1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <crypto/hash.h>
7 #include <linux/kernel.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <linux/swap.h>
31 #include <linux/migrate.h>
32 #include <linux/sched/mm.h>
33 #include <linux/iomap.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "ordered-data.h"
45 #include "compression.h"
47 #include "free-space-cache.h"
48 #include "inode-map.h"
51 #include "delalloc-space.h"
52 #include "block-group.h"
53 #include "space-info.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 struct btrfs_dio_data {
64 struct extent_changeset *data_reserved;
67 static const struct inode_operations btrfs_dir_inode_operations;
68 static const struct inode_operations btrfs_symlink_inode_operations;
69 static const struct inode_operations btrfs_special_inode_operations;
70 static const struct inode_operations btrfs_file_inode_operations;
71 static const struct address_space_operations btrfs_aops;
72 static const struct file_operations btrfs_dir_file_operations;
74 static struct kmem_cache *btrfs_inode_cachep;
75 struct kmem_cache *btrfs_trans_handle_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
78 struct kmem_cache *btrfs_free_space_bitmap_cachep;
80 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
81 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
82 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
83 static noinline int cow_file_range(struct btrfs_inode *inode,
84 struct page *locked_page,
85 u64 start, u64 end, int *page_started,
86 unsigned long *nr_written, int unlock);
87 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
88 u64 len, u64 orig_start, u64 block_start,
89 u64 block_len, u64 orig_block_len,
90 u64 ram_bytes, int compress_type,
93 static void __endio_write_update_ordered(struct btrfs_inode *inode,
94 const u64 offset, const u64 bytes,
98 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
100 * ilock_flags can have the following bit set:
102 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
103 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
106 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
108 if (ilock_flags & BTRFS_ILOCK_SHARED) {
109 if (ilock_flags & BTRFS_ILOCK_TRY) {
110 if (!inode_trylock_shared(inode))
115 inode_lock_shared(inode);
117 if (ilock_flags & BTRFS_ILOCK_TRY) {
118 if (!inode_trylock(inode))
129 * btrfs_inode_unlock - unock inode i_rwsem
131 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
132 * to decide whether the lock acquired is shared or exclusive.
134 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
136 if (ilock_flags & BTRFS_ILOCK_SHARED)
137 inode_unlock_shared(inode);
143 * Cleanup all submitted ordered extents in specified range to handle errors
144 * from the btrfs_run_delalloc_range() callback.
146 * NOTE: caller must ensure that when an error happens, it can not call
147 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
148 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
149 * to be released, which we want to happen only when finishing the ordered
150 * extent (btrfs_finish_ordered_io()).
152 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
153 struct page *locked_page,
154 u64 offset, u64 bytes)
156 unsigned long index = offset >> PAGE_SHIFT;
157 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
158 u64 page_start = page_offset(locked_page);
159 u64 page_end = page_start + PAGE_SIZE - 1;
163 while (index <= end_index) {
164 page = find_get_page(inode->vfs_inode.i_mapping, index);
168 ClearPagePrivate2(page);
173 * In case this page belongs to the delalloc range being instantiated
174 * then skip it, since the first page of a range is going to be
175 * properly cleaned up by the caller of run_delalloc_range
177 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
182 return __endio_write_update_ordered(inode, offset, bytes, false);
185 static int btrfs_dirty_inode(struct inode *inode);
187 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
188 struct inode *inode, struct inode *dir,
189 const struct qstr *qstr)
193 err = btrfs_init_acl(trans, inode, dir);
195 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
200 * this does all the hard work for inserting an inline extent into
201 * the btree. The caller should have done a btrfs_drop_extents so that
202 * no overlapping inline items exist in the btree
204 static int insert_inline_extent(struct btrfs_trans_handle *trans,
205 struct btrfs_path *path, bool extent_inserted,
206 struct btrfs_root *root, struct inode *inode,
207 u64 start, size_t size, size_t compressed_size,
209 struct page **compressed_pages)
211 struct extent_buffer *leaf;
212 struct page *page = NULL;
215 struct btrfs_file_extent_item *ei;
217 size_t cur_size = size;
218 unsigned long offset;
220 ASSERT((compressed_size > 0 && compressed_pages) ||
221 (compressed_size == 0 && !compressed_pages));
223 if (compressed_size && compressed_pages)
224 cur_size = compressed_size;
226 if (!extent_inserted) {
227 struct btrfs_key key;
230 key.objectid = btrfs_ino(BTRFS_I(inode));
232 key.type = BTRFS_EXTENT_DATA_KEY;
234 datasize = btrfs_file_extent_calc_inline_size(cur_size);
235 ret = btrfs_insert_empty_item(trans, root, path, &key,
240 leaf = path->nodes[0];
241 ei = btrfs_item_ptr(leaf, path->slots[0],
242 struct btrfs_file_extent_item);
243 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
244 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
245 btrfs_set_file_extent_encryption(leaf, ei, 0);
246 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
247 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
248 ptr = btrfs_file_extent_inline_start(ei);
250 if (compress_type != BTRFS_COMPRESS_NONE) {
253 while (compressed_size > 0) {
254 cpage = compressed_pages[i];
255 cur_size = min_t(unsigned long, compressed_size,
258 kaddr = kmap_atomic(cpage);
259 write_extent_buffer(leaf, kaddr, ptr, cur_size);
260 kunmap_atomic(kaddr);
264 compressed_size -= cur_size;
266 btrfs_set_file_extent_compression(leaf, ei,
269 page = find_get_page(inode->i_mapping,
270 start >> PAGE_SHIFT);
271 btrfs_set_file_extent_compression(leaf, ei, 0);
272 kaddr = kmap_atomic(page);
273 offset = offset_in_page(start);
274 write_extent_buffer(leaf, kaddr + offset, ptr, size);
275 kunmap_atomic(kaddr);
278 btrfs_mark_buffer_dirty(leaf);
279 btrfs_release_path(path);
282 * We align size to sectorsize for inline extents just for simplicity
285 size = ALIGN(size, root->fs_info->sectorsize);
286 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
291 * we're an inline extent, so nobody can
292 * extend the file past i_size without locking
293 * a page we already have locked.
295 * We must do any isize and inode updates
296 * before we unlock the pages. Otherwise we
297 * could end up racing with unlink.
299 BTRFS_I(inode)->disk_i_size = inode->i_size;
306 * conditionally insert an inline extent into the file. This
307 * does the checks required to make sure the data is small enough
308 * to fit as an inline extent.
310 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
311 u64 end, size_t compressed_size,
313 struct page **compressed_pages)
315 struct btrfs_drop_extents_args drop_args = { 0 };
316 struct btrfs_root *root = inode->root;
317 struct btrfs_fs_info *fs_info = root->fs_info;
318 struct btrfs_trans_handle *trans;
319 u64 isize = i_size_read(&inode->vfs_inode);
320 u64 actual_end = min(end + 1, isize);
321 u64 inline_len = actual_end - start;
322 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
323 u64 data_len = inline_len;
325 struct btrfs_path *path;
328 data_len = compressed_size;
331 actual_end > fs_info->sectorsize ||
332 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
334 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
336 data_len > fs_info->max_inline) {
340 path = btrfs_alloc_path();
344 trans = btrfs_join_transaction(root);
346 btrfs_free_path(path);
347 return PTR_ERR(trans);
349 trans->block_rsv = &inode->block_rsv;
351 drop_args.path = path;
352 drop_args.start = start;
353 drop_args.end = aligned_end;
354 drop_args.drop_cache = true;
355 drop_args.replace_extent = true;
357 if (compressed_size && compressed_pages)
358 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
361 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
364 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
366 btrfs_abort_transaction(trans, ret);
370 if (isize > actual_end)
371 inline_len = min_t(u64, isize, actual_end);
372 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
373 root, &inode->vfs_inode, start,
374 inline_len, compressed_size,
375 compress_type, compressed_pages);
376 if (ret && ret != -ENOSPC) {
377 btrfs_abort_transaction(trans, ret);
379 } else if (ret == -ENOSPC) {
384 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
385 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
386 if (ret && ret != -ENOSPC) {
387 btrfs_abort_transaction(trans, ret);
389 } else if (ret == -ENOSPC) {
394 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
395 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
398 * Don't forget to free the reserved space, as for inlined extent
399 * it won't count as data extent, free them directly here.
400 * And at reserve time, it's always aligned to page size, so
401 * just free one page here.
403 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
404 btrfs_free_path(path);
405 btrfs_end_transaction(trans);
409 struct async_extent {
414 unsigned long nr_pages;
416 struct list_head list;
421 struct page *locked_page;
424 unsigned int write_flags;
425 struct list_head extents;
426 struct cgroup_subsys_state *blkcg_css;
427 struct btrfs_work work;
432 /* Number of chunks in flight; must be first in the structure */
434 struct async_chunk chunks[];
437 static noinline int add_async_extent(struct async_chunk *cow,
438 u64 start, u64 ram_size,
441 unsigned long nr_pages,
444 struct async_extent *async_extent;
446 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
447 BUG_ON(!async_extent); /* -ENOMEM */
448 async_extent->start = start;
449 async_extent->ram_size = ram_size;
450 async_extent->compressed_size = compressed_size;
451 async_extent->pages = pages;
452 async_extent->nr_pages = nr_pages;
453 async_extent->compress_type = compress_type;
454 list_add_tail(&async_extent->list, &cow->extents);
459 * Check if the inode has flags compatible with compression
461 static inline bool inode_can_compress(struct btrfs_inode *inode)
463 if (inode->flags & BTRFS_INODE_NODATACOW ||
464 inode->flags & BTRFS_INODE_NODATASUM)
470 * Check if the inode needs to be submitted to compression, based on mount
471 * options, defragmentation, properties or heuristics.
473 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
476 struct btrfs_fs_info *fs_info = inode->root->fs_info;
478 if (!inode_can_compress(inode)) {
479 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
480 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
485 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
488 if (inode->defrag_compress)
490 /* bad compression ratios */
491 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
493 if (btrfs_test_opt(fs_info, COMPRESS) ||
494 inode->flags & BTRFS_INODE_COMPRESS ||
495 inode->prop_compress)
496 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
500 static inline void inode_should_defrag(struct btrfs_inode *inode,
501 u64 start, u64 end, u64 num_bytes, u64 small_write)
503 /* If this is a small write inside eof, kick off a defrag */
504 if (num_bytes < small_write &&
505 (start > 0 || end + 1 < inode->disk_i_size))
506 btrfs_add_inode_defrag(NULL, inode);
510 * we create compressed extents in two phases. The first
511 * phase compresses a range of pages that have already been
512 * locked (both pages and state bits are locked).
514 * This is done inside an ordered work queue, and the compression
515 * is spread across many cpus. The actual IO submission is step
516 * two, and the ordered work queue takes care of making sure that
517 * happens in the same order things were put onto the queue by
518 * writepages and friends.
520 * If this code finds it can't get good compression, it puts an
521 * entry onto the work queue to write the uncompressed bytes. This
522 * makes sure that both compressed inodes and uncompressed inodes
523 * are written in the same order that the flusher thread sent them
526 static noinline int compress_file_range(struct async_chunk *async_chunk)
528 struct inode *inode = async_chunk->inode;
529 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
530 u64 blocksize = fs_info->sectorsize;
531 u64 start = async_chunk->start;
532 u64 end = async_chunk->end;
536 struct page **pages = NULL;
537 unsigned long nr_pages;
538 unsigned long total_compressed = 0;
539 unsigned long total_in = 0;
542 int compress_type = fs_info->compress_type;
543 int compressed_extents = 0;
546 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
550 * We need to save i_size before now because it could change in between
551 * us evaluating the size and assigning it. This is because we lock and
552 * unlock the page in truncate and fallocate, and then modify the i_size
555 * The barriers are to emulate READ_ONCE, remove that once i_size_read
559 i_size = i_size_read(inode);
561 actual_end = min_t(u64, i_size, end + 1);
564 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
565 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
566 nr_pages = min_t(unsigned long, nr_pages,
567 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
570 * we don't want to send crud past the end of i_size through
571 * compression, that's just a waste of CPU time. So, if the
572 * end of the file is before the start of our current
573 * requested range of bytes, we bail out to the uncompressed
574 * cleanup code that can deal with all of this.
576 * It isn't really the fastest way to fix things, but this is a
577 * very uncommon corner.
579 if (actual_end <= start)
580 goto cleanup_and_bail_uncompressed;
582 total_compressed = actual_end - start;
585 * skip compression for a small file range(<=blocksize) that
586 * isn't an inline extent, since it doesn't save disk space at all.
588 if (total_compressed <= blocksize &&
589 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
590 goto cleanup_and_bail_uncompressed;
592 total_compressed = min_t(unsigned long, total_compressed,
593 BTRFS_MAX_UNCOMPRESSED);
598 * we do compression for mount -o compress and when the
599 * inode has not been flagged as nocompress. This flag can
600 * change at any time if we discover bad compression ratios.
602 if (inode_need_compress(BTRFS_I(inode), start, end)) {
604 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
606 /* just bail out to the uncompressed code */
611 if (BTRFS_I(inode)->defrag_compress)
612 compress_type = BTRFS_I(inode)->defrag_compress;
613 else if (BTRFS_I(inode)->prop_compress)
614 compress_type = BTRFS_I(inode)->prop_compress;
617 * we need to call clear_page_dirty_for_io on each
618 * page in the range. Otherwise applications with the file
619 * mmap'd can wander in and change the page contents while
620 * we are compressing them.
622 * If the compression fails for any reason, we set the pages
623 * dirty again later on.
625 * Note that the remaining part is redirtied, the start pointer
626 * has moved, the end is the original one.
629 extent_range_clear_dirty_for_io(inode, start, end);
633 /* Compression level is applied here and only here */
634 ret = btrfs_compress_pages(
635 compress_type | (fs_info->compress_level << 4),
636 inode->i_mapping, start,
643 unsigned long offset = offset_in_page(total_compressed);
644 struct page *page = pages[nr_pages - 1];
647 /* zero the tail end of the last page, we might be
648 * sending it down to disk
651 kaddr = kmap_atomic(page);
652 memset(kaddr + offset, 0,
654 kunmap_atomic(kaddr);
661 /* lets try to make an inline extent */
662 if (ret || total_in < actual_end) {
663 /* we didn't compress the entire range, try
664 * to make an uncompressed inline extent.
666 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
667 0, BTRFS_COMPRESS_NONE,
670 /* try making a compressed inline extent */
671 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
673 compress_type, pages);
676 unsigned long clear_flags = EXTENT_DELALLOC |
677 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
678 EXTENT_DO_ACCOUNTING;
679 unsigned long page_error_op;
681 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
684 * inline extent creation worked or returned error,
685 * we don't need to create any more async work items.
686 * Unlock and free up our temp pages.
688 * We use DO_ACCOUNTING here because we need the
689 * delalloc_release_metadata to be done _after_ we drop
690 * our outstanding extent for clearing delalloc for this
693 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
703 * Ensure we only free the compressed pages if we have
704 * them allocated, as we can still reach here with
705 * inode_need_compress() == false.
708 for (i = 0; i < nr_pages; i++) {
709 WARN_ON(pages[i]->mapping);
720 * we aren't doing an inline extent round the compressed size
721 * up to a block size boundary so the allocator does sane
724 total_compressed = ALIGN(total_compressed, blocksize);
727 * one last check to make sure the compression is really a
728 * win, compare the page count read with the blocks on disk,
729 * compression must free at least one sector size
731 total_in = ALIGN(total_in, PAGE_SIZE);
732 if (total_compressed + blocksize <= total_in) {
733 compressed_extents++;
736 * The async work queues will take care of doing actual
737 * allocation on disk for these compressed pages, and
738 * will submit them to the elevator.
740 add_async_extent(async_chunk, start, total_in,
741 total_compressed, pages, nr_pages,
744 if (start + total_in < end) {
750 return compressed_extents;
755 * the compression code ran but failed to make things smaller,
756 * free any pages it allocated and our page pointer array
758 for (i = 0; i < nr_pages; i++) {
759 WARN_ON(pages[i]->mapping);
764 total_compressed = 0;
767 /* flag the file so we don't compress in the future */
768 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
769 !(BTRFS_I(inode)->prop_compress)) {
770 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
773 cleanup_and_bail_uncompressed:
775 * No compression, but we still need to write the pages in the file
776 * we've been given so far. redirty the locked page if it corresponds
777 * to our extent and set things up for the async work queue to run
778 * cow_file_range to do the normal delalloc dance.
780 if (async_chunk->locked_page &&
781 (page_offset(async_chunk->locked_page) >= start &&
782 page_offset(async_chunk->locked_page)) <= end) {
783 __set_page_dirty_nobuffers(async_chunk->locked_page);
784 /* unlocked later on in the async handlers */
788 extent_range_redirty_for_io(inode, start, end);
789 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
790 BTRFS_COMPRESS_NONE);
791 compressed_extents++;
793 return compressed_extents;
796 static void free_async_extent_pages(struct async_extent *async_extent)
800 if (!async_extent->pages)
803 for (i = 0; i < async_extent->nr_pages; i++) {
804 WARN_ON(async_extent->pages[i]->mapping);
805 put_page(async_extent->pages[i]);
807 kfree(async_extent->pages);
808 async_extent->nr_pages = 0;
809 async_extent->pages = NULL;
813 * phase two of compressed writeback. This is the ordered portion
814 * of the code, which only gets called in the order the work was
815 * queued. We walk all the async extents created by compress_file_range
816 * and send them down to the disk.
818 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
820 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
821 struct btrfs_fs_info *fs_info = inode->root->fs_info;
822 struct async_extent *async_extent;
824 struct btrfs_key ins;
825 struct extent_map *em;
826 struct btrfs_root *root = inode->root;
827 struct extent_io_tree *io_tree = &inode->io_tree;
831 while (!list_empty(&async_chunk->extents)) {
832 async_extent = list_entry(async_chunk->extents.next,
833 struct async_extent, list);
834 list_del(&async_extent->list);
837 lock_extent(io_tree, async_extent->start,
838 async_extent->start + async_extent->ram_size - 1);
839 /* did the compression code fall back to uncompressed IO? */
840 if (!async_extent->pages) {
841 int page_started = 0;
842 unsigned long nr_written = 0;
844 /* allocate blocks */
845 ret = cow_file_range(inode, async_chunk->locked_page,
847 async_extent->start +
848 async_extent->ram_size - 1,
849 &page_started, &nr_written, 0);
854 * if page_started, cow_file_range inserted an
855 * inline extent and took care of all the unlocking
856 * and IO for us. Otherwise, we need to submit
857 * all those pages down to the drive.
859 if (!page_started && !ret)
860 extent_write_locked_range(&inode->vfs_inode,
862 async_extent->start +
863 async_extent->ram_size - 1,
865 else if (ret && async_chunk->locked_page)
866 unlock_page(async_chunk->locked_page);
872 ret = btrfs_reserve_extent(root, async_extent->ram_size,
873 async_extent->compressed_size,
874 async_extent->compressed_size,
875 0, alloc_hint, &ins, 1, 1);
877 free_async_extent_pages(async_extent);
879 if (ret == -ENOSPC) {
880 unlock_extent(io_tree, async_extent->start,
881 async_extent->start +
882 async_extent->ram_size - 1);
885 * we need to redirty the pages if we decide to
886 * fallback to uncompressed IO, otherwise we
887 * will not submit these pages down to lower
890 extent_range_redirty_for_io(&inode->vfs_inode,
892 async_extent->start +
893 async_extent->ram_size - 1);
900 * here we're doing allocation and writeback of the
903 em = create_io_em(inode, async_extent->start,
904 async_extent->ram_size, /* len */
905 async_extent->start, /* orig_start */
906 ins.objectid, /* block_start */
907 ins.offset, /* block_len */
908 ins.offset, /* orig_block_len */
909 async_extent->ram_size, /* ram_bytes */
910 async_extent->compress_type,
911 BTRFS_ORDERED_COMPRESSED);
913 /* ret value is not necessary due to void function */
914 goto out_free_reserve;
917 ret = btrfs_add_ordered_extent_compress(inode,
920 async_extent->ram_size,
922 BTRFS_ORDERED_COMPRESSED,
923 async_extent->compress_type);
925 btrfs_drop_extent_cache(inode, async_extent->start,
926 async_extent->start +
927 async_extent->ram_size - 1, 0);
928 goto out_free_reserve;
930 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
933 * clear dirty, set writeback and unlock the pages.
935 extent_clear_unlock_delalloc(inode, async_extent->start,
936 async_extent->start +
937 async_extent->ram_size - 1,
938 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
939 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
941 if (btrfs_submit_compressed_write(inode, async_extent->start,
942 async_extent->ram_size,
944 ins.offset, async_extent->pages,
945 async_extent->nr_pages,
946 async_chunk->write_flags,
947 async_chunk->blkcg_css)) {
948 struct page *p = async_extent->pages[0];
949 const u64 start = async_extent->start;
950 const u64 end = start + async_extent->ram_size - 1;
952 p->mapping = inode->vfs_inode.i_mapping;
953 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
956 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
959 free_async_extent_pages(async_extent);
961 alloc_hint = ins.objectid + ins.offset;
967 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
968 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
970 extent_clear_unlock_delalloc(inode, async_extent->start,
971 async_extent->start +
972 async_extent->ram_size - 1,
973 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
974 EXTENT_DELALLOC_NEW |
975 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
976 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
977 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
979 free_async_extent_pages(async_extent);
984 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
987 struct extent_map_tree *em_tree = &inode->extent_tree;
988 struct extent_map *em;
991 read_lock(&em_tree->lock);
992 em = search_extent_mapping(em_tree, start, num_bytes);
995 * if block start isn't an actual block number then find the
996 * first block in this inode and use that as a hint. If that
997 * block is also bogus then just don't worry about it.
999 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1000 free_extent_map(em);
1001 em = search_extent_mapping(em_tree, 0, 0);
1002 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1003 alloc_hint = em->block_start;
1005 free_extent_map(em);
1007 alloc_hint = em->block_start;
1008 free_extent_map(em);
1011 read_unlock(&em_tree->lock);
1017 * when extent_io.c finds a delayed allocation range in the file,
1018 * the call backs end up in this code. The basic idea is to
1019 * allocate extents on disk for the range, and create ordered data structs
1020 * in ram to track those extents.
1022 * locked_page is the page that writepage had locked already. We use
1023 * it to make sure we don't do extra locks or unlocks.
1025 * *page_started is set to one if we unlock locked_page and do everything
1026 * required to start IO on it. It may be clean and already done with
1027 * IO when we return.
1029 static noinline int cow_file_range(struct btrfs_inode *inode,
1030 struct page *locked_page,
1031 u64 start, u64 end, int *page_started,
1032 unsigned long *nr_written, int unlock)
1034 struct btrfs_root *root = inode->root;
1035 struct btrfs_fs_info *fs_info = root->fs_info;
1038 unsigned long ram_size;
1039 u64 cur_alloc_size = 0;
1041 u64 blocksize = fs_info->sectorsize;
1042 struct btrfs_key ins;
1043 struct extent_map *em;
1044 unsigned clear_bits;
1045 unsigned long page_ops;
1046 bool extent_reserved = false;
1049 if (btrfs_is_free_space_inode(inode)) {
1055 num_bytes = ALIGN(end - start + 1, blocksize);
1056 num_bytes = max(blocksize, num_bytes);
1057 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1059 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1062 /* lets try to make an inline extent */
1063 ret = cow_file_range_inline(inode, start, end, 0,
1064 BTRFS_COMPRESS_NONE, NULL);
1067 * We use DO_ACCOUNTING here because we need the
1068 * delalloc_release_metadata to be run _after_ we drop
1069 * our outstanding extent for clearing delalloc for this
1072 extent_clear_unlock_delalloc(inode, start, end, NULL,
1073 EXTENT_LOCKED | EXTENT_DELALLOC |
1074 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1075 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1076 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1077 PAGE_END_WRITEBACK);
1078 *nr_written = *nr_written +
1079 (end - start + PAGE_SIZE) / PAGE_SIZE;
1082 } else if (ret < 0) {
1087 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1088 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1091 * Relocation relies on the relocated extents to have exactly the same
1092 * size as the original extents. Normally writeback for relocation data
1093 * extents follows a NOCOW path because relocation preallocates the
1094 * extents. However, due to an operation such as scrub turning a block
1095 * group to RO mode, it may fallback to COW mode, so we must make sure
1096 * an extent allocated during COW has exactly the requested size and can
1097 * not be split into smaller extents, otherwise relocation breaks and
1098 * fails during the stage where it updates the bytenr of file extent
1101 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1102 min_alloc_size = num_bytes;
1104 min_alloc_size = fs_info->sectorsize;
1106 while (num_bytes > 0) {
1107 cur_alloc_size = num_bytes;
1108 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1109 min_alloc_size, 0, alloc_hint,
1113 cur_alloc_size = ins.offset;
1114 extent_reserved = true;
1116 ram_size = ins.offset;
1117 em = create_io_em(inode, start, ins.offset, /* len */
1118 start, /* orig_start */
1119 ins.objectid, /* block_start */
1120 ins.offset, /* block_len */
1121 ins.offset, /* orig_block_len */
1122 ram_size, /* ram_bytes */
1123 BTRFS_COMPRESS_NONE, /* compress_type */
1124 BTRFS_ORDERED_REGULAR /* type */);
1129 free_extent_map(em);
1131 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1132 ram_size, cur_alloc_size, 0);
1134 goto out_drop_extent_cache;
1136 if (root->root_key.objectid ==
1137 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1138 ret = btrfs_reloc_clone_csums(inode, start,
1141 * Only drop cache here, and process as normal.
1143 * We must not allow extent_clear_unlock_delalloc()
1144 * at out_unlock label to free meta of this ordered
1145 * extent, as its meta should be freed by
1146 * btrfs_finish_ordered_io().
1148 * So we must continue until @start is increased to
1149 * skip current ordered extent.
1152 btrfs_drop_extent_cache(inode, start,
1153 start + ram_size - 1, 0);
1156 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1158 /* we're not doing compressed IO, don't unlock the first
1159 * page (which the caller expects to stay locked), don't
1160 * clear any dirty bits and don't set any writeback bits
1162 * Do set the Private2 bit so we know this page was properly
1163 * setup for writepage
1165 page_ops = unlock ? PAGE_UNLOCK : 0;
1166 page_ops |= PAGE_SET_PRIVATE2;
1168 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1170 EXTENT_LOCKED | EXTENT_DELALLOC,
1172 if (num_bytes < cur_alloc_size)
1175 num_bytes -= cur_alloc_size;
1176 alloc_hint = ins.objectid + ins.offset;
1177 start += cur_alloc_size;
1178 extent_reserved = false;
1181 * btrfs_reloc_clone_csums() error, since start is increased
1182 * extent_clear_unlock_delalloc() at out_unlock label won't
1183 * free metadata of current ordered extent, we're OK to exit.
1191 out_drop_extent_cache:
1192 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1194 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1195 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1197 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1198 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1199 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1202 * If we reserved an extent for our delalloc range (or a subrange) and
1203 * failed to create the respective ordered extent, then it means that
1204 * when we reserved the extent we decremented the extent's size from
1205 * the data space_info's bytes_may_use counter and incremented the
1206 * space_info's bytes_reserved counter by the same amount. We must make
1207 * sure extent_clear_unlock_delalloc() does not try to decrement again
1208 * the data space_info's bytes_may_use counter, therefore we do not pass
1209 * it the flag EXTENT_CLEAR_DATA_RESV.
1211 if (extent_reserved) {
1212 extent_clear_unlock_delalloc(inode, start,
1213 start + cur_alloc_size - 1,
1217 start += cur_alloc_size;
1221 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1222 clear_bits | EXTENT_CLEAR_DATA_RESV,
1228 * work queue call back to started compression on a file and pages
1230 static noinline void async_cow_start(struct btrfs_work *work)
1232 struct async_chunk *async_chunk;
1233 int compressed_extents;
1235 async_chunk = container_of(work, struct async_chunk, work);
1237 compressed_extents = compress_file_range(async_chunk);
1238 if (compressed_extents == 0) {
1239 btrfs_add_delayed_iput(async_chunk->inode);
1240 async_chunk->inode = NULL;
1245 * work queue call back to submit previously compressed pages
1247 static noinline void async_cow_submit(struct btrfs_work *work)
1249 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1251 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1252 unsigned long nr_pages;
1254 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1257 /* atomic_sub_return implies a barrier */
1258 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1260 cond_wake_up_nomb(&fs_info->async_submit_wait);
1263 * ->inode could be NULL if async_chunk_start has failed to compress,
1264 * in which case we don't have anything to submit, yet we need to
1265 * always adjust ->async_delalloc_pages as its paired with the init
1266 * happening in cow_file_range_async
1268 if (async_chunk->inode)
1269 submit_compressed_extents(async_chunk);
1272 static noinline void async_cow_free(struct btrfs_work *work)
1274 struct async_chunk *async_chunk;
1276 async_chunk = container_of(work, struct async_chunk, work);
1277 if (async_chunk->inode)
1278 btrfs_add_delayed_iput(async_chunk->inode);
1279 if (async_chunk->blkcg_css)
1280 css_put(async_chunk->blkcg_css);
1282 * Since the pointer to 'pending' is at the beginning of the array of
1283 * async_chunk's, freeing it ensures the whole array has been freed.
1285 if (atomic_dec_and_test(async_chunk->pending))
1286 kvfree(async_chunk->pending);
1289 static int cow_file_range_async(struct btrfs_inode *inode,
1290 struct writeback_control *wbc,
1291 struct page *locked_page,
1292 u64 start, u64 end, int *page_started,
1293 unsigned long *nr_written)
1295 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1296 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1297 struct async_cow *ctx;
1298 struct async_chunk *async_chunk;
1299 unsigned long nr_pages;
1301 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1303 bool should_compress;
1305 const unsigned int write_flags = wbc_to_write_flags(wbc);
1307 unlock_extent(&inode->io_tree, start, end);
1309 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1310 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1312 should_compress = false;
1314 should_compress = true;
1317 nofs_flag = memalloc_nofs_save();
1318 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1319 memalloc_nofs_restore(nofs_flag);
1322 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1323 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1324 EXTENT_DO_ACCOUNTING;
1325 unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1326 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
1329 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1330 clear_bits, page_ops);
1334 async_chunk = ctx->chunks;
1335 atomic_set(&ctx->num_chunks, num_chunks);
1337 for (i = 0; i < num_chunks; i++) {
1338 if (should_compress)
1339 cur_end = min(end, start + SZ_512K - 1);
1344 * igrab is called higher up in the call chain, take only the
1345 * lightweight reference for the callback lifetime
1347 ihold(&inode->vfs_inode);
1348 async_chunk[i].pending = &ctx->num_chunks;
1349 async_chunk[i].inode = &inode->vfs_inode;
1350 async_chunk[i].start = start;
1351 async_chunk[i].end = cur_end;
1352 async_chunk[i].write_flags = write_flags;
1353 INIT_LIST_HEAD(&async_chunk[i].extents);
1356 * The locked_page comes all the way from writepage and its
1357 * the original page we were actually given. As we spread
1358 * this large delalloc region across multiple async_chunk
1359 * structs, only the first struct needs a pointer to locked_page
1361 * This way we don't need racey decisions about who is supposed
1366 * Depending on the compressibility, the pages might or
1367 * might not go through async. We want all of them to
1368 * be accounted against wbc once. Let's do it here
1369 * before the paths diverge. wbc accounting is used
1370 * only for foreign writeback detection and doesn't
1371 * need full accuracy. Just account the whole thing
1372 * against the first page.
1374 wbc_account_cgroup_owner(wbc, locked_page,
1376 async_chunk[i].locked_page = locked_page;
1379 async_chunk[i].locked_page = NULL;
1382 if (blkcg_css != blkcg_root_css) {
1384 async_chunk[i].blkcg_css = blkcg_css;
1386 async_chunk[i].blkcg_css = NULL;
1389 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1390 async_cow_submit, async_cow_free);
1392 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1393 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1395 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1397 *nr_written += nr_pages;
1398 start = cur_end + 1;
1404 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1405 u64 bytenr, u64 num_bytes)
1408 struct btrfs_ordered_sum *sums;
1411 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1412 bytenr + num_bytes - 1, &list, 0);
1413 if (ret == 0 && list_empty(&list))
1416 while (!list_empty(&list)) {
1417 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1418 list_del(&sums->list);
1426 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1427 const u64 start, const u64 end,
1428 int *page_started, unsigned long *nr_written)
1430 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1431 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1432 BTRFS_DATA_RELOC_TREE_OBJECTID);
1433 const u64 range_bytes = end + 1 - start;
1434 struct extent_io_tree *io_tree = &inode->io_tree;
1435 u64 range_start = start;
1439 * If EXTENT_NORESERVE is set it means that when the buffered write was
1440 * made we had not enough available data space and therefore we did not
1441 * reserve data space for it, since we though we could do NOCOW for the
1442 * respective file range (either there is prealloc extent or the inode
1443 * has the NOCOW bit set).
1445 * However when we need to fallback to COW mode (because for example the
1446 * block group for the corresponding extent was turned to RO mode by a
1447 * scrub or relocation) we need to do the following:
1449 * 1) We increment the bytes_may_use counter of the data space info.
1450 * If COW succeeds, it allocates a new data extent and after doing
1451 * that it decrements the space info's bytes_may_use counter and
1452 * increments its bytes_reserved counter by the same amount (we do
1453 * this at btrfs_add_reserved_bytes()). So we need to increment the
1454 * bytes_may_use counter to compensate (when space is reserved at
1455 * buffered write time, the bytes_may_use counter is incremented);
1457 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1458 * that if the COW path fails for any reason, it decrements (through
1459 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1460 * data space info, which we incremented in the step above.
1462 * If we need to fallback to cow and the inode corresponds to a free
1463 * space cache inode or an inode of the data relocation tree, we must
1464 * also increment bytes_may_use of the data space_info for the same
1465 * reason. Space caches and relocated data extents always get a prealloc
1466 * extent for them, however scrub or balance may have set the block
1467 * group that contains that extent to RO mode and therefore force COW
1468 * when starting writeback.
1470 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1471 EXTENT_NORESERVE, 0);
1472 if (count > 0 || is_space_ino || is_reloc_ino) {
1474 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1475 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1477 if (is_space_ino || is_reloc_ino)
1478 bytes = range_bytes;
1480 spin_lock(&sinfo->lock);
1481 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1482 spin_unlock(&sinfo->lock);
1485 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1489 return cow_file_range(inode, locked_page, start, end, page_started,
1494 * when nowcow writeback call back. This checks for snapshots or COW copies
1495 * of the extents that exist in the file, and COWs the file as required.
1497 * If no cow copies or snapshots exist, we write directly to the existing
1500 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1501 struct page *locked_page,
1502 const u64 start, const u64 end,
1503 int *page_started, int force,
1504 unsigned long *nr_written)
1506 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1507 struct btrfs_root *root = inode->root;
1508 struct btrfs_path *path;
1509 u64 cow_start = (u64)-1;
1510 u64 cur_offset = start;
1512 bool check_prev = true;
1513 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1514 u64 ino = btrfs_ino(inode);
1516 u64 disk_bytenr = 0;
1518 path = btrfs_alloc_path();
1520 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1521 EXTENT_LOCKED | EXTENT_DELALLOC |
1522 EXTENT_DO_ACCOUNTING |
1523 EXTENT_DEFRAG, PAGE_UNLOCK |
1525 PAGE_SET_WRITEBACK |
1526 PAGE_END_WRITEBACK);
1531 struct btrfs_key found_key;
1532 struct btrfs_file_extent_item *fi;
1533 struct extent_buffer *leaf;
1543 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1549 * If there is no extent for our range when doing the initial
1550 * search, then go back to the previous slot as it will be the
1551 * one containing the search offset
1553 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1554 leaf = path->nodes[0];
1555 btrfs_item_key_to_cpu(leaf, &found_key,
1556 path->slots[0] - 1);
1557 if (found_key.objectid == ino &&
1558 found_key.type == BTRFS_EXTENT_DATA_KEY)
1563 /* Go to next leaf if we have exhausted the current one */
1564 leaf = path->nodes[0];
1565 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1566 ret = btrfs_next_leaf(root, path);
1568 if (cow_start != (u64)-1)
1569 cur_offset = cow_start;
1574 leaf = path->nodes[0];
1577 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1579 /* Didn't find anything for our INO */
1580 if (found_key.objectid > ino)
1583 * Keep searching until we find an EXTENT_ITEM or there are no
1584 * more extents for this inode
1586 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1587 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1592 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1593 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1594 found_key.offset > end)
1598 * If the found extent starts after requested offset, then
1599 * adjust extent_end to be right before this extent begins
1601 if (found_key.offset > cur_offset) {
1602 extent_end = found_key.offset;
1608 * Found extent which begins before our range and potentially
1611 fi = btrfs_item_ptr(leaf, path->slots[0],
1612 struct btrfs_file_extent_item);
1613 extent_type = btrfs_file_extent_type(leaf, fi);
1615 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1616 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1617 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1618 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1619 extent_offset = btrfs_file_extent_offset(leaf, fi);
1620 extent_end = found_key.offset +
1621 btrfs_file_extent_num_bytes(leaf, fi);
1623 btrfs_file_extent_disk_num_bytes(leaf, fi);
1625 * If the extent we got ends before our current offset,
1626 * skip to the next extent.
1628 if (extent_end <= cur_offset) {
1633 if (disk_bytenr == 0)
1635 /* Skip compressed/encrypted/encoded extents */
1636 if (btrfs_file_extent_compression(leaf, fi) ||
1637 btrfs_file_extent_encryption(leaf, fi) ||
1638 btrfs_file_extent_other_encoding(leaf, fi))
1641 * If extent is created before the last volume's snapshot
1642 * this implies the extent is shared, hence we can't do
1643 * nocow. This is the same check as in
1644 * btrfs_cross_ref_exist but without calling
1645 * btrfs_search_slot.
1647 if (!freespace_inode &&
1648 btrfs_file_extent_generation(leaf, fi) <=
1649 btrfs_root_last_snapshot(&root->root_item))
1651 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1653 /* If extent is RO, we must COW it */
1654 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1656 ret = btrfs_cross_ref_exist(root, ino,
1658 extent_offset, disk_bytenr, false);
1661 * ret could be -EIO if the above fails to read
1665 if (cow_start != (u64)-1)
1666 cur_offset = cow_start;
1670 WARN_ON_ONCE(freespace_inode);
1673 disk_bytenr += extent_offset;
1674 disk_bytenr += cur_offset - found_key.offset;
1675 num_bytes = min(end + 1, extent_end) - cur_offset;
1677 * If there are pending snapshots for this root, we
1678 * fall into common COW way
1680 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1683 * force cow if csum exists in the range.
1684 * this ensure that csum for a given extent are
1685 * either valid or do not exist.
1687 ret = csum_exist_in_range(fs_info, disk_bytenr,
1691 * ret could be -EIO if the above fails to read
1695 if (cow_start != (u64)-1)
1696 cur_offset = cow_start;
1699 WARN_ON_ONCE(freespace_inode);
1702 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1705 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1706 extent_end = found_key.offset + ram_bytes;
1707 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1708 /* Skip extents outside of our requested range */
1709 if (extent_end <= start) {
1714 /* If this triggers then we have a memory corruption */
1719 * If nocow is false then record the beginning of the range
1720 * that needs to be COWed
1723 if (cow_start == (u64)-1)
1724 cow_start = cur_offset;
1725 cur_offset = extent_end;
1726 if (cur_offset > end)
1732 btrfs_release_path(path);
1735 * COW range from cow_start to found_key.offset - 1. As the key
1736 * will contain the beginning of the first extent that can be
1737 * NOCOW, following one which needs to be COW'ed
1739 if (cow_start != (u64)-1) {
1740 ret = fallback_to_cow(inode, locked_page,
1741 cow_start, found_key.offset - 1,
1742 page_started, nr_written);
1745 cow_start = (u64)-1;
1748 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1749 u64 orig_start = found_key.offset - extent_offset;
1750 struct extent_map *em;
1752 em = create_io_em(inode, cur_offset, num_bytes,
1754 disk_bytenr, /* block_start */
1755 num_bytes, /* block_len */
1756 disk_num_bytes, /* orig_block_len */
1757 ram_bytes, BTRFS_COMPRESS_NONE,
1758 BTRFS_ORDERED_PREALLOC);
1763 free_extent_map(em);
1764 ret = btrfs_add_ordered_extent(inode, cur_offset,
1765 disk_bytenr, num_bytes,
1767 BTRFS_ORDERED_PREALLOC);
1769 btrfs_drop_extent_cache(inode, cur_offset,
1770 cur_offset + num_bytes - 1,
1775 ret = btrfs_add_ordered_extent(inode, cur_offset,
1776 disk_bytenr, num_bytes,
1778 BTRFS_ORDERED_NOCOW);
1784 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1787 if (root->root_key.objectid ==
1788 BTRFS_DATA_RELOC_TREE_OBJECTID)
1790 * Error handled later, as we must prevent
1791 * extent_clear_unlock_delalloc() in error handler
1792 * from freeing metadata of created ordered extent.
1794 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1797 extent_clear_unlock_delalloc(inode, cur_offset,
1798 cur_offset + num_bytes - 1,
1799 locked_page, EXTENT_LOCKED |
1801 EXTENT_CLEAR_DATA_RESV,
1802 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1804 cur_offset = extent_end;
1807 * btrfs_reloc_clone_csums() error, now we're OK to call error
1808 * handler, as metadata for created ordered extent will only
1809 * be freed by btrfs_finish_ordered_io().
1813 if (cur_offset > end)
1816 btrfs_release_path(path);
1818 if (cur_offset <= end && cow_start == (u64)-1)
1819 cow_start = cur_offset;
1821 if (cow_start != (u64)-1) {
1823 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1824 page_started, nr_written);
1831 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1833 if (ret && cur_offset < end)
1834 extent_clear_unlock_delalloc(inode, cur_offset, end,
1835 locked_page, EXTENT_LOCKED |
1836 EXTENT_DELALLOC | EXTENT_DEFRAG |
1837 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1839 PAGE_SET_WRITEBACK |
1840 PAGE_END_WRITEBACK);
1841 btrfs_free_path(path);
1845 static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end)
1848 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1849 !(inode->flags & BTRFS_INODE_PREALLOC))
1853 * @defrag_bytes is a hint value, no spinlock held here,
1854 * if is not zero, it means the file is defragging.
1855 * Force cow if given extent needs to be defragged.
1857 if (inode->defrag_bytes &&
1858 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL))
1865 * Function to process delayed allocation (create CoW) for ranges which are
1866 * being touched for the first time.
1868 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1869 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1870 struct writeback_control *wbc)
1873 int force_cow = need_force_cow(inode, start, end);
1875 if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1876 ret = run_delalloc_nocow(inode, locked_page, start, end,
1877 page_started, 1, nr_written);
1878 } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1879 ret = run_delalloc_nocow(inode, locked_page, start, end,
1880 page_started, 0, nr_written);
1881 } else if (!inode_can_compress(inode) ||
1882 !inode_need_compress(inode, start, end)) {
1883 ret = cow_file_range(inode, locked_page, start, end,
1884 page_started, nr_written, 1);
1886 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1887 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1888 page_started, nr_written);
1891 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1896 void btrfs_split_delalloc_extent(struct inode *inode,
1897 struct extent_state *orig, u64 split)
1901 /* not delalloc, ignore it */
1902 if (!(orig->state & EXTENT_DELALLOC))
1905 size = orig->end - orig->start + 1;
1906 if (size > BTRFS_MAX_EXTENT_SIZE) {
1911 * See the explanation in btrfs_merge_delalloc_extent, the same
1912 * applies here, just in reverse.
1914 new_size = orig->end - split + 1;
1915 num_extents = count_max_extents(new_size);
1916 new_size = split - orig->start;
1917 num_extents += count_max_extents(new_size);
1918 if (count_max_extents(size) >= num_extents)
1922 spin_lock(&BTRFS_I(inode)->lock);
1923 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1924 spin_unlock(&BTRFS_I(inode)->lock);
1928 * Handle merged delayed allocation extents so we can keep track of new extents
1929 * that are just merged onto old extents, such as when we are doing sequential
1930 * writes, so we can properly account for the metadata space we'll need.
1932 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1933 struct extent_state *other)
1935 u64 new_size, old_size;
1938 /* not delalloc, ignore it */
1939 if (!(other->state & EXTENT_DELALLOC))
1942 if (new->start > other->start)
1943 new_size = new->end - other->start + 1;
1945 new_size = other->end - new->start + 1;
1947 /* we're not bigger than the max, unreserve the space and go */
1948 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1949 spin_lock(&BTRFS_I(inode)->lock);
1950 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1951 spin_unlock(&BTRFS_I(inode)->lock);
1956 * We have to add up either side to figure out how many extents were
1957 * accounted for before we merged into one big extent. If the number of
1958 * extents we accounted for is <= the amount we need for the new range
1959 * then we can return, otherwise drop. Think of it like this
1963 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1964 * need 2 outstanding extents, on one side we have 1 and the other side
1965 * we have 1 so they are == and we can return. But in this case
1967 * [MAX_SIZE+4k][MAX_SIZE+4k]
1969 * Each range on their own accounts for 2 extents, but merged together
1970 * they are only 3 extents worth of accounting, so we need to drop in
1973 old_size = other->end - other->start + 1;
1974 num_extents = count_max_extents(old_size);
1975 old_size = new->end - new->start + 1;
1976 num_extents += count_max_extents(old_size);
1977 if (count_max_extents(new_size) >= num_extents)
1980 spin_lock(&BTRFS_I(inode)->lock);
1981 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1982 spin_unlock(&BTRFS_I(inode)->lock);
1985 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1986 struct inode *inode)
1988 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1990 spin_lock(&root->delalloc_lock);
1991 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1992 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1993 &root->delalloc_inodes);
1994 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1995 &BTRFS_I(inode)->runtime_flags);
1996 root->nr_delalloc_inodes++;
1997 if (root->nr_delalloc_inodes == 1) {
1998 spin_lock(&fs_info->delalloc_root_lock);
1999 BUG_ON(!list_empty(&root->delalloc_root));
2000 list_add_tail(&root->delalloc_root,
2001 &fs_info->delalloc_roots);
2002 spin_unlock(&fs_info->delalloc_root_lock);
2005 spin_unlock(&root->delalloc_lock);
2009 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2010 struct btrfs_inode *inode)
2012 struct btrfs_fs_info *fs_info = root->fs_info;
2014 if (!list_empty(&inode->delalloc_inodes)) {
2015 list_del_init(&inode->delalloc_inodes);
2016 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2017 &inode->runtime_flags);
2018 root->nr_delalloc_inodes--;
2019 if (!root->nr_delalloc_inodes) {
2020 ASSERT(list_empty(&root->delalloc_inodes));
2021 spin_lock(&fs_info->delalloc_root_lock);
2022 BUG_ON(list_empty(&root->delalloc_root));
2023 list_del_init(&root->delalloc_root);
2024 spin_unlock(&fs_info->delalloc_root_lock);
2029 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2030 struct btrfs_inode *inode)
2032 spin_lock(&root->delalloc_lock);
2033 __btrfs_del_delalloc_inode(root, inode);
2034 spin_unlock(&root->delalloc_lock);
2038 * Properly track delayed allocation bytes in the inode and to maintain the
2039 * list of inodes that have pending delalloc work to be done.
2041 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2044 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2046 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2049 * set_bit and clear bit hooks normally require _irqsave/restore
2050 * but in this case, we are only testing for the DELALLOC
2051 * bit, which is only set or cleared with irqs on
2053 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2054 struct btrfs_root *root = BTRFS_I(inode)->root;
2055 u64 len = state->end + 1 - state->start;
2056 u32 num_extents = count_max_extents(len);
2057 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2059 spin_lock(&BTRFS_I(inode)->lock);
2060 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2061 spin_unlock(&BTRFS_I(inode)->lock);
2063 /* For sanity tests */
2064 if (btrfs_is_testing(fs_info))
2067 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2068 fs_info->delalloc_batch);
2069 spin_lock(&BTRFS_I(inode)->lock);
2070 BTRFS_I(inode)->delalloc_bytes += len;
2071 if (*bits & EXTENT_DEFRAG)
2072 BTRFS_I(inode)->defrag_bytes += len;
2073 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2074 &BTRFS_I(inode)->runtime_flags))
2075 btrfs_add_delalloc_inodes(root, inode);
2076 spin_unlock(&BTRFS_I(inode)->lock);
2079 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2080 (*bits & EXTENT_DELALLOC_NEW)) {
2081 spin_lock(&BTRFS_I(inode)->lock);
2082 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2084 spin_unlock(&BTRFS_I(inode)->lock);
2089 * Once a range is no longer delalloc this function ensures that proper
2090 * accounting happens.
2092 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2093 struct extent_state *state, unsigned *bits)
2095 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2096 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2097 u64 len = state->end + 1 - state->start;
2098 u32 num_extents = count_max_extents(len);
2100 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2101 spin_lock(&inode->lock);
2102 inode->defrag_bytes -= len;
2103 spin_unlock(&inode->lock);
2107 * set_bit and clear bit hooks normally require _irqsave/restore
2108 * but in this case, we are only testing for the DELALLOC
2109 * bit, which is only set or cleared with irqs on
2111 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2112 struct btrfs_root *root = inode->root;
2113 bool do_list = !btrfs_is_free_space_inode(inode);
2115 spin_lock(&inode->lock);
2116 btrfs_mod_outstanding_extents(inode, -num_extents);
2117 spin_unlock(&inode->lock);
2120 * We don't reserve metadata space for space cache inodes so we
2121 * don't need to call delalloc_release_metadata if there is an
2124 if (*bits & EXTENT_CLEAR_META_RESV &&
2125 root != fs_info->tree_root)
2126 btrfs_delalloc_release_metadata(inode, len, false);
2128 /* For sanity tests. */
2129 if (btrfs_is_testing(fs_info))
2132 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2133 do_list && !(state->state & EXTENT_NORESERVE) &&
2134 (*bits & EXTENT_CLEAR_DATA_RESV))
2135 btrfs_free_reserved_data_space_noquota(fs_info, len);
2137 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2138 fs_info->delalloc_batch);
2139 spin_lock(&inode->lock);
2140 inode->delalloc_bytes -= len;
2141 if (do_list && inode->delalloc_bytes == 0 &&
2142 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2143 &inode->runtime_flags))
2144 btrfs_del_delalloc_inode(root, inode);
2145 spin_unlock(&inode->lock);
2148 if ((state->state & EXTENT_DELALLOC_NEW) &&
2149 (*bits & EXTENT_DELALLOC_NEW)) {
2150 spin_lock(&inode->lock);
2151 ASSERT(inode->new_delalloc_bytes >= len);
2152 inode->new_delalloc_bytes -= len;
2153 if (*bits & EXTENT_ADD_INODE_BYTES)
2154 inode_add_bytes(&inode->vfs_inode, len);
2155 spin_unlock(&inode->lock);
2160 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2161 * in a chunk's stripe. This function ensures that bios do not span a
2164 * @page - The page we are about to add to the bio
2165 * @size - size we want to add to the bio
2166 * @bio - bio we want to ensure is smaller than a stripe
2167 * @bio_flags - flags of the bio
2169 * return 1 if page cannot be added to the bio
2170 * return 0 if page can be added to the bio
2171 * return error otherwise
2173 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2174 unsigned long bio_flags)
2176 struct inode *inode = page->mapping->host;
2177 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2178 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
2182 struct btrfs_io_geometry geom;
2184 if (bio_flags & EXTENT_BIO_COMPRESSED)
2187 length = bio->bi_iter.bi_size;
2188 map_length = length;
2189 ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length,
2194 if (geom.len < length + size)
2200 * in order to insert checksums into the metadata in large chunks,
2201 * we wait until bio submission time. All the pages in the bio are
2202 * checksummed and sums are attached onto the ordered extent record.
2204 * At IO completion time the cums attached on the ordered extent record
2205 * are inserted into the btree
2207 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2210 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2214 * extent_io.c submission hook. This does the right thing for csum calculation
2215 * on write, or reading the csums from the tree before a read.
2217 * Rules about async/sync submit,
2218 * a) read: sync submit
2220 * b) write without checksum: sync submit
2222 * c) write with checksum:
2223 * c-1) if bio is issued by fsync: sync submit
2224 * (sync_writers != 0)
2226 * c-2) if root is reloc root: sync submit
2227 * (only in case of buffered IO)
2229 * c-3) otherwise: async submit
2231 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2232 int mirror_num, unsigned long bio_flags)
2235 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2236 struct btrfs_root *root = BTRFS_I(inode)->root;
2237 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2238 blk_status_t ret = 0;
2240 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2242 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2243 !fs_info->csum_root;
2245 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2246 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2248 if (bio_op(bio) != REQ_OP_WRITE) {
2249 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2253 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2254 ret = btrfs_submit_compressed_read(inode, bio,
2260 * Lookup bio sums does extra checks around whether we
2261 * need to csum or not, which is why we ignore skip_sum
2264 ret = btrfs_lookup_bio_sums(inode, bio, (u64)-1, NULL);
2269 } else if (async && !skip_sum) {
2270 /* csum items have already been cloned */
2271 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2273 /* we're doing a write, do the async checksumming */
2274 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2275 0, btrfs_submit_bio_start);
2277 } else if (!skip_sum) {
2278 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2284 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2288 bio->bi_status = ret;
2295 * given a list of ordered sums record them in the inode. This happens
2296 * at IO completion time based on sums calculated at bio submission time.
2298 static int add_pending_csums(struct btrfs_trans_handle *trans,
2299 struct list_head *list)
2301 struct btrfs_ordered_sum *sum;
2304 list_for_each_entry(sum, list, list) {
2305 trans->adding_csums = true;
2306 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2307 trans->adding_csums = false;
2314 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2317 struct extent_state **cached_state)
2319 u64 search_start = start;
2320 const u64 end = start + len - 1;
2322 while (search_start < end) {
2323 const u64 search_len = end - search_start + 1;
2324 struct extent_map *em;
2328 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2332 if (em->block_start != EXTENT_MAP_HOLE)
2336 if (em->start < search_start)
2337 em_len -= search_start - em->start;
2338 if (em_len > search_len)
2339 em_len = search_len;
2341 ret = set_extent_bit(&inode->io_tree, search_start,
2342 search_start + em_len - 1,
2343 EXTENT_DELALLOC_NEW,
2344 NULL, cached_state, GFP_NOFS);
2346 search_start = extent_map_end(em);
2347 free_extent_map(em);
2354 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2355 unsigned int extra_bits,
2356 struct extent_state **cached_state)
2358 WARN_ON(PAGE_ALIGNED(end));
2360 if (start >= i_size_read(&inode->vfs_inode) &&
2361 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2363 * There can't be any extents following eof in this case so just
2364 * set the delalloc new bit for the range directly.
2366 extra_bits |= EXTENT_DELALLOC_NEW;
2370 ret = btrfs_find_new_delalloc_bytes(inode, start,
2377 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2381 /* see btrfs_writepage_start_hook for details on why this is required */
2382 struct btrfs_writepage_fixup {
2384 struct inode *inode;
2385 struct btrfs_work work;
2388 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2390 struct btrfs_writepage_fixup *fixup;
2391 struct btrfs_ordered_extent *ordered;
2392 struct extent_state *cached_state = NULL;
2393 struct extent_changeset *data_reserved = NULL;
2395 struct btrfs_inode *inode;
2399 bool free_delalloc_space = true;
2401 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2403 inode = BTRFS_I(fixup->inode);
2404 page_start = page_offset(page);
2405 page_end = page_offset(page) + PAGE_SIZE - 1;
2408 * This is similar to page_mkwrite, we need to reserve the space before
2409 * we take the page lock.
2411 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2417 * Before we queued this fixup, we took a reference on the page.
2418 * page->mapping may go NULL, but it shouldn't be moved to a different
2421 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2423 * Unfortunately this is a little tricky, either
2425 * 1) We got here and our page had already been dealt with and
2426 * we reserved our space, thus ret == 0, so we need to just
2427 * drop our space reservation and bail. This can happen the
2428 * first time we come into the fixup worker, or could happen
2429 * while waiting for the ordered extent.
2430 * 2) Our page was already dealt with, but we happened to get an
2431 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2432 * this case we obviously don't have anything to release, but
2433 * because the page was already dealt with we don't want to
2434 * mark the page with an error, so make sure we're resetting
2435 * ret to 0. This is why we have this check _before_ the ret
2436 * check, because we do not want to have a surprise ENOSPC
2437 * when the page was already properly dealt with.
2440 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2441 btrfs_delalloc_release_space(inode, data_reserved,
2442 page_start, PAGE_SIZE,
2450 * We can't mess with the page state unless it is locked, so now that
2451 * it is locked bail if we failed to make our space reservation.
2456 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2458 /* already ordered? We're done */
2459 if (PagePrivate2(page))
2462 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2464 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2467 btrfs_start_ordered_extent(ordered, 1);
2468 btrfs_put_ordered_extent(ordered);
2472 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2478 * Everything went as planned, we're now the owner of a dirty page with
2479 * delayed allocation bits set and space reserved for our COW
2482 * The page was dirty when we started, nothing should have cleaned it.
2484 BUG_ON(!PageDirty(page));
2485 free_delalloc_space = false;
2487 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2488 if (free_delalloc_space)
2489 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2491 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2496 * We hit ENOSPC or other errors. Update the mapping and page
2497 * to reflect the errors and clean the page.
2499 mapping_set_error(page->mapping, ret);
2500 end_extent_writepage(page, ret, page_start, page_end);
2501 clear_page_dirty_for_io(page);
2504 ClearPageChecked(page);
2508 extent_changeset_free(data_reserved);
2510 * As a precaution, do a delayed iput in case it would be the last iput
2511 * that could need flushing space. Recursing back to fixup worker would
2514 btrfs_add_delayed_iput(&inode->vfs_inode);
2518 * There are a few paths in the higher layers of the kernel that directly
2519 * set the page dirty bit without asking the filesystem if it is a
2520 * good idea. This causes problems because we want to make sure COW
2521 * properly happens and the data=ordered rules are followed.
2523 * In our case any range that doesn't have the ORDERED bit set
2524 * hasn't been properly setup for IO. We kick off an async process
2525 * to fix it up. The async helper will wait for ordered extents, set
2526 * the delalloc bit and make it safe to write the page.
2528 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2530 struct inode *inode = page->mapping->host;
2531 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2532 struct btrfs_writepage_fixup *fixup;
2534 /* this page is properly in the ordered list */
2535 if (TestClearPagePrivate2(page))
2539 * PageChecked is set below when we create a fixup worker for this page,
2540 * don't try to create another one if we're already PageChecked()
2542 * The extent_io writepage code will redirty the page if we send back
2545 if (PageChecked(page))
2548 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2553 * We are already holding a reference to this inode from
2554 * write_cache_pages. We need to hold it because the space reservation
2555 * takes place outside of the page lock, and we can't trust
2556 * page->mapping outside of the page lock.
2559 SetPageChecked(page);
2561 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2563 fixup->inode = inode;
2564 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2569 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2570 struct btrfs_inode *inode, u64 file_pos,
2571 struct btrfs_file_extent_item *stack_fi,
2572 const bool update_inode_bytes,
2573 u64 qgroup_reserved)
2575 struct btrfs_root *root = inode->root;
2576 const u64 sectorsize = root->fs_info->sectorsize;
2577 struct btrfs_path *path;
2578 struct extent_buffer *leaf;
2579 struct btrfs_key ins;
2580 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2581 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2582 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2583 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2584 struct btrfs_drop_extents_args drop_args = { 0 };
2587 path = btrfs_alloc_path();
2592 * we may be replacing one extent in the tree with another.
2593 * The new extent is pinned in the extent map, and we don't want
2594 * to drop it from the cache until it is completely in the btree.
2596 * So, tell btrfs_drop_extents to leave this extent in the cache.
2597 * the caller is expected to unpin it and allow it to be merged
2600 drop_args.path = path;
2601 drop_args.start = file_pos;
2602 drop_args.end = file_pos + num_bytes;
2603 drop_args.replace_extent = true;
2604 drop_args.extent_item_size = sizeof(*stack_fi);
2605 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2609 if (!drop_args.extent_inserted) {
2610 ins.objectid = btrfs_ino(inode);
2611 ins.offset = file_pos;
2612 ins.type = BTRFS_EXTENT_DATA_KEY;
2614 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2619 leaf = path->nodes[0];
2620 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2621 write_extent_buffer(leaf, stack_fi,
2622 btrfs_item_ptr_offset(leaf, path->slots[0]),
2623 sizeof(struct btrfs_file_extent_item));
2625 btrfs_mark_buffer_dirty(leaf);
2626 btrfs_release_path(path);
2629 * If we dropped an inline extent here, we know the range where it is
2630 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2631 * number of bytes only for that range contaning the inline extent.
2632 * The remaining of the range will be processed when clearning the
2633 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2635 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2636 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2638 inline_size = drop_args.bytes_found - inline_size;
2639 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2640 drop_args.bytes_found -= inline_size;
2641 num_bytes -= sectorsize;
2644 if (update_inode_bytes)
2645 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2647 ins.objectid = disk_bytenr;
2648 ins.offset = disk_num_bytes;
2649 ins.type = BTRFS_EXTENT_ITEM_KEY;
2651 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2655 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2656 file_pos, qgroup_reserved, &ins);
2658 btrfs_free_path(path);
2663 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2666 struct btrfs_block_group *cache;
2668 cache = btrfs_lookup_block_group(fs_info, start);
2671 spin_lock(&cache->lock);
2672 cache->delalloc_bytes -= len;
2673 spin_unlock(&cache->lock);
2675 btrfs_put_block_group(cache);
2678 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2679 struct btrfs_ordered_extent *oe)
2681 struct btrfs_file_extent_item stack_fi;
2683 bool update_inode_bytes;
2685 memset(&stack_fi, 0, sizeof(stack_fi));
2686 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2687 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2688 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2689 oe->disk_num_bytes);
2690 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2691 logical_len = oe->truncated_len;
2693 logical_len = oe->num_bytes;
2694 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2695 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2696 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2697 /* Encryption and other encoding is reserved and all 0 */
2700 * For delalloc, when completing an ordered extent we update the inode's
2701 * bytes when clearing the range in the inode's io tree, so pass false
2702 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2703 * except if the ordered extent was truncated.
2705 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2706 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2708 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2709 oe->file_offset, &stack_fi,
2710 update_inode_bytes, oe->qgroup_rsv);
2714 * As ordered data IO finishes, this gets called so we can finish
2715 * an ordered extent if the range of bytes in the file it covers are
2718 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2720 struct inode *inode = ordered_extent->inode;
2721 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2722 struct btrfs_root *root = BTRFS_I(inode)->root;
2723 struct btrfs_trans_handle *trans = NULL;
2724 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2725 struct extent_state *cached_state = NULL;
2727 int compress_type = 0;
2729 u64 logical_len = ordered_extent->num_bytes;
2730 bool freespace_inode;
2731 bool truncated = false;
2732 bool clear_reserved_extent = true;
2733 unsigned int clear_bits = EXTENT_DEFRAG;
2735 start = ordered_extent->file_offset;
2736 end = start + ordered_extent->num_bytes - 1;
2738 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2739 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2740 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2741 clear_bits |= EXTENT_DELALLOC_NEW;
2743 freespace_inode = btrfs_is_free_space_inode(BTRFS_I(inode));
2745 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2750 btrfs_free_io_failure_record(BTRFS_I(inode), start, end);
2752 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2754 logical_len = ordered_extent->truncated_len;
2755 /* Truncated the entire extent, don't bother adding */
2760 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2761 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2763 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2764 if (freespace_inode)
2765 trans = btrfs_join_transaction_spacecache(root);
2767 trans = btrfs_join_transaction(root);
2768 if (IS_ERR(trans)) {
2769 ret = PTR_ERR(trans);
2773 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2774 ret = btrfs_update_inode_fallback(trans, root, inode);
2775 if (ret) /* -ENOMEM or corruption */
2776 btrfs_abort_transaction(trans, ret);
2780 clear_bits |= EXTENT_LOCKED;
2781 lock_extent_bits(io_tree, start, end, &cached_state);
2783 if (freespace_inode)
2784 trans = btrfs_join_transaction_spacecache(root);
2786 trans = btrfs_join_transaction(root);
2787 if (IS_ERR(trans)) {
2788 ret = PTR_ERR(trans);
2793 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2795 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2796 compress_type = ordered_extent->compress_type;
2797 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2798 BUG_ON(compress_type);
2799 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2800 ordered_extent->file_offset,
2801 ordered_extent->file_offset +
2804 BUG_ON(root == fs_info->tree_root);
2805 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2807 clear_reserved_extent = false;
2808 btrfs_release_delalloc_bytes(fs_info,
2809 ordered_extent->disk_bytenr,
2810 ordered_extent->disk_num_bytes);
2813 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2814 ordered_extent->file_offset,
2815 ordered_extent->num_bytes, trans->transid);
2817 btrfs_abort_transaction(trans, ret);
2821 ret = add_pending_csums(trans, &ordered_extent->list);
2823 btrfs_abort_transaction(trans, ret);
2828 * If this is a new delalloc range, clear its new delalloc flag to
2829 * update the inode's number of bytes. This needs to be done first
2830 * before updating the inode item.
2832 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
2833 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
2834 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end,
2835 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
2836 0, 0, &cached_state);
2838 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
2839 ret = btrfs_update_inode_fallback(trans, root, inode);
2840 if (ret) { /* -ENOMEM or corruption */
2841 btrfs_abort_transaction(trans, ret);
2846 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, clear_bits,
2847 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2851 btrfs_end_transaction(trans);
2853 if (ret || truncated) {
2854 u64 unwritten_start = start;
2857 unwritten_start += logical_len;
2858 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2860 /* Drop the cache for the part of the extent we didn't write. */
2861 btrfs_drop_extent_cache(BTRFS_I(inode), unwritten_start, end, 0);
2864 * If the ordered extent had an IOERR or something else went
2865 * wrong we need to return the space for this ordered extent
2866 * back to the allocator. We only free the extent in the
2867 * truncated case if we didn't write out the extent at all.
2869 * If we made it past insert_reserved_file_extent before we
2870 * errored out then we don't need to do this as the accounting
2871 * has already been done.
2873 if ((ret || !logical_len) &&
2874 clear_reserved_extent &&
2875 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2876 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2878 * Discard the range before returning it back to the
2881 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
2882 btrfs_discard_extent(fs_info,
2883 ordered_extent->disk_bytenr,
2884 ordered_extent->disk_num_bytes,
2886 btrfs_free_reserved_extent(fs_info,
2887 ordered_extent->disk_bytenr,
2888 ordered_extent->disk_num_bytes, 1);
2893 * This needs to be done to make sure anybody waiting knows we are done
2894 * updating everything for this ordered extent.
2896 btrfs_remove_ordered_extent(BTRFS_I(inode), ordered_extent);
2899 btrfs_put_ordered_extent(ordered_extent);
2900 /* once for the tree */
2901 btrfs_put_ordered_extent(ordered_extent);
2906 static void finish_ordered_fn(struct btrfs_work *work)
2908 struct btrfs_ordered_extent *ordered_extent;
2909 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2910 btrfs_finish_ordered_io(ordered_extent);
2913 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
2914 u64 end, int uptodate)
2916 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2917 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2918 struct btrfs_ordered_extent *ordered_extent = NULL;
2919 struct btrfs_workqueue *wq;
2921 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2923 ClearPagePrivate2(page);
2924 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2925 end - start + 1, uptodate))
2928 if (btrfs_is_free_space_inode(inode))
2929 wq = fs_info->endio_freespace_worker;
2931 wq = fs_info->endio_write_workers;
2933 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2934 btrfs_queue_work(wq, &ordered_extent->work);
2938 * check_data_csum - verify checksum of one sector of uncompressed data
2940 * @io_bio: btrfs_io_bio which contains the csum
2941 * @icsum: checksum index in the io_bio->csum array, size of csum_size
2942 * @page: page where is the data to be verified
2943 * @pgoff: offset inside the page
2945 * The length of such check is always one sector size.
2947 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
2948 int icsum, struct page *page, int pgoff)
2950 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2951 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
2953 u32 len = fs_info->sectorsize;
2954 const u32 csum_size = fs_info->csum_size;
2956 u8 csum[BTRFS_CSUM_SIZE];
2958 ASSERT(pgoff + len <= PAGE_SIZE);
2960 csum_expected = ((u8 *)io_bio->csum) + icsum * csum_size;
2962 kaddr = kmap_atomic(page);
2963 shash->tfm = fs_info->csum_shash;
2965 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
2967 if (memcmp(csum, csum_expected, csum_size))
2970 kunmap_atomic(kaddr);
2973 btrfs_print_data_csum_error(BTRFS_I(inode), page_offset(page) + pgoff,
2974 csum, csum_expected, io_bio->mirror_num);
2976 btrfs_dev_stat_inc_and_print(io_bio->device,
2977 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2978 memset(kaddr + pgoff, 1, len);
2979 flush_dcache_page(page);
2980 kunmap_atomic(kaddr);
2985 * when reads are done, we need to check csums to verify the data is correct
2986 * if there's a match, we allow the bio to finish. If not, the code in
2987 * extent_io.c will try to find good copies for us.
2989 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u64 phy_offset,
2990 struct page *page, u64 start, u64 end, int mirror)
2992 size_t offset = start - page_offset(page);
2993 struct inode *inode = page->mapping->host;
2994 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2995 struct btrfs_root *root = BTRFS_I(inode)->root;
2997 if (PageChecked(page)) {
2998 ClearPageChecked(page);
3002 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3005 if (!root->fs_info->csum_root)
3008 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3009 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3010 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3014 phy_offset >>= root->fs_info->sectorsize_bits;
3015 return check_data_csum(inode, io_bio, phy_offset, page, offset);
3019 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3021 * @inode: The inode we want to perform iput on
3023 * This function uses the generic vfs_inode::i_count to track whether we should
3024 * just decrement it (in case it's > 1) or if this is the last iput then link
3025 * the inode to the delayed iput machinery. Delayed iputs are processed at
3026 * transaction commit time/superblock commit/cleaner kthread.
3028 void btrfs_add_delayed_iput(struct inode *inode)
3030 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3031 struct btrfs_inode *binode = BTRFS_I(inode);
3033 if (atomic_add_unless(&inode->i_count, -1, 1))
3036 atomic_inc(&fs_info->nr_delayed_iputs);
3037 spin_lock(&fs_info->delayed_iput_lock);
3038 ASSERT(list_empty(&binode->delayed_iput));
3039 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3040 spin_unlock(&fs_info->delayed_iput_lock);
3041 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3042 wake_up_process(fs_info->cleaner_kthread);
3045 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3046 struct btrfs_inode *inode)
3048 list_del_init(&inode->delayed_iput);
3049 spin_unlock(&fs_info->delayed_iput_lock);
3050 iput(&inode->vfs_inode);
3051 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3052 wake_up(&fs_info->delayed_iputs_wait);
3053 spin_lock(&fs_info->delayed_iput_lock);
3056 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3057 struct btrfs_inode *inode)
3059 if (!list_empty(&inode->delayed_iput)) {
3060 spin_lock(&fs_info->delayed_iput_lock);
3061 if (!list_empty(&inode->delayed_iput))
3062 run_delayed_iput_locked(fs_info, inode);
3063 spin_unlock(&fs_info->delayed_iput_lock);
3067 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3070 spin_lock(&fs_info->delayed_iput_lock);
3071 while (!list_empty(&fs_info->delayed_iputs)) {
3072 struct btrfs_inode *inode;
3074 inode = list_first_entry(&fs_info->delayed_iputs,
3075 struct btrfs_inode, delayed_iput);
3076 run_delayed_iput_locked(fs_info, inode);
3078 spin_unlock(&fs_info->delayed_iput_lock);
3082 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
3083 * @fs_info - the fs_info for this fs
3084 * @return - EINTR if we were killed, 0 if nothing's pending
3086 * This will wait on any delayed iputs that are currently running with KILLABLE
3087 * set. Once they are all done running we will return, unless we are killed in
3088 * which case we return EINTR. This helps in user operations like fallocate etc
3089 * that might get blocked on the iputs.
3091 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3093 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3094 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3101 * This creates an orphan entry for the given inode in case something goes wrong
3102 * in the middle of an unlink.
3104 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3105 struct btrfs_inode *inode)
3109 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3110 if (ret && ret != -EEXIST) {
3111 btrfs_abort_transaction(trans, ret);
3119 * We have done the delete so we can go ahead and remove the orphan item for
3120 * this particular inode.
3122 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3123 struct btrfs_inode *inode)
3125 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3129 * this cleans up any orphans that may be left on the list from the last use
3132 int btrfs_orphan_cleanup(struct btrfs_root *root)
3134 struct btrfs_fs_info *fs_info = root->fs_info;
3135 struct btrfs_path *path;
3136 struct extent_buffer *leaf;
3137 struct btrfs_key key, found_key;
3138 struct btrfs_trans_handle *trans;
3139 struct inode *inode;
3140 u64 last_objectid = 0;
3141 int ret = 0, nr_unlink = 0;
3143 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3146 path = btrfs_alloc_path();
3151 path->reada = READA_BACK;
3153 key.objectid = BTRFS_ORPHAN_OBJECTID;
3154 key.type = BTRFS_ORPHAN_ITEM_KEY;
3155 key.offset = (u64)-1;
3158 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3163 * if ret == 0 means we found what we were searching for, which
3164 * is weird, but possible, so only screw with path if we didn't
3165 * find the key and see if we have stuff that matches
3169 if (path->slots[0] == 0)
3174 /* pull out the item */
3175 leaf = path->nodes[0];
3176 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3178 /* make sure the item matches what we want */
3179 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3181 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3184 /* release the path since we're done with it */
3185 btrfs_release_path(path);
3188 * this is where we are basically btrfs_lookup, without the
3189 * crossing root thing. we store the inode number in the
3190 * offset of the orphan item.
3193 if (found_key.offset == last_objectid) {
3195 "Error removing orphan entry, stopping orphan cleanup");
3200 last_objectid = found_key.offset;
3202 found_key.objectid = found_key.offset;
3203 found_key.type = BTRFS_INODE_ITEM_KEY;
3204 found_key.offset = 0;
3205 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3206 ret = PTR_ERR_OR_ZERO(inode);
3207 if (ret && ret != -ENOENT)
3210 if (ret == -ENOENT && root == fs_info->tree_root) {
3211 struct btrfs_root *dead_root;
3212 int is_dead_root = 0;
3215 * this is an orphan in the tree root. Currently these
3216 * could come from 2 sources:
3217 * a) a snapshot deletion in progress
3218 * b) a free space cache inode
3219 * We need to distinguish those two, as the snapshot
3220 * orphan must not get deleted.
3221 * find_dead_roots already ran before us, so if this
3222 * is a snapshot deletion, we should find the root
3223 * in the fs_roots radix tree.
3226 spin_lock(&fs_info->fs_roots_radix_lock);
3227 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3228 (unsigned long)found_key.objectid);
3229 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3231 spin_unlock(&fs_info->fs_roots_radix_lock);
3234 /* prevent this orphan from being found again */
3235 key.offset = found_key.objectid - 1;
3242 * If we have an inode with links, there are a couple of
3243 * possibilities. Old kernels (before v3.12) used to create an
3244 * orphan item for truncate indicating that there were possibly
3245 * extent items past i_size that needed to be deleted. In v3.12,
3246 * truncate was changed to update i_size in sync with the extent
3247 * items, but the (useless) orphan item was still created. Since
3248 * v4.18, we don't create the orphan item for truncate at all.
3250 * So, this item could mean that we need to do a truncate, but
3251 * only if this filesystem was last used on a pre-v3.12 kernel
3252 * and was not cleanly unmounted. The odds of that are quite
3253 * slim, and it's a pain to do the truncate now, so just delete
3256 * It's also possible that this orphan item was supposed to be
3257 * deleted but wasn't. The inode number may have been reused,
3258 * but either way, we can delete the orphan item.
3260 if (ret == -ENOENT || inode->i_nlink) {
3263 trans = btrfs_start_transaction(root, 1);
3264 if (IS_ERR(trans)) {
3265 ret = PTR_ERR(trans);
3268 btrfs_debug(fs_info, "auto deleting %Lu",
3269 found_key.objectid);
3270 ret = btrfs_del_orphan_item(trans, root,
3271 found_key.objectid);
3272 btrfs_end_transaction(trans);
3280 /* this will do delete_inode and everything for us */
3283 /* release the path since we're done with it */
3284 btrfs_release_path(path);
3286 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3288 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3289 trans = btrfs_join_transaction(root);
3291 btrfs_end_transaction(trans);
3295 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3299 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3300 btrfs_free_path(path);
3305 * very simple check to peek ahead in the leaf looking for xattrs. If we
3306 * don't find any xattrs, we know there can't be any acls.
3308 * slot is the slot the inode is in, objectid is the objectid of the inode
3310 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3311 int slot, u64 objectid,
3312 int *first_xattr_slot)
3314 u32 nritems = btrfs_header_nritems(leaf);
3315 struct btrfs_key found_key;
3316 static u64 xattr_access = 0;
3317 static u64 xattr_default = 0;
3320 if (!xattr_access) {
3321 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3322 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3323 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3324 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3328 *first_xattr_slot = -1;
3329 while (slot < nritems) {
3330 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3332 /* we found a different objectid, there must not be acls */
3333 if (found_key.objectid != objectid)
3336 /* we found an xattr, assume we've got an acl */
3337 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3338 if (*first_xattr_slot == -1)
3339 *first_xattr_slot = slot;
3340 if (found_key.offset == xattr_access ||
3341 found_key.offset == xattr_default)
3346 * we found a key greater than an xattr key, there can't
3347 * be any acls later on
3349 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3356 * it goes inode, inode backrefs, xattrs, extents,
3357 * so if there are a ton of hard links to an inode there can
3358 * be a lot of backrefs. Don't waste time searching too hard,
3359 * this is just an optimization
3364 /* we hit the end of the leaf before we found an xattr or
3365 * something larger than an xattr. We have to assume the inode
3368 if (*first_xattr_slot == -1)
3369 *first_xattr_slot = slot;
3374 * read an inode from the btree into the in-memory inode
3376 static int btrfs_read_locked_inode(struct inode *inode,
3377 struct btrfs_path *in_path)
3379 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3380 struct btrfs_path *path = in_path;
3381 struct extent_buffer *leaf;
3382 struct btrfs_inode_item *inode_item;
3383 struct btrfs_root *root = BTRFS_I(inode)->root;
3384 struct btrfs_key location;
3389 bool filled = false;
3390 int first_xattr_slot;
3392 ret = btrfs_fill_inode(inode, &rdev);
3397 path = btrfs_alloc_path();
3402 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3404 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3406 if (path != in_path)
3407 btrfs_free_path(path);
3411 leaf = path->nodes[0];
3416 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3417 struct btrfs_inode_item);
3418 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3419 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3420 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3421 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3422 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3423 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3424 round_up(i_size_read(inode), fs_info->sectorsize));
3426 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3427 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3429 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3430 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3432 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3433 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3435 BTRFS_I(inode)->i_otime.tv_sec =
3436 btrfs_timespec_sec(leaf, &inode_item->otime);
3437 BTRFS_I(inode)->i_otime.tv_nsec =
3438 btrfs_timespec_nsec(leaf, &inode_item->otime);
3440 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3441 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3442 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3444 inode_set_iversion_queried(inode,
3445 btrfs_inode_sequence(leaf, inode_item));
3446 inode->i_generation = BTRFS_I(inode)->generation;
3448 rdev = btrfs_inode_rdev(leaf, inode_item);
3450 BTRFS_I(inode)->index_cnt = (u64)-1;
3451 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3455 * If we were modified in the current generation and evicted from memory
3456 * and then re-read we need to do a full sync since we don't have any
3457 * idea about which extents were modified before we were evicted from
3460 * This is required for both inode re-read from disk and delayed inode
3461 * in delayed_nodes_tree.
3463 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3464 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3465 &BTRFS_I(inode)->runtime_flags);
3468 * We don't persist the id of the transaction where an unlink operation
3469 * against the inode was last made. So here we assume the inode might
3470 * have been evicted, and therefore the exact value of last_unlink_trans
3471 * lost, and set it to last_trans to avoid metadata inconsistencies
3472 * between the inode and its parent if the inode is fsync'ed and the log
3473 * replayed. For example, in the scenario:
3476 * ln mydir/foo mydir/bar
3479 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3480 * xfs_io -c fsync mydir/foo
3482 * mount fs, triggers fsync log replay
3484 * We must make sure that when we fsync our inode foo we also log its
3485 * parent inode, otherwise after log replay the parent still has the
3486 * dentry with the "bar" name but our inode foo has a link count of 1
3487 * and doesn't have an inode ref with the name "bar" anymore.
3489 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3490 * but it guarantees correctness at the expense of occasional full
3491 * transaction commits on fsync if our inode is a directory, or if our
3492 * inode is not a directory, logging its parent unnecessarily.
3494 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3497 * Same logic as for last_unlink_trans. We don't persist the generation
3498 * of the last transaction where this inode was used for a reflink
3499 * operation, so after eviction and reloading the inode we must be
3500 * pessimistic and assume the last transaction that modified the inode.
3502 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3505 if (inode->i_nlink != 1 ||
3506 path->slots[0] >= btrfs_header_nritems(leaf))
3509 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3510 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3513 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3514 if (location.type == BTRFS_INODE_REF_KEY) {
3515 struct btrfs_inode_ref *ref;
3517 ref = (struct btrfs_inode_ref *)ptr;
3518 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3519 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3520 struct btrfs_inode_extref *extref;
3522 extref = (struct btrfs_inode_extref *)ptr;
3523 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3528 * try to precache a NULL acl entry for files that don't have
3529 * any xattrs or acls
3531 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3532 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3533 if (first_xattr_slot != -1) {
3534 path->slots[0] = first_xattr_slot;
3535 ret = btrfs_load_inode_props(inode, path);
3538 "error loading props for ino %llu (root %llu): %d",
3539 btrfs_ino(BTRFS_I(inode)),
3540 root->root_key.objectid, ret);
3542 if (path != in_path)
3543 btrfs_free_path(path);
3546 cache_no_acl(inode);
3548 switch (inode->i_mode & S_IFMT) {
3550 inode->i_mapping->a_ops = &btrfs_aops;
3551 inode->i_fop = &btrfs_file_operations;
3552 inode->i_op = &btrfs_file_inode_operations;
3555 inode->i_fop = &btrfs_dir_file_operations;
3556 inode->i_op = &btrfs_dir_inode_operations;
3559 inode->i_op = &btrfs_symlink_inode_operations;
3560 inode_nohighmem(inode);
3561 inode->i_mapping->a_ops = &btrfs_aops;
3564 inode->i_op = &btrfs_special_inode_operations;
3565 init_special_inode(inode, inode->i_mode, rdev);
3569 btrfs_sync_inode_flags_to_i_flags(inode);
3574 * given a leaf and an inode, copy the inode fields into the leaf
3576 static void fill_inode_item(struct btrfs_trans_handle *trans,
3577 struct extent_buffer *leaf,
3578 struct btrfs_inode_item *item,
3579 struct inode *inode)
3581 struct btrfs_map_token token;
3583 btrfs_init_map_token(&token, leaf);
3585 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3586 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3587 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3588 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3589 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3591 btrfs_set_token_timespec_sec(&token, &item->atime,
3592 inode->i_atime.tv_sec);
3593 btrfs_set_token_timespec_nsec(&token, &item->atime,
3594 inode->i_atime.tv_nsec);
3596 btrfs_set_token_timespec_sec(&token, &item->mtime,
3597 inode->i_mtime.tv_sec);
3598 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3599 inode->i_mtime.tv_nsec);
3601 btrfs_set_token_timespec_sec(&token, &item->ctime,
3602 inode->i_ctime.tv_sec);
3603 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3604 inode->i_ctime.tv_nsec);
3606 btrfs_set_token_timespec_sec(&token, &item->otime,
3607 BTRFS_I(inode)->i_otime.tv_sec);
3608 btrfs_set_token_timespec_nsec(&token, &item->otime,
3609 BTRFS_I(inode)->i_otime.tv_nsec);
3611 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3612 btrfs_set_token_inode_generation(&token, item,
3613 BTRFS_I(inode)->generation);
3614 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3615 btrfs_set_token_inode_transid(&token, item, trans->transid);
3616 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3617 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3618 btrfs_set_token_inode_block_group(&token, item, 0);
3622 * copy everything in the in-memory inode into the btree.
3624 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3625 struct btrfs_root *root, struct inode *inode)
3627 struct btrfs_inode_item *inode_item;
3628 struct btrfs_path *path;
3629 struct extent_buffer *leaf;
3632 path = btrfs_alloc_path();
3636 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3644 leaf = path->nodes[0];
3645 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3646 struct btrfs_inode_item);
3648 fill_inode_item(trans, leaf, inode_item, inode);
3649 btrfs_mark_buffer_dirty(leaf);
3650 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
3653 btrfs_free_path(path);
3658 * copy everything in the in-memory inode into the btree.
3660 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3661 struct btrfs_root *root, struct inode *inode)
3663 struct btrfs_fs_info *fs_info = root->fs_info;
3667 * If the inode is a free space inode, we can deadlock during commit
3668 * if we put it into the delayed code.
3670 * The data relocation inode should also be directly updated
3673 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3674 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3675 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3676 btrfs_update_root_times(trans, root);
3678 ret = btrfs_delayed_update_inode(trans, root, inode);
3680 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
3684 return btrfs_update_inode_item(trans, root, inode);
3687 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3688 struct btrfs_root *root,
3689 struct inode *inode)
3693 ret = btrfs_update_inode(trans, root, inode);
3695 return btrfs_update_inode_item(trans, root, inode);
3700 * unlink helper that gets used here in inode.c and in the tree logging
3701 * recovery code. It remove a link in a directory with a given name, and
3702 * also drops the back refs in the inode to the directory
3704 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3705 struct btrfs_root *root,
3706 struct btrfs_inode *dir,
3707 struct btrfs_inode *inode,
3708 const char *name, int name_len)
3710 struct btrfs_fs_info *fs_info = root->fs_info;
3711 struct btrfs_path *path;
3713 struct btrfs_dir_item *di;
3715 u64 ino = btrfs_ino(inode);
3716 u64 dir_ino = btrfs_ino(dir);
3718 path = btrfs_alloc_path();
3724 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3725 name, name_len, -1);
3726 if (IS_ERR_OR_NULL(di)) {
3727 ret = di ? PTR_ERR(di) : -ENOENT;
3730 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3733 btrfs_release_path(path);
3736 * If we don't have dir index, we have to get it by looking up
3737 * the inode ref, since we get the inode ref, remove it directly,
3738 * it is unnecessary to do delayed deletion.
3740 * But if we have dir index, needn't search inode ref to get it.
3741 * Since the inode ref is close to the inode item, it is better
3742 * that we delay to delete it, and just do this deletion when
3743 * we update the inode item.
3745 if (inode->dir_index) {
3746 ret = btrfs_delayed_delete_inode_ref(inode);
3748 index = inode->dir_index;
3753 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3757 "failed to delete reference to %.*s, inode %llu parent %llu",
3758 name_len, name, ino, dir_ino);
3759 btrfs_abort_transaction(trans, ret);
3763 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3765 btrfs_abort_transaction(trans, ret);
3769 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3771 if (ret != 0 && ret != -ENOENT) {
3772 btrfs_abort_transaction(trans, ret);
3776 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3781 btrfs_abort_transaction(trans, ret);
3784 * If we have a pending delayed iput we could end up with the final iput
3785 * being run in btrfs-cleaner context. If we have enough of these built
3786 * up we can end up burning a lot of time in btrfs-cleaner without any
3787 * way to throttle the unlinks. Since we're currently holding a ref on
3788 * the inode we can run the delayed iput here without any issues as the
3789 * final iput won't be done until after we drop the ref we're currently
3792 btrfs_run_delayed_iput(fs_info, inode);
3794 btrfs_free_path(path);
3798 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3799 inode_inc_iversion(&inode->vfs_inode);
3800 inode_inc_iversion(&dir->vfs_inode);
3801 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3802 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3803 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3808 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3809 struct btrfs_root *root,
3810 struct btrfs_inode *dir, struct btrfs_inode *inode,
3811 const char *name, int name_len)
3814 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3816 drop_nlink(&inode->vfs_inode);
3817 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3823 * helper to start transaction for unlink and rmdir.
3825 * unlink and rmdir are special in btrfs, they do not always free space, so
3826 * if we cannot make our reservations the normal way try and see if there is
3827 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3828 * allow the unlink to occur.
3830 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3832 struct btrfs_root *root = BTRFS_I(dir)->root;
3835 * 1 for the possible orphan item
3836 * 1 for the dir item
3837 * 1 for the dir index
3838 * 1 for the inode ref
3841 return btrfs_start_transaction_fallback_global_rsv(root, 5);
3844 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3846 struct btrfs_root *root = BTRFS_I(dir)->root;
3847 struct btrfs_trans_handle *trans;
3848 struct inode *inode = d_inode(dentry);
3851 trans = __unlink_start_trans(dir);
3853 return PTR_ERR(trans);
3855 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
3858 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
3859 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
3860 dentry->d_name.len);
3864 if (inode->i_nlink == 0) {
3865 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3871 btrfs_end_transaction(trans);
3872 btrfs_btree_balance_dirty(root->fs_info);
3876 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3877 struct inode *dir, struct dentry *dentry)
3879 struct btrfs_root *root = BTRFS_I(dir)->root;
3880 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
3881 struct btrfs_path *path;
3882 struct extent_buffer *leaf;
3883 struct btrfs_dir_item *di;
3884 struct btrfs_key key;
3885 const char *name = dentry->d_name.name;
3886 int name_len = dentry->d_name.len;
3890 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
3892 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
3893 objectid = inode->root->root_key.objectid;
3894 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3895 objectid = inode->location.objectid;
3901 path = btrfs_alloc_path();
3905 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3906 name, name_len, -1);
3907 if (IS_ERR_OR_NULL(di)) {
3908 ret = di ? PTR_ERR(di) : -ENOENT;
3912 leaf = path->nodes[0];
3913 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3914 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3915 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3917 btrfs_abort_transaction(trans, ret);
3920 btrfs_release_path(path);
3923 * This is a placeholder inode for a subvolume we didn't have a
3924 * reference to at the time of the snapshot creation. In the meantime
3925 * we could have renamed the real subvol link into our snapshot, so
3926 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
3927 * Instead simply lookup the dir_index_item for this entry so we can
3928 * remove it. Otherwise we know we have a ref to the root and we can
3929 * call btrfs_del_root_ref, and it _shouldn't_ fail.
3931 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3932 di = btrfs_search_dir_index_item(root, path, dir_ino,
3934 if (IS_ERR_OR_NULL(di)) {
3939 btrfs_abort_transaction(trans, ret);
3943 leaf = path->nodes[0];
3944 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3946 btrfs_release_path(path);
3948 ret = btrfs_del_root_ref(trans, objectid,
3949 root->root_key.objectid, dir_ino,
3950 &index, name, name_len);
3952 btrfs_abort_transaction(trans, ret);
3957 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
3959 btrfs_abort_transaction(trans, ret);
3963 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
3964 inode_inc_iversion(dir);
3965 dir->i_mtime = dir->i_ctime = current_time(dir);
3966 ret = btrfs_update_inode_fallback(trans, root, dir);
3968 btrfs_abort_transaction(trans, ret);
3970 btrfs_free_path(path);
3975 * Helper to check if the subvolume references other subvolumes or if it's
3978 static noinline int may_destroy_subvol(struct btrfs_root *root)
3980 struct btrfs_fs_info *fs_info = root->fs_info;
3981 struct btrfs_path *path;
3982 struct btrfs_dir_item *di;
3983 struct btrfs_key key;
3987 path = btrfs_alloc_path();
3991 /* Make sure this root isn't set as the default subvol */
3992 dir_id = btrfs_super_root_dir(fs_info->super_copy);
3993 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
3994 dir_id, "default", 7, 0);
3995 if (di && !IS_ERR(di)) {
3996 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
3997 if (key.objectid == root->root_key.objectid) {
4000 "deleting default subvolume %llu is not allowed",
4004 btrfs_release_path(path);
4007 key.objectid = root->root_key.objectid;
4008 key.type = BTRFS_ROOT_REF_KEY;
4009 key.offset = (u64)-1;
4011 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4017 if (path->slots[0] > 0) {
4019 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4020 if (key.objectid == root->root_key.objectid &&
4021 key.type == BTRFS_ROOT_REF_KEY)
4025 btrfs_free_path(path);
4029 /* Delete all dentries for inodes belonging to the root */
4030 static void btrfs_prune_dentries(struct btrfs_root *root)
4032 struct btrfs_fs_info *fs_info = root->fs_info;
4033 struct rb_node *node;
4034 struct rb_node *prev;
4035 struct btrfs_inode *entry;
4036 struct inode *inode;
4039 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4040 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4042 spin_lock(&root->inode_lock);
4044 node = root->inode_tree.rb_node;
4048 entry = rb_entry(node, struct btrfs_inode, rb_node);
4050 if (objectid < btrfs_ino(entry))
4051 node = node->rb_left;
4052 else if (objectid > btrfs_ino(entry))
4053 node = node->rb_right;
4059 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4060 if (objectid <= btrfs_ino(entry)) {
4064 prev = rb_next(prev);
4068 entry = rb_entry(node, struct btrfs_inode, rb_node);
4069 objectid = btrfs_ino(entry) + 1;
4070 inode = igrab(&entry->vfs_inode);
4072 spin_unlock(&root->inode_lock);
4073 if (atomic_read(&inode->i_count) > 1)
4074 d_prune_aliases(inode);
4076 * btrfs_drop_inode will have it removed from the inode
4077 * cache when its usage count hits zero.
4081 spin_lock(&root->inode_lock);
4085 if (cond_resched_lock(&root->inode_lock))
4088 node = rb_next(node);
4090 spin_unlock(&root->inode_lock);
4093 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4095 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4096 struct btrfs_root *root = BTRFS_I(dir)->root;
4097 struct inode *inode = d_inode(dentry);
4098 struct btrfs_root *dest = BTRFS_I(inode)->root;
4099 struct btrfs_trans_handle *trans;
4100 struct btrfs_block_rsv block_rsv;
4106 * Don't allow to delete a subvolume with send in progress. This is
4107 * inside the inode lock so the error handling that has to drop the bit
4108 * again is not run concurrently.
4110 spin_lock(&dest->root_item_lock);
4111 if (dest->send_in_progress) {
4112 spin_unlock(&dest->root_item_lock);
4114 "attempt to delete subvolume %llu during send",
4115 dest->root_key.objectid);
4118 root_flags = btrfs_root_flags(&dest->root_item);
4119 btrfs_set_root_flags(&dest->root_item,
4120 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4121 spin_unlock(&dest->root_item_lock);
4123 down_write(&fs_info->subvol_sem);
4125 err = may_destroy_subvol(dest);
4129 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4131 * One for dir inode,
4132 * two for dir entries,
4133 * two for root ref/backref.
4135 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4139 trans = btrfs_start_transaction(root, 0);
4140 if (IS_ERR(trans)) {
4141 err = PTR_ERR(trans);
4144 trans->block_rsv = &block_rsv;
4145 trans->bytes_reserved = block_rsv.size;
4147 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4149 ret = btrfs_unlink_subvol(trans, dir, dentry);
4152 btrfs_abort_transaction(trans, ret);
4156 btrfs_record_root_in_trans(trans, dest);
4158 memset(&dest->root_item.drop_progress, 0,
4159 sizeof(dest->root_item.drop_progress));
4160 btrfs_set_root_drop_level(&dest->root_item, 0);
4161 btrfs_set_root_refs(&dest->root_item, 0);
4163 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4164 ret = btrfs_insert_orphan_item(trans,
4166 dest->root_key.objectid);
4168 btrfs_abort_transaction(trans, ret);
4174 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4175 BTRFS_UUID_KEY_SUBVOL,
4176 dest->root_key.objectid);
4177 if (ret && ret != -ENOENT) {
4178 btrfs_abort_transaction(trans, ret);
4182 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4183 ret = btrfs_uuid_tree_remove(trans,
4184 dest->root_item.received_uuid,
4185 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4186 dest->root_key.objectid);
4187 if (ret && ret != -ENOENT) {
4188 btrfs_abort_transaction(trans, ret);
4194 free_anon_bdev(dest->anon_dev);
4197 trans->block_rsv = NULL;
4198 trans->bytes_reserved = 0;
4199 ret = btrfs_end_transaction(trans);
4202 inode->i_flags |= S_DEAD;
4204 btrfs_subvolume_release_metadata(root, &block_rsv);
4206 up_write(&fs_info->subvol_sem);
4208 spin_lock(&dest->root_item_lock);
4209 root_flags = btrfs_root_flags(&dest->root_item);
4210 btrfs_set_root_flags(&dest->root_item,
4211 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4212 spin_unlock(&dest->root_item_lock);
4214 d_invalidate(dentry);
4215 btrfs_prune_dentries(dest);
4216 ASSERT(dest->send_in_progress == 0);
4219 if (dest->ino_cache_inode) {
4220 iput(dest->ino_cache_inode);
4221 dest->ino_cache_inode = NULL;
4228 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4230 struct inode *inode = d_inode(dentry);
4232 struct btrfs_root *root = BTRFS_I(dir)->root;
4233 struct btrfs_trans_handle *trans;
4234 u64 last_unlink_trans;
4236 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4238 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4239 return btrfs_delete_subvolume(dir, dentry);
4241 trans = __unlink_start_trans(dir);
4243 return PTR_ERR(trans);
4245 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4246 err = btrfs_unlink_subvol(trans, dir, dentry);
4250 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4254 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4256 /* now the directory is empty */
4257 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4258 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4259 dentry->d_name.len);
4261 btrfs_i_size_write(BTRFS_I(inode), 0);
4263 * Propagate the last_unlink_trans value of the deleted dir to
4264 * its parent directory. This is to prevent an unrecoverable
4265 * log tree in the case we do something like this:
4267 * 2) create snapshot under dir foo
4268 * 3) delete the snapshot
4271 * 6) fsync foo or some file inside foo
4273 if (last_unlink_trans >= trans->transid)
4274 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4277 btrfs_end_transaction(trans);
4278 btrfs_btree_balance_dirty(root->fs_info);
4284 * Return this if we need to call truncate_block for the last bit of the
4287 #define NEED_TRUNCATE_BLOCK 1
4290 * this can truncate away extent items, csum items and directory items.
4291 * It starts at a high offset and removes keys until it can't find
4292 * any higher than new_size
4294 * csum items that cross the new i_size are truncated to the new size
4297 * min_type is the minimum key type to truncate down to. If set to 0, this
4298 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4300 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4301 struct btrfs_root *root,
4302 struct btrfs_inode *inode,
4303 u64 new_size, u32 min_type)
4305 struct btrfs_fs_info *fs_info = root->fs_info;
4306 struct btrfs_path *path;
4307 struct extent_buffer *leaf;
4308 struct btrfs_file_extent_item *fi;
4309 struct btrfs_key key;
4310 struct btrfs_key found_key;
4311 u64 extent_start = 0;
4312 u64 extent_num_bytes = 0;
4313 u64 extent_offset = 0;
4315 u64 last_size = new_size;
4316 u32 found_type = (u8)-1;
4319 int pending_del_nr = 0;
4320 int pending_del_slot = 0;
4321 int extent_type = -1;
4323 u64 ino = btrfs_ino(inode);
4324 u64 bytes_deleted = 0;
4325 bool be_nice = false;
4326 bool should_throttle = false;
4327 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4328 struct extent_state *cached_state = NULL;
4330 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4333 * For non-free space inodes and non-shareable roots, we want to back
4334 * off from time to time. This means all inodes in subvolume roots,
4335 * reloc roots, and data reloc roots.
4337 if (!btrfs_is_free_space_inode(inode) &&
4338 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4341 path = btrfs_alloc_path();
4344 path->reada = READA_BACK;
4346 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4347 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4351 * We want to drop from the next block forward in case this
4352 * new size is not block aligned since we will be keeping the
4353 * last block of the extent just the way it is.
4355 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4356 fs_info->sectorsize),
4361 * This function is also used to drop the items in the log tree before
4362 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4363 * it is used to drop the logged items. So we shouldn't kill the delayed
4366 if (min_type == 0 && root == inode->root)
4367 btrfs_kill_delayed_inode_items(inode);
4370 key.offset = (u64)-1;
4375 * with a 16K leaf size and 128MB extents, you can actually queue
4376 * up a huge file in a single leaf. Most of the time that
4377 * bytes_deleted is > 0, it will be huge by the time we get here
4379 if (be_nice && bytes_deleted > SZ_32M &&
4380 btrfs_should_end_transaction(trans)) {
4385 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4391 /* there are no items in the tree for us to truncate, we're
4394 if (path->slots[0] == 0)
4400 u64 clear_start = 0, clear_len = 0;
4403 leaf = path->nodes[0];
4404 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4405 found_type = found_key.type;
4407 if (found_key.objectid != ino)
4410 if (found_type < min_type)
4413 item_end = found_key.offset;
4414 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4415 fi = btrfs_item_ptr(leaf, path->slots[0],
4416 struct btrfs_file_extent_item);
4417 extent_type = btrfs_file_extent_type(leaf, fi);
4418 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4420 btrfs_file_extent_num_bytes(leaf, fi);
4422 trace_btrfs_truncate_show_fi_regular(
4423 inode, leaf, fi, found_key.offset);
4424 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4425 item_end += btrfs_file_extent_ram_bytes(leaf,
4428 trace_btrfs_truncate_show_fi_inline(
4429 inode, leaf, fi, path->slots[0],
4434 if (found_type > min_type) {
4437 if (item_end < new_size)
4439 if (found_key.offset >= new_size)
4445 /* FIXME, shrink the extent if the ref count is only 1 */
4446 if (found_type != BTRFS_EXTENT_DATA_KEY)
4449 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4452 clear_start = found_key.offset;
4453 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4455 u64 orig_num_bytes =
4456 btrfs_file_extent_num_bytes(leaf, fi);
4457 extent_num_bytes = ALIGN(new_size -
4459 fs_info->sectorsize);
4460 clear_start = ALIGN(new_size, fs_info->sectorsize);
4461 btrfs_set_file_extent_num_bytes(leaf, fi,
4463 num_dec = (orig_num_bytes -
4465 if (test_bit(BTRFS_ROOT_SHAREABLE,
4468 inode_sub_bytes(&inode->vfs_inode,
4470 btrfs_mark_buffer_dirty(leaf);
4473 btrfs_file_extent_disk_num_bytes(leaf,
4475 extent_offset = found_key.offset -
4476 btrfs_file_extent_offset(leaf, fi);
4478 /* FIXME blocksize != 4096 */
4479 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4480 if (extent_start != 0) {
4482 if (test_bit(BTRFS_ROOT_SHAREABLE,
4484 inode_sub_bytes(&inode->vfs_inode,
4488 clear_len = num_dec;
4489 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4491 * we can't truncate inline items that have had
4495 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4496 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4497 btrfs_file_extent_compression(leaf, fi) == 0) {
4498 u32 size = (u32)(new_size - found_key.offset);
4500 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4501 size = btrfs_file_extent_calc_inline_size(size);
4502 btrfs_truncate_item(path, size, 1);
4503 } else if (!del_item) {
4505 * We have to bail so the last_size is set to
4506 * just before this extent.
4508 ret = NEED_TRUNCATE_BLOCK;
4512 * Inline extents are special, we just treat
4513 * them as a full sector worth in the file
4514 * extent tree just for simplicity sake.
4516 clear_len = fs_info->sectorsize;
4519 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4520 inode_sub_bytes(&inode->vfs_inode,
4521 item_end + 1 - new_size);
4525 * We use btrfs_truncate_inode_items() to clean up log trees for
4526 * multiple fsyncs, and in this case we don't want to clear the
4527 * file extent range because it's just the log.
4529 if (root == inode->root) {
4530 ret = btrfs_inode_clear_file_extent_range(inode,
4531 clear_start, clear_len);
4533 btrfs_abort_transaction(trans, ret);
4539 last_size = found_key.offset;
4541 last_size = new_size;
4543 if (!pending_del_nr) {
4544 /* no pending yet, add ourselves */
4545 pending_del_slot = path->slots[0];
4547 } else if (pending_del_nr &&
4548 path->slots[0] + 1 == pending_del_slot) {
4549 /* hop on the pending chunk */
4551 pending_del_slot = path->slots[0];
4558 should_throttle = false;
4561 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4562 struct btrfs_ref ref = { 0 };
4564 bytes_deleted += extent_num_bytes;
4566 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4567 extent_start, extent_num_bytes, 0);
4568 ref.real_root = root->root_key.objectid;
4569 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4570 ino, extent_offset);
4571 ret = btrfs_free_extent(trans, &ref);
4573 btrfs_abort_transaction(trans, ret);
4577 if (btrfs_should_throttle_delayed_refs(trans))
4578 should_throttle = true;
4582 if (found_type == BTRFS_INODE_ITEM_KEY)
4585 if (path->slots[0] == 0 ||
4586 path->slots[0] != pending_del_slot ||
4588 if (pending_del_nr) {
4589 ret = btrfs_del_items(trans, root, path,
4593 btrfs_abort_transaction(trans, ret);
4598 btrfs_release_path(path);
4601 * We can generate a lot of delayed refs, so we need to
4602 * throttle every once and a while and make sure we're
4603 * adding enough space to keep up with the work we are
4604 * generating. Since we hold a transaction here we
4605 * can't flush, and we don't want to FLUSH_LIMIT because
4606 * we could have generated too many delayed refs to
4607 * actually allocate, so just bail if we're short and
4608 * let the normal reservation dance happen higher up.
4610 if (should_throttle) {
4611 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4612 BTRFS_RESERVE_NO_FLUSH);
4624 if (ret >= 0 && pending_del_nr) {
4627 err = btrfs_del_items(trans, root, path, pending_del_slot,
4630 btrfs_abort_transaction(trans, err);
4634 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4635 ASSERT(last_size >= new_size);
4636 if (!ret && last_size > new_size)
4637 last_size = new_size;
4638 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4639 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4643 btrfs_free_path(path);
4648 * btrfs_truncate_block - read, zero a chunk and write a block
4649 * @inode - inode that we're zeroing
4650 * @from - the offset to start zeroing
4651 * @len - the length to zero, 0 to zero the entire range respective to the
4653 * @front - zero up to the offset instead of from the offset on
4655 * This will find the block for the "from" offset and cow the block and zero the
4656 * part we want to zero. This is used with truncate and hole punching.
4658 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4661 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4662 struct address_space *mapping = inode->i_mapping;
4663 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4664 struct btrfs_ordered_extent *ordered;
4665 struct extent_state *cached_state = NULL;
4666 struct extent_changeset *data_reserved = NULL;
4668 bool only_release_metadata = false;
4669 u32 blocksize = fs_info->sectorsize;
4670 pgoff_t index = from >> PAGE_SHIFT;
4671 unsigned offset = from & (blocksize - 1);
4673 gfp_t mask = btrfs_alloc_write_mask(mapping);
4674 size_t write_bytes = blocksize;
4679 if (IS_ALIGNED(offset, blocksize) &&
4680 (!len || IS_ALIGNED(len, blocksize)))
4683 block_start = round_down(from, blocksize);
4684 block_end = block_start + blocksize - 1;
4686 ret = btrfs_check_data_free_space(BTRFS_I(inode), &data_reserved,
4687 block_start, blocksize);
4689 if (btrfs_check_nocow_lock(BTRFS_I(inode), block_start,
4690 &write_bytes) > 0) {
4691 /* For nocow case, no need to reserve data space */
4692 only_release_metadata = true;
4697 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), blocksize);
4699 if (!only_release_metadata)
4700 btrfs_free_reserved_data_space(BTRFS_I(inode),
4701 data_reserved, block_start, blocksize);
4705 page = find_or_create_page(mapping, index, mask);
4707 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
4708 block_start, blocksize, true);
4709 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4714 if (!PageUptodate(page)) {
4715 ret = btrfs_readpage(NULL, page);
4717 if (page->mapping != mapping) {
4722 if (!PageUptodate(page)) {
4727 wait_on_page_writeback(page);
4729 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4730 set_page_extent_mapped(page);
4732 ordered = btrfs_lookup_ordered_extent(BTRFS_I(inode), block_start);
4734 unlock_extent_cached(io_tree, block_start, block_end,
4738 btrfs_start_ordered_extent(ordered, 1);
4739 btrfs_put_ordered_extent(ordered);
4743 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4744 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4745 0, 0, &cached_state);
4747 ret = btrfs_set_extent_delalloc(BTRFS_I(inode), block_start, block_end, 0,
4750 unlock_extent_cached(io_tree, block_start, block_end,
4755 if (offset != blocksize) {
4757 len = blocksize - offset;
4760 memset(kaddr + (block_start - page_offset(page)),
4763 memset(kaddr + (block_start - page_offset(page)) + offset,
4765 flush_dcache_page(page);
4768 ClearPageChecked(page);
4769 set_page_dirty(page);
4770 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4772 if (only_release_metadata)
4773 set_extent_bit(&BTRFS_I(inode)->io_tree, block_start,
4774 block_end, EXTENT_NORESERVE, NULL, GFP_NOFS);
4778 if (only_release_metadata)
4779 btrfs_delalloc_release_metadata(BTRFS_I(inode),
4782 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved,
4783 block_start, blocksize, true);
4785 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize);
4789 if (only_release_metadata)
4790 btrfs_check_nocow_unlock(BTRFS_I(inode));
4791 extent_changeset_free(data_reserved);
4795 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4796 u64 offset, u64 len)
4798 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4799 struct btrfs_trans_handle *trans;
4800 struct btrfs_drop_extents_args drop_args = { 0 };
4804 * Still need to make sure the inode looks like it's been updated so
4805 * that any holes get logged if we fsync.
4807 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4808 BTRFS_I(inode)->last_trans = fs_info->generation;
4809 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4810 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4815 * 1 - for the one we're dropping
4816 * 1 - for the one we're adding
4817 * 1 - for updating the inode.
4819 trans = btrfs_start_transaction(root, 3);
4821 return PTR_ERR(trans);
4823 drop_args.start = offset;
4824 drop_args.end = offset + len;
4825 drop_args.drop_cache = true;
4827 ret = btrfs_drop_extents(trans, root, BTRFS_I(inode), &drop_args);
4829 btrfs_abort_transaction(trans, ret);
4830 btrfs_end_transaction(trans);
4834 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4835 offset, 0, 0, len, 0, len, 0, 0, 0);
4837 btrfs_abort_transaction(trans, ret);
4839 btrfs_update_inode_bytes(BTRFS_I(inode), 0, drop_args.bytes_found);
4840 btrfs_update_inode(trans, root, inode);
4842 btrfs_end_transaction(trans);
4847 * This function puts in dummy file extents for the area we're creating a hole
4848 * for. So if we are truncating this file to a larger size we need to insert
4849 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4850 * the range between oldsize and size
4852 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4854 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4855 struct btrfs_root *root = BTRFS_I(inode)->root;
4856 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4857 struct extent_map *em = NULL;
4858 struct extent_state *cached_state = NULL;
4859 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4860 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4861 u64 block_end = ALIGN(size, fs_info->sectorsize);
4868 * If our size started in the middle of a block we need to zero out the
4869 * rest of the block before we expand the i_size, otherwise we could
4870 * expose stale data.
4872 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4876 if (size <= hole_start)
4879 btrfs_lock_and_flush_ordered_range(BTRFS_I(inode), hole_start,
4880 block_end - 1, &cached_state);
4881 cur_offset = hole_start;
4883 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4884 block_end - cur_offset);
4890 last_byte = min(extent_map_end(em), block_end);
4891 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4892 hole_size = last_byte - cur_offset;
4894 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4895 struct extent_map *hole_em;
4897 err = maybe_insert_hole(root, inode, cur_offset,
4902 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4903 cur_offset, hole_size);
4907 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4908 cur_offset + hole_size - 1, 0);
4909 hole_em = alloc_extent_map();
4911 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4912 &BTRFS_I(inode)->runtime_flags);
4915 hole_em->start = cur_offset;
4916 hole_em->len = hole_size;
4917 hole_em->orig_start = cur_offset;
4919 hole_em->block_start = EXTENT_MAP_HOLE;
4920 hole_em->block_len = 0;
4921 hole_em->orig_block_len = 0;
4922 hole_em->ram_bytes = hole_size;
4923 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4924 hole_em->generation = fs_info->generation;
4927 write_lock(&em_tree->lock);
4928 err = add_extent_mapping(em_tree, hole_em, 1);
4929 write_unlock(&em_tree->lock);
4932 btrfs_drop_extent_cache(BTRFS_I(inode),
4937 free_extent_map(hole_em);
4939 err = btrfs_inode_set_file_extent_range(BTRFS_I(inode),
4940 cur_offset, hole_size);
4945 free_extent_map(em);
4947 cur_offset = last_byte;
4948 if (cur_offset >= block_end)
4951 free_extent_map(em);
4952 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4956 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4958 struct btrfs_root *root = BTRFS_I(inode)->root;
4959 struct btrfs_trans_handle *trans;
4960 loff_t oldsize = i_size_read(inode);
4961 loff_t newsize = attr->ia_size;
4962 int mask = attr->ia_valid;
4966 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4967 * special case where we need to update the times despite not having
4968 * these flags set. For all other operations the VFS set these flags
4969 * explicitly if it wants a timestamp update.
4971 if (newsize != oldsize) {
4972 inode_inc_iversion(inode);
4973 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4974 inode->i_ctime = inode->i_mtime =
4975 current_time(inode);
4978 if (newsize > oldsize) {
4980 * Don't do an expanding truncate while snapshotting is ongoing.
4981 * This is to ensure the snapshot captures a fully consistent
4982 * state of this file - if the snapshot captures this expanding
4983 * truncation, it must capture all writes that happened before
4986 btrfs_drew_write_lock(&root->snapshot_lock);
4987 ret = btrfs_cont_expand(inode, oldsize, newsize);
4989 btrfs_drew_write_unlock(&root->snapshot_lock);
4993 trans = btrfs_start_transaction(root, 1);
4994 if (IS_ERR(trans)) {
4995 btrfs_drew_write_unlock(&root->snapshot_lock);
4996 return PTR_ERR(trans);
4999 i_size_write(inode, newsize);
5000 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5001 pagecache_isize_extended(inode, oldsize, newsize);
5002 ret = btrfs_update_inode(trans, root, inode);
5003 btrfs_drew_write_unlock(&root->snapshot_lock);
5004 btrfs_end_transaction(trans);
5008 * We're truncating a file that used to have good data down to
5009 * zero. Make sure any new writes to the file get on disk
5013 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5014 &BTRFS_I(inode)->runtime_flags);
5016 truncate_setsize(inode, newsize);
5018 inode_dio_wait(inode);
5020 ret = btrfs_truncate(inode, newsize == oldsize);
5021 if (ret && inode->i_nlink) {
5025 * Truncate failed, so fix up the in-memory size. We
5026 * adjusted disk_i_size down as we removed extents, so
5027 * wait for disk_i_size to be stable and then update the
5028 * in-memory size to match.
5030 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5033 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5040 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5042 struct inode *inode = d_inode(dentry);
5043 struct btrfs_root *root = BTRFS_I(inode)->root;
5046 if (btrfs_root_readonly(root))
5049 err = setattr_prepare(dentry, attr);
5053 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5054 err = btrfs_setsize(inode, attr);
5059 if (attr->ia_valid) {
5060 setattr_copy(inode, attr);
5061 inode_inc_iversion(inode);
5062 err = btrfs_dirty_inode(inode);
5064 if (!err && attr->ia_valid & ATTR_MODE)
5065 err = posix_acl_chmod(inode, inode->i_mode);
5072 * While truncating the inode pages during eviction, we get the VFS calling
5073 * btrfs_invalidatepage() against each page of the inode. This is slow because
5074 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5075 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5076 * extent_state structures over and over, wasting lots of time.
5078 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5079 * those expensive operations on a per page basis and do only the ordered io
5080 * finishing, while we release here the extent_map and extent_state structures,
5081 * without the excessive merging and splitting.
5083 static void evict_inode_truncate_pages(struct inode *inode)
5085 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5086 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5087 struct rb_node *node;
5089 ASSERT(inode->i_state & I_FREEING);
5090 truncate_inode_pages_final(&inode->i_data);
5092 write_lock(&map_tree->lock);
5093 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5094 struct extent_map *em;
5096 node = rb_first_cached(&map_tree->map);
5097 em = rb_entry(node, struct extent_map, rb_node);
5098 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5099 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5100 remove_extent_mapping(map_tree, em);
5101 free_extent_map(em);
5102 if (need_resched()) {
5103 write_unlock(&map_tree->lock);
5105 write_lock(&map_tree->lock);
5108 write_unlock(&map_tree->lock);
5111 * Keep looping until we have no more ranges in the io tree.
5112 * We can have ongoing bios started by readahead that have
5113 * their endio callback (extent_io.c:end_bio_extent_readpage)
5114 * still in progress (unlocked the pages in the bio but did not yet
5115 * unlocked the ranges in the io tree). Therefore this means some
5116 * ranges can still be locked and eviction started because before
5117 * submitting those bios, which are executed by a separate task (work
5118 * queue kthread), inode references (inode->i_count) were not taken
5119 * (which would be dropped in the end io callback of each bio).
5120 * Therefore here we effectively end up waiting for those bios and
5121 * anyone else holding locked ranges without having bumped the inode's
5122 * reference count - if we don't do it, when they access the inode's
5123 * io_tree to unlock a range it may be too late, leading to an
5124 * use-after-free issue.
5126 spin_lock(&io_tree->lock);
5127 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5128 struct extent_state *state;
5129 struct extent_state *cached_state = NULL;
5132 unsigned state_flags;
5134 node = rb_first(&io_tree->state);
5135 state = rb_entry(node, struct extent_state, rb_node);
5136 start = state->start;
5138 state_flags = state->state;
5139 spin_unlock(&io_tree->lock);
5141 lock_extent_bits(io_tree, start, end, &cached_state);
5144 * If still has DELALLOC flag, the extent didn't reach disk,
5145 * and its reserved space won't be freed by delayed_ref.
5146 * So we need to free its reserved space here.
5147 * (Refer to comment in btrfs_invalidatepage, case 2)
5149 * Note, end is the bytenr of last byte, so we need + 1 here.
5151 if (state_flags & EXTENT_DELALLOC)
5152 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5155 clear_extent_bit(io_tree, start, end,
5156 EXTENT_LOCKED | EXTENT_DELALLOC |
5157 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5161 spin_lock(&io_tree->lock);
5163 spin_unlock(&io_tree->lock);
5166 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5167 struct btrfs_block_rsv *rsv)
5169 struct btrfs_fs_info *fs_info = root->fs_info;
5170 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5171 struct btrfs_trans_handle *trans;
5172 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5176 * Eviction should be taking place at some place safe because of our
5177 * delayed iputs. However the normal flushing code will run delayed
5178 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5180 * We reserve the delayed_refs_extra here again because we can't use
5181 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5182 * above. We reserve our extra bit here because we generate a ton of
5183 * delayed refs activity by truncating.
5185 * If we cannot make our reservation we'll attempt to steal from the
5186 * global reserve, because we really want to be able to free up space.
5188 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5189 BTRFS_RESERVE_FLUSH_EVICT);
5192 * Try to steal from the global reserve if there is space for
5195 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5196 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5198 "could not allocate space for delete; will truncate on mount");
5199 return ERR_PTR(-ENOSPC);
5201 delayed_refs_extra = 0;
5204 trans = btrfs_join_transaction(root);
5208 if (delayed_refs_extra) {
5209 trans->block_rsv = &fs_info->trans_block_rsv;
5210 trans->bytes_reserved = delayed_refs_extra;
5211 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5212 delayed_refs_extra, 1);
5217 void btrfs_evict_inode(struct inode *inode)
5219 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5220 struct btrfs_trans_handle *trans;
5221 struct btrfs_root *root = BTRFS_I(inode)->root;
5222 struct btrfs_block_rsv *rsv;
5225 trace_btrfs_inode_evict(inode);
5232 evict_inode_truncate_pages(inode);
5234 if (inode->i_nlink &&
5235 ((btrfs_root_refs(&root->root_item) != 0 &&
5236 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5237 btrfs_is_free_space_inode(BTRFS_I(inode))))
5240 if (is_bad_inode(inode))
5243 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5245 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5248 if (inode->i_nlink > 0) {
5249 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5250 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5254 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5258 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5261 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5264 btrfs_i_size_write(BTRFS_I(inode), 0);
5267 trans = evict_refill_and_join(root, rsv);
5271 trans->block_rsv = rsv;
5273 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5275 trans->block_rsv = &fs_info->trans_block_rsv;
5276 btrfs_end_transaction(trans);
5277 btrfs_btree_balance_dirty(fs_info);
5278 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5285 * Errors here aren't a big deal, it just means we leave orphan items in
5286 * the tree. They will be cleaned up on the next mount. If the inode
5287 * number gets reused, cleanup deletes the orphan item without doing
5288 * anything, and unlink reuses the existing orphan item.
5290 * If it turns out that we are dropping too many of these, we might want
5291 * to add a mechanism for retrying these after a commit.
5293 trans = evict_refill_and_join(root, rsv);
5294 if (!IS_ERR(trans)) {
5295 trans->block_rsv = rsv;
5296 btrfs_orphan_del(trans, BTRFS_I(inode));
5297 trans->block_rsv = &fs_info->trans_block_rsv;
5298 btrfs_end_transaction(trans);
5301 if (!(root == fs_info->tree_root ||
5302 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5303 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5306 btrfs_free_block_rsv(fs_info, rsv);
5309 * If we didn't successfully delete, the orphan item will still be in
5310 * the tree and we'll retry on the next mount. Again, we might also want
5311 * to retry these periodically in the future.
5313 btrfs_remove_delayed_node(BTRFS_I(inode));
5318 * Return the key found in the dir entry in the location pointer, fill @type
5319 * with BTRFS_FT_*, and return 0.
5321 * If no dir entries were found, returns -ENOENT.
5322 * If found a corrupted location in dir entry, returns -EUCLEAN.
5324 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5325 struct btrfs_key *location, u8 *type)
5327 const char *name = dentry->d_name.name;
5328 int namelen = dentry->d_name.len;
5329 struct btrfs_dir_item *di;
5330 struct btrfs_path *path;
5331 struct btrfs_root *root = BTRFS_I(dir)->root;
5334 path = btrfs_alloc_path();
5338 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5340 if (IS_ERR_OR_NULL(di)) {
5341 ret = di ? PTR_ERR(di) : -ENOENT;
5345 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5346 if (location->type != BTRFS_INODE_ITEM_KEY &&
5347 location->type != BTRFS_ROOT_ITEM_KEY) {
5349 btrfs_warn(root->fs_info,
5350 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5351 __func__, name, btrfs_ino(BTRFS_I(dir)),
5352 location->objectid, location->type, location->offset);
5355 *type = btrfs_dir_type(path->nodes[0], di);
5357 btrfs_free_path(path);
5362 * when we hit a tree root in a directory, the btrfs part of the inode
5363 * needs to be changed to reflect the root directory of the tree root. This
5364 * is kind of like crossing a mount point.
5366 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5368 struct dentry *dentry,
5369 struct btrfs_key *location,
5370 struct btrfs_root **sub_root)
5372 struct btrfs_path *path;
5373 struct btrfs_root *new_root;
5374 struct btrfs_root_ref *ref;
5375 struct extent_buffer *leaf;
5376 struct btrfs_key key;
5380 path = btrfs_alloc_path();
5387 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5388 key.type = BTRFS_ROOT_REF_KEY;
5389 key.offset = location->objectid;
5391 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5398 leaf = path->nodes[0];
5399 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5400 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5401 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5404 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5405 (unsigned long)(ref + 1),
5406 dentry->d_name.len);
5410 btrfs_release_path(path);
5412 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5413 if (IS_ERR(new_root)) {
5414 err = PTR_ERR(new_root);
5418 *sub_root = new_root;
5419 location->objectid = btrfs_root_dirid(&new_root->root_item);
5420 location->type = BTRFS_INODE_ITEM_KEY;
5421 location->offset = 0;
5424 btrfs_free_path(path);
5428 static void inode_tree_add(struct inode *inode)
5430 struct btrfs_root *root = BTRFS_I(inode)->root;
5431 struct btrfs_inode *entry;
5433 struct rb_node *parent;
5434 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5435 u64 ino = btrfs_ino(BTRFS_I(inode));
5437 if (inode_unhashed(inode))
5440 spin_lock(&root->inode_lock);
5441 p = &root->inode_tree.rb_node;
5444 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5446 if (ino < btrfs_ino(entry))
5447 p = &parent->rb_left;
5448 else if (ino > btrfs_ino(entry))
5449 p = &parent->rb_right;
5451 WARN_ON(!(entry->vfs_inode.i_state &
5452 (I_WILL_FREE | I_FREEING)));
5453 rb_replace_node(parent, new, &root->inode_tree);
5454 RB_CLEAR_NODE(parent);
5455 spin_unlock(&root->inode_lock);
5459 rb_link_node(new, parent, p);
5460 rb_insert_color(new, &root->inode_tree);
5461 spin_unlock(&root->inode_lock);
5464 static void inode_tree_del(struct btrfs_inode *inode)
5466 struct btrfs_root *root = inode->root;
5469 spin_lock(&root->inode_lock);
5470 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5471 rb_erase(&inode->rb_node, &root->inode_tree);
5472 RB_CLEAR_NODE(&inode->rb_node);
5473 empty = RB_EMPTY_ROOT(&root->inode_tree);
5475 spin_unlock(&root->inode_lock);
5477 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5478 spin_lock(&root->inode_lock);
5479 empty = RB_EMPTY_ROOT(&root->inode_tree);
5480 spin_unlock(&root->inode_lock);
5482 btrfs_add_dead_root(root);
5487 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5489 struct btrfs_iget_args *args = p;
5491 inode->i_ino = args->ino;
5492 BTRFS_I(inode)->location.objectid = args->ino;
5493 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5494 BTRFS_I(inode)->location.offset = 0;
5495 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5496 BUG_ON(args->root && !BTRFS_I(inode)->root);
5500 static int btrfs_find_actor(struct inode *inode, void *opaque)
5502 struct btrfs_iget_args *args = opaque;
5504 return args->ino == BTRFS_I(inode)->location.objectid &&
5505 args->root == BTRFS_I(inode)->root;
5508 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5509 struct btrfs_root *root)
5511 struct inode *inode;
5512 struct btrfs_iget_args args;
5513 unsigned long hashval = btrfs_inode_hash(ino, root);
5518 inode = iget5_locked(s, hashval, btrfs_find_actor,
5519 btrfs_init_locked_inode,
5525 * Get an inode object given its inode number and corresponding root.
5526 * Path can be preallocated to prevent recursing back to iget through
5527 * allocator. NULL is also valid but may require an additional allocation
5530 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5531 struct btrfs_root *root, struct btrfs_path *path)
5533 struct inode *inode;
5535 inode = btrfs_iget_locked(s, ino, root);
5537 return ERR_PTR(-ENOMEM);
5539 if (inode->i_state & I_NEW) {
5542 ret = btrfs_read_locked_inode(inode, path);
5544 inode_tree_add(inode);
5545 unlock_new_inode(inode);
5549 * ret > 0 can come from btrfs_search_slot called by
5550 * btrfs_read_locked_inode, this means the inode item
5555 inode = ERR_PTR(ret);
5562 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5564 return btrfs_iget_path(s, ino, root, NULL);
5567 static struct inode *new_simple_dir(struct super_block *s,
5568 struct btrfs_key *key,
5569 struct btrfs_root *root)
5571 struct inode *inode = new_inode(s);
5574 return ERR_PTR(-ENOMEM);
5576 BTRFS_I(inode)->root = btrfs_grab_root(root);
5577 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5578 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5580 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5582 * We only need lookup, the rest is read-only and there's no inode
5583 * associated with the dentry
5585 inode->i_op = &simple_dir_inode_operations;
5586 inode->i_opflags &= ~IOP_XATTR;
5587 inode->i_fop = &simple_dir_operations;
5588 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5589 inode->i_mtime = current_time(inode);
5590 inode->i_atime = inode->i_mtime;
5591 inode->i_ctime = inode->i_mtime;
5592 BTRFS_I(inode)->i_otime = inode->i_mtime;
5597 static inline u8 btrfs_inode_type(struct inode *inode)
5600 * Compile-time asserts that generic FT_* types still match
5603 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5604 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5605 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5606 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5607 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5608 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5609 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5610 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5612 return fs_umode_to_ftype(inode->i_mode);
5615 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5617 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5618 struct inode *inode;
5619 struct btrfs_root *root = BTRFS_I(dir)->root;
5620 struct btrfs_root *sub_root = root;
5621 struct btrfs_key location;
5625 if (dentry->d_name.len > BTRFS_NAME_LEN)
5626 return ERR_PTR(-ENAMETOOLONG);
5628 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5630 return ERR_PTR(ret);
5632 if (location.type == BTRFS_INODE_ITEM_KEY) {
5633 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5637 /* Do extra check against inode mode with di_type */
5638 if (btrfs_inode_type(inode) != di_type) {
5640 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5641 inode->i_mode, btrfs_inode_type(inode),
5644 return ERR_PTR(-EUCLEAN);
5649 ret = fixup_tree_root_location(fs_info, dir, dentry,
5650 &location, &sub_root);
5653 inode = ERR_PTR(ret);
5655 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5657 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5659 if (root != sub_root)
5660 btrfs_put_root(sub_root);
5662 if (!IS_ERR(inode) && root != sub_root) {
5663 down_read(&fs_info->cleanup_work_sem);
5664 if (!sb_rdonly(inode->i_sb))
5665 ret = btrfs_orphan_cleanup(sub_root);
5666 up_read(&fs_info->cleanup_work_sem);
5669 inode = ERR_PTR(ret);
5676 static int btrfs_dentry_delete(const struct dentry *dentry)
5678 struct btrfs_root *root;
5679 struct inode *inode = d_inode(dentry);
5681 if (!inode && !IS_ROOT(dentry))
5682 inode = d_inode(dentry->d_parent);
5685 root = BTRFS_I(inode)->root;
5686 if (btrfs_root_refs(&root->root_item) == 0)
5689 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5695 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5698 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5700 if (inode == ERR_PTR(-ENOENT))
5702 return d_splice_alias(inode, dentry);
5706 * All this infrastructure exists because dir_emit can fault, and we are holding
5707 * the tree lock when doing readdir. For now just allocate a buffer and copy
5708 * our information into that, and then dir_emit from the buffer. This is
5709 * similar to what NFS does, only we don't keep the buffer around in pagecache
5710 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5711 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5714 static int btrfs_opendir(struct inode *inode, struct file *file)
5716 struct btrfs_file_private *private;
5718 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5721 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5722 if (!private->filldir_buf) {
5726 file->private_data = private;
5737 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5740 struct dir_entry *entry = addr;
5741 char *name = (char *)(entry + 1);
5743 ctx->pos = get_unaligned(&entry->offset);
5744 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5745 get_unaligned(&entry->ino),
5746 get_unaligned(&entry->type)))
5748 addr += sizeof(struct dir_entry) +
5749 get_unaligned(&entry->name_len);
5755 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5757 struct inode *inode = file_inode(file);
5758 struct btrfs_root *root = BTRFS_I(inode)->root;
5759 struct btrfs_file_private *private = file->private_data;
5760 struct btrfs_dir_item *di;
5761 struct btrfs_key key;
5762 struct btrfs_key found_key;
5763 struct btrfs_path *path;
5765 struct list_head ins_list;
5766 struct list_head del_list;
5768 struct extent_buffer *leaf;
5775 struct btrfs_key location;
5777 if (!dir_emit_dots(file, ctx))
5780 path = btrfs_alloc_path();
5784 addr = private->filldir_buf;
5785 path->reada = READA_FORWARD;
5787 INIT_LIST_HEAD(&ins_list);
5788 INIT_LIST_HEAD(&del_list);
5789 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5792 key.type = BTRFS_DIR_INDEX_KEY;
5793 key.offset = ctx->pos;
5794 key.objectid = btrfs_ino(BTRFS_I(inode));
5796 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5801 struct dir_entry *entry;
5803 leaf = path->nodes[0];
5804 slot = path->slots[0];
5805 if (slot >= btrfs_header_nritems(leaf)) {
5806 ret = btrfs_next_leaf(root, path);
5814 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5816 if (found_key.objectid != key.objectid)
5818 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5820 if (found_key.offset < ctx->pos)
5822 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5824 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5825 name_len = btrfs_dir_name_len(leaf, di);
5826 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5828 btrfs_release_path(path);
5829 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5832 addr = private->filldir_buf;
5839 put_unaligned(name_len, &entry->name_len);
5840 name_ptr = (char *)(entry + 1);
5841 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5843 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5845 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5846 put_unaligned(location.objectid, &entry->ino);
5847 put_unaligned(found_key.offset, &entry->offset);
5849 addr += sizeof(struct dir_entry) + name_len;
5850 total_len += sizeof(struct dir_entry) + name_len;
5854 btrfs_release_path(path);
5856 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5860 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5865 * Stop new entries from being returned after we return the last
5868 * New directory entries are assigned a strictly increasing
5869 * offset. This means that new entries created during readdir
5870 * are *guaranteed* to be seen in the future by that readdir.
5871 * This has broken buggy programs which operate on names as
5872 * they're returned by readdir. Until we re-use freed offsets
5873 * we have this hack to stop new entries from being returned
5874 * under the assumption that they'll never reach this huge
5877 * This is being careful not to overflow 32bit loff_t unless the
5878 * last entry requires it because doing so has broken 32bit apps
5881 if (ctx->pos >= INT_MAX)
5882 ctx->pos = LLONG_MAX;
5889 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5890 btrfs_free_path(path);
5895 * This is somewhat expensive, updating the tree every time the
5896 * inode changes. But, it is most likely to find the inode in cache.
5897 * FIXME, needs more benchmarking...there are no reasons other than performance
5898 * to keep or drop this code.
5900 static int btrfs_dirty_inode(struct inode *inode)
5902 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5903 struct btrfs_root *root = BTRFS_I(inode)->root;
5904 struct btrfs_trans_handle *trans;
5907 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5910 trans = btrfs_join_transaction(root);
5912 return PTR_ERR(trans);
5914 ret = btrfs_update_inode(trans, root, inode);
5915 if (ret && ret == -ENOSPC) {
5916 /* whoops, lets try again with the full transaction */
5917 btrfs_end_transaction(trans);
5918 trans = btrfs_start_transaction(root, 1);
5920 return PTR_ERR(trans);
5922 ret = btrfs_update_inode(trans, root, inode);
5924 btrfs_end_transaction(trans);
5925 if (BTRFS_I(inode)->delayed_node)
5926 btrfs_balance_delayed_items(fs_info);
5932 * This is a copy of file_update_time. We need this so we can return error on
5933 * ENOSPC for updating the inode in the case of file write and mmap writes.
5935 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5938 struct btrfs_root *root = BTRFS_I(inode)->root;
5939 bool dirty = flags & ~S_VERSION;
5941 if (btrfs_root_readonly(root))
5944 if (flags & S_VERSION)
5945 dirty |= inode_maybe_inc_iversion(inode, dirty);
5946 if (flags & S_CTIME)
5947 inode->i_ctime = *now;
5948 if (flags & S_MTIME)
5949 inode->i_mtime = *now;
5950 if (flags & S_ATIME)
5951 inode->i_atime = *now;
5952 return dirty ? btrfs_dirty_inode(inode) : 0;
5956 * find the highest existing sequence number in a directory
5957 * and then set the in-memory index_cnt variable to reflect
5958 * free sequence numbers
5960 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5962 struct btrfs_root *root = inode->root;
5963 struct btrfs_key key, found_key;
5964 struct btrfs_path *path;
5965 struct extent_buffer *leaf;
5968 key.objectid = btrfs_ino(inode);
5969 key.type = BTRFS_DIR_INDEX_KEY;
5970 key.offset = (u64)-1;
5972 path = btrfs_alloc_path();
5976 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5979 /* FIXME: we should be able to handle this */
5985 * MAGIC NUMBER EXPLANATION:
5986 * since we search a directory based on f_pos we have to start at 2
5987 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5988 * else has to start at 2
5990 if (path->slots[0] == 0) {
5991 inode->index_cnt = 2;
5997 leaf = path->nodes[0];
5998 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6000 if (found_key.objectid != btrfs_ino(inode) ||
6001 found_key.type != BTRFS_DIR_INDEX_KEY) {
6002 inode->index_cnt = 2;
6006 inode->index_cnt = found_key.offset + 1;
6008 btrfs_free_path(path);
6013 * helper to find a free sequence number in a given directory. This current
6014 * code is very simple, later versions will do smarter things in the btree
6016 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6020 if (dir->index_cnt == (u64)-1) {
6021 ret = btrfs_inode_delayed_dir_index_count(dir);
6023 ret = btrfs_set_inode_index_count(dir);
6029 *index = dir->index_cnt;
6035 static int btrfs_insert_inode_locked(struct inode *inode)
6037 struct btrfs_iget_args args;
6039 args.ino = BTRFS_I(inode)->location.objectid;
6040 args.root = BTRFS_I(inode)->root;
6042 return insert_inode_locked4(inode,
6043 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6044 btrfs_find_actor, &args);
6048 * Inherit flags from the parent inode.
6050 * Currently only the compression flags and the cow flags are inherited.
6052 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6059 flags = BTRFS_I(dir)->flags;
6061 if (flags & BTRFS_INODE_NOCOMPRESS) {
6062 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6063 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6064 } else if (flags & BTRFS_INODE_COMPRESS) {
6065 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6066 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6069 if (flags & BTRFS_INODE_NODATACOW) {
6070 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6071 if (S_ISREG(inode->i_mode))
6072 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6075 btrfs_sync_inode_flags_to_i_flags(inode);
6078 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6079 struct btrfs_root *root,
6081 const char *name, int name_len,
6082 u64 ref_objectid, u64 objectid,
6083 umode_t mode, u64 *index)
6085 struct btrfs_fs_info *fs_info = root->fs_info;
6086 struct inode *inode;
6087 struct btrfs_inode_item *inode_item;
6088 struct btrfs_key *location;
6089 struct btrfs_path *path;
6090 struct btrfs_inode_ref *ref;
6091 struct btrfs_key key[2];
6093 int nitems = name ? 2 : 1;
6095 unsigned int nofs_flag;
6098 path = btrfs_alloc_path();
6100 return ERR_PTR(-ENOMEM);
6102 nofs_flag = memalloc_nofs_save();
6103 inode = new_inode(fs_info->sb);
6104 memalloc_nofs_restore(nofs_flag);
6106 btrfs_free_path(path);
6107 return ERR_PTR(-ENOMEM);
6111 * O_TMPFILE, set link count to 0, so that after this point,
6112 * we fill in an inode item with the correct link count.
6115 set_nlink(inode, 0);
6118 * we have to initialize this early, so we can reclaim the inode
6119 * number if we fail afterwards in this function.
6121 inode->i_ino = objectid;
6124 trace_btrfs_inode_request(dir);
6126 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6128 btrfs_free_path(path);
6130 return ERR_PTR(ret);
6136 * index_cnt is ignored for everything but a dir,
6137 * btrfs_set_inode_index_count has an explanation for the magic
6140 BTRFS_I(inode)->index_cnt = 2;
6141 BTRFS_I(inode)->dir_index = *index;
6142 BTRFS_I(inode)->root = btrfs_grab_root(root);
6143 BTRFS_I(inode)->generation = trans->transid;
6144 inode->i_generation = BTRFS_I(inode)->generation;
6147 * We could have gotten an inode number from somebody who was fsynced
6148 * and then removed in this same transaction, so let's just set full
6149 * sync since it will be a full sync anyway and this will blow away the
6150 * old info in the log.
6152 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6154 key[0].objectid = objectid;
6155 key[0].type = BTRFS_INODE_ITEM_KEY;
6158 sizes[0] = sizeof(struct btrfs_inode_item);
6162 * Start new inodes with an inode_ref. This is slightly more
6163 * efficient for small numbers of hard links since they will
6164 * be packed into one item. Extended refs will kick in if we
6165 * add more hard links than can fit in the ref item.
6167 key[1].objectid = objectid;
6168 key[1].type = BTRFS_INODE_REF_KEY;
6169 key[1].offset = ref_objectid;
6171 sizes[1] = name_len + sizeof(*ref);
6174 location = &BTRFS_I(inode)->location;
6175 location->objectid = objectid;
6176 location->offset = 0;
6177 location->type = BTRFS_INODE_ITEM_KEY;
6179 ret = btrfs_insert_inode_locked(inode);
6185 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6189 inode_init_owner(inode, dir, mode);
6190 inode_set_bytes(inode, 0);
6192 inode->i_mtime = current_time(inode);
6193 inode->i_atime = inode->i_mtime;
6194 inode->i_ctime = inode->i_mtime;
6195 BTRFS_I(inode)->i_otime = inode->i_mtime;
6197 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6198 struct btrfs_inode_item);
6199 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6200 sizeof(*inode_item));
6201 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6204 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6205 struct btrfs_inode_ref);
6206 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6207 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6208 ptr = (unsigned long)(ref + 1);
6209 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6212 btrfs_mark_buffer_dirty(path->nodes[0]);
6213 btrfs_free_path(path);
6215 btrfs_inherit_iflags(inode, dir);
6217 if (S_ISREG(mode)) {
6218 if (btrfs_test_opt(fs_info, NODATASUM))
6219 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6220 if (btrfs_test_opt(fs_info, NODATACOW))
6221 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6222 BTRFS_INODE_NODATASUM;
6225 inode_tree_add(inode);
6227 trace_btrfs_inode_new(inode);
6228 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6230 btrfs_update_root_times(trans, root);
6232 ret = btrfs_inode_inherit_props(trans, inode, dir);
6235 "error inheriting props for ino %llu (root %llu): %d",
6236 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6241 discard_new_inode(inode);
6244 BTRFS_I(dir)->index_cnt--;
6245 btrfs_free_path(path);
6246 return ERR_PTR(ret);
6250 * utility function to add 'inode' into 'parent_inode' with
6251 * a give name and a given sequence number.
6252 * if 'add_backref' is true, also insert a backref from the
6253 * inode to the parent directory.
6255 int btrfs_add_link(struct btrfs_trans_handle *trans,
6256 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6257 const char *name, int name_len, int add_backref, u64 index)
6260 struct btrfs_key key;
6261 struct btrfs_root *root = parent_inode->root;
6262 u64 ino = btrfs_ino(inode);
6263 u64 parent_ino = btrfs_ino(parent_inode);
6265 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6266 memcpy(&key, &inode->root->root_key, sizeof(key));
6269 key.type = BTRFS_INODE_ITEM_KEY;
6273 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6274 ret = btrfs_add_root_ref(trans, key.objectid,
6275 root->root_key.objectid, parent_ino,
6276 index, name, name_len);
6277 } else if (add_backref) {
6278 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6282 /* Nothing to clean up yet */
6286 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6287 btrfs_inode_type(&inode->vfs_inode), index);
6288 if (ret == -EEXIST || ret == -EOVERFLOW)
6291 btrfs_abort_transaction(trans, ret);
6295 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6297 inode_inc_iversion(&parent_inode->vfs_inode);
6299 * If we are replaying a log tree, we do not want to update the mtime
6300 * and ctime of the parent directory with the current time, since the
6301 * log replay procedure is responsible for setting them to their correct
6302 * values (the ones it had when the fsync was done).
6304 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6305 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6307 parent_inode->vfs_inode.i_mtime = now;
6308 parent_inode->vfs_inode.i_ctime = now;
6310 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6312 btrfs_abort_transaction(trans, ret);
6316 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6319 err = btrfs_del_root_ref(trans, key.objectid,
6320 root->root_key.objectid, parent_ino,
6321 &local_index, name, name_len);
6323 btrfs_abort_transaction(trans, err);
6324 } else if (add_backref) {
6328 err = btrfs_del_inode_ref(trans, root, name, name_len,
6329 ino, parent_ino, &local_index);
6331 btrfs_abort_transaction(trans, err);
6334 /* Return the original error code */
6338 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6339 struct btrfs_inode *dir, struct dentry *dentry,
6340 struct btrfs_inode *inode, int backref, u64 index)
6342 int err = btrfs_add_link(trans, dir, inode,
6343 dentry->d_name.name, dentry->d_name.len,
6350 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6351 umode_t mode, dev_t rdev)
6353 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6354 struct btrfs_trans_handle *trans;
6355 struct btrfs_root *root = BTRFS_I(dir)->root;
6356 struct inode *inode = NULL;
6362 * 2 for inode item and ref
6364 * 1 for xattr if selinux is on
6366 trans = btrfs_start_transaction(root, 5);
6368 return PTR_ERR(trans);
6370 err = btrfs_find_free_ino(root, &objectid);
6374 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6375 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6377 if (IS_ERR(inode)) {
6378 err = PTR_ERR(inode);
6384 * If the active LSM wants to access the inode during
6385 * d_instantiate it needs these. Smack checks to see
6386 * if the filesystem supports xattrs by looking at the
6389 inode->i_op = &btrfs_special_inode_operations;
6390 init_special_inode(inode, inode->i_mode, rdev);
6392 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6396 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6401 btrfs_update_inode(trans, root, inode);
6402 d_instantiate_new(dentry, inode);
6405 btrfs_end_transaction(trans);
6406 btrfs_btree_balance_dirty(fs_info);
6408 inode_dec_link_count(inode);
6409 discard_new_inode(inode);
6414 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6415 umode_t mode, bool excl)
6417 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6418 struct btrfs_trans_handle *trans;
6419 struct btrfs_root *root = BTRFS_I(dir)->root;
6420 struct inode *inode = NULL;
6426 * 2 for inode item and ref
6428 * 1 for xattr if selinux is on
6430 trans = btrfs_start_transaction(root, 5);
6432 return PTR_ERR(trans);
6434 err = btrfs_find_free_ino(root, &objectid);
6438 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6439 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6441 if (IS_ERR(inode)) {
6442 err = PTR_ERR(inode);
6447 * If the active LSM wants to access the inode during
6448 * d_instantiate it needs these. Smack checks to see
6449 * if the filesystem supports xattrs by looking at the
6452 inode->i_fop = &btrfs_file_operations;
6453 inode->i_op = &btrfs_file_inode_operations;
6454 inode->i_mapping->a_ops = &btrfs_aops;
6456 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6460 err = btrfs_update_inode(trans, root, inode);
6464 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6469 d_instantiate_new(dentry, inode);
6472 btrfs_end_transaction(trans);
6474 inode_dec_link_count(inode);
6475 discard_new_inode(inode);
6477 btrfs_btree_balance_dirty(fs_info);
6481 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6482 struct dentry *dentry)
6484 struct btrfs_trans_handle *trans = NULL;
6485 struct btrfs_root *root = BTRFS_I(dir)->root;
6486 struct inode *inode = d_inode(old_dentry);
6487 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6492 /* do not allow sys_link's with other subvols of the same device */
6493 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6496 if (inode->i_nlink >= BTRFS_LINK_MAX)
6499 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6504 * 2 items for inode and inode ref
6505 * 2 items for dir items
6506 * 1 item for parent inode
6507 * 1 item for orphan item deletion if O_TMPFILE
6509 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6510 if (IS_ERR(trans)) {
6511 err = PTR_ERR(trans);
6516 /* There are several dir indexes for this inode, clear the cache. */
6517 BTRFS_I(inode)->dir_index = 0ULL;
6519 inode_inc_iversion(inode);
6520 inode->i_ctime = current_time(inode);
6522 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6524 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6530 struct dentry *parent = dentry->d_parent;
6532 err = btrfs_update_inode(trans, root, inode);
6535 if (inode->i_nlink == 1) {
6537 * If new hard link count is 1, it's a file created
6538 * with open(2) O_TMPFILE flag.
6540 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6544 d_instantiate(dentry, inode);
6545 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6550 btrfs_end_transaction(trans);
6552 inode_dec_link_count(inode);
6555 btrfs_btree_balance_dirty(fs_info);
6559 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6561 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6562 struct inode *inode = NULL;
6563 struct btrfs_trans_handle *trans;
6564 struct btrfs_root *root = BTRFS_I(dir)->root;
6570 * 2 items for inode and ref
6571 * 2 items for dir items
6572 * 1 for xattr if selinux is on
6574 trans = btrfs_start_transaction(root, 5);
6576 return PTR_ERR(trans);
6578 err = btrfs_find_free_ino(root, &objectid);
6582 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6583 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6584 S_IFDIR | mode, &index);
6585 if (IS_ERR(inode)) {
6586 err = PTR_ERR(inode);
6591 /* these must be set before we unlock the inode */
6592 inode->i_op = &btrfs_dir_inode_operations;
6593 inode->i_fop = &btrfs_dir_file_operations;
6595 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6599 btrfs_i_size_write(BTRFS_I(inode), 0);
6600 err = btrfs_update_inode(trans, root, inode);
6604 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6605 dentry->d_name.name,
6606 dentry->d_name.len, 0, index);
6610 d_instantiate_new(dentry, inode);
6613 btrfs_end_transaction(trans);
6615 inode_dec_link_count(inode);
6616 discard_new_inode(inode);
6618 btrfs_btree_balance_dirty(fs_info);
6622 static noinline int uncompress_inline(struct btrfs_path *path,
6624 size_t pg_offset, u64 extent_offset,
6625 struct btrfs_file_extent_item *item)
6628 struct extent_buffer *leaf = path->nodes[0];
6631 unsigned long inline_size;
6635 WARN_ON(pg_offset != 0);
6636 compress_type = btrfs_file_extent_compression(leaf, item);
6637 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6638 inline_size = btrfs_file_extent_inline_item_len(leaf,
6639 btrfs_item_nr(path->slots[0]));
6640 tmp = kmalloc(inline_size, GFP_NOFS);
6643 ptr = btrfs_file_extent_inline_start(item);
6645 read_extent_buffer(leaf, tmp, ptr, inline_size);
6647 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6648 ret = btrfs_decompress(compress_type, tmp, page,
6649 extent_offset, inline_size, max_size);
6652 * decompression code contains a memset to fill in any space between the end
6653 * of the uncompressed data and the end of max_size in case the decompressed
6654 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6655 * the end of an inline extent and the beginning of the next block, so we
6656 * cover that region here.
6659 if (max_size + pg_offset < PAGE_SIZE) {
6660 char *map = kmap(page);
6661 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6669 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6670 * @inode: file to search in
6671 * @page: page to read extent data into if the extent is inline
6672 * @pg_offset: offset into @page to copy to
6673 * @start: file offset
6674 * @len: length of range starting at @start
6676 * This returns the first &struct extent_map which overlaps with the given
6677 * range, reading it from the B-tree and caching it if necessary. Note that
6678 * there may be more extents which overlap the given range after the returned
6681 * If @page is not NULL and the extent is inline, this also reads the extent
6682 * data directly into the page and marks the extent up to date in the io_tree.
6684 * Return: ERR_PTR on error, non-NULL extent_map on success.
6686 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6687 struct page *page, size_t pg_offset,
6690 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6692 u64 extent_start = 0;
6694 u64 objectid = btrfs_ino(inode);
6695 int extent_type = -1;
6696 struct btrfs_path *path = NULL;
6697 struct btrfs_root *root = inode->root;
6698 struct btrfs_file_extent_item *item;
6699 struct extent_buffer *leaf;
6700 struct btrfs_key found_key;
6701 struct extent_map *em = NULL;
6702 struct extent_map_tree *em_tree = &inode->extent_tree;
6703 struct extent_io_tree *io_tree = &inode->io_tree;
6705 read_lock(&em_tree->lock);
6706 em = lookup_extent_mapping(em_tree, start, len);
6707 read_unlock(&em_tree->lock);
6710 if (em->start > start || em->start + em->len <= start)
6711 free_extent_map(em);
6712 else if (em->block_start == EXTENT_MAP_INLINE && page)
6713 free_extent_map(em);
6717 em = alloc_extent_map();
6722 em->start = EXTENT_MAP_HOLE;
6723 em->orig_start = EXTENT_MAP_HOLE;
6725 em->block_len = (u64)-1;
6727 path = btrfs_alloc_path();
6733 /* Chances are we'll be called again, so go ahead and do readahead */
6734 path->reada = READA_FORWARD;
6737 * The same explanation in load_free_space_cache applies here as well,
6738 * we only read when we're loading the free space cache, and at that
6739 * point the commit_root has everything we need.
6741 if (btrfs_is_free_space_inode(inode)) {
6742 path->search_commit_root = 1;
6743 path->skip_locking = 1;
6746 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6749 } else if (ret > 0) {
6750 if (path->slots[0] == 0)
6756 leaf = path->nodes[0];
6757 item = btrfs_item_ptr(leaf, path->slots[0],
6758 struct btrfs_file_extent_item);
6759 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6760 if (found_key.objectid != objectid ||
6761 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6763 * If we backup past the first extent we want to move forward
6764 * and see if there is an extent in front of us, otherwise we'll
6765 * say there is a hole for our whole search range which can
6772 extent_type = btrfs_file_extent_type(leaf, item);
6773 extent_start = found_key.offset;
6774 extent_end = btrfs_file_extent_end(path);
6775 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6776 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6777 /* Only regular file could have regular/prealloc extent */
6778 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6781 "regular/prealloc extent found for non-regular inode %llu",
6785 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6787 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6788 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6793 if (start >= extent_end) {
6795 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6796 ret = btrfs_next_leaf(root, path);
6802 leaf = path->nodes[0];
6804 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6805 if (found_key.objectid != objectid ||
6806 found_key.type != BTRFS_EXTENT_DATA_KEY)
6808 if (start + len <= found_key.offset)
6810 if (start > found_key.offset)
6813 /* New extent overlaps with existing one */
6815 em->orig_start = start;
6816 em->len = found_key.offset - start;
6817 em->block_start = EXTENT_MAP_HOLE;
6821 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6823 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6824 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6826 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6830 size_t extent_offset;
6836 size = btrfs_file_extent_ram_bytes(leaf, item);
6837 extent_offset = page_offset(page) + pg_offset - extent_start;
6838 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6839 size - extent_offset);
6840 em->start = extent_start + extent_offset;
6841 em->len = ALIGN(copy_size, fs_info->sectorsize);
6842 em->orig_block_len = em->len;
6843 em->orig_start = em->start;
6844 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6846 if (!PageUptodate(page)) {
6847 if (btrfs_file_extent_compression(leaf, item) !=
6848 BTRFS_COMPRESS_NONE) {
6849 ret = uncompress_inline(path, page, pg_offset,
6850 extent_offset, item);
6855 read_extent_buffer(leaf, map + pg_offset, ptr,
6857 if (pg_offset + copy_size < PAGE_SIZE) {
6858 memset(map + pg_offset + copy_size, 0,
6859 PAGE_SIZE - pg_offset -
6864 flush_dcache_page(page);
6866 set_extent_uptodate(io_tree, em->start,
6867 extent_map_end(em) - 1, NULL, GFP_NOFS);
6872 em->orig_start = start;
6874 em->block_start = EXTENT_MAP_HOLE;
6877 btrfs_release_path(path);
6878 if (em->start > start || extent_map_end(em) <= start) {
6880 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6881 em->start, em->len, start, len);
6886 write_lock(&em_tree->lock);
6887 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6888 write_unlock(&em_tree->lock);
6890 btrfs_free_path(path);
6892 trace_btrfs_get_extent(root, inode, em);
6895 free_extent_map(em);
6896 return ERR_PTR(ret);
6901 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6904 struct extent_map *em;
6905 struct extent_map *hole_em = NULL;
6906 u64 delalloc_start = start;
6912 em = btrfs_get_extent(inode, NULL, 0, start, len);
6916 * If our em maps to:
6918 * - a pre-alloc extent,
6919 * there might actually be delalloc bytes behind it.
6921 if (em->block_start != EXTENT_MAP_HOLE &&
6922 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6927 /* check to see if we've wrapped (len == -1 or similar) */
6936 /* ok, we didn't find anything, lets look for delalloc */
6937 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6938 end, len, EXTENT_DELALLOC, 1);
6939 delalloc_end = delalloc_start + delalloc_len;
6940 if (delalloc_end < delalloc_start)
6941 delalloc_end = (u64)-1;
6944 * We didn't find anything useful, return the original results from
6947 if (delalloc_start > end || delalloc_end <= start) {
6954 * Adjust the delalloc_start to make sure it doesn't go backwards from
6955 * the start they passed in
6957 delalloc_start = max(start, delalloc_start);
6958 delalloc_len = delalloc_end - delalloc_start;
6960 if (delalloc_len > 0) {
6963 const u64 hole_end = extent_map_end(hole_em);
6965 em = alloc_extent_map();
6973 * When btrfs_get_extent can't find anything it returns one
6976 * Make sure what it found really fits our range, and adjust to
6977 * make sure it is based on the start from the caller
6979 if (hole_end <= start || hole_em->start > end) {
6980 free_extent_map(hole_em);
6983 hole_start = max(hole_em->start, start);
6984 hole_len = hole_end - hole_start;
6987 if (hole_em && delalloc_start > hole_start) {
6989 * Our hole starts before our delalloc, so we have to
6990 * return just the parts of the hole that go until the
6993 em->len = min(hole_len, delalloc_start - hole_start);
6994 em->start = hole_start;
6995 em->orig_start = hole_start;
6997 * Don't adjust block start at all, it is fixed at
7000 em->block_start = hole_em->block_start;
7001 em->block_len = hole_len;
7002 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7003 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7006 * Hole is out of passed range or it starts after
7009 em->start = delalloc_start;
7010 em->len = delalloc_len;
7011 em->orig_start = delalloc_start;
7012 em->block_start = EXTENT_MAP_DELALLOC;
7013 em->block_len = delalloc_len;
7020 free_extent_map(hole_em);
7022 free_extent_map(em);
7023 return ERR_PTR(err);
7028 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7031 const u64 orig_start,
7032 const u64 block_start,
7033 const u64 block_len,
7034 const u64 orig_block_len,
7035 const u64 ram_bytes,
7038 struct extent_map *em = NULL;
7041 if (type != BTRFS_ORDERED_NOCOW) {
7042 em = create_io_em(inode, start, len, orig_start, block_start,
7043 block_len, orig_block_len, ram_bytes,
7044 BTRFS_COMPRESS_NONE, /* compress_type */
7049 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7053 free_extent_map(em);
7054 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7063 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7066 struct btrfs_root *root = inode->root;
7067 struct btrfs_fs_info *fs_info = root->fs_info;
7068 struct extent_map *em;
7069 struct btrfs_key ins;
7073 alloc_hint = get_extent_allocation_hint(inode, start, len);
7074 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7075 0, alloc_hint, &ins, 1, 1);
7077 return ERR_PTR(ret);
7079 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7080 ins.objectid, ins.offset, ins.offset,
7081 ins.offset, BTRFS_ORDERED_REGULAR);
7082 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7084 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7091 * Check if we can do nocow write into the range [@offset, @offset + @len)
7093 * @offset: File offset
7094 * @len: The length to write, will be updated to the nocow writeable
7096 * @orig_start: (optional) Return the original file offset of the file extent
7097 * @orig_len: (optional) Return the original on-disk length of the file extent
7098 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7099 * @strict: if true, omit optimizations that might force us into unnecessary
7100 * cow. e.g., don't trust generation number.
7102 * This function will flush ordered extents in the range to ensure proper
7103 * nocow checks for (nowait == false) case.
7106 * >0 and update @len if we can do nocow write
7107 * 0 if we can't do nocow write
7108 * <0 if error happened
7110 * NOTE: This only checks the file extents, caller is responsible to wait for
7111 * any ordered extents.
7113 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7114 u64 *orig_start, u64 *orig_block_len,
7115 u64 *ram_bytes, bool strict)
7117 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7118 struct btrfs_path *path;
7120 struct extent_buffer *leaf;
7121 struct btrfs_root *root = BTRFS_I(inode)->root;
7122 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7123 struct btrfs_file_extent_item *fi;
7124 struct btrfs_key key;
7131 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7133 path = btrfs_alloc_path();
7137 ret = btrfs_lookup_file_extent(NULL, root, path,
7138 btrfs_ino(BTRFS_I(inode)), offset, 0);
7142 slot = path->slots[0];
7145 /* can't find the item, must cow */
7152 leaf = path->nodes[0];
7153 btrfs_item_key_to_cpu(leaf, &key, slot);
7154 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7155 key.type != BTRFS_EXTENT_DATA_KEY) {
7156 /* not our file or wrong item type, must cow */
7160 if (key.offset > offset) {
7161 /* Wrong offset, must cow */
7165 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7166 found_type = btrfs_file_extent_type(leaf, fi);
7167 if (found_type != BTRFS_FILE_EXTENT_REG &&
7168 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7169 /* not a regular extent, must cow */
7173 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7176 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7177 if (extent_end <= offset)
7180 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7181 if (disk_bytenr == 0)
7184 if (btrfs_file_extent_compression(leaf, fi) ||
7185 btrfs_file_extent_encryption(leaf, fi) ||
7186 btrfs_file_extent_other_encoding(leaf, fi))
7190 * Do the same check as in btrfs_cross_ref_exist but without the
7191 * unnecessary search.
7194 (btrfs_file_extent_generation(leaf, fi) <=
7195 btrfs_root_last_snapshot(&root->root_item)))
7198 backref_offset = btrfs_file_extent_offset(leaf, fi);
7201 *orig_start = key.offset - backref_offset;
7202 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7203 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7206 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7209 num_bytes = min(offset + *len, extent_end) - offset;
7210 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7213 range_end = round_up(offset + num_bytes,
7214 root->fs_info->sectorsize) - 1;
7215 ret = test_range_bit(io_tree, offset, range_end,
7216 EXTENT_DELALLOC, 0, NULL);
7223 btrfs_release_path(path);
7226 * look for other files referencing this extent, if we
7227 * find any we must cow
7230 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7231 key.offset - backref_offset, disk_bytenr,
7239 * adjust disk_bytenr and num_bytes to cover just the bytes
7240 * in this extent we are about to write. If there
7241 * are any csums in that range we have to cow in order
7242 * to keep the csums correct
7244 disk_bytenr += backref_offset;
7245 disk_bytenr += offset - key.offset;
7246 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7249 * all of the above have passed, it is safe to overwrite this extent
7255 btrfs_free_path(path);
7259 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7260 struct extent_state **cached_state, bool writing)
7262 struct btrfs_ordered_extent *ordered;
7266 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7269 * We're concerned with the entire range that we're going to be
7270 * doing DIO to, so we need to make sure there's no ordered
7271 * extents in this range.
7273 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7274 lockend - lockstart + 1);
7277 * We need to make sure there are no buffered pages in this
7278 * range either, we could have raced between the invalidate in
7279 * generic_file_direct_write and locking the extent. The
7280 * invalidate needs to happen so that reads after a write do not
7284 (!writing || !filemap_range_has_page(inode->i_mapping,
7285 lockstart, lockend)))
7288 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7293 * If we are doing a DIO read and the ordered extent we
7294 * found is for a buffered write, we can not wait for it
7295 * to complete and retry, because if we do so we can
7296 * deadlock with concurrent buffered writes on page
7297 * locks. This happens only if our DIO read covers more
7298 * than one extent map, if at this point has already
7299 * created an ordered extent for a previous extent map
7300 * and locked its range in the inode's io tree, and a
7301 * concurrent write against that previous extent map's
7302 * range and this range started (we unlock the ranges
7303 * in the io tree only when the bios complete and
7304 * buffered writes always lock pages before attempting
7305 * to lock range in the io tree).
7308 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7309 btrfs_start_ordered_extent(ordered, 1);
7312 btrfs_put_ordered_extent(ordered);
7315 * We could trigger writeback for this range (and wait
7316 * for it to complete) and then invalidate the pages for
7317 * this range (through invalidate_inode_pages2_range()),
7318 * but that can lead us to a deadlock with a concurrent
7319 * call to readahead (a buffered read or a defrag call
7320 * triggered a readahead) on a page lock due to an
7321 * ordered dio extent we created before but did not have
7322 * yet a corresponding bio submitted (whence it can not
7323 * complete), which makes readahead wait for that
7324 * ordered extent to complete while holding a lock on
7339 /* The callers of this must take lock_extent() */
7340 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7341 u64 len, u64 orig_start, u64 block_start,
7342 u64 block_len, u64 orig_block_len,
7343 u64 ram_bytes, int compress_type,
7346 struct extent_map_tree *em_tree;
7347 struct extent_map *em;
7350 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7351 type == BTRFS_ORDERED_COMPRESSED ||
7352 type == BTRFS_ORDERED_NOCOW ||
7353 type == BTRFS_ORDERED_REGULAR);
7355 em_tree = &inode->extent_tree;
7356 em = alloc_extent_map();
7358 return ERR_PTR(-ENOMEM);
7361 em->orig_start = orig_start;
7363 em->block_len = block_len;
7364 em->block_start = block_start;
7365 em->orig_block_len = orig_block_len;
7366 em->ram_bytes = ram_bytes;
7367 em->generation = -1;
7368 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7369 if (type == BTRFS_ORDERED_PREALLOC) {
7370 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7371 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7372 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7373 em->compress_type = compress_type;
7377 btrfs_drop_extent_cache(inode, em->start,
7378 em->start + em->len - 1, 0);
7379 write_lock(&em_tree->lock);
7380 ret = add_extent_mapping(em_tree, em, 1);
7381 write_unlock(&em_tree->lock);
7383 * The caller has taken lock_extent(), who could race with us
7386 } while (ret == -EEXIST);
7389 free_extent_map(em);
7390 return ERR_PTR(ret);
7393 /* em got 2 refs now, callers needs to do free_extent_map once. */
7398 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7399 struct inode *inode,
7400 struct btrfs_dio_data *dio_data,
7403 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7404 struct extent_map *em = *map;
7408 * We don't allocate a new extent in the following cases
7410 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7412 * 2) The extent is marked as PREALLOC. We're good to go here and can
7413 * just use the extent.
7416 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7417 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7418 em->block_start != EXTENT_MAP_HOLE)) {
7420 u64 block_start, orig_start, orig_block_len, ram_bytes;
7422 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7423 type = BTRFS_ORDERED_PREALLOC;
7425 type = BTRFS_ORDERED_NOCOW;
7426 len = min(len, em->len - (start - em->start));
7427 block_start = em->block_start + (start - em->start);
7429 if (can_nocow_extent(inode, start, &len, &orig_start,
7430 &orig_block_len, &ram_bytes, false) == 1 &&
7431 btrfs_inc_nocow_writers(fs_info, block_start)) {
7432 struct extent_map *em2;
7434 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7435 orig_start, block_start,
7436 len, orig_block_len,
7438 btrfs_dec_nocow_writers(fs_info, block_start);
7439 if (type == BTRFS_ORDERED_PREALLOC) {
7440 free_extent_map(em);
7444 if (em2 && IS_ERR(em2)) {
7449 * For inode marked NODATACOW or extent marked PREALLOC,
7450 * use the existing or preallocated extent, so does not
7451 * need to adjust btrfs_space_info's bytes_may_use.
7453 btrfs_free_reserved_data_space_noquota(fs_info, len);
7458 /* this will cow the extent */
7459 free_extent_map(em);
7460 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7466 len = min(len, em->len - (start - em->start));
7470 * Need to update the i_size under the extent lock so buffered
7471 * readers will get the updated i_size when we unlock.
7473 if (start + len > i_size_read(inode))
7474 i_size_write(inode, start + len);
7476 dio_data->reserve -= len;
7481 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7482 loff_t length, unsigned int flags, struct iomap *iomap,
7483 struct iomap *srcmap)
7485 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7486 struct extent_map *em;
7487 struct extent_state *cached_state = NULL;
7488 struct btrfs_dio_data *dio_data = NULL;
7489 u64 lockstart, lockend;
7490 const bool write = !!(flags & IOMAP_WRITE);
7493 bool unlock_extents = false;
7496 len = min_t(u64, len, fs_info->sectorsize);
7499 lockend = start + len - 1;
7502 * The generic stuff only does filemap_write_and_wait_range, which
7503 * isn't enough if we've written compressed pages to this area, so we
7504 * need to flush the dirty pages again to make absolutely sure that any
7505 * outstanding dirty pages are on disk.
7507 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7508 &BTRFS_I(inode)->runtime_flags)) {
7509 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7510 start + length - 1);
7515 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7519 dio_data->length = length;
7521 dio_data->reserve = round_up(length, fs_info->sectorsize);
7522 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7523 &dio_data->data_reserved,
7524 start, dio_data->reserve);
7526 extent_changeset_free(dio_data->data_reserved);
7531 iomap->private = dio_data;
7535 * If this errors out it's because we couldn't invalidate pagecache for
7536 * this range and we need to fallback to buffered.
7538 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7543 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7550 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7551 * io. INLINE is special, and we could probably kludge it in here, but
7552 * it's still buffered so for safety lets just fall back to the generic
7555 * For COMPRESSED we _have_ to read the entire extent in so we can
7556 * decompress it, so there will be buffering required no matter what we
7557 * do, so go ahead and fallback to buffered.
7559 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7560 * to buffered IO. Don't blame me, this is the price we pay for using
7563 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7564 em->block_start == EXTENT_MAP_INLINE) {
7565 free_extent_map(em);
7570 len = min(len, em->len - (start - em->start));
7572 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7576 unlock_extents = true;
7577 /* Recalc len in case the new em is smaller than requested */
7578 len = min(len, em->len - (start - em->start));
7581 * We need to unlock only the end area that we aren't using.
7582 * The rest is going to be unlocked by the endio routine.
7584 lockstart = start + len;
7585 if (lockstart < lockend)
7586 unlock_extents = true;
7590 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7591 lockstart, lockend, &cached_state);
7593 free_extent_state(cached_state);
7596 * Translate extent map information to iomap.
7597 * We trim the extents (and move the addr) even though iomap code does
7598 * that, since we have locked only the parts we are performing I/O in.
7600 if ((em->block_start == EXTENT_MAP_HOLE) ||
7601 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7602 iomap->addr = IOMAP_NULL_ADDR;
7603 iomap->type = IOMAP_HOLE;
7605 iomap->addr = em->block_start + (start - em->start);
7606 iomap->type = IOMAP_MAPPED;
7608 iomap->offset = start;
7609 iomap->bdev = fs_info->fs_devices->latest_bdev;
7610 iomap->length = len;
7612 free_extent_map(em);
7617 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7621 btrfs_delalloc_release_space(BTRFS_I(inode),
7622 dio_data->data_reserved, start,
7623 dio_data->reserve, true);
7624 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7625 extent_changeset_free(dio_data->data_reserved);
7631 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7632 ssize_t written, unsigned int flags, struct iomap *iomap)
7635 struct btrfs_dio_data *dio_data = iomap->private;
7636 size_t submitted = dio_data->submitted;
7637 const bool write = !!(flags & IOMAP_WRITE);
7639 if (!write && (iomap->type == IOMAP_HOLE)) {
7640 /* If reading from a hole, unlock and return */
7641 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7645 if (submitted < length) {
7647 length -= submitted;
7649 __endio_write_update_ordered(BTRFS_I(inode), pos,
7652 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7658 if (dio_data->reserve)
7659 btrfs_delalloc_release_space(BTRFS_I(inode),
7660 dio_data->data_reserved, pos,
7661 dio_data->reserve, true);
7662 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7663 extent_changeset_free(dio_data->data_reserved);
7667 iomap->private = NULL;
7672 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7675 * This implies a barrier so that stores to dio_bio->bi_status before
7676 * this and loads of dio_bio->bi_status after this are fully ordered.
7678 if (!refcount_dec_and_test(&dip->refs))
7681 if (bio_op(dip->dio_bio) == REQ_OP_WRITE) {
7682 __endio_write_update_ordered(BTRFS_I(dip->inode),
7683 dip->logical_offset,
7685 !dip->dio_bio->bi_status);
7687 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7688 dip->logical_offset,
7689 dip->logical_offset + dip->bytes - 1);
7692 bio_endio(dip->dio_bio);
7696 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7698 unsigned long bio_flags)
7700 struct btrfs_dio_private *dip = bio->bi_private;
7701 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7704 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7706 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7710 refcount_inc(&dip->refs);
7711 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7713 refcount_dec(&dip->refs);
7717 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7718 struct btrfs_io_bio *io_bio,
7719 const bool uptodate)
7721 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7722 const u32 sectorsize = fs_info->sectorsize;
7723 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7724 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7725 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7726 struct bio_vec bvec;
7727 struct bvec_iter iter;
7728 u64 start = io_bio->logical;
7730 blk_status_t err = BLK_STS_OK;
7732 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7733 unsigned int i, nr_sectors, pgoff;
7735 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7736 pgoff = bvec.bv_offset;
7737 for (i = 0; i < nr_sectors; i++) {
7738 ASSERT(pgoff < PAGE_SIZE);
7740 (!csum || !check_data_csum(inode, io_bio, icsum,
7741 bvec.bv_page, pgoff))) {
7742 clean_io_failure(fs_info, failure_tree, io_tree,
7743 start, bvec.bv_page,
7744 btrfs_ino(BTRFS_I(inode)),
7747 blk_status_t status;
7749 status = btrfs_submit_read_repair(inode,
7751 start - io_bio->logical,
7752 bvec.bv_page, pgoff,
7754 start + sectorsize - 1,
7756 submit_dio_repair_bio);
7760 start += sectorsize;
7762 pgoff += sectorsize;
7768 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7769 const u64 offset, const u64 bytes,
7770 const bool uptodate)
7772 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7773 struct btrfs_ordered_extent *ordered = NULL;
7774 struct btrfs_workqueue *wq;
7775 u64 ordered_offset = offset;
7776 u64 ordered_bytes = bytes;
7779 if (btrfs_is_free_space_inode(inode))
7780 wq = fs_info->endio_freespace_worker;
7782 wq = fs_info->endio_write_workers;
7784 while (ordered_offset < offset + bytes) {
7785 last_offset = ordered_offset;
7786 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7790 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7792 btrfs_queue_work(wq, &ordered->work);
7795 * If btrfs_dec_test_ordered_pending does not find any ordered
7796 * extent in the range, we can exit.
7798 if (ordered_offset == last_offset)
7801 * Our bio might span multiple ordered extents. In this case
7802 * we keep going until we have accounted the whole dio.
7804 if (ordered_offset < offset + bytes) {
7805 ordered_bytes = offset + bytes - ordered_offset;
7811 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7812 struct bio *bio, u64 offset)
7814 return btrfs_csum_one_bio(BTRFS_I(inode), bio, offset, 1);
7817 static void btrfs_end_dio_bio(struct bio *bio)
7819 struct btrfs_dio_private *dip = bio->bi_private;
7820 blk_status_t err = bio->bi_status;
7823 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7824 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7825 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7827 (unsigned long long)bio->bi_iter.bi_sector,
7828 bio->bi_iter.bi_size, err);
7830 if (bio_op(bio) == REQ_OP_READ) {
7831 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
7836 dip->dio_bio->bi_status = err;
7839 btrfs_dio_private_put(dip);
7842 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7843 struct inode *inode, u64 file_offset, int async_submit)
7845 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7846 struct btrfs_dio_private *dip = bio->bi_private;
7847 bool write = bio_op(bio) == REQ_OP_WRITE;
7850 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7852 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7855 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7860 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7863 if (write && async_submit) {
7864 ret = btrfs_wq_submit_bio(inode, bio, 0, 0,
7866 btrfs_submit_bio_start_direct_io);
7870 * If we aren't doing async submit, calculate the csum of the
7873 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
7879 csum_offset = file_offset - dip->logical_offset;
7880 csum_offset >>= fs_info->sectorsize_bits;
7881 csum_offset *= fs_info->csum_size;
7882 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
7885 ret = btrfs_map_bio(fs_info, bio, 0);
7891 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7892 * or ordered extents whether or not we submit any bios.
7894 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7895 struct inode *inode,
7898 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7899 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7901 struct btrfs_dio_private *dip;
7903 dip_size = sizeof(*dip);
7904 if (!write && csum) {
7905 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7908 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
7909 dip_size += fs_info->csum_size * nblocks;
7912 dip = kzalloc(dip_size, GFP_NOFS);
7917 dip->logical_offset = file_offset;
7918 dip->bytes = dio_bio->bi_iter.bi_size;
7919 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7920 dip->dio_bio = dio_bio;
7921 refcount_set(&dip->refs, 1);
7925 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
7926 struct bio *dio_bio, loff_t file_offset)
7928 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7929 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7930 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7931 BTRFS_BLOCK_GROUP_RAID56_MASK);
7932 struct btrfs_dio_private *dip;
7935 int async_submit = 0;
7937 int clone_offset = 0;
7940 blk_status_t status;
7941 struct btrfs_io_geometry geom;
7942 struct btrfs_dio_data *dio_data = iomap->private;
7944 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7947 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7948 file_offset + dio_bio->bi_iter.bi_size - 1);
7950 dio_bio->bi_status = BLK_STS_RESOURCE;
7952 return BLK_QC_T_NONE;
7957 * Load the csums up front to reduce csum tree searches and
7958 * contention when submitting bios.
7960 * If we have csums disabled this will do nothing.
7962 status = btrfs_lookup_bio_sums(inode, dio_bio, file_offset,
7964 if (status != BLK_STS_OK)
7968 start_sector = dio_bio->bi_iter.bi_sector;
7969 submit_len = dio_bio->bi_iter.bi_size;
7972 ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio),
7973 start_sector << 9, submit_len,
7976 status = errno_to_blk_status(ret);
7979 ASSERT(geom.len <= INT_MAX);
7981 clone_len = min_t(int, submit_len, geom.len);
7984 * This will never fail as it's passing GPF_NOFS and
7985 * the allocation is backed by btrfs_bioset.
7987 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7988 bio->bi_private = dip;
7989 bio->bi_end_io = btrfs_end_dio_bio;
7990 btrfs_io_bio(bio)->logical = file_offset;
7992 ASSERT(submit_len >= clone_len);
7993 submit_len -= clone_len;
7996 * Increase the count before we submit the bio so we know
7997 * the end IO handler won't happen before we increase the
7998 * count. Otherwise, the dip might get freed before we're
7999 * done setting it up.
8001 * We transfer the initial reference to the last bio, so we
8002 * don't need to increment the reference count for the last one.
8004 if (submit_len > 0) {
8005 refcount_inc(&dip->refs);
8007 * If we are submitting more than one bio, submit them
8008 * all asynchronously. The exception is RAID 5 or 6, as
8009 * asynchronous checksums make it difficult to collect
8010 * full stripe writes.
8016 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8021 refcount_dec(&dip->refs);
8025 dio_data->submitted += clone_len;
8026 clone_offset += clone_len;
8027 start_sector += clone_len >> 9;
8028 file_offset += clone_len;
8029 } while (submit_len > 0);
8030 return BLK_QC_T_NONE;
8033 dip->dio_bio->bi_status = status;
8034 btrfs_dio_private_put(dip);
8035 return BLK_QC_T_NONE;
8038 const struct iomap_ops btrfs_dio_iomap_ops = {
8039 .iomap_begin = btrfs_dio_iomap_begin,
8040 .iomap_end = btrfs_dio_iomap_end,
8043 const struct iomap_dio_ops btrfs_dio_ops = {
8044 .submit_io = btrfs_submit_direct,
8047 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8052 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8056 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8059 int btrfs_readpage(struct file *file, struct page *page)
8061 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8062 u64 start = page_offset(page);
8063 u64 end = start + PAGE_SIZE - 1;
8064 unsigned long bio_flags = 0;
8065 struct bio *bio = NULL;
8068 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8070 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8072 ret = submit_one_bio(bio, 0, bio_flags);
8076 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8078 struct inode *inode = page->mapping->host;
8081 if (current->flags & PF_MEMALLOC) {
8082 redirty_page_for_writepage(wbc, page);
8088 * If we are under memory pressure we will call this directly from the
8089 * VM, we need to make sure we have the inode referenced for the ordered
8090 * extent. If not just return like we didn't do anything.
8092 if (!igrab(inode)) {
8093 redirty_page_for_writepage(wbc, page);
8094 return AOP_WRITEPAGE_ACTIVATE;
8096 ret = extent_write_full_page(page, wbc);
8097 btrfs_add_delayed_iput(inode);
8101 static int btrfs_writepages(struct address_space *mapping,
8102 struct writeback_control *wbc)
8104 return extent_writepages(mapping, wbc);
8107 static void btrfs_readahead(struct readahead_control *rac)
8109 extent_readahead(rac);
8112 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8114 int ret = try_release_extent_mapping(page, gfp_flags);
8116 detach_page_private(page);
8120 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8122 if (PageWriteback(page) || PageDirty(page))
8124 return __btrfs_releasepage(page, gfp_flags);
8127 #ifdef CONFIG_MIGRATION
8128 static int btrfs_migratepage(struct address_space *mapping,
8129 struct page *newpage, struct page *page,
8130 enum migrate_mode mode)
8134 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8135 if (ret != MIGRATEPAGE_SUCCESS)
8138 if (page_has_private(page))
8139 attach_page_private(newpage, detach_page_private(page));
8141 if (PagePrivate2(page)) {
8142 ClearPagePrivate2(page);
8143 SetPagePrivate2(newpage);
8146 if (mode != MIGRATE_SYNC_NO_COPY)
8147 migrate_page_copy(newpage, page);
8149 migrate_page_states(newpage, page);
8150 return MIGRATEPAGE_SUCCESS;
8154 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8155 unsigned int length)
8157 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8158 struct extent_io_tree *tree = &inode->io_tree;
8159 struct btrfs_ordered_extent *ordered;
8160 struct extent_state *cached_state = NULL;
8161 u64 page_start = page_offset(page);
8162 u64 page_end = page_start + PAGE_SIZE - 1;
8165 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8166 bool found_ordered = false;
8167 bool completed_ordered = false;
8170 * we have the page locked, so new writeback can't start,
8171 * and the dirty bit won't be cleared while we are here.
8173 * Wait for IO on this page so that we can safely clear
8174 * the PagePrivate2 bit and do ordered accounting
8176 wait_on_page_writeback(page);
8179 btrfs_releasepage(page, GFP_NOFS);
8183 if (!inode_evicting)
8184 lock_extent_bits(tree, page_start, page_end, &cached_state);
8187 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8189 found_ordered = true;
8191 ordered->file_offset + ordered->num_bytes - 1);
8193 * IO on this page will never be started, so we need to account
8194 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8195 * here, must leave that up for the ordered extent completion.
8197 if (!inode_evicting)
8198 clear_extent_bit(tree, start, end,
8200 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8201 EXTENT_DEFRAG, 1, 0, &cached_state);
8203 * whoever cleared the private bit is responsible
8204 * for the finish_ordered_io
8206 if (TestClearPagePrivate2(page)) {
8207 struct btrfs_ordered_inode_tree *tree;
8210 tree = &inode->ordered_tree;
8212 spin_lock_irq(&tree->lock);
8213 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8214 new_len = start - ordered->file_offset;
8215 if (new_len < ordered->truncated_len)
8216 ordered->truncated_len = new_len;
8217 spin_unlock_irq(&tree->lock);
8219 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8221 end - start + 1, 1)) {
8222 btrfs_finish_ordered_io(ordered);
8223 completed_ordered = true;
8226 btrfs_put_ordered_extent(ordered);
8227 if (!inode_evicting) {
8228 cached_state = NULL;
8229 lock_extent_bits(tree, start, end,
8234 if (start < page_end)
8239 * Qgroup reserved space handler
8240 * Page here will be either
8241 * 1) Already written to disk or ordered extent already submitted
8242 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8243 * Qgroup will be handled by its qgroup_record then.
8244 * btrfs_qgroup_free_data() call will do nothing here.
8246 * 2) Not written to disk yet
8247 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8248 * bit of its io_tree, and free the qgroup reserved data space.
8249 * Since the IO will never happen for this page.
8251 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8252 if (!inode_evicting) {
8256 * If there's an ordered extent for this range and we have not
8257 * finished it ourselves, we must leave EXTENT_DELALLOC_NEW set
8258 * in the range for the ordered extent completion. We must also
8259 * not delete the range, otherwise we would lose that bit (and
8260 * any other bits set in the range). Make sure EXTENT_UPTODATE
8261 * is cleared if we don't delete, otherwise it can lead to
8262 * corruptions if the i_size is extented later.
8264 if (found_ordered && !completed_ordered)
8266 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8267 EXTENT_DELALLOC | EXTENT_UPTODATE |
8268 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8269 delete, &cached_state);
8271 __btrfs_releasepage(page, GFP_NOFS);
8274 ClearPageChecked(page);
8275 detach_page_private(page);
8279 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8280 * called from a page fault handler when a page is first dirtied. Hence we must
8281 * be careful to check for EOF conditions here. We set the page up correctly
8282 * for a written page which means we get ENOSPC checking when writing into
8283 * holes and correct delalloc and unwritten extent mapping on filesystems that
8284 * support these features.
8286 * We are not allowed to take the i_mutex here so we have to play games to
8287 * protect against truncate races as the page could now be beyond EOF. Because
8288 * truncate_setsize() writes the inode size before removing pages, once we have
8289 * the page lock we can determine safely if the page is beyond EOF. If it is not
8290 * beyond EOF, then the page is guaranteed safe against truncation until we
8293 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8295 struct page *page = vmf->page;
8296 struct inode *inode = file_inode(vmf->vma->vm_file);
8297 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8298 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8299 struct btrfs_ordered_extent *ordered;
8300 struct extent_state *cached_state = NULL;
8301 struct extent_changeset *data_reserved = NULL;
8303 unsigned long zero_start;
8313 reserved_space = PAGE_SIZE;
8315 sb_start_pagefault(inode->i_sb);
8316 page_start = page_offset(page);
8317 page_end = page_start + PAGE_SIZE - 1;
8321 * Reserving delalloc space after obtaining the page lock can lead to
8322 * deadlock. For example, if a dirty page is locked by this function
8323 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8324 * dirty page write out, then the btrfs_writepage() function could
8325 * end up waiting indefinitely to get a lock on the page currently
8326 * being processed by btrfs_page_mkwrite() function.
8328 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8329 page_start, reserved_space);
8331 ret2 = file_update_time(vmf->vma->vm_file);
8335 ret = vmf_error(ret2);
8341 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8344 size = i_size_read(inode);
8346 if ((page->mapping != inode->i_mapping) ||
8347 (page_start >= size)) {
8348 /* page got truncated out from underneath us */
8351 wait_on_page_writeback(page);
8353 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8354 set_page_extent_mapped(page);
8357 * we can't set the delalloc bits if there are pending ordered
8358 * extents. Drop our locks and wait for them to finish
8360 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8363 unlock_extent_cached(io_tree, page_start, page_end,
8366 btrfs_start_ordered_extent(ordered, 1);
8367 btrfs_put_ordered_extent(ordered);
8371 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8372 reserved_space = round_up(size - page_start,
8373 fs_info->sectorsize);
8374 if (reserved_space < PAGE_SIZE) {
8375 end = page_start + reserved_space - 1;
8376 btrfs_delalloc_release_space(BTRFS_I(inode),
8377 data_reserved, page_start,
8378 PAGE_SIZE - reserved_space, true);
8383 * page_mkwrite gets called when the page is firstly dirtied after it's
8384 * faulted in, but write(2) could also dirty a page and set delalloc
8385 * bits, thus in this case for space account reason, we still need to
8386 * clear any delalloc bits within this page range since we have to
8387 * reserve data&meta space before lock_page() (see above comments).
8389 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8390 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8391 EXTENT_DEFRAG, 0, 0, &cached_state);
8393 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8396 unlock_extent_cached(io_tree, page_start, page_end,
8398 ret = VM_FAULT_SIGBUS;
8402 /* page is wholly or partially inside EOF */
8403 if (page_start + PAGE_SIZE > size)
8404 zero_start = offset_in_page(size);
8406 zero_start = PAGE_SIZE;
8408 if (zero_start != PAGE_SIZE) {
8410 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8411 flush_dcache_page(page);
8414 ClearPageChecked(page);
8415 set_page_dirty(page);
8416 SetPageUptodate(page);
8418 BTRFS_I(inode)->last_trans = fs_info->generation;
8419 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8420 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8422 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8424 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8425 sb_end_pagefault(inode->i_sb);
8426 extent_changeset_free(data_reserved);
8427 return VM_FAULT_LOCKED;
8432 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8433 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8434 reserved_space, (ret != 0));
8436 sb_end_pagefault(inode->i_sb);
8437 extent_changeset_free(data_reserved);
8441 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8443 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8444 struct btrfs_root *root = BTRFS_I(inode)->root;
8445 struct btrfs_block_rsv *rsv;
8447 struct btrfs_trans_handle *trans;
8448 u64 mask = fs_info->sectorsize - 1;
8449 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8451 if (!skip_writeback) {
8452 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8459 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8460 * things going on here:
8462 * 1) We need to reserve space to update our inode.
8464 * 2) We need to have something to cache all the space that is going to
8465 * be free'd up by the truncate operation, but also have some slack
8466 * space reserved in case it uses space during the truncate (thank you
8467 * very much snapshotting).
8469 * And we need these to be separate. The fact is we can use a lot of
8470 * space doing the truncate, and we have no earthly idea how much space
8471 * we will use, so we need the truncate reservation to be separate so it
8472 * doesn't end up using space reserved for updating the inode. We also
8473 * need to be able to stop the transaction and start a new one, which
8474 * means we need to be able to update the inode several times, and we
8475 * have no idea of knowing how many times that will be, so we can't just
8476 * reserve 1 item for the entirety of the operation, so that has to be
8477 * done separately as well.
8479 * So that leaves us with
8481 * 1) rsv - for the truncate reservation, which we will steal from the
8482 * transaction reservation.
8483 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8484 * updating the inode.
8486 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8489 rsv->size = min_size;
8493 * 1 for the truncate slack space
8494 * 1 for updating the inode.
8496 trans = btrfs_start_transaction(root, 2);
8497 if (IS_ERR(trans)) {
8498 ret = PTR_ERR(trans);
8502 /* Migrate the slack space for the truncate to our reserve */
8503 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8508 * So if we truncate and then write and fsync we normally would just
8509 * write the extents that changed, which is a problem if we need to
8510 * first truncate that entire inode. So set this flag so we write out
8511 * all of the extents in the inode to the sync log so we're completely
8514 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8515 trans->block_rsv = rsv;
8518 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8520 BTRFS_EXTENT_DATA_KEY);
8521 trans->block_rsv = &fs_info->trans_block_rsv;
8522 if (ret != -ENOSPC && ret != -EAGAIN)
8525 ret = btrfs_update_inode(trans, root, inode);
8529 btrfs_end_transaction(trans);
8530 btrfs_btree_balance_dirty(fs_info);
8532 trans = btrfs_start_transaction(root, 2);
8533 if (IS_ERR(trans)) {
8534 ret = PTR_ERR(trans);
8539 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8540 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8541 rsv, min_size, false);
8542 BUG_ON(ret); /* shouldn't happen */
8543 trans->block_rsv = rsv;
8547 * We can't call btrfs_truncate_block inside a trans handle as we could
8548 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8549 * we've truncated everything except the last little bit, and can do
8550 * btrfs_truncate_block and then update the disk_i_size.
8552 if (ret == NEED_TRUNCATE_BLOCK) {
8553 btrfs_end_transaction(trans);
8554 btrfs_btree_balance_dirty(fs_info);
8556 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
8559 trans = btrfs_start_transaction(root, 1);
8560 if (IS_ERR(trans)) {
8561 ret = PTR_ERR(trans);
8564 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8570 trans->block_rsv = &fs_info->trans_block_rsv;
8571 ret2 = btrfs_update_inode(trans, root, inode);
8575 ret2 = btrfs_end_transaction(trans);
8578 btrfs_btree_balance_dirty(fs_info);
8581 btrfs_free_block_rsv(fs_info, rsv);
8587 * create a new subvolume directory/inode (helper for the ioctl).
8589 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8590 struct btrfs_root *new_root,
8591 struct btrfs_root *parent_root,
8594 struct inode *inode;
8598 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8599 new_dirid, new_dirid,
8600 S_IFDIR | (~current_umask() & S_IRWXUGO),
8603 return PTR_ERR(inode);
8604 inode->i_op = &btrfs_dir_inode_operations;
8605 inode->i_fop = &btrfs_dir_file_operations;
8607 set_nlink(inode, 1);
8608 btrfs_i_size_write(BTRFS_I(inode), 0);
8609 unlock_new_inode(inode);
8611 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8613 btrfs_err(new_root->fs_info,
8614 "error inheriting subvolume %llu properties: %d",
8615 new_root->root_key.objectid, err);
8617 err = btrfs_update_inode(trans, new_root, inode);
8623 struct inode *btrfs_alloc_inode(struct super_block *sb)
8625 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8626 struct btrfs_inode *ei;
8627 struct inode *inode;
8629 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8636 ei->last_sub_trans = 0;
8637 ei->logged_trans = 0;
8638 ei->delalloc_bytes = 0;
8639 ei->new_delalloc_bytes = 0;
8640 ei->defrag_bytes = 0;
8641 ei->disk_i_size = 0;
8644 ei->index_cnt = (u64)-1;
8646 ei->last_unlink_trans = 0;
8647 ei->last_reflink_trans = 0;
8648 ei->last_log_commit = 0;
8650 spin_lock_init(&ei->lock);
8651 ei->outstanding_extents = 0;
8652 if (sb->s_magic != BTRFS_TEST_MAGIC)
8653 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8654 BTRFS_BLOCK_RSV_DELALLOC);
8655 ei->runtime_flags = 0;
8656 ei->prop_compress = BTRFS_COMPRESS_NONE;
8657 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8659 ei->delayed_node = NULL;
8661 ei->i_otime.tv_sec = 0;
8662 ei->i_otime.tv_nsec = 0;
8664 inode = &ei->vfs_inode;
8665 extent_map_tree_init(&ei->extent_tree);
8666 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8667 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8668 IO_TREE_INODE_IO_FAILURE, inode);
8669 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8670 IO_TREE_INODE_FILE_EXTENT, inode);
8671 ei->io_tree.track_uptodate = true;
8672 ei->io_failure_tree.track_uptodate = true;
8673 atomic_set(&ei->sync_writers, 0);
8674 mutex_init(&ei->log_mutex);
8675 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8676 INIT_LIST_HEAD(&ei->delalloc_inodes);
8677 INIT_LIST_HEAD(&ei->delayed_iput);
8678 RB_CLEAR_NODE(&ei->rb_node);
8683 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8684 void btrfs_test_destroy_inode(struct inode *inode)
8686 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8687 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8691 void btrfs_free_inode(struct inode *inode)
8693 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8696 void btrfs_destroy_inode(struct inode *vfs_inode)
8698 struct btrfs_ordered_extent *ordered;
8699 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8700 struct btrfs_root *root = inode->root;
8702 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8703 WARN_ON(vfs_inode->i_data.nrpages);
8704 WARN_ON(inode->block_rsv.reserved);
8705 WARN_ON(inode->block_rsv.size);
8706 WARN_ON(inode->outstanding_extents);
8707 WARN_ON(inode->delalloc_bytes);
8708 WARN_ON(inode->new_delalloc_bytes);
8709 WARN_ON(inode->csum_bytes);
8710 WARN_ON(inode->defrag_bytes);
8713 * This can happen where we create an inode, but somebody else also
8714 * created the same inode and we need to destroy the one we already
8721 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8725 btrfs_err(root->fs_info,
8726 "found ordered extent %llu %llu on inode cleanup",
8727 ordered->file_offset, ordered->num_bytes);
8728 btrfs_remove_ordered_extent(inode, ordered);
8729 btrfs_put_ordered_extent(ordered);
8730 btrfs_put_ordered_extent(ordered);
8733 btrfs_qgroup_check_reserved_leak(inode);
8734 inode_tree_del(inode);
8735 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8736 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8737 btrfs_put_root(inode->root);
8740 int btrfs_drop_inode(struct inode *inode)
8742 struct btrfs_root *root = BTRFS_I(inode)->root;
8747 /* the snap/subvol tree is on deleting */
8748 if (btrfs_root_refs(&root->root_item) == 0)
8751 return generic_drop_inode(inode);
8754 static void init_once(void *foo)
8756 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8758 inode_init_once(&ei->vfs_inode);
8761 void __cold btrfs_destroy_cachep(void)
8764 * Make sure all delayed rcu free inodes are flushed before we
8768 kmem_cache_destroy(btrfs_inode_cachep);
8769 kmem_cache_destroy(btrfs_trans_handle_cachep);
8770 kmem_cache_destroy(btrfs_path_cachep);
8771 kmem_cache_destroy(btrfs_free_space_cachep);
8772 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8775 int __init btrfs_init_cachep(void)
8777 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8778 sizeof(struct btrfs_inode), 0,
8779 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8781 if (!btrfs_inode_cachep)
8784 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8785 sizeof(struct btrfs_trans_handle), 0,
8786 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8787 if (!btrfs_trans_handle_cachep)
8790 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8791 sizeof(struct btrfs_path), 0,
8792 SLAB_MEM_SPREAD, NULL);
8793 if (!btrfs_path_cachep)
8796 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8797 sizeof(struct btrfs_free_space), 0,
8798 SLAB_MEM_SPREAD, NULL);
8799 if (!btrfs_free_space_cachep)
8802 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8803 PAGE_SIZE, PAGE_SIZE,
8804 SLAB_RED_ZONE, NULL);
8805 if (!btrfs_free_space_bitmap_cachep)
8810 btrfs_destroy_cachep();
8814 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8815 u32 request_mask, unsigned int flags)
8819 struct inode *inode = d_inode(path->dentry);
8820 u32 blocksize = inode->i_sb->s_blocksize;
8821 u32 bi_flags = BTRFS_I(inode)->flags;
8823 stat->result_mask |= STATX_BTIME;
8824 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8825 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8826 if (bi_flags & BTRFS_INODE_APPEND)
8827 stat->attributes |= STATX_ATTR_APPEND;
8828 if (bi_flags & BTRFS_INODE_COMPRESS)
8829 stat->attributes |= STATX_ATTR_COMPRESSED;
8830 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8831 stat->attributes |= STATX_ATTR_IMMUTABLE;
8832 if (bi_flags & BTRFS_INODE_NODUMP)
8833 stat->attributes |= STATX_ATTR_NODUMP;
8835 stat->attributes_mask |= (STATX_ATTR_APPEND |
8836 STATX_ATTR_COMPRESSED |
8837 STATX_ATTR_IMMUTABLE |
8840 generic_fillattr(inode, stat);
8841 stat->dev = BTRFS_I(inode)->root->anon_dev;
8843 spin_lock(&BTRFS_I(inode)->lock);
8844 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8845 inode_bytes = inode_get_bytes(inode);
8846 spin_unlock(&BTRFS_I(inode)->lock);
8847 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8848 ALIGN(delalloc_bytes, blocksize)) >> 9;
8852 static int btrfs_rename_exchange(struct inode *old_dir,
8853 struct dentry *old_dentry,
8854 struct inode *new_dir,
8855 struct dentry *new_dentry)
8857 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8858 struct btrfs_trans_handle *trans;
8859 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8860 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8861 struct inode *new_inode = new_dentry->d_inode;
8862 struct inode *old_inode = old_dentry->d_inode;
8863 struct timespec64 ctime = current_time(old_inode);
8864 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8865 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8870 bool root_log_pinned = false;
8871 bool dest_log_pinned = false;
8873 /* we only allow rename subvolume link between subvolumes */
8874 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8877 /* close the race window with snapshot create/destroy ioctl */
8878 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8879 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8880 down_read(&fs_info->subvol_sem);
8883 * We want to reserve the absolute worst case amount of items. So if
8884 * both inodes are subvols and we need to unlink them then that would
8885 * require 4 item modifications, but if they are both normal inodes it
8886 * would require 5 item modifications, so we'll assume their normal
8887 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
8888 * should cover the worst case number of items we'll modify.
8890 trans = btrfs_start_transaction(root, 12);
8891 if (IS_ERR(trans)) {
8892 ret = PTR_ERR(trans);
8897 btrfs_record_root_in_trans(trans, dest);
8900 * We need to find a free sequence number both in the source and
8901 * in the destination directory for the exchange.
8903 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8906 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8910 BTRFS_I(old_inode)->dir_index = 0ULL;
8911 BTRFS_I(new_inode)->dir_index = 0ULL;
8913 /* Reference for the source. */
8914 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8915 /* force full log commit if subvolume involved. */
8916 btrfs_set_log_full_commit(trans);
8918 btrfs_pin_log_trans(root);
8919 root_log_pinned = true;
8920 ret = btrfs_insert_inode_ref(trans, dest,
8921 new_dentry->d_name.name,
8922 new_dentry->d_name.len,
8924 btrfs_ino(BTRFS_I(new_dir)),
8930 /* And now for the dest. */
8931 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8932 /* force full log commit if subvolume involved. */
8933 btrfs_set_log_full_commit(trans);
8935 btrfs_pin_log_trans(dest);
8936 dest_log_pinned = true;
8937 ret = btrfs_insert_inode_ref(trans, root,
8938 old_dentry->d_name.name,
8939 old_dentry->d_name.len,
8941 btrfs_ino(BTRFS_I(old_dir)),
8947 /* Update inode version and ctime/mtime. */
8948 inode_inc_iversion(old_dir);
8949 inode_inc_iversion(new_dir);
8950 inode_inc_iversion(old_inode);
8951 inode_inc_iversion(new_inode);
8952 old_dir->i_ctime = old_dir->i_mtime = ctime;
8953 new_dir->i_ctime = new_dir->i_mtime = ctime;
8954 old_inode->i_ctime = ctime;
8955 new_inode->i_ctime = ctime;
8957 if (old_dentry->d_parent != new_dentry->d_parent) {
8958 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8959 BTRFS_I(old_inode), 1);
8960 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8961 BTRFS_I(new_inode), 1);
8964 /* src is a subvolume */
8965 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8966 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
8967 } else { /* src is an inode */
8968 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
8969 BTRFS_I(old_dentry->d_inode),
8970 old_dentry->d_name.name,
8971 old_dentry->d_name.len);
8973 ret = btrfs_update_inode(trans, root, old_inode);
8976 btrfs_abort_transaction(trans, ret);
8980 /* dest is a subvolume */
8981 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8982 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
8983 } else { /* dest is an inode */
8984 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
8985 BTRFS_I(new_dentry->d_inode),
8986 new_dentry->d_name.name,
8987 new_dentry->d_name.len);
8989 ret = btrfs_update_inode(trans, dest, new_inode);
8992 btrfs_abort_transaction(trans, ret);
8996 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8997 new_dentry->d_name.name,
8998 new_dentry->d_name.len, 0, old_idx);
9000 btrfs_abort_transaction(trans, ret);
9004 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9005 old_dentry->d_name.name,
9006 old_dentry->d_name.len, 0, new_idx);
9008 btrfs_abort_transaction(trans, ret);
9012 if (old_inode->i_nlink == 1)
9013 BTRFS_I(old_inode)->dir_index = old_idx;
9014 if (new_inode->i_nlink == 1)
9015 BTRFS_I(new_inode)->dir_index = new_idx;
9017 if (root_log_pinned) {
9018 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9019 new_dentry->d_parent);
9020 btrfs_end_log_trans(root);
9021 root_log_pinned = false;
9023 if (dest_log_pinned) {
9024 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9025 old_dentry->d_parent);
9026 btrfs_end_log_trans(dest);
9027 dest_log_pinned = false;
9031 * If we have pinned a log and an error happened, we unpin tasks
9032 * trying to sync the log and force them to fallback to a transaction
9033 * commit if the log currently contains any of the inodes involved in
9034 * this rename operation (to ensure we do not persist a log with an
9035 * inconsistent state for any of these inodes or leading to any
9036 * inconsistencies when replayed). If the transaction was aborted, the
9037 * abortion reason is propagated to userspace when attempting to commit
9038 * the transaction. If the log does not contain any of these inodes, we
9039 * allow the tasks to sync it.
9041 if (ret && (root_log_pinned || dest_log_pinned)) {
9042 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9043 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9044 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9046 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9047 btrfs_set_log_full_commit(trans);
9049 if (root_log_pinned) {
9050 btrfs_end_log_trans(root);
9051 root_log_pinned = false;
9053 if (dest_log_pinned) {
9054 btrfs_end_log_trans(dest);
9055 dest_log_pinned = false;
9058 ret2 = btrfs_end_transaction(trans);
9059 ret = ret ? ret : ret2;
9061 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9062 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9063 up_read(&fs_info->subvol_sem);
9068 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9069 struct btrfs_root *root,
9071 struct dentry *dentry)
9074 struct inode *inode;
9078 ret = btrfs_find_free_ino(root, &objectid);
9082 inode = btrfs_new_inode(trans, root, dir,
9083 dentry->d_name.name,
9085 btrfs_ino(BTRFS_I(dir)),
9087 S_IFCHR | WHITEOUT_MODE,
9090 if (IS_ERR(inode)) {
9091 ret = PTR_ERR(inode);
9095 inode->i_op = &btrfs_special_inode_operations;
9096 init_special_inode(inode, inode->i_mode,
9099 ret = btrfs_init_inode_security(trans, inode, dir,
9104 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9105 BTRFS_I(inode), 0, index);
9109 ret = btrfs_update_inode(trans, root, inode);
9111 unlock_new_inode(inode);
9113 inode_dec_link_count(inode);
9119 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9120 struct inode *new_dir, struct dentry *new_dentry,
9123 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9124 struct btrfs_trans_handle *trans;
9125 unsigned int trans_num_items;
9126 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9127 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9128 struct inode *new_inode = d_inode(new_dentry);
9129 struct inode *old_inode = d_inode(old_dentry);
9133 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9134 bool log_pinned = false;
9136 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9139 /* we only allow rename subvolume link between subvolumes */
9140 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9143 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9144 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9147 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9148 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9152 /* check for collisions, even if the name isn't there */
9153 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9154 new_dentry->d_name.name,
9155 new_dentry->d_name.len);
9158 if (ret == -EEXIST) {
9160 * eexist without a new_inode */
9161 if (WARN_ON(!new_inode)) {
9165 /* maybe -EOVERFLOW */
9172 * we're using rename to replace one file with another. Start IO on it
9173 * now so we don't add too much work to the end of the transaction
9175 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9176 filemap_flush(old_inode->i_mapping);
9178 /* close the racy window with snapshot create/destroy ioctl */
9179 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9180 down_read(&fs_info->subvol_sem);
9182 * We want to reserve the absolute worst case amount of items. So if
9183 * both inodes are subvols and we need to unlink them then that would
9184 * require 4 item modifications, but if they are both normal inodes it
9185 * would require 5 item modifications, so we'll assume they are normal
9186 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9187 * should cover the worst case number of items we'll modify.
9188 * If our rename has the whiteout flag, we need more 5 units for the
9189 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9190 * when selinux is enabled).
9192 trans_num_items = 11;
9193 if (flags & RENAME_WHITEOUT)
9194 trans_num_items += 5;
9195 trans = btrfs_start_transaction(root, trans_num_items);
9196 if (IS_ERR(trans)) {
9197 ret = PTR_ERR(trans);
9202 btrfs_record_root_in_trans(trans, dest);
9204 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9208 BTRFS_I(old_inode)->dir_index = 0ULL;
9209 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9210 /* force full log commit if subvolume involved. */
9211 btrfs_set_log_full_commit(trans);
9213 btrfs_pin_log_trans(root);
9215 ret = btrfs_insert_inode_ref(trans, dest,
9216 new_dentry->d_name.name,
9217 new_dentry->d_name.len,
9219 btrfs_ino(BTRFS_I(new_dir)), index);
9224 inode_inc_iversion(old_dir);
9225 inode_inc_iversion(new_dir);
9226 inode_inc_iversion(old_inode);
9227 old_dir->i_ctime = old_dir->i_mtime =
9228 new_dir->i_ctime = new_dir->i_mtime =
9229 old_inode->i_ctime = current_time(old_dir);
9231 if (old_dentry->d_parent != new_dentry->d_parent)
9232 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9233 BTRFS_I(old_inode), 1);
9235 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9236 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9238 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9239 BTRFS_I(d_inode(old_dentry)),
9240 old_dentry->d_name.name,
9241 old_dentry->d_name.len);
9243 ret = btrfs_update_inode(trans, root, old_inode);
9246 btrfs_abort_transaction(trans, ret);
9251 inode_inc_iversion(new_inode);
9252 new_inode->i_ctime = current_time(new_inode);
9253 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9254 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9255 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9256 BUG_ON(new_inode->i_nlink == 0);
9258 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9259 BTRFS_I(d_inode(new_dentry)),
9260 new_dentry->d_name.name,
9261 new_dentry->d_name.len);
9263 if (!ret && new_inode->i_nlink == 0)
9264 ret = btrfs_orphan_add(trans,
9265 BTRFS_I(d_inode(new_dentry)));
9267 btrfs_abort_transaction(trans, ret);
9272 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9273 new_dentry->d_name.name,
9274 new_dentry->d_name.len, 0, index);
9276 btrfs_abort_transaction(trans, ret);
9280 if (old_inode->i_nlink == 1)
9281 BTRFS_I(old_inode)->dir_index = index;
9284 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9285 new_dentry->d_parent);
9286 btrfs_end_log_trans(root);
9290 if (flags & RENAME_WHITEOUT) {
9291 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9295 btrfs_abort_transaction(trans, ret);
9301 * If we have pinned the log and an error happened, we unpin tasks
9302 * trying to sync the log and force them to fallback to a transaction
9303 * commit if the log currently contains any of the inodes involved in
9304 * this rename operation (to ensure we do not persist a log with an
9305 * inconsistent state for any of these inodes or leading to any
9306 * inconsistencies when replayed). If the transaction was aborted, the
9307 * abortion reason is propagated to userspace when attempting to commit
9308 * the transaction. If the log does not contain any of these inodes, we
9309 * allow the tasks to sync it.
9311 if (ret && log_pinned) {
9312 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9313 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9314 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9316 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9317 btrfs_set_log_full_commit(trans);
9319 btrfs_end_log_trans(root);
9322 ret2 = btrfs_end_transaction(trans);
9323 ret = ret ? ret : ret2;
9325 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9326 up_read(&fs_info->subvol_sem);
9331 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9332 struct inode *new_dir, struct dentry *new_dentry,
9335 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9338 if (flags & RENAME_EXCHANGE)
9339 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9342 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9345 struct btrfs_delalloc_work {
9346 struct inode *inode;
9347 struct completion completion;
9348 struct list_head list;
9349 struct btrfs_work work;
9352 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9354 struct btrfs_delalloc_work *delalloc_work;
9355 struct inode *inode;
9357 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9359 inode = delalloc_work->inode;
9360 filemap_flush(inode->i_mapping);
9361 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9362 &BTRFS_I(inode)->runtime_flags))
9363 filemap_flush(inode->i_mapping);
9366 complete(&delalloc_work->completion);
9369 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9371 struct btrfs_delalloc_work *work;
9373 work = kmalloc(sizeof(*work), GFP_NOFS);
9377 init_completion(&work->completion);
9378 INIT_LIST_HEAD(&work->list);
9379 work->inode = inode;
9380 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9386 * some fairly slow code that needs optimization. This walks the list
9387 * of all the inodes with pending delalloc and forces them to disk.
9389 static int start_delalloc_inodes(struct btrfs_root *root, u64 *nr, bool snapshot)
9391 struct btrfs_inode *binode;
9392 struct inode *inode;
9393 struct btrfs_delalloc_work *work, *next;
9394 struct list_head works;
9395 struct list_head splice;
9398 INIT_LIST_HEAD(&works);
9399 INIT_LIST_HEAD(&splice);
9401 mutex_lock(&root->delalloc_mutex);
9402 spin_lock(&root->delalloc_lock);
9403 list_splice_init(&root->delalloc_inodes, &splice);
9404 while (!list_empty(&splice)) {
9405 binode = list_entry(splice.next, struct btrfs_inode,
9408 list_move_tail(&binode->delalloc_inodes,
9409 &root->delalloc_inodes);
9410 inode = igrab(&binode->vfs_inode);
9412 cond_resched_lock(&root->delalloc_lock);
9415 spin_unlock(&root->delalloc_lock);
9418 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9419 &binode->runtime_flags);
9420 work = btrfs_alloc_delalloc_work(inode);
9426 list_add_tail(&work->list, &works);
9427 btrfs_queue_work(root->fs_info->flush_workers,
9429 if (*nr != U64_MAX) {
9435 spin_lock(&root->delalloc_lock);
9437 spin_unlock(&root->delalloc_lock);
9440 list_for_each_entry_safe(work, next, &works, list) {
9441 list_del_init(&work->list);
9442 wait_for_completion(&work->completion);
9446 if (!list_empty(&splice)) {
9447 spin_lock(&root->delalloc_lock);
9448 list_splice_tail(&splice, &root->delalloc_inodes);
9449 spin_unlock(&root->delalloc_lock);
9451 mutex_unlock(&root->delalloc_mutex);
9455 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9457 struct btrfs_fs_info *fs_info = root->fs_info;
9460 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9463 return start_delalloc_inodes(root, &nr, true);
9466 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, u64 nr)
9468 struct btrfs_root *root;
9469 struct list_head splice;
9472 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9475 INIT_LIST_HEAD(&splice);
9477 mutex_lock(&fs_info->delalloc_root_mutex);
9478 spin_lock(&fs_info->delalloc_root_lock);
9479 list_splice_init(&fs_info->delalloc_roots, &splice);
9480 while (!list_empty(&splice) && nr) {
9481 root = list_first_entry(&splice, struct btrfs_root,
9483 root = btrfs_grab_root(root);
9485 list_move_tail(&root->delalloc_root,
9486 &fs_info->delalloc_roots);
9487 spin_unlock(&fs_info->delalloc_root_lock);
9489 ret = start_delalloc_inodes(root, &nr, false);
9490 btrfs_put_root(root);
9493 spin_lock(&fs_info->delalloc_root_lock);
9495 spin_unlock(&fs_info->delalloc_root_lock);
9499 if (!list_empty(&splice)) {
9500 spin_lock(&fs_info->delalloc_root_lock);
9501 list_splice_tail(&splice, &fs_info->delalloc_roots);
9502 spin_unlock(&fs_info->delalloc_root_lock);
9504 mutex_unlock(&fs_info->delalloc_root_mutex);
9508 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9509 const char *symname)
9511 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9512 struct btrfs_trans_handle *trans;
9513 struct btrfs_root *root = BTRFS_I(dir)->root;
9514 struct btrfs_path *path;
9515 struct btrfs_key key;
9516 struct inode *inode = NULL;
9523 struct btrfs_file_extent_item *ei;
9524 struct extent_buffer *leaf;
9526 name_len = strlen(symname);
9527 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9528 return -ENAMETOOLONG;
9531 * 2 items for inode item and ref
9532 * 2 items for dir items
9533 * 1 item for updating parent inode item
9534 * 1 item for the inline extent item
9535 * 1 item for xattr if selinux is on
9537 trans = btrfs_start_transaction(root, 7);
9539 return PTR_ERR(trans);
9541 err = btrfs_find_free_ino(root, &objectid);
9545 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9546 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9547 objectid, S_IFLNK|S_IRWXUGO, &index);
9548 if (IS_ERR(inode)) {
9549 err = PTR_ERR(inode);
9555 * If the active LSM wants to access the inode during
9556 * d_instantiate it needs these. Smack checks to see
9557 * if the filesystem supports xattrs by looking at the
9560 inode->i_fop = &btrfs_file_operations;
9561 inode->i_op = &btrfs_file_inode_operations;
9562 inode->i_mapping->a_ops = &btrfs_aops;
9564 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9568 path = btrfs_alloc_path();
9573 key.objectid = btrfs_ino(BTRFS_I(inode));
9575 key.type = BTRFS_EXTENT_DATA_KEY;
9576 datasize = btrfs_file_extent_calc_inline_size(name_len);
9577 err = btrfs_insert_empty_item(trans, root, path, &key,
9580 btrfs_free_path(path);
9583 leaf = path->nodes[0];
9584 ei = btrfs_item_ptr(leaf, path->slots[0],
9585 struct btrfs_file_extent_item);
9586 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9587 btrfs_set_file_extent_type(leaf, ei,
9588 BTRFS_FILE_EXTENT_INLINE);
9589 btrfs_set_file_extent_encryption(leaf, ei, 0);
9590 btrfs_set_file_extent_compression(leaf, ei, 0);
9591 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9592 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9594 ptr = btrfs_file_extent_inline_start(ei);
9595 write_extent_buffer(leaf, symname, ptr, name_len);
9596 btrfs_mark_buffer_dirty(leaf);
9597 btrfs_free_path(path);
9599 inode->i_op = &btrfs_symlink_inode_operations;
9600 inode_nohighmem(inode);
9601 inode_set_bytes(inode, name_len);
9602 btrfs_i_size_write(BTRFS_I(inode), name_len);
9603 err = btrfs_update_inode(trans, root, inode);
9605 * Last step, add directory indexes for our symlink inode. This is the
9606 * last step to avoid extra cleanup of these indexes if an error happens
9610 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9611 BTRFS_I(inode), 0, index);
9615 d_instantiate_new(dentry, inode);
9618 btrfs_end_transaction(trans);
9620 inode_dec_link_count(inode);
9621 discard_new_inode(inode);
9623 btrfs_btree_balance_dirty(fs_info);
9627 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9628 struct btrfs_trans_handle *trans_in,
9629 struct btrfs_inode *inode,
9630 struct btrfs_key *ins,
9633 struct btrfs_file_extent_item stack_fi;
9634 struct btrfs_replace_extent_info extent_info;
9635 struct btrfs_trans_handle *trans = trans_in;
9636 struct btrfs_path *path;
9637 u64 start = ins->objectid;
9638 u64 len = ins->offset;
9641 memset(&stack_fi, 0, sizeof(stack_fi));
9643 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9644 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9645 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9646 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9647 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9648 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9649 /* Encryption and other encoding is reserved and all 0 */
9651 ret = btrfs_qgroup_release_data(inode, file_offset, len);
9653 return ERR_PTR(ret);
9656 ret = insert_reserved_file_extent(trans, inode,
9657 file_offset, &stack_fi,
9660 return ERR_PTR(ret);
9664 extent_info.disk_offset = start;
9665 extent_info.disk_len = len;
9666 extent_info.data_offset = 0;
9667 extent_info.data_len = len;
9668 extent_info.file_offset = file_offset;
9669 extent_info.extent_buf = (char *)&stack_fi;
9670 extent_info.is_new_extent = true;
9671 extent_info.qgroup_reserved = ret;
9672 extent_info.insertions = 0;
9674 path = btrfs_alloc_path();
9676 return ERR_PTR(-ENOMEM);
9678 ret = btrfs_replace_file_extents(&inode->vfs_inode, path, file_offset,
9679 file_offset + len - 1, &extent_info,
9681 btrfs_free_path(path);
9683 return ERR_PTR(ret);
9688 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9689 u64 start, u64 num_bytes, u64 min_size,
9690 loff_t actual_len, u64 *alloc_hint,
9691 struct btrfs_trans_handle *trans)
9693 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9694 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9695 struct extent_map *em;
9696 struct btrfs_root *root = BTRFS_I(inode)->root;
9697 struct btrfs_key ins;
9698 u64 cur_offset = start;
9699 u64 clear_offset = start;
9702 u64 last_alloc = (u64)-1;
9704 bool own_trans = true;
9705 u64 end = start + num_bytes - 1;
9709 while (num_bytes > 0) {
9710 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9711 cur_bytes = max(cur_bytes, min_size);
9713 * If we are severely fragmented we could end up with really
9714 * small allocations, so if the allocator is returning small
9715 * chunks lets make its job easier by only searching for those
9718 cur_bytes = min(cur_bytes, last_alloc);
9719 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9720 min_size, 0, *alloc_hint, &ins, 1, 0);
9725 * We've reserved this space, and thus converted it from
9726 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9727 * from here on out we will only need to clear our reservation
9728 * for the remaining unreserved area, so advance our
9729 * clear_offset by our extent size.
9731 clear_offset += ins.offset;
9733 last_alloc = ins.offset;
9734 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9737 * Now that we inserted the prealloc extent we can finally
9738 * decrement the number of reservations in the block group.
9739 * If we did it before, we could race with relocation and have
9740 * relocation miss the reserved extent, making it fail later.
9742 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9743 if (IS_ERR(trans)) {
9744 ret = PTR_ERR(trans);
9745 btrfs_free_reserved_extent(fs_info, ins.objectid,
9750 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9751 cur_offset + ins.offset -1, 0);
9753 em = alloc_extent_map();
9755 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9756 &BTRFS_I(inode)->runtime_flags);
9760 em->start = cur_offset;
9761 em->orig_start = cur_offset;
9762 em->len = ins.offset;
9763 em->block_start = ins.objectid;
9764 em->block_len = ins.offset;
9765 em->orig_block_len = ins.offset;
9766 em->ram_bytes = ins.offset;
9767 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9768 em->generation = trans->transid;
9771 write_lock(&em_tree->lock);
9772 ret = add_extent_mapping(em_tree, em, 1);
9773 write_unlock(&em_tree->lock);
9776 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9777 cur_offset + ins.offset - 1,
9780 free_extent_map(em);
9782 num_bytes -= ins.offset;
9783 cur_offset += ins.offset;
9784 *alloc_hint = ins.objectid + ins.offset;
9786 inode_inc_iversion(inode);
9787 inode->i_ctime = current_time(inode);
9788 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9789 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9790 (actual_len > inode->i_size) &&
9791 (cur_offset > inode->i_size)) {
9792 if (cur_offset > actual_len)
9793 i_size = actual_len;
9795 i_size = cur_offset;
9796 i_size_write(inode, i_size);
9797 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9800 ret = btrfs_update_inode(trans, root, inode);
9803 btrfs_abort_transaction(trans, ret);
9805 btrfs_end_transaction(trans);
9810 btrfs_end_transaction(trans);
9814 if (clear_offset < end)
9815 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9816 end - clear_offset + 1);
9820 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9821 u64 start, u64 num_bytes, u64 min_size,
9822 loff_t actual_len, u64 *alloc_hint)
9824 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9825 min_size, actual_len, alloc_hint,
9829 int btrfs_prealloc_file_range_trans(struct inode *inode,
9830 struct btrfs_trans_handle *trans, int mode,
9831 u64 start, u64 num_bytes, u64 min_size,
9832 loff_t actual_len, u64 *alloc_hint)
9834 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9835 min_size, actual_len, alloc_hint, trans);
9838 static int btrfs_set_page_dirty(struct page *page)
9840 return __set_page_dirty_nobuffers(page);
9843 static int btrfs_permission(struct inode *inode, int mask)
9845 struct btrfs_root *root = BTRFS_I(inode)->root;
9846 umode_t mode = inode->i_mode;
9848 if (mask & MAY_WRITE &&
9849 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9850 if (btrfs_root_readonly(root))
9852 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9855 return generic_permission(inode, mask);
9858 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9860 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9861 struct btrfs_trans_handle *trans;
9862 struct btrfs_root *root = BTRFS_I(dir)->root;
9863 struct inode *inode = NULL;
9869 * 5 units required for adding orphan entry
9871 trans = btrfs_start_transaction(root, 5);
9873 return PTR_ERR(trans);
9875 ret = btrfs_find_free_ino(root, &objectid);
9879 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9880 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
9881 if (IS_ERR(inode)) {
9882 ret = PTR_ERR(inode);
9887 inode->i_fop = &btrfs_file_operations;
9888 inode->i_op = &btrfs_file_inode_operations;
9890 inode->i_mapping->a_ops = &btrfs_aops;
9892 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9896 ret = btrfs_update_inode(trans, root, inode);
9899 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
9904 * We set number of links to 0 in btrfs_new_inode(), and here we set
9905 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9908 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9910 set_nlink(inode, 1);
9911 d_tmpfile(dentry, inode);
9912 unlock_new_inode(inode);
9913 mark_inode_dirty(inode);
9915 btrfs_end_transaction(trans);
9917 discard_new_inode(inode);
9918 btrfs_btree_balance_dirty(fs_info);
9922 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
9924 struct inode *inode = tree->private_data;
9925 unsigned long index = start >> PAGE_SHIFT;
9926 unsigned long end_index = end >> PAGE_SHIFT;
9929 while (index <= end_index) {
9930 page = find_get_page(inode->i_mapping, index);
9931 ASSERT(page); /* Pages should be in the extent_io_tree */
9932 set_page_writeback(page);
9940 * Add an entry indicating a block group or device which is pinned by a
9941 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
9942 * negative errno on failure.
9944 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
9945 bool is_block_group)
9947 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9948 struct btrfs_swapfile_pin *sp, *entry;
9950 struct rb_node *parent = NULL;
9952 sp = kmalloc(sizeof(*sp), GFP_NOFS);
9957 sp->is_block_group = is_block_group;
9959 spin_lock(&fs_info->swapfile_pins_lock);
9960 p = &fs_info->swapfile_pins.rb_node;
9963 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
9964 if (sp->ptr < entry->ptr ||
9965 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
9967 } else if (sp->ptr > entry->ptr ||
9968 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
9969 p = &(*p)->rb_right;
9971 spin_unlock(&fs_info->swapfile_pins_lock);
9976 rb_link_node(&sp->node, parent, p);
9977 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
9978 spin_unlock(&fs_info->swapfile_pins_lock);
9982 /* Free all of the entries pinned by this swapfile. */
9983 static void btrfs_free_swapfile_pins(struct inode *inode)
9985 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9986 struct btrfs_swapfile_pin *sp;
9987 struct rb_node *node, *next;
9989 spin_lock(&fs_info->swapfile_pins_lock);
9990 node = rb_first(&fs_info->swapfile_pins);
9992 next = rb_next(node);
9993 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
9994 if (sp->inode == inode) {
9995 rb_erase(&sp->node, &fs_info->swapfile_pins);
9996 if (sp->is_block_group)
9997 btrfs_put_block_group(sp->ptr);
10002 spin_unlock(&fs_info->swapfile_pins_lock);
10005 struct btrfs_swap_info {
10011 unsigned long nr_pages;
10015 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10016 struct btrfs_swap_info *bsi)
10018 unsigned long nr_pages;
10019 u64 first_ppage, first_ppage_reported, next_ppage;
10022 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10023 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10024 PAGE_SIZE) >> PAGE_SHIFT;
10026 if (first_ppage >= next_ppage)
10028 nr_pages = next_ppage - first_ppage;
10030 first_ppage_reported = first_ppage;
10031 if (bsi->start == 0)
10032 first_ppage_reported++;
10033 if (bsi->lowest_ppage > first_ppage_reported)
10034 bsi->lowest_ppage = first_ppage_reported;
10035 if (bsi->highest_ppage < (next_ppage - 1))
10036 bsi->highest_ppage = next_ppage - 1;
10038 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10041 bsi->nr_extents += ret;
10042 bsi->nr_pages += nr_pages;
10046 static void btrfs_swap_deactivate(struct file *file)
10048 struct inode *inode = file_inode(file);
10050 btrfs_free_swapfile_pins(inode);
10051 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10054 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10057 struct inode *inode = file_inode(file);
10058 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10059 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10060 struct extent_state *cached_state = NULL;
10061 struct extent_map *em = NULL;
10062 struct btrfs_device *device = NULL;
10063 struct btrfs_swap_info bsi = {
10064 .lowest_ppage = (sector_t)-1ULL,
10071 * If the swap file was just created, make sure delalloc is done. If the
10072 * file changes again after this, the user is doing something stupid and
10073 * we don't really care.
10075 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10080 * The inode is locked, so these flags won't change after we check them.
10082 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10083 btrfs_warn(fs_info, "swapfile must not be compressed");
10086 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10087 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10090 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10091 btrfs_warn(fs_info, "swapfile must not be checksummed");
10096 * Balance or device remove/replace/resize can move stuff around from
10097 * under us. The exclop protection makes sure they aren't running/won't
10098 * run concurrently while we are mapping the swap extents, and
10099 * fs_info->swapfile_pins prevents them from running while the swap
10100 * file is active and moving the extents. Note that this also prevents
10101 * a concurrent device add which isn't actually necessary, but it's not
10102 * really worth the trouble to allow it.
10104 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10105 btrfs_warn(fs_info,
10106 "cannot activate swapfile while exclusive operation is running");
10110 * Snapshots can create extents which require COW even if NODATACOW is
10111 * set. We use this counter to prevent snapshots. We must increment it
10112 * before walking the extents because we don't want a concurrent
10113 * snapshot to run after we've already checked the extents.
10115 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10117 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10119 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10121 while (start < isize) {
10122 u64 logical_block_start, physical_block_start;
10123 struct btrfs_block_group *bg;
10124 u64 len = isize - start;
10126 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10132 if (em->block_start == EXTENT_MAP_HOLE) {
10133 btrfs_warn(fs_info, "swapfile must not have holes");
10137 if (em->block_start == EXTENT_MAP_INLINE) {
10139 * It's unlikely we'll ever actually find ourselves
10140 * here, as a file small enough to fit inline won't be
10141 * big enough to store more than the swap header, but in
10142 * case something changes in the future, let's catch it
10143 * here rather than later.
10145 btrfs_warn(fs_info, "swapfile must not be inline");
10149 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10150 btrfs_warn(fs_info, "swapfile must not be compressed");
10155 logical_block_start = em->block_start + (start - em->start);
10156 len = min(len, em->len - (start - em->start));
10157 free_extent_map(em);
10160 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10166 btrfs_warn(fs_info,
10167 "swapfile must not be copy-on-write");
10172 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10178 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10179 btrfs_warn(fs_info,
10180 "swapfile must have single data profile");
10185 if (device == NULL) {
10186 device = em->map_lookup->stripes[0].dev;
10187 ret = btrfs_add_swapfile_pin(inode, device, false);
10192 } else if (device != em->map_lookup->stripes[0].dev) {
10193 btrfs_warn(fs_info, "swapfile must be on one device");
10198 physical_block_start = (em->map_lookup->stripes[0].physical +
10199 (logical_block_start - em->start));
10200 len = min(len, em->len - (logical_block_start - em->start));
10201 free_extent_map(em);
10204 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10206 btrfs_warn(fs_info,
10207 "could not find block group containing swapfile");
10212 ret = btrfs_add_swapfile_pin(inode, bg, true);
10214 btrfs_put_block_group(bg);
10221 if (bsi.block_len &&
10222 bsi.block_start + bsi.block_len == physical_block_start) {
10223 bsi.block_len += len;
10225 if (bsi.block_len) {
10226 ret = btrfs_add_swap_extent(sis, &bsi);
10231 bsi.block_start = physical_block_start;
10232 bsi.block_len = len;
10239 ret = btrfs_add_swap_extent(sis, &bsi);
10242 if (!IS_ERR_OR_NULL(em))
10243 free_extent_map(em);
10245 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10248 btrfs_swap_deactivate(file);
10250 btrfs_exclop_finish(fs_info);
10256 sis->bdev = device->bdev;
10257 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10258 sis->max = bsi.nr_pages;
10259 sis->pages = bsi.nr_pages - 1;
10260 sis->highest_bit = bsi.nr_pages - 1;
10261 return bsi.nr_extents;
10264 static void btrfs_swap_deactivate(struct file *file)
10268 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10271 return -EOPNOTSUPP;
10276 * Update the number of bytes used in the VFS' inode. When we replace extents in
10277 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10278 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10279 * always get a correct value.
10281 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10282 const u64 add_bytes,
10283 const u64 del_bytes)
10285 if (add_bytes == del_bytes)
10288 spin_lock(&inode->lock);
10290 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10292 inode_add_bytes(&inode->vfs_inode, add_bytes);
10293 spin_unlock(&inode->lock);
10296 static const struct inode_operations btrfs_dir_inode_operations = {
10297 .getattr = btrfs_getattr,
10298 .lookup = btrfs_lookup,
10299 .create = btrfs_create,
10300 .unlink = btrfs_unlink,
10301 .link = btrfs_link,
10302 .mkdir = btrfs_mkdir,
10303 .rmdir = btrfs_rmdir,
10304 .rename = btrfs_rename2,
10305 .symlink = btrfs_symlink,
10306 .setattr = btrfs_setattr,
10307 .mknod = btrfs_mknod,
10308 .listxattr = btrfs_listxattr,
10309 .permission = btrfs_permission,
10310 .get_acl = btrfs_get_acl,
10311 .set_acl = btrfs_set_acl,
10312 .update_time = btrfs_update_time,
10313 .tmpfile = btrfs_tmpfile,
10316 static const struct file_operations btrfs_dir_file_operations = {
10317 .llseek = generic_file_llseek,
10318 .read = generic_read_dir,
10319 .iterate_shared = btrfs_real_readdir,
10320 .open = btrfs_opendir,
10321 .unlocked_ioctl = btrfs_ioctl,
10322 #ifdef CONFIG_COMPAT
10323 .compat_ioctl = btrfs_compat_ioctl,
10325 .release = btrfs_release_file,
10326 .fsync = btrfs_sync_file,
10330 * btrfs doesn't support the bmap operation because swapfiles
10331 * use bmap to make a mapping of extents in the file. They assume
10332 * these extents won't change over the life of the file and they
10333 * use the bmap result to do IO directly to the drive.
10335 * the btrfs bmap call would return logical addresses that aren't
10336 * suitable for IO and they also will change frequently as COW
10337 * operations happen. So, swapfile + btrfs == corruption.
10339 * For now we're avoiding this by dropping bmap.
10341 static const struct address_space_operations btrfs_aops = {
10342 .readpage = btrfs_readpage,
10343 .writepage = btrfs_writepage,
10344 .writepages = btrfs_writepages,
10345 .readahead = btrfs_readahead,
10346 .direct_IO = noop_direct_IO,
10347 .invalidatepage = btrfs_invalidatepage,
10348 .releasepage = btrfs_releasepage,
10349 #ifdef CONFIG_MIGRATION
10350 .migratepage = btrfs_migratepage,
10352 .set_page_dirty = btrfs_set_page_dirty,
10353 .error_remove_page = generic_error_remove_page,
10354 .swap_activate = btrfs_swap_activate,
10355 .swap_deactivate = btrfs_swap_deactivate,
10358 static const struct inode_operations btrfs_file_inode_operations = {
10359 .getattr = btrfs_getattr,
10360 .setattr = btrfs_setattr,
10361 .listxattr = btrfs_listxattr,
10362 .permission = btrfs_permission,
10363 .fiemap = btrfs_fiemap,
10364 .get_acl = btrfs_get_acl,
10365 .set_acl = btrfs_set_acl,
10366 .update_time = btrfs_update_time,
10368 static const struct inode_operations btrfs_special_inode_operations = {
10369 .getattr = btrfs_getattr,
10370 .setattr = btrfs_setattr,
10371 .permission = btrfs_permission,
10372 .listxattr = btrfs_listxattr,
10373 .get_acl = btrfs_get_acl,
10374 .set_acl = btrfs_set_acl,
10375 .update_time = btrfs_update_time,
10377 static const struct inode_operations btrfs_symlink_inode_operations = {
10378 .get_link = page_get_link,
10379 .getattr = btrfs_getattr,
10380 .setattr = btrfs_setattr,
10381 .permission = btrfs_permission,
10382 .listxattr = btrfs_listxattr,
10383 .update_time = btrfs_update_time,
10386 const struct dentry_operations btrfs_dentry_operations = {
10387 .d_delete = btrfs_dentry_delete,