_index = 0ULL;
6750 inode_inc_iversion(inode);
6751 inode->i_ctime = current_time(inode);
6753 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6755 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6761 struct dentry *parent = dentry->d_parent;
6762 err = btrfs_update_inode(trans, root, inode);
6765 if (inode->i_nlink == 1) {
6767 * If new hard link count is 1, it's a file created
6768 * with open(2) O_TMPFILE flag.
6770 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6774 d_instantiate(dentry, inode);
6775 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6780 btrfs_end_transaction(trans);
6782 inode_dec_link_count(inode);
6785 btrfs_btree_balance_dirty(fs_info);
6789 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6791 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6792 struct inode *inode = NULL;
6793 struct btrfs_trans_handle *trans;
6794 struct btrfs_root *root = BTRFS_I(dir)->root;
6796 int drop_on_err = 0;
6801 * 2 items for inode and ref
6802 * 2 items for dir items
6803 * 1 for xattr if selinux is on
6805 trans = btrfs_start_transaction(root, 5);
6807 return PTR_ERR(trans);
6809 err = btrfs_find_free_ino(root, &objectid);
6813 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6814 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6815 S_IFDIR | mode, &index);
6816 if (IS_ERR(inode)) {
6817 err = PTR_ERR(inode);
6822 /* these must be set before we unlock the inode */
6823 inode->i_op = &btrfs_dir_inode_operations;
6824 inode->i_fop = &btrfs_dir_file_operations;
6826 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6828 goto out_fail_inode;
6830 btrfs_i_size_write(BTRFS_I(inode), 0);
6831 err = btrfs_update_inode(trans, root, inode);
6833 goto out_fail_inode;
6835 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6836 dentry->d_name.name,
6837 dentry->d_name.len, 0, index);
6839 goto out_fail_inode;
6841 d_instantiate(dentry, inode);
6843 * mkdir is special. We're unlocking after we call d_instantiate
6844 * to avoid a race with nfsd calling d_instantiate.
6846 unlock_new_inode(inode);
6850 btrfs_end_transaction(trans);
6852 inode_dec_link_count(inode);
6855 btrfs_btree_balance_dirty(fs_info);
6859 unlock_new_inode(inode);
6863 static noinline int uncompress_inline(struct btrfs_path *path,
6865 size_t pg_offset, u64 extent_offset,
6866 struct btrfs_file_extent_item *item)
6869 struct extent_buffer *leaf = path->nodes[0];
6872 unsigned long inline_size;
6876 WARN_ON(pg_offset != 0);
6877 compress_type = btrfs_file_extent_compression(leaf, item);
6878 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6879 inline_size = btrfs_file_extent_inline_item_len(leaf,
6880 btrfs_item_nr(path->slots[0]));
6881 tmp = kmalloc(inline_size, GFP_NOFS);
6884 ptr = btrfs_file_extent_inline_start(item);
6886 read_extent_buffer(leaf, tmp, ptr, inline_size);
6888 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6889 ret = btrfs_decompress(compress_type, tmp, page,
6890 extent_offset, inline_size, max_size);
6893 * decompression code contains a memset to fill in any space between the end
6894 * of the uncompressed data and the end of max_size in case the decompressed
6895 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6896 * the end of an inline extent and the beginning of the next block, so we
6897 * cover that region here.
6900 if (max_size + pg_offset < PAGE_SIZE) {
6901 char *map = kmap(page);
6902 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6910 * a bit scary, this does extent mapping from logical file offset to the disk.
6911 * the ugly parts come from merging extents from the disk with the in-ram
6912 * representation. This gets more complex because of the data=ordered code,
6913 * where the in-ram extents might be locked pending data=ordered completion.
6915 * This also copies inline extents directly into the page.
6917 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6919 size_t pg_offset, u64 start, u64 len,
6922 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6925 u64 extent_start = 0;
6927 u64 objectid = btrfs_ino(inode);
6929 struct btrfs_path *path = NULL;
6930 struct btrfs_root *root = inode->root;
6931 struct btrfs_file_extent_item *item;
6932 struct extent_buffer *leaf;
6933 struct btrfs_key found_key;
6934 struct extent_map *em = NULL;
6935 struct extent_map_tree *em_tree = &inode->extent_tree;
6936 struct extent_io_tree *io_tree = &inode->io_tree;
6937 const bool new_inline = !page || create;
6939 read_lock(&em_tree->lock);
6940 em = lookup_extent_mapping(em_tree, start, len);
6942 em->bdev = fs_info->fs_devices->latest_bdev;
6943 read_unlock(&em_tree->lock);
6946 if (em->start > start || em->start + em->len <= start)
6947 free_extent_map(em);
6948 else if (em->block_start == EXTENT_MAP_INLINE && page)
6949 free_extent_map(em);
6953 em = alloc_extent_map();
6958 em->bdev = fs_info->fs_devices->latest_bdev;
6959 em->start = EXTENT_MAP_HOLE;
6960 em->orig_start = EXTENT_MAP_HOLE;
6962 em->block_len = (u64)-1;
6965 path = btrfs_alloc_path();
6971 * Chances are we'll be called again, so go ahead and do
6974 path->reada = READA_FORWARD;
6977 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6984 if (path->slots[0] == 0)
6989 leaf = path->nodes[0];
6990 item = btrfs_item_ptr(leaf, path->slots[0],
6991 struct btrfs_file_extent_item);
6992 /* are we inside the extent that was found? */
6993 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6994 found_type = found_key.type;
6995 if (found_key.objectid != objectid ||
6996 found_type != BTRFS_EXTENT_DATA_KEY) {
6998 * If we backup past the first extent we want to move forward
6999 * and see if there is an extent in front of us, otherwise we'll
7000 * say there is a hole for our whole search range which can
7007 found_type = btrfs_file_extent_type(leaf, item);
7008 extent_start = found_key.offset;
7009 if (found_type == BTRFS_FILE_EXTENT_REG ||
7010 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7011 extent_end = extent_start +
7012 btrfs_file_extent_num_bytes(leaf, item);
7014 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
7016 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7018 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7019 extent_end = ALIGN(extent_start + size,
7020 fs_info->sectorsize);
7022 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
7027 if (start >= extent_end) {
7029 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
7030 ret = btrfs_next_leaf(root, path);
7037 leaf = path->nodes[0];
7039 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
7040 if (found_key.objectid != objectid ||
7041 found_key.type != BTRFS_EXTENT_DATA_KEY)
7043 if (start + len <= found_key.offset)
7045 if (start > found_key.offset)
7048 em->orig_start = start;
7049 em->len = found_key.offset - start;
7053 btrfs_extent_item_to_extent_map(inode, path, item,
7056 if (found_type == BTRFS_FILE_EXTENT_REG ||
7057 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7059 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7063 size_t extent_offset;
7069 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7070 extent_offset = page_offset(page) + pg_offset - extent_start;
7071 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7072 size - extent_offset);
7073 em->start = extent_start + extent_offset;
7074 em->len = ALIGN(copy_size, fs_info->sectorsize);
7075 em->orig_block_len = em->len;
7076 em->orig_start = em->start;
7077 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7078 if (!PageUptodate(page)) {
7079 if (btrfs_file_extent_compression(leaf, item) !=
7080 BTRFS_COMPRESS_NONE) {
7081 ret = uncompress_inline(path, page, pg_offset,
7082 extent_offset, item);
7089 read_extent_buffer(leaf, map + pg_offset, ptr,
7091 if (pg_offset + copy_size < PAGE_SIZE) {
7092 memset(map + pg_offset + copy_size, 0,
7093 PAGE_SIZE - pg_offset -
7098 flush_dcache_page(page);
7100 set_extent_uptodate(io_tree, em->start,
7101 extent_map_end(em) - 1, NULL, GFP_NOFS);
7106 em->orig_start = start;
7109 em->block_start = EXTENT_MAP_HOLE;
7111 btrfs_release_path(path);
7112 if (em->start > start || extent_map_end(em) <= start) {
7114 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7115 em->start, em->len, start, len);
7121 write_lock(&em_tree->lock);
7122 err = btrfs_add_extent_mapping(em_tree, &em, start, len);
7123 write_unlock(&em_tree->lock);
7126 trace_btrfs_get_extent(root, inode, em);
7128 btrfs_free_path(path);
7130 free_extent_map(em);
7131 return ERR_PTR(err);
7133 BUG_ON(!em); /* Error is always set */
7137 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7139 size_t pg_offset, u64 start, u64 len,
7142 struct extent_map *em;
7143 struct extent_map *hole_em = NULL;
7144 u64 range_start = start;
7150 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7154 * If our em maps to:
7156 * - a pre-alloc extent,
7157 * there might actually be delalloc bytes behind it.
7159 if (em->block_start != EXTENT_MAP_HOLE &&
7160 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7165 /* check to see if we've wrapped (len == -1 or similar) */
7174 /* ok, we didn't find anything, lets look for delalloc */
7175 found = count_range_bits(&inode->io_tree, &range_start,
7176 end, len, EXTENT_DELALLOC, 1);
7177 found_end = range_start + found;
7178 if (found_end < range_start)
7179 found_end = (u64)-1;
7182 * we didn't find anything useful, return
7183 * the original results from get_extent()
7185 if (range_start > end || found_end <= start) {
7191 /* adjust the range_start to make sure it doesn't
7192 * go backwards from the start they passed in
7194 range_start = max(start, range_start);
7195 found = found_end - range_start;
7198 u64 hole_start = start;
7201 em = alloc_extent_map();
7207 * when btrfs_get_extent can't find anything it
7208 * returns one huge hole
7210 * make sure what it found really fits our range, and
7211 * adjust to make sure it is based on the start from
7215 u64 calc_end = extent_map_end(hole_em);
7217 if (calc_end <= start || (hole_em->start > end)) {
7218 free_extent_map(hole_em);
7221 hole_start = max(hole_em->start, start);
7222 hole_len = calc_end - hole_start;
7226 if (hole_em && range_start > hole_start) {
7227 /* our hole starts before our delalloc, so we
7228 * have to return just the parts of the hole
7229 * that go until the delalloc starts
7231 em->len = min(hole_len,
7232 range_start - hole_start);
7233 em->start = hole_start;
7234 em->orig_start = hole_start;
7236 * don't adjust block start at all,
7237 * it is fixed at EXTENT_MAP_HOLE
7239 em->block_start = hole_em->block_start;
7240 em->block_len = hole_len;
7241 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7242 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7244 em->start = range_start;
7246 em->orig_start = range_start;
7247 em->block_start = EXTENT_MAP_DELALLOC;
7248 em->block_len = found;
7255 free_extent_map(hole_em);
7257 free_extent_map(em);
7258 return ERR_PTR(err);
7263 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7266 const u64 orig_start,
7267 const u64 block_start,
7268 const u64 block_len,
7269 const u64 orig_block_len,
7270 const u64 ram_bytes,
7273 struct extent_map *em = NULL;
7276 if (type != BTRFS_ORDERED_NOCOW) {
7277 em = create_io_em(inode, start, len, orig_start,
7278 block_start, block_len, orig_block_len,
7280 BTRFS_COMPRESS_NONE, /* compress_type */
7285 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7286 len, block_len, type);
7289 free_extent_map(em);
7290 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7291 start + len - 1, 0);
7300 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7303 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7304 struct btrfs_root *root = BTRFS_I(inode)->root;
7305 struct extent_map *em;
7306 struct btrfs_key ins;
7310 alloc_hint = get_extent_allocation_hint(inode, start, len);
7311 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7312 0, alloc_hint, &ins, 1, 1);
7314 return ERR_PTR(ret);
7316 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7317 ins.objectid, ins.offset, ins.offset,
7318 ins.offset, BTRFS_ORDERED_REGULAR);
7319 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7321 btrfs_free_reserved_extent(fs_info, ins.objectid,
7328 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7329 * block must be cow'd
7331 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7332 u64 *orig_start, u64 *orig_block_len,
7335 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7336 struct btrfs_path *path;
7338 struct extent_buffer *leaf;
7339 struct btrfs_root *root = BTRFS_I(inode)->root;
7340 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7341 struct btrfs_file_extent_item *fi;
7342 struct btrfs_key key;
7349 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7351 path = btrfs_alloc_path();
7355 ret = btrfs_lookup_file_extent(NULL, root, path,
7356 btrfs_ino(BTRFS_I(inode)), offset, 0);
7360 slot = path->slots[0];
7363 /* can't find the item, must cow */
7370 leaf = path->nodes[0];
7371 btrfs_item_key_to_cpu(leaf, &key, slot);
7372 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7373 key.type != BTRFS_EXTENT_DATA_KEY) {
7374 /* not our file or wrong item type, must cow */
7378 if (key.offset > offset) {
7379 /* Wrong offset, must cow */
7383 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7384 found_type = btrfs_file_extent_type(leaf, fi);
7385 if (found_type != BTRFS_FILE_EXTENT_REG &&
7386 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7387 /* not a regular extent, must cow */
7391 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7394 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7395 if (extent_end <= offset)
7398 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7399 if (disk_bytenr == 0)
7402 if (btrfs_file_extent_compression(leaf, fi) ||
7403 btrfs_file_extent_encryption(leaf, fi) ||
7404 btrfs_file_extent_other_encoding(leaf, fi))
7407 backref_offset = btrfs_file_extent_offset(leaf, fi);
7410 *orig_start = key.offset - backref_offset;
7411 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7412 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7415 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7418 num_bytes = min(offset + *len, extent_end) - offset;
7419 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7422 range_end = round_up(offset + num_bytes,
7423 root->fs_info->sectorsize) - 1;
7424 ret = test_range_bit(io_tree, offset, range_end,
7425 EXTENT_DELALLOC, 0, NULL);
7432 btrfs_release_path(path);
7435 * look for other files referencing this extent, if we
7436 * find any we must cow
7439 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7440 key.offset - backref_offset, disk_bytenr);
7447 * adjust disk_bytenr and num_bytes to cover just the bytes
7448 * in this extent we are about to write. If there
7449 * are any csums in that range we have to cow in order
7450 * to keep the csums correct
7452 disk_bytenr += backref_offset;
7453 disk_bytenr += offset - key.offset;
7454 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7457 * all of the above have passed, it is safe to overwrite this extent
7463 btrfs_free_path(path);
7467 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7468 struct extent_state **cached_state, int writing)
7470 struct btrfs_ordered_extent *ordered;
7474 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7477 * We're concerned with the entire range that we're going to be
7478 * doing DIO to, so we need to make sure there's no ordered
7479 * extents in this range.
7481 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7482 lockend - lockstart + 1);
7485 * We need to make sure there are no buffered pages in this
7486 * range either, we could have raced between the invalidate in
7487 * generic_file_direct_write and locking the extent. The
7488 * invalidate needs to happen so that reads after a write do not
7492 (!writing || !filemap_range_has_page(inode->i_mapping,
7493 lockstart, lockend)))
7496 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7501 * If we are doing a DIO read and the ordered extent we
7502 * found is for a buffered write, we can not wait for it
7503 * to complete and retry, because if we do so we can
7504 * deadlock with concurrent buffered writes on page
7505 * locks. This happens only if our DIO read covers more
7506 * than one extent map, if at this point has already
7507 * created an ordered extent for a previous extent map
7508 * and locked its range in the inode's io tree, and a
7509 * concurrent write against that previous extent map's
7510 * range and this range started (we unlock the ranges
7511 * in the io tree only when the bios complete and
7512 * buffered writes always lock pages before attempting
7513 * to lock range in the io tree).
7516 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7517 btrfs_start_ordered_extent(inode, ordered, 1);
7520 btrfs_put_ordered_extent(ordered);
7523 * We could trigger writeback for this range (and wait
7524 * for it to complete) and then invalidate the pages for
7525 * this range (through invalidate_inode_pages2_range()),
7526 * but that can lead us to a deadlock with a concurrent
7527 * call to readpages() (a buffered read or a defrag call
7528 * triggered a readahead) on a page lock due to an
7529 * ordered dio extent we created before but did not have
7530 * yet a corresponding bio submitted (whence it can not
7531 * complete), which makes readpages() wait for that
7532 * ordered extent to complete while holding a lock on
7547 /* The callers of this must take lock_extent() */
7548 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7549 u64 orig_start, u64 block_start,
7550 u64 block_len, u64 orig_block_len,
7551 u64 ram_bytes, int compress_type,
7554 struct extent_map_tree *em_tree;
7555 struct extent_map *em;
7556 struct btrfs_root *root = BTRFS_I(inode)->root;
7559 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7560 type == BTRFS_ORDERED_COMPRESSED ||
7561 type == BTRFS_ORDERED_NOCOW ||
7562 type == BTRFS_ORDERED_REGULAR);
7564 em_tree = &BTRFS_I(inode)->extent_tree;
7565 em = alloc_extent_map();
7567 return ERR_PTR(-ENOMEM);
7570 em->orig_start = orig_start;
7572 em->block_len = block_len;
7573 em->block_start = block_start;
7574 em->bdev = root->fs_info->fs_devices->latest_bdev;
7575 em->orig_block_len = orig_block_len;
7576 em->ram_bytes = ram_bytes;
7577 em->generation = -1;
7578 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7579 if (type == BTRFS_ORDERED_PREALLOC) {
7580 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7581 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7582 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7583 em->compress_type = compress_type;
7587 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7588 em->start + em->len - 1, 0);
7589 write_lock(&em_tree->lock);
7590 ret = add_extent_mapping(em_tree, em, 1);
7591 write_unlock(&em_tree->lock);
7593 * The caller has taken lock_extent(), who could race with us
7596 } while (ret == -EEXIST);
7599 free_extent_map(em);
7600 return ERR_PTR(ret);
7603 /* em got 2 refs now, callers needs to do free_extent_map once. */
7607 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7608 struct buffer_head *bh_result, int create)
7610 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7611 struct extent_map *em;
7612 struct extent_state *cached_state = NULL;
7613 struct btrfs_dio_data *dio_data = NULL;
7614 u64 start = iblock << inode->i_blkbits;
7615 u64 lockstart, lockend;
7616 u64 len = bh_result->b_size;
7617 int unlock_bits = EXTENT_LOCKED;
7621 unlock_bits |= EXTENT_DIRTY;
7623 len = min_t(u64, len, fs_info->sectorsize);
7626 lockend = start + len - 1;
7628 if (current->journal_info) {
7630 * Need to pull our outstanding extents and set journal_info to NULL so
7631 * that anything that needs to check if there's a transaction doesn't get
7634 dio_data = current->journal_info;
7635 current->journal_info = NULL;
7639 * If this errors out it's because we couldn't invalidate pagecache for
7640 * this range and we need to fallback to buffered.
7642 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7648 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7655 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7656 * io. INLINE is special, and we could probably kludge it in here, but
7657 * it's still buffered so for safety lets just fall back to the generic
7660 * For COMPRESSED we _have_ to read the entire extent in so we can
7661 * decompress it, so there will be buffering required no matter what we
7662 * do, so go ahead and fallback to buffered.
7664 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7665 * to buffered IO. Don't blame me, this is the price we pay for using
7668 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7669 em->block_start == EXTENT_MAP_INLINE) {
7670 free_extent_map(em);
7675 /* Just a good old fashioned hole, return */
7676 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7677 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7678 free_extent_map(em);
7683 * We don't allocate a new extent in the following cases
7685 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7687 * 2) The extent is marked as PREALLOC. We're good to go here and can
7688 * just use the extent.
7692 len = min(len, em->len - (start - em->start));
7693 lockstart = start + len;
7697 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7698 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7699 em->block_start != EXTENT_MAP_HOLE)) {
7701 u64 block_start, orig_start, orig_block_len, ram_bytes;
7703 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7704 type = BTRFS_ORDERED_PREALLOC;
7706 type = BTRFS_ORDERED_NOCOW;
7707 len = min(len, em->len - (start - em->start));
7708 block_start = em->block_start + (start - em->start);
7710 if (can_nocow_extent(inode, start, &len, &orig_start,
7711 &orig_block_len, &ram_bytes) == 1 &&
7712 btrfs_inc_nocow_writers(fs_info, block_start)) {
7713 struct extent_map *em2;
7715 em2 = btrfs_create_dio_extent(inode, start, len,
7716 orig_start, block_start,
7717 len, orig_block_len,
7719 btrfs_dec_nocow_writers(fs_info, block_start);
7720 if (type == BTRFS_ORDERED_PREALLOC) {
7721 free_extent_map(em);
7724 if (em2 && IS_ERR(em2)) {
7729 * For inode marked NODATACOW or extent marked PREALLOC,
7730 * use the existing or preallocated extent, so does not
7731 * need to adjust btrfs_space_info's bytes_may_use.
7733 btrfs_free_reserved_data_space_noquota(inode,
7740 * this will cow the extent, reset the len in case we changed
7743 len = bh_result->b_size;
7744 free_extent_map(em);
7745 em = btrfs_new_extent_direct(inode, start, len);
7750 len = min(len, em->len - (start - em->start));
7752 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7754 bh_result->b_size = len;
7755 bh_result->b_bdev = em->bdev;
7756 set_buffer_mapped(bh_result);
7758 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7759 set_buffer_new(bh_result);
7762 * Need to update the i_size under the extent lock so buffered
7763 * readers will get the updated i_size when we unlock.
7765 if (!dio_data->overwrite && start + len > i_size_read(inode))
7766 i_size_write(inode, start + len);
7768 WARN_ON(dio_data->reserve < len);
7769 dio_data->reserve -= len;
7770 dio_data->unsubmitted_oe_range_end = start + len;
7771 current->journal_info = dio_data;
7775 * In the case of write we need to clear and unlock the entire range,
7776 * in the case of read we need to unlock only the end area that we
7777 * aren't using if there is any left over space.
7779 if (lockstart < lockend) {
7780 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7781 lockend, unlock_bits, 1, 0,
7784 free_extent_state(cached_state);
7787 free_extent_map(em);
7792 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7793 unlock_bits, 1, 0, &cached_state);
7796 current->journal_info = dio_data;
7800 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7804 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7807 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7809 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7813 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7818 static int btrfs_check_dio_repairable(struct inode *inode,
7819 struct bio *failed_bio,
7820 struct io_failure_record *failrec,
7823 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7826 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7827 if (num_copies == 1) {
7829 * we only have a single copy of the data, so don't bother with
7830 * all the retry and error correction code that follows. no
7831 * matter what the error is, it is very likely to persist.
7833 btrfs_debug(fs_info,
7834 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7835 num_copies, failrec->this_mirror, failed_mirror);
7839 failrec->failed_mirror = failed_mirror;
7840 failrec->this_mirror++;
7841 if (failrec->this_mirror == failed_mirror)
7842 failrec->this_mirror++;
7844 if (failrec->this_mirror > num_copies) {
7845 btrfs_debug(fs_info,
7846 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7847 num_copies, failrec->this_mirror, failed_mirror);
7854 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7855 struct page *page, unsigned int pgoff,
7856 u64 start, u64 end, int failed_mirror,
7857 bio_end_io_t *repair_endio, void *repair_arg)
7859 struct io_failure_record *failrec;
7860 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7861 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7864 unsigned int read_mode = 0;
7867 blk_status_t status;
7868 struct bio_vec bvec;
7870 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7872 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7874 return errno_to_blk_status(ret);
7876 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7879 free_io_failure(failure_tree, io_tree, failrec);
7880 return BLK_STS_IOERR;
7883 segs = bio_segments(failed_bio);
7884 bio_get_first_bvec(failed_bio, &bvec);
7886 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7887 read_mode |= REQ_FAILFAST_DEV;
7889 isector = start - btrfs_io_bio(failed_bio)->logical;
7890 isector >>= inode->i_sb->s_blocksize_bits;
7891 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7892 pgoff, isector, repair_endio, repair_arg);
7893 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7895 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7896 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7897 read_mode, failrec->this_mirror, failrec->in_validation);
7899 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7901 free_io_failure(failure_tree, io_tree, failrec);
7908 struct btrfs_retry_complete {
7909 struct completion done;
7910 struct inode *inode;
7915 static void btrfs_retry_endio_nocsum(struct bio *bio)
7917 struct btrfs_retry_complete *done = bio->bi_private;
7918 struct inode *inode = done->inode;
7919 struct bio_vec *bvec;
7920 struct extent_io_tree *io_tree, *failure_tree;
7926 ASSERT(bio->bi_vcnt == 1);
7927 io_tree = &BTRFS_I(inode)->io_tree;
7928 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7929 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7932 ASSERT(!bio_flagged(bio, BIO_CLONED));
7933 bio_for_each_segment_all(bvec, bio, i)
7934 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7935 io_tree, done->start, bvec->bv_page,
7936 btrfs_ino(BTRFS_I(inode)), 0);
7938 complete(&done->done);
7942 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7943 struct btrfs_io_bio *io_bio)
7945 struct btrfs_fs_info *fs_info;
7946 struct bio_vec bvec;
7947 struct bvec_iter iter;
7948 struct btrfs_retry_complete done;
7954 blk_status_t err = BLK_STS_OK;
7956 fs_info = BTRFS_I(inode)->root->fs_info;
7957 sectorsize = fs_info->sectorsize;
7959 start = io_bio->logical;
7961 io_bio->bio.bi_iter = io_bio->iter;
7963 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7964 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7965 pgoff = bvec.bv_offset;
7967 next_block_or_try_again:
7970 init_completion(&done.done);
7972 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7973 pgoff, start, start + sectorsize - 1,
7975 btrfs_retry_endio_nocsum, &done);
7981 wait_for_completion_io(&done.done);
7983 if (!done.uptodate) {
7984 /* We might have another mirror, so try again */
7985 goto next_block_or_try_again;
7989 start += sectorsize;
7993 pgoff += sectorsize;
7994 ASSERT(pgoff < PAGE_SIZE);
7995 goto next_block_or_try_again;
8002 static void btrfs_retry_endio(struct bio *bio)
8004 struct btrfs_retry_complete *done = bio->bi_private;
8005 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8006 struct extent_io_tree *io_tree, *failure_tree;
8007 struct inode *inode = done->inode;
8008 struct bio_vec *bvec;
8018 ASSERT(bio->bi_vcnt == 1);
8019 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
8021 io_tree = &BTRFS_I(inode)->io_tree;
8022 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8024 ASSERT(!bio_flagged(bio, BIO_CLONED));
8025 bio_for_each_segment_all(bvec, bio, i) {
8026 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8027 bvec->bv_offset, done->start,
8030 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8031 failure_tree, io_tree, done->start,
8033 btrfs_ino(BTRFS_I(inode)),
8039 done->uptodate = uptodate;
8041 complete(&done->done);
8045 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8046 struct btrfs_io_bio *io_bio, blk_status_t err)
8048 struct btrfs_fs_info *fs_info;
8049 struct bio_vec bvec;
8050 struct bvec_iter iter;
8051 struct btrfs_retry_complete done;
8058 bool uptodate = (err == 0);
8060 blk_status_t status;
8062 fs_info = BTRFS_I(inode)->root->fs_info;
8063 sectorsize = fs_info->sectorsize;
8066 start = io_bio->logical;
8068 io_bio->bio.bi_iter = io_bio->iter;
8070 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8071 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8073 pgoff = bvec.bv_offset;
8076 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8077 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8078 bvec.bv_page, pgoff, start, sectorsize);
8085 init_completion(&done.done);
8087 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8088 pgoff, start, start + sectorsize - 1,
8089 io_bio->mirror_num, btrfs_retry_endio,
8096 wait_for_completion_io(&done.done);
8098 if (!done.uptodate) {
8099 /* We might have another mirror, so try again */
8103 offset += sectorsize;
8104 start += sectorsize;
8110 pgoff += sectorsize;
8111 ASSERT(pgoff < PAGE_SIZE);
8119 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8120 struct btrfs_io_bio *io_bio, blk_status_t err)
8122 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8126 return __btrfs_correct_data_nocsum(inode, io_bio);
8130 return __btrfs_subio_endio_read(inode, io_bio, err);
8134 static void btrfs_endio_direct_read(struct bio *bio)
8136 struct btrfs_dio_private *dip = bio->bi_private;
8137 struct inode *inode = dip->inode;
8138 struct bio *dio_bio;
8139 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8140 blk_status_t err = bio->bi_status;
8142 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8143 err = btrfs_subio_endio_read(inode, io_bio, err);
8145 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8146 dip->logical_offset + dip->bytes - 1);
8147 dio_bio = dip->dio_bio;
8151 dio_bio->bi_status = err;
8152 dio_end_io(dio_bio);
8155 io_bio->end_io(io_bio, blk_status_to_errno(err));
8159 static void __endio_write_update_ordered(struct inode *inode,
8160 const u64 offset, const u64 bytes,
8161 const bool uptodate)
8163 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8164 struct btrfs_ordered_extent *ordered = NULL;
8165 struct btrfs_workqueue *wq;
8166 btrfs_work_func_t func;
8167 u64 ordered_offset = offset;
8168 u64 ordered_bytes = bytes;
8172 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8173 wq = fs_info->endio_freespace_worker;
8174 func = btrfs_freespace_write_helper;
8176 wq = fs_info->endio_write_workers;
8177 func = btrfs_endio_write_helper;
8181 last_offset = ordered_offset;
8182 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8189 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8190 btrfs_queue_work(wq, &ordered->work);
8193 * If btrfs_dec_test_ordered_pending does not find any ordered extent
8194 * in the range, we can exit.
8196 if (ordered_offset == last_offset)
8199 * our bio might span multiple ordered extents. If we haven't
8200 * completed the accounting for the whole dio, go back and try again
8202 if (ordered_offset < offset + bytes) {
8203 ordered_bytes = offset + bytes - ordered_offset;
8209 static void btrfs_endio_direct_write(struct bio *bio)
8211 struct btrfs_dio_private *dip = bio->bi_private;
8212 struct bio *dio_bio = dip->dio_bio;
8214 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8215 dip->bytes, !bio->bi_status);
8219 dio_bio->bi_status = bio->bi_status;
8220 dio_end_io(dio_bio);
8224 static blk_status_t __btrfs_submit_bio_start_direct_io(void *private_data,
8225 struct bio *bio, int mirror_num,
8226 unsigned long bio_flags, u64 offset)
8228 struct inode *inode = private_data;
8230 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8231 BUG_ON(ret); /* -ENOMEM */
8235 static void btrfs_end_dio_bio(struct bio *bio)
8237 struct btrfs_dio_private *dip = bio->bi_private;
8238 blk_status_t err = bio->bi_status;
8241 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8242 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8243 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8245 (unsigned long long)bio->bi_iter.bi_sector,
8246 bio->bi_iter.bi_size, err);
8248 if (dip->subio_endio)
8249 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8253 * We want to perceive the errors flag being set before
8254 * decrementing the reference count. We don't need a barrier
8255 * since atomic operations with a return value are fully
8256 * ordered as per atomic_t.txt
8261 /* if there are more bios still pending for this dio, just exit */
8262 if (!atomic_dec_and_test(&dip->pending_bios))
8266 bio_io_error(dip->orig_bio);
8268 dip->dio_bio->bi_status = BLK_STS_OK;
8269 bio_endio(dip->orig_bio);
8275 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8276 struct btrfs_dio_private *dip,
8280 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8281 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8285 * We load all the csum data we need when we submit
8286 * the first bio to reduce the csum tree search and
8289 if (dip->logical_offset == file_offset) {
8290 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8296 if (bio == dip->orig_bio)
8299 file_offset -= dip->logical_offset;
8300 file_offset >>= inode->i_sb->s_blocksize_bits;
8301 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8306 static inline blk_status_t
8307 __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode, u64 file_offset,
8310 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8311 struct btrfs_dio_private *dip = bio->bi_private;
8312 bool write = bio_op(bio) == REQ_OP_WRITE;
8315 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8317 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8320 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8325 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8328 if (write && async_submit) {
8329 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8331 __btrfs_submit_bio_start_direct_io,
8332 __btrfs_submit_bio_done);
8336 * If we aren't doing async submit, calculate the csum of the
8339 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8343 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8349 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8354 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8356 struct inode *inode = dip->inode;
8357 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8359 struct bio *orig_bio = dip->orig_bio;
8360 u64 start_sector = orig_bio->bi_iter.bi_sector;
8361 u64 file_offset = dip->logical_offset;
8363 int async_submit = 0;
8365 int clone_offset = 0;
8368 blk_status_t status;
8370 map_length = orig_bio->bi_iter.bi_size;
8371 submit_len = map_length;
8372 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8373 &map_length, NULL, 0);
8377 if (map_length >= submit_len) {
8379 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8383 /* async crcs make it difficult to collect full stripe writes. */
8384 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8390 ASSERT(map_length <= INT_MAX);
8391 atomic_inc(&dip->pending_bios);
8393 clone_len = min_t(int, submit_len, map_length);
8396 * This will never fail as it's passing GPF_NOFS and
8397 * the allocation is backed by btrfs_bioset.
8399 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8401 bio->bi_private = dip;
8402 bio->bi_end_io = btrfs_end_dio_bio;
8403 btrfs_io_bio(bio)->logical = file_offset;
8405 ASSERT(submit_len >= clone_len);
8406 submit_len -= clone_len;
8407 if (submit_len == 0)
8411 * Increase the count before we submit the bio so we know
8412 * the end IO handler won't happen before we increase the
8413 * count. Otherwise, the dip might get freed before we're
8414 * done setting it up.
8416 atomic_inc(&dip->pending_bios);
8418 status = __btrfs_submit_dio_bio(bio, inode, file_offset,
8422 atomic_dec(&dip->pending_bios);
8426 clone_offset += clone_len;
8427 start_sector += clone_len >> 9;
8428 file_offset += clone_len;
8430 map_length = submit_len;
8431 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8432 start_sector << 9, &map_length, NULL, 0);
8435 } while (submit_len > 0);
8438 status = __btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8446 * Before atomic variable goto zero, we must make sure dip->errors is
8447 * perceived to be set. This ordering is ensured by the fact that an
8448 * atomic operations with a return value are fully ordered as per
8451 if (atomic_dec_and_test(&dip->pending_bios))
8452 bio_io_error(dip->orig_bio);
8454 /* bio_end_io() will handle error, so we needn't return it */
8458 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8461 struct btrfs_dio_private *dip = NULL;
8462 struct bio *bio = NULL;
8463 struct btrfs_io_bio *io_bio;
8464 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8467 bio = btrfs_bio_clone(dio_bio);
8469 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8475 dip->private = dio_bio->bi_private;
8477 dip->logical_offset = file_offset;
8478 dip->bytes = dio_bio->bi_iter.bi_size;
8479 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8480 bio->bi_private = dip;
8481 dip->orig_bio = bio;
8482 dip->dio_bio = dio_bio;
8483 atomic_set(&dip->pending_bios, 0);
8484 io_bio = btrfs_io_bio(bio);
8485 io_bio->logical = file_offset;
8488 bio->bi_end_io = btrfs_endio_direct_write;
8490 bio->bi_end_io = btrfs_endio_direct_read;
8491 dip->subio_endio = btrfs_subio_endio_read;
8495 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8496 * even if we fail to submit a bio, because in such case we do the
8497 * corresponding error handling below and it must not be done a second
8498 * time by btrfs_direct_IO().
8501 struct btrfs_dio_data *dio_data = current->journal_info;
8503 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8505 dio_data->unsubmitted_oe_range_start =
8506 dio_data->unsubmitted_oe_range_end;
8509 ret = btrfs_submit_direct_hook(dip);
8514 io_bio->end_io(io_bio, ret);
8518 * If we arrived here it means either we failed to submit the dip
8519 * or we either failed to clone the dio_bio or failed to allocate the
8520 * dip. If we cloned the dio_bio and allocated the dip, we can just
8521 * call bio_endio against our io_bio so that we get proper resource
8522 * cleanup if we fail to submit the dip, otherwise, we must do the
8523 * same as btrfs_endio_direct_[write|read] because we can't call these
8524 * callbacks - they require an allocated dip and a clone of dio_bio.
8529 * The end io callbacks free our dip, do the final put on bio
8530 * and all the cleanup and final put for dio_bio (through
8537 __endio_write_update_ordered(inode,
8539 dio_bio->bi_iter.bi_size,
8542 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8543 file_offset + dio_bio->bi_iter.bi_size - 1);
8545 dio_bio->bi_status = BLK_STS_IOERR;
8547 * Releases and cleans up our dio_bio, no need to bio_put()
8548 * nor bio_endio()/bio_io_error() against dio_bio.
8550 dio_end_io(dio_bio);
8557 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8558 const struct iov_iter *iter, loff_t offset)
8562 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8563 ssize_t retval = -EINVAL;
8565 if (offset & blocksize_mask)
8568 if (iov_iter_alignment(iter) & blocksize_mask)
8571 /* If this is a write we don't need to check anymore */
8572 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8575 * Check to make sure we don't have duplicate iov_base's in this
8576 * iovec, if so return EINVAL, otherwise we'll get csum errors
8577 * when reading back.
8579 for (seg = 0; seg < iter->nr_segs; seg++) {
8580 for (i = seg + 1; i < iter->nr_segs; i++) {
8581 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8590 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8592 struct file *file = iocb->ki_filp;
8593 struct inode *inode = file->f_mapping->host;
8594 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8595 struct btrfs_dio_data dio_data = { 0 };
8596 struct extent_changeset *data_reserved = NULL;
8597 loff_t offset = iocb->ki_pos;
8601 bool relock = false;
8604 if (check_direct_IO(fs_info, iter, offset))
8607 inode_dio_begin(inode);
8610 * The generic stuff only does filemap_write_and_wait_range, which
8611 * isn't enough if we've written compressed pages to this area, so
8612 * we need to flush the dirty pages again to make absolutely sure
8613 * that any outstanding dirty pages are on disk.
8615 count = iov_iter_count(iter);
8616 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8617 &BTRFS_I(inode)->runtime_flags))
8618 filemap_fdatawrite_range(inode->i_mapping, offset,
8619 offset + count - 1);
8621 if (iov_iter_rw(iter) == WRITE) {
8623 * If the write DIO is beyond the EOF, we need update
8624 * the isize, but it is protected by i_mutex. So we can
8625 * not unlock the i_mutex at this case.
8627 if (offset + count <= inode->i_size) {
8628 dio_data.overwrite = 1;
8629 inode_unlock(inode);
8631 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8635 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8641 * We need to know how many extents we reserved so that we can
8642 * do the accounting properly if we go over the number we
8643 * originally calculated. Abuse current->journal_info for this.
8645 dio_data.reserve = round_up(count,
8646 fs_info->sectorsize);
8647 dio_data.unsubmitted_oe_range_start = (u64)offset;
8648 dio_data.unsubmitted_oe_range_end = (u64)offset;
8649 current->journal_info = &dio_data;
8650 down_read(&BTRFS_I(inode)->dio_sem);
8651 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8652 &BTRFS_I(inode)->runtime_flags)) {
8653 inode_dio_end(inode);
8654 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8658 ret = __blockdev_direct_IO(iocb, inode,
8659 fs_info->fs_devices->latest_bdev,
8660 iter, btrfs_get_blocks_direct, NULL,
8661 btrfs_submit_direct, flags);
8662 if (iov_iter_rw(iter) == WRITE) {
8663 up_read(&BTRFS_I(inode)->dio_sem);
8664 current->journal_info = NULL;
8665 if (ret < 0 && ret != -EIOCBQUEUED) {
8666 if (dio_data.reserve)
8667 btrfs_delalloc_release_space(inode, data_reserved,
8668 offset, dio_data.reserve);
8670 * On error we might have left some ordered extents
8671 * without submitting corresponding bios for them, so
8672 * cleanup them up to avoid other tasks getting them
8673 * and waiting for them to complete forever.
8675 if (dio_data.unsubmitted_oe_range_start <
8676 dio_data.unsubmitted_oe_range_end)
8677 __endio_write_update_ordered(inode,
8678 dio_data.unsubmitted_oe_range_start,
8679 dio_data.unsubmitted_oe_range_end -
8680 dio_data.unsubmitted_oe_range_start,
8682 } else if (ret >= 0 && (size_t)ret < count)
8683 btrfs_delalloc_release_space(inode, data_reserved,
8684 offset, count - (size_t)ret);
8685 btrfs_delalloc_release_extents(BTRFS_I(inode), count);
8689 inode_dio_end(inode);
8693 extent_changeset_free(data_reserved);
8697 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8699 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8700 __u64 start, __u64 len)
8704 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8708 return extent_fiemap(inode, fieinfo, start, len);
8711 int btrfs_readpage(struct file *file, struct page *page)
8713 struct extent_io_tree *tree;
8714 tree = &BTRFS_I(page->mapping->host)->io_tree;
8715 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8718 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8720 struct inode *inode = page->mapping->host;
8723 if (current->flags & PF_MEMALLOC) {
8724 redirty_page_for_writepage(wbc, page);
8730 * If we are under memory pressure we will call this directly from the
8731 * VM, we need to make sure we have the inode referenced for the ordered
8732 * extent. If not just return like we didn't do anything.
8734 if (!igrab(inode)) {
8735 redirty_page_for_writepage(wbc, page);
8736 return AOP_WRITEPAGE_ACTIVATE;
8738 ret = extent_write_full_page(page, wbc);
8739 btrfs_add_delayed_iput(inode);
8743 static int btrfs_writepages(struct address_space *mapping,
8744 struct writeback_control *wbc)
8746 struct extent_io_tree *tree;
8748 tree = &BTRFS_I(mapping->host)->io_tree;
8749 return extent_writepages(tree, mapping, wbc);
8753 btrfs_readpages(struct file *file, struct address_space *mapping,
8754 struct list_head *pages, unsigned nr_pages)
8756 struct extent_io_tree *tree;
8757 tree = &BTRFS_I(mapping->host)->io_tree;
8758 return extent_readpages(tree, mapping, pages, nr_pages);
8760 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8762 struct extent_io_tree *tree;
8763 struct extent_map_tree *map;
8766 tree = &BTRFS_I(page->mapping->host)->io_tree;
8767 map = &BTRFS_I(page->mapping->host)->extent_tree;
8768 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8770 ClearPagePrivate(page);
8771 set_page_private(page, 0);
8777 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8779 if (PageWriteback(page) || PageDirty(page))
8781 return __btrfs_releasepage(page, gfp_flags);
8784 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8785 unsigned int length)
8787 struct inode *inode = page->mapping->host;
8788 struct extent_io_tree *tree;
8789 struct btrfs_ordered_extent *ordered;
8790 struct extent_state *cached_state = NULL;
8791 u64 page_start = page_offset(page);
8792 u64 page_end = page_start + PAGE_SIZE - 1;
8795 int inode_evicting = inode->i_state & I_FREEING;
8798 * we have the page locked, so new writeback can't start,
8799 * and the dirty bit won't be cleared while we are here.
8801 * Wait for IO on this page so that we can safely clear
8802 * the PagePrivate2 bit and do ordered accounting
8804 wait_on_page_writeback(page);
8806 tree = &BTRFS_I(inode)->io_tree;
8808 btrfs_releasepage(page, GFP_NOFS);
8812 if (!inode_evicting)
8813 lock_extent_bits(tree, page_start, page_end, &cached_state);
8816 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8817 page_end - start + 1);
8819 end = min(page_end, ordered->file_offset + ordered->len - 1);
8821 * IO on this page will never be started, so we need
8822 * to account for any ordered extents now
8824 if (!inode_evicting)
8825 clear_extent_bit(tree, start, end,
8826 EXTENT_DIRTY | EXTENT_DELALLOC |
8827 EXTENT_DELALLOC_NEW |
8828 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8829 EXTENT_DEFRAG, 1, 0, &cached_state);
8831 * whoever cleared the private bit is responsible
8832 * for the finish_ordered_io
8834 if (TestClearPagePrivate2(page)) {
8835 struct btrfs_ordered_inode_tree *tree;
8838 tree = &BTRFS_I(inode)->ordered_tree;
8840 spin_lock_irq(&tree->lock);
8841 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8842 new_len = start - ordered->file_offset;
8843 if (new_len < ordered->truncated_len)
8844 ordered->truncated_len = new_len;
8845 spin_unlock_irq(&tree->lock);
8847 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8849 end - start + 1, 1))
8850 btrfs_finish_ordered_io(ordered);
8852 btrfs_put_ordered_extent(ordered);
8853 if (!inode_evicting) {
8854 cached_state = NULL;
8855 lock_extent_bits(tree, start, end,
8860 if (start < page_end)
8865 * Qgroup reserved space handler
8866 * Page here will be either
8867 * 1) Already written to disk
8868 * In this case, its reserved space is released from data rsv map
8869 * and will be freed by delayed_ref handler finally.
8870 * So even we call qgroup_free_data(), it won't decrease reserved
8872 * 2) Not written to disk
8873 * This means the reserved space should be freed here. However,
8874 * if a truncate invalidates the page (by clearing PageDirty)
8875 * and the page is accounted for while allocating extent
8876 * in btrfs_check_data_free_space() we let delayed_ref to
8877 * free the entire extent.
8879 if (PageDirty(page))
8880 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8881 if (!inode_evicting) {
8882 clear_extent_bit(tree, page_start, page_end,
8883 EXTENT_LOCKED | EXTENT_DIRTY |
8884 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8885 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8888 __btrfs_releasepage(page, GFP_NOFS);
8891 ClearPageChecked(page);
8892 if (PagePrivate(page)) {
8893 ClearPagePrivate(page);
8894 set_page_private(page, 0);
8900 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8901 * called from a page fault handler when a page is first dirtied. Hence we must
8902 * be careful to check for EOF conditions here. We set the page up correctly
8903 * for a written page which means we get ENOSPC checking when writing into
8904 * holes and correct delalloc and unwritten extent mapping on filesystems that
8905 * support these features.
8907 * We are not allowed to take the i_mutex here so we have to play games to
8908 * protect against truncate races as the page could now be beyond EOF. Because
8909 * vmtruncate() writes the inode size before removing pages, once we have the
8910 * page lock we can determine safely if the page is beyond EOF. If it is not
8911 * beyond EOF, then the page is guaranteed safe against truncation until we
8914 int btrfs_page_mkwrite(struct vm_fault *vmf)
8916 struct page *page = vmf->page;
8917 struct inode *inode = file_inode(vmf->vma->vm_file);
8918 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8919 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8920 struct btrfs_ordered_extent *ordered;
8921 struct extent_state *cached_state = NULL;
8922 struct extent_changeset *data_reserved = NULL;
8924 unsigned long zero_start;
8933 reserved_space = PAGE_SIZE;
8935 sb_start_pagefault(inode->i_sb);
8936 page_start = page_offset(page);
8937 page_end = page_start + PAGE_SIZE - 1;
8941 * Reserving delalloc space after obtaining the page lock can lead to
8942 * deadlock. For example, if a dirty page is locked by this function
8943 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8944 * dirty page write out, then the btrfs_writepage() function could
8945 * end up waiting indefinitely to get a lock on the page currently
8946 * being processed by btrfs_page_mkwrite() function.
8948 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8951 ret = file_update_time(vmf->vma->vm_file);
8957 else /* -ENOSPC, -EIO, etc */
8958 ret = VM_FAULT_SIGBUS;
8964 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8967 size = i_size_read(inode);
8969 if ((page->mapping != inode->i_mapping) ||
8970 (page_start >= size)) {
8971 /* page got truncated out from underneath us */
8974 wait_on_page_writeback(page);
8976 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8977 set_page_extent_mapped(page);
8980 * we can't set the delalloc bits if there are pending ordered
8981 * extents. Drop our locks and wait for them to finish
8983 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8986 unlock_extent_cached(io_tree, page_start, page_end,
8989 btrfs_start_ordered_extent(inode, ordered, 1);
8990 btrfs_put_ordered_extent(ordered);
8994 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8995 reserved_space = round_up(size - page_start,
8996 fs_info->sectorsize);
8997 if (reserved_space < PAGE_SIZE) {
8998 end = page_start + reserved_space - 1;
8999 btrfs_delalloc_release_space(inode, data_reserved,
9000 page_start, PAGE_SIZE - reserved_space);
9005 * page_mkwrite gets called when the page is firstly dirtied after it's
9006 * faulted in, but write(2) could also dirty a page and set delalloc
9007 * bits, thus in this case for space account reason, we still need to
9008 * clear any delalloc bits within this page range since we have to
9009 * reserve data&meta space before lock_page() (see above comments).
9011 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9012 EXTENT_DIRTY | EXTENT_DELALLOC |
9013 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9014 0, 0, &cached_state);
9016 ret = btrfs_set_extent_delalloc(inode, page_start, end, 0,
9019 unlock_extent_cached(io_tree, page_start, page_end,
9021 ret = VM_FAULT_SIGBUS;
9026 /* page is wholly or partially inside EOF */
9027 if (page_start + PAGE_SIZE > size)
9028 zero_start = size & ~PAGE_MASK;
9030 zero_start = PAGE_SIZE;
9032 if (zero_start != PAGE_SIZE) {
9034 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9035 flush_dcache_page(page);
9038 ClearPageChecked(page);
9039 set_page_dirty(page);
9040 SetPageUptodate(page);
9042 BTRFS_I(inode)->last_trans = fs_info->generation;
9043 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9044 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9046 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
9050 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9051 sb_end_pagefault(inode->i_sb);
9052 extent_changeset_free(data_reserved);
9053 return VM_FAULT_LOCKED;
9057 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
9058 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9061 sb_end_pagefault(inode->i_sb);
9062 extent_changeset_free(data_reserved);
9066 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
9068 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9069 struct btrfs_root *root = BTRFS_I(inode)->root;
9070 struct btrfs_block_rsv *rsv;
9073 struct btrfs_trans_handle *trans;
9074 u64 mask = fs_info->sectorsize - 1;
9075 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9077 if (!skip_writeback) {
9078 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9085 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9086 * 3 things going on here
9088 * 1) We need to reserve space for our orphan item and the space to
9089 * delete our orphan item. Lord knows we don't want to have a dangling
9090 * orphan item because we didn't reserve space to remove it.
9092 * 2) We need to reserve space to update our inode.
9094 * 3) We need to have something to cache all the space that is going to
9095 * be free'd up by the truncate operation, but also have some slack
9096 * space reserved in case it uses space during the truncate (thank you
9097 * very much snapshotting).
9099 * And we need these to all be separate. The fact is we can use a lot of
9100 * space doing the truncate, and we have no earthly idea how much space
9101 * we will use, so we need the truncate reservation to be separate so it
9102 * doesn't end up using space reserved for updating the inode or
9103 * removing the orphan item. We also need to be able to stop the
9104 * transaction and start a new one, which means we need to be able to
9105 * update the inode several times, and we have no idea of knowing how
9106 * many times that will be, so we can't just reserve 1 item for the
9107 * entirety of the operation, so that has to be done separately as well.
9108 * Then there is the orphan item, which does indeed need to be held on
9109 * to for the whole operation, and we need nobody to touch this reserved
9110 * space except the orphan code.
9112 * So that leaves us with
9114 * 1) root->orphan_block_rsv - for the orphan deletion.
9115 * 2) rsv - for the truncate reservation, which we will steal from the
9116 * transaction reservation.
9117 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9118 * updating the inode.
9120 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9123 rsv->size = min_size;
9127 * 1 for the truncate slack space
9128 * 1 for updating the inode.
9130 trans = btrfs_start_transaction(root, 2);
9131 if (IS_ERR(trans)) {
9132 err = PTR_ERR(trans);
9136 /* Migrate the slack space for the truncate to our reserve */
9137 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9142 * So if we truncate and then write and fsync we normally would just
9143 * write the extents that changed, which is a problem if we need to
9144 * first truncate that entire inode. So set this flag so we write out
9145 * all of the extents in the inode to the sync log so we're completely
9148 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9149 trans->block_rsv = rsv;
9152 ret = btrfs_truncate_inode_items(trans, root, inode,
9154 BTRFS_EXTENT_DATA_KEY);
9155 trans->block_rsv = &fs_info->trans_block_rsv;
9156 if (ret != -ENOSPC && ret != -EAGAIN) {
9161 ret = btrfs_update_inode(trans, root, inode);
9167 btrfs_end_transaction(trans);
9168 btrfs_btree_balance_dirty(fs_info);
9170 trans = btrfs_start_transaction(root, 2);
9171 if (IS_ERR(trans)) {
9172 ret = err = PTR_ERR(trans);
9177 btrfs_block_rsv_release(fs_info, rsv, -1);
9178 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9180 BUG_ON(ret); /* shouldn't happen */
9181 trans->block_rsv = rsv;
9185 * We can't call btrfs_truncate_block inside a trans handle as we could
9186 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9187 * we've truncated everything except the last little bit, and can do
9188 * btrfs_truncate_block and then update the disk_i_size.
9190 if (ret == NEED_TRUNCATE_BLOCK) {
9191 btrfs_end_transaction(trans);
9192 btrfs_btree_balance_dirty(fs_info);
9194 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9197 trans = btrfs_start_transaction(root, 1);
9198 if (IS_ERR(trans)) {
9199 ret = PTR_ERR(trans);
9202 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9205 if (ret == 0 && inode->i_nlink > 0) {
9206 trans->block_rsv = root->orphan_block_rsv;
9207 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9213 trans->block_rsv = &fs_info->trans_block_rsv;
9214 ret = btrfs_update_inode(trans, root, inode);
9218 ret = btrfs_end_transaction(trans);
9219 btrfs_btree_balance_dirty(fs_info);
9222 btrfs_free_block_rsv(fs_info, rsv);
9231 * create a new subvolume directory/inode (helper for the ioctl).
9233 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9234 struct btrfs_root *new_root,
9235 struct btrfs_root *parent_root,
9238 struct inode *inode;
9242 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9243 new_dirid, new_dirid,
9244 S_IFDIR | (~current_umask() & S_IRWXUGO),
9247 return PTR_ERR(inode);
9248 inode->i_op = &btrfs_dir_inode_operations;
9249 inode->i_fop = &btrfs_dir_file_operations;
9251 set_nlink(inode, 1);
9252 btrfs_i_size_write(BTRFS_I(inode), 0);
9253 unlock_new_inode(inode);
9255 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9257 btrfs_err(new_root->fs_info,
9258 "error inheriting subvolume %llu properties: %d",
9259 new_root->root_key.objectid, err);
9261 err = btrfs_update_inode(trans, new_root, inode);
9267 struct inode *btrfs_alloc_inode(struct super_block *sb)
9269 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9270 struct btrfs_inode *ei;
9271 struct inode *inode;
9273 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9280 ei->last_sub_trans = 0;
9281 ei->logged_trans = 0;
9282 ei->delalloc_bytes = 0;
9283 ei->new_delalloc_bytes = 0;
9284 ei->defrag_bytes = 0;
9285 ei->disk_i_size = 0;
9288 ei->index_cnt = (u64)-1;
9290 ei->last_unlink_trans = 0;
9291 ei->last_log_commit = 0;
9293 spin_lock_init(&ei->lock);
9294 ei->outstanding_extents = 0;
9295 if (sb->s_magic != BTRFS_TEST_MAGIC)
9296 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9297 BTRFS_BLOCK_RSV_DELALLOC);
9298 ei->runtime_flags = 0;
9299 ei->prop_compress = BTRFS_COMPRESS_NONE;
9300 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9302 ei->delayed_node = NULL;
9304 ei->i_otime.tv_sec = 0;
9305 ei->i_otime.tv_nsec = 0;
9307 inode = &ei->vfs_inode;
9308 extent_map_tree_init(&ei->extent_tree);
9309 extent_io_tree_init(&ei->io_tree, inode);
9310 extent_io_tree_init(&ei->io_failure_tree, inode);
9311 ei->io_tree.track_uptodate = 1;
9312 ei->io_failure_tree.track_uptodate = 1;
9313 atomic_set(&ei->sync_writers, 0);
9314 mutex_init(&ei->log_mutex);
9315 mutex_init(&ei->delalloc_mutex);
9316 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9317 INIT_LIST_HEAD(&ei->delalloc_inodes);
9318 INIT_LIST_HEAD(&ei->delayed_iput);
9319 RB_CLEAR_NODE(&ei->rb_node);
9320 init_rwsem(&ei->dio_sem);
9325 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9326 void btrfs_test_destroy_inode(struct inode *inode)
9328 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9329 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9333 static void btrfs_i_callback(struct rcu_head *head)
9335 struct inode *inode = container_of(head, struct inode, i_rcu);
9336 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9339 void btrfs_destroy_inode(struct inode *inode)
9341 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9342 struct btrfs_ordered_extent *ordered;
9343 struct btrfs_root *root = BTRFS_I(inode)->root;
9345 WARN_ON(!hlist_empty(&inode->i_dentry));
9346 WARN_ON(inode->i_data.nrpages);
9347 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9348 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9349 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9350 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9351 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9352 WARN_ON(BTRFS_I(inode)->csum_bytes);
9353 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9356 * This can happen where we create an inode, but somebody else also
9357 * created the same inode and we need to destroy the one we already
9363 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9364 &BTRFS_I(inode)->runtime_flags)) {
9365 btrfs_info(fs_info, "inode %llu still on the orphan list",
9366 btrfs_ino(BTRFS_I(inode)));
9367 atomic_dec(&root->orphan_inodes);
9371 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9376 "found ordered extent %llu %llu on inode cleanup",
9377 ordered->file_offset, ordered->len);
9378 btrfs_remove_ordered_extent(inode, ordered);
9379 btrfs_put_ordered_extent(ordered);
9380 btrfs_put_ordered_extent(ordered);
9383 btrfs_qgroup_check_reserved_leak(inode);
9384 inode_tree_del(inode);
9385 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9387 call_rcu(&inode->i_rcu, btrfs_i_callback);
9390 int btrfs_drop_inode(struct inode *inode)
9392 struct btrfs_root *root = BTRFS_I(inode)->root;
9397 /* the snap/subvol tree is on deleting */
9398 if (btrfs_root_refs(&root->root_item) == 0)
9401 return generic_drop_inode(inode);
9404 static void init_once(void *foo)
9406 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9408 inode_init_once(&ei->vfs_inode);
9411 void __cold btrfs_destroy_cachep(void)
9414 * Make sure all delayed rcu free inodes are flushed before we
9418 kmem_cache_destroy(btrfs_inode_cachep);
9419 kmem_cache_destroy(btrfs_trans_handle_cachep);
9420 kmem_cache_destroy(btrfs_path_cachep);
9421 kmem_cache_destroy(btrfs_free_space_cachep);
9424 int __init btrfs_init_cachep(void)
9426 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9427 sizeof(struct btrfs_inode), 0,
9428 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9430 if (!btrfs_inode_cachep)
9433 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9434 sizeof(struct btrfs_trans_handle), 0,
9435 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9436 if (!btrfs_trans_handle_cachep)
9439 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9440 sizeof(struct btrfs_path), 0,
9441 SLAB_MEM_SPREAD, NULL);
9442 if (!btrfs_path_cachep)
9445 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9446 sizeof(struct btrfs_free_space), 0,
9447 SLAB_MEM_SPREAD, NULL);
9448 if (!btrfs_free_space_cachep)
9453 btrfs_destroy_cachep();
9457 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9458 u32 request_mask, unsigned int flags)
9461 struct inode *inode = d_inode(path->dentry);
9462 u32 blocksize = inode->i_sb->s_blocksize;
9463 u32 bi_flags = BTRFS_I(inode)->flags;
9465 stat->result_mask |= STATX_BTIME;
9466 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9467 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9468 if (bi_flags & BTRFS_INODE_APPEND)
9469 stat->attributes |= STATX_ATTR_APPEND;
9470 if (bi_flags & BTRFS_INODE_COMPRESS)
9471 stat->attributes |= STATX_ATTR_COMPRESSED;
9472 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9473 stat->attributes |= STATX_ATTR_IMMUTABLE;
9474 if (bi_flags & BTRFS_INODE_NODUMP)
9475 stat->attributes |= STATX_ATTR_NODUMP;
9477 stat->attributes_mask |= (STATX_ATTR_APPEND |
9478 STATX_ATTR_COMPRESSED |
9479 STATX_ATTR_IMMUTABLE |
9482 generic_fillattr(inode, stat);
9483 stat->dev = BTRFS_I(inode)->root->anon_dev;
9485 spin_lock(&BTRFS_I(inode)->lock);
9486 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9487 spin_unlock(&BTRFS_I(inode)->lock);
9488 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9489 ALIGN(delalloc_bytes, blocksize)) >> 9;
9493 static int btrfs_rename_exchange(struct inode *old_dir,
9494 struct dentry *old_dentry,
9495 struct inode *new_dir,
9496 struct dentry *new_dentry)
9498 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9499 struct btrfs_trans_handle *trans;
9500 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9501 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9502 struct inode *new_inode = new_dentry->d_inode;
9503 struct inode *old_inode = old_dentry->d_inode;
9504 struct timespec ctime = current_time(old_inode);
9505 struct dentry *parent;
9506 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9507 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9512 bool root_log_pinned = false;
9513 bool dest_log_pinned = false;
9515 /* we only allow rename subvolume link between subvolumes */
9516 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9519 /* close the race window with snapshot create/destroy ioctl */
9520 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9521 down_read(&fs_info->subvol_sem);
9522 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9523 down_read(&fs_info->subvol_sem);
9526 * We want to reserve the absolute worst case amount of items. So if
9527 * both inodes are subvols and we need to unlink them then that would
9528 * require 4 item modifications, but if they are both normal inodes it
9529 * would require 5 item modifications, so we'll assume their normal
9530 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9531 * should cover the worst case number of items we'll modify.
9533 trans = btrfs_start_transaction(root, 12);
9534 if (IS_ERR(trans)) {
9535 ret = PTR_ERR(trans);
9540 * We need to find a free sequence number both in the source and
9541 * in the destination directory for the exchange.
9543 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9546 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9550 BTRFS_I(old_inode)->dir_index = 0ULL;
9551 BTRFS_I(new_inode)->dir_index = 0ULL;
9553 /* Reference for the source. */
9554 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9555 /* force full log commit if subvolume involved. */
9556 btrfs_set_log_full_commit(fs_info, trans);
9558 btrfs_pin_log_trans(root);
9559 root_log_pinned = true;
9560 ret = btrfs_insert_inode_ref(trans, dest,
9561 new_dentry->d_name.name,
9562 new_dentry->d_name.len,
9564 btrfs_ino(BTRFS_I(new_dir)),
9570 /* And now for the dest. */
9571 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9572 /* force full log commit if subvolume involved. */
9573 btrfs_set_log_full_commit(fs_info, trans);
9575 btrfs_pin_log_trans(dest);
9576 dest_log_pinned = true;
9577 ret = btrfs_insert_inode_ref(trans, root,
9578 old_dentry->d_name.name,
9579 old_dentry->d_name.len,
9581 btrfs_ino(BTRFS_I(old_dir)),
9587 /* Update inode version and ctime/mtime. */
9588 inode_inc_iversion(old_dir);
9589 inode_inc_iversion(new_dir);
9590 inode_inc_iversion(old_inode);
9591 inode_inc_iversion(new_inode);
9592 old_dir->i_ctime = old_dir->i_mtime = ctime;
9593 new_dir->i_ctime = new_dir->i_mtime = ctime;
9594 old_inode->i_ctime = ctime;
9595 new_inode->i_ctime = ctime;
9597 if (old_dentry->d_parent != new_dentry->d_parent) {
9598 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9599 BTRFS_I(old_inode), 1);
9600 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9601 BTRFS_I(new_inode), 1);
9604 /* src is a subvolume */
9605 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9606 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9607 ret = btrfs_unlink_subvol(trans, root, old_dir,
9609 old_dentry->d_name.name,
9610 old_dentry->d_name.len);
9611 } else { /* src is an inode */
9612 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9613 BTRFS_I(old_dentry->d_inode),
9614 old_dentry->d_name.name,
9615 old_dentry->d_name.len);
9617 ret = btrfs_update_inode(trans, root, old_inode);
9620 btrfs_abort_transaction(trans, ret);
9624 /* dest is a subvolume */
9625 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9626 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9627 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9629 new_dentry->d_name.name,
9630 new_dentry->d_name.len);
9631 } else { /* dest is an inode */
9632 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9633 BTRFS_I(new_dentry->d_inode),
9634 new_dentry->d_name.name,
9635 new_dentry->d_name.len);
9637 ret = btrfs_update_inode(trans, dest, new_inode);
9640 btrfs_abort_transaction(trans, ret);
9644 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9645 new_dentry->d_name.name,
9646 new_dentry->d_name.len, 0, old_idx);
9648 btrfs_abort_transaction(trans, ret);
9652 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9653 old_dentry->d_name.name,
9654 old_dentry->d_name.len, 0, new_idx);
9656 btrfs_abort_transaction(trans, ret);
9660 if (old_inode->i_nlink == 1)
9661 BTRFS_I(old_inode)->dir_index = old_idx;
9662 if (new_inode->i_nlink == 1)
9663 BTRFS_I(new_inode)->dir_index = new_idx;
9665 if (root_log_pinned) {
9666 parent = new_dentry->d_parent;
9667 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9669 btrfs_end_log_trans(root);
9670 root_log_pinned = false;
9672 if (dest_log_pinned) {
9673 parent = old_dentry->d_parent;
9674 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9676 btrfs_end_log_trans(dest);
9677 dest_log_pinned = false;
9681 * If we have pinned a log and an error happened, we unpin tasks
9682 * trying to sync the log and force them to fallback to a transaction
9683 * commit if the log currently contains any of the inodes involved in
9684 * this rename operation (to ensure we do not persist a log with an
9685 * inconsistent state for any of these inodes or leading to any
9686 * inconsistencies when replayed). If the transaction was aborted, the
9687 * abortion reason is propagated to userspace when attempting to commit
9688 * the transaction. If the log does not contain any of these inodes, we
9689 * allow the tasks to sync it.
9691 if (ret && (root_log_pinned || dest_log_pinned)) {
9692 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9693 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9694 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9696 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9697 btrfs_set_log_full_commit(fs_info, trans);
9699 if (root_log_pinned) {
9700 btrfs_end_log_trans(root);
9701 root_log_pinned = false;
9703 if (dest_log_pinned) {
9704 btrfs_end_log_trans(dest);
9705 dest_log_pinned = false;
9708 ret = btrfs_end_transaction(trans);
9710 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9711 up_read(&fs_info->subvol_sem);
9712 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9713 up_read(&fs_info->subvol_sem);
9718 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9719 struct btrfs_root *root,
9721 struct dentry *dentry)
9724 struct inode *inode;
9728 ret = btrfs_find_free_ino(root, &objectid);
9732 inode = btrfs_new_inode(trans, root, dir,
9733 dentry->d_name.name,
9735 btrfs_ino(BTRFS_I(dir)),
9737 S_IFCHR | WHITEOUT_MODE,
9740 if (IS_ERR(inode)) {
9741 ret = PTR_ERR(inode);
9745 inode->i_op = &btrfs_special_inode_operations;
9746 init_special_inode(inode, inode->i_mode,
9749 ret = btrfs_init_inode_security(trans, inode, dir,
9754 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9755 BTRFS_I(inode), 0, index);
9759 ret = btrfs_update_inode(trans, root, inode);
9761 unlock_new_inode(inode);
9763 inode_dec_link_count(inode);
9769 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9770 struct inode *new_dir, struct dentry *new_dentry,
9773 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9774 struct btrfs_trans_handle *trans;
9775 unsigned int trans_num_items;
9776 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9777 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9778 struct inode *new_inode = d_inode(new_dentry);
9779 struct inode *old_inode = d_inode(old_dentry);
9783 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9784 bool log_pinned = false;
9786 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9789 /* we only allow rename subvolume link between subvolumes */
9790 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9793 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9794 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9797 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9798 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9802 /* check for collisions, even if the name isn't there */
9803 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9804 new_dentry->d_name.name,
9805 new_dentry->d_name.len);
9808 if (ret == -EEXIST) {
9810 * eexist without a new_inode */
9811 if (WARN_ON(!new_inode)) {
9815 /* maybe -EOVERFLOW */
9822 * we're using rename to replace one file with another. Start IO on it
9823 * now so we don't add too much work to the end of the transaction
9825 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9826 filemap_flush(old_inode->i_mapping);
9828 /* close the racy window with snapshot create/destroy ioctl */
9829 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9830 down_read(&fs_info->subvol_sem);
9832 * We want to reserve the absolute worst case amount of items. So if
9833 * both inodes are subvols and we need to unlink them then that would
9834 * require 4 item modifications, but if they are both normal inodes it
9835 * would require 5 item modifications, so we'll assume they are normal
9836 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9837 * should cover the worst case number of items we'll modify.
9838 * If our rename has the whiteout flag, we need more 5 units for the
9839 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9840 * when selinux is enabled).
9842 trans_num_items = 11;
9843 if (flags & RENAME_WHITEOUT)
9844 trans_num_items += 5;
9845 trans = btrfs_start_transaction(root, trans_num_items);
9846 if (IS_ERR(trans)) {
9847 ret = PTR_ERR(trans);
9852 btrfs_record_root_in_trans(trans, dest);
9854 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9858 BTRFS_I(old_inode)->dir_index = 0ULL;
9859 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9860 /* force full log commit if subvolume involved. */
9861 btrfs_set_log_full_commit(fs_info, trans);
9863 btrfs_pin_log_trans(root);
9865 ret = btrfs_insert_inode_ref(trans, dest,
9866 new_dentry->d_name.name,
9867 new_dentry->d_name.len,
9869 btrfs_ino(BTRFS_I(new_dir)), index);
9874 inode_inc_iversion(old_dir);
9875 inode_inc_iversion(new_dir);
9876 inode_inc_iversion(old_inode);
9877 old_dir->i_ctime = old_dir->i_mtime =
9878 new_dir->i_ctime = new_dir->i_mtime =
9879 old_inode->i_ctime = current_time(old_dir);
9881 if (old_dentry->d_parent != new_dentry->d_parent)
9882 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9883 BTRFS_I(old_inode), 1);
9885 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9886 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9887 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9888 old_dentry->d_name.name,
9889 old_dentry->d_name.len);
9891 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9892 BTRFS_I(d_inode(old_dentry)),
9893 old_dentry->d_name.name,
9894 old_dentry->d_name.len);
9896 ret = btrfs_update_inode(trans, root, old_inode);
9899 btrfs_abort_transaction(trans, ret);
9904 inode_inc_iversion(new_inode);
9905 new_inode->i_ctime = current_time(new_inode);
9906 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9907 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9908 root_objectid = BTRFS_I(new_inode)->location.objectid;
9909 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9911 new_dentry->d_name.name,
9912 new_dentry->d_name.len);
9913 BUG_ON(new_inode->i_nlink == 0);
9915 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9916 BTRFS_I(d_inode(new_dentry)),
9917 new_dentry->d_name.name,
9918 new_dentry->d_name.len);
9920 if (!ret && new_inode->i_nlink == 0)
9921 ret = btrfs_orphan_add(trans,
9922 BTRFS_I(d_inode(new_dentry)));
9924 btrfs_abort_transaction(trans, ret);
9929 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9930 new_dentry->d_name.name,
9931 new_dentry->d_name.len, 0, index);
9933 btrfs_abort_transaction(trans, ret);
9937 if (old_inode->i_nlink == 1)
9938 BTRFS_I(old_inode)->dir_index = index;
9941 struct dentry *parent = new_dentry->d_parent;
9943 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9945 btrfs_end_log_trans(root);
9949 if (flags & RENAME_WHITEOUT) {
9950 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9954 btrfs_abort_transaction(trans, ret);
9960 * If we have pinned the log and an error happened, we unpin tasks
9961 * trying to sync the log and force them to fallback to a transaction
9962 * commit if the log currently contains any of the inodes involved in
9963 * this rename operation (to ensure we do not persist a log with an
9964 * inconsistent state for any of these inodes or leading to any
9965 * inconsistencies when replayed). If the transaction was aborted, the
9966 * abortion reason is propagated to userspace when attempting to commit
9967 * the transaction. If the log does not contain any of these inodes, we
9968 * allow the tasks to sync it.
9970 if (ret && log_pinned) {
9971 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9972 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9973 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9975 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9976 btrfs_set_log_full_commit(fs_info, trans);
9978 btrfs_end_log_trans(root);
9981 btrfs_end_transaction(trans);
9983 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9984 up_read(&fs_info->subvol_sem);
9989 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9990 struct inode *new_dir, struct dentry *new_dentry,
9993 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9996 if (flags & RENAME_EXCHANGE)
9997 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10000 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10003 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10005 struct btrfs_delalloc_work *delalloc_work;
10006 struct inode *inode;
10008 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10010 inode = delalloc_work->inode;
10011 filemap_flush(inode->i_mapping);
10012 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10013 &BTRFS_I(inode)->runtime_flags))
10014 filemap_flush(inode->i_mapping);
10016 if (delalloc_work->delay_iput)
10017 btrfs_add_delayed_iput(inode);
10020 complete(&delalloc_work->completion);
10023 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10026 struct btrfs_delalloc_work *work;
10028 work = kmalloc(sizeof(*work), GFP_NOFS);
10032 init_completion(&work->completion);
10033 INIT_LIST_HEAD(&work->list);
10034 work->inode = inode;
10035 work->delay_iput = delay_iput;
10036 WARN_ON_ONCE(!inode);
10037 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10038 btrfs_run_delalloc_work, NULL, NULL);
10043 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10045 wait_for_completion(&work->completion);
10050 * some fairly slow code that needs optimization. This walks the list
10051 * of all the inodes with pending delalloc and forces them to disk.
10053 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10056 struct btrfs_inode *binode;
10057 struct inode *inode;
10058 struct btrfs_delalloc_work *work, *next;
10059 struct list_head works;
10060 struct list_head splice;
10063 INIT_LIST_HEAD(&works);
10064 INIT_LIST_HEAD(&splice);
10066 mutex_lock(&root->delalloc_mutex);
10067 spin_lock(&root->delalloc_lock);
10068 list_splice_init(&root->delalloc_inodes, &splice);
10069 while (!list_empty(&splice)) {
10070 binode = list_entry(splice.next, struct btrfs_inode,
10073 list_move_tail(&binode->delalloc_inodes,
10074 &root->delalloc_inodes);
10075 inode = igrab(&binode->vfs_inode);
10077 cond_resched_lock(&root->delalloc_lock);
10080 spin_unlock(&root->delalloc_lock);
10082 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10085 btrfs_add_delayed_iput(inode);
10091 list_add_tail(&work->list, &works);
10092 btrfs_queue_work(root->fs_info->flush_workers,
10095 if (nr != -1 && ret >= nr)
10098 spin_lock(&root->delalloc_lock);
10100 spin_unlock(&root->delalloc_lock);
10103 list_for_each_entry_safe(work, next, &works, list) {
10104 list_del_init(&work->list);
10105 btrfs_wait_and_free_delalloc_work(work);
10108 if (!list_empty_careful(&splice)) {
10109 spin_lock(&root->delalloc_lock);
10110 list_splice_tail(&splice, &root->delalloc_inodes);
10111 spin_unlock(&root->delalloc_lock);
10113 mutex_unlock(&root->delalloc_mutex);
10117 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10119 struct btrfs_fs_info *fs_info = root->fs_info;
10122 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10125 ret = __start_delalloc_inodes(root, delay_iput, -1);
10131 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10134 struct btrfs_root *root;
10135 struct list_head splice;
10138 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10141 INIT_LIST_HEAD(&splice);
10143 mutex_lock(&fs_info->delalloc_root_mutex);
10144 spin_lock(&fs_info->delalloc_root_lock);
10145 list_splice_init(&fs_info->delalloc_roots, &splice);
10146 while (!list_empty(&splice) && nr) {
10147 root = list_first_entry(&splice, struct btrfs_root,
10149 root = btrfs_grab_fs_root(root);
10151 list_move_tail(&root->delalloc_root,
10152 &fs_info->delalloc_roots);
10153 spin_unlock(&fs_info->delalloc_root_lock);
10155 ret = __start_delalloc_inodes(root, delay_iput, nr);
10156 btrfs_put_fs_root(root);
10164 spin_lock(&fs_info->delalloc_root_lock);
10166 spin_unlock(&fs_info->delalloc_root_lock);
10170 if (!list_empty_careful(&splice)) {
10171 spin_lock(&fs_info->delalloc_root_lock);
10172 list_splice_tail(&splice, &fs_info->delalloc_roots);
10173 spin_unlock(&fs_info->delalloc_root_lock);
10175 mutex_unlock(&fs_info->delalloc_root_mutex);
10179 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10180 const char *symname)
10182 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10183 struct btrfs_trans_handle *trans;
10184 struct btrfs_root *root = BTRFS_I(dir)->root;
10185 struct btrfs_path *path;
10186 struct btrfs_key key;
10187 struct inode *inode = NULL;
10189 int drop_inode = 0;
10195 struct btrfs_file_extent_item *ei;
10196 struct extent_buffer *leaf;
10198 name_len = strlen(symname);
10199 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10200 return -ENAMETOOLONG;
10203 * 2 items for inode item and ref
10204 * 2 items for dir items
10205 * 1 item for updating parent inode item
10206 * 1 item for the inline extent item
10207 * 1 item for xattr if selinux is on
10209 trans = btrfs_start_transaction(root, 7);
10211 return PTR_ERR(trans);
10213 err = btrfs_find_free_ino(root, &objectid);
10217 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10218 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10219 objectid, S_IFLNK|S_IRWXUGO, &index);
10220 if (IS_ERR(inode)) {
10221 err = PTR_ERR(inode);
10226 * If the active LSM wants to access the inode during
10227 * d_instantiate it needs these. Smack checks to see
10228 * if the filesystem supports xattrs by looking at the
10231 inode->i_fop = &btrfs_file_operations;
10232 inode->i_op = &btrfs_file_inode_operations;
10233 inode->i_mapping->a_ops = &btrfs_aops;
10234 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10236 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10238 goto out_unlock_inode;
10240 path = btrfs_alloc_path();
10243 goto out_unlock_inode;
10245 key.objectid = btrfs_ino(BTRFS_I(inode));
10247 key.type = BTRFS_EXTENT_DATA_KEY;
10248 datasize = btrfs_file_extent_calc_inline_size(name_len);
10249 err = btrfs_insert_empty_item(trans, root, path, &key,
10252 btrfs_free_path(path);
10253 goto out_unlock_inode;
10255 leaf = path->nodes[0];
10256 ei = btrfs_item_ptr(leaf, path->slots[0],
10257 struct btrfs_file_extent_item);
10258 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10259 btrfs_set_file_extent_type(leaf, ei,
10260 BTRFS_FILE_EXTENT_INLINE);
10261 btrfs_set_file_extent_encryption(leaf, ei, 0);
10262 btrfs_set_file_extent_compression(leaf, ei, 0);
10263 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10264 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10266 ptr = btrfs_file_extent_inline_start(ei);
10267 write_extent_buffer(leaf, symname, ptr, name_len);
10268 btrfs_mark_buffer_dirty(leaf);
10269 btrfs_free_path(path);
10271 inode->i_op = &btrfs_symlink_inode_operations;
10272 inode_nohighmem(inode);
10273 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10274 inode_set_bytes(inode, name_len);
10275 btrfs_i_size_write(BTRFS_I(inode), name_len);
10276 err = btrfs_update_inode(trans, root, inode);
10278 * Last step, add directory indexes for our symlink inode. This is the
10279 * last step to avoid extra cleanup of these indexes if an error happens
10283 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10284 BTRFS_I(inode), 0, index);
10287 goto out_unlock_inode;
10290 unlock_new_inode(inode);
10291 d_instantiate(dentry, inode);
10294 btrfs_end_transaction(trans);
10296 inode_dec_link_count(inode);
10299 btrfs_btree_balance_dirty(fs_info);
10304 unlock_new_inode(inode);
10308 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10309 u64 start, u64 num_bytes, u64 min_size,
10310 loff_t actual_len, u64 *alloc_hint,
10311 struct btrfs_trans_handle *trans)
10313 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10314 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10315 struct extent_map *em;
10316 struct btrfs_root *root = BTRFS_I(inode)->root;
10317 struct btrfs_key ins;
10318 u64 cur_offset = start;
10321 u64 last_alloc = (u64)-1;
10323 bool own_trans = true;
10324 u64 end = start + num_bytes - 1;
10328 while (num_bytes > 0) {
10330 trans = btrfs_start_transaction(root, 3);
10331 if (IS_ERR(trans)) {
10332 ret = PTR_ERR(trans);
10337 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10338 cur_bytes = max(cur_bytes, min_size);
10340 * If we are severely fragmented we could end up with really
10341 * small allocations, so if the allocator is returning small
10342 * chunks lets make its job easier by only searching for those
10345 cur_bytes = min(cur_bytes, last_alloc);
10346 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10347 min_size, 0, *alloc_hint, &ins, 1, 0);
10350 btrfs_end_transaction(trans);
10353 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10355 last_alloc = ins.offset;
10356 ret = insert_reserved_file_extent(trans, inode,
10357 cur_offset, ins.objectid,
10358 ins.offset, ins.offset,
10359 ins.offset, 0, 0, 0,
10360 BTRFS_FILE_EXTENT_PREALLOC);
10362 btrfs_free_reserved_extent(fs_info, ins.objectid,
10364 btrfs_abort_transaction(trans, ret);
10366 btrfs_end_transaction(trans);
10370 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10371 cur_offset + ins.offset -1, 0);
10373 em = alloc_extent_map();
10375 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10376 &BTRFS_I(inode)->runtime_flags);
10380 em->start = cur_offset;
10381 em->orig_start = cur_offset;
10382 em->len = ins.offset;
10383 em->block_start = ins.objectid;
10384 em->block_len = ins.offset;
10385 em->orig_block_len = ins.offset;
10386 em->ram_bytes = ins.offset;
10387 em->bdev = fs_info->fs_devices->latest_bdev;
10388 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10389 em->generation = trans->transid;
10392 write_lock(&em_tree->lock);
10393 ret = add_extent_mapping(em_tree, em, 1);
10394 write_unlock(&em_tree->lock);
10395 if (ret != -EEXIST)
10397 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10398 cur_offset + ins.offset - 1,
10401 free_extent_map(em);
10403 num_bytes -= ins.offset;
10404 cur_offset += ins.offset;
10405 *alloc_hint = ins.objectid + ins.offset;
10407 inode_inc_iversion(inode);
10408 inode->i_ctime = current_time(inode);
10409 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10410 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10411 (actual_len > inode->i_size) &&
10412 (cur_offset > inode->i_size)) {
10413 if (cur_offset > actual_len)
10414 i_size = actual_len;
10416 i_size = cur_offset;
10417 i_size_write(inode, i_size);
10418 btrfs_ordered_update_i_size(inode, i_size, NULL);
10421 ret = btrfs_update_inode(trans, root, inode);
10424 btrfs_abort_transaction(trans, ret);
10426 btrfs_end_transaction(trans);
10431 btrfs_end_transaction(trans);
10433 if (cur_offset < end)
10434 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10435 end - cur_offset + 1);
10439 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10440 u64 start, u64 num_bytes, u64 min_size,
10441 loff_t actual_len, u64 *alloc_hint)
10443 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10444 min_size, actual_len, alloc_hint,
10448 int btrfs_prealloc_file_range_trans(struct inode *inode,
10449 struct btrfs_trans_handle *trans, int mode,
10450 u64 start, u64 num_bytes, u64 min_size,
10451 loff_t actual_len, u64 *alloc_hint)
10453 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10454 min_size, actual_len, alloc_hint, trans);
10457 static int btrfs_set_page_dirty(struct page *page)
10459 return __set_page_dirty_nobuffers(page);
10462 static int btrfs_permission(struct inode *inode, int mask)
10464 struct btrfs_root *root = BTRFS_I(inode)->root;
10465 umode_t mode = inode->i_mode;
10467 if (mask & MAY_WRITE &&
10468 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10469 if (btrfs_root_readonly(root))
10471 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10474 return generic_permission(inode, mask);
10477 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10479 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10480 struct btrfs_trans_handle *trans;
10481 struct btrfs_root *root = BTRFS_I(dir)->root;
10482 struct inode *inode = NULL;
10488 * 5 units required for adding orphan entry
10490 trans = btrfs_start_transaction(root, 5);
10492 return PTR_ERR(trans);
10494 ret = btrfs_find_free_ino(root, &objectid);
10498 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10499 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10500 if (IS_ERR(inode)) {
10501 ret = PTR_ERR(inode);
10506 inode->i_fop = &btrfs_file_operations;
10507 inode->i_op = &btrfs_file_inode_operations;
10509 inode->i_mapping->a_ops = &btrfs_aops;
10510 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10512 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10516 ret = btrfs_update_inode(trans, root, inode);
10519 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10524 * We set number of links to 0 in btrfs_new_inode(), and here we set
10525 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10528 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10530 set_nlink(inode, 1);
10531 unlock_new_inode(inode);
10532 d_tmpfile(dentry, inode);
10533 mark_inode_dirty(inode);
10536 btrfs_end_transaction(trans);
10539 btrfs_btree_balance_dirty(fs_info);
10543 unlock_new_inode(inode);
10548 __attribute__((const))
10549 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10554 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10556 struct inode *inode = private_data;
10557 return btrfs_sb(inode->i_sb);
10560 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10561 u64 start, u64 end)
10563 struct inode *inode = private_data;
10566 isize = i_size_read(inode);
10567 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10568 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10569 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10570 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10574 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10576 struct inode *inode = private_data;
10577 unsigned long index = start >> PAGE_SHIFT;
10578 unsigned long end_index = end >> PAGE_SHIFT;
10581 while (index <= end_index) {
10582 page = find_get_page(inode->i_mapping, index);
10583 ASSERT(page); /* Pages should be in the extent_io_tree */
10584 set_page_writeback(page);
10590 static const struct inode_operations btrfs_dir_inode_operations = {
10591 .getattr = btrfs_getattr,
10592 .lookup = btrfs_lookup,
10593 .create = btrfs_create,
10594 .unlink = btrfs_unlink,
10595 .link = btrfs_link,
10596 .mkdir = btrfs_mkdir,
10597 .rmdir = btrfs_rmdir,
10598 .rename = btrfs_rename2,
10599 .symlink = btrfs_symlink,
10600 .setattr = btrfs_setattr,
10601 .mknod = btrfs_mknod,
10602 .listxattr = btrfs_listxattr,
10603 .permission = btrfs_permission,
10604 .get_acl = btrfs_get_acl,
10605 .set_acl = btrfs_set_acl,
10606 .update_time = btrfs_update_time,
10607 .tmpfile = btrfs_tmpfile,
10609 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10610 .lookup = btrfs_lookup,
10611 .permission = btrfs_permission,
10612 .update_time = btrfs_update_time,
10615 static const struct file_operations btrfs_dir_file_operations = {
10616 .llseek = generic_file_llseek,
10617 .read = generic_read_dir,
10618 .iterate_shared = btrfs_real_readdir,
10619 .open = btrfs_opendir,
10620 .unlocked_ioctl = btrfs_ioctl,
10621 #ifdef CONFIG_COMPAT
10622 .compat_ioctl = btrfs_compat_ioctl,
10624 .release = btrfs_release_file,
10625 .fsync = btrfs_sync_file,
10628 static const struct extent_io_ops btrfs_extent_io_ops = {
10629 /* mandatory callbacks */
10630 .submit_bio_hook = btrfs_submit_bio_hook,
10631 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10632 .merge_bio_hook = btrfs_merge_bio_hook,
10633 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10634 .tree_fs_info = iotree_fs_info,
10635 .set_range_writeback = btrfs_set_range_writeback,
10637 /* optional callbacks */
10638 .fill_delalloc = run_delalloc_range,
10639 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10640 .writepage_start_hook = btrfs_writepage_start_hook,
10641 .set_bit_hook = btrfs_set_bit_hook,
10642 .clear_bit_hook = btrfs_clear_bit_hook,
10643 .merge_extent_hook = btrfs_merge_extent_hook,
10644 .split_extent_hook = btrfs_split_extent_hook,
10645 .check_extent_io_range = btrfs_check_extent_io_range,
10649 * btrfs doesn't support the bmap operation because swapfiles
10650 * use bmap to make a mapping of extents in the file. They assume
10651 * these extents won't change over the life of the file and they
10652 * use the bmap result to do IO directly to the drive.
10654 * the btrfs bmap call would return logical addresses that aren't
10655 * suitable for IO and they also will change frequently as COW
10656 * operations happen. So, swapfile + btrfs == corruption.
10658 * For now we're avoiding this by dropping bmap.
10660 static const struct address_space_operations btrfs_aops = {
10661 .readpage = btrfs_readpage,
10662 .writepage = btrfs_writepage,
10663 .writepages = btrfs_writepages,
10664 .readpages = btrfs_readpages,
10665 .direct_IO = btrfs_direct_IO,
10666 .invalidatepage = btrfs_invalidatepage,
10667 .releasepage = btrfs_releasepage,
10668 .set_page_dirty = btrfs_set_page_dirty,
10669 .error_remove_page = generic_error_remove_page,
10672 static const struct address_space_operations btrfs_symlink_aops = {
10673 .readpage = btrfs_readpage,
10674 .writepage = btrfs_writepage,
10675 .invalidatepage = btrfs_invalidatepage,
10676 .releasepage = btrfs_releasepage,
10679 static const struct inode_operations btrfs_file_inode_operations = {
10680 .getattr = btrfs_getattr,
10681 .setattr = btrfs_setattr,
10682 .listxattr = btrfs_listxattr,
10683 .permission = btrfs_permission,
10684 .fiemap = btrfs_fiemap,
10685 .get_acl = btrfs_get_acl,
10686 .set_acl = btrfs_set_acl,
10687 .update_time = btrfs_update_time,
10689 static const struct inode_operations btrfs_special_inode_operations = {
10690 .getattr = btrfs_getattr,
10691 .setattr = btrfs_setattr,
10692 .permission = btrfs_permission,
10693 .listxattr = btrfs_listxattr,
10694 .get_acl = btrfs_get_acl,
10695 .set_acl = btrfs_set_acl,
10696 .update_time = btrfs_update_time,
10698 static const struct inode_operations btrfs_symlink_inode_operations = {
10699 .get_link = page_get_link,
10700 .getattr = btrfs_getattr,
10701 .setattr = btrfs_setattr,
10702 .permission = btrfs_permission,
10703 .listxattr = btrfs_listxattr,
10704 .update_time = btrfs_update_time,
10707 const struct dentry_operations btrfs_dentry_operations = {
10708 .d_delete = btrfs_dentry_delete,
10709 .d_release = btrfs_dentry_release,