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,
701 * Ensure we only free the compressed pages if we have
702 * them allocated, as we can still reach here with
703 * inode_need_compress() == false.
706 for (i = 0; i < nr_pages; i++) {
707 WARN_ON(pages[i]->mapping);
718 * we aren't doing an inline extent round the compressed size
719 * up to a block size boundary so the allocator does sane
722 total_compressed = ALIGN(total_compressed, blocksize);
725 * one last check to make sure the compression is really a
726 * win, compare the page count read with the blocks on disk,
727 * compression must free at least one sector size
729 total_in = ALIGN(total_in, PAGE_SIZE);
730 if (total_compressed + blocksize <= total_in) {
731 compressed_extents++;
734 * The async work queues will take care of doing actual
735 * allocation on disk for these compressed pages, and
736 * will submit them to the elevator.
738 add_async_extent(async_chunk, start, total_in,
739 total_compressed, pages, nr_pages,
742 if (start + total_in < end) {
748 return compressed_extents;
753 * the compression code ran but failed to make things smaller,
754 * free any pages it allocated and our page pointer array
756 for (i = 0; i < nr_pages; i++) {
757 WARN_ON(pages[i]->mapping);
762 total_compressed = 0;
765 /* flag the file so we don't compress in the future */
766 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
767 !(BTRFS_I(inode)->prop_compress)) {
768 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
771 cleanup_and_bail_uncompressed:
773 * No compression, but we still need to write the pages in the file
774 * we've been given so far. redirty the locked page if it corresponds
775 * to our extent and set things up for the async work queue to run
776 * cow_file_range to do the normal delalloc dance.
778 if (async_chunk->locked_page &&
779 (page_offset(async_chunk->locked_page) >= start &&
780 page_offset(async_chunk->locked_page)) <= end) {
781 __set_page_dirty_nobuffers(async_chunk->locked_page);
782 /* unlocked later on in the async handlers */
786 extent_range_redirty_for_io(inode, start, end);
787 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
788 BTRFS_COMPRESS_NONE);
789 compressed_extents++;
791 return compressed_extents;
794 static void free_async_extent_pages(struct async_extent *async_extent)
798 if (!async_extent->pages)
801 for (i = 0; i < async_extent->nr_pages; i++) {
802 WARN_ON(async_extent->pages[i]->mapping);
803 put_page(async_extent->pages[i]);
805 kfree(async_extent->pages);
806 async_extent->nr_pages = 0;
807 async_extent->pages = NULL;
811 * phase two of compressed writeback. This is the ordered portion
812 * of the code, which only gets called in the order the work was
813 * queued. We walk all the async extents created by compress_file_range
814 * and send them down to the disk.
816 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
818 struct btrfs_inode *inode = BTRFS_I(async_chunk->inode);
819 struct btrfs_fs_info *fs_info = inode->root->fs_info;
820 struct async_extent *async_extent;
822 struct btrfs_key ins;
823 struct extent_map *em;
824 struct btrfs_root *root = inode->root;
825 struct extent_io_tree *io_tree = &inode->io_tree;
829 while (!list_empty(&async_chunk->extents)) {
830 async_extent = list_entry(async_chunk->extents.next,
831 struct async_extent, list);
832 list_del(&async_extent->list);
835 lock_extent(io_tree, async_extent->start,
836 async_extent->start + async_extent->ram_size - 1);
837 /* did the compression code fall back to uncompressed IO? */
838 if (!async_extent->pages) {
839 int page_started = 0;
840 unsigned long nr_written = 0;
842 /* allocate blocks */
843 ret = cow_file_range(inode, async_chunk->locked_page,
845 async_extent->start +
846 async_extent->ram_size - 1,
847 &page_started, &nr_written, 0);
852 * if page_started, cow_file_range inserted an
853 * inline extent and took care of all the unlocking
854 * and IO for us. Otherwise, we need to submit
855 * all those pages down to the drive.
857 if (!page_started && !ret)
858 extent_write_locked_range(&inode->vfs_inode,
860 async_extent->start +
861 async_extent->ram_size - 1,
863 else if (ret && async_chunk->locked_page)
864 unlock_page(async_chunk->locked_page);
870 ret = btrfs_reserve_extent(root, async_extent->ram_size,
871 async_extent->compressed_size,
872 async_extent->compressed_size,
873 0, alloc_hint, &ins, 1, 1);
875 free_async_extent_pages(async_extent);
877 if (ret == -ENOSPC) {
878 unlock_extent(io_tree, async_extent->start,
879 async_extent->start +
880 async_extent->ram_size - 1);
883 * we need to redirty the pages if we decide to
884 * fallback to uncompressed IO, otherwise we
885 * will not submit these pages down to lower
888 extent_range_redirty_for_io(&inode->vfs_inode,
890 async_extent->start +
891 async_extent->ram_size - 1);
898 * here we're doing allocation and writeback of the
901 em = create_io_em(inode, async_extent->start,
902 async_extent->ram_size, /* len */
903 async_extent->start, /* orig_start */
904 ins.objectid, /* block_start */
905 ins.offset, /* block_len */
906 ins.offset, /* orig_block_len */
907 async_extent->ram_size, /* ram_bytes */
908 async_extent->compress_type,
909 BTRFS_ORDERED_COMPRESSED);
911 /* ret value is not necessary due to void function */
912 goto out_free_reserve;
915 ret = btrfs_add_ordered_extent_compress(inode,
918 async_extent->ram_size,
920 BTRFS_ORDERED_COMPRESSED,
921 async_extent->compress_type);
923 btrfs_drop_extent_cache(inode, async_extent->start,
924 async_extent->start +
925 async_extent->ram_size - 1, 0);
926 goto out_free_reserve;
928 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
931 * clear dirty, set writeback and unlock the pages.
933 extent_clear_unlock_delalloc(inode, async_extent->start,
934 async_extent->start +
935 async_extent->ram_size - 1,
936 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
937 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
939 if (btrfs_submit_compressed_write(inode, async_extent->start,
940 async_extent->ram_size,
942 ins.offset, async_extent->pages,
943 async_extent->nr_pages,
944 async_chunk->write_flags,
945 async_chunk->blkcg_css)) {
946 struct page *p = async_extent->pages[0];
947 const u64 start = async_extent->start;
948 const u64 end = start + async_extent->ram_size - 1;
950 p->mapping = inode->vfs_inode.i_mapping;
951 btrfs_writepage_endio_finish_ordered(p, start, end, 0);
954 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
957 free_async_extent_pages(async_extent);
959 alloc_hint = ins.objectid + ins.offset;
965 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
966 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
968 extent_clear_unlock_delalloc(inode, async_extent->start,
969 async_extent->start +
970 async_extent->ram_size - 1,
971 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
972 EXTENT_DELALLOC_NEW |
973 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
974 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
975 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
977 free_async_extent_pages(async_extent);
982 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
985 struct extent_map_tree *em_tree = &inode->extent_tree;
986 struct extent_map *em;
989 read_lock(&em_tree->lock);
990 em = search_extent_mapping(em_tree, start, num_bytes);
993 * if block start isn't an actual block number then find the
994 * first block in this inode and use that as a hint. If that
995 * block is also bogus then just don't worry about it.
997 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
999 em = search_extent_mapping(em_tree, 0, 0);
1000 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1001 alloc_hint = em->block_start;
1003 free_extent_map(em);
1005 alloc_hint = em->block_start;
1006 free_extent_map(em);
1009 read_unlock(&em_tree->lock);
1015 * when extent_io.c finds a delayed allocation range in the file,
1016 * the call backs end up in this code. The basic idea is to
1017 * allocate extents on disk for the range, and create ordered data structs
1018 * in ram to track those extents.
1020 * locked_page is the page that writepage had locked already. We use
1021 * it to make sure we don't do extra locks or unlocks.
1023 * *page_started is set to one if we unlock locked_page and do everything
1024 * required to start IO on it. It may be clean and already done with
1025 * IO when we return.
1027 static noinline int cow_file_range(struct btrfs_inode *inode,
1028 struct page *locked_page,
1029 u64 start, u64 end, int *page_started,
1030 unsigned long *nr_written, int unlock)
1032 struct btrfs_root *root = inode->root;
1033 struct btrfs_fs_info *fs_info = root->fs_info;
1036 unsigned long ram_size;
1037 u64 cur_alloc_size = 0;
1039 u64 blocksize = fs_info->sectorsize;
1040 struct btrfs_key ins;
1041 struct extent_map *em;
1042 unsigned clear_bits;
1043 unsigned long page_ops;
1044 bool extent_reserved = false;
1047 if (btrfs_is_free_space_inode(inode)) {
1053 num_bytes = ALIGN(end - start + 1, blocksize);
1054 num_bytes = max(blocksize, num_bytes);
1055 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1057 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1060 /* lets try to make an inline extent */
1061 ret = cow_file_range_inline(inode, start, end, 0,
1062 BTRFS_COMPRESS_NONE, NULL);
1065 * We use DO_ACCOUNTING here because we need the
1066 * delalloc_release_metadata to be run _after_ we drop
1067 * our outstanding extent for clearing delalloc for this
1070 extent_clear_unlock_delalloc(inode, start, end, NULL,
1071 EXTENT_LOCKED | EXTENT_DELALLOC |
1072 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1073 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1074 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1075 PAGE_END_WRITEBACK);
1076 *nr_written = *nr_written +
1077 (end - start + PAGE_SIZE) / PAGE_SIZE;
1080 } else if (ret < 0) {
1085 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1086 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
1089 * Relocation relies on the relocated extents to have exactly the same
1090 * size as the original extents. Normally writeback for relocation data
1091 * extents follows a NOCOW path because relocation preallocates the
1092 * extents. However, due to an operation such as scrub turning a block
1093 * group to RO mode, it may fallback to COW mode, so we must make sure
1094 * an extent allocated during COW has exactly the requested size and can
1095 * not be split into smaller extents, otherwise relocation breaks and
1096 * fails during the stage where it updates the bytenr of file extent
1099 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1100 min_alloc_size = num_bytes;
1102 min_alloc_size = fs_info->sectorsize;
1104 while (num_bytes > 0) {
1105 cur_alloc_size = num_bytes;
1106 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1107 min_alloc_size, 0, alloc_hint,
1111 cur_alloc_size = ins.offset;
1112 extent_reserved = true;
1114 ram_size = ins.offset;
1115 em = create_io_em(inode, start, ins.offset, /* len */
1116 start, /* orig_start */
1117 ins.objectid, /* block_start */
1118 ins.offset, /* block_len */
1119 ins.offset, /* orig_block_len */
1120 ram_size, /* ram_bytes */
1121 BTRFS_COMPRESS_NONE, /* compress_type */
1122 BTRFS_ORDERED_REGULAR /* type */);
1127 free_extent_map(em);
1129 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1130 ram_size, cur_alloc_size, 0);
1132 goto out_drop_extent_cache;
1134 if (root->root_key.objectid ==
1135 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1136 ret = btrfs_reloc_clone_csums(inode, start,
1139 * Only drop cache here, and process as normal.
1141 * We must not allow extent_clear_unlock_delalloc()
1142 * at out_unlock label to free meta of this ordered
1143 * extent, as its meta should be freed by
1144 * btrfs_finish_ordered_io().
1146 * So we must continue until @start is increased to
1147 * skip current ordered extent.
1150 btrfs_drop_extent_cache(inode, start,
1151 start + ram_size - 1, 0);
1154 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1156 /* we're not doing compressed IO, don't unlock the first
1157 * page (which the caller expects to stay locked), don't
1158 * clear any dirty bits and don't set any writeback bits
1160 * Do set the Private2 bit so we know this page was properly
1161 * setup for writepage
1163 page_ops = unlock ? PAGE_UNLOCK : 0;
1164 page_ops |= PAGE_SET_PRIVATE2;
1166 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1168 EXTENT_LOCKED | EXTENT_DELALLOC,
1170 if (num_bytes < cur_alloc_size)
1173 num_bytes -= cur_alloc_size;
1174 alloc_hint = ins.objectid + ins.offset;
1175 start += cur_alloc_size;
1176 extent_reserved = false;
1179 * btrfs_reloc_clone_csums() error, since start is increased
1180 * extent_clear_unlock_delalloc() at out_unlock label won't
1181 * free metadata of current ordered extent, we're OK to exit.
1189 out_drop_extent_cache:
1190 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1192 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1193 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1195 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1196 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1197 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1200 * If we reserved an extent for our delalloc range (or a subrange) and
1201 * failed to create the respective ordered extent, then it means that
1202 * when we reserved the extent we decremented the extent's size from
1203 * the data space_info's bytes_may_use counter and incremented the
1204 * space_info's bytes_reserved counter by the same amount. We must make
1205 * sure extent_clear_unlock_delalloc() does not try to decrement again
1206 * the data space_info's bytes_may_use counter, therefore we do not pass
1207 * it the flag EXTENT_CLEAR_DATA_RESV.
1209 if (extent_reserved) {
1210 extent_clear_unlock_delalloc(inode, start,
1211 start + cur_alloc_size - 1,
1215 start += cur_alloc_size;
1219 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1220 clear_bits | EXTENT_CLEAR_DATA_RESV,
1226 * work queue call back to started compression on a file and pages
1228 static noinline void async_cow_start(struct btrfs_work *work)
1230 struct async_chunk *async_chunk;
1231 int compressed_extents;
1233 async_chunk = container_of(work, struct async_chunk, work);
1235 compressed_extents = compress_file_range(async_chunk);
1236 if (compressed_extents == 0) {
1237 btrfs_add_delayed_iput(async_chunk->inode);
1238 async_chunk->inode = NULL;
1243 * work queue call back to submit previously compressed pages
1245 static noinline void async_cow_submit(struct btrfs_work *work)
1247 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1249 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1250 unsigned long nr_pages;
1252 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1255 /* atomic_sub_return implies a barrier */
1256 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1258 cond_wake_up_nomb(&fs_info->async_submit_wait);
1261 * ->inode could be NULL if async_chunk_start has failed to compress,
1262 * in which case we don't have anything to submit, yet we need to
1263 * always adjust ->async_delalloc_pages as its paired with the init
1264 * happening in cow_file_range_async
1266 if (async_chunk->inode)
1267 submit_compressed_extents(async_chunk);
1270 static noinline void async_cow_free(struct btrfs_work *work)
1272 struct async_chunk *async_chunk;
1274 async_chunk = container_of(work, struct async_chunk, work);
1275 if (async_chunk->inode)
1276 btrfs_add_delayed_iput(async_chunk->inode);
1277 if (async_chunk->blkcg_css)
1278 css_put(async_chunk->blkcg_css);
1280 * Since the pointer to 'pending' is at the beginning of the array of
1281 * async_chunk's, freeing it ensures the whole array has been freed.
1283 if (atomic_dec_and_test(async_chunk->pending))
1284 kvfree(async_chunk->pending);
1287 static int cow_file_range_async(struct btrfs_inode *inode,
1288 struct writeback_control *wbc,
1289 struct page *locked_page,
1290 u64 start, u64 end, int *page_started,
1291 unsigned long *nr_written)
1293 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1294 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1295 struct async_cow *ctx;
1296 struct async_chunk *async_chunk;
1297 unsigned long nr_pages;
1299 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1301 bool should_compress;
1303 const unsigned int write_flags = wbc_to_write_flags(wbc);
1305 unlock_extent(&inode->io_tree, start, end);
1307 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1308 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1310 should_compress = false;
1312 should_compress = true;
1315 nofs_flag = memalloc_nofs_save();
1316 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1317 memalloc_nofs_restore(nofs_flag);
1320 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1321 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1322 EXTENT_DO_ACCOUNTING;
1323 unsigned long page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1324 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
1327 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1328 clear_bits, page_ops);
1332 async_chunk = ctx->chunks;
1333 atomic_set(&ctx->num_chunks, num_chunks);
1335 for (i = 0; i < num_chunks; i++) {
1336 if (should_compress)
1337 cur_end = min(end, start + SZ_512K - 1);
1342 * igrab is called higher up in the call chain, take only the
1343 * lightweight reference for the callback lifetime
1345 ihold(&inode->vfs_inode);
1346 async_chunk[i].pending = &ctx->num_chunks;
1347 async_chunk[i].inode = &inode->vfs_inode;
1348 async_chunk[i].start = start;
1349 async_chunk[i].end = cur_end;
1350 async_chunk[i].write_flags = write_flags;
1351 INIT_LIST_HEAD(&async_chunk[i].extents);
1354 * The locked_page comes all the way from writepage and its
1355 * the original page we were actually given. As we spread
1356 * this large delalloc region across multiple async_chunk
1357 * structs, only the first struct needs a pointer to locked_page
1359 * This way we don't need racey decisions about who is supposed
1364 * Depending on the compressibility, the pages might or
1365 * might not go through async. We want all of them to
1366 * be accounted against wbc once. Let's do it here
1367 * before the paths diverge. wbc accounting is used
1368 * only for foreign writeback detection and doesn't
1369 * need full accuracy. Just account the whole thing
1370 * against the first page.
1372 wbc_account_cgroup_owner(wbc, locked_page,
1374 async_chunk[i].locked_page = locked_page;
1377 async_chunk[i].locked_page = NULL;
1380 if (blkcg_css != blkcg_root_css) {
1382 async_chunk[i].blkcg_css = blkcg_css;
1384 async_chunk[i].blkcg_css = NULL;
1387 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1388 async_cow_submit, async_cow_free);
1390 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1391 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1393 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1395 *nr_written += nr_pages;
1396 start = cur_end + 1;
1402 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1403 u64 bytenr, u64 num_bytes)
1406 struct btrfs_ordered_sum *sums;
1409 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1410 bytenr + num_bytes - 1, &list, 0);
1411 if (ret == 0 && list_empty(&list))
1414 while (!list_empty(&list)) {
1415 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1416 list_del(&sums->list);
1424 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1425 const u64 start, const u64 end,
1426 int *page_started, unsigned long *nr_written)
1428 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1429 const bool is_reloc_ino = (inode->root->root_key.objectid ==
1430 BTRFS_DATA_RELOC_TREE_OBJECTID);
1431 const u64 range_bytes = end + 1 - start;
1432 struct extent_io_tree *io_tree = &inode->io_tree;
1433 u64 range_start = start;
1437 * If EXTENT_NORESERVE is set it means that when the buffered write was
1438 * made we had not enough available data space and therefore we did not
1439 * reserve data space for it, since we though we could do NOCOW for the
1440 * respective file range (either there is prealloc extent or the inode
1441 * has the NOCOW bit set).
1443 * However when we need to fallback to COW mode (because for example the
1444 * block group for the corresponding extent was turned to RO mode by a
1445 * scrub or relocation) we need to do the following:
1447 * 1) We increment the bytes_may_use counter of the data space info.
1448 * If COW succeeds, it allocates a new data extent and after doing
1449 * that it decrements the space info's bytes_may_use counter and
1450 * increments its bytes_reserved counter by the same amount (we do
1451 * this at btrfs_add_reserved_bytes()). So we need to increment the
1452 * bytes_may_use counter to compensate (when space is reserved at
1453 * buffered write time, the bytes_may_use counter is incremented);
1455 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1456 * that if the COW path fails for any reason, it decrements (through
1457 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1458 * data space info, which we incremented in the step above.
1460 * If we need to fallback to cow and the inode corresponds to a free
1461 * space cache inode or an inode of the data relocation tree, we must
1462 * also increment bytes_may_use of the data space_info for the same
1463 * reason. Space caches and relocated data extents always get a prealloc
1464 * extent for them, however scrub or balance may have set the block
1465 * group that contains that extent to RO mode and therefore force COW
1466 * when starting writeback.
1468 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1469 EXTENT_NORESERVE, 0);
1470 if (count > 0 || is_space_ino || is_reloc_ino) {
1472 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1473 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1475 if (is_space_ino || is_reloc_ino)
1476 bytes = range_bytes;
1478 spin_lock(&sinfo->lock);
1479 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1480 spin_unlock(&sinfo->lock);
1483 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1487 return cow_file_range(inode, locked_page, start, end, page_started,
1492 * when nowcow writeback call back. This checks for snapshots or COW copies
1493 * of the extents that exist in the file, and COWs the file as required.
1495 * If no cow copies or snapshots exist, we write directly to the existing
1498 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1499 struct page *locked_page,
1500 const u64 start, const u64 end,
1501 int *page_started, int force,
1502 unsigned long *nr_written)
1504 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1505 struct btrfs_root *root = inode->root;
1506 struct btrfs_path *path;
1507 u64 cow_start = (u64)-1;
1508 u64 cur_offset = start;
1510 bool check_prev = true;
1511 const bool freespace_inode = btrfs_is_free_space_inode(inode);
1512 u64 ino = btrfs_ino(inode);
1514 u64 disk_bytenr = 0;
1516 path = btrfs_alloc_path();
1518 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1519 EXTENT_LOCKED | EXTENT_DELALLOC |
1520 EXTENT_DO_ACCOUNTING |
1521 EXTENT_DEFRAG, PAGE_UNLOCK |
1523 PAGE_SET_WRITEBACK |
1524 PAGE_END_WRITEBACK);
1529 struct btrfs_key found_key;
1530 struct btrfs_file_extent_item *fi;
1531 struct extent_buffer *leaf;
1541 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1547 * If there is no extent for our range when doing the initial
1548 * search, then go back to the previous slot as it will be the
1549 * one containing the search offset
1551 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1552 leaf = path->nodes[0];
1553 btrfs_item_key_to_cpu(leaf, &found_key,
1554 path->slots[0] - 1);
1555 if (found_key.objectid == ino &&
1556 found_key.type == BTRFS_EXTENT_DATA_KEY)
1561 /* Go to next leaf if we have exhausted the current one */
1562 leaf = path->nodes[0];
1563 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1564 ret = btrfs_next_leaf(root, path);
1566 if (cow_start != (u64)-1)
1567 cur_offset = cow_start;
1572 leaf = path->nodes[0];
1575 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1577 /* Didn't find anything for our INO */
1578 if (found_key.objectid > ino)
1581 * Keep searching until we find an EXTENT_ITEM or there are no
1582 * more extents for this inode
1584 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1585 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1590 /* Found key is not EXTENT_DATA_KEY or starts after req range */
1591 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1592 found_key.offset > end)
1596 * If the found extent starts after requested offset, then
1597 * adjust extent_end to be right before this extent begins
1599 if (found_key.offset > cur_offset) {
1600 extent_end = found_key.offset;
1606 * Found extent which begins before our range and potentially
1609 fi = btrfs_item_ptr(leaf, path->slots[0],
1610 struct btrfs_file_extent_item);
1611 extent_type = btrfs_file_extent_type(leaf, fi);
1613 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1614 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1615 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1616 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1617 extent_offset = btrfs_file_extent_offset(leaf, fi);
1618 extent_end = found_key.offset +
1619 btrfs_file_extent_num_bytes(leaf, fi);
1621 btrfs_file_extent_disk_num_bytes(leaf, fi);
1623 * If the extent we got ends before our current offset,
1624 * skip to the next extent.
1626 if (extent_end <= cur_offset) {
1631 if (disk_bytenr == 0)
1633 /* Skip compressed/encrypted/encoded extents */
1634 if (btrfs_file_extent_compression(leaf, fi) ||
1635 btrfs_file_extent_encryption(leaf, fi) ||
1636 btrfs_file_extent_other_encoding(leaf, fi))
1639 * If extent is created before the last volume's snapshot
1640 * this implies the extent is shared, hence we can't do
1641 * nocow. This is the same check as in
1642 * btrfs_cross_ref_exist but without calling
1643 * btrfs_search_slot.
1645 if (!freespace_inode &&
1646 btrfs_file_extent_generation(leaf, fi) <=
1647 btrfs_root_last_snapshot(&root->root_item))
1649 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1653 * The following checks can be expensive, as they need to
1654 * take other locks and do btree or rbtree searches, so
1655 * release the path to avoid blocking other tasks for too
1658 btrfs_release_path(path);
1660 /* If extent is RO, we must COW it */
1661 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1663 ret = btrfs_cross_ref_exist(root, ino,
1665 extent_offset, disk_bytenr, false);
1668 * ret could be -EIO if the above fails to read
1672 if (cow_start != (u64)-1)
1673 cur_offset = cow_start;
1677 WARN_ON_ONCE(freespace_inode);
1680 disk_bytenr += extent_offset;
1681 disk_bytenr += cur_offset - found_key.offset;
1682 num_bytes = min(end + 1, extent_end) - cur_offset;
1684 * If there are pending snapshots for this root, we
1685 * fall into common COW way
1687 if (!freespace_inode && atomic_read(&root->snapshot_force_cow))
1690 * force cow if csum exists in the range.
1691 * this ensure that csum for a given extent are
1692 * either valid or do not exist.
1694 ret = csum_exist_in_range(fs_info, disk_bytenr,
1698 * ret could be -EIO if the above fails to read
1702 if (cow_start != (u64)-1)
1703 cur_offset = cow_start;
1706 WARN_ON_ONCE(freespace_inode);
1709 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1712 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1713 extent_end = found_key.offset + ram_bytes;
1714 extent_end = ALIGN(extent_end, fs_info->sectorsize);
1715 /* Skip extents outside of our requested range */
1716 if (extent_end <= start) {
1721 /* If this triggers then we have a memory corruption */
1726 * If nocow is false then record the beginning of the range
1727 * that needs to be COWed
1730 if (cow_start == (u64)-1)
1731 cow_start = cur_offset;
1732 cur_offset = extent_end;
1733 if (cur_offset > end)
1735 if (!path->nodes[0])
1742 * COW range from cow_start to found_key.offset - 1. As the key
1743 * will contain the beginning of the first extent that can be
1744 * NOCOW, following one which needs to be COW'ed
1746 if (cow_start != (u64)-1) {
1747 ret = fallback_to_cow(inode, locked_page,
1748 cow_start, found_key.offset - 1,
1749 page_started, nr_written);
1752 cow_start = (u64)-1;
1755 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1756 u64 orig_start = found_key.offset - extent_offset;
1757 struct extent_map *em;
1759 em = create_io_em(inode, cur_offset, num_bytes,
1761 disk_bytenr, /* block_start */
1762 num_bytes, /* block_len */
1763 disk_num_bytes, /* orig_block_len */
1764 ram_bytes, BTRFS_COMPRESS_NONE,
1765 BTRFS_ORDERED_PREALLOC);
1770 free_extent_map(em);
1771 ret = btrfs_add_ordered_extent(inode, cur_offset,
1772 disk_bytenr, num_bytes,
1774 BTRFS_ORDERED_PREALLOC);
1776 btrfs_drop_extent_cache(inode, cur_offset,
1777 cur_offset + num_bytes - 1,
1782 ret = btrfs_add_ordered_extent(inode, cur_offset,
1783 disk_bytenr, num_bytes,
1785 BTRFS_ORDERED_NOCOW);
1791 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1794 if (root->root_key.objectid ==
1795 BTRFS_DATA_RELOC_TREE_OBJECTID)
1797 * Error handled later, as we must prevent
1798 * extent_clear_unlock_delalloc() in error handler
1799 * from freeing metadata of created ordered extent.
1801 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1804 extent_clear_unlock_delalloc(inode, cur_offset,
1805 cur_offset + num_bytes - 1,
1806 locked_page, EXTENT_LOCKED |
1808 EXTENT_CLEAR_DATA_RESV,
1809 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1811 cur_offset = extent_end;
1814 * btrfs_reloc_clone_csums() error, now we're OK to call error
1815 * handler, as metadata for created ordered extent will only
1816 * be freed by btrfs_finish_ordered_io().
1820 if (cur_offset > end)
1823 btrfs_release_path(path);
1825 if (cur_offset <= end && cow_start == (u64)-1)
1826 cow_start = cur_offset;
1828 if (cow_start != (u64)-1) {
1830 ret = fallback_to_cow(inode, locked_page, cow_start, end,
1831 page_started, nr_written);
1838 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1840 if (ret && cur_offset < end)
1841 extent_clear_unlock_delalloc(inode, cur_offset, end,
1842 locked_page, EXTENT_LOCKED |
1843 EXTENT_DELALLOC | EXTENT_DEFRAG |
1844 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1846 PAGE_SET_WRITEBACK |
1847 PAGE_END_WRITEBACK);
1848 btrfs_free_path(path);
1852 static inline int need_force_cow(struct btrfs_inode *inode, u64 start, u64 end)
1855 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1856 !(inode->flags & BTRFS_INODE_PREALLOC))
1860 * @defrag_bytes is a hint value, no spinlock held here,
1861 * if is not zero, it means the file is defragging.
1862 * Force cow if given extent needs to be defragged.
1864 if (inode->defrag_bytes &&
1865 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG, 0, NULL))
1872 * Function to process delayed allocation (create CoW) for ranges which are
1873 * being touched for the first time.
1875 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
1876 u64 start, u64 end, int *page_started, unsigned long *nr_written,
1877 struct writeback_control *wbc)
1880 int force_cow = need_force_cow(inode, start, end);
1882 if (inode->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1883 ret = run_delalloc_nocow(inode, locked_page, start, end,
1884 page_started, 1, nr_written);
1885 } else if (inode->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1886 ret = run_delalloc_nocow(inode, locked_page, start, end,
1887 page_started, 0, nr_written);
1888 } else if (!inode_can_compress(inode) ||
1889 !inode_need_compress(inode, start, end)) {
1890 ret = cow_file_range(inode, locked_page, start, end,
1891 page_started, nr_written, 1);
1893 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1894 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
1895 page_started, nr_written);
1898 btrfs_cleanup_ordered_extents(inode, locked_page, start,
1903 void btrfs_split_delalloc_extent(struct inode *inode,
1904 struct extent_state *orig, u64 split)
1908 /* not delalloc, ignore it */
1909 if (!(orig->state & EXTENT_DELALLOC))
1912 size = orig->end - orig->start + 1;
1913 if (size > BTRFS_MAX_EXTENT_SIZE) {
1918 * See the explanation in btrfs_merge_delalloc_extent, the same
1919 * applies here, just in reverse.
1921 new_size = orig->end - split + 1;
1922 num_extents = count_max_extents(new_size);
1923 new_size = split - orig->start;
1924 num_extents += count_max_extents(new_size);
1925 if (count_max_extents(size) >= num_extents)
1929 spin_lock(&BTRFS_I(inode)->lock);
1930 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1931 spin_unlock(&BTRFS_I(inode)->lock);
1935 * Handle merged delayed allocation extents so we can keep track of new extents
1936 * that are just merged onto old extents, such as when we are doing sequential
1937 * writes, so we can properly account for the metadata space we'll need.
1939 void btrfs_merge_delalloc_extent(struct inode *inode, struct extent_state *new,
1940 struct extent_state *other)
1942 u64 new_size, old_size;
1945 /* not delalloc, ignore it */
1946 if (!(other->state & EXTENT_DELALLOC))
1949 if (new->start > other->start)
1950 new_size = new->end - other->start + 1;
1952 new_size = other->end - new->start + 1;
1954 /* we're not bigger than the max, unreserve the space and go */
1955 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1956 spin_lock(&BTRFS_I(inode)->lock);
1957 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1958 spin_unlock(&BTRFS_I(inode)->lock);
1963 * We have to add up either side to figure out how many extents were
1964 * accounted for before we merged into one big extent. If the number of
1965 * extents we accounted for is <= the amount we need for the new range
1966 * then we can return, otherwise drop. Think of it like this
1970 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1971 * need 2 outstanding extents, on one side we have 1 and the other side
1972 * we have 1 so they are == and we can return. But in this case
1974 * [MAX_SIZE+4k][MAX_SIZE+4k]
1976 * Each range on their own accounts for 2 extents, but merged together
1977 * they are only 3 extents worth of accounting, so we need to drop in
1980 old_size = other->end - other->start + 1;
1981 num_extents = count_max_extents(old_size);
1982 old_size = new->end - new->start + 1;
1983 num_extents += count_max_extents(old_size);
1984 if (count_max_extents(new_size) >= num_extents)
1987 spin_lock(&BTRFS_I(inode)->lock);
1988 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1989 spin_unlock(&BTRFS_I(inode)->lock);
1992 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1993 struct inode *inode)
1995 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1997 spin_lock(&root->delalloc_lock);
1998 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1999 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
2000 &root->delalloc_inodes);
2001 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2002 &BTRFS_I(inode)->runtime_flags);
2003 root->nr_delalloc_inodes++;
2004 if (root->nr_delalloc_inodes == 1) {
2005 spin_lock(&fs_info->delalloc_root_lock);
2006 BUG_ON(!list_empty(&root->delalloc_root));
2007 list_add_tail(&root->delalloc_root,
2008 &fs_info->delalloc_roots);
2009 spin_unlock(&fs_info->delalloc_root_lock);
2012 spin_unlock(&root->delalloc_lock);
2016 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2017 struct btrfs_inode *inode)
2019 struct btrfs_fs_info *fs_info = root->fs_info;
2021 if (!list_empty(&inode->delalloc_inodes)) {
2022 list_del_init(&inode->delalloc_inodes);
2023 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2024 &inode->runtime_flags);
2025 root->nr_delalloc_inodes--;
2026 if (!root->nr_delalloc_inodes) {
2027 ASSERT(list_empty(&root->delalloc_inodes));
2028 spin_lock(&fs_info->delalloc_root_lock);
2029 BUG_ON(list_empty(&root->delalloc_root));
2030 list_del_init(&root->delalloc_root);
2031 spin_unlock(&fs_info->delalloc_root_lock);
2036 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2037 struct btrfs_inode *inode)
2039 spin_lock(&root->delalloc_lock);
2040 __btrfs_del_delalloc_inode(root, inode);
2041 spin_unlock(&root->delalloc_lock);
2045 * Properly track delayed allocation bytes in the inode and to maintain the
2046 * list of inodes that have pending delalloc work to be done.
2048 void btrfs_set_delalloc_extent(struct inode *inode, struct extent_state *state,
2051 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2053 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
2056 * set_bit and clear bit hooks normally require _irqsave/restore
2057 * but in this case, we are only testing for the DELALLOC
2058 * bit, which is only set or cleared with irqs on
2060 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2061 struct btrfs_root *root = BTRFS_I(inode)->root;
2062 u64 len = state->end + 1 - state->start;
2063 u32 num_extents = count_max_extents(len);
2064 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
2066 spin_lock(&BTRFS_I(inode)->lock);
2067 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
2068 spin_unlock(&BTRFS_I(inode)->lock);
2070 /* For sanity tests */
2071 if (btrfs_is_testing(fs_info))
2074 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2075 fs_info->delalloc_batch);
2076 spin_lock(&BTRFS_I(inode)->lock);
2077 BTRFS_I(inode)->delalloc_bytes += len;
2078 if (*bits & EXTENT_DEFRAG)
2079 BTRFS_I(inode)->defrag_bytes += len;
2080 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2081 &BTRFS_I(inode)->runtime_flags))
2082 btrfs_add_delalloc_inodes(root, inode);
2083 spin_unlock(&BTRFS_I(inode)->lock);
2086 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2087 (*bits & EXTENT_DELALLOC_NEW)) {
2088 spin_lock(&BTRFS_I(inode)->lock);
2089 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
2091 spin_unlock(&BTRFS_I(inode)->lock);
2096 * Once a range is no longer delalloc this function ensures that proper
2097 * accounting happens.
2099 void btrfs_clear_delalloc_extent(struct inode *vfs_inode,
2100 struct extent_state *state, unsigned *bits)
2102 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
2103 struct btrfs_fs_info *fs_info = btrfs_sb(vfs_inode->i_sb);
2104 u64 len = state->end + 1 - state->start;
2105 u32 num_extents = count_max_extents(len);
2107 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
2108 spin_lock(&inode->lock);
2109 inode->defrag_bytes -= len;
2110 spin_unlock(&inode->lock);
2114 * set_bit and clear bit hooks normally require _irqsave/restore
2115 * but in this case, we are only testing for the DELALLOC
2116 * bit, which is only set or cleared with irqs on
2118 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
2119 struct btrfs_root *root = inode->root;
2120 bool do_list = !btrfs_is_free_space_inode(inode);
2122 spin_lock(&inode->lock);
2123 btrfs_mod_outstanding_extents(inode, -num_extents);
2124 spin_unlock(&inode->lock);
2127 * We don't reserve metadata space for space cache inodes so we
2128 * don't need to call delalloc_release_metadata if there is an
2131 if (*bits & EXTENT_CLEAR_META_RESV &&
2132 root != fs_info->tree_root)
2133 btrfs_delalloc_release_metadata(inode, len, false);
2135 /* For sanity tests. */
2136 if (btrfs_is_testing(fs_info))
2139 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
2140 do_list && !(state->state & EXTENT_NORESERVE) &&
2141 (*bits & EXTENT_CLEAR_DATA_RESV))
2142 btrfs_free_reserved_data_space_noquota(fs_info, len);
2144 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2145 fs_info->delalloc_batch);
2146 spin_lock(&inode->lock);
2147 inode->delalloc_bytes -= len;
2148 if (do_list && inode->delalloc_bytes == 0 &&
2149 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2150 &inode->runtime_flags))
2151 btrfs_del_delalloc_inode(root, inode);
2152 spin_unlock(&inode->lock);
2155 if ((state->state & EXTENT_DELALLOC_NEW) &&
2156 (*bits & EXTENT_DELALLOC_NEW)) {
2157 spin_lock(&inode->lock);
2158 ASSERT(inode->new_delalloc_bytes >= len);
2159 inode->new_delalloc_bytes -= len;
2160 if (*bits & EXTENT_ADD_INODE_BYTES)
2161 inode_add_bytes(&inode->vfs_inode, len);
2162 spin_unlock(&inode->lock);
2167 * btrfs_bio_fits_in_stripe - Checks whether the size of the given bio will fit
2168 * in a chunk's stripe. This function ensures that bios do not span a
2171 * @page - The page we are about to add to the bio
2172 * @size - size we want to add to the bio
2173 * @bio - bio we want to ensure is smaller than a stripe
2174 * @bio_flags - flags of the bio
2176 * return 1 if page cannot be added to the bio
2177 * return 0 if page can be added to the bio
2178 * return error otherwise
2180 int btrfs_bio_fits_in_stripe(struct page *page, size_t size, struct bio *bio,
2181 unsigned long bio_flags)
2183 struct inode *inode = page->mapping->host;
2184 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2185 u64 logical = bio->bi_iter.bi_sector << 9;
2189 struct btrfs_io_geometry geom;
2191 if (bio_flags & EXTENT_BIO_COMPRESSED)
2194 length = bio->bi_iter.bi_size;
2195 map_length = length;
2196 ret = btrfs_get_io_geometry(fs_info, btrfs_op(bio), logical, map_length,
2201 if (geom.len < length + size)
2207 * in order to insert checksums into the metadata in large chunks,
2208 * we wait until bio submission time. All the pages in the bio are
2209 * checksummed and sums are attached onto the ordered extent record.
2211 * At IO completion time the cums attached on the ordered extent record
2212 * are inserted into the btree
2214 static blk_status_t btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
2215 u64 dio_file_offset)
2217 return btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2221 * extent_io.c submission hook. This does the right thing for csum calculation
2222 * on write, or reading the csums from the tree before a read.
2224 * Rules about async/sync submit,
2225 * a) read: sync submit
2227 * b) write without checksum: sync submit
2229 * c) write with checksum:
2230 * c-1) if bio is issued by fsync: sync submit
2231 * (sync_writers != 0)
2233 * c-2) if root is reloc root: sync submit
2234 * (only in case of buffered IO)
2236 * c-3) otherwise: async submit
2238 blk_status_t btrfs_submit_data_bio(struct inode *inode, struct bio *bio,
2239 int mirror_num, unsigned long bio_flags)
2242 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2243 struct btrfs_root *root = BTRFS_I(inode)->root;
2244 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2245 blk_status_t ret = 0;
2247 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2249 skip_sum = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM) ||
2250 !fs_info->csum_root;
2252 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2253 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2255 if (bio_op(bio) != REQ_OP_WRITE) {
2256 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2260 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2261 ret = btrfs_submit_compressed_read(inode, bio,
2267 * Lookup bio sums does extra checks around whether we
2268 * need to csum or not, which is why we ignore skip_sum
2271 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2276 } else if (async && !skip_sum) {
2277 /* csum items have already been cloned */
2278 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2280 /* we're doing a write, do the async checksumming */
2281 ret = btrfs_wq_submit_bio(inode, bio, mirror_num, bio_flags,
2282 0, btrfs_submit_bio_start);
2284 } else if (!skip_sum) {
2285 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, 0, 0);
2291 ret = btrfs_map_bio(fs_info, bio, mirror_num);
2295 bio->bi_status = ret;
2302 * given a list of ordered sums record them in the inode. This happens
2303 * at IO completion time based on sums calculated at bio submission time.
2305 static int add_pending_csums(struct btrfs_trans_handle *trans,
2306 struct list_head *list)
2308 struct btrfs_ordered_sum *sum;
2311 list_for_each_entry(sum, list, list) {
2312 trans->adding_csums = true;
2313 ret = btrfs_csum_file_blocks(trans, trans->fs_info->csum_root, sum);
2314 trans->adding_csums = false;
2321 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2324 struct extent_state **cached_state)
2326 u64 search_start = start;
2327 const u64 end = start + len - 1;
2329 while (search_start < end) {
2330 const u64 search_len = end - search_start + 1;
2331 struct extent_map *em;
2335 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2339 if (em->block_start != EXTENT_MAP_HOLE)
2343 if (em->start < search_start)
2344 em_len -= search_start - em->start;
2345 if (em_len > search_len)
2346 em_len = search_len;
2348 ret = set_extent_bit(&inode->io_tree, search_start,
2349 search_start + em_len - 1,
2350 EXTENT_DELALLOC_NEW, 0, NULL, cached_state,
2353 search_start = extent_map_end(em);
2354 free_extent_map(em);
2361 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2362 unsigned int extra_bits,
2363 struct extent_state **cached_state)
2365 WARN_ON(PAGE_ALIGNED(end));
2367 if (start >= i_size_read(&inode->vfs_inode) &&
2368 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2370 * There can't be any extents following eof in this case so just
2371 * set the delalloc new bit for the range directly.
2373 extra_bits |= EXTENT_DELALLOC_NEW;
2377 ret = btrfs_find_new_delalloc_bytes(inode, start,
2384 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2388 /* see btrfs_writepage_start_hook for details on why this is required */
2389 struct btrfs_writepage_fixup {
2391 struct inode *inode;
2392 struct btrfs_work work;
2395 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2397 struct btrfs_writepage_fixup *fixup;
2398 struct btrfs_ordered_extent *ordered;
2399 struct extent_state *cached_state = NULL;
2400 struct extent_changeset *data_reserved = NULL;
2402 struct btrfs_inode *inode;
2406 bool free_delalloc_space = true;
2408 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2410 inode = BTRFS_I(fixup->inode);
2411 page_start = page_offset(page);
2412 page_end = page_offset(page) + PAGE_SIZE - 1;
2415 * This is similar to page_mkwrite, we need to reserve the space before
2416 * we take the page lock.
2418 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2424 * Before we queued this fixup, we took a reference on the page.
2425 * page->mapping may go NULL, but it shouldn't be moved to a different
2428 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2430 * Unfortunately this is a little tricky, either
2432 * 1) We got here and our page had already been dealt with and
2433 * we reserved our space, thus ret == 0, so we need to just
2434 * drop our space reservation and bail. This can happen the
2435 * first time we come into the fixup worker, or could happen
2436 * while waiting for the ordered extent.
2437 * 2) Our page was already dealt with, but we happened to get an
2438 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2439 * this case we obviously don't have anything to release, but
2440 * because the page was already dealt with we don't want to
2441 * mark the page with an error, so make sure we're resetting
2442 * ret to 0. This is why we have this check _before_ the ret
2443 * check, because we do not want to have a surprise ENOSPC
2444 * when the page was already properly dealt with.
2447 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2448 btrfs_delalloc_release_space(inode, data_reserved,
2449 page_start, PAGE_SIZE,
2457 * We can't mess with the page state unless it is locked, so now that
2458 * it is locked bail if we failed to make our space reservation.
2463 lock_extent_bits(&inode->io_tree, page_start, page_end, &cached_state);
2465 /* already ordered? We're done */
2466 if (PagePrivate2(page))
2469 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2471 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2474 btrfs_start_ordered_extent(ordered, 1);
2475 btrfs_put_ordered_extent(ordered);
2479 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2485 * Everything went as planned, we're now the owner of a dirty page with
2486 * delayed allocation bits set and space reserved for our COW
2489 * The page was dirty when we started, nothing should have cleaned it.
2491 BUG_ON(!PageDirty(page));
2492 free_delalloc_space = false;
2494 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2495 if (free_delalloc_space)
2496 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2498 unlock_extent_cached(&inode->io_tree, page_start, page_end,
2503 * We hit ENOSPC or other errors. Update the mapping and page
2504 * to reflect the errors and clean the page.
2506 mapping_set_error(page->mapping, ret);
2507 end_extent_writepage(page, ret, page_start, page_end);
2508 clear_page_dirty_for_io(page);
2511 ClearPageChecked(page);
2515 extent_changeset_free(data_reserved);
2517 * As a precaution, do a delayed iput in case it would be the last iput
2518 * that could need flushing space. Recursing back to fixup worker would
2521 btrfs_add_delayed_iput(&inode->vfs_inode);
2525 * There are a few paths in the higher layers of the kernel that directly
2526 * set the page dirty bit without asking the filesystem if it is a
2527 * good idea. This causes problems because we want to make sure COW
2528 * properly happens and the data=ordered rules are followed.
2530 * In our case any range that doesn't have the ORDERED bit set
2531 * hasn't been properly setup for IO. We kick off an async process
2532 * to fix it up. The async helper will wait for ordered extents, set
2533 * the delalloc bit and make it safe to write the page.
2535 int btrfs_writepage_cow_fixup(struct page *page, u64 start, u64 end)
2537 struct inode *inode = page->mapping->host;
2538 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2539 struct btrfs_writepage_fixup *fixup;
2541 /* this page is properly in the ordered list */
2542 if (TestClearPagePrivate2(page))
2546 * PageChecked is set below when we create a fixup worker for this page,
2547 * don't try to create another one if we're already PageChecked()
2549 * The extent_io writepage code will redirty the page if we send back
2552 if (PageChecked(page))
2555 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2560 * We are already holding a reference to this inode from
2561 * write_cache_pages. We need to hold it because the space reservation
2562 * takes place outside of the page lock, and we can't trust
2563 * page->mapping outside of the page lock.
2566 SetPageChecked(page);
2568 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2570 fixup->inode = inode;
2571 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2576 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2577 struct btrfs_inode *inode, u64 file_pos,
2578 struct btrfs_file_extent_item *stack_fi,
2579 const bool update_inode_bytes,
2580 u64 qgroup_reserved)
2582 struct btrfs_root *root = inode->root;
2583 const u64 sectorsize = root->fs_info->sectorsize;
2584 struct btrfs_path *path;
2585 struct extent_buffer *leaf;
2586 struct btrfs_key ins;
2587 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2588 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2589 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2590 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2591 struct btrfs_drop_extents_args drop_args = { 0 };
2594 path = btrfs_alloc_path();
2599 * we may be replacing one extent in the tree with another.
2600 * The new extent is pinned in the extent map, and we don't want
2601 * to drop it from the cache until it is completely in the btree.
2603 * So, tell btrfs_drop_extents to leave this extent in the cache.
2604 * the caller is expected to unpin it and allow it to be merged
2607 drop_args.path = path;
2608 drop_args.start = file_pos;
2609 drop_args.end = file_pos + num_bytes;
2610 drop_args.replace_extent = true;
2611 drop_args.extent_item_size = sizeof(*stack_fi);
2612 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2616 if (!drop_args.extent_inserted) {
2617 ins.objectid = btrfs_ino(inode);
2618 ins.offset = file_pos;
2619 ins.type = BTRFS_EXTENT_DATA_KEY;
2621 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2626 leaf = path->nodes[0];
2627 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2628 write_extent_buffer(leaf, stack_fi,
2629 btrfs_item_ptr_offset(leaf, path->slots[0]),
2630 sizeof(struct btrfs_file_extent_item));
2632 btrfs_mark_buffer_dirty(leaf);
2633 btrfs_release_path(path);
2636 * If we dropped an inline extent here, we know the range where it is
2637 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2638 * number of bytes only for that range contaning the inline extent.
2639 * The remaining of the range will be processed when clearning the
2640 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2642 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2643 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2645 inline_size = drop_args.bytes_found - inline_size;
2646 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2647 drop_args.bytes_found -= inline_size;
2648 num_bytes -= sectorsize;
2651 if (update_inode_bytes)
2652 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2654 ins.objectid = disk_bytenr;
2655 ins.offset = disk_num_bytes;
2656 ins.type = BTRFS_EXTENT_ITEM_KEY;
2658 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2662 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2663 file_pos, qgroup_reserved, &ins);
2665 btrfs_free_path(path);
2670 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2673 struct btrfs_block_group *cache;
2675 cache = btrfs_lookup_block_group(fs_info, start);
2678 spin_lock(&cache->lock);
2679 cache->delalloc_bytes -= len;
2680 spin_unlock(&cache->lock);
2682 btrfs_put_block_group(cache);
2685 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2686 struct btrfs_ordered_extent *oe)
2688 struct btrfs_file_extent_item stack_fi;
2690 bool update_inode_bytes;
2692 memset(&stack_fi, 0, sizeof(stack_fi));
2693 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2694 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2695 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2696 oe->disk_num_bytes);
2697 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags))
2698 logical_len = oe->truncated_len;
2700 logical_len = oe->num_bytes;
2701 btrfs_set_stack_file_extent_num_bytes(&stack_fi, logical_len);
2702 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, logical_len);
2703 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2704 /* Encryption and other encoding is reserved and all 0 */
2707 * For delalloc, when completing an ordered extent we update the inode's
2708 * bytes when clearing the range in the inode's io tree, so pass false
2709 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2710 * except if the ordered extent was truncated.
2712 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
2713 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
2715 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
2716 oe->file_offset, &stack_fi,
2717 update_inode_bytes, oe->qgroup_rsv);
2721 * As ordered data IO finishes, this gets called so we can finish
2722 * an ordered extent if the range of bytes in the file it covers are
2725 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2727 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
2728 struct btrfs_root *root = inode->root;
2729 struct btrfs_fs_info *fs_info = root->fs_info;
2730 struct btrfs_trans_handle *trans = NULL;
2731 struct extent_io_tree *io_tree = &inode->io_tree;
2732 struct extent_state *cached_state = NULL;
2734 int compress_type = 0;
2736 u64 logical_len = ordered_extent->num_bytes;
2737 bool freespace_inode;
2738 bool truncated = false;
2739 bool clear_reserved_extent = true;
2740 unsigned int clear_bits = EXTENT_DEFRAG;
2742 start = ordered_extent->file_offset;
2743 end = start + ordered_extent->num_bytes - 1;
2745 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2746 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2747 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2748 clear_bits |= EXTENT_DELALLOC_NEW;
2750 freespace_inode = btrfs_is_free_space_inode(inode);
2752 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2757 btrfs_free_io_failure_record(inode, start, end);
2759 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2761 logical_len = ordered_extent->truncated_len;
2762 /* Truncated the entire extent, don't bother adding */
2767 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2768 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2770 btrfs_inode_safe_disk_i_size_write(inode, 0);
2771 if (freespace_inode)
2772 trans = btrfs_join_transaction_spacecache(root);
2774 trans = btrfs_join_transaction(root);
2775 if (IS_ERR(trans)) {
2776 ret = PTR_ERR(trans);
2780 trans->block_rsv = &inode->block_rsv;
2781 ret = btrfs_update_inode_fallback(trans, root, inode);
2782 if (ret) /* -ENOMEM or corruption */
2783 btrfs_abort_transaction(trans, ret);
2787 clear_bits |= EXTENT_LOCKED;
2788 lock_extent_bits(io_tree, start, end, &cached_state);
2790 if (freespace_inode)
2791 trans = btrfs_join_transaction_spacecache(root);
2793 trans = btrfs_join_transaction(root);
2794 if (IS_ERR(trans)) {
2795 ret = PTR_ERR(trans);
2800 trans->block_rsv = &inode->block_rsv;
2802 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2803 compress_type = ordered_extent->compress_type;
2804 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2805 BUG_ON(compress_type);
2806 ret = btrfs_mark_extent_written(trans, inode,
2807 ordered_extent->file_offset,
2808 ordered_extent->file_offset +
2811 BUG_ON(root == fs_info->tree_root);
2812 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
2814 clear_reserved_extent = false;
2815 btrfs_release_delalloc_bytes(fs_info,
2816 ordered_extent->disk_bytenr,
2817 ordered_extent->disk_num_bytes);
2820 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
2821 ordered_extent->num_bytes, trans->transid);
2823 btrfs_abort_transaction(trans, ret);
2827 ret = add_pending_csums(trans, &ordered_extent->list);
2829 btrfs_abort_transaction(trans, ret);
2834 * If this is a new delalloc range, clear its new delalloc flag to
2835 * update the inode's number of bytes. This needs to be done first
2836 * before updating the inode item.
2838 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
2839 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
2840 clear_extent_bit(&inode->io_tree, start, end,
2841 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
2842 0, 0, &cached_state);
2844 btrfs_inode_safe_disk_i_size_write(inode, 0);
2845 ret = btrfs_update_inode_fallback(trans, root, inode);
2846 if (ret) { /* -ENOMEM or corruption */
2847 btrfs_abort_transaction(trans, ret);
2852 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
2853 (clear_bits & EXTENT_LOCKED) ? 1 : 0, 0,
2857 btrfs_end_transaction(trans);
2859 if (ret || truncated) {
2860 u64 unwritten_start = start;
2863 unwritten_start += logical_len;
2864 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
2866 /* Drop the cache for the part of the extent we didn't write. */
2867 btrfs_drop_extent_cache(inode, unwritten_start, end, 0);
2870 * If the ordered extent had an IOERR or something else went
2871 * wrong we need to return the space for this ordered extent
2872 * back to the allocator. We only free the extent in the
2873 * truncated case if we didn't write out the extent at all.
2875 * If we made it past insert_reserved_file_extent before we
2876 * errored out then we don't need to do this as the accounting
2877 * has already been done.
2879 if ((ret || !logical_len) &&
2880 clear_reserved_extent &&
2881 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2882 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2884 * Discard the range before returning it back to the
2887 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
2888 btrfs_discard_extent(fs_info,
2889 ordered_extent->disk_bytenr,
2890 ordered_extent->disk_num_bytes,
2892 btrfs_free_reserved_extent(fs_info,
2893 ordered_extent->disk_bytenr,
2894 ordered_extent->disk_num_bytes, 1);
2899 * This needs to be done to make sure anybody waiting knows we are done
2900 * updating everything for this ordered extent.
2902 btrfs_remove_ordered_extent(inode, ordered_extent);
2905 btrfs_put_ordered_extent(ordered_extent);
2906 /* once for the tree */
2907 btrfs_put_ordered_extent(ordered_extent);
2912 static void finish_ordered_fn(struct btrfs_work *work)
2914 struct btrfs_ordered_extent *ordered_extent;
2915 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2916 btrfs_finish_ordered_io(ordered_extent);
2919 void btrfs_writepage_endio_finish_ordered(struct page *page, u64 start,
2920 u64 end, int uptodate)
2922 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
2923 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2924 struct btrfs_ordered_extent *ordered_extent = NULL;
2925 struct btrfs_workqueue *wq;
2927 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2929 ClearPagePrivate2(page);
2930 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2931 end - start + 1, uptodate))
2934 if (btrfs_is_free_space_inode(inode))
2935 wq = fs_info->endio_freespace_worker;
2937 wq = fs_info->endio_write_workers;
2939 btrfs_init_work(&ordered_extent->work, finish_ordered_fn, NULL, NULL);
2940 btrfs_queue_work(wq, &ordered_extent->work);
2944 * check_data_csum - verify checksum of one sector of uncompressed data
2946 * @io_bio: btrfs_io_bio which contains the csum
2947 * @bio_offset: offset to the beginning of the bio (in bytes)
2948 * @page: page where is the data to be verified
2949 * @pgoff: offset inside the page
2951 * The length of such check is always one sector size.
2953 static int check_data_csum(struct inode *inode, struct btrfs_io_bio *io_bio,
2954 u32 bio_offset, struct page *page, u32 pgoff)
2956 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2957 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
2959 u32 len = fs_info->sectorsize;
2960 const u32 csum_size = fs_info->csum_size;
2961 unsigned int offset_sectors;
2963 u8 csum[BTRFS_CSUM_SIZE];
2965 ASSERT(pgoff + len <= PAGE_SIZE);
2967 offset_sectors = bio_offset >> fs_info->sectorsize_bits;
2968 csum_expected = ((u8 *)io_bio->csum) + offset_sectors * csum_size;
2970 kaddr = kmap_atomic(page);
2971 shash->tfm = fs_info->csum_shash;
2973 crypto_shash_digest(shash, kaddr + pgoff, len, csum);
2975 if (memcmp(csum, csum_expected, csum_size))
2978 kunmap_atomic(kaddr);
2981 btrfs_print_data_csum_error(BTRFS_I(inode), page_offset(page) + pgoff,
2982 csum, csum_expected, io_bio->mirror_num);
2984 btrfs_dev_stat_inc_and_print(io_bio->device,
2985 BTRFS_DEV_STAT_CORRUPTION_ERRS);
2986 memset(kaddr + pgoff, 1, len);
2987 flush_dcache_page(page);
2988 kunmap_atomic(kaddr);
2993 * When reads are done, we need to check csums to verify the data is correct.
2994 * if there's a match, we allow the bio to finish. If not, the code in
2995 * extent_io.c will try to find good copies for us.
2997 * @bio_offset: offset to the beginning of the bio (in bytes)
2998 * @start: file offset of the range start
2999 * @end: file offset of the range end (inclusive)
3000 * @mirror: mirror number
3002 int btrfs_verify_data_csum(struct btrfs_io_bio *io_bio, u32 bio_offset,
3003 struct page *page, u64 start, u64 end, int mirror)
3005 struct inode *inode = page->mapping->host;
3006 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3007 struct btrfs_root *root = BTRFS_I(inode)->root;
3008 const u32 sectorsize = root->fs_info->sectorsize;
3011 if (PageChecked(page)) {
3012 ClearPageChecked(page);
3016 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3019 if (!root->fs_info->csum_root)
3022 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3023 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3024 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3028 ASSERT(page_offset(page) <= start &&
3029 end <= page_offset(page) + PAGE_SIZE - 1);
3030 for (pg_off = offset_in_page(start);
3031 pg_off < offset_in_page(end);
3032 pg_off += sectorsize, bio_offset += sectorsize) {
3035 ret = check_data_csum(inode, io_bio, bio_offset, page, pg_off);
3043 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3045 * @inode: The inode we want to perform iput on
3047 * This function uses the generic vfs_inode::i_count to track whether we should
3048 * just decrement it (in case it's > 1) or if this is the last iput then link
3049 * the inode to the delayed iput machinery. Delayed iputs are processed at
3050 * transaction commit time/superblock commit/cleaner kthread.
3052 void btrfs_add_delayed_iput(struct inode *inode)
3054 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3055 struct btrfs_inode *binode = BTRFS_I(inode);
3057 if (atomic_add_unless(&inode->i_count, -1, 1))
3060 atomic_inc(&fs_info->nr_delayed_iputs);
3061 spin_lock(&fs_info->delayed_iput_lock);
3062 ASSERT(list_empty(&binode->delayed_iput));
3063 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3064 spin_unlock(&fs_info->delayed_iput_lock);
3065 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3066 wake_up_process(fs_info->cleaner_kthread);
3069 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3070 struct btrfs_inode *inode)
3072 list_del_init(&inode->delayed_iput);
3073 spin_unlock(&fs_info->delayed_iput_lock);
3074 iput(&inode->vfs_inode);
3075 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3076 wake_up(&fs_info->delayed_iputs_wait);
3077 spin_lock(&fs_info->delayed_iput_lock);
3080 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3081 struct btrfs_inode *inode)
3083 if (!list_empty(&inode->delayed_iput)) {
3084 spin_lock(&fs_info->delayed_iput_lock);
3085 if (!list_empty(&inode->delayed_iput))
3086 run_delayed_iput_locked(fs_info, inode);
3087 spin_unlock(&fs_info->delayed_iput_lock);
3091 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3094 spin_lock(&fs_info->delayed_iput_lock);
3095 while (!list_empty(&fs_info->delayed_iputs)) {
3096 struct btrfs_inode *inode;
3098 inode = list_first_entry(&fs_info->delayed_iputs,
3099 struct btrfs_inode, delayed_iput);
3100 run_delayed_iput_locked(fs_info, inode);
3102 spin_unlock(&fs_info->delayed_iput_lock);
3106 * btrfs_wait_on_delayed_iputs - wait on the delayed iputs to be done running
3107 * @fs_info - the fs_info for this fs
3108 * @return - EINTR if we were killed, 0 if nothing's pending
3110 * This will wait on any delayed iputs that are currently running with KILLABLE
3111 * set. Once they are all done running we will return, unless we are killed in
3112 * which case we return EINTR. This helps in user operations like fallocate etc
3113 * that might get blocked on the iputs.
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);
4724 if (!PageUptodate(page)) {
4725 ret = btrfs_readpage(NULL, page);
4727 if (page->mapping != mapping) {
4732 if (!PageUptodate(page)) {
4737 wait_on_page_writeback(page);
4739 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4740 set_page_extent_mapped(page);
4742 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4744 unlock_extent_cached(io_tree, block_start, block_end,
4748 btrfs_start_ordered_extent(ordered, 1);
4749 btrfs_put_ordered_extent(ordered);
4753 clear_extent_bit(&inode->io_tree, block_start, block_end,
4754 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4755 0, 0, &cached_state);
4757 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4760 unlock_extent_cached(io_tree, block_start, block_end,
4765 if (offset != blocksize) {
4767 len = blocksize - offset;
4770 memset(kaddr + (block_start - page_offset(page)),
4773 memset(kaddr + (block_start - page_offset(page)) + offset,
4775 flush_dcache_page(page);
4778 ClearPageChecked(page);
4779 set_page_dirty(page);
4780 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4782 if (only_release_metadata)
4783 set_extent_bit(&inode->io_tree, block_start, block_end,
4784 EXTENT_NORESERVE, 0, NULL, NULL, GFP_NOFS, NULL);
4788 if (only_release_metadata)
4789 btrfs_delalloc_release_metadata(inode, blocksize, true);
4791 btrfs_delalloc_release_space(inode, data_reserved,
4792 block_start, blocksize, true);
4794 btrfs_delalloc_release_extents(inode, blocksize);
4798 if (only_release_metadata)
4799 btrfs_check_nocow_unlock(inode);
4800 extent_changeset_free(data_reserved);
4804 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4805 u64 offset, u64 len)
4807 struct btrfs_fs_info *fs_info = root->fs_info;
4808 struct btrfs_trans_handle *trans;
4809 struct btrfs_drop_extents_args drop_args = { 0 };
4813 * Still need to make sure the inode looks like it's been updated so
4814 * that any holes get logged if we fsync.
4816 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4817 inode->last_trans = fs_info->generation;
4818 inode->last_sub_trans = root->log_transid;
4819 inode->last_log_commit = root->last_log_commit;
4824 * 1 - for the one we're dropping
4825 * 1 - for the one we're adding
4826 * 1 - for updating the inode.
4828 trans = btrfs_start_transaction(root, 3);
4830 return PTR_ERR(trans);
4832 drop_args.start = offset;
4833 drop_args.end = offset + len;
4834 drop_args.drop_cache = true;
4836 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4838 btrfs_abort_transaction(trans, ret);
4839 btrfs_end_transaction(trans);
4843 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode),
4844 offset, 0, 0, len, 0, len, 0, 0, 0);
4846 btrfs_abort_transaction(trans, ret);
4848 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4849 btrfs_update_inode(trans, root, inode);
4851 btrfs_end_transaction(trans);
4856 * This function puts in dummy file extents for the area we're creating a hole
4857 * for. So if we are truncating this file to a larger size we need to insert
4858 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4859 * the range between oldsize and size
4861 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4863 struct btrfs_root *root = inode->root;
4864 struct btrfs_fs_info *fs_info = root->fs_info;
4865 struct extent_io_tree *io_tree = &inode->io_tree;
4866 struct extent_map *em = NULL;
4867 struct extent_state *cached_state = NULL;
4868 struct extent_map_tree *em_tree = &inode->extent_tree;
4869 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4870 u64 block_end = ALIGN(size, fs_info->sectorsize);
4877 * If our size started in the middle of a block we need to zero out the
4878 * rest of the block before we expand the i_size, otherwise we could
4879 * expose stale data.
4881 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4885 if (size <= hole_start)
4888 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4890 cur_offset = hole_start;
4892 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4893 block_end - cur_offset);
4899 last_byte = min(extent_map_end(em), block_end);
4900 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4901 hole_size = last_byte - cur_offset;
4903 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4904 struct extent_map *hole_em;
4906 err = maybe_insert_hole(root, inode, cur_offset,
4911 err = btrfs_inode_set_file_extent_range(inode,
4912 cur_offset, hole_size);
4916 btrfs_drop_extent_cache(inode, cur_offset,
4917 cur_offset + hole_size - 1, 0);
4918 hole_em = alloc_extent_map();
4920 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4921 &inode->runtime_flags);
4924 hole_em->start = cur_offset;
4925 hole_em->len = hole_size;
4926 hole_em->orig_start = cur_offset;
4928 hole_em->block_start = EXTENT_MAP_HOLE;
4929 hole_em->block_len = 0;
4930 hole_em->orig_block_len = 0;
4931 hole_em->ram_bytes = hole_size;
4932 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4933 hole_em->generation = fs_info->generation;
4936 write_lock(&em_tree->lock);
4937 err = add_extent_mapping(em_tree, hole_em, 1);
4938 write_unlock(&em_tree->lock);
4941 btrfs_drop_extent_cache(inode, cur_offset,
4945 free_extent_map(hole_em);
4947 err = btrfs_inode_set_file_extent_range(inode,
4948 cur_offset, hole_size);
4953 free_extent_map(em);
4955 cur_offset = last_byte;
4956 if (cur_offset >= block_end)
4959 free_extent_map(em);
4960 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
4964 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4966 struct btrfs_root *root = BTRFS_I(inode)->root;
4967 struct btrfs_trans_handle *trans;
4968 loff_t oldsize = i_size_read(inode);
4969 loff_t newsize = attr->ia_size;
4970 int mask = attr->ia_valid;
4974 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4975 * special case where we need to update the times despite not having
4976 * these flags set. For all other operations the VFS set these flags
4977 * explicitly if it wants a timestamp update.
4979 if (newsize != oldsize) {
4980 inode_inc_iversion(inode);
4981 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4982 inode->i_ctime = inode->i_mtime =
4983 current_time(inode);
4986 if (newsize > oldsize) {
4988 * Don't do an expanding truncate while snapshotting is ongoing.
4989 * This is to ensure the snapshot captures a fully consistent
4990 * state of this file - if the snapshot captures this expanding
4991 * truncation, it must capture all writes that happened before
4994 btrfs_drew_write_lock(&root->snapshot_lock);
4995 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4997 btrfs_drew_write_unlock(&root->snapshot_lock);
5001 trans = btrfs_start_transaction(root, 1);
5002 if (IS_ERR(trans)) {
5003 btrfs_drew_write_unlock(&root->snapshot_lock);
5004 return PTR_ERR(trans);
5007 i_size_write(inode, newsize);
5008 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5009 pagecache_isize_extended(inode, oldsize, newsize);
5010 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5011 btrfs_drew_write_unlock(&root->snapshot_lock);
5012 btrfs_end_transaction(trans);
5016 * We're truncating a file that used to have good data down to
5017 * zero. Make sure any new writes to the file get on disk
5021 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5022 &BTRFS_I(inode)->runtime_flags);
5024 truncate_setsize(inode, newsize);
5026 inode_dio_wait(inode);
5028 ret = btrfs_truncate(inode, newsize == oldsize);
5029 if (ret && inode->i_nlink) {
5033 * Truncate failed, so fix up the in-memory size. We
5034 * adjusted disk_i_size down as we removed extents, so
5035 * wait for disk_i_size to be stable and then update the
5036 * in-memory size to match.
5038 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5041 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5048 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5050 struct inode *inode = d_inode(dentry);
5051 struct btrfs_root *root = BTRFS_I(inode)->root;
5054 if (btrfs_root_readonly(root))
5057 err = setattr_prepare(dentry, attr);
5061 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5062 err = btrfs_setsize(inode, attr);
5067 if (attr->ia_valid) {
5068 setattr_copy(inode, attr);
5069 inode_inc_iversion(inode);
5070 err = btrfs_dirty_inode(inode);
5072 if (!err && attr->ia_valid & ATTR_MODE)
5073 err = posix_acl_chmod(inode, inode->i_mode);
5080 * While truncating the inode pages during eviction, we get the VFS calling
5081 * btrfs_invalidatepage() against each page of the inode. This is slow because
5082 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5083 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5084 * extent_state structures over and over, wasting lots of time.
5086 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5087 * those expensive operations on a per page basis and do only the ordered io
5088 * finishing, while we release here the extent_map and extent_state structures,
5089 * without the excessive merging and splitting.
5091 static void evict_inode_truncate_pages(struct inode *inode)
5093 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5094 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5095 struct rb_node *node;
5097 ASSERT(inode->i_state & I_FREEING);
5098 truncate_inode_pages_final(&inode->i_data);
5100 write_lock(&map_tree->lock);
5101 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5102 struct extent_map *em;
5104 node = rb_first_cached(&map_tree->map);
5105 em = rb_entry(node, struct extent_map, rb_node);
5106 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5107 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5108 remove_extent_mapping(map_tree, em);
5109 free_extent_map(em);
5110 if (need_resched()) {
5111 write_unlock(&map_tree->lock);
5113 write_lock(&map_tree->lock);
5116 write_unlock(&map_tree->lock);
5119 * Keep looping until we have no more ranges in the io tree.
5120 * We can have ongoing bios started by readahead that have
5121 * their endio callback (extent_io.c:end_bio_extent_readpage)
5122 * still in progress (unlocked the pages in the bio but did not yet
5123 * unlocked the ranges in the io tree). Therefore this means some
5124 * ranges can still be locked and eviction started because before
5125 * submitting those bios, which are executed by a separate task (work
5126 * queue kthread), inode references (inode->i_count) were not taken
5127 * (which would be dropped in the end io callback of each bio).
5128 * Therefore here we effectively end up waiting for those bios and
5129 * anyone else holding locked ranges without having bumped the inode's
5130 * reference count - if we don't do it, when they access the inode's
5131 * io_tree to unlock a range it may be too late, leading to an
5132 * use-after-free issue.
5134 spin_lock(&io_tree->lock);
5135 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5136 struct extent_state *state;
5137 struct extent_state *cached_state = NULL;
5140 unsigned state_flags;
5142 node = rb_first(&io_tree->state);
5143 state = rb_entry(node, struct extent_state, rb_node);
5144 start = state->start;
5146 state_flags = state->state;
5147 spin_unlock(&io_tree->lock);
5149 lock_extent_bits(io_tree, start, end, &cached_state);
5152 * If still has DELALLOC flag, the extent didn't reach disk,
5153 * and its reserved space won't be freed by delayed_ref.
5154 * So we need to free its reserved space here.
5155 * (Refer to comment in btrfs_invalidatepage, case 2)
5157 * Note, end is the bytenr of last byte, so we need + 1 here.
5159 if (state_flags & EXTENT_DELALLOC)
5160 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5163 clear_extent_bit(io_tree, start, end,
5164 EXTENT_LOCKED | EXTENT_DELALLOC |
5165 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
5169 spin_lock(&io_tree->lock);
5171 spin_unlock(&io_tree->lock);
5174 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5175 struct btrfs_block_rsv *rsv)
5177 struct btrfs_fs_info *fs_info = root->fs_info;
5178 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5179 struct btrfs_trans_handle *trans;
5180 u64 delayed_refs_extra = btrfs_calc_insert_metadata_size(fs_info, 1);
5184 * Eviction should be taking place at some place safe because of our
5185 * delayed iputs. However the normal flushing code will run delayed
5186 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5188 * We reserve the delayed_refs_extra here again because we can't use
5189 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5190 * above. We reserve our extra bit here because we generate a ton of
5191 * delayed refs activity by truncating.
5193 * If we cannot make our reservation we'll attempt to steal from the
5194 * global reserve, because we really want to be able to free up space.
5196 ret = btrfs_block_rsv_refill(root, rsv, rsv->size + delayed_refs_extra,
5197 BTRFS_RESERVE_FLUSH_EVICT);
5200 * Try to steal from the global reserve if there is space for
5203 if (btrfs_check_space_for_delayed_refs(fs_info) ||
5204 btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, 0)) {
5206 "could not allocate space for delete; will truncate on mount");
5207 return ERR_PTR(-ENOSPC);
5209 delayed_refs_extra = 0;
5212 trans = btrfs_join_transaction(root);
5216 if (delayed_refs_extra) {
5217 trans->block_rsv = &fs_info->trans_block_rsv;
5218 trans->bytes_reserved = delayed_refs_extra;
5219 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5220 delayed_refs_extra, 1);
5225 void btrfs_evict_inode(struct inode *inode)
5227 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5228 struct btrfs_trans_handle *trans;
5229 struct btrfs_root *root = BTRFS_I(inode)->root;
5230 struct btrfs_block_rsv *rsv;
5233 trace_btrfs_inode_evict(inode);
5240 evict_inode_truncate_pages(inode);
5242 if (inode->i_nlink &&
5243 ((btrfs_root_refs(&root->root_item) != 0 &&
5244 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5245 btrfs_is_free_space_inode(BTRFS_I(inode))))
5248 if (is_bad_inode(inode))
5251 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5253 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5256 if (inode->i_nlink > 0) {
5257 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5258 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5262 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5266 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5269 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5272 btrfs_i_size_write(BTRFS_I(inode), 0);
5275 trans = evict_refill_and_join(root, rsv);
5279 trans->block_rsv = rsv;
5281 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
5283 trans->block_rsv = &fs_info->trans_block_rsv;
5284 btrfs_end_transaction(trans);
5285 btrfs_btree_balance_dirty(fs_info);
5286 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5293 * Errors here aren't a big deal, it just means we leave orphan items in
5294 * the tree. They will be cleaned up on the next mount. If the inode
5295 * number gets reused, cleanup deletes the orphan item without doing
5296 * anything, and unlink reuses the existing orphan item.
5298 * If it turns out that we are dropping too many of these, we might want
5299 * to add a mechanism for retrying these after a commit.
5301 trans = evict_refill_and_join(root, rsv);
5302 if (!IS_ERR(trans)) {
5303 trans->block_rsv = rsv;
5304 btrfs_orphan_del(trans, BTRFS_I(inode));
5305 trans->block_rsv = &fs_info->trans_block_rsv;
5306 btrfs_end_transaction(trans);
5310 btrfs_free_block_rsv(fs_info, rsv);
5313 * If we didn't successfully delete, the orphan item will still be in
5314 * the tree and we'll retry on the next mount. Again, we might also want
5315 * to retry these periodically in the future.
5317 btrfs_remove_delayed_node(BTRFS_I(inode));
5322 * Return the key found in the dir entry in the location pointer, fill @type
5323 * with BTRFS_FT_*, and return 0.
5325 * If no dir entries were found, returns -ENOENT.
5326 * If found a corrupted location in dir entry, returns -EUCLEAN.
5328 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5329 struct btrfs_key *location, u8 *type)
5331 const char *name = dentry->d_name.name;
5332 int namelen = dentry->d_name.len;
5333 struct btrfs_dir_item *di;
5334 struct btrfs_path *path;
5335 struct btrfs_root *root = BTRFS_I(dir)->root;
5338 path = btrfs_alloc_path();
5342 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5344 if (IS_ERR_OR_NULL(di)) {
5345 ret = di ? PTR_ERR(di) : -ENOENT;
5349 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5350 if (location->type != BTRFS_INODE_ITEM_KEY &&
5351 location->type != BTRFS_ROOT_ITEM_KEY) {
5353 btrfs_warn(root->fs_info,
5354 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5355 __func__, name, btrfs_ino(BTRFS_I(dir)),
5356 location->objectid, location->type, location->offset);
5359 *type = btrfs_dir_type(path->nodes[0], di);
5361 btrfs_free_path(path);
5366 * when we hit a tree root in a directory, the btrfs part of the inode
5367 * needs to be changed to reflect the root directory of the tree root. This
5368 * is kind of like crossing a mount point.
5370 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5372 struct dentry *dentry,
5373 struct btrfs_key *location,
5374 struct btrfs_root **sub_root)
5376 struct btrfs_path *path;
5377 struct btrfs_root *new_root;
5378 struct btrfs_root_ref *ref;
5379 struct extent_buffer *leaf;
5380 struct btrfs_key key;
5384 path = btrfs_alloc_path();
5391 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5392 key.type = BTRFS_ROOT_REF_KEY;
5393 key.offset = location->objectid;
5395 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5402 leaf = path->nodes[0];
5403 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5404 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5405 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5408 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5409 (unsigned long)(ref + 1),
5410 dentry->d_name.len);
5414 btrfs_release_path(path);
5416 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5417 if (IS_ERR(new_root)) {
5418 err = PTR_ERR(new_root);
5422 *sub_root = new_root;
5423 location->objectid = btrfs_root_dirid(&new_root->root_item);
5424 location->type = BTRFS_INODE_ITEM_KEY;
5425 location->offset = 0;
5428 btrfs_free_path(path);
5432 static void inode_tree_add(struct inode *inode)
5434 struct btrfs_root *root = BTRFS_I(inode)->root;
5435 struct btrfs_inode *entry;
5437 struct rb_node *parent;
5438 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5439 u64 ino = btrfs_ino(BTRFS_I(inode));
5441 if (inode_unhashed(inode))
5444 spin_lock(&root->inode_lock);
5445 p = &root->inode_tree.rb_node;
5448 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5450 if (ino < btrfs_ino(entry))
5451 p = &parent->rb_left;
5452 else if (ino > btrfs_ino(entry))
5453 p = &parent->rb_right;
5455 WARN_ON(!(entry->vfs_inode.i_state &
5456 (I_WILL_FREE | I_FREEING)));
5457 rb_replace_node(parent, new, &root->inode_tree);
5458 RB_CLEAR_NODE(parent);
5459 spin_unlock(&root->inode_lock);
5463 rb_link_node(new, parent, p);
5464 rb_insert_color(new, &root->inode_tree);
5465 spin_unlock(&root->inode_lock);
5468 static void inode_tree_del(struct btrfs_inode *inode)
5470 struct btrfs_root *root = inode->root;
5473 spin_lock(&root->inode_lock);
5474 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5475 rb_erase(&inode->rb_node, &root->inode_tree);
5476 RB_CLEAR_NODE(&inode->rb_node);
5477 empty = RB_EMPTY_ROOT(&root->inode_tree);
5479 spin_unlock(&root->inode_lock);
5481 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5482 spin_lock(&root->inode_lock);
5483 empty = RB_EMPTY_ROOT(&root->inode_tree);
5484 spin_unlock(&root->inode_lock);
5486 btrfs_add_dead_root(root);
5491 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5493 struct btrfs_iget_args *args = p;
5495 inode->i_ino = args->ino;
5496 BTRFS_I(inode)->location.objectid = args->ino;
5497 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5498 BTRFS_I(inode)->location.offset = 0;
5499 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5500 BUG_ON(args->root && !BTRFS_I(inode)->root);
5504 static int btrfs_find_actor(struct inode *inode, void *opaque)
5506 struct btrfs_iget_args *args = opaque;
5508 return args->ino == BTRFS_I(inode)->location.objectid &&
5509 args->root == BTRFS_I(inode)->root;
5512 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5513 struct btrfs_root *root)
5515 struct inode *inode;
5516 struct btrfs_iget_args args;
5517 unsigned long hashval = btrfs_inode_hash(ino, root);
5522 inode = iget5_locked(s, hashval, btrfs_find_actor,
5523 btrfs_init_locked_inode,
5529 * Get an inode object given its inode number and corresponding root.
5530 * Path can be preallocated to prevent recursing back to iget through
5531 * allocator. NULL is also valid but may require an additional allocation
5534 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5535 struct btrfs_root *root, struct btrfs_path *path)
5537 struct inode *inode;
5539 inode = btrfs_iget_locked(s, ino, root);
5541 return ERR_PTR(-ENOMEM);
5543 if (inode->i_state & I_NEW) {
5546 ret = btrfs_read_locked_inode(inode, path);
5548 inode_tree_add(inode);
5549 unlock_new_inode(inode);
5553 * ret > 0 can come from btrfs_search_slot called by
5554 * btrfs_read_locked_inode, this means the inode item
5559 inode = ERR_PTR(ret);
5566 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5568 return btrfs_iget_path(s, ino, root, NULL);
5571 static struct inode *new_simple_dir(struct super_block *s,
5572 struct btrfs_key *key,
5573 struct btrfs_root *root)
5575 struct inode *inode = new_inode(s);
5578 return ERR_PTR(-ENOMEM);
5580 BTRFS_I(inode)->root = btrfs_grab_root(root);
5581 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5582 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5584 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5586 * We only need lookup, the rest is read-only and there's no inode
5587 * associated with the dentry
5589 inode->i_op = &simple_dir_inode_operations;
5590 inode->i_opflags &= ~IOP_XATTR;
5591 inode->i_fop = &simple_dir_operations;
5592 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5593 inode->i_mtime = current_time(inode);
5594 inode->i_atime = inode->i_mtime;
5595 inode->i_ctime = inode->i_mtime;
5596 BTRFS_I(inode)->i_otime = inode->i_mtime;
5601 static inline u8 btrfs_inode_type(struct inode *inode)
5604 * Compile-time asserts that generic FT_* types still match
5607 BUILD_BUG_ON(BTRFS_FT_UNKNOWN != FT_UNKNOWN);
5608 BUILD_BUG_ON(BTRFS_FT_REG_FILE != FT_REG_FILE);
5609 BUILD_BUG_ON(BTRFS_FT_DIR != FT_DIR);
5610 BUILD_BUG_ON(BTRFS_FT_CHRDEV != FT_CHRDEV);
5611 BUILD_BUG_ON(BTRFS_FT_BLKDEV != FT_BLKDEV);
5612 BUILD_BUG_ON(BTRFS_FT_FIFO != FT_FIFO);
5613 BUILD_BUG_ON(BTRFS_FT_SOCK != FT_SOCK);
5614 BUILD_BUG_ON(BTRFS_FT_SYMLINK != FT_SYMLINK);
5616 return fs_umode_to_ftype(inode->i_mode);
5619 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5621 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5622 struct inode *inode;
5623 struct btrfs_root *root = BTRFS_I(dir)->root;
5624 struct btrfs_root *sub_root = root;
5625 struct btrfs_key location;
5629 if (dentry->d_name.len > BTRFS_NAME_LEN)
5630 return ERR_PTR(-ENAMETOOLONG);
5632 ret = btrfs_inode_by_name(dir, dentry, &location, &di_type);
5634 return ERR_PTR(ret);
5636 if (location.type == BTRFS_INODE_ITEM_KEY) {
5637 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5641 /* Do extra check against inode mode with di_type */
5642 if (btrfs_inode_type(inode) != di_type) {
5644 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5645 inode->i_mode, btrfs_inode_type(inode),
5648 return ERR_PTR(-EUCLEAN);
5653 ret = fixup_tree_root_location(fs_info, dir, dentry,
5654 &location, &sub_root);
5657 inode = ERR_PTR(ret);
5659 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5661 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5663 if (root != sub_root)
5664 btrfs_put_root(sub_root);
5666 if (!IS_ERR(inode) && root != sub_root) {
5667 down_read(&fs_info->cleanup_work_sem);
5668 if (!sb_rdonly(inode->i_sb))
5669 ret = btrfs_orphan_cleanup(sub_root);
5670 up_read(&fs_info->cleanup_work_sem);
5673 inode = ERR_PTR(ret);
5680 static int btrfs_dentry_delete(const struct dentry *dentry)
5682 struct btrfs_root *root;
5683 struct inode *inode = d_inode(dentry);
5685 if (!inode && !IS_ROOT(dentry))
5686 inode = d_inode(dentry->d_parent);
5689 root = BTRFS_I(inode)->root;
5690 if (btrfs_root_refs(&root->root_item) == 0)
5693 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5699 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5702 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5704 if (inode == ERR_PTR(-ENOENT))
5706 return d_splice_alias(inode, dentry);
5710 * All this infrastructure exists because dir_emit can fault, and we are holding
5711 * the tree lock when doing readdir. For now just allocate a buffer and copy
5712 * our information into that, and then dir_emit from the buffer. This is
5713 * similar to what NFS does, only we don't keep the buffer around in pagecache
5714 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5715 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5718 static int btrfs_opendir(struct inode *inode, struct file *file)
5720 struct btrfs_file_private *private;
5722 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5725 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5726 if (!private->filldir_buf) {
5730 file->private_data = private;
5741 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5744 struct dir_entry *entry = addr;
5745 char *name = (char *)(entry + 1);
5747 ctx->pos = get_unaligned(&entry->offset);
5748 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5749 get_unaligned(&entry->ino),
5750 get_unaligned(&entry->type)))
5752 addr += sizeof(struct dir_entry) +
5753 get_unaligned(&entry->name_len);
5759 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5761 struct inode *inode = file_inode(file);
5762 struct btrfs_root *root = BTRFS_I(inode)->root;
5763 struct btrfs_file_private *private = file->private_data;
5764 struct btrfs_dir_item *di;
5765 struct btrfs_key key;
5766 struct btrfs_key found_key;
5767 struct btrfs_path *path;
5769 struct list_head ins_list;
5770 struct list_head del_list;
5772 struct extent_buffer *leaf;
5779 struct btrfs_key location;
5781 if (!dir_emit_dots(file, ctx))
5784 path = btrfs_alloc_path();
5788 addr = private->filldir_buf;
5789 path->reada = READA_FORWARD;
5791 INIT_LIST_HEAD(&ins_list);
5792 INIT_LIST_HEAD(&del_list);
5793 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5796 key.type = BTRFS_DIR_INDEX_KEY;
5797 key.offset = ctx->pos;
5798 key.objectid = btrfs_ino(BTRFS_I(inode));
5800 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5805 struct dir_entry *entry;
5807 leaf = path->nodes[0];
5808 slot = path->slots[0];
5809 if (slot >= btrfs_header_nritems(leaf)) {
5810 ret = btrfs_next_leaf(root, path);
5818 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5820 if (found_key.objectid != key.objectid)
5822 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5824 if (found_key.offset < ctx->pos)
5826 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5828 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5829 name_len = btrfs_dir_name_len(leaf, di);
5830 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5832 btrfs_release_path(path);
5833 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5836 addr = private->filldir_buf;
5843 put_unaligned(name_len, &entry->name_len);
5844 name_ptr = (char *)(entry + 1);
5845 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5847 put_unaligned(fs_ftype_to_dtype(btrfs_dir_type(leaf, di)),
5849 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5850 put_unaligned(location.objectid, &entry->ino);
5851 put_unaligned(found_key.offset, &entry->offset);
5853 addr += sizeof(struct dir_entry) + name_len;
5854 total_len += sizeof(struct dir_entry) + name_len;
5858 btrfs_release_path(path);
5860 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5864 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5869 * Stop new entries from being returned after we return the last
5872 * New directory entries are assigned a strictly increasing
5873 * offset. This means that new entries created during readdir
5874 * are *guaranteed* to be seen in the future by that readdir.
5875 * This has broken buggy programs which operate on names as
5876 * they're returned by readdir. Until we re-use freed offsets
5877 * we have this hack to stop new entries from being returned
5878 * under the assumption that they'll never reach this huge
5881 * This is being careful not to overflow 32bit loff_t unless the
5882 * last entry requires it because doing so has broken 32bit apps
5885 if (ctx->pos >= INT_MAX)
5886 ctx->pos = LLONG_MAX;
5893 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5894 btrfs_free_path(path);
5899 * This is somewhat expensive, updating the tree every time the
5900 * inode changes. But, it is most likely to find the inode in cache.
5901 * FIXME, needs more benchmarking...there are no reasons other than performance
5902 * to keep or drop this code.
5904 static int btrfs_dirty_inode(struct inode *inode)
5906 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5907 struct btrfs_root *root = BTRFS_I(inode)->root;
5908 struct btrfs_trans_handle *trans;
5911 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5914 trans = btrfs_join_transaction(root);
5916 return PTR_ERR(trans);
5918 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5919 if (ret && ret == -ENOSPC) {
5920 /* whoops, lets try again with the full transaction */
5921 btrfs_end_transaction(trans);
5922 trans = btrfs_start_transaction(root, 1);
5924 return PTR_ERR(trans);
5926 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5928 btrfs_end_transaction(trans);
5929 if (BTRFS_I(inode)->delayed_node)
5930 btrfs_balance_delayed_items(fs_info);
5936 * This is a copy of file_update_time. We need this so we can return error on
5937 * ENOSPC for updating the inode in the case of file write and mmap writes.
5939 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5942 struct btrfs_root *root = BTRFS_I(inode)->root;
5943 bool dirty = flags & ~S_VERSION;
5945 if (btrfs_root_readonly(root))
5948 if (flags & S_VERSION)
5949 dirty |= inode_maybe_inc_iversion(inode, dirty);
5950 if (flags & S_CTIME)
5951 inode->i_ctime = *now;
5952 if (flags & S_MTIME)
5953 inode->i_mtime = *now;
5954 if (flags & S_ATIME)
5955 inode->i_atime = *now;
5956 return dirty ? btrfs_dirty_inode(inode) : 0;
5960 * find the highest existing sequence number in a directory
5961 * and then set the in-memory index_cnt variable to reflect
5962 * free sequence numbers
5964 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5966 struct btrfs_root *root = inode->root;
5967 struct btrfs_key key, found_key;
5968 struct btrfs_path *path;
5969 struct extent_buffer *leaf;
5972 key.objectid = btrfs_ino(inode);
5973 key.type = BTRFS_DIR_INDEX_KEY;
5974 key.offset = (u64)-1;
5976 path = btrfs_alloc_path();
5980 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5983 /* FIXME: we should be able to handle this */
5989 * MAGIC NUMBER EXPLANATION:
5990 * since we search a directory based on f_pos we have to start at 2
5991 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5992 * else has to start at 2
5994 if (path->slots[0] == 0) {
5995 inode->index_cnt = 2;
6001 leaf = path->nodes[0];
6002 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6004 if (found_key.objectid != btrfs_ino(inode) ||
6005 found_key.type != BTRFS_DIR_INDEX_KEY) {
6006 inode->index_cnt = 2;
6010 inode->index_cnt = found_key.offset + 1;
6012 btrfs_free_path(path);
6017 * helper to find a free sequence number in a given directory. This current
6018 * code is very simple, later versions will do smarter things in the btree
6020 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6024 if (dir->index_cnt == (u64)-1) {
6025 ret = btrfs_inode_delayed_dir_index_count(dir);
6027 ret = btrfs_set_inode_index_count(dir);
6033 *index = dir->index_cnt;
6039 static int btrfs_insert_inode_locked(struct inode *inode)
6041 struct btrfs_iget_args args;
6043 args.ino = BTRFS_I(inode)->location.objectid;
6044 args.root = BTRFS_I(inode)->root;
6046 return insert_inode_locked4(inode,
6047 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6048 btrfs_find_actor, &args);
6052 * Inherit flags from the parent inode.
6054 * Currently only the compression flags and the cow flags are inherited.
6056 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6063 flags = BTRFS_I(dir)->flags;
6065 if (flags & BTRFS_INODE_NOCOMPRESS) {
6066 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6067 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6068 } else if (flags & BTRFS_INODE_COMPRESS) {
6069 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6070 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6073 if (flags & BTRFS_INODE_NODATACOW) {
6074 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6075 if (S_ISREG(inode->i_mode))
6076 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6079 btrfs_sync_inode_flags_to_i_flags(inode);
6082 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6083 struct btrfs_root *root,
6085 const char *name, int name_len,
6086 u64 ref_objectid, u64 objectid,
6087 umode_t mode, u64 *index)
6089 struct btrfs_fs_info *fs_info = root->fs_info;
6090 struct inode *inode;
6091 struct btrfs_inode_item *inode_item;
6092 struct btrfs_key *location;
6093 struct btrfs_path *path;
6094 struct btrfs_inode_ref *ref;
6095 struct btrfs_key key[2];
6097 int nitems = name ? 2 : 1;
6099 unsigned int nofs_flag;
6102 path = btrfs_alloc_path();
6104 return ERR_PTR(-ENOMEM);
6106 nofs_flag = memalloc_nofs_save();
6107 inode = new_inode(fs_info->sb);
6108 memalloc_nofs_restore(nofs_flag);
6110 btrfs_free_path(path);
6111 return ERR_PTR(-ENOMEM);
6115 * O_TMPFILE, set link count to 0, so that after this point,
6116 * we fill in an inode item with the correct link count.
6119 set_nlink(inode, 0);
6122 * we have to initialize this early, so we can reclaim the inode
6123 * number if we fail afterwards in this function.
6125 inode->i_ino = objectid;
6128 trace_btrfs_inode_request(dir);
6130 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6132 btrfs_free_path(path);
6134 return ERR_PTR(ret);
6140 * index_cnt is ignored for everything but a dir,
6141 * btrfs_set_inode_index_count has an explanation for the magic
6144 BTRFS_I(inode)->index_cnt = 2;
6145 BTRFS_I(inode)->dir_index = *index;
6146 BTRFS_I(inode)->root = btrfs_grab_root(root);
6147 BTRFS_I(inode)->generation = trans->transid;
6148 inode->i_generation = BTRFS_I(inode)->generation;
6151 * We could have gotten an inode number from somebody who was fsynced
6152 * and then removed in this same transaction, so let's just set full
6153 * sync since it will be a full sync anyway and this will blow away the
6154 * old info in the log.
6156 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6158 key[0].objectid = objectid;
6159 key[0].type = BTRFS_INODE_ITEM_KEY;
6162 sizes[0] = sizeof(struct btrfs_inode_item);
6166 * Start new inodes with an inode_ref. This is slightly more
6167 * efficient for small numbers of hard links since they will
6168 * be packed into one item. Extended refs will kick in if we
6169 * add more hard links than can fit in the ref item.
6171 key[1].objectid = objectid;
6172 key[1].type = BTRFS_INODE_REF_KEY;
6173 key[1].offset = ref_objectid;
6175 sizes[1] = name_len + sizeof(*ref);
6178 location = &BTRFS_I(inode)->location;
6179 location->objectid = objectid;
6180 location->offset = 0;
6181 location->type = BTRFS_INODE_ITEM_KEY;
6183 ret = btrfs_insert_inode_locked(inode);
6189 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6193 inode_init_owner(inode, dir, mode);
6194 inode_set_bytes(inode, 0);
6196 inode->i_mtime = current_time(inode);
6197 inode->i_atime = inode->i_mtime;
6198 inode->i_ctime = inode->i_mtime;
6199 BTRFS_I(inode)->i_otime = inode->i_mtime;
6201 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6202 struct btrfs_inode_item);
6203 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6204 sizeof(*inode_item));
6205 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6208 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6209 struct btrfs_inode_ref);
6210 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6211 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6212 ptr = (unsigned long)(ref + 1);
6213 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6216 btrfs_mark_buffer_dirty(path->nodes[0]);
6217 btrfs_free_path(path);
6219 btrfs_inherit_iflags(inode, dir);
6221 if (S_ISREG(mode)) {
6222 if (btrfs_test_opt(fs_info, NODATASUM))
6223 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6224 if (btrfs_test_opt(fs_info, NODATACOW))
6225 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6226 BTRFS_INODE_NODATASUM;
6229 inode_tree_add(inode);
6231 trace_btrfs_inode_new(inode);
6232 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6234 btrfs_update_root_times(trans, root);
6236 ret = btrfs_inode_inherit_props(trans, inode, dir);
6239 "error inheriting props for ino %llu (root %llu): %d",
6240 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6245 discard_new_inode(inode);
6248 BTRFS_I(dir)->index_cnt--;
6249 btrfs_free_path(path);
6250 return ERR_PTR(ret);
6254 * utility function to add 'inode' into 'parent_inode' with
6255 * a give name and a given sequence number.
6256 * if 'add_backref' is true, also insert a backref from the
6257 * inode to the parent directory.
6259 int btrfs_add_link(struct btrfs_trans_handle *trans,
6260 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6261 const char *name, int name_len, int add_backref, u64 index)
6264 struct btrfs_key key;
6265 struct btrfs_root *root = parent_inode->root;
6266 u64 ino = btrfs_ino(inode);
6267 u64 parent_ino = btrfs_ino(parent_inode);
6269 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6270 memcpy(&key, &inode->root->root_key, sizeof(key));
6273 key.type = BTRFS_INODE_ITEM_KEY;
6277 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6278 ret = btrfs_add_root_ref(trans, key.objectid,
6279 root->root_key.objectid, parent_ino,
6280 index, name, name_len);
6281 } else if (add_backref) {
6282 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6286 /* Nothing to clean up yet */
6290 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6291 btrfs_inode_type(&inode->vfs_inode), index);
6292 if (ret == -EEXIST || ret == -EOVERFLOW)
6295 btrfs_abort_transaction(trans, ret);
6299 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6301 inode_inc_iversion(&parent_inode->vfs_inode);
6303 * If we are replaying a log tree, we do not want to update the mtime
6304 * and ctime of the parent directory with the current time, since the
6305 * log replay procedure is responsible for setting them to their correct
6306 * values (the ones it had when the fsync was done).
6308 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6309 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6311 parent_inode->vfs_inode.i_mtime = now;
6312 parent_inode->vfs_inode.i_ctime = now;
6314 ret = btrfs_update_inode(trans, root, parent_inode);
6316 btrfs_abort_transaction(trans, ret);
6320 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6323 err = btrfs_del_root_ref(trans, key.objectid,
6324 root->root_key.objectid, parent_ino,
6325 &local_index, name, name_len);
6327 btrfs_abort_transaction(trans, err);
6328 } else if (add_backref) {
6332 err = btrfs_del_inode_ref(trans, root, name, name_len,
6333 ino, parent_ino, &local_index);
6335 btrfs_abort_transaction(trans, err);
6338 /* Return the original error code */
6342 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6343 struct btrfs_inode *dir, struct dentry *dentry,
6344 struct btrfs_inode *inode, int backref, u64 index)
6346 int err = btrfs_add_link(trans, dir, inode,
6347 dentry->d_name.name, dentry->d_name.len,
6354 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6355 umode_t mode, dev_t rdev)
6357 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6358 struct btrfs_trans_handle *trans;
6359 struct btrfs_root *root = BTRFS_I(dir)->root;
6360 struct inode *inode = NULL;
6366 * 2 for inode item and ref
6368 * 1 for xattr if selinux is on
6370 trans = btrfs_start_transaction(root, 5);
6372 return PTR_ERR(trans);
6374 err = btrfs_get_free_objectid(root, &objectid);
6378 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6379 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6381 if (IS_ERR(inode)) {
6382 err = PTR_ERR(inode);
6388 * If the active LSM wants to access the inode during
6389 * d_instantiate it needs these. Smack checks to see
6390 * if the filesystem supports xattrs by looking at the
6393 inode->i_op = &btrfs_special_inode_operations;
6394 init_special_inode(inode, inode->i_mode, rdev);
6396 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6400 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6405 btrfs_update_inode(trans, root, BTRFS_I(inode));
6406 d_instantiate_new(dentry, inode);
6409 btrfs_end_transaction(trans);
6410 btrfs_btree_balance_dirty(fs_info);
6412 inode_dec_link_count(inode);
6413 discard_new_inode(inode);
6418 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6419 umode_t mode, bool excl)
6421 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6422 struct btrfs_trans_handle *trans;
6423 struct btrfs_root *root = BTRFS_I(dir)->root;
6424 struct inode *inode = NULL;
6430 * 2 for inode item and ref
6432 * 1 for xattr if selinux is on
6434 trans = btrfs_start_transaction(root, 5);
6436 return PTR_ERR(trans);
6438 err = btrfs_get_free_objectid(root, &objectid);
6442 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6443 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6445 if (IS_ERR(inode)) {
6446 err = PTR_ERR(inode);
6451 * If the active LSM wants to access the inode during
6452 * d_instantiate it needs these. Smack checks to see
6453 * if the filesystem supports xattrs by looking at the
6456 inode->i_fop = &btrfs_file_operations;
6457 inode->i_op = &btrfs_file_inode_operations;
6458 inode->i_mapping->a_ops = &btrfs_aops;
6460 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6464 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6468 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6473 d_instantiate_new(dentry, inode);
6476 btrfs_end_transaction(trans);
6478 inode_dec_link_count(inode);
6479 discard_new_inode(inode);
6481 btrfs_btree_balance_dirty(fs_info);
6485 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6486 struct dentry *dentry)
6488 struct btrfs_trans_handle *trans = NULL;
6489 struct btrfs_root *root = BTRFS_I(dir)->root;
6490 struct inode *inode = d_inode(old_dentry);
6491 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6496 /* do not allow sys_link's with other subvols of the same device */
6497 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6500 if (inode->i_nlink >= BTRFS_LINK_MAX)
6503 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6508 * 2 items for inode and inode ref
6509 * 2 items for dir items
6510 * 1 item for parent inode
6511 * 1 item for orphan item deletion if O_TMPFILE
6513 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6514 if (IS_ERR(trans)) {
6515 err = PTR_ERR(trans);
6520 /* There are several dir indexes for this inode, clear the cache. */
6521 BTRFS_I(inode)->dir_index = 0ULL;
6523 inode_inc_iversion(inode);
6524 inode->i_ctime = current_time(inode);
6526 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6528 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6534 struct dentry *parent = dentry->d_parent;
6536 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6539 if (inode->i_nlink == 1) {
6541 * If new hard link count is 1, it's a file created
6542 * with open(2) O_TMPFILE flag.
6544 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6548 d_instantiate(dentry, inode);
6549 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6554 btrfs_end_transaction(trans);
6556 inode_dec_link_count(inode);
6559 btrfs_btree_balance_dirty(fs_info);
6563 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6565 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6566 struct inode *inode = NULL;
6567 struct btrfs_trans_handle *trans;
6568 struct btrfs_root *root = BTRFS_I(dir)->root;
6574 * 2 items for inode and ref
6575 * 2 items for dir items
6576 * 1 for xattr if selinux is on
6578 trans = btrfs_start_transaction(root, 5);
6580 return PTR_ERR(trans);
6582 err = btrfs_get_free_objectid(root, &objectid);
6586 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6587 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6588 S_IFDIR | mode, &index);
6589 if (IS_ERR(inode)) {
6590 err = PTR_ERR(inode);
6595 /* these must be set before we unlock the inode */
6596 inode->i_op = &btrfs_dir_inode_operations;
6597 inode->i_fop = &btrfs_dir_file_operations;
6599 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6603 btrfs_i_size_write(BTRFS_I(inode), 0);
6604 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6608 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6609 dentry->d_name.name,
6610 dentry->d_name.len, 0, index);
6614 d_instantiate_new(dentry, inode);
6617 btrfs_end_transaction(trans);
6619 inode_dec_link_count(inode);
6620 discard_new_inode(inode);
6622 btrfs_btree_balance_dirty(fs_info);
6626 static noinline int uncompress_inline(struct btrfs_path *path,
6628 size_t pg_offset, u64 extent_offset,
6629 struct btrfs_file_extent_item *item)
6632 struct extent_buffer *leaf = path->nodes[0];
6635 unsigned long inline_size;
6639 WARN_ON(pg_offset != 0);
6640 compress_type = btrfs_file_extent_compression(leaf, item);
6641 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6642 inline_size = btrfs_file_extent_inline_item_len(leaf,
6643 btrfs_item_nr(path->slots[0]));
6644 tmp = kmalloc(inline_size, GFP_NOFS);
6647 ptr = btrfs_file_extent_inline_start(item);
6649 read_extent_buffer(leaf, tmp, ptr, inline_size);
6651 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6652 ret = btrfs_decompress(compress_type, tmp, page,
6653 extent_offset, inline_size, max_size);
6656 * decompression code contains a memset to fill in any space between the end
6657 * of the uncompressed data and the end of max_size in case the decompressed
6658 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6659 * the end of an inline extent and the beginning of the next block, so we
6660 * cover that region here.
6663 if (max_size + pg_offset < PAGE_SIZE) {
6664 char *map = kmap(page);
6665 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6673 * btrfs_get_extent - Lookup the first extent overlapping a range in a file.
6674 * @inode: file to search in
6675 * @page: page to read extent data into if the extent is inline
6676 * @pg_offset: offset into @page to copy to
6677 * @start: file offset
6678 * @len: length of range starting at @start
6680 * This returns the first &struct extent_map which overlaps with the given
6681 * range, reading it from the B-tree and caching it if necessary. Note that
6682 * there may be more extents which overlap the given range after the returned
6685 * If @page is not NULL and the extent is inline, this also reads the extent
6686 * data directly into the page and marks the extent up to date in the io_tree.
6688 * Return: ERR_PTR on error, non-NULL extent_map on success.
6690 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6691 struct page *page, size_t pg_offset,
6694 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6696 u64 extent_start = 0;
6698 u64 objectid = btrfs_ino(inode);
6699 int extent_type = -1;
6700 struct btrfs_path *path = NULL;
6701 struct btrfs_root *root = inode->root;
6702 struct btrfs_file_extent_item *item;
6703 struct extent_buffer *leaf;
6704 struct btrfs_key found_key;
6705 struct extent_map *em = NULL;
6706 struct extent_map_tree *em_tree = &inode->extent_tree;
6707 struct extent_io_tree *io_tree = &inode->io_tree;
6709 read_lock(&em_tree->lock);
6710 em = lookup_extent_mapping(em_tree, start, len);
6711 read_unlock(&em_tree->lock);
6714 if (em->start > start || em->start + em->len <= start)
6715 free_extent_map(em);
6716 else if (em->block_start == EXTENT_MAP_INLINE && page)
6717 free_extent_map(em);
6721 em = alloc_extent_map();
6726 em->start = EXTENT_MAP_HOLE;
6727 em->orig_start = EXTENT_MAP_HOLE;
6729 em->block_len = (u64)-1;
6731 path = btrfs_alloc_path();
6737 /* Chances are we'll be called again, so go ahead and do readahead */
6738 path->reada = READA_FORWARD;
6741 * The same explanation in load_free_space_cache applies here as well,
6742 * we only read when we're loading the free space cache, and at that
6743 * point the commit_root has everything we need.
6745 if (btrfs_is_free_space_inode(inode)) {
6746 path->search_commit_root = 1;
6747 path->skip_locking = 1;
6750 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6753 } else if (ret > 0) {
6754 if (path->slots[0] == 0)
6760 leaf = path->nodes[0];
6761 item = btrfs_item_ptr(leaf, path->slots[0],
6762 struct btrfs_file_extent_item);
6763 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6764 if (found_key.objectid != objectid ||
6765 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6767 * If we backup past the first extent we want to move forward
6768 * and see if there is an extent in front of us, otherwise we'll
6769 * say there is a hole for our whole search range which can
6776 extent_type = btrfs_file_extent_type(leaf, item);
6777 extent_start = found_key.offset;
6778 extent_end = btrfs_file_extent_end(path);
6779 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6780 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6781 /* Only regular file could have regular/prealloc extent */
6782 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6785 "regular/prealloc extent found for non-regular inode %llu",
6789 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6791 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6792 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6797 if (start >= extent_end) {
6799 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6800 ret = btrfs_next_leaf(root, path);
6806 leaf = path->nodes[0];
6808 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6809 if (found_key.objectid != objectid ||
6810 found_key.type != BTRFS_EXTENT_DATA_KEY)
6812 if (start + len <= found_key.offset)
6814 if (start > found_key.offset)
6817 /* New extent overlaps with existing one */
6819 em->orig_start = start;
6820 em->len = found_key.offset - start;
6821 em->block_start = EXTENT_MAP_HOLE;
6825 btrfs_extent_item_to_extent_map(inode, path, item, !page, em);
6827 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6828 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6830 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6834 size_t extent_offset;
6840 size = btrfs_file_extent_ram_bytes(leaf, item);
6841 extent_offset = page_offset(page) + pg_offset - extent_start;
6842 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6843 size - extent_offset);
6844 em->start = extent_start + extent_offset;
6845 em->len = ALIGN(copy_size, fs_info->sectorsize);
6846 em->orig_block_len = em->len;
6847 em->orig_start = em->start;
6848 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6850 if (!PageUptodate(page)) {
6851 if (btrfs_file_extent_compression(leaf, item) !=
6852 BTRFS_COMPRESS_NONE) {
6853 ret = uncompress_inline(path, page, pg_offset,
6854 extent_offset, item);
6859 read_extent_buffer(leaf, map + pg_offset, ptr,
6861 if (pg_offset + copy_size < PAGE_SIZE) {
6862 memset(map + pg_offset + copy_size, 0,
6863 PAGE_SIZE - pg_offset -
6868 flush_dcache_page(page);
6870 set_extent_uptodate(io_tree, em->start,
6871 extent_map_end(em) - 1, NULL, GFP_NOFS);
6876 em->orig_start = start;
6878 em->block_start = EXTENT_MAP_HOLE;
6881 btrfs_release_path(path);
6882 if (em->start > start || extent_map_end(em) <= start) {
6884 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6885 em->start, em->len, start, len);
6890 write_lock(&em_tree->lock);
6891 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6892 write_unlock(&em_tree->lock);
6894 btrfs_free_path(path);
6896 trace_btrfs_get_extent(root, inode, em);
6899 free_extent_map(em);
6900 return ERR_PTR(ret);
6905 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6908 struct extent_map *em;
6909 struct extent_map *hole_em = NULL;
6910 u64 delalloc_start = start;
6916 em = btrfs_get_extent(inode, NULL, 0, start, len);
6920 * If our em maps to:
6922 * - a pre-alloc extent,
6923 * there might actually be delalloc bytes behind it.
6925 if (em->block_start != EXTENT_MAP_HOLE &&
6926 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6931 /* check to see if we've wrapped (len == -1 or similar) */
6940 /* ok, we didn't find anything, lets look for delalloc */
6941 delalloc_len = count_range_bits(&inode->io_tree, &delalloc_start,
6942 end, len, EXTENT_DELALLOC, 1);
6943 delalloc_end = delalloc_start + delalloc_len;
6944 if (delalloc_end < delalloc_start)
6945 delalloc_end = (u64)-1;
6948 * We didn't find anything useful, return the original results from
6951 if (delalloc_start > end || delalloc_end <= start) {
6958 * Adjust the delalloc_start to make sure it doesn't go backwards from
6959 * the start they passed in
6961 delalloc_start = max(start, delalloc_start);
6962 delalloc_len = delalloc_end - delalloc_start;
6964 if (delalloc_len > 0) {
6967 const u64 hole_end = extent_map_end(hole_em);
6969 em = alloc_extent_map();
6977 * When btrfs_get_extent can't find anything it returns one
6980 * Make sure what it found really fits our range, and adjust to
6981 * make sure it is based on the start from the caller
6983 if (hole_end <= start || hole_em->start > end) {
6984 free_extent_map(hole_em);
6987 hole_start = max(hole_em->start, start);
6988 hole_len = hole_end - hole_start;
6991 if (hole_em && delalloc_start > hole_start) {
6993 * Our hole starts before our delalloc, so we have to
6994 * return just the parts of the hole that go until the
6997 em->len = min(hole_len, delalloc_start - hole_start);
6998 em->start = hole_start;
6999 em->orig_start = hole_start;
7001 * Don't adjust block start at all, it is fixed at
7004 em->block_start = hole_em->block_start;
7005 em->block_len = hole_len;
7006 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7007 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7010 * Hole is out of passed range or it starts after
7013 em->start = delalloc_start;
7014 em->len = delalloc_len;
7015 em->orig_start = delalloc_start;
7016 em->block_start = EXTENT_MAP_DELALLOC;
7017 em->block_len = delalloc_len;
7024 free_extent_map(hole_em);
7026 free_extent_map(em);
7027 return ERR_PTR(err);
7032 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7035 const u64 orig_start,
7036 const u64 block_start,
7037 const u64 block_len,
7038 const u64 orig_block_len,
7039 const u64 ram_bytes,
7042 struct extent_map *em = NULL;
7045 if (type != BTRFS_ORDERED_NOCOW) {
7046 em = create_io_em(inode, start, len, orig_start, block_start,
7047 block_len, orig_block_len, ram_bytes,
7048 BTRFS_COMPRESS_NONE, /* compress_type */
7053 ret = btrfs_add_ordered_extent_dio(inode, start, block_start, len,
7057 free_extent_map(em);
7058 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
7067 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7070 struct btrfs_root *root = inode->root;
7071 struct btrfs_fs_info *fs_info = root->fs_info;
7072 struct extent_map *em;
7073 struct btrfs_key ins;
7077 alloc_hint = get_extent_allocation_hint(inode, start, len);
7078 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7079 0, alloc_hint, &ins, 1, 1);
7081 return ERR_PTR(ret);
7083 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7084 ins.objectid, ins.offset, ins.offset,
7085 ins.offset, BTRFS_ORDERED_REGULAR);
7086 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7088 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7095 * Check if we can do nocow write into the range [@offset, @offset + @len)
7097 * @offset: File offset
7098 * @len: The length to write, will be updated to the nocow writeable
7100 * @orig_start: (optional) Return the original file offset of the file extent
7101 * @orig_len: (optional) Return the original on-disk length of the file extent
7102 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7103 * @strict: if true, omit optimizations that might force us into unnecessary
7104 * cow. e.g., don't trust generation number.
7106 * This function will flush ordered extents in the range to ensure proper
7107 * nocow checks for (nowait == false) case.
7110 * >0 and update @len if we can do nocow write
7111 * 0 if we can't do nocow write
7112 * <0 if error happened
7114 * NOTE: This only checks the file extents, caller is responsible to wait for
7115 * any ordered extents.
7117 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7118 u64 *orig_start, u64 *orig_block_len,
7119 u64 *ram_bytes, bool strict)
7121 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7122 struct btrfs_path *path;
7124 struct extent_buffer *leaf;
7125 struct btrfs_root *root = BTRFS_I(inode)->root;
7126 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7127 struct btrfs_file_extent_item *fi;
7128 struct btrfs_key key;
7135 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7137 path = btrfs_alloc_path();
7141 ret = btrfs_lookup_file_extent(NULL, root, path,
7142 btrfs_ino(BTRFS_I(inode)), offset, 0);
7146 slot = path->slots[0];
7149 /* can't find the item, must cow */
7156 leaf = path->nodes[0];
7157 btrfs_item_key_to_cpu(leaf, &key, slot);
7158 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7159 key.type != BTRFS_EXTENT_DATA_KEY) {
7160 /* not our file or wrong item type, must cow */
7164 if (key.offset > offset) {
7165 /* Wrong offset, must cow */
7169 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7170 found_type = btrfs_file_extent_type(leaf, fi);
7171 if (found_type != BTRFS_FILE_EXTENT_REG &&
7172 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7173 /* not a regular extent, must cow */
7177 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7180 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7181 if (extent_end <= offset)
7184 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7185 if (disk_bytenr == 0)
7188 if (btrfs_file_extent_compression(leaf, fi) ||
7189 btrfs_file_extent_encryption(leaf, fi) ||
7190 btrfs_file_extent_other_encoding(leaf, fi))
7194 * Do the same check as in btrfs_cross_ref_exist but without the
7195 * unnecessary search.
7198 (btrfs_file_extent_generation(leaf, fi) <=
7199 btrfs_root_last_snapshot(&root->root_item)))
7202 backref_offset = btrfs_file_extent_offset(leaf, fi);
7205 *orig_start = key.offset - backref_offset;
7206 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7207 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7210 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7213 num_bytes = min(offset + *len, extent_end) - offset;
7214 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7217 range_end = round_up(offset + num_bytes,
7218 root->fs_info->sectorsize) - 1;
7219 ret = test_range_bit(io_tree, offset, range_end,
7220 EXTENT_DELALLOC, 0, NULL);
7227 btrfs_release_path(path);
7230 * look for other files referencing this extent, if we
7231 * find any we must cow
7234 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7235 key.offset - backref_offset, disk_bytenr,
7243 * adjust disk_bytenr and num_bytes to cover just the bytes
7244 * in this extent we are about to write. If there
7245 * are any csums in that range we have to cow in order
7246 * to keep the csums correct
7248 disk_bytenr += backref_offset;
7249 disk_bytenr += offset - key.offset;
7250 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7253 * all of the above have passed, it is safe to overwrite this extent
7259 btrfs_free_path(path);
7263 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7264 struct extent_state **cached_state, bool writing)
7266 struct btrfs_ordered_extent *ordered;
7270 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7273 * We're concerned with the entire range that we're going to be
7274 * doing DIO to, so we need to make sure there's no ordered
7275 * extents in this range.
7277 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7278 lockend - lockstart + 1);
7281 * We need to make sure there are no buffered pages in this
7282 * range either, we could have raced between the invalidate in
7283 * generic_file_direct_write and locking the extent. The
7284 * invalidate needs to happen so that reads after a write do not
7288 (!writing || !filemap_range_has_page(inode->i_mapping,
7289 lockstart, lockend)))
7292 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7297 * If we are doing a DIO read and the ordered extent we
7298 * found is for a buffered write, we can not wait for it
7299 * to complete and retry, because if we do so we can
7300 * deadlock with concurrent buffered writes on page
7301 * locks. This happens only if our DIO read covers more
7302 * than one extent map, if at this point has already
7303 * created an ordered extent for a previous extent map
7304 * and locked its range in the inode's io tree, and a
7305 * concurrent write against that previous extent map's
7306 * range and this range started (we unlock the ranges
7307 * in the io tree only when the bios complete and
7308 * buffered writes always lock pages before attempting
7309 * to lock range in the io tree).
7312 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7313 btrfs_start_ordered_extent(ordered, 1);
7316 btrfs_put_ordered_extent(ordered);
7319 * We could trigger writeback for this range (and wait
7320 * for it to complete) and then invalidate the pages for
7321 * this range (through invalidate_inode_pages2_range()),
7322 * but that can lead us to a deadlock with a concurrent
7323 * call to readahead (a buffered read or a defrag call
7324 * triggered a readahead) on a page lock due to an
7325 * ordered dio extent we created before but did not have
7326 * yet a corresponding bio submitted (whence it can not
7327 * complete), which makes readahead wait for that
7328 * ordered extent to complete while holding a lock on
7343 /* The callers of this must take lock_extent() */
7344 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7345 u64 len, u64 orig_start, u64 block_start,
7346 u64 block_len, u64 orig_block_len,
7347 u64 ram_bytes, int compress_type,
7350 struct extent_map_tree *em_tree;
7351 struct extent_map *em;
7354 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7355 type == BTRFS_ORDERED_COMPRESSED ||
7356 type == BTRFS_ORDERED_NOCOW ||
7357 type == BTRFS_ORDERED_REGULAR);
7359 em_tree = &inode->extent_tree;
7360 em = alloc_extent_map();
7362 return ERR_PTR(-ENOMEM);
7365 em->orig_start = orig_start;
7367 em->block_len = block_len;
7368 em->block_start = block_start;
7369 em->orig_block_len = orig_block_len;
7370 em->ram_bytes = ram_bytes;
7371 em->generation = -1;
7372 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7373 if (type == BTRFS_ORDERED_PREALLOC) {
7374 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7375 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7376 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7377 em->compress_type = compress_type;
7381 btrfs_drop_extent_cache(inode, em->start,
7382 em->start + em->len - 1, 0);
7383 write_lock(&em_tree->lock);
7384 ret = add_extent_mapping(em_tree, em, 1);
7385 write_unlock(&em_tree->lock);
7387 * The caller has taken lock_extent(), who could race with us
7390 } while (ret == -EEXIST);
7393 free_extent_map(em);
7394 return ERR_PTR(ret);
7397 /* em got 2 refs now, callers needs to do free_extent_map once. */
7402 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7403 struct inode *inode,
7404 struct btrfs_dio_data *dio_data,
7407 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7408 struct extent_map *em = *map;
7412 * We don't allocate a new extent in the following cases
7414 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7416 * 2) The extent is marked as PREALLOC. We're good to go here and can
7417 * just use the extent.
7420 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7421 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7422 em->block_start != EXTENT_MAP_HOLE)) {
7424 u64 block_start, orig_start, orig_block_len, ram_bytes;
7426 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7427 type = BTRFS_ORDERED_PREALLOC;
7429 type = BTRFS_ORDERED_NOCOW;
7430 len = min(len, em->len - (start - em->start));
7431 block_start = em->block_start + (start - em->start);
7433 if (can_nocow_extent(inode, start, &len, &orig_start,
7434 &orig_block_len, &ram_bytes, false) == 1 &&
7435 btrfs_inc_nocow_writers(fs_info, block_start)) {
7436 struct extent_map *em2;
7438 em2 = btrfs_create_dio_extent(BTRFS_I(inode), start, len,
7439 orig_start, block_start,
7440 len, orig_block_len,
7442 btrfs_dec_nocow_writers(fs_info, block_start);
7443 if (type == BTRFS_ORDERED_PREALLOC) {
7444 free_extent_map(em);
7448 if (em2 && IS_ERR(em2)) {
7453 * For inode marked NODATACOW or extent marked PREALLOC,
7454 * use the existing or preallocated extent, so does not
7455 * need to adjust btrfs_space_info's bytes_may_use.
7457 btrfs_free_reserved_data_space_noquota(fs_info, len);
7462 /* this will cow the extent */
7463 free_extent_map(em);
7464 *map = em = btrfs_new_extent_direct(BTRFS_I(inode), start, len);
7470 len = min(len, em->len - (start - em->start));
7474 * Need to update the i_size under the extent lock so buffered
7475 * readers will get the updated i_size when we unlock.
7477 if (start + len > i_size_read(inode))
7478 i_size_write(inode, start + len);
7480 dio_data->reserve -= len;
7485 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7486 loff_t length, unsigned int flags, struct iomap *iomap,
7487 struct iomap *srcmap)
7489 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7490 struct extent_map *em;
7491 struct extent_state *cached_state = NULL;
7492 struct btrfs_dio_data *dio_data = NULL;
7493 u64 lockstart, lockend;
7494 const bool write = !!(flags & IOMAP_WRITE);
7497 bool unlock_extents = false;
7500 len = min_t(u64, len, fs_info->sectorsize);
7503 lockend = start + len - 1;
7506 * The generic stuff only does filemap_write_and_wait_range, which
7507 * isn't enough if we've written compressed pages to this area, so we
7508 * need to flush the dirty pages again to make absolutely sure that any
7509 * outstanding dirty pages are on disk.
7511 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7512 &BTRFS_I(inode)->runtime_flags)) {
7513 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7514 start + length - 1);
7519 dio_data = kzalloc(sizeof(*dio_data), GFP_NOFS);
7523 dio_data->length = length;
7525 dio_data->reserve = round_up(length, fs_info->sectorsize);
7526 ret = btrfs_delalloc_reserve_space(BTRFS_I(inode),
7527 &dio_data->data_reserved,
7528 start, dio_data->reserve);
7530 extent_changeset_free(dio_data->data_reserved);
7535 iomap->private = dio_data;
7539 * If this errors out it's because we couldn't invalidate pagecache for
7540 * this range and we need to fallback to buffered.
7542 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, write)) {
7547 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7554 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7555 * io. INLINE is special, and we could probably kludge it in here, but
7556 * it's still buffered so for safety lets just fall back to the generic
7559 * For COMPRESSED we _have_ to read the entire extent in so we can
7560 * decompress it, so there will be buffering required no matter what we
7561 * do, so go ahead and fallback to buffered.
7563 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7564 * to buffered IO. Don't blame me, this is the price we pay for using
7567 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7568 em->block_start == EXTENT_MAP_INLINE) {
7569 free_extent_map(em);
7574 len = min(len, em->len - (start - em->start));
7576 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7580 unlock_extents = true;
7581 /* Recalc len in case the new em is smaller than requested */
7582 len = min(len, em->len - (start - em->start));
7585 * We need to unlock only the end area that we aren't using.
7586 * The rest is going to be unlocked by the endio routine.
7588 lockstart = start + len;
7589 if (lockstart < lockend)
7590 unlock_extents = true;
7594 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7595 lockstart, lockend, &cached_state);
7597 free_extent_state(cached_state);
7600 * Translate extent map information to iomap.
7601 * We trim the extents (and move the addr) even though iomap code does
7602 * that, since we have locked only the parts we are performing I/O in.
7604 if ((em->block_start == EXTENT_MAP_HOLE) ||
7605 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7606 iomap->addr = IOMAP_NULL_ADDR;
7607 iomap->type = IOMAP_HOLE;
7609 iomap->addr = em->block_start + (start - em->start);
7610 iomap->type = IOMAP_MAPPED;
7612 iomap->offset = start;
7613 iomap->bdev = fs_info->fs_devices->latest_bdev;
7614 iomap->length = len;
7616 free_extent_map(em);
7621 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7625 btrfs_delalloc_release_space(BTRFS_I(inode),
7626 dio_data->data_reserved, start,
7627 dio_data->reserve, true);
7628 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->reserve);
7629 extent_changeset_free(dio_data->data_reserved);
7635 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7636 ssize_t written, unsigned int flags, struct iomap *iomap)
7639 struct btrfs_dio_data *dio_data = iomap->private;
7640 size_t submitted = dio_data->submitted;
7641 const bool write = !!(flags & IOMAP_WRITE);
7643 if (!write && (iomap->type == IOMAP_HOLE)) {
7644 /* If reading from a hole, unlock and return */
7645 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1);
7649 if (submitted < length) {
7651 length -= submitted;
7653 __endio_write_update_ordered(BTRFS_I(inode), pos,
7656 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7662 if (dio_data->reserve)
7663 btrfs_delalloc_release_space(BTRFS_I(inode),
7664 dio_data->data_reserved, pos,
7665 dio_data->reserve, true);
7666 btrfs_delalloc_release_extents(BTRFS_I(inode), dio_data->length);
7667 extent_changeset_free(dio_data->data_reserved);
7671 iomap->private = NULL;
7676 static void btrfs_dio_private_put(struct btrfs_dio_private *dip)
7679 * This implies a barrier so that stores to dio_bio->bi_status before
7680 * this and loads of dio_bio->bi_status after this are fully ordered.
7682 if (!refcount_dec_and_test(&dip->refs))
7685 if (bio_op(dip->dio_bio) == REQ_OP_WRITE) {
7686 __endio_write_update_ordered(BTRFS_I(dip->inode),
7687 dip->logical_offset,
7689 !dip->dio_bio->bi_status);
7691 unlock_extent(&BTRFS_I(dip->inode)->io_tree,
7692 dip->logical_offset,
7693 dip->logical_offset + dip->bytes - 1);
7696 bio_endio(dip->dio_bio);
7700 static blk_status_t submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7702 unsigned long bio_flags)
7704 struct btrfs_dio_private *dip = bio->bi_private;
7705 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7708 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7710 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7714 refcount_inc(&dip->refs);
7715 ret = btrfs_map_bio(fs_info, bio, mirror_num);
7717 refcount_dec(&dip->refs);
7721 static blk_status_t btrfs_check_read_dio_bio(struct inode *inode,
7722 struct btrfs_io_bio *io_bio,
7723 const bool uptodate)
7725 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
7726 const u32 sectorsize = fs_info->sectorsize;
7727 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7728 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7729 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7730 struct bio_vec bvec;
7731 struct bvec_iter iter;
7732 u64 start = io_bio->logical;
7734 blk_status_t err = BLK_STS_OK;
7736 __bio_for_each_segment(bvec, &io_bio->bio, iter, io_bio->iter) {
7737 unsigned int i, nr_sectors, pgoff;
7739 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7740 pgoff = bvec.bv_offset;
7741 for (i = 0; i < nr_sectors; i++) {
7742 ASSERT(pgoff < PAGE_SIZE);
7744 (!csum || !check_data_csum(inode, io_bio,
7745 bio_offset, bvec.bv_page, pgoff))) {
7746 clean_io_failure(fs_info, failure_tree, io_tree,
7747 start, bvec.bv_page,
7748 btrfs_ino(BTRFS_I(inode)),
7751 blk_status_t status;
7753 ASSERT((start - io_bio->logical) < UINT_MAX);
7754 status = btrfs_submit_read_repair(inode,
7756 start - io_bio->logical,
7757 bvec.bv_page, pgoff,
7759 start + sectorsize - 1,
7761 submit_dio_repair_bio);
7765 start += sectorsize;
7766 ASSERT(bio_offset + sectorsize > bio_offset);
7767 bio_offset += sectorsize;
7768 pgoff += sectorsize;
7774 static void __endio_write_update_ordered(struct btrfs_inode *inode,
7775 const u64 offset, const u64 bytes,
7776 const bool uptodate)
7778 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7779 struct btrfs_ordered_extent *ordered = NULL;
7780 struct btrfs_workqueue *wq;
7781 u64 ordered_offset = offset;
7782 u64 ordered_bytes = bytes;
7785 if (btrfs_is_free_space_inode(inode))
7786 wq = fs_info->endio_freespace_worker;
7788 wq = fs_info->endio_write_workers;
7790 while (ordered_offset < offset + bytes) {
7791 last_offset = ordered_offset;
7792 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
7796 btrfs_init_work(&ordered->work, finish_ordered_fn, NULL,
7798 btrfs_queue_work(wq, &ordered->work);
7801 /* No ordered extent found in the range, exit */
7802 if (ordered_offset == last_offset)
7805 * Our bio might span multiple ordered extents. In this case
7806 * we keep going until we have accounted the whole dio.
7808 if (ordered_offset < offset + bytes) {
7809 ordered_bytes = offset + bytes - ordered_offset;
7815 static blk_status_t btrfs_submit_bio_start_direct_io(struct inode *inode,
7817 u64 dio_file_offset)
7819 return btrfs_csum_one_bio(BTRFS_I(inode), bio, dio_file_offset, 1);
7822 static void btrfs_end_dio_bio(struct bio *bio)
7824 struct btrfs_dio_private *dip = bio->bi_private;
7825 blk_status_t err = bio->bi_status;
7828 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7829 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
7830 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
7831 bio->bi_opf, bio->bi_iter.bi_sector,
7832 bio->bi_iter.bi_size, err);
7834 if (bio_op(bio) == REQ_OP_READ) {
7835 err = btrfs_check_read_dio_bio(dip->inode, btrfs_io_bio(bio),
7840 dip->dio_bio->bi_status = err;
7843 btrfs_dio_private_put(dip);
7846 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
7847 struct inode *inode, u64 file_offset, int async_submit)
7849 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7850 struct btrfs_dio_private *dip = bio->bi_private;
7851 bool write = bio_op(bio) == REQ_OP_WRITE;
7854 /* Check btrfs_submit_bio_hook() for rules about async submit. */
7856 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7859 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
7864 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
7867 if (write && async_submit) {
7868 ret = btrfs_wq_submit_bio(inode, bio, 0, 0, file_offset,
7869 btrfs_submit_bio_start_direct_io);
7873 * If we aren't doing async submit, calculate the csum of the
7876 ret = btrfs_csum_one_bio(BTRFS_I(inode), bio, file_offset, 1);
7882 csum_offset = file_offset - dip->logical_offset;
7883 csum_offset >>= fs_info->sectorsize_bits;
7884 csum_offset *= fs_info->csum_size;
7885 btrfs_io_bio(bio)->csum = dip->csums + csum_offset;
7888 ret = btrfs_map_bio(fs_info, bio, 0);
7894 * If this succeeds, the btrfs_dio_private is responsible for cleaning up locked
7895 * or ordered extents whether or not we submit any bios.
7897 static struct btrfs_dio_private *btrfs_create_dio_private(struct bio *dio_bio,
7898 struct inode *inode,
7901 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7902 const bool csum = !(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM);
7904 struct btrfs_dio_private *dip;
7906 dip_size = sizeof(*dip);
7907 if (!write && csum) {
7908 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7911 nblocks = dio_bio->bi_iter.bi_size >> fs_info->sectorsize_bits;
7912 dip_size += fs_info->csum_size * nblocks;
7915 dip = kzalloc(dip_size, GFP_NOFS);
7920 dip->logical_offset = file_offset;
7921 dip->bytes = dio_bio->bi_iter.bi_size;
7922 dip->disk_bytenr = dio_bio->bi_iter.bi_sector << 9;
7923 dip->dio_bio = dio_bio;
7924 refcount_set(&dip->refs, 1);
7928 static blk_qc_t btrfs_submit_direct(struct inode *inode, struct iomap *iomap,
7929 struct bio *dio_bio, loff_t file_offset)
7931 const bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
7932 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7933 const bool raid56 = (btrfs_data_alloc_profile(fs_info) &
7934 BTRFS_BLOCK_GROUP_RAID56_MASK);
7935 struct btrfs_dio_private *dip;
7938 int async_submit = 0;
7940 int clone_offset = 0;
7943 blk_status_t status;
7944 struct btrfs_io_geometry geom;
7945 struct btrfs_dio_data *dio_data = iomap->private;
7947 dip = btrfs_create_dio_private(dio_bio, inode, file_offset);
7950 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
7951 file_offset + dio_bio->bi_iter.bi_size - 1);
7953 dio_bio->bi_status = BLK_STS_RESOURCE;
7955 return BLK_QC_T_NONE;
7960 * Load the csums up front to reduce csum tree searches and
7961 * contention when submitting bios.
7963 * If we have csums disabled this will do nothing.
7965 status = btrfs_lookup_bio_sums(inode, dio_bio, dip->csums);
7966 if (status != BLK_STS_OK)
7970 start_sector = dio_bio->bi_iter.bi_sector;
7971 submit_len = dio_bio->bi_iter.bi_size;
7974 ret = btrfs_get_io_geometry(fs_info, btrfs_op(dio_bio),
7975 start_sector << 9, submit_len,
7978 status = errno_to_blk_status(ret);
7981 ASSERT(geom.len <= INT_MAX);
7983 clone_len = min_t(int, submit_len, geom.len);
7986 * This will never fail as it's passing GPF_NOFS and
7987 * the allocation is backed by btrfs_bioset.
7989 bio = btrfs_bio_clone_partial(dio_bio, clone_offset, clone_len);
7990 bio->bi_private = dip;
7991 bio->bi_end_io = btrfs_end_dio_bio;
7992 btrfs_io_bio(bio)->logical = file_offset;
7994 ASSERT(submit_len >= clone_len);
7995 submit_len -= clone_len;
7998 * Increase the count before we submit the bio so we know
7999 * the end IO handler won't happen before we increase the
8000 * count. Otherwise, the dip might get freed before we're
8001 * done setting it up.
8003 * We transfer the initial reference to the last bio, so we
8004 * don't need to increment the reference count for the last one.
8006 if (submit_len > 0) {
8007 refcount_inc(&dip->refs);
8009 * If we are submitting more than one bio, submit them
8010 * all asynchronously. The exception is RAID 5 or 6, as
8011 * asynchronous checksums make it difficult to collect
8012 * full stripe writes.
8018 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8023 refcount_dec(&dip->refs);
8027 dio_data->submitted += clone_len;
8028 clone_offset += clone_len;
8029 start_sector += clone_len >> 9;
8030 file_offset += clone_len;
8031 } while (submit_len > 0);
8032 return BLK_QC_T_NONE;
8035 dip->dio_bio->bi_status = status;
8036 btrfs_dio_private_put(dip);
8037 return BLK_QC_T_NONE;
8040 const struct iomap_ops btrfs_dio_iomap_ops = {
8041 .iomap_begin = btrfs_dio_iomap_begin,
8042 .iomap_end = btrfs_dio_iomap_end,
8045 const struct iomap_dio_ops btrfs_dio_ops = {
8046 .submit_io = btrfs_submit_direct,
8049 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8054 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
8058 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
8061 int btrfs_readpage(struct file *file, struct page *page)
8063 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8064 u64 start = page_offset(page);
8065 u64 end = start + PAGE_SIZE - 1;
8066 unsigned long bio_flags = 0;
8067 struct bio *bio = NULL;
8070 btrfs_lock_and_flush_ordered_range(inode, start, end, NULL);
8072 ret = btrfs_do_readpage(page, NULL, &bio, &bio_flags, 0, NULL);
8074 ret = submit_one_bio(bio, 0, bio_flags);
8078 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8080 struct inode *inode = page->mapping->host;
8083 if (current->flags & PF_MEMALLOC) {
8084 redirty_page_for_writepage(wbc, page);
8090 * If we are under memory pressure we will call this directly from the
8091 * VM, we need to make sure we have the inode referenced for the ordered
8092 * extent. If not just return like we didn't do anything.
8094 if (!igrab(inode)) {
8095 redirty_page_for_writepage(wbc, page);
8096 return AOP_WRITEPAGE_ACTIVATE;
8098 ret = extent_write_full_page(page, wbc);
8099 btrfs_add_delayed_iput(inode);
8103 static int btrfs_writepages(struct address_space *mapping,
8104 struct writeback_control *wbc)
8106 return extent_writepages(mapping, wbc);
8109 static void btrfs_readahead(struct readahead_control *rac)
8111 extent_readahead(rac);
8114 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8116 int ret = try_release_extent_mapping(page, gfp_flags);
8118 detach_page_private(page);
8122 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8124 if (PageWriteback(page) || PageDirty(page))
8126 return __btrfs_releasepage(page, gfp_flags);
8129 #ifdef CONFIG_MIGRATION
8130 static int btrfs_migratepage(struct address_space *mapping,
8131 struct page *newpage, struct page *page,
8132 enum migrate_mode mode)
8136 ret = migrate_page_move_mapping(mapping, newpage, page, 0);
8137 if (ret != MIGRATEPAGE_SUCCESS)
8140 if (page_has_private(page))
8141 attach_page_private(newpage, detach_page_private(page));
8143 if (PagePrivate2(page)) {
8144 ClearPagePrivate2(page);
8145 SetPagePrivate2(newpage);
8148 if (mode != MIGRATE_SYNC_NO_COPY)
8149 migrate_page_copy(newpage, page);
8151 migrate_page_states(newpage, page);
8152 return MIGRATEPAGE_SUCCESS;
8156 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8157 unsigned int length)
8159 struct btrfs_inode *inode = BTRFS_I(page->mapping->host);
8160 struct extent_io_tree *tree = &inode->io_tree;
8161 struct btrfs_ordered_extent *ordered;
8162 struct extent_state *cached_state = NULL;
8163 u64 page_start = page_offset(page);
8164 u64 page_end = page_start + PAGE_SIZE - 1;
8167 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8168 bool found_ordered = false;
8169 bool completed_ordered = false;
8172 * we have the page locked, so new writeback can't start,
8173 * and the dirty bit won't be cleared while we are here.
8175 * Wait for IO on this page so that we can safely clear
8176 * the PagePrivate2 bit and do ordered accounting
8178 wait_on_page_writeback(page);
8181 btrfs_releasepage(page, GFP_NOFS);
8185 if (!inode_evicting)
8186 lock_extent_bits(tree, page_start, page_end, &cached_state);
8189 ordered = btrfs_lookup_ordered_range(inode, start, page_end - start + 1);
8191 found_ordered = true;
8193 ordered->file_offset + ordered->num_bytes - 1);
8195 * IO on this page will never be started, so we need to account
8196 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8197 * here, must leave that up for the ordered extent completion.
8199 if (!inode_evicting)
8200 clear_extent_bit(tree, start, end,
8202 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8203 EXTENT_DEFRAG, 1, 0, &cached_state);
8205 * whoever cleared the private bit is responsible
8206 * for the finish_ordered_io
8208 if (TestClearPagePrivate2(page)) {
8209 struct btrfs_ordered_inode_tree *tree;
8212 tree = &inode->ordered_tree;
8214 spin_lock_irq(&tree->lock);
8215 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8216 new_len = start - ordered->file_offset;
8217 if (new_len < ordered->truncated_len)
8218 ordered->truncated_len = new_len;
8219 spin_unlock_irq(&tree->lock);
8221 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8223 end - start + 1, 1)) {
8224 btrfs_finish_ordered_io(ordered);
8225 completed_ordered = true;
8228 btrfs_put_ordered_extent(ordered);
8229 if (!inode_evicting) {
8230 cached_state = NULL;
8231 lock_extent_bits(tree, start, end,
8236 if (start < page_end)
8241 * Qgroup reserved space handler
8242 * Page here will be either
8243 * 1) Already written to disk or ordered extent already submitted
8244 * Then its QGROUP_RESERVED bit in io_tree is already cleaned.
8245 * Qgroup will be handled by its qgroup_record then.
8246 * btrfs_qgroup_free_data() call will do nothing here.
8248 * 2) Not written to disk yet
8249 * Then btrfs_qgroup_free_data() call will clear the QGROUP_RESERVED
8250 * bit of its io_tree, and free the qgroup reserved data space.
8251 * Since the IO will never happen for this page.
8253 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8254 if (!inode_evicting) {
8258 * If there's an ordered extent for this range and we have not
8259 * finished it ourselves, we must leave EXTENT_DELALLOC_NEW set
8260 * in the range for the ordered extent completion. We must also
8261 * not delete the range, otherwise we would lose that bit (and
8262 * any other bits set in the range). Make sure EXTENT_UPTODATE
8263 * is cleared if we don't delete, otherwise it can lead to
8264 * corruptions if the i_size is extented later.
8266 if (found_ordered && !completed_ordered)
8268 clear_extent_bit(tree, page_start, page_end, EXTENT_LOCKED |
8269 EXTENT_DELALLOC | EXTENT_UPTODATE |
8270 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1,
8271 delete, &cached_state);
8273 __btrfs_releasepage(page, GFP_NOFS);
8276 ClearPageChecked(page);
8277 detach_page_private(page);
8281 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8282 * called from a page fault handler when a page is first dirtied. Hence we must
8283 * be careful to check for EOF conditions here. We set the page up correctly
8284 * for a written page which means we get ENOSPC checking when writing into
8285 * holes and correct delalloc and unwritten extent mapping on filesystems that
8286 * support these features.
8288 * We are not allowed to take the i_mutex here so we have to play games to
8289 * protect against truncate races as the page could now be beyond EOF. Because
8290 * truncate_setsize() writes the inode size before removing pages, once we have
8291 * the page lock we can determine safely if the page is beyond EOF. If it is not
8292 * beyond EOF, then the page is guaranteed safe against truncation until we
8295 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8297 struct page *page = vmf->page;
8298 struct inode *inode = file_inode(vmf->vma->vm_file);
8299 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8300 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8301 struct btrfs_ordered_extent *ordered;
8302 struct extent_state *cached_state = NULL;
8303 struct extent_changeset *data_reserved = NULL;
8305 unsigned long zero_start;
8315 reserved_space = PAGE_SIZE;
8317 sb_start_pagefault(inode->i_sb);
8318 page_start = page_offset(page);
8319 page_end = page_start + PAGE_SIZE - 1;
8323 * Reserving delalloc space after obtaining the page lock can lead to
8324 * deadlock. For example, if a dirty page is locked by this function
8325 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8326 * dirty page write out, then the btrfs_writepage() function could
8327 * end up waiting indefinitely to get a lock on the page currently
8328 * being processed by btrfs_page_mkwrite() function.
8330 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8331 page_start, reserved_space);
8333 ret2 = file_update_time(vmf->vma->vm_file);
8337 ret = vmf_error(ret2);
8343 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8346 size = i_size_read(inode);
8348 if ((page->mapping != inode->i_mapping) ||
8349 (page_start >= size)) {
8350 /* page got truncated out from underneath us */
8353 wait_on_page_writeback(page);
8355 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8356 set_page_extent_mapped(page);
8359 * we can't set the delalloc bits if there are pending ordered
8360 * extents. Drop our locks and wait for them to finish
8362 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8365 unlock_extent_cached(io_tree, page_start, page_end,
8368 btrfs_start_ordered_extent(ordered, 1);
8369 btrfs_put_ordered_extent(ordered);
8373 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8374 reserved_space = round_up(size - page_start,
8375 fs_info->sectorsize);
8376 if (reserved_space < PAGE_SIZE) {
8377 end = page_start + reserved_space - 1;
8378 btrfs_delalloc_release_space(BTRFS_I(inode),
8379 data_reserved, page_start,
8380 PAGE_SIZE - reserved_space, true);
8385 * page_mkwrite gets called when the page is firstly dirtied after it's
8386 * faulted in, but write(2) could also dirty a page and set delalloc
8387 * bits, thus in this case for space account reason, we still need to
8388 * clear any delalloc bits within this page range since we have to
8389 * reserve data&meta space before lock_page() (see above comments).
8391 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8392 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8393 EXTENT_DEFRAG, 0, 0, &cached_state);
8395 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8398 unlock_extent_cached(io_tree, page_start, page_end,
8400 ret = VM_FAULT_SIGBUS;
8404 /* page is wholly or partially inside EOF */
8405 if (page_start + PAGE_SIZE > size)
8406 zero_start = offset_in_page(size);
8408 zero_start = PAGE_SIZE;
8410 if (zero_start != PAGE_SIZE) {
8412 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8413 flush_dcache_page(page);
8416 ClearPageChecked(page);
8417 set_page_dirty(page);
8418 SetPageUptodate(page);
8420 BTRFS_I(inode)->last_trans = fs_info->generation;
8421 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8422 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8424 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8426 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8427 sb_end_pagefault(inode->i_sb);
8428 extent_changeset_free(data_reserved);
8429 return VM_FAULT_LOCKED;
8434 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8435 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8436 reserved_space, (ret != 0));
8438 sb_end_pagefault(inode->i_sb);
8439 extent_changeset_free(data_reserved);
8443 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8445 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8446 struct btrfs_root *root = BTRFS_I(inode)->root;
8447 struct btrfs_block_rsv *rsv;
8449 struct btrfs_trans_handle *trans;
8450 u64 mask = fs_info->sectorsize - 1;
8451 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8453 if (!skip_writeback) {
8454 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8461 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8462 * things going on here:
8464 * 1) We need to reserve space to update our inode.
8466 * 2) We need to have something to cache all the space that is going to
8467 * be free'd up by the truncate operation, but also have some slack
8468 * space reserved in case it uses space during the truncate (thank you
8469 * very much snapshotting).
8471 * And we need these to be separate. The fact is we can use a lot of
8472 * space doing the truncate, and we have no earthly idea how much space
8473 * we will use, so we need the truncate reservation to be separate so it
8474 * doesn't end up using space reserved for updating the inode. We also
8475 * need to be able to stop the transaction and start a new one, which
8476 * means we need to be able to update the inode several times, and we
8477 * have no idea of knowing how many times that will be, so we can't just
8478 * reserve 1 item for the entirety of the operation, so that has to be
8479 * done separately as well.
8481 * So that leaves us with
8483 * 1) rsv - for the truncate reservation, which we will steal from the
8484 * transaction reservation.
8485 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8486 * updating the inode.
8488 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8491 rsv->size = min_size;
8495 * 1 for the truncate slack space
8496 * 1 for updating the inode.
8498 trans = btrfs_start_transaction(root, 2);
8499 if (IS_ERR(trans)) {
8500 ret = PTR_ERR(trans);
8504 /* Migrate the slack space for the truncate to our reserve */
8505 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8510 * So if we truncate and then write and fsync we normally would just
8511 * write the extents that changed, which is a problem if we need to
8512 * first truncate that entire inode. So set this flag so we write out
8513 * all of the extents in the inode to the sync log so we're completely
8516 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8517 trans->block_rsv = rsv;
8520 ret = btrfs_truncate_inode_items(trans, root, BTRFS_I(inode),
8522 BTRFS_EXTENT_DATA_KEY);
8523 trans->block_rsv = &fs_info->trans_block_rsv;
8524 if (ret != -ENOSPC && ret != -EAGAIN)
8527 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
8531 btrfs_end_transaction(trans);
8532 btrfs_btree_balance_dirty(fs_info);
8534 trans = btrfs_start_transaction(root, 2);
8535 if (IS_ERR(trans)) {
8536 ret = PTR_ERR(trans);
8541 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8542 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8543 rsv, min_size, false);
8544 BUG_ON(ret); /* shouldn't happen */
8545 trans->block_rsv = rsv;
8549 * We can't call btrfs_truncate_block inside a trans handle as we could
8550 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
8551 * we've truncated everything except the last little bit, and can do
8552 * btrfs_truncate_block and then update the disk_i_size.
8554 if (ret == NEED_TRUNCATE_BLOCK) {
8555 btrfs_end_transaction(trans);
8556 btrfs_btree_balance_dirty(fs_info);
8558 ret = btrfs_truncate_block(BTRFS_I(inode), inode->i_size, 0, 0);
8561 trans = btrfs_start_transaction(root, 1);
8562 if (IS_ERR(trans)) {
8563 ret = PTR_ERR(trans);
8566 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
8572 trans->block_rsv = &fs_info->trans_block_rsv;
8573 ret2 = btrfs_update_inode(trans, root, BTRFS_I(inode));
8577 ret2 = btrfs_end_transaction(trans);
8580 btrfs_btree_balance_dirty(fs_info);
8583 btrfs_free_block_rsv(fs_info, rsv);
8589 * create a new subvolume directory/inode (helper for the ioctl).
8591 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8592 struct btrfs_root *new_root,
8593 struct btrfs_root *parent_root)
8595 struct inode *inode;
8600 err = btrfs_get_free_objectid(new_root, &ino);
8604 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, ino, ino,
8605 S_IFDIR | (~current_umask() & S_IRWXUGO),
8608 return PTR_ERR(inode);
8609 inode->i_op = &btrfs_dir_inode_operations;
8610 inode->i_fop = &btrfs_dir_file_operations;
8612 set_nlink(inode, 1);
8613 btrfs_i_size_write(BTRFS_I(inode), 0);
8614 unlock_new_inode(inode);
8616 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8618 btrfs_err(new_root->fs_info,
8619 "error inheriting subvolume %llu properties: %d",
8620 new_root->root_key.objectid, err);
8622 err = btrfs_update_inode(trans, new_root, BTRFS_I(inode));
8628 struct inode *btrfs_alloc_inode(struct super_block *sb)
8630 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8631 struct btrfs_inode *ei;
8632 struct inode *inode;
8634 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
8641 ei->last_sub_trans = 0;
8642 ei->logged_trans = 0;
8643 ei->delalloc_bytes = 0;
8644 ei->new_delalloc_bytes = 0;
8645 ei->defrag_bytes = 0;
8646 ei->disk_i_size = 0;
8649 ei->index_cnt = (u64)-1;
8651 ei->last_unlink_trans = 0;
8652 ei->last_reflink_trans = 0;
8653 ei->last_log_commit = 0;
8655 spin_lock_init(&ei->lock);
8656 ei->outstanding_extents = 0;
8657 if (sb->s_magic != BTRFS_TEST_MAGIC)
8658 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8659 BTRFS_BLOCK_RSV_DELALLOC);
8660 ei->runtime_flags = 0;
8661 ei->prop_compress = BTRFS_COMPRESS_NONE;
8662 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8664 ei->delayed_node = NULL;
8666 ei->i_otime.tv_sec = 0;
8667 ei->i_otime.tv_nsec = 0;
8669 inode = &ei->vfs_inode;
8670 extent_map_tree_init(&ei->extent_tree);
8671 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO, inode);
8672 extent_io_tree_init(fs_info, &ei->io_failure_tree,
8673 IO_TREE_INODE_IO_FAILURE, inode);
8674 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8675 IO_TREE_INODE_FILE_EXTENT, inode);
8676 ei->io_tree.track_uptodate = true;
8677 ei->io_failure_tree.track_uptodate = true;
8678 atomic_set(&ei->sync_writers, 0);
8679 mutex_init(&ei->log_mutex);
8680 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8681 INIT_LIST_HEAD(&ei->delalloc_inodes);
8682 INIT_LIST_HEAD(&ei->delayed_iput);
8683 RB_CLEAR_NODE(&ei->rb_node);
8688 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8689 void btrfs_test_destroy_inode(struct inode *inode)
8691 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
8692 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8696 void btrfs_free_inode(struct inode *inode)
8698 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8701 void btrfs_destroy_inode(struct inode *vfs_inode)
8703 struct btrfs_ordered_extent *ordered;
8704 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8705 struct btrfs_root *root = inode->root;
8707 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8708 WARN_ON(vfs_inode->i_data.nrpages);
8709 WARN_ON(inode->block_rsv.reserved);
8710 WARN_ON(inode->block_rsv.size);
8711 WARN_ON(inode->outstanding_extents);
8712 WARN_ON(inode->delalloc_bytes);
8713 WARN_ON(inode->new_delalloc_bytes);
8714 WARN_ON(inode->csum_bytes);
8715 WARN_ON(inode->defrag_bytes);
8718 * This can happen where we create an inode, but somebody else also
8719 * created the same inode and we need to destroy the one we already
8726 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8730 btrfs_err(root->fs_info,
8731 "found ordered extent %llu %llu on inode cleanup",
8732 ordered->file_offset, ordered->num_bytes);
8733 btrfs_remove_ordered_extent(inode, ordered);
8734 btrfs_put_ordered_extent(ordered);
8735 btrfs_put_ordered_extent(ordered);
8738 btrfs_qgroup_check_reserved_leak(inode);
8739 inode_tree_del(inode);
8740 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8741 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8742 btrfs_put_root(inode->root);
8745 int btrfs_drop_inode(struct inode *inode)
8747 struct btrfs_root *root = BTRFS_I(inode)->root;
8752 /* the snap/subvol tree is on deleting */
8753 if (btrfs_root_refs(&root->root_item) == 0)
8756 return generic_drop_inode(inode);
8759 static void init_once(void *foo)
8761 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8763 inode_init_once(&ei->vfs_inode);
8766 void __cold btrfs_destroy_cachep(void)
8769 * Make sure all delayed rcu free inodes are flushed before we
8773 kmem_cache_destroy(btrfs_inode_cachep);
8774 kmem_cache_destroy(btrfs_trans_handle_cachep);
8775 kmem_cache_destroy(btrfs_path_cachep);
8776 kmem_cache_destroy(btrfs_free_space_cachep);
8777 kmem_cache_destroy(btrfs_free_space_bitmap_cachep);
8780 int __init btrfs_init_cachep(void)
8782 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8783 sizeof(struct btrfs_inode), 0,
8784 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8786 if (!btrfs_inode_cachep)
8789 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8790 sizeof(struct btrfs_trans_handle), 0,
8791 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
8792 if (!btrfs_trans_handle_cachep)
8795 btrfs_path_cachep = kmem_cache_create("btrfs_path",
8796 sizeof(struct btrfs_path), 0,
8797 SLAB_MEM_SPREAD, NULL);
8798 if (!btrfs_path_cachep)
8801 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8802 sizeof(struct btrfs_free_space), 0,
8803 SLAB_MEM_SPREAD, NULL);
8804 if (!btrfs_free_space_cachep)
8807 btrfs_free_space_bitmap_cachep = kmem_cache_create("btrfs_free_space_bitmap",
8808 PAGE_SIZE, PAGE_SIZE,
8809 SLAB_RED_ZONE, NULL);
8810 if (!btrfs_free_space_bitmap_cachep)
8815 btrfs_destroy_cachep();
8819 static int btrfs_getattr(const struct path *path, struct kstat *stat,
8820 u32 request_mask, unsigned int flags)
8824 struct inode *inode = d_inode(path->dentry);
8825 u32 blocksize = inode->i_sb->s_blocksize;
8826 u32 bi_flags = BTRFS_I(inode)->flags;
8828 stat->result_mask |= STATX_BTIME;
8829 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8830 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8831 if (bi_flags & BTRFS_INODE_APPEND)
8832 stat->attributes |= STATX_ATTR_APPEND;
8833 if (bi_flags & BTRFS_INODE_COMPRESS)
8834 stat->attributes |= STATX_ATTR_COMPRESSED;
8835 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8836 stat->attributes |= STATX_ATTR_IMMUTABLE;
8837 if (bi_flags & BTRFS_INODE_NODUMP)
8838 stat->attributes |= STATX_ATTR_NODUMP;
8840 stat->attributes_mask |= (STATX_ATTR_APPEND |
8841 STATX_ATTR_COMPRESSED |
8842 STATX_ATTR_IMMUTABLE |
8845 generic_fillattr(inode, stat);
8846 stat->dev = BTRFS_I(inode)->root->anon_dev;
8848 spin_lock(&BTRFS_I(inode)->lock);
8849 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8850 inode_bytes = inode_get_bytes(inode);
8851 spin_unlock(&BTRFS_I(inode)->lock);
8852 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8853 ALIGN(delalloc_bytes, blocksize)) >> 9;
8857 static int btrfs_rename_exchange(struct inode *old_dir,
8858 struct dentry *old_dentry,
8859 struct inode *new_dir,
8860 struct dentry *new_dentry)
8862 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8863 struct btrfs_trans_handle *trans;
8864 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8865 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8866 struct inode *new_inode = new_dentry->d_inode;
8867 struct inode *old_inode = old_dentry->d_inode;
8868 struct timespec64 ctime = current_time(old_inode);
8869 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8870 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8875 bool root_log_pinned = false;
8876 bool dest_log_pinned = false;
8878 /* we only allow rename subvolume link between subvolumes */
8879 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8882 /* close the race window with snapshot create/destroy ioctl */
8883 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8884 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8885 down_read(&fs_info->subvol_sem);
8888 * We want to reserve the absolute worst case amount of items. So if
8889 * both inodes are subvols and we need to unlink them then that would
8890 * require 4 item modifications, but if they are both normal inodes it
8891 * would require 5 item modifications, so we'll assume their normal
8892 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
8893 * should cover the worst case number of items we'll modify.
8895 trans = btrfs_start_transaction(root, 12);
8896 if (IS_ERR(trans)) {
8897 ret = PTR_ERR(trans);
8902 btrfs_record_root_in_trans(trans, dest);
8905 * We need to find a free sequence number both in the source and
8906 * in the destination directory for the exchange.
8908 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8911 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8915 BTRFS_I(old_inode)->dir_index = 0ULL;
8916 BTRFS_I(new_inode)->dir_index = 0ULL;
8918 /* Reference for the source. */
8919 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8920 /* force full log commit if subvolume involved. */
8921 btrfs_set_log_full_commit(trans);
8923 btrfs_pin_log_trans(root);
8924 root_log_pinned = true;
8925 ret = btrfs_insert_inode_ref(trans, dest,
8926 new_dentry->d_name.name,
8927 new_dentry->d_name.len,
8929 btrfs_ino(BTRFS_I(new_dir)),
8935 /* And now for the dest. */
8936 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8937 /* force full log commit if subvolume involved. */
8938 btrfs_set_log_full_commit(trans);
8940 btrfs_pin_log_trans(dest);
8941 dest_log_pinned = true;
8942 ret = btrfs_insert_inode_ref(trans, root,
8943 old_dentry->d_name.name,
8944 old_dentry->d_name.len,
8946 btrfs_ino(BTRFS_I(old_dir)),
8952 /* Update inode version and ctime/mtime. */
8953 inode_inc_iversion(old_dir);
8954 inode_inc_iversion(new_dir);
8955 inode_inc_iversion(old_inode);
8956 inode_inc_iversion(new_inode);
8957 old_dir->i_ctime = old_dir->i_mtime = ctime;
8958 new_dir->i_ctime = new_dir->i_mtime = ctime;
8959 old_inode->i_ctime = ctime;
8960 new_inode->i_ctime = ctime;
8962 if (old_dentry->d_parent != new_dentry->d_parent) {
8963 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8964 BTRFS_I(old_inode), 1);
8965 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8966 BTRFS_I(new_inode), 1);
8969 /* src is a subvolume */
8970 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8971 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
8972 } else { /* src is an inode */
8973 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
8974 BTRFS_I(old_dentry->d_inode),
8975 old_dentry->d_name.name,
8976 old_dentry->d_name.len);
8978 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8981 btrfs_abort_transaction(trans, ret);
8985 /* dest is a subvolume */
8986 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8987 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
8988 } else { /* dest is an inode */
8989 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
8990 BTRFS_I(new_dentry->d_inode),
8991 new_dentry->d_name.name,
8992 new_dentry->d_name.len);
8994 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8997 btrfs_abort_transaction(trans, ret);
9001 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9002 new_dentry->d_name.name,
9003 new_dentry->d_name.len, 0, old_idx);
9005 btrfs_abort_transaction(trans, ret);
9009 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9010 old_dentry->d_name.name,
9011 old_dentry->d_name.len, 0, new_idx);
9013 btrfs_abort_transaction(trans, ret);
9017 if (old_inode->i_nlink == 1)
9018 BTRFS_I(old_inode)->dir_index = old_idx;
9019 if (new_inode->i_nlink == 1)
9020 BTRFS_I(new_inode)->dir_index = new_idx;
9022 if (root_log_pinned) {
9023 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9024 new_dentry->d_parent);
9025 btrfs_end_log_trans(root);
9026 root_log_pinned = false;
9028 if (dest_log_pinned) {
9029 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9030 old_dentry->d_parent);
9031 btrfs_end_log_trans(dest);
9032 dest_log_pinned = false;
9036 * If we have pinned a log and an error happened, we unpin tasks
9037 * trying to sync the log and force them to fallback to a transaction
9038 * commit if the log currently contains any of the inodes involved in
9039 * this rename operation (to ensure we do not persist a log with an
9040 * inconsistent state for any of these inodes or leading to any
9041 * inconsistencies when replayed). If the transaction was aborted, the
9042 * abortion reason is propagated to userspace when attempting to commit
9043 * the transaction. If the log does not contain any of these inodes, we
9044 * allow the tasks to sync it.
9046 if (ret && (root_log_pinned || dest_log_pinned)) {
9047 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9048 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9049 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9051 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9052 btrfs_set_log_full_commit(trans);
9054 if (root_log_pinned) {
9055 btrfs_end_log_trans(root);
9056 root_log_pinned = false;
9058 if (dest_log_pinned) {
9059 btrfs_end_log_trans(dest);
9060 dest_log_pinned = false;
9063 ret2 = btrfs_end_transaction(trans);
9064 ret = ret ? ret : ret2;
9066 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9067 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9068 up_read(&fs_info->subvol_sem);
9073 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9074 struct btrfs_root *root,
9076 struct dentry *dentry)
9079 struct inode *inode;
9083 ret = btrfs_get_free_objectid(root, &objectid);
9087 inode = btrfs_new_inode(trans, root, dir,
9088 dentry->d_name.name,
9090 btrfs_ino(BTRFS_I(dir)),
9092 S_IFCHR | WHITEOUT_MODE,
9095 if (IS_ERR(inode)) {
9096 ret = PTR_ERR(inode);
9100 inode->i_op = &btrfs_special_inode_operations;
9101 init_special_inode(inode, inode->i_mode,
9104 ret = btrfs_init_inode_security(trans, inode, dir,
9109 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9110 BTRFS_I(inode), 0, index);
9114 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9116 unlock_new_inode(inode);
9118 inode_dec_link_count(inode);
9124 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9125 struct inode *new_dir, struct dentry *new_dentry,
9128 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9129 struct btrfs_trans_handle *trans;
9130 unsigned int trans_num_items;
9131 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9132 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9133 struct inode *new_inode = d_inode(new_dentry);
9134 struct inode *old_inode = d_inode(old_dentry);
9138 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9139 bool log_pinned = false;
9141 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9144 /* we only allow rename subvolume link between subvolumes */
9145 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9148 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9149 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9152 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9153 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9157 /* check for collisions, even if the name isn't there */
9158 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9159 new_dentry->d_name.name,
9160 new_dentry->d_name.len);
9163 if (ret == -EEXIST) {
9165 * eexist without a new_inode */
9166 if (WARN_ON(!new_inode)) {
9170 /* maybe -EOVERFLOW */
9177 * we're using rename to replace one file with another. Start IO on it
9178 * now so we don't add too much work to the end of the transaction
9180 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9181 filemap_flush(old_inode->i_mapping);
9183 /* close the racy window with snapshot create/destroy ioctl */
9184 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9185 down_read(&fs_info->subvol_sem);
9187 * We want to reserve the absolute worst case amount of items. So if
9188 * both inodes are subvols and we need to unlink them then that would
9189 * require 4 item modifications, but if they are both normal inodes it
9190 * would require 5 item modifications, so we'll assume they are normal
9191 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9192 * should cover the worst case number of items we'll modify.
9193 * If our rename has the whiteout flag, we need more 5 units for the
9194 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9195 * when selinux is enabled).
9197 trans_num_items = 11;
9198 if (flags & RENAME_WHITEOUT)
9199 trans_num_items += 5;
9200 trans = btrfs_start_transaction(root, trans_num_items);
9201 if (IS_ERR(trans)) {
9202 ret = PTR_ERR(trans);
9207 btrfs_record_root_in_trans(trans, dest);
9209 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9213 BTRFS_I(old_inode)->dir_index = 0ULL;
9214 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9215 /* force full log commit if subvolume involved. */
9216 btrfs_set_log_full_commit(trans);
9218 btrfs_pin_log_trans(root);
9220 ret = btrfs_insert_inode_ref(trans, dest,
9221 new_dentry->d_name.name,
9222 new_dentry->d_name.len,
9224 btrfs_ino(BTRFS_I(new_dir)), index);
9229 inode_inc_iversion(old_dir);
9230 inode_inc_iversion(new_dir);
9231 inode_inc_iversion(old_inode);
9232 old_dir->i_ctime = old_dir->i_mtime =
9233 new_dir->i_ctime = new_dir->i_mtime =
9234 old_inode->i_ctime = current_time(old_dir);
9236 if (old_dentry->d_parent != new_dentry->d_parent)
9237 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9238 BTRFS_I(old_inode), 1);
9240 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9241 ret = btrfs_unlink_subvol(trans, old_dir, old_dentry);
9243 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9244 BTRFS_I(d_inode(old_dentry)),
9245 old_dentry->d_name.name,
9246 old_dentry->d_name.len);
9248 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9251 btrfs_abort_transaction(trans, ret);
9256 inode_inc_iversion(new_inode);
9257 new_inode->i_ctime = current_time(new_inode);
9258 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9259 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9260 ret = btrfs_unlink_subvol(trans, new_dir, new_dentry);
9261 BUG_ON(new_inode->i_nlink == 0);
9263 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9264 BTRFS_I(d_inode(new_dentry)),
9265 new_dentry->d_name.name,
9266 new_dentry->d_name.len);
9268 if (!ret && new_inode->i_nlink == 0)
9269 ret = btrfs_orphan_add(trans,
9270 BTRFS_I(d_inode(new_dentry)));
9272 btrfs_abort_transaction(trans, ret);
9277 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9278 new_dentry->d_name.name,
9279 new_dentry->d_name.len, 0, index);
9281 btrfs_abort_transaction(trans, ret);
9285 if (old_inode->i_nlink == 1)
9286 BTRFS_I(old_inode)->dir_index = index;
9289 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9290 new_dentry->d_parent);
9291 btrfs_end_log_trans(root);
9295 if (flags & RENAME_WHITEOUT) {
9296 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9300 btrfs_abort_transaction(trans, ret);
9306 * If we have pinned the log and an error happened, we unpin tasks
9307 * trying to sync the log and force them to fallback to a transaction
9308 * commit if the log currently contains any of the inodes involved in
9309 * this rename operation (to ensure we do not persist a log with an
9310 * inconsistent state for any of these inodes or leading to any
9311 * inconsistencies when replayed). If the transaction was aborted, the
9312 * abortion reason is propagated to userspace when attempting to commit
9313 * the transaction. If the log does not contain any of these inodes, we
9314 * allow the tasks to sync it.
9316 if (ret && log_pinned) {
9317 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9318 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9319 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9321 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9322 btrfs_set_log_full_commit(trans);
9324 btrfs_end_log_trans(root);
9327 ret2 = btrfs_end_transaction(trans);
9328 ret = ret ? ret : ret2;
9330 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9331 up_read(&fs_info->subvol_sem);
9336 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9337 struct inode *new_dir, struct dentry *new_dentry,
9340 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9343 if (flags & RENAME_EXCHANGE)
9344 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9347 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9350 struct btrfs_delalloc_work {
9351 struct inode *inode;
9352 struct completion completion;
9353 struct list_head list;
9354 struct btrfs_work work;
9357 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9359 struct btrfs_delalloc_work *delalloc_work;
9360 struct inode *inode;
9362 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9364 inode = delalloc_work->inode;
9365 filemap_flush(inode->i_mapping);
9366 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9367 &BTRFS_I(inode)->runtime_flags))
9368 filemap_flush(inode->i_mapping);
9371 complete(&delalloc_work->completion);
9374 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9376 struct btrfs_delalloc_work *work;
9378 work = kmalloc(sizeof(*work), GFP_NOFS);
9382 init_completion(&work->completion);
9383 INIT_LIST_HEAD(&work->list);
9384 work->inode = inode;
9385 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9391 * some fairly slow code that needs optimization. This walks the list
9392 * of all the inodes with pending delalloc and forces them to disk.
9394 static int start_delalloc_inodes(struct btrfs_root *root,
9395 struct writeback_control *wbc, bool snapshot,
9396 bool in_reclaim_context)
9398 struct btrfs_inode *binode;
9399 struct inode *inode;
9400 struct btrfs_delalloc_work *work, *next;
9401 struct list_head works;
9402 struct list_head splice;
9404 bool full_flush = wbc->nr_to_write == LONG_MAX;
9406 INIT_LIST_HEAD(&works);
9407 INIT_LIST_HEAD(&splice);
9409 mutex_lock(&root->delalloc_mutex);
9410 spin_lock(&root->delalloc_lock);
9411 list_splice_init(&root->delalloc_inodes, &splice);
9412 while (!list_empty(&splice)) {
9413 binode = list_entry(splice.next, struct btrfs_inode,
9416 list_move_tail(&binode->delalloc_inodes,
9417 &root->delalloc_inodes);
9419 if (in_reclaim_context &&
9420 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9423 inode = igrab(&binode->vfs_inode);
9425 cond_resched_lock(&root->delalloc_lock);
9428 spin_unlock(&root->delalloc_lock);
9431 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9432 &binode->runtime_flags);
9434 work = btrfs_alloc_delalloc_work(inode);
9440 list_add_tail(&work->list, &works);
9441 btrfs_queue_work(root->fs_info->flush_workers,
9444 ret = sync_inode(inode, wbc);
9446 test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9447 &BTRFS_I(inode)->runtime_flags))
9448 ret = sync_inode(inode, wbc);
9449 btrfs_add_delayed_iput(inode);
9450 if (ret || wbc->nr_to_write <= 0)
9454 spin_lock(&root->delalloc_lock);
9456 spin_unlock(&root->delalloc_lock);
9459 list_for_each_entry_safe(work, next, &works, list) {
9460 list_del_init(&work->list);
9461 wait_for_completion(&work->completion);
9465 if (!list_empty(&splice)) {
9466 spin_lock(&root->delalloc_lock);
9467 list_splice_tail(&splice, &root->delalloc_inodes);
9468 spin_unlock(&root->delalloc_lock);
9470 mutex_unlock(&root->delalloc_mutex);
9474 int btrfs_start_delalloc_snapshot(struct btrfs_root *root)
9476 struct writeback_control wbc = {
9477 .nr_to_write = LONG_MAX,
9478 .sync_mode = WB_SYNC_NONE,
9480 .range_end = LLONG_MAX,
9482 struct btrfs_fs_info *fs_info = root->fs_info;
9484 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9487 return start_delalloc_inodes(root, &wbc, true, false);
9490 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9491 bool in_reclaim_context)
9493 struct writeback_control wbc = {
9495 .sync_mode = WB_SYNC_NONE,
9497 .range_end = LLONG_MAX,
9499 struct btrfs_root *root;
9500 struct list_head splice;
9503 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9506 INIT_LIST_HEAD(&splice);
9508 mutex_lock(&fs_info->delalloc_root_mutex);
9509 spin_lock(&fs_info->delalloc_root_lock);
9510 list_splice_init(&fs_info->delalloc_roots, &splice);
9511 while (!list_empty(&splice)) {
9513 * Reset nr_to_write here so we know that we're doing a full
9517 wbc.nr_to_write = LONG_MAX;
9519 root = list_first_entry(&splice, struct btrfs_root,
9521 root = btrfs_grab_root(root);
9523 list_move_tail(&root->delalloc_root,
9524 &fs_info->delalloc_roots);
9525 spin_unlock(&fs_info->delalloc_root_lock);
9527 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9528 btrfs_put_root(root);
9529 if (ret < 0 || wbc.nr_to_write <= 0)
9531 spin_lock(&fs_info->delalloc_root_lock);
9533 spin_unlock(&fs_info->delalloc_root_lock);
9537 if (!list_empty(&splice)) {
9538 spin_lock(&fs_info->delalloc_root_lock);
9539 list_splice_tail(&splice, &fs_info->delalloc_roots);
9540 spin_unlock(&fs_info->delalloc_root_lock);
9542 mutex_unlock(&fs_info->delalloc_root_mutex);
9546 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9547 const char *symname)
9549 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9550 struct btrfs_trans_handle *trans;
9551 struct btrfs_root *root = BTRFS_I(dir)->root;
9552 struct btrfs_path *path;
9553 struct btrfs_key key;
9554 struct inode *inode = NULL;
9561 struct btrfs_file_extent_item *ei;
9562 struct extent_buffer *leaf;
9564 name_len = strlen(symname);
9565 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9566 return -ENAMETOOLONG;
9569 * 2 items for inode item and ref
9570 * 2 items for dir items
9571 * 1 item for updating parent inode item
9572 * 1 item for the inline extent item
9573 * 1 item for xattr if selinux is on
9575 trans = btrfs_start_transaction(root, 7);
9577 return PTR_ERR(trans);
9579 err = btrfs_get_free_objectid(root, &objectid);
9583 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9584 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
9585 objectid, S_IFLNK|S_IRWXUGO, &index);
9586 if (IS_ERR(inode)) {
9587 err = PTR_ERR(inode);
9593 * If the active LSM wants to access the inode during
9594 * d_instantiate it needs these. Smack checks to see
9595 * if the filesystem supports xattrs by looking at the
9598 inode->i_fop = &btrfs_file_operations;
9599 inode->i_op = &btrfs_file_inode_operations;
9600 inode->i_mapping->a_ops = &btrfs_aops;
9602 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9606 path = btrfs_alloc_path();
9611 key.objectid = btrfs_ino(BTRFS_I(inode));
9613 key.type = BTRFS_EXTENT_DATA_KEY;
9614 datasize = btrfs_file_extent_calc_inline_size(name_len);
9615 err = btrfs_insert_empty_item(trans, root, path, &key,
9618 btrfs_free_path(path);
9621 leaf = path->nodes[0];
9622 ei = btrfs_item_ptr(leaf, path->slots[0],
9623 struct btrfs_file_extent_item);
9624 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9625 btrfs_set_file_extent_type(leaf, ei,
9626 BTRFS_FILE_EXTENT_INLINE);
9627 btrfs_set_file_extent_encryption(leaf, ei, 0);
9628 btrfs_set_file_extent_compression(leaf, ei, 0);
9629 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9630 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9632 ptr = btrfs_file_extent_inline_start(ei);
9633 write_extent_buffer(leaf, symname, ptr, name_len);
9634 btrfs_mark_buffer_dirty(leaf);
9635 btrfs_free_path(path);
9637 inode->i_op = &btrfs_symlink_inode_operations;
9638 inode_nohighmem(inode);
9639 inode_set_bytes(inode, name_len);
9640 btrfs_i_size_write(BTRFS_I(inode), name_len);
9641 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
9643 * Last step, add directory indexes for our symlink inode. This is the
9644 * last step to avoid extra cleanup of these indexes if an error happens
9648 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9649 BTRFS_I(inode), 0, index);
9653 d_instantiate_new(dentry, inode);
9656 btrfs_end_transaction(trans);
9658 inode_dec_link_count(inode);
9659 discard_new_inode(inode);
9661 btrfs_btree_balance_dirty(fs_info);
9665 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9666 struct btrfs_trans_handle *trans_in,
9667 struct btrfs_inode *inode,
9668 struct btrfs_key *ins,
9671 struct btrfs_file_extent_item stack_fi;
9672 struct btrfs_replace_extent_info extent_info;
9673 struct btrfs_trans_handle *trans = trans_in;
9674 struct btrfs_path *path;
9675 u64 start = ins->objectid;
9676 u64 len = ins->offset;
9679 memset(&stack_fi, 0, sizeof(stack_fi));
9681 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9682 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9683 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9684 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9685 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9686 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9687 /* Encryption and other encoding is reserved and all 0 */
9689 ret = btrfs_qgroup_release_data(inode, file_offset, len);
9691 return ERR_PTR(ret);
9694 ret = insert_reserved_file_extent(trans, inode,
9695 file_offset, &stack_fi,
9698 return ERR_PTR(ret);
9702 extent_info.disk_offset = start;
9703 extent_info.disk_len = len;
9704 extent_info.data_offset = 0;
9705 extent_info.data_len = len;
9706 extent_info.file_offset = file_offset;
9707 extent_info.extent_buf = (char *)&stack_fi;
9708 extent_info.is_new_extent = true;
9709 extent_info.qgroup_reserved = ret;
9710 extent_info.insertions = 0;
9712 path = btrfs_alloc_path();
9714 return ERR_PTR(-ENOMEM);
9716 ret = btrfs_replace_file_extents(&inode->vfs_inode, path, file_offset,
9717 file_offset + len - 1, &extent_info,
9719 btrfs_free_path(path);
9721 return ERR_PTR(ret);
9726 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9727 u64 start, u64 num_bytes, u64 min_size,
9728 loff_t actual_len, u64 *alloc_hint,
9729 struct btrfs_trans_handle *trans)
9731 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9732 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9733 struct extent_map *em;
9734 struct btrfs_root *root = BTRFS_I(inode)->root;
9735 struct btrfs_key ins;
9736 u64 cur_offset = start;
9737 u64 clear_offset = start;
9740 u64 last_alloc = (u64)-1;
9742 bool own_trans = true;
9743 u64 end = start + num_bytes - 1;
9747 while (num_bytes > 0) {
9748 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9749 cur_bytes = max(cur_bytes, min_size);
9751 * If we are severely fragmented we could end up with really
9752 * small allocations, so if the allocator is returning small
9753 * chunks lets make its job easier by only searching for those
9756 cur_bytes = min(cur_bytes, last_alloc);
9757 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9758 min_size, 0, *alloc_hint, &ins, 1, 0);
9763 * We've reserved this space, and thus converted it from
9764 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9765 * from here on out we will only need to clear our reservation
9766 * for the remaining unreserved area, so advance our
9767 * clear_offset by our extent size.
9769 clear_offset += ins.offset;
9771 last_alloc = ins.offset;
9772 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9775 * Now that we inserted the prealloc extent we can finally
9776 * decrement the number of reservations in the block group.
9777 * If we did it before, we could race with relocation and have
9778 * relocation miss the reserved extent, making it fail later.
9780 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9781 if (IS_ERR(trans)) {
9782 ret = PTR_ERR(trans);
9783 btrfs_free_reserved_extent(fs_info, ins.objectid,
9788 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9789 cur_offset + ins.offset -1, 0);
9791 em = alloc_extent_map();
9793 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9794 &BTRFS_I(inode)->runtime_flags);
9798 em->start = cur_offset;
9799 em->orig_start = cur_offset;
9800 em->len = ins.offset;
9801 em->block_start = ins.objectid;
9802 em->block_len = ins.offset;
9803 em->orig_block_len = ins.offset;
9804 em->ram_bytes = ins.offset;
9805 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9806 em->generation = trans->transid;
9809 write_lock(&em_tree->lock);
9810 ret = add_extent_mapping(em_tree, em, 1);
9811 write_unlock(&em_tree->lock);
9814 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
9815 cur_offset + ins.offset - 1,
9818 free_extent_map(em);
9820 num_bytes -= ins.offset;
9821 cur_offset += ins.offset;
9822 *alloc_hint = ins.objectid + ins.offset;
9824 inode_inc_iversion(inode);
9825 inode->i_ctime = current_time(inode);
9826 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9827 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9828 (actual_len > inode->i_size) &&
9829 (cur_offset > inode->i_size)) {
9830 if (cur_offset > actual_len)
9831 i_size = actual_len;
9833 i_size = cur_offset;
9834 i_size_write(inode, i_size);
9835 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9838 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9841 btrfs_abort_transaction(trans, ret);
9843 btrfs_end_transaction(trans);
9848 btrfs_end_transaction(trans);
9852 if (clear_offset < end)
9853 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9854 end - clear_offset + 1);
9858 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9859 u64 start, u64 num_bytes, u64 min_size,
9860 loff_t actual_len, u64 *alloc_hint)
9862 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9863 min_size, actual_len, alloc_hint,
9867 int btrfs_prealloc_file_range_trans(struct inode *inode,
9868 struct btrfs_trans_handle *trans, int mode,
9869 u64 start, u64 num_bytes, u64 min_size,
9870 loff_t actual_len, u64 *alloc_hint)
9872 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9873 min_size, actual_len, alloc_hint, trans);
9876 static int btrfs_set_page_dirty(struct page *page)
9878 return __set_page_dirty_nobuffers(page);
9881 static int btrfs_permission(struct inode *inode, int mask)
9883 struct btrfs_root *root = BTRFS_I(inode)->root;
9884 umode_t mode = inode->i_mode;
9886 if (mask & MAY_WRITE &&
9887 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9888 if (btrfs_root_readonly(root))
9890 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9893 return generic_permission(inode, mask);
9896 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9898 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9899 struct btrfs_trans_handle *trans;
9900 struct btrfs_root *root = BTRFS_I(dir)->root;
9901 struct inode *inode = NULL;
9907 * 5 units required for adding orphan entry
9909 trans = btrfs_start_transaction(root, 5);
9911 return PTR_ERR(trans);
9913 ret = btrfs_get_free_objectid(root, &objectid);
9917 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9918 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
9919 if (IS_ERR(inode)) {
9920 ret = PTR_ERR(inode);
9925 inode->i_fop = &btrfs_file_operations;
9926 inode->i_op = &btrfs_file_inode_operations;
9928 inode->i_mapping->a_ops = &btrfs_aops;
9930 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9934 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9937 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
9942 * We set number of links to 0 in btrfs_new_inode(), and here we set
9943 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9946 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9948 set_nlink(inode, 1);
9949 d_tmpfile(dentry, inode);
9950 unlock_new_inode(inode);
9951 mark_inode_dirty(inode);
9953 btrfs_end_transaction(trans);
9955 discard_new_inode(inode);
9956 btrfs_btree_balance_dirty(fs_info);
9960 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
9962 struct inode *inode = tree->private_data;
9963 unsigned long index = start >> PAGE_SHIFT;
9964 unsigned long end_index = end >> PAGE_SHIFT;
9967 while (index <= end_index) {
9968 page = find_get_page(inode->i_mapping, index);
9969 ASSERT(page); /* Pages should be in the extent_io_tree */
9970 set_page_writeback(page);
9978 * Add an entry indicating a block group or device which is pinned by a
9979 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
9980 * negative errno on failure.
9982 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
9983 bool is_block_group)
9985 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
9986 struct btrfs_swapfile_pin *sp, *entry;
9988 struct rb_node *parent = NULL;
9990 sp = kmalloc(sizeof(*sp), GFP_NOFS);
9995 sp->is_block_group = is_block_group;
9997 spin_lock(&fs_info->swapfile_pins_lock);
9998 p = &fs_info->swapfile_pins.rb_node;
10001 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10002 if (sp->ptr < entry->ptr ||
10003 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10004 p = &(*p)->rb_left;
10005 } else if (sp->ptr > entry->ptr ||
10006 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10007 p = &(*p)->rb_right;
10009 spin_unlock(&fs_info->swapfile_pins_lock);
10014 rb_link_node(&sp->node, parent, p);
10015 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10016 spin_unlock(&fs_info->swapfile_pins_lock);
10020 /* Free all of the entries pinned by this swapfile. */
10021 static void btrfs_free_swapfile_pins(struct inode *inode)
10023 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10024 struct btrfs_swapfile_pin *sp;
10025 struct rb_node *node, *next;
10027 spin_lock(&fs_info->swapfile_pins_lock);
10028 node = rb_first(&fs_info->swapfile_pins);
10030 next = rb_next(node);
10031 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10032 if (sp->inode == inode) {
10033 rb_erase(&sp->node, &fs_info->swapfile_pins);
10034 if (sp->is_block_group)
10035 btrfs_put_block_group(sp->ptr);
10040 spin_unlock(&fs_info->swapfile_pins_lock);
10043 struct btrfs_swap_info {
10049 unsigned long nr_pages;
10053 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10054 struct btrfs_swap_info *bsi)
10056 unsigned long nr_pages;
10057 u64 first_ppage, first_ppage_reported, next_ppage;
10060 first_ppage = ALIGN(bsi->block_start, PAGE_SIZE) >> PAGE_SHIFT;
10061 next_ppage = ALIGN_DOWN(bsi->block_start + bsi->block_len,
10062 PAGE_SIZE) >> PAGE_SHIFT;
10064 if (first_ppage >= next_ppage)
10066 nr_pages = next_ppage - first_ppage;
10068 first_ppage_reported = first_ppage;
10069 if (bsi->start == 0)
10070 first_ppage_reported++;
10071 if (bsi->lowest_ppage > first_ppage_reported)
10072 bsi->lowest_ppage = first_ppage_reported;
10073 if (bsi->highest_ppage < (next_ppage - 1))
10074 bsi->highest_ppage = next_ppage - 1;
10076 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10079 bsi->nr_extents += ret;
10080 bsi->nr_pages += nr_pages;
10084 static void btrfs_swap_deactivate(struct file *file)
10086 struct inode *inode = file_inode(file);
10088 btrfs_free_swapfile_pins(inode);
10089 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10092 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10095 struct inode *inode = file_inode(file);
10096 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10097 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10098 struct extent_state *cached_state = NULL;
10099 struct extent_map *em = NULL;
10100 struct btrfs_device *device = NULL;
10101 struct btrfs_swap_info bsi = {
10102 .lowest_ppage = (sector_t)-1ULL,
10109 * If the swap file was just created, make sure delalloc is done. If the
10110 * file changes again after this, the user is doing something stupid and
10111 * we don't really care.
10113 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10118 * The inode is locked, so these flags won't change after we check them.
10120 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10121 btrfs_warn(fs_info, "swapfile must not be compressed");
10124 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10125 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10128 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10129 btrfs_warn(fs_info, "swapfile must not be checksummed");
10134 * Balance or device remove/replace/resize can move stuff around from
10135 * under us. The exclop protection makes sure they aren't running/won't
10136 * run concurrently while we are mapping the swap extents, and
10137 * fs_info->swapfile_pins prevents them from running while the swap
10138 * file is active and moving the extents. Note that this also prevents
10139 * a concurrent device add which isn't actually necessary, but it's not
10140 * really worth the trouble to allow it.
10142 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10143 btrfs_warn(fs_info,
10144 "cannot activate swapfile while exclusive operation is running");
10148 * Snapshots can create extents which require COW even if NODATACOW is
10149 * set. We use this counter to prevent snapshots. We must increment it
10150 * before walking the extents because we don't want a concurrent
10151 * snapshot to run after we've already checked the extents.
10153 atomic_inc(&BTRFS_I(inode)->root->nr_swapfiles);
10155 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10157 lock_extent_bits(io_tree, 0, isize - 1, &cached_state);
10159 while (start < isize) {
10160 u64 logical_block_start, physical_block_start;
10161 struct btrfs_block_group *bg;
10162 u64 len = isize - start;
10164 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10170 if (em->block_start == EXTENT_MAP_HOLE) {
10171 btrfs_warn(fs_info, "swapfile must not have holes");
10175 if (em->block_start == EXTENT_MAP_INLINE) {
10177 * It's unlikely we'll ever actually find ourselves
10178 * here, as a file small enough to fit inline won't be
10179 * big enough to store more than the swap header, but in
10180 * case something changes in the future, let's catch it
10181 * here rather than later.
10183 btrfs_warn(fs_info, "swapfile must not be inline");
10187 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10188 btrfs_warn(fs_info, "swapfile must not be compressed");
10193 logical_block_start = em->block_start + (start - em->start);
10194 len = min(len, em->len - (start - em->start));
10195 free_extent_map(em);
10198 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, true);
10204 btrfs_warn(fs_info,
10205 "swapfile must not be copy-on-write");
10210 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10216 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10217 btrfs_warn(fs_info,
10218 "swapfile must have single data profile");
10223 if (device == NULL) {
10224 device = em->map_lookup->stripes[0].dev;
10225 ret = btrfs_add_swapfile_pin(inode, device, false);
10230 } else if (device != em->map_lookup->stripes[0].dev) {
10231 btrfs_warn(fs_info, "swapfile must be on one device");
10236 physical_block_start = (em->map_lookup->stripes[0].physical +
10237 (logical_block_start - em->start));
10238 len = min(len, em->len - (logical_block_start - em->start));
10239 free_extent_map(em);
10242 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10244 btrfs_warn(fs_info,
10245 "could not find block group containing swapfile");
10250 ret = btrfs_add_swapfile_pin(inode, bg, true);
10252 btrfs_put_block_group(bg);
10259 if (bsi.block_len &&
10260 bsi.block_start + bsi.block_len == physical_block_start) {
10261 bsi.block_len += len;
10263 if (bsi.block_len) {
10264 ret = btrfs_add_swap_extent(sis, &bsi);
10269 bsi.block_start = physical_block_start;
10270 bsi.block_len = len;
10277 ret = btrfs_add_swap_extent(sis, &bsi);
10280 if (!IS_ERR_OR_NULL(em))
10281 free_extent_map(em);
10283 unlock_extent_cached(io_tree, 0, isize - 1, &cached_state);
10286 btrfs_swap_deactivate(file);
10288 btrfs_exclop_finish(fs_info);
10294 sis->bdev = device->bdev;
10295 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10296 sis->max = bsi.nr_pages;
10297 sis->pages = bsi.nr_pages - 1;
10298 sis->highest_bit = bsi.nr_pages - 1;
10299 return bsi.nr_extents;
10302 static void btrfs_swap_deactivate(struct file *file)
10306 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10309 return -EOPNOTSUPP;
10314 * Update the number of bytes used in the VFS' inode. When we replace extents in
10315 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10316 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10317 * always get a correct value.
10319 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10320 const u64 add_bytes,
10321 const u64 del_bytes)
10323 if (add_bytes == del_bytes)
10326 spin_lock(&inode->lock);
10328 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10330 inode_add_bytes(&inode->vfs_inode, add_bytes);
10331 spin_unlock(&inode->lock);
10334 static const struct inode_operations btrfs_dir_inode_operations = {
10335 .getattr = btrfs_getattr,
10336 .lookup = btrfs_lookup,
10337 .create = btrfs_create,
10338 .unlink = btrfs_unlink,
10339 .link = btrfs_link,
10340 .mkdir = btrfs_mkdir,
10341 .rmdir = btrfs_rmdir,
10342 .rename = btrfs_rename2,
10343 .symlink = btrfs_symlink,
10344 .setattr = btrfs_setattr,
10345 .mknod = btrfs_mknod,
10346 .listxattr = btrfs_listxattr,
10347 .permission = btrfs_permission,
10348 .get_acl = btrfs_get_acl,
10349 .set_acl = btrfs_set_acl,
10350 .update_time = btrfs_update_time,
10351 .tmpfile = btrfs_tmpfile,
10354 static const struct file_operations btrfs_dir_file_operations = {
10355 .llseek = generic_file_llseek,
10356 .read = generic_read_dir,
10357 .iterate_shared = btrfs_real_readdir,
10358 .open = btrfs_opendir,
10359 .unlocked_ioctl = btrfs_ioctl,
10360 #ifdef CONFIG_COMPAT
10361 .compat_ioctl = btrfs_compat_ioctl,
10363 .release = btrfs_release_file,
10364 .fsync = btrfs_sync_file,
10368 * btrfs doesn't support the bmap operation because swapfiles
10369 * use bmap to make a mapping of extents in the file. They assume
10370 * these extents won't change over the life of the file and they
10371 * use the bmap result to do IO directly to the drive.
10373 * the btrfs bmap call would return logical addresses that aren't
10374 * suitable for IO and they also will change frequently as COW
10375 * operations happen. So, swapfile + btrfs == corruption.
10377 * For now we're avoiding this by dropping bmap.
10379 static const struct address_space_operations btrfs_aops = {
10380 .readpage = btrfs_readpage,
10381 .writepage = btrfs_writepage,
10382 .writepages = btrfs_writepages,
10383 .readahead = btrfs_readahead,
10384 .direct_IO = noop_direct_IO,
10385 .invalidatepage = btrfs_invalidatepage,
10386 .releasepage = btrfs_releasepage,
10387 #ifdef CONFIG_MIGRATION
10388 .migratepage = btrfs_migratepage,
10390 .set_page_dirty = btrfs_set_page_dirty,
10391 .error_remove_page = generic_error_remove_page,
10392 .swap_activate = btrfs_swap_activate,
10393 .swap_deactivate = btrfs_swap_deactivate,
10396 static const struct inode_operations btrfs_file_inode_operations = {
10397 .getattr = btrfs_getattr,
10398 .setattr = btrfs_setattr,
10399 .listxattr = btrfs_listxattr,
10400 .permission = btrfs_permission,
10401 .fiemap = btrfs_fiemap,
10402 .get_acl = btrfs_get_acl,
10403 .set_acl = btrfs_set_acl,
10404 .update_time = btrfs_update_time,
10406 static const struct inode_operations btrfs_special_inode_operations = {
10407 .getattr = btrfs_getattr,
10408 .setattr = btrfs_setattr,
10409 .permission = btrfs_permission,
10410 .listxattr = btrfs_listxattr,
10411 .get_acl = btrfs_get_acl,
10412 .set_acl = btrfs_set_acl,
10413 .update_time = btrfs_update_time,
10415 static const struct inode_operations btrfs_symlink_inode_operations = {
10416 .get_link = page_get_link,
10417 .getattr = btrfs_getattr,
10418 .setattr = btrfs_setattr,
10419 .permission = btrfs_permission,
10420 .listxattr = btrfs_listxattr,
10421 .update_time = btrfs_update_time,
10424 const struct dentry_operations btrfs_dentry_operations = {
10425 .d_delete = btrfs_dentry_delete,