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"
50 #include "delalloc-space.h"
51 #include "block-group.h"
52 #include "space-info.h"
54 struct btrfs_iget_args {
56 struct btrfs_root *root;
59 struct btrfs_dio_data {
63 struct extent_changeset *data_reserved;
66 static const struct inode_operations btrfs_dir_inode_operations;
67 static const struct inode_operations btrfs_symlink_inode_operations;
68 static const struct inode_operations btrfs_special_inode_operations;
69 static const struct inode_operations btrfs_file_inode_operations;
70 static const struct address_space_operations btrfs_aops;
71 static const struct file_operations btrfs_dir_file_operations;
73 static struct kmem_cache *btrfs_inode_cachep;
74 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_path_cachep;
76 struct kmem_cache *btrfs_free_space_cachep;
77 struct kmem_cache *btrfs_free_space_bitmap_cachep;
79 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
80 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
81 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
82 static noinline int cow_file_range(struct btrfs_inode *inode,
83 struct page *locked_page,
84 u64 start, u64 end, int *page_started,
85 unsigned long *nr_written, int unlock);
86 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
87 u64 len, u64 orig_start, u64 block_start,
88 u64 block_len, u64 orig_block_len,
89 u64 ram_bytes, int compress_type,
92 static void __endio_write_update_ordered(struct btrfs_inode *inode,
93 const u64 offset, const u64 bytes,
97 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
99 * ilock_flags can have the following bit set:
101 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
102 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
105 int btrfs_inode_lock(struct inode *inode, unsigned int ilock_flags)
107 if (ilock_flags & BTRFS_ILOCK_SHARED) {
108 if (ilock_flags & BTRFS_ILOCK_TRY) {
109 if (!inode_trylock_shared(inode))
114 inode_lock_shared(inode);
116 if (ilock_flags & BTRFS_ILOCK_TRY) {
117 if (!inode_trylock(inode))
128 * btrfs_inode_unlock - unock inode i_rwsem
130 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
131 * to decide whether the lock acquired is shared or exclusive.
133 void btrfs_inode_unlock(struct inode *inode, unsigned int ilock_flags)
135 if (ilock_flags & BTRFS_ILOCK_SHARED)
136 inode_unlock_shared(inode);
142 * Cleanup all submitted ordered extents in specified range to handle errors
143 * from the btrfs_run_delalloc_range() callback.
145 * NOTE: caller must ensure that when an error happens, it can not call
146 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
147 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
148 * to be released, which we want to happen only when finishing the ordered
149 * extent (btrfs_finish_ordered_io()).
151 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
152 struct page *locked_page,
153 u64 offset, u64 bytes)
155 unsigned long index = offset >> PAGE_SHIFT;
156 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
157 u64 page_start = page_offset(locked_page);
158 u64 page_end = page_start + PAGE_SIZE - 1;
162 while (index <= end_index) {
163 page = find_get_page(inode->vfs_inode.i_mapping, index);
167 ClearPagePrivate2(page);
172 * In case this page belongs to the delalloc range being instantiated
173 * then skip it, since the first page of a range is going to be
174 * properly cleaned up by the caller of run_delalloc_range
176 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
181 return __endio_write_update_ordered(inode, offset, bytes, false);
184 static int btrfs_dirty_inode(struct inode *inode);
186 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
187 struct inode *inode, struct inode *dir,
188 const struct qstr *qstr)
192 err = btrfs_init_acl(trans, inode, dir);
194 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
199 * this does all the hard work for inserting an inline extent into
200 * the btree. The caller should have done a btrfs_drop_extents so that
201 * no overlapping inline items exist in the btree
203 static int insert_inline_extent(struct btrfs_trans_handle *trans,
204 struct btrfs_path *path, bool extent_inserted,
205 struct btrfs_root *root, struct inode *inode,
206 u64 start, size_t size, size_t compressed_size,
208 struct page **compressed_pages)
210 struct extent_buffer *leaf;
211 struct page *page = NULL;
214 struct btrfs_file_extent_item *ei;
216 size_t cur_size = size;
217 unsigned long offset;
219 ASSERT((compressed_size > 0 && compressed_pages) ||
220 (compressed_size == 0 && !compressed_pages));
222 if (compressed_size && compressed_pages)
223 cur_size = compressed_size;
225 if (!extent_inserted) {
226 struct btrfs_key key;
229 key.objectid = btrfs_ino(BTRFS_I(inode));
231 key.type = BTRFS_EXTENT_DATA_KEY;
233 datasize = btrfs_file_extent_calc_inline_size(cur_size);
234 ret = btrfs_insert_empty_item(trans, root, path, &key,
239 leaf = path->nodes[0];
240 ei = btrfs_item_ptr(leaf, path->slots[0],
241 struct btrfs_file_extent_item);
242 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
243 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
244 btrfs_set_file_extent_encryption(leaf, ei, 0);
245 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
246 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
247 ptr = btrfs_file_extent_inline_start(ei);
249 if (compress_type != BTRFS_COMPRESS_NONE) {
252 while (compressed_size > 0) {
253 cpage = compressed_pages[i];
254 cur_size = min_t(unsigned long, compressed_size,
257 kaddr = kmap_atomic(cpage);
258 write_extent_buffer(leaf, kaddr, ptr, cur_size);
259 kunmap_atomic(kaddr);
263 compressed_size -= cur_size;
265 btrfs_set_file_extent_compression(leaf, ei,
268 page = find_get_page(inode->i_mapping,
269 start >> PAGE_SHIFT);
270 btrfs_set_file_extent_compression(leaf, ei, 0);
271 kaddr = kmap_atomic(page);
272 offset = offset_in_page(start);
273 write_extent_buffer(leaf, kaddr + offset, ptr, size);
274 kunmap_atomic(kaddr);
277 btrfs_mark_buffer_dirty(leaf);
278 btrfs_release_path(path);
281 * We align size to sectorsize for inline extents just for simplicity
284 size = ALIGN(size, root->fs_info->sectorsize);
285 ret = btrfs_inode_set_file_extent_range(BTRFS_I(inode), start, size);
290 * we're an inline extent, so nobody can
291 * extend the file past i_size without locking
292 * a page we already have locked.
294 * We must do any isize and inode updates
295 * before we unlock the pages. Otherwise we
296 * could end up racing with unlink.
298 BTRFS_I(inode)->disk_i_size = inode->i_size;
305 * conditionally insert an inline extent into the file. This
306 * does the checks required to make sure the data is small enough
307 * to fit as an inline extent.
309 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 start,
310 u64 end, size_t compressed_size,
312 struct page **compressed_pages)
314 struct btrfs_drop_extents_args drop_args = { 0 };
315 struct btrfs_root *root = inode->root;
316 struct btrfs_fs_info *fs_info = root->fs_info;
317 struct btrfs_trans_handle *trans;
318 u64 isize = i_size_read(&inode->vfs_inode);
319 u64 actual_end = min(end + 1, isize);
320 u64 inline_len = actual_end - start;
321 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
322 u64 data_len = inline_len;
324 struct btrfs_path *path;
327 data_len = compressed_size;
330 actual_end > fs_info->sectorsize ||
331 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
333 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
335 data_len > fs_info->max_inline) {
339 path = btrfs_alloc_path();
343 trans = btrfs_join_transaction(root);
345 btrfs_free_path(path);
346 return PTR_ERR(trans);
348 trans->block_rsv = &inode->block_rsv;
350 drop_args.path = path;
351 drop_args.start = start;
352 drop_args.end = aligned_end;
353 drop_args.drop_cache = true;
354 drop_args.replace_extent = true;
356 if (compressed_size && compressed_pages)
357 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
360 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(
363 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
365 btrfs_abort_transaction(trans, ret);
369 if (isize > actual_end)
370 inline_len = min_t(u64, isize, actual_end);
371 ret = insert_inline_extent(trans, path, drop_args.extent_inserted,
372 root, &inode->vfs_inode, start,
373 inline_len, compressed_size,
374 compress_type, compressed_pages);
375 if (ret && ret != -ENOSPC) {
376 btrfs_abort_transaction(trans, ret);
378 } else if (ret == -ENOSPC) {
383 btrfs_update_inode_bytes(inode, inline_len, drop_args.bytes_found);
384 ret = btrfs_update_inode(trans, root, inode);
385 if (ret && ret != -ENOSPC) {
386 btrfs_abort_transaction(trans, ret);
388 } else if (ret == -ENOSPC) {
393 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
396 * Don't forget to free the reserved space, as for inlined extent
397 * it won't count as data extent, free them directly here.
398 * And at reserve time, it's always aligned to page size, so
399 * just free one page here.
401 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
402 btrfs_free_path(path);
403 btrfs_end_transaction(trans);
407 struct async_extent {
412 unsigned long nr_pages;
414 struct list_head list;
419 struct page *locked_page;
422 unsigned int write_flags;
423 struct list_head extents;
424 struct cgroup_subsys_state *blkcg_css;
425 struct btrfs_work work;
430 /* Number of chunks in flight; must be first in the structure */
432 struct async_chunk chunks[];
435 static noinline int add_async_extent(struct async_chunk *cow,
436 u64 start, u64 ram_size,
439 unsigned long nr_pages,
442 struct async_extent *async_extent;
444 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
445 BUG_ON(!async_extent); /* -ENOMEM */
446 async_extent->start = start;
447 async_extent->ram_size = ram_size;
448 async_extent->compressed_size = compressed_size;
449 async_extent->pages = pages;
450 async_extent->nr_pages = nr_pages;
451 async_extent->compress_type = compress_type;
452 list_add_tail(&async_extent->list, &cow->extents);
457 * Check if the inode has flags compatible with compression
459 static inline bool inode_can_compress(struct btrfs_inode *inode)
461 if (inode->flags & BTRFS_INODE_NODATACOW ||
462 inode->flags & BTRFS_INODE_NODATASUM)
468 * Check if the inode needs to be submitted to compression, based on mount
469 * options, defragmentation, properties or heuristics.
471 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
474 struct btrfs_fs_info *fs_info = inode->root->fs_info;
476 if (!inode_can_compress(inode)) {
477 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
478 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
483 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
486 if (inode->defrag_compress)
488 /* bad compression ratios */
489 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
491 if (btrfs_test_opt(fs_info, COMPRESS) ||
492 inode->flags & BTRFS_INODE_COMPRESS ||
493 inode->prop_compress)
494 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
498 static inline void inode_should_defrag(struct btrfs_inode *inode,
499 u64 start, u64 end, u64 num_bytes, u64 small_write)
501 /* If this is a small write inside eof, kick off a defrag */
502 if (num_bytes < small_write &&
503 (start > 0 || end + 1 < inode->disk_i_size))
504 btrfs_add_inode_defrag(NULL, inode);
508 * we create compressed extents in two phases. The first
509 * phase compresses a range of pages that have already been
510 * locked (both pages and state bits are locked).
512 * This is done inside an ordered work queue, and the compression
513 * is spread across many cpus. The actual IO submission is step
514 * two, and the ordered work queue takes care of making sure that
515 * happens in the same order things were put onto the queue by
516 * writepages and friends.
518 * If this code finds it can't get good compression, it puts an
519 * entry onto the work queue to write the uncompressed bytes. This
520 * makes sure that both compressed inodes and uncompressed inodes
521 * are written in the same order that the flusher thread sent them
524 static noinline int compress_file_range(struct async_chunk *async_chunk)
526 struct inode *inode = async_chunk->inode;
527 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
528 u64 blocksize = fs_info->sectorsize;
529 u64 start = async_chunk->start;
530 u64 end = async_chunk->end;
534 struct page **pages = NULL;
535 unsigned long nr_pages;
536 unsigned long total_compressed = 0;
537 unsigned long total_in = 0;
540 int compress_type = fs_info->compress_type;
541 int compressed_extents = 0;
544 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
548 * We need to save i_size before now because it could change in between
549 * us evaluating the size and assigning it. This is because we lock and
550 * unlock the page in truncate and fallocate, and then modify the i_size
553 * The barriers are to emulate READ_ONCE, remove that once i_size_read
557 i_size = i_size_read(inode);
559 actual_end = min_t(u64, i_size, end + 1);
562 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
563 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
564 nr_pages = min_t(unsigned long, nr_pages,
565 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
568 * we don't want to send crud past the end of i_size through
569 * compression, that's just a waste of CPU time. So, if the
570 * end of the file is before the start of our current
571 * requested range of bytes, we bail out to the uncompressed
572 * cleanup code that can deal with all of this.
574 * It isn't really the fastest way to fix things, but this is a
575 * very uncommon corner.
577 if (actual_end <= start)
578 goto cleanup_and_bail_uncompressed;
580 total_compressed = actual_end - start;
583 * skip compression for a small file range(<=blocksize) that
584 * isn't an inline extent, since it doesn't save disk space at all.
586 if (total_compressed <= blocksize &&
587 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
588 goto cleanup_and_bail_uncompressed;
590 total_compressed = min_t(unsigned long, total_compressed,
591 BTRFS_MAX_UNCOMPRESSED);
596 * we do compression for mount -o compress and when the
597 * inode has not been flagged as nocompress. This flag can
598 * change at any time if we discover bad compression ratios.
600 if (inode_need_compress(BTRFS_I(inode), start, end)) {
602 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
604 /* just bail out to the uncompressed code */
609 if (BTRFS_I(inode)->defrag_compress)
610 compress_type = BTRFS_I(inode)->defrag_compress;
611 else if (BTRFS_I(inode)->prop_compress)
612 compress_type = BTRFS_I(inode)->prop_compress;
615 * we need to call clear_page_dirty_for_io on each
616 * page in the range. Otherwise applications with the file
617 * mmap'd can wander in and change the page contents while
618 * we are compressing them.
620 * If the compression fails for any reason, we set the pages
621 * dirty again later on.
623 * Note that the remaining part is redirtied, the start pointer
624 * has moved, the end is the original one.
627 extent_range_clear_dirty_for_io(inode, start, end);
631 /* Compression level is applied here and only here */
632 ret = btrfs_compress_pages(
633 compress_type | (fs_info->compress_level << 4),
634 inode->i_mapping, start,
641 unsigned long offset = offset_in_page(total_compressed);
642 struct page *page = pages[nr_pages - 1];
645 /* zero the tail end of the last page, we might be
646 * sending it down to disk
649 kaddr = kmap_atomic(page);
650 memset(kaddr + offset, 0,
652 kunmap_atomic(kaddr);
659 /* lets try to make an inline extent */
660 if (ret || total_in < actual_end) {
661 /* we didn't compress the entire range, try
662 * to make an uncompressed inline extent.
664 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
665 0, BTRFS_COMPRESS_NONE,
668 /* try making a compressed inline extent */
669 ret = cow_file_range_inline(BTRFS_I(inode), start, end,
671 compress_type, pages);
674 unsigned long clear_flags = EXTENT_DELALLOC |
675 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
676 EXTENT_DO_ACCOUNTING;
677 unsigned long page_error_op;
679 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
682 * inline extent creation worked or returned error,
683 * we don't need to create any more async work items.
684 * Unlock and free up our temp pages.
686 * We use DO_ACCOUNTING here because we need the
687 * delalloc_release_metadata to be done _after_ we drop
688 * our outstanding extent for clearing delalloc for this
691 extent_clear_unlock_delalloc(BTRFS_I(inode), start, end,
695 PAGE_START_WRITEBACK |
700 * Ensure we only free the compressed pages if we have
701 * them allocated, as we can still reach here with
702 * inode_need_compress() == false.
705 for (i = 0; i < nr_pages; i++) {
706 WARN_ON(pages[i]->mapping);
717 * we aren't doing an inline extent round the compressed size
718 * up to a block size boundary so the allocator does sane
721 total_compressed = ALIGN(total_compressed, blocksize);
724 * one last check to make sure the compression is really a
725 * win, compare the page count read with the blocks on disk,
726 * compression must free at least one sector size
728 total_in = ALIGN(total_in, PAGE_SIZE);
729 if (total_compressed + blocksize <= total_in) {
730 compressed_extents++;
733 * The async work queues will take care of doing actual
734 * allocation on disk for these compressed pages, and
735 * will submit them to the elevator.
737 add_async_extent(async_chunk, start, total_in,
738 total_compressed, pages, nr_pages,
741 if (start + total_in < end) {
747 return compressed_extents;
752 * the compression code ran but failed to make things smaller,
753 * free any pages it allocated and our page pointer array
755 for (i = 0; i < nr_pages; i++) {
756 WARN_ON(pages[i]->mapping);
761 total_compressed = 0;
764 /* flag the file so we don't compress in the future */
765 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
766 !(BTRFS_I(inode)->prop_compress)) {
767 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
770 cleanup_and_bail_uncompressed:
772 * No compression, but we still need to write the pages in the file
773 * we've been given so far. redirty the locked page if it corresponds
774 * to our extent and set things up for the async work queue to run
775 * cow_file_range to do the normal delalloc dance.
777 if (async_chunk->locked_page &&
778 (page_offset(async_chunk->locked_page) >= start &&
779 page_offset(async_chunk->locked_page)) <= end) {
780 __set_page_dirty_nobuffers(async_chunk->locked_page);
781 /* unlocked later on in the async handlers */
785 extent_range_redirty_for_io(inode, start, end);
786 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
787 BTRFS_COMPRESS_NONE);
788 compressed_extents++;
790 return compressed_extents;
793 static void free_async_extent_pages(struct async_extent *async_extent)
797 if (!async_extent->pages)
800 for (i = 0; i < async_extent->nr_pages; i++) {
801 WARN_ON(async_extent->pages[i]->mapping);
802 put_page(async_extent->pages[i]);
804 kfree(async_extent->pages);
805 async_extent->nr_pages = 0;
806 async_extent->pages = NULL;
810 * phase two of compressed writeback. This is the ordered portion
811 * of the code, which only gets called in the order the work was
812 * queued. We walk all the async extents created by compress_file_range
813 * and send them down to the disk.
815 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
817 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
818 struct btrfs_fs_info *fs_info = inode->root->fs_info;
819 struct async_extent *async_extent;
821 struct btrfs_key ins;
822 struct extent_map *em;
823 struct btrfs_root *root = inode->root;
824 struct extent_io_tree *io_tree = &inode->io_tree;
828 while (!list_empty(&async_chunk->extents)) {
829 async_extent = list_entry(async_chunk->extents.next,
830 struct async_extent, list);
831 list_del(&async_extent->list);
834 lock_extent(io_tree, async_extent->start,
835 async_extent->start + async_extent->ram_size - 1);
836 /* did the compression code fall back to uncompressed IO? */
837 if (!async_extent->pages) {
838 int page_started = 0;
839 unsigned long nr_written = 0;
841 /* allocate blocks */
842 ret = cow_file_range(inode, async_chunk->locked_page,
844 async_extent->start +
845 async_extent->ram_size - 1,
846 &page_started, &nr_written, 0);
851 * if page_started, cow_file_range inserted an
852 * inline extent and took care of all the unlocking
853 * and IO for us. Otherwise, we need to submit
854 * all those pages down to the drive.
856 if (!page_started && !ret)
857 extent_write_locked_range(&inode->vfs_inode,
859 async_extent->start +
860 async_extent->ram_size - 1,
862 else if (ret && async_chunk->locked_page)
863 unlock_page(async_chunk->locked_page);
869 ret = btrfs_reserve_extent(root, async_extent->ram_size,
870 async_extent->compressed_size,
871 async_extent->compressed_size,
872 0, alloc_hint, &ins, 1, 1);
874 free_async_extent_pages(async_extent);
876 if (ret == -ENOSPC) {
877 unlock_extent(io_tree, async_extent->start,
878 async_extent->start +
879 async_extent->ram_size - 1);
882 * we need to redirty the pages if we decide to
883 * fallback to uncompressed IO, otherwise we
884 * will not submit these pages down to lower
887 extent_range_redirty_for_io(&inode->vfs_inode,
889 async_extent->start +
890 async_extent->ram_size - 1);
897 * here we're doing allocation and writeback of the
900 em = create_io_em(inode, async_extent->start,
901 async_extent->ram_size, /* len */
902 async_extent->start, /* orig_start */
903 ins.objectid, /* block_start */
904 ins.offset, /* block_len */
905 ins.offset, /* orig_block_len */
906 async_extent->ram_size, /* ram_bytes */
907 async_extent->compress_type,
908 BTRFS_ORDERED_COMPRESSED);
910 /* ret value is not necessary due to void function */
911 goto out_free_reserve;
914 ret = btrfs_add_ordered_extent_compress(inode,
917 async_extent->ram_size,
919 async_extent->compress_type);
921 btrfs_drop_extent_cache(inode, async_extent->start,
922 async_extent->start +
923 async_extent->ram_size - 1, 0);
924 goto out_free_reserve;
926 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
929 * clear dirty, set writeback and unlock the pages.
931 extent_clear_unlock_delalloc(inode, async_extent->start,
932 async_extent->start +
933 async_extent->ram_size - 1,
934 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
935 PAGE_UNLOCK | PAGE_START_WRITEBACK);
936 if (btrfs_submit_compressed_write(inode, async_extent->start,
937 async_extent->ram_size,
939 ins.offset, async_extent->pages,
940 async_extent->nr_pages,
941 async_chunk->write_flags,
942 async_chunk->blkcg_css)) {
943 struct page *p = async_extent->pages[0];
944 const u64 start = async_extent->start;
945 const u64 end = start + async_extent->ram_size - 1;
947 p->mapping = inode->vfs_inode.i_mapping;
948 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
951 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
954 free_async_extent_pages(async_extent);
956 alloc_hint = ins.objectid + ins.offset;
962 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
963 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
965 extent_clear_unlock_delalloc(inode, async_extent->start,
966 async_extent->start +
967 async_extent->ram_size - 1,
968 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
969 EXTENT_DELALLOC_NEW |
970 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
971 PAGE_UNLOCK | PAGE_START_WRITEBACK |
972 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
973 free_async_extent_pages(async_extent);
978 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
981 struct extent_map_tree *em_tree = &inode->extent_tree;
982 struct extent_map *em;
985 read_lock(&em_tree->lock);
986 em = search_extent_mapping(em_tree, start, num_bytes);
989 * if block start isn't an actual block number then find the
990 * first block in this inode and use that as a hint. If that
991 * block is also bogus then just don't worry about it.
993 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
995 em = search_extent_mapping(em_tree, 0, 0);
996 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
997 alloc_hint = em->block_start;
1001 alloc_hint = em->block_start;
1002 free_extent_map(em);
1005 read_unlock(&em_tree->lock);
1011 * when extent_io.c finds a delayed allocation range in the file,
1012 * the call backs end up in this code. The basic idea is to
1013 * allocate extents on disk for the range, and create ordered data structs
1014 * in ram to track those extents.
1016 * locked_page is the page that writepage had locked already. We use
1017 * it to make sure we don't do extra locks or unlocks.
1019 * *page_started is set to one if we unlock locked_page and do everything
1020 * required to start IO on it. It may be clean and already done with
1021 * IO when we return.
1023 static noinline int cow_file_range(struct btrfs_inode *inode,
1024 struct page *locked_page,
1025 u64 start, u64 end, int *page_started,
1026 unsigned long *nr_written, int unlock)
1028 struct btrfs_root *root = inode->root;
1029 struct btrfs_fs_info *fs_info = root->fs_info;
1032 unsigned long ram_size;
1033 u64 cur_alloc_size = 0;
1035 u64 blocksize = fs_info->sectorsize;
1036 struct btrfs_key ins;
1037 struct extent_map *em;
1038 unsigned clear_bits;
1039 unsigned long page_ops;
1040 bool extent_reserved = false;
1043 if (btrfs_is_free_space_inode(inode)) {
1049 num_bytes = ALIGN(end - start + 1, blocksize);
1050 num_bytes = max(blocksize, num_bytes);
1051 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1053 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1056 /* lets try to make an inline extent */
1057 ret = cow_file_range_inline(inode, start, end, 0,
1058 BTRFS_COMPRESS_NONE, NULL);
1061 * We use DO_ACCOUNTING here because we need the
1062 * delalloc_release_metadata to be run _after_ we drop
1063 * our outstanding extent for clearing delalloc for this
1066 extent_clear_unlock_delalloc(inode, start, end, NULL,
1067 EXTENT_LOCKED | EXTENT_DELALLOC |
1068 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1069 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1070 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1071 *nr_written = *nr_written +
1072 (end - start + PAGE_SIZE) / PAGE_SIZE;
1075 } else if (ret < 0) {
1080 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1081 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1084 * Relocation relies on the relocated extents to have exactly the same
1085 * size as the original extents. Normally writeback for relocation data
1086 * extents follows a NOCOW path because relocation preallocates the
1087 * extents. However, due to an operation such as scrub turning a block
1088 * group to RO mode, it may fallback to COW mode, so we must make sure
1089 * an extent allocated during COW has exactly the requested size and can
1090 * not be split into smaller extents, otherwise relocation breaks and
1091 * fails during the stage where it updates the bytenr of file extent
1094 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1095 min_alloc_size = num_bytes;
1097 min_alloc_size = fs_info->sectorsize;
1099 while (num_bytes > 0) {
1100 cur_alloc_size = num_bytes;
1101 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1102 min_alloc_size, 0, alloc_hint,
1106 cur_alloc_size = ins.offset;
1107 extent_reserved = true;
1109 ram_size = ins.offset;
1110 em = create_io_em(inode, start, ins.offset, /* len */
1111 start, /* orig_start */
1112 ins.objectid, /* block_start */
1113 ins.offset, /* block_len */
1114 ins.offset, /* orig_block_len */
1115 ram_size, /* ram_bytes */
1116 BTRFS_COMPRESS_NONE, /* compress_type */
1117 BTRFS_ORDERED_REGULAR /* type */);
1122 free_extent_map(em);
1124 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1125 ram_size, cur_alloc_size,
1126 BTRFS_ORDERED_REGULAR);
1128 goto out_drop_extent_cache;
1130 if (root->root_key.objectid ==
1131 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1132 ret = btrfs_reloc_clone_csums(inode, start,
1135 * Only drop cache here, and process as normal.
1137 * We must not allow extent_clear_unlock_delalloc()
1138 * at out_unlock label to free meta of this ordered
1139 * extent, as its meta should be freed by
1140 * btrfs_finish_ordered_io().
1142 * So we must continue until @start is increased to
1143 * skip current ordered extent.
1146 btrfs_drop_extent_cache(inode, start,
1147 start + ram_size - 1, 0);
1150 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1152 /* we're not doing compressed IO, don't unlock the first
1153 * page (which the caller expects to stay locked), don't
1154 * clear any dirty bits and don't set any writeback bits
1156 * Do set the Private2 bit so we know this page was properly
1157 * setup for writepage
1159 page_ops = unlock ? PAGE_UNLOCK : 0;
1160 page_ops |= PAGE_SET_PRIVATE2;
1162 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1164 EXTENT_LOCKED | EXTENT_DELALLOC,
1166 if (num_bytes < cur_alloc_size)
1169 num_bytes -= cur_alloc_size;
1170 alloc_hint = ins.objectid + ins.offset;
1171 start += cur_alloc_size;
1172 extent_reserved = false;
1175 * btrfs_reloc_clone_csums() error, since start is increased
1176 * extent_clear_unlock_delalloc() at out_unlock label won't
1177 * free metadata of current ordered extent, we're OK to exit.
1185 out_drop_extent_cache:
1186 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1188 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1189 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1191 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1192 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1193 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1195 * If we reserved an extent for our delalloc range (or a subrange) and
1196 * failed to create the respective ordered extent, then it means that
1197 * when we reserved the extent we decremented the extent's size from
1198 * the data space_info's bytes_may_use counter and incremented the
1199 * space_info's bytes_reserved counter by the same amount. We must make
1200 * sure extent_clear_unlock_delalloc() does not try to decrement again
1201 * the data space_info's bytes_may_use counter, therefore we do not pass
1202 * it the flag EXTENT_CLEAR_DATA_RESV.
1204 if (extent_reserved) {
1205 extent_clear_unlock_delalloc(inode, start,
1206 start + cur_alloc_size - 1,
1210 start += cur_alloc_size;
1214 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1215 clear_bits | EXTENT_CLEAR_DATA_RESV,
1221 * work queue call back to started compression on a file and pages
1223 static noinline void async_cow_start(struct btrfs_work *work)
1225 struct async_chunk *async_chunk;
1226 int compressed_extents;
1228 async_chunk = container_of(work, struct async_chunk, work);
1230 compressed_extents = compress_file_range(async_chunk);
1231 if (compressed_extents == 0) {
1232 btrfs_add_delayed_iput(async_chunk->inode);
1233 async_chunk->inode = NULL;
1238 * work queue call back to submit previously compressed pages
1240 static noinline void async_cow_submit(struct btrfs_work *work)
1242 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1244 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1245 unsigned long nr_pages;
1247 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1250 /* atomic_sub_return implies a barrier */
1251 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1253 cond_wake_up_nomb(&fs_info->async_submit_wait);
1256 * ->inode could be NULL if async_chunk_start has failed to compress,
1257 * in which case we don't have anything to submit, yet we need to
1258 * always adjust ->async_delalloc_pages as its paired with the init
1259 * happening in cow_file_range_async
1261 if (async_chunk->inode)
1262 submit_compressed_extents(async_chunk);
1265 static noinline void async_cow_free(struct btrfs_work *work)
1267 struct async_chunk *async_chunk;
1269 async_chunk = container_of(work, struct async_chunk, work);
1270 if (async_chunk->inode)
1271 btrfs_add_delayed_iput(async_chunk->inode);
1272 if (async_chunk->blkcg_css)
1273 css_put(async_chunk->blkcg_css);
1275 * Since the pointer to 'pending' is at the beginning of the array of
1276 * async_chunk's, freeing it ensures the whole array has been freed.
1278 if (atomic_dec_and_test(async_chunk->pending))
1279 kvfree(async_chunk->pending);
1282 static int cow_file_range_async(struct btrfs_inode *inode,
1283 struct writeback_control *wbc,
1284 struct page *locked_page,
1285 u64 start, u64 end, int *page_started,
1286 unsigned long *nr_written)
1288 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1289 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1290 struct async_cow *ctx;
1291 struct async_chunk *async_chunk;
1292 unsigned long nr_pages;
1294 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1296 bool should_compress;
1298 const unsigned int write_flags = wbc_to_write_flags(wbc);
1300 unlock_extent(&inode->io_tree, start, end);
1302 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1303 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1305 should_compress = false;
1307 should_compress = true;
1310 nofs_flag = memalloc_nofs_save();
1311 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1312 memalloc_nofs_restore(nofs_flag);
1315 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1316 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1317 EXTENT_DO_ACCOUNTING;
1318 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1319 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1321 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1322 clear_bits, page_ops);
1326 async_chunk = ctx->chunks;
1327 atomic_set(&ctx->num_chunks, num_chunks);
1329 for (i = 0; i < num_chunks; i++) {
1330 if (should_compress)
1331 cur_end = min(end, start + SZ_512K - 1);
1336 * igrab is called higher up in the call chain, take only the
1337 * lightweight reference for the callback lifetime
1339 ihold(&inode->vfs_inode);
1340 async_chunk[i].pending = &ctx->num_chunks;
1341 async_chunk[i].inode = &inode->vfs_inode;
1342 async_chunk[i].start = start;
1343 async_chunk[i].end = cur_end;
1344 async_chunk[i].write_flags = write_flags;
1345 INIT_LIST_HEAD(&async_chunk[i].extents);
1348 * The locked_page comes all the way from writepage and its
1349 * the original page we were actually given. As we spread
1350 * this large delalloc region across multiple async_chunk
1351 * structs, only the first struct needs a pointer to locked_page
1353 * This way we don't need racey decisions about who is supposed
1358 * Depending on the compressibility, the pages might or
1359 * might not go through async. We want all of them to
1360 * be accounted against wbc once. Let's do it here
1361 * before the paths diverge. wbc accounting is used
1362 * only for foreign writeback detection and doesn't
1363 * need full accuracy. Just account the whole thing
1364 * against the first page.
1366 wbc_account_cgroup_owner(wbc, locked_page,
1368 async_chunk[i].locked_page = locked_page;
1371 async_chunk[i].locked_page = NULL;
1374 if (blkcg_css != blkcg_root_css) {
1376 async_chunk[i].blkcg_css = blkcg_css;
1378 async_chunk[i].blkcg_css = NULL;
1381 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1382 async_cow_submit, async_cow_free);
1384 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1385 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1387 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1389 *nr_written += nr_pages;
1390 start = cur_end + 1;
1396 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1397 u64 bytenr, u64 num_bytes)
1400 struct btrfs_ordered_sum *sums;
1403 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1404 bytenr + num_bytes - 1, &list, 0);
1405 if (ret == 0 && list_empty(&list))
1408 while (!list_empty(&list)) {
1409 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1410 list_del(&sums->list);
1418 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1419 const u64 start, const u64 end,
1420 int *page_started, unsigned long *nr_written)
1422 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1423 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1424 BTRFS_DATA_RELOC_TREE_OBJECTID);
1425 const u64 range_bytes = end + 1 - start;
1426 struct extent_io_tree *io_tree = &inode->io_tree;
1427 u64 range_start = start;
1431 * If EXTENT_NORESERVE is set it means that when the buffered write was
1432 * made we had not enough available data space and therefore we did not
1433 * reserve data space for it, since we though we could do NOCOW for the
1434 * respective file range (either there is prealloc extent or the inode
1435 * has the NOCOW bit set).
1437 * However when we need to fallback to COW mode (because for example the
1438 * block group for the corresponding extent was turned to RO mode by a
1439 * scrub or relocation) we need to do the following:
1441 * 1) We increment the bytes_may_use counter of the data space info.
1442 * If COW succeeds, it allocates a new data extent and after doing
1443 * that it decrements the space info's bytes_may_use counter and
1444 * increments its bytes_reserved counter by the same amount (we do
1445 * this at btrfs_add_reserved_bytes()). So we need to increment the
1446 * bytes_may_use counter to compensate (when space is reserved at
1447 * buffered write time, the bytes_may_use counter is incremented);
1449 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1450 * that if the COW path fails for any reason, it decrements (through
1451 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1452 * data space info, which we incremented in the step above.
1454 * If we need to fallback to cow and the inode corresponds to a free
1455 * space cache inode or an inode of the data relocation tree, we must
1456 * also increment bytes_may_use of the data space_info for the same
1457 * reason. Space caches and relocated data extents always get a prealloc
1458 * extent for them, however scrub or balance may have set the block
1459 * group that contains that extent to RO mode and therefore force COW
1460 * when starting writeback.
1462 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1463 EXTENT_NORESERVE, 0);
1464 if (count > 0 || is_space_ino || is_reloc_ino) {
1466 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1467 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1469 if (is_space_ino || is_reloc_ino)
1470 bytes = range_bytes;
1472 spin_lock(&sinfo->lock);
1473 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1474 spin_unlock(&sinfo->lock);
1477 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1481 return cow_file_range(inode, locked_page, start, end, page_started,
1486 * when nowcow writeback call back. This checks for snapshots or COW copies
1487 * of the extents that exist in the file, and COWs the file as required.
1489 * If no cow copies or snapshots exist, we write directly to the existing
1492 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1493 struct page *locked_page,
1494 const u64 start, const u64 end,
1495 int *page_started, int force,
1496 unsigned long *nr_written)
1498 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1499 struct btrfs_root *root = inode->root;
1500 struct btrfs_path *path;
1501 u64 cow_start = (u64)-1;
1502 u64 cur_offset = start;
1504 bool check_prev = true;
1505 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1506 u64 ino = btrfs_ino(inode);
1508 u64 disk_bytenr = 0;
1510 path = btrfs_alloc_path();
1512 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1513 EXTENT_LOCKED | EXTENT_DELALLOC |
1514 EXTENT_DO_ACCOUNTING |
1515 EXTENT_DEFRAG, PAGE_UNLOCK |
1516 PAGE_START_WRITEBACK |
1517 PAGE_END_WRITEBACK);
1522 struct btrfs_key found_key;
1523 struct btrfs_file_extent_item *fi;
1524 struct extent_buffer *leaf;
1534 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1540 * If there is no extent for our range when doing the initial
1541 * search, then go back to the previous slot as it will be the
1542 * one containing the search offset
1544 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1545 leaf = path->nodes[0];
1546 btrfs_item_key_to_cpu(leaf, &found_key,
1547 path->slots[0] - 1);
1548 if (found_key.objectid == ino &&
1549 found_key.type == BTRFS_EXTENT_DATA_KEY)
1554 /* Go to next leaf if we have exhausted the current one */
1555 leaf = path->nodes[0];
1556 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1557 ret = btrfs_next_leaf(root, path);
1559 if (cow_start != (u64)-1)
1560 cur_offset = cow_start;
1565 leaf = path->nodes[0];
1568 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1570 /* Didn't find anything for our INO */
1571 if (found_key.objectid > ino)
1574 * Keep searching until we find an EXTENT_ITEM or there are no
1575 * more extents for this inode
1577 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1578 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1583 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1584 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1585 found_key.offset > end)
1589 * If the found extent starts after requested offset, then
1590 * adjust extent_end to be right before this extent begins
1592 if (found_key.offset > cur_offset) {
1593 extent_end = found_key.offset;
1599 * Found extent which begins before our range and potentially
1602 fi = btrfs_item_ptr(leaf, path->slots[0],
1603 struct btrfs_file_extent_item);
1604 extent_type = btrfs_file_extent_type(leaf, fi);
1606 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1607 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1608 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1609 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1610 extent_offset = btrfs_file_extent_offset(leaf, fi);
1611 extent_end = found_key.offset +
1612 btrfs_file_extent_num_bytes(leaf, fi);
1614 btrfs_file_extent_disk_num_bytes(leaf, fi);
1616 * If the extent we got ends before our current offset,
1617 * skip to the next extent.
1619 if (extent_end <= cur_offset) {
1624 if (disk_bytenr == 0)
1626 /* Skip compressed/encrypted/encoded extents */
1627 if (btrfs_file_extent_compression(leaf, fi) ||
1628 btrfs_file_extent_encryption(leaf, fi) ||
1629 btrfs_file_extent_other_encoding(leaf, fi))
1632 * If extent is created before the last volume's snapshot
1633 * this implies the extent is shared, hence we can't do
1634 * nocow. This is the same check as in
1635 * btrfs_cross_ref_exist but without calling
1636 * btrfs_search_slot.
1638 if (!freespace_inode &&
1639 btrfs_file_extent_generation(leaf, fi) <=
1640 btrfs_root_last_snapshot(&root->root_item))
1642 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1646 * The following checks can be expensive, as they need to
1647 * take other locks and do btree or rbtree searches, so
1648 * release the path to avoid blocking other tasks for too
1651 btrfs_release_path(path);
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)
1728 if (!path->nodes[0])
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 |
1838 PAGE_START_WRITEBACK |
1839 PAGE_END_WRITEBACK);
1840 btrfs_free_path(path);
1844 static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end)
1847 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1848 !(inode->flags & BTRFS_INODE_PREALLOC))
1852 * @defrag_bytes is a hint value, no spinlock held here,
1853 * if is not zero, it means the file is defragging.
1854 * Force cow if given extent needs to be defragged.
1856 if (inode->defrag_bytes &&
1857 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL))
1864 * Function to process delayed allocation (create CoW) for ranges which are
1865 * being touched for the first time.
1867 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1868 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1869 struct writeback_control *wbc)
1872 int force_cow = need_force_cow(inode, start, end);
1874 if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1875 ret = run_delalloc_nocow(inode, locked_page, start, end,
1876 page_started, 1, nr_written);
1877 } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1878 ret = run_delalloc_nocow(inode, locked_page, start, end,
1879 page_started, 0, nr_written);
1880 } else if (!inode_can_compress(inode) ||
1881 !inode_need_compress(inode, start, end)) {
1882 ret = cow_file_range(inode, locked_page, start, end,
1883 page_started, nr_written, 1);
1885 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1886 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1887 page_started, nr_written);
1890 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1895 void btrfs_split_delalloc_extent(struct inode *inode,
1896 struct extent_state *orig, u64 split)
1900 /* not delalloc, ignore it */
1901 if (!(orig->state & EXTENT_DELALLOC))
1904 size = orig->end - orig->start + 1;
1905 if (size > BTRFS_MAX_EXTENT_SIZE) {
1910 * See the explanation in btrfs_merge_delalloc_extent, the same
1911 * applies here, just in reverse.
1913 new_size = orig->end - split + 1;
1914 num_extents = count_max_extents(new_size);
1915 new_size = split - orig->start;
1916 num_extents += count_max_extents(new_size);
1917 if (count_max_extents(size) >= num_extents)
1921 spin_lock(&BTRFS_I(inode)->lock);
1922 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1923 spin_unlock(&BTRFS_I(inode)->lock);
1927 * Handle merged delayed allocation extents so we can keep track of new extents
1928 * that are just merged onto old extents, such as when we are doing sequential
1929 * writes, so we can properly account for the metadata space we'll need.
1931 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1932 struct extent_state *other)
1934 u64 new_size, old_size;
1937 /* not delalloc, ignore it */
1938 if (!(other->state & EXTENT_DELALLOC))
1941 if (new->start > other->start)
1942 new_size = new->end - other->start + 1;
1944 new_size = other->end - new->start + 1;
1946 /* we're not bigger than the max, unreserve the space and go */
1947 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1948 spin_lock(&BTRFS_I(inode)->lock);
1949 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1950 spin_unlock(&BTRFS_I(inode)->lock);
1955 * We have to add up either side to figure out how many extents were
1956 * accounted for before we merged into one big extent. If the number of
1957 * extents we accounted for is <= the amount we need for the new range
1958 * then we can return, otherwise drop. Think of it like this
1962 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1963 * need 2 outstanding extents, on one side we have 1 and the other side
1964 * we have 1 so they are == and we can return. But in this case
1966 * [MAX_SIZE+4k][MAX_SIZE+4k]
1968 * Each range on their own accounts for 2 extents, but merged together
1969 * they are only 3 extents worth of accounting, so we need to drop in
1972 old_size = other->end - other->start + 1;
1973 num_extents = count_max_extents(old_size);
1974 old_size = new->end - new->start + 1;
1975 num_extents += count_max_extents(old_size);
1976 if (count_max_extents(new_size) >= num_extents)
1979 spin_lock(&BTRFS_I(inode)->lock);
1980 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1981 spin_unlock(&BTRFS_I(inode)->lock);
1984 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1985 struct inode *inode)
1987 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1989 spin_lock(&root->delalloc_lock);
1990 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1991 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1992 &root->delalloc_inodes);
1993 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1994 &BTRFS_I(inode)->runtime_flags);
1995 root->nr_delalloc_inodes++;
1996 if (root->nr_delalloc_inodes == 1) {
1997 spin_lock(&fs_info->delalloc_root_lock);
1998 BUG_ON(!list_empty(&root->delalloc_root));
1999 list_add_tail(&root->delalloc_root,
2000 &fs_info->delalloc_roots);
2001 spin_unlock(&fs_info->delalloc_root_lock);
2004 spin_unlock(&root->delalloc_lock);
2008 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2009 struct btrfs_inode *inode)
2011 struct btrfs_fs_info *fs_info = root->fs_info;
2013 if (!list_empty(&inode->delalloc_inodes)) {
2014 list_del_init(&inode->delalloc_inodes);
2015 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2016 &inode->runtime_flags);
2017 root->nr_delalloc_inodes--;
2018 if (!root->nr_delalloc_inodes) {
2019 ASSERT(list_empty(&root->delalloc_inodes));
2020 spin_lock(&fs_info->delalloc_root_lock);
2021 BUG_ON(list_empty(&root->delalloc_root));
2022 list_del_init(&root->delalloc_root);
2023 spin_unlock(&fs_info->delalloc_root_lock);
2028 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2029 struct btrfs_inode *inode)
2031 spin_lock(&root->delalloc_lock);
2032 __btrfs_del_delalloc_inode(root, inode);
2033 spin_unlock(&root->delalloc_lock);
2037 * Properly track delayed allocation bytes in the inode and to maintain the
2038 * list of inodes that have pending delalloc work to be done.
2040 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2043 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2045 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2048 * set_bit and clear bit hooks normally require _irqsave/restore
2049 * but in this case, we are only testing for the DELALLOC
2050 * bit, which is only set or cleared with irqs on
2052 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2053 struct btrfs_root *root = BTRFS_I(inode)->root;
2054 u64 len = state->end + 1 - state->start;
2055 u32 num_extents = count_max_extents(len);
2056 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2058 spin_lock(&BTRFS_I(inode)->lock);
2059 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2060 spin_unlock(&BTRFS_I(inode)->lock);
2062 /* For sanity tests */
2063 if (btrfs_is_testing(fs_info))
2066 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2067 fs_info->delalloc_batch);
2068 spin_lock(&BTRFS_I(inode)->lock);
2069 BTRFS_I(inode)->delalloc_bytes += len;
2070 if (*bits & EXTENT_DEFRAG)
2071 BTRFS_I(inode)->defrag_bytes += len;
2072 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2073 &BTRFS_I(inode)->runtime_flags))
2074 btrfs_add_delalloc_inodes(root, inode);
2075 spin_unlock(&BTRFS_I(inode)->lock);
2078 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2079 (*bits & EXTENT_DELALLOC_NEW)) {
2080 spin_lock(&BTRFS_I(inode)->lock);
2081 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2083 spin_unlock(&BTRFS_I(inode)->lock);
2088 * Once a range is no longer delalloc this function ensures that proper
2089 * accounting happens.
2091 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2092 struct extent_state *state, unsigned *bits)
2094 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2095 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2096 u64 len = state->end + 1 - state->start;
2097 u32 num_extents = count_max_extents(len);
2099 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2100 spin_lock(&inode->lock);
2101 inode->defrag_bytes -= len;
2102 spin_unlock(&inode->lock);
2106 * set_bit and clear bit hooks normally require _irqsave/restore
2107 * but in this case, we are only testing for the DELALLOC
2108 * bit, which is only set or cleared with irqs on
2110 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2111 struct btrfs_root *root = inode->root;
2112 bool do_list = !btrfs_is_free_space_inode(inode);
2114 spin_lock(&inode->lock);
2115 btrfs_mod_outstanding_extents(inode, -num_extents);
2116 spin_unlock(&inode->lock);
2119 * We don't reserve metadata space for space cache inodes so we
2120 * don't need to call delalloc_release_metadata if there is an
2123 if (*bits & EXTENT_CLEAR_META_RESV &&
2124 root != fs_info->tree_root)
2125 btrfs_delalloc_release_metadata(inode, len, false);
2127 /* For sanity tests. */
2128 if (btrfs_is_testing(fs_info))
2131 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2132 do_list && !(state->state & EXTENT_NORESERVE) &&
2133 (*bits & EXTENT_CLEAR_DATA_RESV))
2134 btrfs_free_reserved_data_space_noquota(fs_info, len);
2136 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2137 fs_info->delalloc_batch);
2138 spin_lock(&inode->lock);
2139 inode->delalloc_bytes -= len;
2140 if (do_list && inode->delalloc_bytes == 0 &&
2141 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2142 &inode->runtime_flags))
2143 btrfs_del_delalloc_inode(root, inode);
2144 spin_unlock(&inode->lock);
2147 if ((state->state & EXTENT_DELALLOC_NEW) &&
2148 (*bits & EXTENT_DELALLOC_NEW)) {
2149 spin_lock(&inode->lock);
2150 ASSERT(inode->new_delalloc_bytes >= len);
2151 inode->new_delalloc_bytes -= len;
2152 if (*bits & EXTENT_ADD_INODE_BYTES)
2153 inode_add_bytes(&inode->vfs_inode, len);
2154 spin_unlock(&inode->lock);
2159 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2160 * in a chunk's stripe. This function ensures that bios do not span a
2163 * @page - The page we are about to add to the bio
2164 * @size - size we want to add to the bio
2165 * @bio - bio we want to ensure is smaller than a stripe
2166 * @bio_flags - flags of the bio
2168 * return 1 if page cannot be added to the bio
2169 * return 0 if page can be added to the bio
2170 * return error otherwise
2172 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2173 unsigned long bio_flags)
2175 struct inode *inode = page->mapping->host;
2176 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2177 u64 logical = bio->bi_iter.bi_sector << 9;
2178 struct extent_map *em;
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 em = btrfs_get_chunk_map(fs_info, logical, map_length);
2192 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(bio), logical,
2197 if (geom.len < length + size)
2200 free_extent_map(em);
2205 * in order to insert checksums into the metadata in large chunks,
2206 * we wait until bio submission time. All the pages in the bio are
2207 * checksummed and sums are attached onto the ordered extent record.
2209 * At IO completion time the cums attached on the ordered extent record
2210 * are inserted into the btree
2212 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2213 u64 dio_file_offset)
2215 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2219 * extent_io.c submission hook. This does the right thing for csum calculation
2220 * on write, or reading the csums from the tree before a read.
2222 * Rules about async/sync submit,
2223 * a) read: sync submit
2225 * b) write without checksum: sync submit
2227 * c) write with checksum:
2228 * c-1) if bio is issued by fsync: sync submit
2229 * (sync_writers != 0)
2231 * c-2) if root is reloc root: sync submit
2232 * (only in case of buffered IO)
2234 * c-3) otherwise: async submit
2236 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2237 int mirror_num, unsigned long bio_flags)
2240 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2241 struct btrfs_root *root = BTRFS_I(inode)->root;
2242 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2243 blk_status_t ret = 0;
2245 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2247 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2248 !fs_info->csum_root;
2250 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2251 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2253 if (bio_op(bio) != REQ_OP_WRITE) {
2254 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2258 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2259 ret = btrfs_submit_compressed_read(inode, bio,
2265 * Lookup bio sums does extra checks around whether we
2266 * need to csum or not, which is why we ignore skip_sum
2269 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2274 } else if (async && !skip_sum) {
2275 /* csum items have already been cloned */
2276 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2278 /* we're doing a write, do the async checksumming */
2279 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2280 0, btrfs_submit_bio_start);
2282 } else if (!skip_sum) {
2283 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2289 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2293 bio->bi_status = ret;
2300 * given a list of ordered sums record them in the inode. This happens
2301 * at IO completion time based on sums calculated at bio submission time.
2303 static int add_pending_csums(struct btrfs_trans_handle *trans,
2304 struct list_head *list)
2306 struct btrfs_ordered_sum *sum;
2309 list_for_each_entry(sum, list, list) {
2310 trans->adding_csums = true;
2311 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2312 trans->adding_csums = false;
2319 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2322 struct extent_state **cached_state)
2324 u64 search_start = start;
2325 const u64 end = start + len - 1;
2327 while (search_start < end) {
2328 const u64 search_len = end - search_start + 1;
2329 struct extent_map *em;
2333 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2337 if (em->block_start != EXTENT_MAP_HOLE)
2341 if (em->start < search_start)
2342 em_len -= search_start - em->start;
2343 if (em_len > search_len)
2344 em_len = search_len;
2346 ret = set_extent_bit(&inode->io_tree, search_start,
2347 search_start + em_len - 1,
2348 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2351 search_start = extent_map_end(em);
2352 free_extent_map(em);
2359 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2360 unsigned int extra_bits,
2361 struct extent_state **cached_state)
2363 WARN_ON(PAGE_ALIGNED(end));
2365 if (start >= i_size_read(&inode->vfs_inode) &&
2366 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2368 * There can't be any extents following eof in this case so just
2369 * set the delalloc new bit for the range directly.
2371 extra_bits |= EXTENT_DELALLOC_NEW;
2375 ret = btrfs_find_new_delalloc_bytes(inode, start,
2382 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2386 /* see btrfs_writepage_start_hook for details on why this is required */
2387 struct btrfs_writepage_fixup {
2389 struct inode *inode;
2390 struct btrfs_work work;
2393 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2395 struct btrfs_writepage_fixup *fixup;
2396 struct btrfs_ordered_extent *ordered;
2397 struct extent_state *cached_state = NULL;
2398 struct extent_changeset *data_reserved = NULL;
2400 struct btrfs_inode *inode;
2404 bool free_delalloc_space = true;
2406 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2408 inode = BTRFS_I(fixup->inode);
2409 page_start = page_offset(page);
2410 page_end = page_offset(page) + PAGE_SIZE - 1;
2413 * This is similar to page_mkwrite, we need to reserve the space before
2414 * we take the page lock.
2416 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2422 * Before we queued this fixup, we took a reference on the page.
2423 * page->mapping may go NULL, but it shouldn't be moved to a different
2426 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2428 * Unfortunately this is a little tricky, either
2430 * 1) We got here and our page had already been dealt with and
2431 * we reserved our space, thus ret == 0, so we need to just
2432 * drop our space reservation and bail. This can happen the
2433 * first time we come into the fixup worker, or could happen
2434 * while waiting for the ordered extent.
2435 * 2) Our page was already dealt with, but we happened to get an
2436 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2437 * this case we obviously don't have anything to release, but
2438 * because the page was already dealt with we don't want to
2439 * mark the page with an error, so make sure we're resetting
2440 * ret to 0. This is why we have this check _before_ the ret
2441 * check, because we do not want to have a surprise ENOSPC
2442 * when the page was already properly dealt with.
2445 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2446 btrfs_delalloc_release_space(inode, data_reserved,
2447 page_start, PAGE_SIZE,
2455 * We can't mess with the page state unless it is locked, so now that
2456 * it is locked bail if we failed to make our space reservation.
2461 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2463 /* already ordered? We're done */
2464 if (PagePrivate2(page))
2467 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2469 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2472 btrfs_start_ordered_extent(ordered, 1);
2473 btrfs_put_ordered_extent(ordered);
2477 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2483 * Everything went as planned, we're now the owner of a dirty page with
2484 * delayed allocation bits set and space reserved for our COW
2487 * The page was dirty when we started, nothing should have cleaned it.
2489 BUG_ON(!PageDirty(page));
2490 free_delalloc_space = false;
2492 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2493 if (free_delalloc_space)
2494 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2496 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2501 * We hit ENOSPC or other errors. Update the mapping and page
2502 * to reflect the errors and clean the page.
2504 mapping_set_error(page->mapping, ret);
2505 end_extent_writepage(page, ret, page_start, page_end);
2506 clear_page_dirty_for_io(page);
2509 ClearPageChecked(page);
2513 extent_changeset_free(data_reserved);
2515 * As a precaution, do a delayed iput in case it would be the last iput
2516 * that could need flushing space. Recursing back to fixup worker would
2519 btrfs_add_delayed_iput(&inode->vfs_inode);
2523 * There are a few paths in the higher layers of the kernel that directly
2524 * set the page dirty bit without asking the filesystem if it is a
2525 * good idea. This causes problems because we want to make sure COW
2526 * properly happens and the data=ordered rules are followed.
2528 * In our case any range that doesn't have the ORDERED bit set
2529 * hasn't been properly setup for IO. We kick off an async process
2530 * to fix it up. The async helper will wait for ordered extents, set
2531 * the delalloc bit and make it safe to write the page.
2533 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2535 struct inode *inode = page->mapping->host;
2536 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2537 struct btrfs_writepage_fixup *fixup;
2539 /* this page is properly in the ordered list */
2540 if (TestClearPagePrivate2(page))
2544 * PageChecked is set below when we create a fixup worker for this page,
2545 * don't try to create another one if we're already PageChecked()
2547 * The extent_io writepage code will redirty the page if we send back
2550 if (PageChecked(page))
2553 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2558 * We are already holding a reference to this inode from
2559 * write_cache_pages. We need to hold it because the space reservation
2560 * takes place outside of the page lock, and we can't trust
2561 * page->mapping outside of the page lock.
2564 SetPageChecked(page);
2566 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2568 fixup->inode = inode;
2569 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2574 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2575 struct btrfs_inode *inode, u64 file_pos,
2576 struct btrfs_file_extent_item *stack_fi,
2577 const bool update_inode_bytes,
2578 u64 qgroup_reserved)
2580 struct btrfs_root *root = inode->root;
2581 const u64 sectorsize = root->fs_info->sectorsize;
2582 struct btrfs_path *path;
2583 struct extent_buffer *leaf;
2584 struct btrfs_key ins;
2585 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2586 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2587 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2588 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2589 struct btrfs_drop_extents_args drop_args = { 0 };
2592 path = btrfs_alloc_path();
2597 * we may be replacing one extent in the tree with another.
2598 * The new extent is pinned in the extent map, and we don't want
2599 * to drop it from the cache until it is completely in the btree.
2601 * So, tell btrfs_drop_extents to leave this extent in the cache.
2602 * the caller is expected to unpin it and allow it to be merged
2605 drop_args.path = path;
2606 drop_args.start = file_pos;
2607 drop_args.end = file_pos + num_bytes;
2608 drop_args.replace_extent = true;
2609 drop_args.extent_item_size = sizeof(*stack_fi);
2610 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2614 if (!drop_args.extent_inserted) {
2615 ins.objectid = btrfs_ino(inode);
2616 ins.offset = file_pos;
2617 ins.type = BTRFS_EXTENT_DATA_KEY;
2619 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2624 leaf = path->nodes[0];
2625 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2626 write_extent_buffer(leaf, stack_fi,
2627 btrfs_item_ptr_offset(leaf, path->slots[0]),
2628 sizeof(struct btrfs_file_extent_item));
2630 btrfs_mark_buffer_dirty(leaf);
2631 btrfs_release_path(path);
2634 * If we dropped an inline extent here, we know the range where it is
2635 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2636 * number of bytes only for that range contaning the inline extent.
2637 * The remaining of the range will be processed when clearning the
2638 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2640 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2641 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2643 inline_size = drop_args.bytes_found - inline_size;
2644 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2645 drop_args.bytes_found -= inline_size;
2646 num_bytes -= sectorsize;
2649 if (update_inode_bytes)
2650 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2652 ins.objectid = disk_bytenr;
2653 ins.offset = disk_num_bytes;
2654 ins.type = BTRFS_EXTENT_ITEM_KEY;
2656 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2660 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2661 file_pos, qgroup_reserved, &ins);
2663 btrfs_free_path(path);
2668 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2671 struct btrfs_block_group *cache;
2673 cache = btrfs_lookup_block_group(fs_info, start);
2676 spin_lock(&cache->lock);
2677 cache->delalloc_bytes -= len;
2678 spin_unlock(&cache->lock);
2680 btrfs_put_block_group(cache);
2683 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2684 struct btrfs_ordered_extent *oe)
2686 struct btrfs_file_extent_item stack_fi;
2688 bool update_inode_bytes;
2690 memset(&stack_fi, 0, sizeof(stack_fi));
2691 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2692 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2693 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2694 oe->disk_num_bytes);
2695 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2696 logical_len = oe->truncated_len;
2698 logical_len = oe->num_bytes;
2699 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2700 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2701 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2702 /* Encryption and other encoding is reserved and all 0 */
2705 * For delalloc, when completing an ordered extent we update the inode's
2706 * bytes when clearing the range in the inode's io tree, so pass false
2707 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2708 * except if the ordered extent was truncated.
2710 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2711 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2713 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2714 oe->file_offset, &stack_fi,
2715 update_inode_bytes, oe->qgroup_rsv);
2719 * As ordered data IO finishes, this gets called so we can finish
2720 * an ordered extent if the range of bytes in the file it covers are
2723 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2725 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2726 struct btrfs_root *root = inode->root;
2727 struct btrfs_fs_info *fs_info = root->fs_info;
2728 struct btrfs_trans_handle *trans = NULL;
2729 struct extent_io_tree *io_tree = &inode->io_tree;
2730 struct extent_state *cached_state = NULL;
2732 int compress_type = 0;
2734 u64 logical_len = ordered_extent->num_bytes;
2735 bool freespace_inode;
2736 bool truncated = false;
2737 bool clear_reserved_extent = true;
2738 unsigned int clear_bits = EXTENT_DEFRAG;
2740 start = ordered_extent->file_offset;
2741 end = start + ordered_extent->num_bytes - 1;
2743 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2744 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2745 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2746 clear_bits |= EXTENT_DELALLOC_NEW;
2748 freespace_inode = btrfs_is_free_space_inode(inode);
2750 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2755 btrfs_free_io_failure_record(inode, start, end);
2757 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2759 logical_len = ordered_extent->truncated_len;
2760 /* Truncated the entire extent, don't bother adding */
2765 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2766 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2768 btrfs_inode_safe_disk_i_size_write(inode, 0);
2769 if (freespace_inode)
2770 trans = btrfs_join_transaction_spacecache(root);
2772 trans = btrfs_join_transaction(root);
2773 if (IS_ERR(trans)) {
2774 ret = PTR_ERR(trans);
2778 trans->block_rsv = &inode->block_rsv;
2779 ret = btrfs_update_inode_fallback(trans, root, inode);
2780 if (ret) /* -ENOMEM or corruption */
2781 btrfs_abort_transaction(trans, ret);
2785 clear_bits |= EXTENT_LOCKED;
2786 lock_extent_bits(io_tree, start, end, &cached_state);
2788 if (freespace_inode)
2789 trans = btrfs_join_transaction_spacecache(root);
2791 trans = btrfs_join_transaction(root);
2792 if (IS_ERR(trans)) {
2793 ret = PTR_ERR(trans);
2798 trans->block_rsv = &inode->block_rsv;
2800 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2801 compress_type = ordered_extent->compress_type;
2802 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2803 BUG_ON(compress_type);
2804 ret = btrfs_mark_extent_written(trans, inode,
2805 ordered_extent->file_offset,
2806 ordered_extent->file_offset +
2809 BUG_ON(root == fs_info->tree_root);
2810 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2812 clear_reserved_extent = false;
2813 btrfs_release_delalloc_bytes(fs_info,
2814 ordered_extent->disk_bytenr,
2815 ordered_extent->disk_num_bytes);
2818 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
2819 ordered_extent->num_bytes, trans->transid);
2821 btrfs_abort_transaction(trans, ret);
2825 ret = add_pending_csums(trans, &ordered_extent->list);
2827 btrfs_abort_transaction(trans, ret);
2832 * If this is a new delalloc range, clear its new delalloc flag to
2833 * update the inode's number of bytes. This needs to be done first
2834 * before updating the inode item.
2836 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
2837 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
2838 clear_extent_bit(&inode->io_tree, start, end,
2839 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
2840 0, 0, &cached_state);
2842 btrfs_inode_safe_disk_i_size_write(inode, 0);
2843 ret = btrfs_update_inode_fallback(trans, root, inode);
2844 if (ret) { /* -ENOMEM or corruption */
2845 btrfs_abort_transaction(trans, ret);
2850 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
2851 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2855 btrfs_end_transaction(trans);
2857 if (ret || truncated) {
2858 u64 unwritten_start = start;
2861 unwritten_start += logical_len;
2862 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2864 /* Drop the cache for the part of the extent we didn't write. */
2865 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
2868 * If the ordered extent had an IOERR or something else went
2869 * wrong we need to return the space for this ordered extent
2870 * back to the allocator. We only free the extent in the
2871 * truncated case if we didn't write out the extent at all.
2873 * If we made it past insert_reserved_file_extent before we
2874 * errored out then we don't need to do this as the accounting
2875 * has already been done.
2877 if ((ret || !logical_len) &&
2878 clear_reserved_extent &&
2879 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2880 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2882 * Discard the range before returning it back to the
2885 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
2886 btrfs_discard_extent(fs_info,
2887 ordered_extent->disk_bytenr,
2888 ordered_extent->disk_num_bytes,
2890 btrfs_free_reserved_extent(fs_info,
2891 ordered_extent->disk_bytenr,
2892 ordered_extent->disk_num_bytes, 1);
2897 * This needs to be done to make sure anybody waiting knows we are done
2898 * updating everything for this ordered extent.
2900 btrfs_remove_ordered_extent(inode, ordered_extent);
2903 btrfs_put_ordered_extent(ordered_extent);
2904 /* once for the tree */
2905 btrfs_put_ordered_extent(ordered_extent);
2910 static void finish_ordered_fn(struct btrfs_work *work)
2912 struct btrfs_ordered_extent *ordered_extent;
2913 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2914 btrfs_finish_ordered_io(ordered_extent);
2917 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
2918 u64 end, int uptodate)
2920 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2921 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2922 struct btrfs_ordered_extent *ordered_extent = NULL;
2923 struct btrfs_workqueue *wq;
2925 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2927 ClearPagePrivate2(page);
2928 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2929 end - start + 1, uptodate))
2932 if (btrfs_is_free_space_inode(inode))
2933 wq = fs_info->endio_freespace_worker;
2935 wq = fs_info->endio_write_workers;
2937 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2938 btrfs_queue_work(wq, &ordered_extent->work);
2942 * check_data_csum - verify checksum of one sector of uncompressed data
2944 * @io_bio: btrfs_io_bio which contains the csum
2945 * @bio_offset: offset to the beginning of the bio (in bytes)
2946 * @page: page where is the data to be verified
2947 * @pgoff: offset inside the page
2949 * The length of such check is always one sector size.
2951 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
2952 u32 bio_offset, struct page *page, u32 pgoff)
2954 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2955 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
2957 u32 len = fs_info->sectorsize;
2958 const u32 csum_size = fs_info->csum_size;
2959 unsigned int offset_sectors;
2961 u8 csum[BTRFS_CSUM_SIZE];
2963 ASSERT(pgoff + len <= PAGE_SIZE);
2965 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
2966 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
2968 kaddr = kmap_atomic(page);
2969 shash->tfm = fs_info->csum_shash;
2971 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
2973 if (memcmp(csum, csum_expected, csum_size))
2976 kunmap_atomic(kaddr);
2979 btrfs_print_data_csum_error(BTRFS_I(inode), page_offset(page) + pgoff,
2980 csum, csum_expected, io_bio->mirror_num);
2982 btrfs_dev_stat_inc_and_print(io_bio->device,
2983 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2984 memset(kaddr + pgoff, 1, len);
2985 flush_dcache_page(page);
2986 kunmap_atomic(kaddr);
2991 * When reads are done, we need to check csums to verify the data is correct.
2992 * if there's a match, we allow the bio to finish. If not, the code in
2993 * extent_io.c will try to find good copies for us.
2995 * @bio_offset: offset to the beginning of the bio (in bytes)
2996 * @start: file offset of the range start
2997 * @end: file offset of the range end (inclusive)
2998 * @mirror: mirror number
3000 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3001 struct page *page, u64 start, u64 end, int mirror)
3003 struct inode *inode = page->mapping->host;
3004 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3005 struct btrfs_root *root = BTRFS_I(inode)->root;
3006 const u32 sectorsize = root->fs_info->sectorsize;
3009 if (PageChecked(page)) {
3010 ClearPageChecked(page);
3014 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3017 if (!root->fs_info->csum_root)
3020 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3021 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3022 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3026 ASSERT(page_offset(page) <= start &&
3027 end <= page_offset(page) + PAGE_SIZE - 1);
3028 for (pg_off = offset_in_page(start);
3029 pg_off < offset_in_page(end);
3030 pg_off += sectorsize, bio_offset += sectorsize) {
3033 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off);
3041 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3043 * @inode: The inode we want to perform iput on
3045 * This function uses the generic vfs_inode::i_count to track whether we should
3046 * just decrement it (in case it's > 1) or if this is the last iput then link
3047 * the inode to the delayed iput machinery. Delayed iputs are processed at
3048 * transaction commit time/superblock commit/cleaner kthread.
3050 void btrfs_add_delayed_iput(struct inode *inode)
3052 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3053 struct btrfs_inode *binode = BTRFS_I(inode);
3055 if (atomic_add_unless(&inode->i_count, -1, 1))
3058 atomic_inc(&fs_info->nr_delayed_iputs);
3059 spin_lock(&fs_info->delayed_iput_lock);
3060 ASSERT(list_empty(&binode->delayed_iput));
3061 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3062 spin_unlock(&fs_info->delayed_iput_lock);
3063 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3064 wake_up_process(fs_info->cleaner_kthread);
3067 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3068 struct btrfs_inode *inode)
3070 list_del_init(&inode->delayed_iput);
3071 spin_unlock(&fs_info->delayed_iput_lock);
3072 iput(&inode->vfs_inode);
3073 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3074 wake_up(&fs_info->delayed_iputs_wait);
3075 spin_lock(&fs_info->delayed_iput_lock);
3078 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3079 struct btrfs_inode *inode)
3081 if (!list_empty(&inode->delayed_iput)) {
3082 spin_lock(&fs_info->delayed_iput_lock);
3083 if (!list_empty(&inode->delayed_iput))
3084 run_delayed_iput_locked(fs_info, inode);
3085 spin_unlock(&fs_info->delayed_iput_lock);
3089 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3092 spin_lock(&fs_info->delayed_iput_lock);
3093 while (!list_empty(&fs_info->delayed_iputs)) {
3094 struct btrfs_inode *inode;
3096 inode = list_first_entry(&fs_info->delayed_iputs,
3097 struct btrfs_inode, delayed_iput);
3098 run_delayed_iput_locked(fs_info, inode);
3100 spin_unlock(&fs_info->delayed_iput_lock);
3104 * Wait for flushing all delayed iputs
3106 * @fs_info: the filesystem
3108 * This will wait on any delayed iputs that are currently running with KILLABLE
3109 * set. Once they are all done running we will return, unless we are killed in
3110 * which case we return EINTR. This helps in user operations like fallocate etc
3111 * that might get blocked on the iputs.
3113 * Return EINTR if we were killed, 0 if nothing's pending
3115 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3117 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3118 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3125 * This creates an orphan entry for the given inode in case something goes wrong
3126 * in the middle of an unlink.
3128 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3129 struct btrfs_inode *inode)
3133 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3134 if (ret && ret != -EEXIST) {
3135 btrfs_abort_transaction(trans, ret);
3143 * We have done the delete so we can go ahead and remove the orphan item for
3144 * this particular inode.
3146 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3147 struct btrfs_inode *inode)
3149 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3153 * this cleans up any orphans that may be left on the list from the last use
3156 int btrfs_orphan_cleanup(struct btrfs_root *root)
3158 struct btrfs_fs_info *fs_info = root->fs_info;
3159 struct btrfs_path *path;
3160 struct extent_buffer *leaf;
3161 struct btrfs_key key, found_key;
3162 struct btrfs_trans_handle *trans;
3163 struct inode *inode;
3164 u64 last_objectid = 0;
3165 int ret = 0, nr_unlink = 0;
3167 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3170 path = btrfs_alloc_path();
3175 path->reada = READA_BACK;
3177 key.objectid = BTRFS_ORPHAN_OBJECTID;
3178 key.type = BTRFS_ORPHAN_ITEM_KEY;
3179 key.offset = (u64)-1;
3182 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3187 * if ret == 0 means we found what we were searching for, which
3188 * is weird, but possible, so only screw with path if we didn't
3189 * find the key and see if we have stuff that matches
3193 if (path->slots[0] == 0)
3198 /* pull out the item */
3199 leaf = path->nodes[0];
3200 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3202 /* make sure the item matches what we want */
3203 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3205 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3208 /* release the path since we're done with it */
3209 btrfs_release_path(path);
3212 * this is where we are basically btrfs_lookup, without the
3213 * crossing root thing. we store the inode number in the
3214 * offset of the orphan item.
3217 if (found_key.offset == last_objectid) {
3219 "Error removing orphan entry, stopping orphan cleanup");
3224 last_objectid = found_key.offset;
3226 found_key.objectid = found_key.offset;
3227 found_key.type = BTRFS_INODE_ITEM_KEY;
3228 found_key.offset = 0;
3229 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3230 ret = PTR_ERR_OR_ZERO(inode);
3231 if (ret && ret != -ENOENT)
3234 if (ret == -ENOENT && root == fs_info->tree_root) {
3235 struct btrfs_root *dead_root;
3236 int is_dead_root = 0;
3239 * this is an orphan in the tree root. Currently these
3240 * could come from 2 sources:
3241 * a) a snapshot deletion in progress
3242 * b) a free space cache inode
3243 * We need to distinguish those two, as the snapshot
3244 * orphan must not get deleted.
3245 * find_dead_roots already ran before us, so if this
3246 * is a snapshot deletion, we should find the root
3247 * in the fs_roots radix tree.
3250 spin_lock(&fs_info->fs_roots_radix_lock);
3251 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3252 (unsigned long)found_key.objectid);
3253 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3255 spin_unlock(&fs_info->fs_roots_radix_lock);
3258 /* prevent this orphan from being found again */
3259 key.offset = found_key.objectid - 1;
3266 * If we have an inode with links, there are a couple of
3267 * possibilities. Old kernels (before v3.12) used to create an
3268 * orphan item for truncate indicating that there were possibly
3269 * extent items past i_size that needed to be deleted. In v3.12,
3270 * truncate was changed to update i_size in sync with the extent
3271 * items, but the (useless) orphan item was still created. Since
3272 * v4.18, we don't create the orphan item for truncate at all.
3274 * So, this item could mean that we need to do a truncate, but
3275 * only if this filesystem was last used on a pre-v3.12 kernel
3276 * and was not cleanly unmounted. The odds of that are quite
3277 * slim, and it's a pain to do the truncate now, so just delete
3280 * It's also possible that this orphan item was supposed to be
3281 * deleted but wasn't. The inode number may have been reused,
3282 * but either way, we can delete the orphan item.
3284 if (ret == -ENOENT || inode->i_nlink) {
3287 trans = btrfs_start_transaction(root, 1);
3288 if (IS_ERR(trans)) {
3289 ret = PTR_ERR(trans);
3292 btrfs_debug(fs_info, "auto deleting %Lu",
3293 found_key.objectid);
3294 ret = btrfs_del_orphan_item(trans, root,
3295 found_key.objectid);
3296 btrfs_end_transaction(trans);
3304 /* this will do delete_inode and everything for us */
3307 /* release the path since we're done with it */
3308 btrfs_release_path(path);
3310 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3312 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3313 trans = btrfs_join_transaction(root);
3315 btrfs_end_transaction(trans);
3319 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3323 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3324 btrfs_free_path(path);
3329 * very simple check to peek ahead in the leaf looking for xattrs. If we
3330 * don't find any xattrs, we know there can't be any acls.
3332 * slot is the slot the inode is in, objectid is the objectid of the inode
3334 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3335 int slot, u64 objectid,
3336 int *first_xattr_slot)
3338 u32 nritems = btrfs_header_nritems(leaf);
3339 struct btrfs_key found_key;
3340 static u64 xattr_access = 0;
3341 static u64 xattr_default = 0;
3344 if (!xattr_access) {
3345 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3346 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3347 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3348 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3352 *first_xattr_slot = -1;
3353 while (slot < nritems) {
3354 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3356 /* we found a different objectid, there must not be acls */
3357 if (found_key.objectid != objectid)
3360 /* we found an xattr, assume we've got an acl */
3361 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3362 if (*first_xattr_slot == -1)
3363 *first_xattr_slot = slot;
3364 if (found_key.offset == xattr_access ||
3365 found_key.offset == xattr_default)
3370 * we found a key greater than an xattr key, there can't
3371 * be any acls later on
3373 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3380 * it goes inode, inode backrefs, xattrs, extents,
3381 * so if there are a ton of hard links to an inode there can
3382 * be a lot of backrefs. Don't waste time searching too hard,
3383 * this is just an optimization
3388 /* we hit the end of the leaf before we found an xattr or
3389 * something larger than an xattr. We have to assume the inode
3392 if (*first_xattr_slot == -1)
3393 *first_xattr_slot = slot;
3398 * read an inode from the btree into the in-memory inode
3400 static int btrfs_read_locked_inode(struct inode *inode,
3401 struct btrfs_path *in_path)
3403 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3404 struct btrfs_path *path = in_path;
3405 struct extent_buffer *leaf;
3406 struct btrfs_inode_item *inode_item;
3407 struct btrfs_root *root = BTRFS_I(inode)->root;
3408 struct btrfs_key location;
3413 bool filled = false;
3414 int first_xattr_slot;
3416 ret = btrfs_fill_inode(inode, &rdev);
3421 path = btrfs_alloc_path();
3426 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3428 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3430 if (path != in_path)
3431 btrfs_free_path(path);
3435 leaf = path->nodes[0];
3440 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3441 struct btrfs_inode_item);
3442 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3443 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3444 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3445 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3446 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3447 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3448 round_up(i_size_read(inode), fs_info->sectorsize));
3450 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3451 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3453 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3454 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3456 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3457 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3459 BTRFS_I(inode)->i_otime.tv_sec =
3460 btrfs_timespec_sec(leaf, &inode_item->otime);
3461 BTRFS_I(inode)->i_otime.tv_nsec =
3462 btrfs_timespec_nsec(leaf, &inode_item->otime);
3464 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3465 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3466 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3468 inode_set_iversion_queried(inode,
3469 btrfs_inode_sequence(leaf, inode_item));
3470 inode->i_generation = BTRFS_I(inode)->generation;
3472 rdev = btrfs_inode_rdev(leaf, inode_item);
3474 BTRFS_I(inode)->index_cnt = (u64)-1;
3475 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3479 * If we were modified in the current generation and evicted from memory
3480 * and then re-read we need to do a full sync since we don't have any
3481 * idea about which extents were modified before we were evicted from
3484 * This is required for both inode re-read from disk and delayed inode
3485 * in delayed_nodes_tree.
3487 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3488 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3489 &BTRFS_I(inode)->runtime_flags);
3492 * We don't persist the id of the transaction where an unlink operation
3493 * against the inode was last made. So here we assume the inode might
3494 * have been evicted, and therefore the exact value of last_unlink_trans
3495 * lost, and set it to last_trans to avoid metadata inconsistencies
3496 * between the inode and its parent if the inode is fsync'ed and the log
3497 * replayed. For example, in the scenario:
3500 * ln mydir/foo mydir/bar
3503 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3504 * xfs_io -c fsync mydir/foo
3506 * mount fs, triggers fsync log replay
3508 * We must make sure that when we fsync our inode foo we also log its
3509 * parent inode, otherwise after log replay the parent still has the
3510 * dentry with the "bar" name but our inode foo has a link count of 1
3511 * and doesn't have an inode ref with the name "bar" anymore.
3513 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3514 * but it guarantees correctness at the expense of occasional full
3515 * transaction commits on fsync if our inode is a directory, or if our
3516 * inode is not a directory, logging its parent unnecessarily.
3518 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3521 * Same logic as for last_unlink_trans. We don't persist the generation
3522 * of the last transaction where this inode was used for a reflink
3523 * operation, so after eviction and reloading the inode we must be
3524 * pessimistic and assume the last transaction that modified the inode.
3526 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3529 if (inode->i_nlink != 1 ||
3530 path->slots[0] >= btrfs_header_nritems(leaf))
3533 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3534 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3537 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3538 if (location.type == BTRFS_INODE_REF_KEY) {
3539 struct btrfs_inode_ref *ref;
3541 ref = (struct btrfs_inode_ref *)ptr;
3542 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3543 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3544 struct btrfs_inode_extref *extref;
3546 extref = (struct btrfs_inode_extref *)ptr;
3547 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3552 * try to precache a NULL acl entry for files that don't have
3553 * any xattrs or acls
3555 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3556 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3557 if (first_xattr_slot != -1) {
3558 path->slots[0] = first_xattr_slot;
3559 ret = btrfs_load_inode_props(inode, path);
3562 "error loading props for ino %llu (root %llu): %d",
3563 btrfs_ino(BTRFS_I(inode)),
3564 root->root_key.objectid, ret);
3566 if (path != in_path)
3567 btrfs_free_path(path);
3570 cache_no_acl(inode);
3572 switch (inode->i_mode & S_IFMT) {
3574 inode->i_mapping->a_ops = &btrfs_aops;
3575 inode->i_fop = &btrfs_file_operations;
3576 inode->i_op = &btrfs_file_inode_operations;
3579 inode->i_fop = &btrfs_dir_file_operations;
3580 inode->i_op = &btrfs_dir_inode_operations;
3583 inode->i_op = &btrfs_symlink_inode_operations;
3584 inode_nohighmem(inode);
3585 inode->i_mapping->a_ops = &btrfs_aops;
3588 inode->i_op = &btrfs_special_inode_operations;
3589 init_special_inode(inode, inode->i_mode, rdev);
3593 btrfs_sync_inode_flags_to_i_flags(inode);
3598 * given a leaf and an inode, copy the inode fields into the leaf
3600 static void fill_inode_item(struct btrfs_trans_handle *trans,
3601 struct extent_buffer *leaf,
3602 struct btrfs_inode_item *item,
3603 struct inode *inode)
3605 struct btrfs_map_token token;
3607 btrfs_init_map_token(&token, leaf);
3609 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3610 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3611 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3612 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3613 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3615 btrfs_set_token_timespec_sec(&token, &item->atime,
3616 inode->i_atime.tv_sec);
3617 btrfs_set_token_timespec_nsec(&token, &item->atime,
3618 inode->i_atime.tv_nsec);
3620 btrfs_set_token_timespec_sec(&token, &item->mtime,
3621 inode->i_mtime.tv_sec);
3622 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3623 inode->i_mtime.tv_nsec);
3625 btrfs_set_token_timespec_sec(&token, &item->ctime,
3626 inode->i_ctime.tv_sec);
3627 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3628 inode->i_ctime.tv_nsec);
3630 btrfs_set_token_timespec_sec(&token, &item->otime,
3631 BTRFS_I(inode)->i_otime.tv_sec);
3632 btrfs_set_token_timespec_nsec(&token, &item->otime,
3633 BTRFS_I(inode)->i_otime.tv_nsec);
3635 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3636 btrfs_set_token_inode_generation(&token, item,
3637 BTRFS_I(inode)->generation);
3638 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3639 btrfs_set_token_inode_transid(&token, item, trans->transid);
3640 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3641 btrfs_set_token_inode_flags(&token, item, BTRFS_I(inode)->flags);
3642 btrfs_set_token_inode_block_group(&token, item, 0);
3646 * copy everything in the in-memory inode into the btree.
3648 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3649 struct btrfs_root *root,
3650 struct btrfs_inode *inode)
3652 struct btrfs_inode_item *inode_item;
3653 struct btrfs_path *path;
3654 struct extent_buffer *leaf;
3657 path = btrfs_alloc_path();
3661 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3668 leaf = path->nodes[0];
3669 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3670 struct btrfs_inode_item);
3672 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3673 btrfs_mark_buffer_dirty(leaf);
3674 btrfs_set_inode_last_trans(trans, inode);
3677 btrfs_free_path(path);
3682 * copy everything in the in-memory inode into the btree.
3684 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3685 struct btrfs_root *root,
3686 struct btrfs_inode *inode)
3688 struct btrfs_fs_info *fs_info = root->fs_info;
3692 * If the inode is a free space inode, we can deadlock during commit
3693 * if we put it into the delayed code.
3695 * The data relocation inode should also be directly updated
3698 if (!btrfs_is_free_space_inode(inode)
3699 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3700 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3701 btrfs_update_root_times(trans, root);
3703 ret = btrfs_delayed_update_inode(trans, root, inode);
3705 btrfs_set_inode_last_trans(trans, inode);
3709 return btrfs_update_inode_item(trans, root, inode);
3712 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3713 struct btrfs_root *root, struct btrfs_inode *inode)
3717 ret = btrfs_update_inode(trans, root, inode);
3719 return btrfs_update_inode_item(trans, root, inode);
3724 * unlink helper that gets used here in inode.c and in the tree logging
3725 * recovery code. It remove a link in a directory with a given name, and
3726 * also drops the back refs in the inode to the directory
3728 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3729 struct btrfs_root *root,
3730 struct btrfs_inode *dir,
3731 struct btrfs_inode *inode,
3732 const char *name, int name_len)
3734 struct btrfs_fs_info *fs_info = root->fs_info;
3735 struct btrfs_path *path;
3737 struct btrfs_dir_item *di;
3739 u64 ino = btrfs_ino(inode);
3740 u64 dir_ino = btrfs_ino(dir);
3742 path = btrfs_alloc_path();
3748 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3749 name, name_len, -1);
3750 if (IS_ERR_OR_NULL(di)) {
3751 ret = di ? PTR_ERR(di) : -ENOENT;
3754 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3757 btrfs_release_path(path);
3760 * If we don't have dir index, we have to get it by looking up
3761 * the inode ref, since we get the inode ref, remove it directly,
3762 * it is unnecessary to do delayed deletion.
3764 * But if we have dir index, needn't search inode ref to get it.
3765 * Since the inode ref is close to the inode item, it is better
3766 * that we delay to delete it, and just do this deletion when
3767 * we update the inode item.
3769 if (inode->dir_index) {
3770 ret = btrfs_delayed_delete_inode_ref(inode);
3772 index = inode->dir_index;
3777 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3781 "failed to delete reference to %.*s, inode %llu parent %llu",
3782 name_len, name, ino, dir_ino);
3783 btrfs_abort_transaction(trans, ret);
3787 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3789 btrfs_abort_transaction(trans, ret);
3793 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3795 if (ret != 0 && ret != -ENOENT) {
3796 btrfs_abort_transaction(trans, ret);
3800 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3805 btrfs_abort_transaction(trans, ret);
3808 * If we have a pending delayed iput we could end up with the final iput
3809 * being run in btrfs-cleaner context. If we have enough of these built
3810 * up we can end up burning a lot of time in btrfs-cleaner without any
3811 * way to throttle the unlinks. Since we're currently holding a ref on
3812 * the inode we can run the delayed iput here without any issues as the
3813 * final iput won't be done until after we drop the ref we're currently
3816 btrfs_run_delayed_iput(fs_info, inode);
3818 btrfs_free_path(path);
3822 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3823 inode_inc_iversion(&inode->vfs_inode);
3824 inode_inc_iversion(&dir->vfs_inode);
3825 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3826 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3827 ret = btrfs_update_inode(trans, root, dir);
3832 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3833 struct btrfs_root *root,
3834 struct btrfs_inode *dir, struct btrfs_inode *inode,
3835 const char *name, int name_len)
3838 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3840 drop_nlink(&inode->vfs_inode);
3841 ret = btrfs_update_inode(trans, root, inode);
3847 * helper to start transaction for unlink and rmdir.
3849 * unlink and rmdir are special in btrfs, they do not always free space, so
3850 * if we cannot make our reservations the normal way try and see if there is
3851 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3852 * allow the unlink to occur.
3854 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3856 struct btrfs_root *root = BTRFS_I(dir)->root;
3859 * 1 for the possible orphan item
3860 * 1 for the dir item
3861 * 1 for the dir index
3862 * 1 for the inode ref
3865 return btrfs_start_transaction_fallback_global_rsv(root, 5);
3868 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3870 struct btrfs_root *root = BTRFS_I(dir)->root;
3871 struct btrfs_trans_handle *trans;
3872 struct inode *inode = d_inode(dentry);
3875 trans = __unlink_start_trans(dir);
3877 return PTR_ERR(trans);
3879 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
3882 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
3883 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
3884 dentry->d_name.len);
3888 if (inode->i_nlink == 0) {
3889 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3895 btrfs_end_transaction(trans);
3896 btrfs_btree_balance_dirty(root->fs_info);
3900 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3901 struct inode *dir, struct dentry *dentry)
3903 struct btrfs_root *root = BTRFS_I(dir)->root;
3904 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
3905 struct btrfs_path *path;
3906 struct extent_buffer *leaf;
3907 struct btrfs_dir_item *di;
3908 struct btrfs_key key;
3909 const char *name = dentry->d_name.name;
3910 int name_len = dentry->d_name.len;
3914 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
3916 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
3917 objectid = inode->root->root_key.objectid;
3918 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3919 objectid = inode->location.objectid;
3925 path = btrfs_alloc_path();
3929 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3930 name, name_len, -1);
3931 if (IS_ERR_OR_NULL(di)) {
3932 ret = di ? PTR_ERR(di) : -ENOENT;
3936 leaf = path->nodes[0];
3937 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3938 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3939 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3941 btrfs_abort_transaction(trans, ret);
3944 btrfs_release_path(path);
3947 * This is a placeholder inode for a subvolume we didn't have a
3948 * reference to at the time of the snapshot creation. In the meantime
3949 * we could have renamed the real subvol link into our snapshot, so
3950 * depending on btrfs_del_root_ref to return -ENOENT here is incorret.
3951 * Instead simply lookup the dir_index_item for this entry so we can
3952 * remove it. Otherwise we know we have a ref to the root and we can
3953 * call btrfs_del_root_ref, and it _shouldn't_ fail.
3955 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
3956 di = btrfs_search_dir_index_item(root, path, dir_ino,
3958 if (IS_ERR_OR_NULL(di)) {
3963 btrfs_abort_transaction(trans, ret);
3967 leaf = path->nodes[0];
3968 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3970 btrfs_release_path(path);
3972 ret = btrfs_del_root_ref(trans, objectid,
3973 root->root_key.objectid, dir_ino,
3974 &index, name, name_len);
3976 btrfs_abort_transaction(trans, ret);
3981 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
3983 btrfs_abort_transaction(trans, ret);
3987 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
3988 inode_inc_iversion(dir);
3989 dir->i_mtime = dir->i_ctime = current_time(dir);
3990 ret = btrfs_update_inode_fallback(trans, root, BTRFS_I(dir));
3992 btrfs_abort_transaction(trans, ret);
3994 btrfs_free_path(path);
3999 * Helper to check if the subvolume references other subvolumes or if it's
4002 static noinline int may_destroy_subvol(struct btrfs_root *root)
4004 struct btrfs_fs_info *fs_info = root->fs_info;
4005 struct btrfs_path *path;
4006 struct btrfs_dir_item *di;
4007 struct btrfs_key key;
4011 path = btrfs_alloc_path();
4015 /* Make sure this root isn't set as the default subvol */
4016 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4017 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4018 dir_id, "default", 7, 0);
4019 if (di && !IS_ERR(di)) {
4020 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4021 if (key.objectid == root->root_key.objectid) {
4024 "deleting default subvolume %llu is not allowed",
4028 btrfs_release_path(path);
4031 key.objectid = root->root_key.objectid;
4032 key.type = BTRFS_ROOT_REF_KEY;
4033 key.offset = (u64)-1;
4035 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4041 if (path->slots[0] > 0) {
4043 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4044 if (key.objectid == root->root_key.objectid &&
4045 key.type == BTRFS_ROOT_REF_KEY)
4049 btrfs_free_path(path);
4053 /* Delete all dentries for inodes belonging to the root */
4054 static void btrfs_prune_dentries(struct btrfs_root *root)
4056 struct btrfs_fs_info *fs_info = root->fs_info;
4057 struct rb_node *node;
4058 struct rb_node *prev;
4059 struct btrfs_inode *entry;
4060 struct inode *inode;
4063 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4064 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4066 spin_lock(&root->inode_lock);
4068 node = root->inode_tree.rb_node;
4072 entry = rb_entry(node, struct btrfs_inode, rb_node);
4074 if (objectid < btrfs_ino(entry))
4075 node = node->rb_left;
4076 else if (objectid > btrfs_ino(entry))
4077 node = node->rb_right;
4083 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4084 if (objectid <= btrfs_ino(entry)) {
4088 prev = rb_next(prev);
4092 entry = rb_entry(node, struct btrfs_inode, rb_node);
4093 objectid = btrfs_ino(entry) + 1;
4094 inode = igrab(&entry->vfs_inode);
4096 spin_unlock(&root->inode_lock);
4097 if (atomic_read(&inode->i_count) > 1)
4098 d_prune_aliases(inode);
4100 * btrfs_drop_inode will have it removed from the inode
4101 * cache when its usage count hits zero.
4105 spin_lock(&root->inode_lock);
4109 if (cond_resched_lock(&root->inode_lock))
4112 node = rb_next(node);
4114 spin_unlock(&root->inode_lock);
4117 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4119 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4120 struct btrfs_root *root = BTRFS_I(dir)->root;
4121 struct inode *inode = d_inode(dentry);
4122 struct btrfs_root *dest = BTRFS_I(inode)->root;
4123 struct btrfs_trans_handle *trans;
4124 struct btrfs_block_rsv block_rsv;
4129 * Don't allow to delete a subvolume with send in progress. This is
4130 * inside the inode lock so the error handling that has to drop the bit
4131 * again is not run concurrently.
4133 spin_lock(&dest->root_item_lock);
4134 if (dest->send_in_progress) {
4135 spin_unlock(&dest->root_item_lock);
4137 "attempt to delete subvolume %llu during send",
4138 dest->root_key.objectid);
4141 root_flags = btrfs_root_flags(&dest->root_item);
4142 btrfs_set_root_flags(&dest->root_item,
4143 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4144 spin_unlock(&dest->root_item_lock);
4146 down_write(&fs_info->subvol_sem);
4148 ret = may_destroy_subvol(dest);
4152 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4154 * One for dir inode,
4155 * two for dir entries,
4156 * two for root ref/backref.
4158 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4162 trans = btrfs_start_transaction(root, 0);
4163 if (IS_ERR(trans)) {
4164 ret = PTR_ERR(trans);
4167 trans->block_rsv = &block_rsv;
4168 trans->bytes_reserved = block_rsv.size;
4170 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4172 ret = btrfs_unlink_subvol(trans, dir, dentry);
4174 btrfs_abort_transaction(trans, ret);
4178 btrfs_record_root_in_trans(trans, dest);
4180 memset(&dest->root_item.drop_progress, 0,
4181 sizeof(dest->root_item.drop_progress));
4182 btrfs_set_root_drop_level(&dest->root_item, 0);
4183 btrfs_set_root_refs(&dest->root_item, 0);
4185 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4186 ret = btrfs_insert_orphan_item(trans,
4188 dest->root_key.objectid);
4190 btrfs_abort_transaction(trans, ret);
4195 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4196 BTRFS_UUID_KEY_SUBVOL,
4197 dest->root_key.objectid);
4198 if (ret && ret != -ENOENT) {
4199 btrfs_abort_transaction(trans, ret);
4202 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4203 ret = btrfs_uuid_tree_remove(trans,
4204 dest->root_item.received_uuid,
4205 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4206 dest->root_key.objectid);
4207 if (ret && ret != -ENOENT) {
4208 btrfs_abort_transaction(trans, ret);
4213 free_anon_bdev(dest->anon_dev);
4216 trans->block_rsv = NULL;
4217 trans->bytes_reserved = 0;
4218 ret = btrfs_end_transaction(trans);
4219 inode->i_flags |= S_DEAD;
4221 btrfs_subvolume_release_metadata(root, &block_rsv);
4223 up_write(&fs_info->subvol_sem);
4225 spin_lock(&dest->root_item_lock);
4226 root_flags = btrfs_root_flags(&dest->root_item);
4227 btrfs_set_root_flags(&dest->root_item,
4228 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4229 spin_unlock(&dest->root_item_lock);
4231 d_invalidate(dentry);
4232 btrfs_prune_dentries(dest);
4233 ASSERT(dest->send_in_progress == 0);
4239 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4241 struct inode *inode = d_inode(dentry);
4243 struct btrfs_root *root = BTRFS_I(dir)->root;
4244 struct btrfs_trans_handle *trans;
4245 u64 last_unlink_trans;
4247 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4249 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4250 return btrfs_delete_subvolume(dir, dentry);
4252 trans = __unlink_start_trans(dir);
4254 return PTR_ERR(trans);
4256 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4257 err = btrfs_unlink_subvol(trans, dir, dentry);
4261 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4265 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4267 /* now the directory is empty */
4268 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4269 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4270 dentry->d_name.len);
4272 btrfs_i_size_write(BTRFS_I(inode), 0);
4274 * Propagate the last_unlink_trans value of the deleted dir to
4275 * its parent directory. This is to prevent an unrecoverable
4276 * log tree in the case we do something like this:
4278 * 2) create snapshot under dir foo
4279 * 3) delete the snapshot
4282 * 6) fsync foo or some file inside foo
4284 if (last_unlink_trans >= trans->transid)
4285 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4288 btrfs_end_transaction(trans);
4289 btrfs_btree_balance_dirty(root->fs_info);
4295 * Return this if we need to call truncate_block for the last bit of the
4298 #define NEED_TRUNCATE_BLOCK 1
4301 * this can truncate away extent items, csum items and directory items.
4302 * It starts at a high offset and removes keys until it can't find
4303 * any higher than new_size
4305 * csum items that cross the new i_size are truncated to the new size
4308 * min_type is the minimum key type to truncate down to. If set to 0, this
4309 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4311 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4312 struct btrfs_root *root,
4313 struct btrfs_inode *inode,
4314 u64 new_size, u32 min_type)
4316 struct btrfs_fs_info *fs_info = root->fs_info;
4317 struct btrfs_path *path;
4318 struct extent_buffer *leaf;
4319 struct btrfs_file_extent_item *fi;
4320 struct btrfs_key key;
4321 struct btrfs_key found_key;
4322 u64 extent_start = 0;
4323 u64 extent_num_bytes = 0;
4324 u64 extent_offset = 0;
4326 u64 last_size = new_size;
4327 u32 found_type = (u8)-1;
4330 int pending_del_nr = 0;
4331 int pending_del_slot = 0;
4332 int extent_type = -1;
4334 u64 ino = btrfs_ino(inode);
4335 u64 bytes_deleted = 0;
4336 bool be_nice = false;
4337 bool should_throttle = false;
4338 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
4339 struct extent_state *cached_state = NULL;
4341 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4344 * For non-free space inodes and non-shareable roots, we want to back
4345 * off from time to time. This means all inodes in subvolume roots,
4346 * reloc roots, and data reloc roots.
4348 if (!btrfs_is_free_space_inode(inode) &&
4349 test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4352 path = btrfs_alloc_path();
4355 path->reada = READA_BACK;
4357 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4358 lock_extent_bits(&inode->io_tree, lock_start, (u64)-1,
4362 * We want to drop from the next block forward in case this
4363 * new size is not block aligned since we will be keeping the
4364 * last block of the extent just the way it is.
4366 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4367 fs_info->sectorsize),
4372 * This function is also used to drop the items in the log tree before
4373 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4374 * it is used to drop the logged items. So we shouldn't kill the delayed
4377 if (min_type == 0 && root == inode->root)
4378 btrfs_kill_delayed_inode_items(inode);
4381 key.offset = (u64)-1;
4386 * with a 16K leaf size and 128MB extents, you can actually queue
4387 * up a huge file in a single leaf. Most of the time that
4388 * bytes_deleted is > 0, it will be huge by the time we get here
4390 if (be_nice && bytes_deleted > SZ_32M &&
4391 btrfs_should_end_transaction(trans)) {
4396 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4402 /* there are no items in the tree for us to truncate, we're
4405 if (path->slots[0] == 0)
4411 u64 clear_start = 0, clear_len = 0;
4414 leaf = path->nodes[0];
4415 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4416 found_type = found_key.type;
4418 if (found_key.objectid != ino)
4421 if (found_type < min_type)
4424 item_end = found_key.offset;
4425 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4426 fi = btrfs_item_ptr(leaf, path->slots[0],
4427 struct btrfs_file_extent_item);
4428 extent_type = btrfs_file_extent_type(leaf, fi);
4429 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4431 btrfs_file_extent_num_bytes(leaf, fi);
4433 trace_btrfs_truncate_show_fi_regular(
4434 inode, leaf, fi, found_key.offset);
4435 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4436 item_end += btrfs_file_extent_ram_bytes(leaf,
4439 trace_btrfs_truncate_show_fi_inline(
4440 inode, leaf, fi, path->slots[0],
4445 if (found_type > min_type) {
4448 if (item_end < new_size)
4450 if (found_key.offset >= new_size)
4456 /* FIXME, shrink the extent if the ref count is only 1 */
4457 if (found_type != BTRFS_EXTENT_DATA_KEY)
4460 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4463 clear_start = found_key.offset;
4464 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4466 u64 orig_num_bytes =
4467 btrfs_file_extent_num_bytes(leaf, fi);
4468 extent_num_bytes = ALIGN(new_size -
4470 fs_info->sectorsize);
4471 clear_start = ALIGN(new_size, fs_info->sectorsize);
4472 btrfs_set_file_extent_num_bytes(leaf, fi,
4474 num_dec = (orig_num_bytes -
4476 if (test_bit(BTRFS_ROOT_SHAREABLE,
4479 inode_sub_bytes(&inode->vfs_inode,
4481 btrfs_mark_buffer_dirty(leaf);
4484 btrfs_file_extent_disk_num_bytes(leaf,
4486 extent_offset = found_key.offset -
4487 btrfs_file_extent_offset(leaf, fi);
4489 /* FIXME blocksize != 4096 */
4490 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4491 if (extent_start != 0) {
4493 if (test_bit(BTRFS_ROOT_SHAREABLE,
4495 inode_sub_bytes(&inode->vfs_inode,
4499 clear_len = num_dec;
4500 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4502 * we can't truncate inline items that have had
4506 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4507 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4508 btrfs_file_extent_compression(leaf, fi) == 0) {
4509 u32 size = (u32)(new_size - found_key.offset);
4511 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4512 size = btrfs_file_extent_calc_inline_size(size);
4513 btrfs_truncate_item(path, size, 1);
4514 } else if (!del_item) {
4516 * We have to bail so the last_size is set to
4517 * just before this extent.
4519 ret = NEED_TRUNCATE_BLOCK;
4523 * Inline extents are special, we just treat
4524 * them as a full sector worth in the file
4525 * extent tree just for simplicity sake.
4527 clear_len = fs_info->sectorsize;
4530 if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
4531 inode_sub_bytes(&inode->vfs_inode,
4532 item_end + 1 - new_size);
4536 * We use btrfs_truncate_inode_items() to clean up log trees for
4537 * multiple fsyncs, and in this case we don't want to clear the
4538 * file extent range because it's just the log.
4540 if (root == inode->root) {
4541 ret = btrfs_inode_clear_file_extent_range(inode,
4542 clear_start, clear_len);
4544 btrfs_abort_transaction(trans, ret);
4550 last_size = found_key.offset;
4552 last_size = new_size;
4554 if (!pending_del_nr) {
4555 /* no pending yet, add ourselves */
4556 pending_del_slot = path->slots[0];
4558 } else if (pending_del_nr &&
4559 path->slots[0] + 1 == pending_del_slot) {
4560 /* hop on the pending chunk */
4562 pending_del_slot = path->slots[0];
4569 should_throttle = false;
4572 root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4573 struct btrfs_ref ref = { 0 };
4575 bytes_deleted += extent_num_bytes;
4577 btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF,
4578 extent_start, extent_num_bytes, 0);
4579 ref.real_root = root->root_key.objectid;
4580 btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
4581 ino, extent_offset);
4582 ret = btrfs_free_extent(trans, &ref);
4584 btrfs_abort_transaction(trans, ret);
4588 if (btrfs_should_throttle_delayed_refs(trans))
4589 should_throttle = true;
4593 if (found_type == BTRFS_INODE_ITEM_KEY)
4596 if (path->slots[0] == 0 ||
4597 path->slots[0] != pending_del_slot ||
4599 if (pending_del_nr) {
4600 ret = btrfs_del_items(trans, root, path,
4604 btrfs_abort_transaction(trans, ret);
4609 btrfs_release_path(path);
4612 * We can generate a lot of delayed refs, so we need to
4613 * throttle every once and a while and make sure we're
4614 * adding enough space to keep up with the work we are
4615 * generating. Since we hold a transaction here we
4616 * can't flush, and we don't want to FLUSH_LIMIT because
4617 * we could have generated too many delayed refs to
4618 * actually allocate, so just bail if we're short and
4619 * let the normal reservation dance happen higher up.
4621 if (should_throttle) {
4622 ret = btrfs_delayed_refs_rsv_refill(fs_info,
4623 BTRFS_RESERVE_NO_FLUSH);
4635 if (ret >= 0 && pending_del_nr) {
4638 err = btrfs_del_items(trans, root, path, pending_del_slot,
4641 btrfs_abort_transaction(trans, err);
4645 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4646 ASSERT(last_size >= new_size);
4647 if (!ret && last_size > new_size)
4648 last_size = new_size;
4649 btrfs_inode_safe_disk_i_size_write(inode, last_size);
4650 unlock_extent_cached(&inode->io_tree, lock_start, (u64)-1,
4654 btrfs_free_path(path);
4659 * btrfs_truncate_block - read, zero a chunk and write a block
4660 * @inode - inode that we're zeroing
4661 * @from - the offset to start zeroing
4662 * @len - the length to zero, 0 to zero the entire range respective to the
4664 * @front - zero up to the offset instead of from the offset on
4666 * This will find the block for the "from" offset and cow the block and zero the
4667 * part we want to zero. This is used with truncate and hole punching.
4669 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4672 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4673 struct address_space *mapping = inode->vfs_inode.i_mapping;
4674 struct extent_io_tree *io_tree = &inode->io_tree;
4675 struct btrfs_ordered_extent *ordered;
4676 struct extent_state *cached_state = NULL;
4677 struct extent_changeset *data_reserved = NULL;
4679 bool only_release_metadata = false;
4680 u32 blocksize = fs_info->sectorsize;
4681 pgoff_t index = from >> PAGE_SHIFT;
4682 unsigned offset = from & (blocksize - 1);
4684 gfp_t mask = btrfs_alloc_write_mask(mapping);
4685 size_t write_bytes = blocksize;
4690 if (IS_ALIGNED(offset, blocksize) &&
4691 (!len || IS_ALIGNED(len, blocksize)))
4694 block_start = round_down(from, blocksize);
4695 block_end = block_start + blocksize - 1;
4697 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4700 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes) > 0) {
4701 /* For nocow case, no need to reserve data space */
4702 only_release_metadata = true;
4707 ret = btrfs_delalloc_reserve_metadata(inode, blocksize);
4709 if (!only_release_metadata)
4710 btrfs_free_reserved_data_space(inode, data_reserved,
4711 block_start, blocksize);
4715 page = find_or_create_page(mapping, index, mask);
4717 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4719 btrfs_delalloc_release_extents(inode, blocksize);
4723 ret = set_page_extent_mapped(page);
4727 if (!PageUptodate(page)) {
4728 ret = btrfs_readpage(NULL, page);
4730 if (page->mapping != mapping) {
4735 if (!PageUptodate(page)) {
4740 wait_on_page_writeback(page);
4742 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4744 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4746 unlock_extent_cached(io_tree, block_start, block_end,
4750 btrfs_start_ordered_extent(ordered, 1);
4751 btrfs_put_ordered_extent(ordered);
4755 clear_extent_bit(&inode->io_tree, block_start, block_end,
4756 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4757 0, 0, &cached_state);
4759 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4762 unlock_extent_cached(io_tree, block_start, block_end,
4767 if (offset != blocksize) {
4769 len = blocksize - offset;
4772 memset(kaddr + (block_start - page_offset(page)),
4775 memset(kaddr + (block_start - page_offset(page)) + offset,
4777 flush_dcache_page(page);
4780 ClearPageChecked(page);
4781 set_page_dirty(page);
4782 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4784 if (only_release_metadata)
4785 set_extent_bit(&inode->io_tree, block_start, block_end,
4786 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4790 if (only_release_metadata)
4791 btrfs_delalloc_release_metadata(inode, blocksize, true);
4793 btrfs_delalloc_release_space(inode, data_reserved,
4794 block_start, blocksize, true);
4796 btrfs_delalloc_release_extents(inode, blocksize);
4800 if (only_release_metadata)
4801 btrfs_check_nocow_unlock(inode);
4802 extent_changeset_free(data_reserved);
4806 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4807 u64 offset, u64 len)
4809 struct btrfs_fs_info *fs_info = root->fs_info;
4810 struct btrfs_trans_handle *trans;
4811 struct btrfs_drop_extents_args drop_args = { 0 };
4815 * Still need to make sure the inode looks like it's been updated so
4816 * that any holes get logged if we fsync.
4818 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4819 inode->last_trans = fs_info->generation;
4820 inode->last_sub_trans = root->log_transid;
4821 inode->last_log_commit = root->last_log_commit;
4826 * 1 - for the one we're dropping
4827 * 1 - for the one we're adding
4828 * 1 - for updating the inode.
4830 trans = btrfs_start_transaction(root, 3);
4832 return PTR_ERR(trans);
4834 drop_args.start = offset;
4835 drop_args.end = offset + len;
4836 drop_args.drop_cache = true;
4838 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4840 btrfs_abort_transaction(trans, ret);
4841 btrfs_end_transaction(trans);
4845 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4846 offset, 0, 0, len, 0, len, 0, 0, 0);
4848 btrfs_abort_transaction(trans, ret);
4850 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4851 btrfs_update_inode(trans, root, inode);
4853 btrfs_end_transaction(trans);
4858 * This function puts in dummy file extents for the area we're creating a hole
4859 * for. So if we are truncating this file to a larger size we need to insert
4860 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4861 * the range between oldsize and size
4863 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4865 struct btrfs_root *root = inode->root;
4866 struct btrfs_fs_info *fs_info = root->fs_info;
4867 struct extent_io_tree *io_tree = &inode->io_tree;
4868 struct extent_map *em = NULL;
4869 struct extent_state *cached_state = NULL;
4870 struct extent_map_tree *em_tree = &inode->extent_tree;
4871 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4872 u64 block_end = ALIGN(size, fs_info->sectorsize);
4879 * If our size started in the middle of a block we need to zero out the
4880 * rest of the block before we expand the i_size, otherwise we could
4881 * expose stale data.
4883 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4887 if (size <= hole_start)
4890 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4892 cur_offset = hole_start;
4894 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4895 block_end - cur_offset);
4901 last_byte = min(extent_map_end(em), block_end);
4902 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4903 hole_size = last_byte - cur_offset;
4905 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4906 struct extent_map *hole_em;
4908 err = maybe_insert_hole(root, inode, cur_offset,
4913 err = btrfs_inode_set_file_extent_range(inode,
4914 cur_offset, hole_size);
4918 btrfs_drop_extent_cache(inode, cur_offset,
4919 cur_offset + hole_size - 1, 0);
4920 hole_em = alloc_extent_map();
4922 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4923 &inode->runtime_flags);
4926 hole_em->start = cur_offset;
4927 hole_em->len = hole_size;
4928 hole_em->orig_start = cur_offset;
4930 hole_em->block_start = EXTENT_MAP_HOLE;
4931 hole_em->block_len = 0;
4932 hole_em->orig_block_len = 0;
4933 hole_em->ram_bytes = hole_size;
4934 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4935 hole_em->generation = fs_info->generation;
4938 write_lock(&em_tree->lock);
4939 err = add_extent_mapping(em_tree, hole_em, 1);
4940 write_unlock(&em_tree->lock);
4943 btrfs_drop_extent_cache(inode, cur_offset,
4947 free_extent_map(hole_em);
4949 err = btrfs_inode_set_file_extent_range(inode,
4950 cur_offset, hole_size);
4955 free_extent_map(em);
4957 cur_offset = last_byte;
4958 if (cur_offset >= block_end)
4961 free_extent_map(em);
4962 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4966 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4968 struct btrfs_root *root = BTRFS_I(inode)->root;
4969 struct btrfs_trans_handle *trans;
4970 loff_t oldsize = i_size_read(inode);
4971 loff_t newsize = attr->ia_size;
4972 int mask = attr->ia_valid;
4976 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4977 * special case where we need to update the times despite not having
4978 * these flags set. For all other operations the VFS set these flags
4979 * explicitly if it wants a timestamp update.
4981 if (newsize != oldsize) {
4982 inode_inc_iversion(inode);
4983 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4984 inode->i_ctime = inode->i_mtime =
4985 current_time(inode);
4988 if (newsize > oldsize) {
4990 * Don't do an expanding truncate while snapshotting is ongoing.
4991 * This is to ensure the snapshot captures a fully consistent
4992 * state of this file - if the snapshot captures this expanding
4993 * truncation, it must capture all writes that happened before
4996 btrfs_drew_write_lock(&root->snapshot_lock);
4997 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4999 btrfs_drew_write_unlock(&root->snapshot_lock);
5003 trans = btrfs_start_transaction(root, 1);
5004 if (IS_ERR(trans)) {
5005 btrfs_drew_write_unlock(&root->snapshot_lock);
5006 return PTR_ERR(trans);
5009 i_size_write(inode, newsize);
5010 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5011 pagecache_isize_extended(inode, oldsize, newsize);
5012 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5013 btrfs_drew_write_unlock(&root->snapshot_lock);
5014 btrfs_end_transaction(trans);
5018 * We're truncating a file that used to have good data down to
5019 * zero. Make sure any new writes to the file get on disk
5023 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5024 &BTRFS_I(inode)->runtime_flags);
5026 truncate_setsize(inode, newsize);
5028 inode_dio_wait(inode);
5030 ret = btrfs_truncate(inode, newsize == oldsize);
5031 if (ret && inode->i_nlink) {
5035 * Truncate failed, so fix up the in-memory size. We
5036 * adjusted disk_i_size down as we removed extents, so
5037 * wait for disk_i_size to be stable and then update the
5038 * in-memory size to match.
5040 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5043 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5050 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5052 struct inode *inode = d_inode(dentry);
5053 struct btrfs_root *root = BTRFS_I(inode)->root;
5056 if (btrfs_root_readonly(root))
5059 err = setattr_prepare(dentry, attr);
5063 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5064 err = btrfs_setsize(inode, attr);
5069 if (attr->ia_valid) {
5070 setattr_copy(inode, attr);
5071 inode_inc_iversion(inode);
5072 err = btrfs_dirty_inode(inode);
5074 if (!err && attr->ia_valid & ATTR_MODE)
5075 err = posix_acl_chmod(inode, inode->i_mode);
5082 * While truncating the inode pages during eviction, we get the VFS calling
5083 * btrfs_invalidatepage() against each page of the inode. This is slow because
5084 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5085 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5086 * extent_state structures over and over, wasting lots of time.
5088 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5089 * those expensive operations on a per page basis and do only the ordered io
5090 * finishing, while we release here the extent_map and extent_state structures,
5091 * without the excessive merging and splitting.
5093 static void evict_inode_truncate_pages(struct inode *inode)
5095 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5096 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5097 struct rb_node *node;
5099 ASSERT(inode->i_state & I_FREEING);
5100 truncate_inode_pages_final(&inode->i_data);
5102 write_lock(&map_tree->lock);
5103 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5104 struct extent_map *em;
5106 node = rb_first_cached(&map_tree->map);
5107 em = rb_entry(node, struct extent_map, rb_node);
5108 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5109 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5110 remove_extent_mapping(map_tree, em);
5111 free_extent_map(em);
5112 if (need_resched()) {
5113 write_unlock(&map_tree->lock);
5115 write_lock(&map_tree->lock);
5118 write_unlock(&map_tree->lock);
5121 * Keep looping until we have no more ranges in the io tree.
5122 * We can have ongoing bios started by readahead that have
5123 * their endio callback (extent_io.c:end_bio_extent_readpage)
5124 * still in progress (unlocked the pages in the bio but did not yet
5125 * unlocked the ranges in the io tree). Therefore this means some
5126 * ranges can still be locked and eviction started because before
5127 * submitting those bios, which are executed by a separate task (work
5128 * queue kthread), inode references (inode->i_count) were not taken
5129 * (which would be dropped in the end io callback of each bio).
5130 * Therefore here we effectively end up waiting for those bios and
5131 * anyone else holding locked ranges without having bumped the inode's
5132 * reference count - if we don't do it, when they access the inode's
5133 * io_tree to unlock a range it may be too late, leading to an
5134 * use-after-free issue.
5136 spin_lock(&io_tree->lock);
5137 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5138 struct extent_state *state;
5139 struct extent_state *cached_state = NULL;
5142 unsigned state_flags;
5144 node = rb_first(&io_tree->state);
5145 state = rb_entry(node, struct extent_state, rb_node);
5146 start = state->start;
5148 state_flags = state->state;
5149 spin_unlock(&io_tree->lock);
5151 lock_extent_bits(io_tree, start, end, &cached_state);
5154 * If still has DELALLOC flag, the extent didn't reach disk,
5155 * and its reserved space won't be freed by delayed_ref.
5156 * So we need to free its reserved space here.
5157 * (Refer to comment in btrfs_invalidatepage, case 2)
5159 * Note, end is the bytenr of last byte, so we need + 1 here.
5161 if (state_flags & EXTENT_DELALLOC)
5162 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5165 clear_extent_bit(io_tree, start, end,
5166 EXTENT_LOCKED | EXTENT_DELALLOC |
5167 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5171 spin_lock(&io_tree->lock);
5173 spin_unlock(&io_tree->lock);
5176 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5177 struct btrfs_block_rsv *rsv)
5179 struct btrfs_fs_info *fs_info = root->fs_info;
5180 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5181 struct btrfs_trans_handle *trans;
5182 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5186 * Eviction should be taking place at some place safe because of our
5187 * delayed iputs. However the normal flushing code will run delayed
5188 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5190 * We reserve the delayed_refs_extra here again because we can't use
5191 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5192 * above. We reserve our extra bit here because we generate a ton of
5193 * delayed refs activity by truncating.
5195 * If we cannot make our reservation we'll attempt to steal from the
5196 * global reserve, because we really want to be able to free up space.
5198 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5199 BTRFS_RESERVE_FLUSH_EVICT);
5202 * Try to steal from the global reserve if there is space for
5205 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5206 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5208 "could not allocate space for delete; will truncate on mount");
5209 return ERR_PTR(-ENOSPC);
5211 delayed_refs_extra = 0;
5214 trans = btrfs_join_transaction(root);
5218 if (delayed_refs_extra) {
5219 trans->block_rsv = &fs_info->trans_block_rsv;
5220 trans->bytes_reserved = delayed_refs_extra;
5221 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5222 delayed_refs_extra, 1);
5227 void btrfs_evict_inode(struct inode *inode)
5229 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5230 struct btrfs_trans_handle *trans;
5231 struct btrfs_root *root = BTRFS_I(inode)->root;
5232 struct btrfs_block_rsv *rsv;
5235 trace_btrfs_inode_evict(inode);
5242 evict_inode_truncate_pages(inode);
5244 if (inode->i_nlink &&
5245 ((btrfs_root_refs(&root->root_item) != 0 &&
5246 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5247 btrfs_is_free_space_inode(BTRFS_I(inode))))
5250 if (is_bad_inode(inode))
5253 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5255 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5258 if (inode->i_nlink > 0) {
5259 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5260 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5264 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5268 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5271 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5274 btrfs_i_size_write(BTRFS_I(inode), 0);
5277 trans = evict_refill_and_join(root, rsv);
5281 trans->block_rsv = rsv;
5283 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5285 trans->block_rsv = &fs_info->trans_block_rsv;
5286 btrfs_end_transaction(trans);
5287 btrfs_btree_balance_dirty(fs_info);
5288 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5295 * Errors here aren't a big deal, it just means we leave orphan items in
5296 * the tree. They will be cleaned up on the next mount. If the inode
5297 * number gets reused, cleanup deletes the orphan item without doing
5298 * anything, and unlink reuses the existing orphan item.
5300 * If it turns out that we are dropping too many of these, we might want
5301 * to add a mechanism for retrying these after a commit.
5303 trans = evict_refill_and_join(root, rsv);
5304 if (!IS_ERR(trans)) {
5305 trans->block_rsv = rsv;
5306 btrfs_orphan_del(trans, BTRFS_I(inode));
5307 trans->block_rsv = &fs_info->trans_block_rsv;
5308 btrfs_end_transaction(trans);
5312 btrfs_free_block_rsv(fs_info, rsv);
5315 * If we didn't successfully delete, the orphan item will still be in
5316 * the tree and we'll retry on the next mount. Again, we might also want
5317 * to retry these periodically in the future.
5319 btrfs_remove_delayed_node(BTRFS_I(inode));
5324 * Return the key found in the dir entry in the location pointer, fill @type
5325 * with BTRFS_FT_*, and return 0.
5327 * If no dir entries were found, returns -ENOENT.
5328 * If found a corrupted location in dir entry, returns -EUCLEAN.
5330 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5331 struct btrfs_key *location, u8 *type)
5333 const char *name = dentry->d_name.name;
5334 int namelen = dentry->d_name.len;
5335 struct btrfs_dir_item *di;
5336 struct btrfs_path *path;
5337 struct btrfs_root *root = BTRFS_I(dir)->root;
5340 path = btrfs_alloc_path();
5344 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5346 if (IS_ERR_OR_NULL(di)) {
5347 ret = di ? PTR_ERR(di) : -ENOENT;
5351 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5352 if (location->type != BTRFS_INODE_ITEM_KEY &&
5353 location->type != BTRFS_ROOT_ITEM_KEY) {
5355 btrfs_warn(root->fs_info,
5356 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5357 __func__, name, btrfs_ino(BTRFS_I(dir)),
5358 location->objectid, location->type, location->offset);
5361 *type = btrfs_dir_type(path->nodes[0], di);
5363 btrfs_free_path(path);
5368 * when we hit a tree root in a directory, the btrfs part of the inode
5369 * needs to be changed to reflect the root directory of the tree root. This
5370 * is kind of like crossing a mount point.
5372 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5374 struct dentry *dentry,
5375 struct btrfs_key *location,
5376 struct btrfs_root **sub_root)
5378 struct btrfs_path *path;
5379 struct btrfs_root *new_root;
5380 struct btrfs_root_ref *ref;
5381 struct extent_buffer *leaf;
5382 struct btrfs_key key;
5386 path = btrfs_alloc_path();
5393 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5394 key.type = BTRFS_ROOT_REF_KEY;
5395 key.offset = location->objectid;
5397 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5404 leaf = path->nodes[0];
5405 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5406 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5407 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5410 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5411 (unsigned long)(ref + 1),
5412 dentry->d_name.len);
5416 btrfs_release_path(path);
5418 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5419 if (IS_ERR(new_root)) {
5420 err = PTR_ERR(new_root);
5424 *sub_root = new_root;
5425 location->objectid = btrfs_root_dirid(&new_root->root_item);
5426 location->type = BTRFS_INODE_ITEM_KEY;
5427 location->offset = 0;
5430 btrfs_free_path(path);
5434 static void inode_tree_add(struct inode *inode)
5436 struct btrfs_root *root = BTRFS_I(inode)->root;
5437 struct btrfs_inode *entry;
5439 struct rb_node *parent;
5440 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5441 u64 ino = btrfs_ino(BTRFS_I(inode));
5443 if (inode_unhashed(inode))
5446 spin_lock(&root->inode_lock);
5447 p = &root->inode_tree.rb_node;
5450 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5452 if (ino < btrfs_ino(entry))
5453 p = &parent->rb_left;
5454 else if (ino > btrfs_ino(entry))
5455 p = &parent->rb_right;
5457 WARN_ON(!(entry->vfs_inode.i_state &
5458 (I_WILL_FREE | I_FREEING)));
5459 rb_replace_node(parent, new, &root->inode_tree);
5460 RB_CLEAR_NODE(parent);
5461 spin_unlock(&root->inode_lock);
5465 rb_link_node(new, parent, p);
5466 rb_insert_color(new, &root->inode_tree);
5467 spin_unlock(&root->inode_lock);
5470 static void inode_tree_del(struct btrfs_inode *inode)
5472 struct btrfs_root *root = inode->root;
5475 spin_lock(&root->inode_lock);
5476 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5477 rb_erase(&inode->rb_node, &root->inode_tree);
5478 RB_CLEAR_NODE(&inode->rb_node);
5479 empty = RB_EMPTY_ROOT(&root->inode_tree);
5481 spin_unlock(&root->inode_lock);
5483 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5484 spin_lock(&root->inode_lock);
5485 empty = RB_EMPTY_ROOT(&root->inode_tree);
5486 spin_unlock(&root->inode_lock);
5488 btrfs_add_dead_root(root);
5493 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5495 struct btrfs_iget_args *args = p;
5497 inode->i_ino = args->ino;
5498 BTRFS_I(inode)->location.objectid = args->ino;
5499 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5500 BTRFS_I(inode)->location.offset = 0;
5501 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5502 BUG_ON(args->root && !BTRFS_I(inode)->root);
5506 static int btrfs_find_actor(struct inode *inode, void *opaque)
5508 struct btrfs_iget_args *args = opaque;
5510 return args->ino == BTRFS_I(inode)->location.objectid &&
5511 args->root == BTRFS_I(inode)->root;
5514 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5515 struct btrfs_root *root)
5517 struct inode *inode;
5518 struct btrfs_iget_args args;
5519 unsigned long hashval = btrfs_inode_hash(ino, root);
5524 inode = iget5_locked(s, hashval, btrfs_find_actor,
5525 btrfs_init_locked_inode,
5531 * Get an inode object given its inode number and corresponding root.
5532 * Path can be preallocated to prevent recursing back to iget through
5533 * allocator. NULL is also valid but may require an additional allocation
5536 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5537 struct btrfs_root *root, struct btrfs_path *path)
5539 struct inode *inode;
5541 inode = btrfs_iget_locked(s, ino, root);
5543 return ERR_PTR(-ENOMEM);
5545 if (inode->i_state & I_NEW) {
5548 ret = btrfs_read_locked_inode(inode, path);
5550 inode_tree_add(inode);
5551 unlock_new_inode(inode);
5555 * ret > 0 can come from btrfs_search_slot called by
5556 * btrfs_read_locked_inode, this means the inode item
5561 inode = ERR_PTR(ret);
5568 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5570 return btrfs_iget_path(s, ino, root, NULL);
5573 static struct inode *new_simple_dir(struct super_block *s,
5574 struct btrfs_key *key,
5575 struct btrfs_root *root)
5577 struct inode *inode = new_inode(s);
5580 return ERR_PTR(-ENOMEM);
5582 BTRFS_I(inode)->root = btrfs_grab_root(root);
5583 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5584 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5586 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5588 * We only need lookup, the rest is read-only and there's no inode
5589 * associated with the dentry
5591 inode->i_op = &simple_dir_inode_operations;
5592 inode->i_opflags &= ~IOP_XATTR;
5593 inode->i_fop = &simple_dir_operations;
5594 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5595 inode->i_mtime = current_time(inode);
5596 inode->i_atime = inode->i_mtime;
5597 inode->i_ctime = inode->i_mtime;
5598 BTRFS_I(inode)->i_otime = inode->i_mtime;
5603 static inline u8 btrfs_inode_type(struct inode *inode)
5606 * Compile-time asserts that generic FT_* types still match
5609 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5610 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5611 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5612 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5613 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5614 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5615 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5616 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5618 return fs_umode_to_ftype(inode->i_mode);
5621 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5623 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5624 struct inode *inode;
5625 struct btrfs_root *root = BTRFS_I(dir)->root;
5626 struct btrfs_root *sub_root = root;
5627 struct btrfs_key location;
5631 if (dentry->d_name.len > BTRFS_NAME_LEN)
5632 return ERR_PTR(-ENAMETOOLONG);
5634 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5636 return ERR_PTR(ret);
5638 if (location.type == BTRFS_INODE_ITEM_KEY) {
5639 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5643 /* Do extra check against inode mode with di_type */
5644 if (btrfs_inode_type(inode) != di_type) {
5646 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5647 inode->i_mode, btrfs_inode_type(inode),
5650 return ERR_PTR(-EUCLEAN);
5655 ret = fixup_tree_root_location(fs_info, dir, dentry,
5656 &location, &sub_root);
5659 inode = ERR_PTR(ret);
5661 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5663 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5665 if (root != sub_root)
5666 btrfs_put_root(sub_root);
5668 if (!IS_ERR(inode) && root != sub_root) {
5669 down_read(&fs_info->cleanup_work_sem);
5670 if (!sb_rdonly(inode->i_sb))
5671 ret = btrfs_orphan_cleanup(sub_root);
5672 up_read(&fs_info->cleanup_work_sem);
5675 inode = ERR_PTR(ret);
5682 static int btrfs_dentry_delete(const struct dentry *dentry)
5684 struct btrfs_root *root;
5685 struct inode *inode = d_inode(dentry);
5687 if (!inode && !IS_ROOT(dentry))
5688 inode = d_inode(dentry->d_parent);
5691 root = BTRFS_I(inode)->root;
5692 if (btrfs_root_refs(&root->root_item) == 0)
5695 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5701 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5704 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5706 if (inode == ERR_PTR(-ENOENT))
5708 return d_splice_alias(inode, dentry);
5712 * All this infrastructure exists because dir_emit can fault, and we are holding
5713 * the tree lock when doing readdir. For now just allocate a buffer and copy
5714 * our information into that, and then dir_emit from the buffer. This is
5715 * similar to what NFS does, only we don't keep the buffer around in pagecache
5716 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5717 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5720 static int btrfs_opendir(struct inode *inode, struct file *file)
5722 struct btrfs_file_private *private;
5724 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5727 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5728 if (!private->filldir_buf) {
5732 file->private_data = private;
5743 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5746 struct dir_entry *entry = addr;
5747 char *name = (char *)(entry + 1);
5749 ctx->pos = get_unaligned(&entry->offset);
5750 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5751 get_unaligned(&entry->ino),
5752 get_unaligned(&entry->type)))
5754 addr += sizeof(struct dir_entry) +
5755 get_unaligned(&entry->name_len);
5761 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5763 struct inode *inode = file_inode(file);
5764 struct btrfs_root *root = BTRFS_I(inode)->root;
5765 struct btrfs_file_private *private = file->private_data;
5766 struct btrfs_dir_item *di;
5767 struct btrfs_key key;
5768 struct btrfs_key found_key;
5769 struct btrfs_path *path;
5771 struct list_head ins_list;
5772 struct list_head del_list;
5774 struct extent_buffer *leaf;
5781 struct btrfs_key location;
5783 if (!dir_emit_dots(file, ctx))
5786 path = btrfs_alloc_path();
5790 addr = private->filldir_buf;
5791 path->reada = READA_FORWARD;
5793 INIT_LIST_HEAD(&ins_list);
5794 INIT_LIST_HEAD(&del_list);
5795 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5798 key.type = BTRFS_DIR_INDEX_KEY;
5799 key.offset = ctx->pos;
5800 key.objectid = btrfs_ino(BTRFS_I(inode));
5802 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5807 struct dir_entry *entry;
5809 leaf = path->nodes[0];
5810 slot = path->slots[0];
5811 if (slot >= btrfs_header_nritems(leaf)) {
5812 ret = btrfs_next_leaf(root, path);
5820 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5822 if (found_key.objectid != key.objectid)
5824 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5826 if (found_key.offset < ctx->pos)
5828 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5830 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5831 name_len = btrfs_dir_name_len(leaf, di);
5832 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5834 btrfs_release_path(path);
5835 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5838 addr = private->filldir_buf;
5845 put_unaligned(name_len, &entry->name_len);
5846 name_ptr = (char *)(entry + 1);
5847 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5849 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5851 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5852 put_unaligned(location.objectid, &entry->ino);
5853 put_unaligned(found_key.offset, &entry->offset);
5855 addr += sizeof(struct dir_entry) + name_len;
5856 total_len += sizeof(struct dir_entry) + name_len;
5860 btrfs_release_path(path);
5862 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5866 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5871 * Stop new entries from being returned after we return the last
5874 * New directory entries are assigned a strictly increasing
5875 * offset. This means that new entries created during readdir
5876 * are *guaranteed* to be seen in the future by that readdir.
5877 * This has broken buggy programs which operate on names as
5878 * they're returned by readdir. Until we re-use freed offsets
5879 * we have this hack to stop new entries from being returned
5880 * under the assumption that they'll never reach this huge
5883 * This is being careful not to overflow 32bit loff_t unless the
5884 * last entry requires it because doing so has broken 32bit apps
5887 if (ctx->pos >= INT_MAX)
5888 ctx->pos = LLONG_MAX;
5895 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5896 btrfs_free_path(path);
5901 * This is somewhat expensive, updating the tree every time the
5902 * inode changes. But, it is most likely to find the inode in cache.
5903 * FIXME, needs more benchmarking...there are no reasons other than performance
5904 * to keep or drop this code.
5906 static int btrfs_dirty_inode(struct inode *inode)
5908 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5909 struct btrfs_root *root = BTRFS_I(inode)->root;
5910 struct btrfs_trans_handle *trans;
5913 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5916 trans = btrfs_join_transaction(root);
5918 return PTR_ERR(trans);
5920 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5921 if (ret && ret == -ENOSPC) {
5922 /* whoops, lets try again with the full transaction */
5923 btrfs_end_transaction(trans);
5924 trans = btrfs_start_transaction(root, 1);
5926 return PTR_ERR(trans);
5928 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5930 btrfs_end_transaction(trans);
5931 if (BTRFS_I(inode)->delayed_node)
5932 btrfs_balance_delayed_items(fs_info);
5938 * This is a copy of file_update_time. We need this so we can return error on
5939 * ENOSPC for updating the inode in the case of file write and mmap writes.
5941 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5944 struct btrfs_root *root = BTRFS_I(inode)->root;
5945 bool dirty = flags & ~S_VERSION;
5947 if (btrfs_root_readonly(root))
5950 if (flags & S_VERSION)
5951 dirty |= inode_maybe_inc_iversion(inode, dirty);
5952 if (flags & S_CTIME)
5953 inode->i_ctime = *now;
5954 if (flags & S_MTIME)
5955 inode->i_mtime = *now;
5956 if (flags & S_ATIME)
5957 inode->i_atime = *now;
5958 return dirty ? btrfs_dirty_inode(inode) : 0;
5962 * find the highest existing sequence number in a directory
5963 * and then set the in-memory index_cnt variable to reflect
5964 * free sequence numbers
5966 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5968 struct btrfs_root *root = inode->root;
5969 struct btrfs_key key, found_key;
5970 struct btrfs_path *path;
5971 struct extent_buffer *leaf;
5974 key.objectid = btrfs_ino(inode);
5975 key.type = BTRFS_DIR_INDEX_KEY;
5976 key.offset = (u64)-1;
5978 path = btrfs_alloc_path();
5982 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5985 /* FIXME: we should be able to handle this */
5991 * MAGIC NUMBER EXPLANATION:
5992 * since we search a directory based on f_pos we have to start at 2
5993 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5994 * else has to start at 2
5996 if (path->slots[0] == 0) {
5997 inode->index_cnt = 2;
6003 leaf = path->nodes[0];
6004 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6006 if (found_key.objectid != btrfs_ino(inode) ||
6007 found_key.type != BTRFS_DIR_INDEX_KEY) {
6008 inode->index_cnt = 2;
6012 inode->index_cnt = found_key.offset + 1;
6014 btrfs_free_path(path);
6019 * helper to find a free sequence number in a given directory. This current
6020 * code is very simple, later versions will do smarter things in the btree
6022 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6026 if (dir->index_cnt == (u64)-1) {
6027 ret = btrfs_inode_delayed_dir_index_count(dir);
6029 ret = btrfs_set_inode_index_count(dir);
6035 *index = dir->index_cnt;
6041 static int btrfs_insert_inode_locked(struct inode *inode)
6043 struct btrfs_iget_args args;
6045 args.ino = BTRFS_I(inode)->location.objectid;
6046 args.root = BTRFS_I(inode)->root;
6048 return insert_inode_locked4(inode,
6049 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6050 btrfs_find_actor, &args);
6054 * Inherit flags from the parent inode.
6056 * Currently only the compression flags and the cow flags are inherited.
6058 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6065 flags = BTRFS_I(dir)->flags;
6067 if (flags & BTRFS_INODE_NOCOMPRESS) {
6068 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6069 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6070 } else if (flags & BTRFS_INODE_COMPRESS) {
6071 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6072 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6075 if (flags & BTRFS_INODE_NODATACOW) {
6076 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6077 if (S_ISREG(inode->i_mode))
6078 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6081 btrfs_sync_inode_flags_to_i_flags(inode);
6084 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6085 struct btrfs_root *root,
6087 const char *name, int name_len,
6088 u64 ref_objectid, u64 objectid,
6089 umode_t mode, u64 *index)
6091 struct btrfs_fs_info *fs_info = root->fs_info;
6092 struct inode *inode;
6093 struct btrfs_inode_item *inode_item;
6094 struct btrfs_key *location;
6095 struct btrfs_path *path;
6096 struct btrfs_inode_ref *ref;
6097 struct btrfs_key key[2];
6099 int nitems = name ? 2 : 1;
6101 unsigned int nofs_flag;
6104 path = btrfs_alloc_path();
6106 return ERR_PTR(-ENOMEM);
6108 nofs_flag = memalloc_nofs_save();
6109 inode = new_inode(fs_info->sb);
6110 memalloc_nofs_restore(nofs_flag);
6112 btrfs_free_path(path);
6113 return ERR_PTR(-ENOMEM);
6117 * O_TMPFILE, set link count to 0, so that after this point,
6118 * we fill in an inode item with the correct link count.
6121 set_nlink(inode, 0);
6124 * we have to initialize this early, so we can reclaim the inode
6125 * number if we fail afterwards in this function.
6127 inode->i_ino = objectid;
6130 trace_btrfs_inode_request(dir);
6132 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6134 btrfs_free_path(path);
6136 return ERR_PTR(ret);
6142 * index_cnt is ignored for everything but a dir,
6143 * btrfs_set_inode_index_count has an explanation for the magic
6146 BTRFS_I(inode)->index_cnt = 2;
6147 BTRFS_I(inode)->dir_index = *index;
6148 BTRFS_I(inode)->root = btrfs_grab_root(root);
6149 BTRFS_I(inode)->generation = trans->transid;
6150 inode->i_generation = BTRFS_I(inode)->generation;
6153 * We could have gotten an inode number from somebody who was fsynced
6154 * and then removed in this same transaction, so let's just set full
6155 * sync since it will be a full sync anyway and this will blow away the
6156 * old info in the log.
6158 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6160 key[0].objectid = objectid;
6161 key[0].type = BTRFS_INODE_ITEM_KEY;
6164 sizes[0] = sizeof(struct btrfs_inode_item);
6168 * Start new inodes with an inode_ref. This is slightly more
6169 * efficient for small numbers of hard links since they will
6170 * be packed into one item. Extended refs will kick in if we
6171 * add more hard links than can fit in the ref item.
6173 key[1].objectid = objectid;
6174 key[1].type = BTRFS_INODE_REF_KEY;
6175 key[1].offset = ref_objectid;
6177 sizes[1] = name_len + sizeof(*ref);
6180 location = &BTRFS_I(inode)->location;
6181 location->objectid = objectid;
6182 location->offset = 0;
6183 location->type = BTRFS_INODE_ITEM_KEY;
6185 ret = btrfs_insert_inode_locked(inode);
6191 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6195 inode_init_owner(inode, dir, mode);
6196 inode_set_bytes(inode, 0);
6198 inode->i_mtime = current_time(inode);
6199 inode->i_atime = inode->i_mtime;
6200 inode->i_ctime = inode->i_mtime;
6201 BTRFS_I(inode)->i_otime = inode->i_mtime;
6203 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6204 struct btrfs_inode_item);
6205 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6206 sizeof(*inode_item));
6207 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6210 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6211 struct btrfs_inode_ref);
6212 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6213 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6214 ptr = (unsigned long)(ref + 1);
6215 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6218 btrfs_mark_buffer_dirty(path->nodes[0]);
6219 btrfs_free_path(path);
6221 btrfs_inherit_iflags(inode, dir);
6223 if (S_ISREG(mode)) {
6224 if (btrfs_test_opt(fs_info, NODATASUM))
6225 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6226 if (btrfs_test_opt(fs_info, NODATACOW))
6227 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6228 BTRFS_INODE_NODATASUM;
6231 inode_tree_add(inode);
6233 trace_btrfs_inode_new(inode);
6234 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6236 btrfs_update_root_times(trans, root);
6238 ret = btrfs_inode_inherit_props(trans, inode, dir);
6241 "error inheriting props for ino %llu (root %llu): %d",
6242 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6247 discard_new_inode(inode);
6250 BTRFS_I(dir)->index_cnt--;
6251 btrfs_free_path(path);
6252 return ERR_PTR(ret);
6256 * utility function to add 'inode' into 'parent_inode' with
6257 * a give name and a given sequence number.
6258 * if 'add_backref' is true, also insert a backref from the
6259 * inode to the parent directory.
6261 int btrfs_add_link(struct btrfs_trans_handle *trans,
6262 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6263 const char *name, int name_len, int add_backref, u64 index)
6266 struct btrfs_key key;
6267 struct btrfs_root *root = parent_inode->root;
6268 u64 ino = btrfs_ino(inode);
6269 u64 parent_ino = btrfs_ino(parent_inode);
6271 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6272 memcpy(&key, &inode->root->root_key, sizeof(key));
6275 key.type = BTRFS_INODE_ITEM_KEY;
6279 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6280 ret = btrfs_add_root_ref(trans, key.objectid,
6281 root->root_key.objectid, parent_ino,
6282 index, name, name_len);
6283 } else if (add_backref) {
6284 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6288 /* Nothing to clean up yet */
6292 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6293 btrfs_inode_type(&inode->vfs_inode), index);
6294 if (ret == -EEXIST || ret == -EOVERFLOW)
6297 btrfs_abort_transaction(trans, ret);
6301 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6303 inode_inc_iversion(&parent_inode->vfs_inode);
6305 * If we are replaying a log tree, we do not want to update the mtime
6306 * and ctime of the parent directory with the current time, since the
6307 * log replay procedure is responsible for setting them to their correct
6308 * values (the ones it had when the fsync was done).
6310 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6311 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6313 parent_inode->vfs_inode.i_mtime = now;
6314 parent_inode->vfs_inode.i_ctime = now;
6316 ret = btrfs_update_inode(trans, root, parent_inode);
6318 btrfs_abort_transaction(trans, ret);
6322 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6325 err = btrfs_del_root_ref(trans, key.objectid,
6326 root->root_key.objectid, parent_ino,
6327 &local_index, name, name_len);
6329 btrfs_abort_transaction(trans, err);
6330 } else if (add_backref) {
6334 err = btrfs_del_inode_ref(trans, root, name, name_len,
6335 ino, parent_ino, &local_index);
6337 btrfs_abort_transaction(trans, err);
6340 /* Return the original error code */
6344 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6345 struct btrfs_inode *dir, struct dentry *dentry,
6346 struct btrfs_inode *inode, int backref, u64 index)
6348 int err = btrfs_add_link(trans, dir, inode,
6349 dentry->d_name.name, dentry->d_name.len,
6356 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6357 umode_t mode, dev_t rdev)
6359 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6360 struct btrfs_trans_handle *trans;
6361 struct btrfs_root *root = BTRFS_I(dir)->root;
6362 struct inode *inode = NULL;
6368 * 2 for inode item and ref
6370 * 1 for xattr if selinux is on
6372 trans = btrfs_start_transaction(root, 5);
6374 return PTR_ERR(trans);
6376 err = btrfs_get_free_objectid(root, &objectid);
6380 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6381 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6383 if (IS_ERR(inode)) {
6384 err = PTR_ERR(inode);
6390 * If the active LSM wants to access the inode during
6391 * d_instantiate it needs these. Smack checks to see
6392 * if the filesystem supports xattrs by looking at the
6395 inode->i_op = &btrfs_special_inode_operations;
6396 init_special_inode(inode, inode->i_mode, rdev);
6398 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6402 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6407 btrfs_update_inode(trans, root, BTRFS_I(inode));
6408 d_instantiate_new(dentry, inode);
6411 btrfs_end_transaction(trans);
6412 btrfs_btree_balance_dirty(fs_info);
6414 inode_dec_link_count(inode);
6415 discard_new_inode(inode);
6420 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6421 umode_t mode, bool excl)
6423 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6424 struct btrfs_trans_handle *trans;
6425 struct btrfs_root *root = BTRFS_I(dir)->root;
6426 struct inode *inode = NULL;
6432 * 2 for inode item and ref
6434 * 1 for xattr if selinux is on
6436 trans = btrfs_start_transaction(root, 5);
6438 return PTR_ERR(trans);
6440 err = btrfs_get_free_objectid(root, &objectid);
6444 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6445 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6447 if (IS_ERR(inode)) {
6448 err = PTR_ERR(inode);
6453 * If the active LSM wants to access the inode during
6454 * d_instantiate it needs these. Smack checks to see
6455 * if the filesystem supports xattrs by looking at the
6458 inode->i_fop = &btrfs_file_operations;
6459 inode->i_op = &btrfs_file_inode_operations;
6460 inode->i_mapping->a_ops = &btrfs_aops;
6462 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6466 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6470 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6475 d_instantiate_new(dentry, inode);
6478 btrfs_end_transaction(trans);
6480 inode_dec_link_count(inode);
6481 discard_new_inode(inode);
6483 btrfs_btree_balance_dirty(fs_info);
6487 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6488 struct dentry *dentry)
6490 struct btrfs_trans_handle *trans = NULL;
6491 struct btrfs_root *root = BTRFS_I(dir)->root;
6492 struct inode *inode = d_inode(old_dentry);
6493 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6498 /* do not allow sys_link's with other subvols of the same device */
6499 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6502 if (inode->i_nlink >= BTRFS_LINK_MAX)
6505 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6510 * 2 items for inode and inode ref
6511 * 2 items for dir items
6512 * 1 item for parent inode
6513 * 1 item for orphan item deletion if O_TMPFILE
6515 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6516 if (IS_ERR(trans)) {
6517 err = PTR_ERR(trans);
6522 /* There are several dir indexes for this inode, clear the cache. */
6523 BTRFS_I(inode)->dir_index = 0ULL;
6525 inode_inc_iversion(inode);
6526 inode->i_ctime = current_time(inode);
6528 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6530 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6536 struct dentry *parent = dentry->d_parent;
6538 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6541 if (inode->i_nlink == 1) {
6543 * If new hard link count is 1, it's a file created
6544 * with open(2) O_TMPFILE flag.
6546 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6550 d_instantiate(dentry, inode);
6551 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6556 btrfs_end_transaction(trans);
6558 inode_dec_link_count(inode);
6561 btrfs_btree_balance_dirty(fs_info);
6565 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6567 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6568 struct inode *inode = NULL;
6569 struct btrfs_trans_handle *trans;
6570 struct btrfs_root *root = BTRFS_I(dir)->root;
6576 * 2 items for inode and ref
6577 * 2 items for dir items
6578 * 1 for xattr if selinux is on
6580 trans = btrfs_start_transaction(root, 5);
6582 return PTR_ERR(trans);
6584 err = btrfs_get_free_objectid(root, &objectid);
6588 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6589 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6590 S_IFDIR | mode, &index);
6591 if (IS_ERR(inode)) {
6592 err = PTR_ERR(inode);
6597 /* these must be set before we unlock the inode */
6598 inode->i_op = &btrfs_dir_inode_operations;
6599 inode->i_fop = &btrfs_dir_file_operations;
6601 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6605 btrfs_i_size_write(BTRFS_I(inode), 0);
6606 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6610 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6611 dentry->d_name.name,
6612 dentry->d_name.len, 0, index);
6616 d_instantiate_new(dentry, inode);
6619 btrfs_end_transaction(trans);
6621 inode_dec_link_count(inode);
6622 discard_new_inode(inode);
6624 btrfs_btree_balance_dirty(fs_info);
6628 static noinline int uncompress_inline(struct btrfs_path *path,
6630 size_t pg_offset, u64 extent_offset,
6631 struct btrfs_file_extent_item *item)
6634 struct extent_buffer *leaf = path->nodes[0];
6637 unsigned long inline_size;
6641 WARN_ON(pg_offset != 0);
6642 compress_type = btrfs_file_extent_compression(leaf, item);
6643 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6644 inline_size = btrfs_file_extent_inline_item_len(leaf,
6645 btrfs_item_nr(path->slots[0]));
6646 tmp = kmalloc(inline_size, GFP_NOFS);
6649 ptr = btrfs_file_extent_inline_start(item);
6651 read_extent_buffer(leaf, tmp, ptr, inline_size);
6653 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6654 ret = btrfs_decompress(compress_type, tmp, page,
6655 extent_offset, inline_size, max_size);
6658 * decompression code contains a memset to fill in any space between the end
6659 * of the uncompressed data and the end of max_size in case the decompressed
6660 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6661 * the end of an inline extent and the beginning of the next block, so we
6662 * cover that region here.
6665 if (max_size + pg_offset < PAGE_SIZE) {
6666 char *map = kmap(page);
6667 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6675 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6676 * @inode: file to search in
6677 * @page: page to read extent data into if the extent is inline
6678 * @pg_offset: offset into @page to copy to
6679 * @start: file offset
6680 * @len: length of range starting at @start
6682 * This returns the first &struct extent_map which overlaps with the given
6683 * range, reading it from the B-tree and caching it if necessary. Note that
6684 * there may be more extents which overlap the given range after the returned
6687 * If @page is not NULL and the extent is inline, this also reads the extent
6688 * data directly into the page and marks the extent up to date in the io_tree.
6690 * Return: ERR_PTR on error, non-NULL extent_map on success.
6692 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6693 struct page *page, size_t pg_offset,
6696 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6698 u64 extent_start = 0;
6700 u64 objectid = btrfs_ino(inode);
6701 int extent_type = -1;
6702 struct btrfs_path *path = NULL;
6703 struct btrfs_root *root = inode->root;
6704 struct btrfs_file_extent_item *item;
6705 struct extent_buffer *leaf;
6706 struct btrfs_key found_key;
6707 struct extent_map *em = NULL;
6708 struct extent_map_tree *em_tree = &inode->extent_tree;
6709 struct extent_io_tree *io_tree = &inode->io_tree;
6711 read_lock(&em_tree->lock);
6712 em = lookup_extent_mapping(em_tree, start, len);
6713 read_unlock(&em_tree->lock);
6716 if (em->start > start || em->start + em->len <= start)
6717 free_extent_map(em);
6718 else if (em->block_start == EXTENT_MAP_INLINE && page)
6719 free_extent_map(em);
6723 em = alloc_extent_map();
6728 em->start = EXTENT_MAP_HOLE;
6729 em->orig_start = EXTENT_MAP_HOLE;
6731 em->block_len = (u64)-1;
6733 path = btrfs_alloc_path();
6739 /* Chances are we'll be called again, so go ahead and do readahead */
6740 path->reada = READA_FORWARD;
6743 * The same explanation in load_free_space_cache applies here as well,
6744 * we only read when we're loading the free space cache, and at that
6745 * point the commit_root has everything we need.
6747 if (btrfs_is_free_space_inode(inode)) {
6748 path->search_commit_root = 1;
6749 path->skip_locking = 1;
6752 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6755 } else if (ret > 0) {
6756 if (path->slots[0] == 0)
6762 leaf = path->nodes[0];
6763 item = btrfs_item_ptr(leaf, path->slots[0],
6764 struct btrfs_file_extent_item);
6765 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6766 if (found_key.objectid != objectid ||
6767 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6769 * If we backup past the first extent we want to move forward
6770 * and see if there is an extent in front of us, otherwise we'll
6771 * say there is a hole for our whole search range which can
6778 extent_type = btrfs_file_extent_type(leaf, item);
6779 extent_start = found_key.offset;
6780 extent_end = btrfs_file_extent_end(path);
6781 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6782 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6783 /* Only regular file could have regular/prealloc extent */
6784 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6787 "regular/prealloc extent found for non-regular inode %llu",
6791 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6793 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6794 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6799 if (start >= extent_end) {
6801 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6802 ret = btrfs_next_leaf(root, path);
6808 leaf = path->nodes[0];
6810 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6811 if (found_key.objectid != objectid ||
6812 found_key.type != BTRFS_EXTENT_DATA_KEY)
6814 if (start + len <= found_key.offset)
6816 if (start > found_key.offset)
6819 /* New extent overlaps with existing one */
6821 em->orig_start = start;
6822 em->len = found_key.offset - start;
6823 em->block_start = EXTENT_MAP_HOLE;
6827 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6829 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6830 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6832 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6836 size_t extent_offset;
6842 size = btrfs_file_extent_ram_bytes(leaf, item);
6843 extent_offset = page_offset(page) + pg_offset - extent_start;
6844 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6845 size - extent_offset);
6846 em->start = extent_start + extent_offset;
6847 em->len = ALIGN(copy_size, fs_info->sectorsize);
6848 em->orig_block_len = em->len;
6849 em->orig_start = em->start;
6850 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6852 if (!PageUptodate(page)) {
6853 if (btrfs_file_extent_compression(leaf, item) !=
6854 BTRFS_COMPRESS_NONE) {
6855 ret = uncompress_inline(path, page, pg_offset,
6856 extent_offset, item);
6861 read_extent_buffer(leaf, map + pg_offset, ptr,
6863 if (pg_offset + copy_size < PAGE_SIZE) {
6864 memset(map + pg_offset + copy_size, 0,
6865 PAGE_SIZE - pg_offset -
6870 flush_dcache_page(page);
6872 set_extent_uptodate(io_tree, em->start,
6873 extent_map_end(em) - 1, NULL, GFP_NOFS);
6878 em->orig_start = start;
6880 em->block_start = EXTENT_MAP_HOLE;
6883 btrfs_release_path(path);
6884 if (em->start > start || extent_map_end(em) <= start) {
6886 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6887 em->start, em->len, start, len);
6892 write_lock(&em_tree->lock);
6893 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6894 write_unlock(&em_tree->lock);
6896 btrfs_free_path(path);
6898 trace_btrfs_get_extent(root, inode, em);
6901 free_extent_map(em);
6902 return ERR_PTR(ret);
6907 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6910 struct extent_map *em;
6911 struct extent_map *hole_em = NULL;
6912 u64 delalloc_start = start;
6918 em = btrfs_get_extent(inode, NULL, 0, start, len);
6922 * If our em maps to:
6924 * - a pre-alloc extent,
6925 * there might actually be delalloc bytes behind it.
6927 if (em->block_start != EXTENT_MAP_HOLE &&
6928 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6933 /* check to see if we've wrapped (len == -1 or similar) */
6942 /* ok, we didn't find anything, lets look for delalloc */
6943 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6944 end, len, EXTENT_DELALLOC, 1);
6945 delalloc_end = delalloc_start + delalloc_len;
6946 if (delalloc_end < delalloc_start)
6947 delalloc_end = (u64)-1;
6950 * We didn't find anything useful, return the original results from
6953 if (delalloc_start > end || delalloc_end <= start) {
6960 * Adjust the delalloc_start to make sure it doesn't go backwards from
6961 * the start they passed in
6963 delalloc_start = max(start, delalloc_start);
6964 delalloc_len = delalloc_end - delalloc_start;
6966 if (delalloc_len > 0) {
6969 const u64 hole_end = extent_map_end(hole_em);
6971 em = alloc_extent_map();
6979 * When btrfs_get_extent can't find anything it returns one
6982 * Make sure what it found really fits our range, and adjust to
6983 * make sure it is based on the start from the caller
6985 if (hole_end <= start || hole_em->start > end) {
6986 free_extent_map(hole_em);
6989 hole_start = max(hole_em->start, start);
6990 hole_len = hole_end - hole_start;
6993 if (hole_em && delalloc_start > hole_start) {
6995 * Our hole starts before our delalloc, so we have to
6996 * return just the parts of the hole that go until the
6999 em->len = min(hole_len, delalloc_start - hole_start);
7000 em->start = hole_start;
7001 em->orig_start = hole_start;
7003 * Don't adjust block start at all, it is fixed at
7006 em->block_start = hole_em->block_start;
7007 em->block_len = hole_len;
7008 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7009 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7012 * Hole is out of passed range or it starts after
7015 em->start = delalloc_start;
7016 em->len = delalloc_len;
7017 em->orig_start = delalloc_start;
7018 em->block_start = EXTENT_MAP_DELALLOC;
7019 em->block_len = delalloc_len;
7026 free_extent_map(hole_em);
7028 free_extent_map(em);
7029 return ERR_PTR(err);
7034 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7037 const u64 orig_start,
7038 const u64 block_start,
7039 const u64 block_len,
7040 const u64 orig_block_len,
7041 const u64 ram_bytes,
7044 struct extent_map *em = NULL;
7047 if (type != BTRFS_ORDERED_NOCOW) {
7048 em = create_io_em(inode, start, len, orig_start, block_start,
7049 block_len, orig_block_len, ram_bytes,
7050 BTRFS_COMPRESS_NONE, /* compress_type */
7055 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7059 free_extent_map(em);
7060 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7069 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7072 struct btrfs_root *root = inode->root;
7073 struct btrfs_fs_info *fs_info = root->fs_info;
7074 struct extent_map *em;
7075 struct btrfs_key ins;
7079 alloc_hint = get_extent_allocation_hint(inode, start, len);
7080 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7081 0, alloc_hint, &ins, 1, 1);
7083 return ERR_PTR(ret);
7085 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7086 ins.objectid, ins.offset, ins.offset,
7087 ins.offset, BTRFS_ORDERED_REGULAR);
7088 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7090 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7097 * Check if we can do nocow write into the range [@offset, @offset + @len)
7099 * @offset: File offset
7100 * @len: The length to write, will be updated to the nocow writeable
7102 * @orig_start: (optional) Return the original file offset of the file extent
7103 * @orig_len: (optional) Return the original on-disk length of the file extent
7104 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7105 * @strict: if true, omit optimizations that might force us into unnecessary
7106 * cow. e.g., don't trust generation number.
7109 * >0 and update @len if we can do nocow write
7110 * 0 if we can't do nocow write
7111 * <0 if error happened
7113 * NOTE: This only checks the file extents, caller is responsible to wait for
7114 * any ordered extents.
7116 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7117 u64 *orig_start, u64 *orig_block_len,
7118 u64 *ram_bytes, bool strict)
7120 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7121 struct btrfs_path *path;
7123 struct extent_buffer *leaf;
7124 struct btrfs_root *root = BTRFS_I(inode)->root;
7125 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7126 struct btrfs_file_extent_item *fi;
7127 struct btrfs_key key;
7134 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7136 path = btrfs_alloc_path();
7140 ret = btrfs_lookup_file_extent(NULL, root, path,
7141 btrfs_ino(BTRFS_I(inode)), offset, 0);
7145 slot = path->slots[0];
7148 /* can't find the item, must cow */
7155 leaf = path->nodes[0];
7156 btrfs_item_key_to_cpu(leaf, &key, slot);
7157 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7158 key.type != BTRFS_EXTENT_DATA_KEY) {
7159 /* not our file or wrong item type, must cow */
7163 if (key.offset > offset) {
7164 /* Wrong offset, must cow */
7168 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7169 found_type = btrfs_file_extent_type(leaf, fi);
7170 if (found_type != BTRFS_FILE_EXTENT_REG &&
7171 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7172 /* not a regular extent, must cow */
7176 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7179 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7180 if (extent_end <= offset)
7183 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7184 if (disk_bytenr == 0)
7187 if (btrfs_file_extent_compression(leaf, fi) ||
7188 btrfs_file_extent_encryption(leaf, fi) ||
7189 btrfs_file_extent_other_encoding(leaf, fi))
7193 * Do the same check as in btrfs_cross_ref_exist but without the
7194 * unnecessary search.
7197 (btrfs_file_extent_generation(leaf, fi) <=
7198 btrfs_root_last_snapshot(&root->root_item)))
7201 backref_offset = btrfs_file_extent_offset(leaf, fi);
7204 *orig_start = key.offset - backref_offset;
7205 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7206 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7209 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7212 num_bytes = min(offset + *len, extent_end) - offset;
7213 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7216 range_end = round_up(offset + num_bytes,
7217 root->fs_info->sectorsize) - 1;
7218 ret = test_range_bit(io_tree, offset, range_end,
7219 EXTENT_DELALLOC, 0, NULL);
7226 btrfs_release_path(path);
7229 * look for other files referencing this extent, if we
7230 * find any we must cow
7233 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7234 key.offset - backref_offset, disk_bytenr,
7242 * adjust disk_bytenr and num_bytes to cover just the bytes
7243 * in this extent we are about to write. If there
7244 * are any csums in that range we have to cow in order
7245 * to keep the csums correct
7247 disk_bytenr += backref_offset;
7248 disk_bytenr += offset - key.offset;
7249 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7252 * all of the above have passed, it is safe to overwrite this extent
7258 btrfs_free_path(path);
7262 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7263 struct extent_state **cached_state, bool writing)
7265 struct btrfs_ordered_extent *ordered;
7269 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7272 * We're concerned with the entire range that we're going to be
7273 * doing DIO to, so we need to make sure there's no ordered
7274 * extents in this range.
7276 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7277 lockend - lockstart + 1);
7280 * We need to make sure there are no buffered pages in this
7281 * range either, we could have raced between the invalidate in
7282 * generic_file_direct_write and locking the extent. The
7283 * invalidate needs to happen so that reads after a write do not
7287 (!writing || !filemap_range_has_page(inode->i_mapping,
7288 lockstart, lockend)))
7291 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7296 * If we are doing a DIO read and the ordered extent we
7297 * found is for a buffered write, we can not wait for it
7298 * to complete and retry, because if we do so we can
7299 * deadlock with concurrent buffered writes on page
7300 * locks. This happens only if our DIO read covers more
7301 * than one extent map, if at this point has already
7302 * created an ordered extent for a previous extent map
7303 * and locked its range in the inode's io tree, and a
7304 * concurrent write against that previous extent map's
7305 * range and this range started (we unlock the ranges
7306 * in the io tree only when the bios complete and
7307 * buffered writes always lock pages before attempting
7308 * to lock range in the io tree).
7311 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7312 btrfs_start_ordered_extent(ordered, 1);
7315 btrfs_put_ordered_extent(ordered);
7318 * We could trigger writeback for this range (and wait
7319 * for it to complete) and then invalidate the pages for
7320 * this range (through invalidate_inode_pages2_range()),
7321 * but that can lead us to a deadlock with a concurrent
7322 * call to readahead (a buffered read or a defrag call
7323 * triggered a readahead) on a page lock due to an
7324 * ordered dio extent we created before but did not have
7325 * yet a corresponding bio submitted (whence it can not
7326 * complete), which makes readahead wait for that
7327 * ordered extent to complete while holding a lock on
7342 /* The callers of this must take lock_extent() */
7343 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7344 u64 len, u64 orig_start, u64 block_start,
7345 u64 block_len, u64 orig_block_len,
7346 u64 ram_bytes, int compress_type,
7349 struct extent_map_tree *em_tree;
7350 struct extent_map *em;
7353 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7354 type == BTRFS_ORDERED_COMPRESSED ||
7355 type == BTRFS_ORDERED_NOCOW ||
7356 type == BTRFS_ORDERED_REGULAR);
7358 em_tree = &inode->extent_tree;
7359 em = alloc_extent_map();
7361 return ERR_PTR(-ENOMEM);
7364 em->orig_start = orig_start;
7366 em->block_len = block_len;
7367 em->block_start = block_start;
7368 em->orig_block_len = orig_block_len;
7369 em->ram_bytes = ram_bytes;
7370 em->generation = -1;
7371 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7372 if (type == BTRFS_ORDERED_PREALLOC) {
7373 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7374 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7375 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7376 em->compress_type = compress_type;
7380 btrfs_drop_extent_cache(inode, em->start,
7381 em->start + em->len - 1, 0);
7382 write_lock(&em_tree->lock);
7383 ret = add_extent_mapping(em_tree, em, 1);
7384 write_unlock(&em_tree->lock);
7386 * The caller has taken lock_extent(), who could race with us
7389 } while (ret == -EEXIST);
7392 free_extent_map(em);
7393 return ERR_PTR(ret);
7396 /* em got 2 refs now, callers needs to do free_extent_map once. */
7401 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7402 struct inode *inode,
7403 struct btrfs_dio_data *dio_data,
7406 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7407 struct extent_map *em = *map;
7411 * We don't allocate a new extent in the following cases
7413 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7415 * 2) The extent is marked as PREALLOC. We're good to go here and can
7416 * just use the extent.
7419 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7420 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7421 em->block_start != EXTENT_MAP_HOLE)) {
7423 u64 block_start, orig_start, orig_block_len, ram_bytes;
7425 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7426 type = BTRFS_ORDERED_PREALLOC;
7428 type = BTRFS_ORDERED_NOCOW;
7429 len = min(len, em->len - (start - em->start));
7430 block_start = em->block_start + (start - em->start);
7432 if (can_nocow_extent(inode, start, &len, &orig_start,
7433 &orig_block_len, &ram_bytes, false) == 1 &&
7434 btrfs_inc_nocow_writers(fs_info, block_start)) {
7435 struct extent_map *em2;
7437 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7438 orig_start, block_start,
7439 len, orig_block_len,
7441 btrfs_dec_nocow_writers(fs_info, block_start);
7442 if (type == BTRFS_ORDERED_PREALLOC) {
7443 free_extent_map(em);
7447 if (em2 && IS_ERR(em2)) {
7452 * For inode marked NODATACOW or extent marked PREALLOC,
7453 * use the existing or preallocated extent, so does not
7454 * need to adjust btrfs_space_info's bytes_may_use.
7456 btrfs_free_reserved_data_space_noquota(fs_info, len);
7461 /* this will cow the extent */
7462 free_extent_map(em);
7463 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7469 len = min(len, em->len - (start - em->start));
7473 * Need to update the i_size under the extent lock so buffered
7474 * readers will get the updated i_size when we unlock.
7476 if (start + len > i_size_read(inode))
7477 i_size_write(inode, start + len);
7479 dio_data->reserve -= len;
7484 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7485 loff_t length, unsigned int flags, struct iomap *iomap,
7486 struct iomap *srcmap)
7488 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7489 struct extent_map *em;
7490 struct extent_state *cached_state = NULL;
7491 struct btrfs_dio_data *dio_data = NULL;
7492 u64 lockstart, lockend;
7493 const bool write = !!(flags & IOMAP_WRITE);
7496 bool unlock_extents = false;
7499 len = min_t(u64, len, fs_info->sectorsize);
7502 lockend = start + len - 1;
7505 * The generic stuff only does filemap_write_and_wait_range, which
7506 * isn't enough if we've written compressed pages to this area, so we
7507 * need to flush the dirty pages again to make absolutely sure that any
7508 * outstanding dirty pages are on disk.
7510 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7511 &BTRFS_I(inode)->runtime_flags)) {
7512 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7513 start + length - 1);
7518 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7522 dio_data->length = length;
7524 dio_data->reserve = round_up(length, fs_info->sectorsize);
7525 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7526 &dio_data->data_reserved,
7527 start, dio_data->reserve);
7529 extent_changeset_free(dio_data->data_reserved);
7534 iomap->private = dio_data;
7538 * If this errors out it's because we couldn't invalidate pagecache for
7539 * this range and we need to fallback to buffered.
7541 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7546 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7553 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7554 * io. INLINE is special, and we could probably kludge it in here, but
7555 * it's still buffered so for safety lets just fall back to the generic
7558 * For COMPRESSED we _have_ to read the entire extent in so we can
7559 * decompress it, so there will be buffering required no matter what we
7560 * do, so go ahead and fallback to buffered.
7562 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7563 * to buffered IO. Don't blame me, this is the price we pay for using
7566 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7567 em->block_start == EXTENT_MAP_INLINE) {
7568 free_extent_map(em);
7573 len = min(len, em->len - (start - em->start));
7575 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7579 unlock_extents = true;
7580 /* Recalc len in case the new em is smaller than requested */
7581 len = min(len, em->len - (start - em->start));
7584 * We need to unlock only the end area that we aren't using.
7585 * The rest is going to be unlocked by the endio routine.
7587 lockstart = start + len;
7588 if (lockstart < lockend)
7589 unlock_extents = true;
7593 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7594 lockstart, lockend, &cached_state);
7596 free_extent_state(cached_state);
7599 * Translate extent map information to iomap.
7600 * We trim the extents (and move the addr) even though iomap code does
7601 * that, since we have locked only the parts we are performing I/O in.
7603 if ((em->block_start == EXTENT_MAP_HOLE) ||
7604 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7605 iomap->addr = IOMAP_NULL_ADDR;
7606 iomap->type = IOMAP_HOLE;
7608 iomap->addr = em->block_start + (start - em->start);
7609 iomap->type = IOMAP_MAPPED;
7611 iomap->offset = start;
7612 iomap->bdev = fs_info->fs_devices->latest_bdev;
7613 iomap->length = len;
7615 free_extent_map(em);
7620 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7624 btrfs_delalloc_release_space(BTRFS_I(inode),
7625 dio_data->data_reserved, start,
7626 dio_data->reserve, true);
7627 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7628 extent_changeset_free(dio_data->data_reserved);
7634 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7635 ssize_t written, unsigned int flags, struct iomap *iomap)
7638 struct btrfs_dio_data *dio_data = iomap->private;
7639 size_t submitted = dio_data->submitted;
7640 const bool write = !!(flags & IOMAP_WRITE);
7642 if (!write && (iomap->type == IOMAP_HOLE)) {
7643 /* If reading from a hole, unlock and return */
7644 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7648 if (submitted < length) {
7650 length -= submitted;
7652 __endio_write_update_ordered(BTRFS_I(inode), pos,
7655 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7661 if (dio_data->reserve)
7662 btrfs_delalloc_release_space(BTRFS_I(inode),
7663 dio_data->data_reserved, pos,
7664 dio_data->reserve, true);
7665 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7666 extent_changeset_free(dio_data->data_reserved);
7670 iomap->private = NULL;
7675 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7678 * This implies a barrier so that stores to dio_bio->bi_status before
7679 * this and loads of dio_bio->bi_status after this are fully ordered.
7681 if (!refcount_dec_and_test(&dip->refs))
7684 if (bio_op(dip->dio_bio) == REQ_OP_WRITE) {
7685 __endio_write_update_ordered(BTRFS_I(dip->inode),
7686 dip->logical_offset,
7688 !dip->dio_bio->bi_status);
7690 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7691 dip->logical_offset,
7692 dip->logical_offset + dip->bytes - 1);
7695 bio_endio(dip->dio_bio);
7699 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7701 unsigned long bio_flags)
7703 struct btrfs_dio_private *dip = bio->bi_private;
7704 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7707 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7709 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7713 refcount_inc(&dip->refs);
7714 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7716 refcount_dec(&dip->refs);
7720 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7721 struct btrfs_io_bio *io_bio,
7722 const bool uptodate)
7724 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7725 const u32 sectorsize = fs_info->sectorsize;
7726 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7727 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7728 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7729 struct bio_vec bvec;
7730 struct bvec_iter iter;
7731 u64 start = io_bio->logical;
7733 blk_status_t err = BLK_STS_OK;
7735 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7736 unsigned int i, nr_sectors, pgoff;
7738 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7739 pgoff = bvec.bv_offset;
7740 for (i = 0; i < nr_sectors; i++) {
7741 ASSERT(pgoff < PAGE_SIZE);
7743 (!csum || !check_data_csum(inode, io_bio,
7744 bio_offset, bvec.bv_page, pgoff))) {
7745 clean_io_failure(fs_info, failure_tree, io_tree,
7746 start, bvec.bv_page,
7747 btrfs_ino(BTRFS_I(inode)),
7750 blk_status_t status;
7752 ASSERT((start - io_bio->logical) < UINT_MAX);
7753 status = btrfs_submit_read_repair(inode,
7755 start - io_bio->logical,
7756 bvec.bv_page, pgoff,
7758 start + sectorsize - 1,
7760 submit_dio_repair_bio);
7764 start += sectorsize;
7765 ASSERT(bio_offset + sectorsize > bio_offset);
7766 bio_offset += sectorsize;
7767 pgoff += sectorsize;
7773 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7774 const u64 offset, const u64 bytes,
7775 const bool uptodate)
7777 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7778 struct btrfs_ordered_extent *ordered = NULL;
7779 struct btrfs_workqueue *wq;
7780 u64 ordered_offset = offset;
7781 u64 ordered_bytes = bytes;
7784 if (btrfs_is_free_space_inode(inode))
7785 wq = fs_info->endio_freespace_worker;
7787 wq = fs_info->endio_write_workers;
7789 while (ordered_offset < offset + bytes) {
7790 last_offset = ordered_offset;
7791 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7795 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7797 btrfs_queue_work(wq, &ordered->work);
7800 /* No ordered extent found in the range, exit */
7801 if (ordered_offset == last_offset)
7804 * Our bio might span multiple ordered extents. In this case
7805 * we keep going until we have accounted the whole dio.
7807 if (ordered_offset < offset + bytes) {
7808 ordered_bytes = offset + bytes - ordered_offset;
7814 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7816 u64 dio_file_offset)
7818 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
7821 static void btrfs_end_dio_bio(struct bio *bio)
7823 struct btrfs_dio_private *dip = bio->bi_private;
7824 blk_status_t err = bio->bi_status;
7827 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7828 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7829 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7830 bio->bi_opf, bio->bi_iter.bi_sector,
7831 bio->bi_iter.bi_size, err);
7833 if (bio_op(bio) == REQ_OP_READ) {
7834 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
7839 dip->dio_bio->bi_status = err;
7842 btrfs_dio_private_put(dip);
7845 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7846 struct inode *inode, u64 file_offset, int async_submit)
7848 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7849 struct btrfs_dio_private *dip = bio->bi_private;
7850 bool write = bio_op(bio) == REQ_OP_WRITE;
7853 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7855 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7858 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7863 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7866 if (write && async_submit) {
7867 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
7868 btrfs_submit_bio_start_direct_io);
7872 * If we aren't doing async submit, calculate the csum of the
7875 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
7881 csum_offset = file_offset - dip->logical_offset;
7882 csum_offset >>= fs_info->sectorsize_bits;
7883 csum_offset *= fs_info->csum_size;
7884 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
7887 ret = btrfs_map_bio(fs_info, bio, 0);
7893 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7894 * or ordered extents whether or not we submit any bios.
7896 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7897 struct inode *inode,
7900 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7901 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7903 struct btrfs_dio_private *dip;
7905 dip_size = sizeof(*dip);
7906 if (!write && csum) {
7907 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7910 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
7911 dip_size += fs_info->csum_size * nblocks;
7914 dip = kzalloc(dip_size, GFP_NOFS);
7919 dip->logical_offset = file_offset;
7920 dip->bytes = dio_bio->bi_iter.bi_size;
7921 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
7922 dip->dio_bio = dio_bio;
7923 refcount_set(&dip->refs, 1);
7927 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
7928 struct bio *dio_bio, loff_t file_offset)
7930 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7931 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7932 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7933 BTRFS_BLOCK_GROUP_RAID56_MASK);
7934 struct btrfs_dio_private *dip;
7937 int async_submit = 0;
7939 int clone_offset = 0;
7943 blk_status_t status;
7944 struct btrfs_io_geometry geom;
7945 struct btrfs_dio_data *dio_data = iomap->private;
7946 struct extent_map *em = NULL;
7948 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7951 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7952 file_offset + dio_bio->bi_iter.bi_size - 1);
7954 dio_bio->bi_status = BLK_STS_RESOURCE;
7956 return BLK_QC_T_NONE;
7961 * Load the csums up front to reduce csum tree searches and
7962 * contention when submitting bios.
7964 * If we have csums disabled this will do nothing.
7966 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
7967 if (status != BLK_STS_OK)
7971 start_sector = dio_bio->bi_iter.bi_sector;
7972 submit_len = dio_bio->bi_iter.bi_size;
7975 logical = start_sector << 9;
7976 em = btrfs_get_chunk_map(fs_info, logical, submit_len);
7978 status = errno_to_blk_status(PTR_ERR(em));
7982 ret = btrfs_get_io_geometry(fs_info, em, btrfs_op(dio_bio),
7983 logical, submit_len, &geom);
7985 status = errno_to_blk_status(ret);
7988 ASSERT(geom.len <= INT_MAX);
7990 clone_len = min_t(int, submit_len, geom.len);
7993 * This will never fail as it's passing GPF_NOFS and
7994 * the allocation is backed by btrfs_bioset.
7996 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7997 bio->bi_private = dip;
7998 bio->bi_end_io = btrfs_end_dio_bio;
7999 btrfs_io_bio(bio)->logical = file_offset;
8001 ASSERT(submit_len >= clone_len);
8002 submit_len -= clone_len;
8005 * Increase the count before we submit the bio so we know
8006 * the end IO handler won't happen before we increase the
8007 * count. Otherwise, the dip might get freed before we're
8008 * done setting it up.
8010 * We transfer the initial reference to the last bio, so we
8011 * don't need to increment the reference count for the last one.
8013 if (submit_len > 0) {
8014 refcount_inc(&dip->refs);
8016 * If we are submitting more than one bio, submit them
8017 * all asynchronously. The exception is RAID 5 or 6, as
8018 * asynchronous checksums make it difficult to collect
8019 * full stripe writes.
8025 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8030 refcount_dec(&dip->refs);
8034 dio_data->submitted += clone_len;
8035 clone_offset += clone_len;
8036 start_sector += clone_len >> 9;
8037 file_offset += clone_len;
8039 free_extent_map(em);
8040 } while (submit_len > 0);
8041 return BLK_QC_T_NONE;
8044 free_extent_map(em);
8046 dip->dio_bio->bi_status = status;
8047 btrfs_dio_private_put(dip);
8049 return BLK_QC_T_NONE;
8052 const struct iomap_ops btrfs_dio_iomap_ops = {
8053 .iomap_begin = btrfs_dio_iomap_begin,
8054 .iomap_end = btrfs_dio_iomap_end,
8057 const struct iomap_dio_ops btrfs_dio_ops = {
8058 .submit_io = btrfs_submit_direct,
8061 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8066 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8070 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8073 int btrfs_readpage(struct file *file, struct page *page)
8075 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8076 u64 start = page_offset(page);
8077 u64 end = start + PAGE_SIZE - 1;
8078 unsigned long bio_flags = 0;
8079 struct bio *bio = NULL;
8082 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8084 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8086 ret = submit_one_bio(bio, 0, bio_flags);
8090 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8092 struct inode *inode = page->mapping->host;
8095 if (current->flags & PF_MEMALLOC) {
8096 redirty_page_for_writepage(wbc, page);
8102 * If we are under memory pressure we will call this directly from the
8103 * VM, we need to make sure we have the inode referenced for the ordered
8104 * extent. If not just return like we didn't do anything.
8106 if (!igrab(inode)) {
8107 redirty_page_for_writepage(wbc, page);
8108 return AOP_WRITEPAGE_ACTIVATE;
8110 ret = extent_write_full_page(page, wbc);
8111 btrfs_add_delayed_iput(inode);
8115 static int btrfs_writepages(struct address_space *mapping,
8116 struct writeback_control *wbc)
8118 return extent_writepages(mapping, wbc);
8121 static void btrfs_readahead(struct readahead_control *rac)
8123 extent_readahead(rac);
8126 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8128 int ret = try_release_extent_mapping(page, gfp_flags);
8130 clear_page_extent_mapped(page);
8134 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8136 if (PageWriteback(page) || PageDirty(page))
8138 return __btrfs_releasepage(page, gfp_flags);
8141 #ifdef CONFIG_MIGRATION
8142 static int btrfs_migratepage(struct address_space *mapping,
8143 struct page *newpage, struct page *page,
8144 enum migrate_mode mode)
8148 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8149 if (ret != MIGRATEPAGE_SUCCESS)
8152 if (page_has_private(page))
8153 attach_page_private(newpage, detach_page_private(page));
8155 if (PagePrivate2(page)) {
8156 ClearPagePrivate2(page);
8157 SetPagePrivate2(newpage);
8160 if (mode != MIGRATE_SYNC_NO_COPY)
8161 migrate_page_copy(newpage, page);
8163 migrate_page_states(newpage, page);
8164 return MIGRATEPAGE_SUCCESS;
8168 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8169 unsigned int length)
8171 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8172 struct extent_io_tree *tree = &inode->io_tree;
8173 struct btrfs_ordered_extent *ordered;
8174 struct extent_state *cached_state = NULL;
8175 u64 page_start = page_offset(page);
8176 u64 page_end = page_start + PAGE_SIZE - 1;
8179 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8180 bool found_ordered = false;
8181 bool completed_ordered = false;
8184 * we have the page locked, so new writeback can't start,
8185 * and the dirty bit won't be cleared while we are here.
8187 * Wait for IO on this page so that we can safely clear
8188 * the PagePrivate2 bit and do ordered accounting
8190 wait_on_page_writeback(page);
8193 btrfs_releasepage(page, GFP_NOFS);
8197 if (!inode_evicting)
8198 lock_extent_bits(tree, page_start, page_end, &cached_state);
8202 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8204 found_ordered = true;
8206 ordered->file_offset + ordered->num_bytes - 1);
8208 * IO on this page will never be started, so we need to account
8209 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8210 * here, must leave that up for the ordered extent completion.
8212 if (!inode_evicting)
8213 clear_extent_bit(tree, start, end,
8215 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8216 EXTENT_DEFRAG, 1, 0, &cached_state);
8218 * whoever cleared the private bit is responsible
8219 * for the finish_ordered_io
8221 if (TestClearPagePrivate2(page)) {
8222 struct btrfs_ordered_inode_tree *tree;
8225 tree = &inode->ordered_tree;
8227 spin_lock_irq(&tree->lock);
8228 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8229 new_len = start - ordered->file_offset;
8230 if (new_len < ordered->truncated_len)
8231 ordered->truncated_len = new_len;
8232 spin_unlock_irq(&tree->lock);
8234 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8236 end - start + 1, 1)) {
8237 btrfs_finish_ordered_io(ordered);
8238 completed_ordered = true;
8241 btrfs_put_ordered_extent(ordered);
8242 if (!inode_evicting) {
8243 cached_state = NULL;
8244 lock_extent_bits(tree, start, end,
8249 if (start < page_end)
8254 * Qgroup reserved space handler
8255 * Page here will be either
8256 * 1) Already written to disk or ordered extent already submitted
8257 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8258 * Qgroup will be handled by its qgroup_record then.
8259 * btrfs_qgroup_free_data() call will do nothing here.
8261 * 2) Not written to disk yet
8262 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8263 * bit of its io_tree, and free the qgroup reserved data space.
8264 * Since the IO will never happen for this page.
8266 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8267 if (!inode_evicting) {
8271 * If there's an ordered extent for this range and we have not
8272 * finished it ourselves, we must leave EXTENT_DELALLOC_NEW set
8273 * in the range for the ordered extent completion. We must also
8274 * not delete the range, otherwise we would lose that bit (and
8275 * any other bits set in the range). Make sure EXTENT_UPTODATE
8276 * is cleared if we don't delete, otherwise it can lead to
8277 * corruptions if the i_size is extented later.
8279 if (found_ordered && !completed_ordered)
8281 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8282 EXTENT_DELALLOC | EXTENT_UPTODATE |
8283 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8284 delete, &cached_state);
8286 __btrfs_releasepage(page, GFP_NOFS);
8289 ClearPageChecked(page);
8290 clear_page_extent_mapped(page);
8294 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8295 * called from a page fault handler when a page is first dirtied. Hence we must
8296 * be careful to check for EOF conditions here. We set the page up correctly
8297 * for a written page which means we get ENOSPC checking when writing into
8298 * holes and correct delalloc and unwritten extent mapping on filesystems that
8299 * support these features.
8301 * We are not allowed to take the i_mutex here so we have to play games to
8302 * protect against truncate races as the page could now be beyond EOF. Because
8303 * truncate_setsize() writes the inode size before removing pages, once we have
8304 * the page lock we can determine safely if the page is beyond EOF. If it is not
8305 * beyond EOF, then the page is guaranteed safe against truncation until we
8308 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8310 struct page *page = vmf->page;
8311 struct inode *inode = file_inode(vmf->vma->vm_file);
8312 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8313 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8314 struct btrfs_ordered_extent *ordered;
8315 struct extent_state *cached_state = NULL;
8316 struct extent_changeset *data_reserved = NULL;
8318 unsigned long zero_start;
8328 reserved_space = PAGE_SIZE;
8330 sb_start_pagefault(inode->i_sb);
8331 page_start = page_offset(page);
8332 page_end = page_start + PAGE_SIZE - 1;
8336 * Reserving delalloc space after obtaining the page lock can lead to
8337 * deadlock. For example, if a dirty page is locked by this function
8338 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8339 * dirty page write out, then the btrfs_writepage() function could
8340 * end up waiting indefinitely to get a lock on the page currently
8341 * being processed by btrfs_page_mkwrite() function.
8343 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8344 page_start, reserved_space);
8346 ret2 = file_update_time(vmf->vma->vm_file);
8350 ret = vmf_error(ret2);
8356 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8359 size = i_size_read(inode);
8361 if ((page->mapping != inode->i_mapping) ||
8362 (page_start >= size)) {
8363 /* page got truncated out from underneath us */
8366 wait_on_page_writeback(page);
8368 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8369 ret2 = set_page_extent_mapped(page);
8371 ret = vmf_error(ret2);
8372 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8377 * we can't set the delalloc bits if there are pending ordered
8378 * extents. Drop our locks and wait for them to finish
8380 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8383 unlock_extent_cached(io_tree, page_start, page_end,
8386 btrfs_start_ordered_extent(ordered, 1);
8387 btrfs_put_ordered_extent(ordered);
8391 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8392 reserved_space = round_up(size - page_start,
8393 fs_info->sectorsize);
8394 if (reserved_space < PAGE_SIZE) {
8395 end = page_start + reserved_space - 1;
8396 btrfs_delalloc_release_space(BTRFS_I(inode),
8397 data_reserved, page_start,
8398 PAGE_SIZE - reserved_space, true);
8403 * page_mkwrite gets called when the page is firstly dirtied after it's
8404 * faulted in, but write(2) could also dirty a page and set delalloc
8405 * bits, thus in this case for space account reason, we still need to
8406 * clear any delalloc bits within this page range since we have to
8407 * reserve data&meta space before lock_page() (see above comments).
8409 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8410 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8411 EXTENT_DEFRAG, 0, 0, &cached_state);
8413 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8416 unlock_extent_cached(io_tree, page_start, page_end,
8418 ret = VM_FAULT_SIGBUS;
8422 /* page is wholly or partially inside EOF */
8423 if (page_start + PAGE_SIZE > size)
8424 zero_start = offset_in_page(size);
8426 zero_start = PAGE_SIZE;
8428 if (zero_start != PAGE_SIZE) {
8430 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8431 flush_dcache_page(page);
8434 ClearPageChecked(page);
8435 set_page_dirty(page);
8436 SetPageUptodate(page);
8438 BTRFS_I(inode)->last_trans = fs_info->generation;
8439 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8440 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8442 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8444 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8445 sb_end_pagefault(inode->i_sb);
8446 extent_changeset_free(data_reserved);
8447 return VM_FAULT_LOCKED;
8452 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8453 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8454 reserved_space, (ret != 0));
8456 sb_end_pagefault(inode->i_sb);
8457 extent_changeset_free(data_reserved);
8461 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8463 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8464 struct btrfs_root *root = BTRFS_I(inode)->root;
8465 struct btrfs_block_rsv *rsv;
8467 struct btrfs_trans_handle *trans;
8468 u64 mask = fs_info->sectorsize - 1;
8469 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8471 if (!skip_writeback) {
8472 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8479 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8480 * things going on here:
8482 * 1) We need to reserve space to update our inode.
8484 * 2) We need to have something to cache all the space that is going to
8485 * be free'd up by the truncate operation, but also have some slack
8486 * space reserved in case it uses space during the truncate (thank you
8487 * very much snapshotting).
8489 * And we need these to be separate. The fact is we can use a lot of
8490 * space doing the truncate, and we have no earthly idea how much space
8491 * we will use, so we need the truncate reservation to be separate so it
8492 * doesn't end up using space reserved for updating the inode. We also
8493 * need to be able to stop the transaction and start a new one, which
8494 * means we need to be able to update the inode several times, and we
8495 * have no idea of knowing how many times that will be, so we can't just
8496 * reserve 1 item for the entirety of the operation, so that has to be
8497 * done separately as well.
8499 * So that leaves us with
8501 * 1) rsv - for the truncate reservation, which we will steal from the
8502 * transaction reservation.
8503 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8504 * updating the inode.
8506 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8509 rsv->size = min_size;
8513 * 1 for the truncate slack space
8514 * 1 for updating the inode.
8516 trans = btrfs_start_transaction(root, 2);
8517 if (IS_ERR(trans)) {
8518 ret = PTR_ERR(trans);
8522 /* Migrate the slack space for the truncate to our reserve */
8523 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8528 * So if we truncate and then write and fsync we normally would just
8529 * write the extents that changed, which is a problem if we need to
8530 * first truncate that entire inode. So set this flag so we write out
8531 * all of the extents in the inode to the sync log so we're completely
8534 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8535 trans->block_rsv = rsv;
8538 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8540 BTRFS_EXTENT_DATA_KEY);
8541 trans->block_rsv = &fs_info->trans_block_rsv;
8542 if (ret != -ENOSPC && ret != -EAGAIN)
8545 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8549 btrfs_end_transaction(trans);
8550 btrfs_btree_balance_dirty(fs_info);
8552 trans = btrfs_start_transaction(root, 2);
8553 if (IS_ERR(trans)) {
8554 ret = PTR_ERR(trans);
8559 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8560 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8561 rsv, min_size, false);
8562 BUG_ON(ret); /* shouldn't happen */
8563 trans->block_rsv = rsv;
8567 * We can't call btrfs_truncate_block inside a trans handle as we could
8568 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8569 * we've truncated everything except the last little bit, and can do
8570 * btrfs_truncate_block and then update the disk_i_size.
8572 if (ret == NEED_TRUNCATE_BLOCK) {
8573 btrfs_end_transaction(trans);
8574 btrfs_btree_balance_dirty(fs_info);
8576 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8579 trans = btrfs_start_transaction(root, 1);
8580 if (IS_ERR(trans)) {
8581 ret = PTR_ERR(trans);
8584 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8590 trans->block_rsv = &fs_info->trans_block_rsv;
8591 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8595 ret2 = btrfs_end_transaction(trans);
8598 btrfs_btree_balance_dirty(fs_info);
8601 btrfs_free_block_rsv(fs_info, rsv);
8607 * create a new subvolume directory/inode (helper for the ioctl).
8609 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8610 struct btrfs_root *new_root,
8611 struct btrfs_root *parent_root)
8613 struct inode *inode;
8618 err = btrfs_get_free_objectid(new_root, &ino);
8622 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino,
8623 S_IFDIR | (~current_umask() & S_IRWXUGO),
8626 return PTR_ERR(inode);
8627 inode->i_op = &btrfs_dir_inode_operations;
8628 inode->i_fop = &btrfs_dir_file_operations;
8630 set_nlink(inode, 1);
8631 btrfs_i_size_write(BTRFS_I(inode), 0);
8632 unlock_new_inode(inode);
8634 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8636 btrfs_err(new_root->fs_info,
8637 "error inheriting subvolume %llu properties: %d",
8638 new_root->root_key.objectid, err);
8640 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8646 struct inode *btrfs_alloc_inode(struct super_block *sb)
8648 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8649 struct btrfs_inode *ei;
8650 struct inode *inode;
8652 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8659 ei->last_sub_trans = 0;
8660 ei->logged_trans = 0;
8661 ei->delalloc_bytes = 0;
8662 ei->new_delalloc_bytes = 0;
8663 ei->defrag_bytes = 0;
8664 ei->disk_i_size = 0;
8667 ei->index_cnt = (u64)-1;
8669 ei->last_unlink_trans = 0;
8670 ei->last_reflink_trans = 0;
8671 ei->last_log_commit = 0;
8673 spin_lock_init(&ei->lock);
8674 ei->outstanding_extents = 0;
8675 if (sb->s_magic != BTRFS_TEST_MAGIC)
8676 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8677 BTRFS_BLOCK_RSV_DELALLOC);
8678 ei->runtime_flags = 0;
8679 ei->prop_compress = BTRFS_COMPRESS_NONE;
8680 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8682 ei->delayed_node = NULL;
8684 ei->i_otime.tv_sec = 0;
8685 ei->i_otime.tv_nsec = 0;
8687 inode = &ei->vfs_inode;
8688 extent_map_tree_init(&ei->extent_tree);
8689 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8690 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8691 IO_TREE_INODE_IO_FAILURE, inode);
8692 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8693 IO_TREE_INODE_FILE_EXTENT, inode);
8694 ei->io_tree.track_uptodate = true;
8695 ei->io_failure_tree.track_uptodate = true;
8696 atomic_set(&ei->sync_writers, 0);
8697 mutex_init(&ei->log_mutex);
8698 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8699 INIT_LIST_HEAD(&ei->delalloc_inodes);
8700 INIT_LIST_HEAD(&ei->delayed_iput);
8701 RB_CLEAR_NODE(&ei->rb_node);
8706 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8707 void btrfs_test_destroy_inode(struct inode *inode)
8709 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8710 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8714 void btrfs_free_inode(struct inode *inode)
8716 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8719 void btrfs_destroy_inode(struct inode *vfs_inode)
8721 struct btrfs_ordered_extent *ordered;
8722 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8723 struct btrfs_root *root = inode->root;
8725 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8726 WARN_ON(vfs_inode->i_data.nrpages);
8727 WARN_ON(inode->block_rsv.reserved);
8728 WARN_ON(inode->block_rsv.size);
8729 WARN_ON(inode->outstanding_extents);
8730 WARN_ON(inode->delalloc_bytes);
8731 WARN_ON(inode->new_delalloc_bytes);
8732 WARN_ON(inode->csum_bytes);
8733 WARN_ON(inode->defrag_bytes);
8736 * This can happen where we create an inode, but somebody else also
8737 * created the same inode and we need to destroy the one we already
8744 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8748 btrfs_err(root->fs_info,
8749 "found ordered extent %llu %llu on inode cleanup",
8750 ordered->file_offset, ordered->num_bytes);
8751 btrfs_remove_ordered_extent(inode, ordered);
8752 btrfs_put_ordered_extent(ordered);
8753 btrfs_put_ordered_extent(ordered);
8756 btrfs_qgroup_check_reserved_leak(inode);
8757 inode_tree_del(inode);
8758 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8759 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8760 btrfs_put_root(inode->root);
8763 int btrfs_drop_inode(struct inode *inode)
8765 struct btrfs_root *root = BTRFS_I(inode)->root;
8770 /* the snap/subvol tree is on deleting */
8771 if (btrfs_root_refs(&root->root_item) == 0)
8774 return generic_drop_inode(inode);
8777 static void init_once(void *foo)
8779 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8781 inode_init_once(&ei->vfs_inode);
8784 void __cold btrfs_destroy_cachep(void)
8787 * Make sure all delayed rcu free inodes are flushed before we
8791 kmem_cache_destroy(btrfs_inode_cachep);
8792 kmem_cache_destroy(btrfs_trans_handle_cachep);
8793 kmem_cache_destroy(btrfs_path_cachep);
8794 kmem_cache_destroy(btrfs_free_space_cachep);
8795 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8798 int __init btrfs_init_cachep(void)
8800 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8801 sizeof(struct btrfs_inode), 0,
8802 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8804 if (!btrfs_inode_cachep)
8807 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8808 sizeof(struct btrfs_trans_handle), 0,
8809 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8810 if (!btrfs_trans_handle_cachep)
8813 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8814 sizeof(struct btrfs_path), 0,
8815 SLAB_MEM_SPREAD, NULL);
8816 if (!btrfs_path_cachep)
8819 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8820 sizeof(struct btrfs_free_space), 0,
8821 SLAB_MEM_SPREAD, NULL);
8822 if (!btrfs_free_space_cachep)
8825 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8826 PAGE_SIZE, PAGE_SIZE,
8827 SLAB_RED_ZONE, NULL);
8828 if (!btrfs_free_space_bitmap_cachep)
8833 btrfs_destroy_cachep();
8837 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8838 u32 request_mask, unsigned int flags)
8842 struct inode *inode = d_inode(path->dentry);
8843 u32 blocksize = inode->i_sb->s_blocksize;
8844 u32 bi_flags = BTRFS_I(inode)->flags;
8846 stat->result_mask |= STATX_BTIME;
8847 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8848 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8849 if (bi_flags & BTRFS_INODE_APPEND)
8850 stat->attributes |= STATX_ATTR_APPEND;
8851 if (bi_flags & BTRFS_INODE_COMPRESS)
8852 stat->attributes |= STATX_ATTR_COMPRESSED;
8853 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8854 stat->attributes |= STATX_ATTR_IMMUTABLE;
8855 if (bi_flags & BTRFS_INODE_NODUMP)
8856 stat->attributes |= STATX_ATTR_NODUMP;
8858 stat->attributes_mask |= (STATX_ATTR_APPEND |
8859 STATX_ATTR_COMPRESSED |
8860 STATX_ATTR_IMMUTABLE |
8863 generic_fillattr(inode, stat);
8864 stat->dev = BTRFS_I(inode)->root->anon_dev;
8866 spin_lock(&BTRFS_I(inode)->lock);
8867 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8868 inode_bytes = inode_get_bytes(inode);
8869 spin_unlock(&BTRFS_I(inode)->lock);
8870 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8871 ALIGN(delalloc_bytes, blocksize)) >> 9;
8875 static int btrfs_rename_exchange(struct inode *old_dir,
8876 struct dentry *old_dentry,
8877 struct inode *new_dir,
8878 struct dentry *new_dentry)
8880 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8881 struct btrfs_trans_handle *trans;
8882 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8883 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8884 struct inode *new_inode = new_dentry->d_inode;
8885 struct inode *old_inode = old_dentry->d_inode;
8886 struct timespec64 ctime = current_time(old_inode);
8887 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8888 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8893 bool root_log_pinned = false;
8894 bool dest_log_pinned = false;
8896 /* we only allow rename subvolume link between subvolumes */
8897 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8900 /* close the race window with snapshot create/destroy ioctl */
8901 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8902 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8903 down_read(&fs_info->subvol_sem);
8906 * We want to reserve the absolute worst case amount of items. So if
8907 * both inodes are subvols and we need to unlink them then that would
8908 * require 4 item modifications, but if they are both normal inodes it
8909 * would require 5 item modifications, so we'll assume their normal
8910 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
8911 * should cover the worst case number of items we'll modify.
8913 trans = btrfs_start_transaction(root, 12);
8914 if (IS_ERR(trans)) {
8915 ret = PTR_ERR(trans);
8920 btrfs_record_root_in_trans(trans, dest);
8923 * We need to find a free sequence number both in the source and
8924 * in the destination directory for the exchange.
8926 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8929 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8933 BTRFS_I(old_inode)->dir_index = 0ULL;
8934 BTRFS_I(new_inode)->dir_index = 0ULL;
8936 /* Reference for the source. */
8937 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8938 /* force full log commit if subvolume involved. */
8939 btrfs_set_log_full_commit(trans);
8941 btrfs_pin_log_trans(root);
8942 root_log_pinned = true;
8943 ret = btrfs_insert_inode_ref(trans, dest,
8944 new_dentry->d_name.name,
8945 new_dentry->d_name.len,
8947 btrfs_ino(BTRFS_I(new_dir)),
8953 /* And now for the dest. */
8954 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8955 /* force full log commit if subvolume involved. */
8956 btrfs_set_log_full_commit(trans);
8958 btrfs_pin_log_trans(dest);
8959 dest_log_pinned = true;
8960 ret = btrfs_insert_inode_ref(trans, root,
8961 old_dentry->d_name.name,
8962 old_dentry->d_name.len,
8964 btrfs_ino(BTRFS_I(old_dir)),
8970 /* Update inode version and ctime/mtime. */
8971 inode_inc_iversion(old_dir);
8972 inode_inc_iversion(new_dir);
8973 inode_inc_iversion(old_inode);
8974 inode_inc_iversion(new_inode);
8975 old_dir->i_ctime = old_dir->i_mtime = ctime;
8976 new_dir->i_ctime = new_dir->i_mtime = ctime;
8977 old_inode->i_ctime = ctime;
8978 new_inode->i_ctime = ctime;
8980 if (old_dentry->d_parent != new_dentry->d_parent) {
8981 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8982 BTRFS_I(old_inode), 1);
8983 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8984 BTRFS_I(new_inode), 1);
8987 /* src is a subvolume */
8988 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8989 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
8990 } else { /* src is an inode */
8991 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
8992 BTRFS_I(old_dentry->d_inode),
8993 old_dentry->d_name.name,
8994 old_dentry->d_name.len);
8996 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8999 btrfs_abort_transaction(trans, ret);
9003 /* dest is a subvolume */
9004 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9005 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9006 } else { /* dest is an inode */
9007 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9008 BTRFS_I(new_dentry->d_inode),
9009 new_dentry->d_name.name,
9010 new_dentry->d_name.len);
9012 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9015 btrfs_abort_transaction(trans, ret);
9019 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9020 new_dentry->d_name.name,
9021 new_dentry->d_name.len, 0, old_idx);
9023 btrfs_abort_transaction(trans, ret);
9027 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9028 old_dentry->d_name.name,
9029 old_dentry->d_name.len, 0, new_idx);
9031 btrfs_abort_transaction(trans, ret);
9035 if (old_inode->i_nlink == 1)
9036 BTRFS_I(old_inode)->dir_index = old_idx;
9037 if (new_inode->i_nlink == 1)
9038 BTRFS_I(new_inode)->dir_index = new_idx;
9040 if (root_log_pinned) {
9041 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9042 new_dentry->d_parent);
9043 btrfs_end_log_trans(root);
9044 root_log_pinned = false;
9046 if (dest_log_pinned) {
9047 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9048 old_dentry->d_parent);
9049 btrfs_end_log_trans(dest);
9050 dest_log_pinned = false;
9054 * If we have pinned a log and an error happened, we unpin tasks
9055 * trying to sync the log and force them to fallback to a transaction
9056 * commit if the log currently contains any of the inodes involved in
9057 * this rename operation (to ensure we do not persist a log with an
9058 * inconsistent state for any of these inodes or leading to any
9059 * inconsistencies when replayed). If the transaction was aborted, the
9060 * abortion reason is propagated to userspace when attempting to commit
9061 * the transaction. If the log does not contain any of these inodes, we
9062 * allow the tasks to sync it.
9064 if (ret && (root_log_pinned || dest_log_pinned)) {
9065 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9066 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9067 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9069 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9070 btrfs_set_log_full_commit(trans);
9072 if (root_log_pinned) {
9073 btrfs_end_log_trans(root);
9074 root_log_pinned = false;
9076 if (dest_log_pinned) {
9077 btrfs_end_log_trans(dest);
9078 dest_log_pinned = false;
9081 ret2 = btrfs_end_transaction(trans);
9082 ret = ret ? ret : ret2;
9084 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9085 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9086 up_read(&fs_info->subvol_sem);
9091 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9092 struct btrfs_root *root,
9094 struct dentry *dentry)
9097 struct inode *inode;
9101 ret = btrfs_get_free_objectid(root, &objectid);
9105 inode = btrfs_new_inode(trans, root, dir,
9106 dentry->d_name.name,
9108 btrfs_ino(BTRFS_I(dir)),
9110 S_IFCHR | WHITEOUT_MODE,
9113 if (IS_ERR(inode)) {
9114 ret = PTR_ERR(inode);
9118 inode->i_op = &btrfs_special_inode_operations;
9119 init_special_inode(inode, inode->i_mode,
9122 ret = btrfs_init_inode_security(trans, inode, dir,
9127 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9128 BTRFS_I(inode), 0, index);
9132 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9134 unlock_new_inode(inode);
9136 inode_dec_link_count(inode);
9142 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9143 struct inode *new_dir, struct dentry *new_dentry,
9146 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9147 struct btrfs_trans_handle *trans;
9148 unsigned int trans_num_items;
9149 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9150 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9151 struct inode *new_inode = d_inode(new_dentry);
9152 struct inode *old_inode = d_inode(old_dentry);
9156 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9157 bool log_pinned = false;
9159 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9162 /* we only allow rename subvolume link between subvolumes */
9163 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9166 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9167 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9170 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9171 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9175 /* check for collisions, even if the name isn't there */
9176 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9177 new_dentry->d_name.name,
9178 new_dentry->d_name.len);
9181 if (ret == -EEXIST) {
9183 * eexist without a new_inode */
9184 if (WARN_ON(!new_inode)) {
9188 /* maybe -EOVERFLOW */
9195 * we're using rename to replace one file with another. Start IO on it
9196 * now so we don't add too much work to the end of the transaction
9198 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9199 filemap_flush(old_inode->i_mapping);
9201 /* close the racy window with snapshot create/destroy ioctl */
9202 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9203 down_read(&fs_info->subvol_sem);
9205 * We want to reserve the absolute worst case amount of items. So if
9206 * both inodes are subvols and we need to unlink them then that would
9207 * require 4 item modifications, but if they are both normal inodes it
9208 * would require 5 item modifications, so we'll assume they are normal
9209 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9210 * should cover the worst case number of items we'll modify.
9211 * If our rename has the whiteout flag, we need more 5 units for the
9212 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9213 * when selinux is enabled).
9215 trans_num_items = 11;
9216 if (flags & RENAME_WHITEOUT)
9217 trans_num_items += 5;
9218 trans = btrfs_start_transaction(root, trans_num_items);
9219 if (IS_ERR(trans)) {
9220 ret = PTR_ERR(trans);
9225 btrfs_record_root_in_trans(trans, dest);
9227 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9231 BTRFS_I(old_inode)->dir_index = 0ULL;
9232 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9233 /* force full log commit if subvolume involved. */
9234 btrfs_set_log_full_commit(trans);
9236 btrfs_pin_log_trans(root);
9238 ret = btrfs_insert_inode_ref(trans, dest,
9239 new_dentry->d_name.name,
9240 new_dentry->d_name.len,
9242 btrfs_ino(BTRFS_I(new_dir)), index);
9247 inode_inc_iversion(old_dir);
9248 inode_inc_iversion(new_dir);
9249 inode_inc_iversion(old_inode);
9250 old_dir->i_ctime = old_dir->i_mtime =
9251 new_dir->i_ctime = new_dir->i_mtime =
9252 old_inode->i_ctime = current_time(old_dir);
9254 if (old_dentry->d_parent != new_dentry->d_parent)
9255 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9256 BTRFS_I(old_inode), 1);
9258 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9259 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9261 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9262 BTRFS_I(d_inode(old_dentry)),
9263 old_dentry->d_name.name,
9264 old_dentry->d_name.len);
9266 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9269 btrfs_abort_transaction(trans, ret);
9274 inode_inc_iversion(new_inode);
9275 new_inode->i_ctime = current_time(new_inode);
9276 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9277 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9278 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9279 BUG_ON(new_inode->i_nlink == 0);
9281 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9282 BTRFS_I(d_inode(new_dentry)),
9283 new_dentry->d_name.name,
9284 new_dentry->d_name.len);
9286 if (!ret && new_inode->i_nlink == 0)
9287 ret = btrfs_orphan_add(trans,
9288 BTRFS_I(d_inode(new_dentry)));
9290 btrfs_abort_transaction(trans, ret);
9295 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9296 new_dentry->d_name.name,
9297 new_dentry->d_name.len, 0, index);
9299 btrfs_abort_transaction(trans, ret);
9303 if (old_inode->i_nlink == 1)
9304 BTRFS_I(old_inode)->dir_index = index;
9307 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9308 new_dentry->d_parent);
9309 btrfs_end_log_trans(root);
9313 if (flags & RENAME_WHITEOUT) {
9314 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9318 btrfs_abort_transaction(trans, ret);
9324 * If we have pinned the log and an error happened, we unpin tasks
9325 * trying to sync the log and force them to fallback to a transaction
9326 * commit if the log currently contains any of the inodes involved in
9327 * this rename operation (to ensure we do not persist a log with an
9328 * inconsistent state for any of these inodes or leading to any
9329 * inconsistencies when replayed). If the transaction was aborted, the
9330 * abortion reason is propagated to userspace when attempting to commit
9331 * the transaction. If the log does not contain any of these inodes, we
9332 * allow the tasks to sync it.
9334 if (ret && log_pinned) {
9335 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9336 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9337 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9339 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9340 btrfs_set_log_full_commit(trans);
9342 btrfs_end_log_trans(root);
9345 ret2 = btrfs_end_transaction(trans);
9346 ret = ret ? ret : ret2;
9348 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9349 up_read(&fs_info->subvol_sem);
9354 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9355 struct inode *new_dir, struct dentry *new_dentry,
9358 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9361 if (flags & RENAME_EXCHANGE)
9362 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9365 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9368 struct btrfs_delalloc_work {
9369 struct inode *inode;
9370 struct completion completion;
9371 struct list_head list;
9372 struct btrfs_work work;
9375 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9377 struct btrfs_delalloc_work *delalloc_work;
9378 struct inode *inode;
9380 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9382 inode = delalloc_work->inode;
9383 filemap_flush(inode->i_mapping);
9384 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9385 &BTRFS_I(inode)->runtime_flags))
9386 filemap_flush(inode->i_mapping);
9389 complete(&delalloc_work->completion);
9392 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9394 struct btrfs_delalloc_work *work;
9396 work = kmalloc(sizeof(*work), GFP_NOFS);
9400 init_completion(&work->completion);
9401 INIT_LIST_HEAD(&work->list);
9402 work->inode = inode;
9403 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9409 * some fairly slow code that needs optimization. This walks the list
9410 * of all the inodes with pending delalloc and forces them to disk.
9412 static int start_delalloc_inodes(struct btrfs_root *root,
9413 struct writeback_control *wbc, bool snapshot,
9414 bool in_reclaim_context)
9416 struct btrfs_inode *binode;
9417 struct inode *inode;
9418 struct btrfs_delalloc_work *work, *next;
9419 struct list_head works;
9420 struct list_head splice;
9422 bool full_flush = wbc->nr_to_write == LONG_MAX;
9424 INIT_LIST_HEAD(&works);
9425 INIT_LIST_HEAD(&splice);
9427 mutex_lock(&root->delalloc_mutex);
9428 spin_lock(&root->delalloc_lock);
9429 list_splice_init(&root->delalloc_inodes, &splice);
9430 while (!list_empty(&splice)) {
9431 binode = list_entry(splice.next, struct btrfs_inode,
9434 list_move_tail(&binode->delalloc_inodes,
9435 &root->delalloc_inodes);
9437 if (in_reclaim_context &&
9438 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9441 inode = igrab(&binode->vfs_inode);
9443 cond_resched_lock(&root->delalloc_lock);
9446 spin_unlock(&root->delalloc_lock);
9449 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9450 &binode->runtime_flags);
9452 work = btrfs_alloc_delalloc_work(inode);
9458 list_add_tail(&work->list, &works);
9459 btrfs_queue_work(root->fs_info->flush_workers,
9462 ret = sync_inode(inode, wbc);
9464 test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9465 &BTRFS_I(inode)->runtime_flags))
9466 ret = sync_inode(inode, wbc);
9467 btrfs_add_delayed_iput(inode);
9468 if (ret || wbc->nr_to_write <= 0)
9472 spin_lock(&root->delalloc_lock);
9474 spin_unlock(&root->delalloc_lock);
9477 list_for_each_entry_safe(work, next, &works, list) {
9478 list_del_init(&work->list);
9479 wait_for_completion(&work->completion);
9483 if (!list_empty(&splice)) {
9484 spin_lock(&root->delalloc_lock);
9485 list_splice_tail(&splice, &root->delalloc_inodes);
9486 spin_unlock(&root->delalloc_lock);
9488 mutex_unlock(&root->delalloc_mutex);
9492 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9494 struct writeback_control wbc = {
9495 .nr_to_write = LONG_MAX,
9496 .sync_mode = WB_SYNC_NONE,
9498 .range_end = LLONG_MAX,
9500 struct btrfs_fs_info *fs_info = root->fs_info;
9502 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9505 return start_delalloc_inodes(root, &wbc, true, false);
9508 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9509 bool in_reclaim_context)
9511 struct writeback_control wbc = {
9513 .sync_mode = WB_SYNC_NONE,
9515 .range_end = LLONG_MAX,
9517 struct btrfs_root *root;
9518 struct list_head splice;
9521 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9524 INIT_LIST_HEAD(&splice);
9526 mutex_lock(&fs_info->delalloc_root_mutex);
9527 spin_lock(&fs_info->delalloc_root_lock);
9528 list_splice_init(&fs_info->delalloc_roots, &splice);
9529 while (!list_empty(&splice)) {
9531 * Reset nr_to_write here so we know that we're doing a full
9535 wbc.nr_to_write = LONG_MAX;
9537 root = list_first_entry(&splice, struct btrfs_root,
9539 root = btrfs_grab_root(root);
9541 list_move_tail(&root->delalloc_root,
9542 &fs_info->delalloc_roots);
9543 spin_unlock(&fs_info->delalloc_root_lock);
9545 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9546 btrfs_put_root(root);
9547 if (ret < 0 || wbc.nr_to_write <= 0)
9549 spin_lock(&fs_info->delalloc_root_lock);
9551 spin_unlock(&fs_info->delalloc_root_lock);
9555 if (!list_empty(&splice)) {
9556 spin_lock(&fs_info->delalloc_root_lock);
9557 list_splice_tail(&splice, &fs_info->delalloc_roots);
9558 spin_unlock(&fs_info->delalloc_root_lock);
9560 mutex_unlock(&fs_info->delalloc_root_mutex);
9564 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9565 const char *symname)
9567 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9568 struct btrfs_trans_handle *trans;
9569 struct btrfs_root *root = BTRFS_I(dir)->root;
9570 struct btrfs_path *path;
9571 struct btrfs_key key;
9572 struct inode *inode = NULL;
9579 struct btrfs_file_extent_item *ei;
9580 struct extent_buffer *leaf;
9582 name_len = strlen(symname);
9583 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9584 return -ENAMETOOLONG;
9587 * 2 items for inode item and ref
9588 * 2 items for dir items
9589 * 1 item for updating parent inode item
9590 * 1 item for the inline extent item
9591 * 1 item for xattr if selinux is on
9593 trans = btrfs_start_transaction(root, 7);
9595 return PTR_ERR(trans);
9597 err = btrfs_get_free_objectid(root, &objectid);
9601 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9602 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9603 objectid, S_IFLNK|S_IRWXUGO, &index);
9604 if (IS_ERR(inode)) {
9605 err = PTR_ERR(inode);
9611 * If the active LSM wants to access the inode during
9612 * d_instantiate it needs these. Smack checks to see
9613 * if the filesystem supports xattrs by looking at the
9616 inode->i_fop = &btrfs_file_operations;
9617 inode->i_op = &btrfs_file_inode_operations;
9618 inode->i_mapping->a_ops = &btrfs_aops;
9620 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9624 path = btrfs_alloc_path();
9629 key.objectid = btrfs_ino(BTRFS_I(inode));
9631 key.type = BTRFS_EXTENT_DATA_KEY;
9632 datasize = btrfs_file_extent_calc_inline_size(name_len);
9633 err = btrfs_insert_empty_item(trans, root, path, &key,
9636 btrfs_free_path(path);
9639 leaf = path->nodes[0];
9640 ei = btrfs_item_ptr(leaf, path->slots[0],
9641 struct btrfs_file_extent_item);
9642 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9643 btrfs_set_file_extent_type(leaf, ei,
9644 BTRFS_FILE_EXTENT_INLINE);
9645 btrfs_set_file_extent_encryption(leaf, ei, 0);
9646 btrfs_set_file_extent_compression(leaf, ei, 0);
9647 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9648 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9650 ptr = btrfs_file_extent_inline_start(ei);
9651 write_extent_buffer(leaf, symname, ptr, name_len);
9652 btrfs_mark_buffer_dirty(leaf);
9653 btrfs_free_path(path);
9655 inode->i_op = &btrfs_symlink_inode_operations;
9656 inode_nohighmem(inode);
9657 inode_set_bytes(inode, name_len);
9658 btrfs_i_size_write(BTRFS_I(inode), name_len);
9659 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9661 * Last step, add directory indexes for our symlink inode. This is the
9662 * last step to avoid extra cleanup of these indexes if an error happens
9666 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9667 BTRFS_I(inode), 0, index);
9671 d_instantiate_new(dentry, inode);
9674 btrfs_end_transaction(trans);
9676 inode_dec_link_count(inode);
9677 discard_new_inode(inode);
9679 btrfs_btree_balance_dirty(fs_info);
9683 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9684 struct btrfs_trans_handle *trans_in,
9685 struct btrfs_inode *inode,
9686 struct btrfs_key *ins,
9689 struct btrfs_file_extent_item stack_fi;
9690 struct btrfs_replace_extent_info extent_info;
9691 struct btrfs_trans_handle *trans = trans_in;
9692 struct btrfs_path *path;
9693 u64 start = ins->objectid;
9694 u64 len = ins->offset;
9697 memset(&stack_fi, 0, sizeof(stack_fi));
9699 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9700 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9701 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9702 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9703 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9704 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9705 /* Encryption and other encoding is reserved and all 0 */
9707 ret = btrfs_qgroup_release_data(inode, file_offset, len);
9709 return ERR_PTR(ret);
9712 ret = insert_reserved_file_extent(trans, inode,
9713 file_offset, &stack_fi,
9716 return ERR_PTR(ret);
9720 extent_info.disk_offset = start;
9721 extent_info.disk_len = len;
9722 extent_info.data_offset = 0;
9723 extent_info.data_len = len;
9724 extent_info.file_offset = file_offset;
9725 extent_info.extent_buf = (char *)&stack_fi;
9726 extent_info.is_new_extent = true;
9727 extent_info.qgroup_reserved = ret;
9728 extent_info.insertions = 0;
9730 path = btrfs_alloc_path();
9732 return ERR_PTR(-ENOMEM);
9734 ret = btrfs_replace_file_extents(&inode->vfs_inode, path, file_offset,
9735 file_offset + len - 1, &extent_info,
9737 btrfs_free_path(path);
9739 return ERR_PTR(ret);
9744 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9745 u64 start, u64 num_bytes, u64 min_size,
9746 loff_t actual_len, u64 *alloc_hint,
9747 struct btrfs_trans_handle *trans)
9749 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9750 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9751 struct extent_map *em;
9752 struct btrfs_root *root = BTRFS_I(inode)->root;
9753 struct btrfs_key ins;
9754 u64 cur_offset = start;
9755 u64 clear_offset = start;
9758 u64 last_alloc = (u64)-1;
9760 bool own_trans = true;
9761 u64 end = start + num_bytes - 1;
9765 while (num_bytes > 0) {
9766 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9767 cur_bytes = max(cur_bytes, min_size);
9769 * If we are severely fragmented we could end up with really
9770 * small allocations, so if the allocator is returning small
9771 * chunks lets make its job easier by only searching for those
9774 cur_bytes = min(cur_bytes, last_alloc);
9775 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9776 min_size, 0, *alloc_hint, &ins, 1, 0);
9781 * We've reserved this space, and thus converted it from
9782 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9783 * from here on out we will only need to clear our reservation
9784 * for the remaining unreserved area, so advance our
9785 * clear_offset by our extent size.
9787 clear_offset += ins.offset;
9789 last_alloc = ins.offset;
9790 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9793 * Now that we inserted the prealloc extent we can finally
9794 * decrement the number of reservations in the block group.
9795 * If we did it before, we could race with relocation and have
9796 * relocation miss the reserved extent, making it fail later.
9798 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9799 if (IS_ERR(trans)) {
9800 ret = PTR_ERR(trans);
9801 btrfs_free_reserved_extent(fs_info, ins.objectid,
9806 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9807 cur_offset + ins.offset -1, 0);
9809 em = alloc_extent_map();
9811 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9812 &BTRFS_I(inode)->runtime_flags);
9816 em->start = cur_offset;
9817 em->orig_start = cur_offset;
9818 em->len = ins.offset;
9819 em->block_start = ins.objectid;
9820 em->block_len = ins.offset;
9821 em->orig_block_len = ins.offset;
9822 em->ram_bytes = ins.offset;
9823 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9824 em->generation = trans->transid;
9827 write_lock(&em_tree->lock);
9828 ret = add_extent_mapping(em_tree, em, 1);
9829 write_unlock(&em_tree->lock);
9832 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9833 cur_offset + ins.offset - 1,
9836 free_extent_map(em);
9838 num_bytes -= ins.offset;
9839 cur_offset += ins.offset;
9840 *alloc_hint = ins.objectid + ins.offset;
9842 inode_inc_iversion(inode);
9843 inode->i_ctime = current_time(inode);
9844 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9845 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9846 (actual_len > inode->i_size) &&
9847 (cur_offset > inode->i_size)) {
9848 if (cur_offset > actual_len)
9849 i_size = actual_len;
9851 i_size = cur_offset;
9852 i_size_write(inode, i_size);
9853 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9856 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9859 btrfs_abort_transaction(trans, ret);
9861 btrfs_end_transaction(trans);
9866 btrfs_end_transaction(trans);
9870 if (clear_offset < end)
9871 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9872 end - clear_offset + 1);
9876 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9877 u64 start, u64 num_bytes, u64 min_size,
9878 loff_t actual_len, u64 *alloc_hint)
9880 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9881 min_size, actual_len, alloc_hint,
9885 int btrfs_prealloc_file_range_trans(struct inode *inode,
9886 struct btrfs_trans_handle *trans, int mode,
9887 u64 start, u64 num_bytes, u64 min_size,
9888 loff_t actual_len, u64 *alloc_hint)
9890 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9891 min_size, actual_len, alloc_hint, trans);
9894 static int btrfs_set_page_dirty(struct page *page)
9896 return __set_page_dirty_nobuffers(page);
9899 static int btrfs_permission(struct inode *inode, int mask)
9901 struct btrfs_root *root = BTRFS_I(inode)->root;
9902 umode_t mode = inode->i_mode;
9904 if (mask & MAY_WRITE &&
9905 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9906 if (btrfs_root_readonly(root))
9908 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9911 return generic_permission(inode, mask);
9914 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9916 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9917 struct btrfs_trans_handle *trans;
9918 struct btrfs_root *root = BTRFS_I(dir)->root;
9919 struct inode *inode = NULL;
9925 * 5 units required for adding orphan entry
9927 trans = btrfs_start_transaction(root, 5);
9929 return PTR_ERR(trans);
9931 ret = btrfs_get_free_objectid(root, &objectid);
9935 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9936 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
9937 if (IS_ERR(inode)) {
9938 ret = PTR_ERR(inode);
9943 inode->i_fop = &btrfs_file_operations;
9944 inode->i_op = &btrfs_file_inode_operations;
9946 inode->i_mapping->a_ops = &btrfs_aops;
9948 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9952 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9955 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
9960 * We set number of links to 0 in btrfs_new_inode(), and here we set
9961 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9964 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9966 set_nlink(inode, 1);
9967 d_tmpfile(dentry, inode);
9968 unlock_new_inode(inode);
9969 mark_inode_dirty(inode);
9971 btrfs_end_transaction(trans);
9973 discard_new_inode(inode);
9974 btrfs_btree_balance_dirty(fs_info);
9978 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
9980 struct inode *inode = tree->private_data;
9981 unsigned long index = start >> PAGE_SHIFT;
9982 unsigned long end_index = end >> PAGE_SHIFT;
9985 while (index <= end_index) {
9986 page = find_get_page(inode->i_mapping, index);
9987 ASSERT(page); /* Pages should be in the extent_io_tree */
9988 set_page_writeback(page);
9996 * Add an entry indicating a block group or device which is pinned by a
9997 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
9998 * negative errno on failure.
10000 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10001 bool is_block_group)
10003 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10004 struct btrfs_swapfile_pin *sp, *entry;
10005 struct rb_node **p;
10006 struct rb_node *parent = NULL;
10008 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10013 sp->is_block_group = is_block_group;
10015 spin_lock(&fs_info->swapfile_pins_lock);
10016 p = &fs_info->swapfile_pins.rb_node;
10019 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10020 if (sp->ptr < entry->ptr ||
10021 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10022 p = &(*p)->rb_left;
10023 } else if (sp->ptr > entry->ptr ||
10024 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10025 p = &(*p)->rb_right;
10027 spin_unlock(&fs_info->swapfile_pins_lock);
10032 rb_link_node(&sp->node, parent, p);
10033 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10034 spin_unlock(&fs_info->swapfile_pins_lock);
10038 /* Free all of the entries pinned by this swapfile. */
10039 static void btrfs_free_swapfile_pins(struct inode *inode)
10041 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10042 struct btrfs_swapfile_pin *sp;
10043 struct rb_node *node, *next;
10045 spin_lock(&fs_info->swapfile_pins_lock);
10046 node = rb_first(&fs_info->swapfile_pins);
10048 next = rb_next(node);
10049 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10050 if (sp->inode == inode) {
10051 rb_erase(&sp->node, &fs_info->swapfile_pins);
10052 if (sp->is_block_group)
10053 btrfs_put_block_group(sp->ptr);
10058 spin_unlock(&fs_info->swapfile_pins_lock);
10061 struct btrfs_swap_info {
10067 unsigned long nr_pages;
10071 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10072 struct btrfs_swap_info *bsi)
10074 unsigned long nr_pages;
10075 u64 first_ppage, first_ppage_reported, next_ppage;
10078 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10079 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10080 PAGE_SIZE) >> PAGE_SHIFT;
10082 if (first_ppage >= next_ppage)
10084 nr_pages = next_ppage - first_ppage;
10086 first_ppage_reported = first_ppage;
10087 if (bsi->start == 0)
10088 first_ppage_reported++;
10089 if (bsi->lowest_ppage > first_ppage_reported)
10090 bsi->lowest_ppage = first_ppage_reported;
10091 if (bsi->highest_ppage < (next_ppage - 1))
10092 bsi->highest_ppage = next_ppage - 1;
10094 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10097 bsi->nr_extents += ret;
10098 bsi->nr_pages += nr_pages;
10102 static void btrfs_swap_deactivate(struct file *file)
10104 struct inode *inode = file_inode(file);
10106 btrfs_free_swapfile_pins(inode);
10107 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10110 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10113 struct inode *inode = file_inode(file);
10114 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10115 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10116 struct extent_state *cached_state = NULL;
10117 struct extent_map *em = NULL;
10118 struct btrfs_device *device = NULL;
10119 struct btrfs_swap_info bsi = {
10120 .lowest_ppage = (sector_t)-1ULL,
10127 * If the swap file was just created, make sure delalloc is done. If the
10128 * file changes again after this, the user is doing something stupid and
10129 * we don't really care.
10131 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10136 * The inode is locked, so these flags won't change after we check them.
10138 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10139 btrfs_warn(fs_info, "swapfile must not be compressed");
10142 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10143 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10146 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10147 btrfs_warn(fs_info, "swapfile must not be checksummed");
10152 * Balance or device remove/replace/resize can move stuff around from
10153 * under us. The exclop protection makes sure they aren't running/won't
10154 * run concurrently while we are mapping the swap extents, and
10155 * fs_info->swapfile_pins prevents them from running while the swap
10156 * file is active and moving the extents. Note that this also prevents
10157 * a concurrent device add which isn't actually necessary, but it's not
10158 * really worth the trouble to allow it.
10160 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10161 btrfs_warn(fs_info,
10162 "cannot activate swapfile while exclusive operation is running");
10166 * Snapshots can create extents which require COW even if NODATACOW is
10167 * set. We use this counter to prevent snapshots. We must increment it
10168 * before walking the extents because we don't want a concurrent
10169 * snapshot to run after we've already checked the extents.
10171 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10173 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10175 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10177 while (start < isize) {
10178 u64 logical_block_start, physical_block_start;
10179 struct btrfs_block_group *bg;
10180 u64 len = isize - start;
10182 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10188 if (em->block_start == EXTENT_MAP_HOLE) {
10189 btrfs_warn(fs_info, "swapfile must not have holes");
10193 if (em->block_start == EXTENT_MAP_INLINE) {
10195 * It's unlikely we'll ever actually find ourselves
10196 * here, as a file small enough to fit inline won't be
10197 * big enough to store more than the swap header, but in
10198 * case something changes in the future, let's catch it
10199 * here rather than later.
10201 btrfs_warn(fs_info, "swapfile must not be inline");
10205 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10206 btrfs_warn(fs_info, "swapfile must not be compressed");
10211 logical_block_start = em->block_start + (start - em->start);
10212 len = min(len, em->len - (start - em->start));
10213 free_extent_map(em);
10216 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10222 btrfs_warn(fs_info,
10223 "swapfile must not be copy-on-write");
10228 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10234 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10235 btrfs_warn(fs_info,
10236 "swapfile must have single data profile");
10241 if (device == NULL) {
10242 device = em->map_lookup->stripes[0].dev;
10243 ret = btrfs_add_swapfile_pin(inode, device, false);
10248 } else if (device != em->map_lookup->stripes[0].dev) {
10249 btrfs_warn(fs_info, "swapfile must be on one device");
10254 physical_block_start = (em->map_lookup->stripes[0].physical +
10255 (logical_block_start - em->start));
10256 len = min(len, em->len - (logical_block_start - em->start));
10257 free_extent_map(em);
10260 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10262 btrfs_warn(fs_info,
10263 "could not find block group containing swapfile");
10268 ret = btrfs_add_swapfile_pin(inode, bg, true);
10270 btrfs_put_block_group(bg);
10277 if (bsi.block_len &&
10278 bsi.block_start + bsi.block_len == physical_block_start) {
10279 bsi.block_len += len;
10281 if (bsi.block_len) {
10282 ret = btrfs_add_swap_extent(sis, &bsi);
10287 bsi.block_start = physical_block_start;
10288 bsi.block_len = len;
10295 ret = btrfs_add_swap_extent(sis, &bsi);
10298 if (!IS_ERR_OR_NULL(em))
10299 free_extent_map(em);
10301 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10304 btrfs_swap_deactivate(file);
10306 btrfs_exclop_finish(fs_info);
10312 sis->bdev = device->bdev;
10313 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10314 sis->max = bsi.nr_pages;
10315 sis->pages = bsi.nr_pages - 1;
10316 sis->highest_bit = bsi.nr_pages - 1;
10317 return bsi.nr_extents;
10320 static void btrfs_swap_deactivate(struct file *file)
10324 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10327 return -EOPNOTSUPP;
10332 * Update the number of bytes used in the VFS' inode. When we replace extents in
10333 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10334 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10335 * always get a correct value.
10337 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10338 const u64 add_bytes,
10339 const u64 del_bytes)
10341 if (add_bytes == del_bytes)
10344 spin_lock(&inode->lock);
10346 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10348 inode_add_bytes(&inode->vfs_inode, add_bytes);
10349 spin_unlock(&inode->lock);
10352 static const struct inode_operations btrfs_dir_inode_operations = {
10353 .getattr = btrfs_getattr,
10354 .lookup = btrfs_lookup,
10355 .create = btrfs_create,
10356 .unlink = btrfs_unlink,
10357 .link = btrfs_link,
10358 .mkdir = btrfs_mkdir,
10359 .rmdir = btrfs_rmdir,
10360 .rename = btrfs_rename2,
10361 .symlink = btrfs_symlink,
10362 .setattr = btrfs_setattr,
10363 .mknod = btrfs_mknod,
10364 .listxattr = btrfs_listxattr,
10365 .permission = btrfs_permission,
10366 .get_acl = btrfs_get_acl,
10367 .set_acl = btrfs_set_acl,
10368 .update_time = btrfs_update_time,
10369 .tmpfile = btrfs_tmpfile,
10372 static const struct file_operations btrfs_dir_file_operations = {
10373 .llseek = generic_file_llseek,
10374 .read = generic_read_dir,
10375 .iterate_shared = btrfs_real_readdir,
10376 .open = btrfs_opendir,
10377 .unlocked_ioctl = btrfs_ioctl,
10378 #ifdef CONFIG_COMPAT
10379 .compat_ioctl = btrfs_compat_ioctl,
10381 .release = btrfs_release_file,
10382 .fsync = btrfs_sync_file,
10386 * btrfs doesn't support the bmap operation because swapfiles
10387 * use bmap to make a mapping of extents in the file. They assume
10388 * these extents won't change over the life of the file and they
10389 * use the bmap result to do IO directly to the drive.
10391 * the btrfs bmap call would return logical addresses that aren't
10392 * suitable for IO and they also will change frequently as COW
10393 * operations happen. So, swapfile + btrfs == corruption.
10395 * For now we're avoiding this by dropping bmap.
10397 static const struct address_space_operations btrfs_aops = {
10398 .readpage = btrfs_readpage,
10399 .writepage = btrfs_writepage,
10400 .writepages = btrfs_writepages,
10401 .readahead = btrfs_readahead,
10402 .direct_IO = noop_direct_IO,
10403 .invalidatepage = btrfs_invalidatepage,
10404 .releasepage = btrfs_releasepage,
10405 #ifdef CONFIG_MIGRATION
10406 .migratepage = btrfs_migratepage,
10408 .set_page_dirty = btrfs_set_page_dirty,
10409 .error_remove_page = generic_error_remove_page,
10410 .swap_activate = btrfs_swap_activate,
10411 .swap_deactivate = btrfs_swap_deactivate,
10414 static const struct inode_operations btrfs_file_inode_operations = {
10415 .getattr = btrfs_getattr,
10416 .setattr = btrfs_setattr,
10417 .listxattr = btrfs_listxattr,
10418 .permission = btrfs_permission,
10419 .fiemap = btrfs_fiemap,
10420 .get_acl = btrfs_get_acl,
10421 .set_acl = btrfs_set_acl,
10422 .update_time = btrfs_update_time,
10424 static const struct inode_operations btrfs_special_inode_operations = {
10425 .getattr = btrfs_getattr,
10426 .setattr = btrfs_setattr,
10427 .permission = btrfs_permission,
10428 .listxattr = btrfs_listxattr,
10429 .get_acl = btrfs_get_acl,
10430 .set_acl = btrfs_set_acl,
10431 .update_time = btrfs_update_time,
10433 static const struct inode_operations btrfs_symlink_inode_operations = {
10434 .get_link = page_get_link,
10435 .getattr = btrfs_getattr,
10436 .setattr = btrfs_setattr,
10437 .permission = btrfs_permission,
10438 .listxattr = btrfs_listxattr,
10439 .update_time = btrfs_update_time,
10442 const struct dentry_operations btrfs_dentry_operations = {
10443 .d_delete = btrfs_dentry_delete,