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/blk-cgroup.h>
10 #include <linux/file.h>
12 #include <linux/pagemap.h>
13 #include <linux/highmem.h>
14 #include <linux/time.h>
15 #include <linux/init.h>
16 #include <linux/string.h>
17 #include <linux/backing-dev.h>
18 #include <linux/writeback.h>
19 #include <linux/compat.h>
20 #include <linux/xattr.h>
21 #include <linux/posix_acl.h>
22 #include <linux/falloc.h>
23 #include <linux/slab.h>
24 #include <linux/ratelimit.h>
25 #include <linux/btrfs.h>
26 #include <linux/blkdev.h>
27 #include <linux/posix_acl_xattr.h>
28 #include <linux/uio.h>
29 #include <linux/magic.h>
30 #include <linux/iversion.h>
31 #include <linux/swap.h>
32 #include <linux/migrate.h>
33 #include <linux/sched/mm.h>
34 #include <linux/iomap.h>
35 #include <asm/unaligned.h>
36 #include <linux/fsverity.h>
40 #include "transaction.h"
41 #include "btrfs_inode.h"
42 #include "print-tree.h"
43 #include "ordered-data.h"
47 #include "compression.h"
49 #include "free-space-cache.h"
52 #include "delalloc-space.h"
53 #include "block-group.h"
54 #include "space-info.h"
57 #include "inode-item.h"
59 #include "accessors.h"
60 #include "extent-tree.h"
61 #include "root-tree.h"
64 #include "file-item.h"
65 #include "uuid-tree.h"
69 #include "relocation.h"
74 struct btrfs_iget_args {
76 struct btrfs_root *root;
79 struct btrfs_dio_data {
81 struct extent_changeset *data_reserved;
82 struct btrfs_ordered_extent *ordered;
83 bool data_space_reserved;
87 struct btrfs_dio_private {
92 /* This must be last */
93 struct btrfs_bio bbio;
96 static struct bio_set btrfs_dio_bioset;
98 struct btrfs_rename_ctx {
99 /* Output field. Stores the index number of the old directory entry. */
103 static const struct inode_operations btrfs_dir_inode_operations;
104 static const struct inode_operations btrfs_symlink_inode_operations;
105 static const struct inode_operations btrfs_special_inode_operations;
106 static const struct inode_operations btrfs_file_inode_operations;
107 static const struct address_space_operations btrfs_aops;
108 static const struct file_operations btrfs_dir_file_operations;
110 static struct kmem_cache *btrfs_inode_cachep;
112 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
113 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
114 static noinline int cow_file_range(struct btrfs_inode *inode,
115 struct page *locked_page,
116 u64 start, u64 end, int *page_started,
117 unsigned long *nr_written, int unlock,
119 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
120 u64 len, u64 orig_start, u64 block_start,
121 u64 block_len, u64 orig_block_len,
122 u64 ram_bytes, int compress_type,
125 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
126 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
128 struct btrfs_root *root = inode->root;
129 const u32 csum_size = root->fs_info->csum_size;
131 /* Output without objectid, which is more meaningful */
132 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
133 btrfs_warn_rl(root->fs_info,
134 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
135 root->root_key.objectid, btrfs_ino(inode),
137 CSUM_FMT_VALUE(csum_size, csum),
138 CSUM_FMT_VALUE(csum_size, csum_expected),
141 btrfs_warn_rl(root->fs_info,
142 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
143 root->root_key.objectid, btrfs_ino(inode),
145 CSUM_FMT_VALUE(csum_size, csum),
146 CSUM_FMT_VALUE(csum_size, csum_expected),
152 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
154 * ilock_flags can have the following bit set:
156 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
157 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
159 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
161 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
163 if (ilock_flags & BTRFS_ILOCK_SHARED) {
164 if (ilock_flags & BTRFS_ILOCK_TRY) {
165 if (!inode_trylock_shared(&inode->vfs_inode))
170 inode_lock_shared(&inode->vfs_inode);
172 if (ilock_flags & BTRFS_ILOCK_TRY) {
173 if (!inode_trylock(&inode->vfs_inode))
178 inode_lock(&inode->vfs_inode);
180 if (ilock_flags & BTRFS_ILOCK_MMAP)
181 down_write(&inode->i_mmap_lock);
186 * btrfs_inode_unlock - unock inode i_rwsem
188 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
189 * to decide whether the lock acquired is shared or exclusive.
191 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
193 if (ilock_flags & BTRFS_ILOCK_MMAP)
194 up_write(&inode->i_mmap_lock);
195 if (ilock_flags & BTRFS_ILOCK_SHARED)
196 inode_unlock_shared(&inode->vfs_inode);
198 inode_unlock(&inode->vfs_inode);
202 * Cleanup all submitted ordered extents in specified range to handle errors
203 * from the btrfs_run_delalloc_range() callback.
205 * NOTE: caller must ensure that when an error happens, it can not call
206 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
207 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
208 * to be released, which we want to happen only when finishing the ordered
209 * extent (btrfs_finish_ordered_io()).
211 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
212 struct page *locked_page,
213 u64 offset, u64 bytes)
215 unsigned long index = offset >> PAGE_SHIFT;
216 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
217 u64 page_start = 0, page_end = 0;
221 page_start = page_offset(locked_page);
222 page_end = page_start + PAGE_SIZE - 1;
225 while (index <= end_index) {
227 * For locked page, we will call end_extent_writepage() on it
228 * in run_delalloc_range() for the error handling. That
229 * end_extent_writepage() function will call
230 * btrfs_mark_ordered_io_finished() to clear page Ordered and
231 * run the ordered extent accounting.
233 * Here we can't just clear the Ordered bit, or
234 * btrfs_mark_ordered_io_finished() would skip the accounting
235 * for the page range, and the ordered extent will never finish.
237 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
241 page = find_get_page(inode->vfs_inode.i_mapping, index);
247 * Here we just clear all Ordered bits for every page in the
248 * range, then btrfs_mark_ordered_io_finished() will handle
249 * the ordered extent accounting for the range.
251 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
257 /* The locked page covers the full range, nothing needs to be done */
258 if (bytes + offset <= page_start + PAGE_SIZE)
261 * In case this page belongs to the delalloc range being
262 * instantiated then skip it, since the first page of a range is
263 * going to be properly cleaned up by the caller of
266 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
267 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
268 offset = page_offset(locked_page) + PAGE_SIZE;
272 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
275 static int btrfs_dirty_inode(struct btrfs_inode *inode);
277 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
278 struct btrfs_new_inode_args *args)
282 if (args->default_acl) {
283 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
289 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
293 if (!args->default_acl && !args->acl)
294 cache_no_acl(args->inode);
295 return btrfs_xattr_security_init(trans, args->inode, args->dir,
296 &args->dentry->d_name);
300 * this does all the hard work for inserting an inline extent into
301 * the btree. The caller should have done a btrfs_drop_extents so that
302 * no overlapping inline items exist in the btree
304 static int insert_inline_extent(struct btrfs_trans_handle *trans,
305 struct btrfs_path *path,
306 struct btrfs_inode *inode, bool extent_inserted,
307 size_t size, size_t compressed_size,
309 struct page **compressed_pages,
312 struct btrfs_root *root = inode->root;
313 struct extent_buffer *leaf;
314 struct page *page = NULL;
317 struct btrfs_file_extent_item *ei;
319 size_t cur_size = size;
322 ASSERT((compressed_size > 0 && compressed_pages) ||
323 (compressed_size == 0 && !compressed_pages));
325 if (compressed_size && compressed_pages)
326 cur_size = compressed_size;
328 if (!extent_inserted) {
329 struct btrfs_key key;
332 key.objectid = btrfs_ino(inode);
334 key.type = BTRFS_EXTENT_DATA_KEY;
336 datasize = btrfs_file_extent_calc_inline_size(cur_size);
337 ret = btrfs_insert_empty_item(trans, root, path, &key,
342 leaf = path->nodes[0];
343 ei = btrfs_item_ptr(leaf, path->slots[0],
344 struct btrfs_file_extent_item);
345 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
346 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
347 btrfs_set_file_extent_encryption(leaf, ei, 0);
348 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
349 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
350 ptr = btrfs_file_extent_inline_start(ei);
352 if (compress_type != BTRFS_COMPRESS_NONE) {
355 while (compressed_size > 0) {
356 cpage = compressed_pages[i];
357 cur_size = min_t(unsigned long, compressed_size,
360 kaddr = kmap_local_page(cpage);
361 write_extent_buffer(leaf, kaddr, ptr, cur_size);
366 compressed_size -= cur_size;
368 btrfs_set_file_extent_compression(leaf, ei,
371 page = find_get_page(inode->vfs_inode.i_mapping, 0);
372 btrfs_set_file_extent_compression(leaf, ei, 0);
373 kaddr = kmap_local_page(page);
374 write_extent_buffer(leaf, kaddr, ptr, size);
378 btrfs_mark_buffer_dirty(leaf);
379 btrfs_release_path(path);
382 * We align size to sectorsize for inline extents just for simplicity
385 ret = btrfs_inode_set_file_extent_range(inode, 0,
386 ALIGN(size, root->fs_info->sectorsize));
391 * We're an inline extent, so nobody can extend the file past i_size
392 * without locking a page we already have locked.
394 * We must do any i_size and inode updates before we unlock the pages.
395 * Otherwise we could end up racing with unlink.
397 i_size = i_size_read(&inode->vfs_inode);
398 if (update_i_size && size > i_size) {
399 i_size_write(&inode->vfs_inode, size);
402 inode->disk_i_size = i_size;
410 * conditionally insert an inline extent into the file. This
411 * does the checks required to make sure the data is small enough
412 * to fit as an inline extent.
414 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
415 size_t compressed_size,
417 struct page **compressed_pages,
420 struct btrfs_drop_extents_args drop_args = { 0 };
421 struct btrfs_root *root = inode->root;
422 struct btrfs_fs_info *fs_info = root->fs_info;
423 struct btrfs_trans_handle *trans;
424 u64 data_len = (compressed_size ?: size);
426 struct btrfs_path *path;
429 * We can create an inline extent if it ends at or beyond the current
430 * i_size, is no larger than a sector (decompressed), and the (possibly
431 * compressed) data fits in a leaf and the configured maximum inline
434 if (size < i_size_read(&inode->vfs_inode) ||
435 size > fs_info->sectorsize ||
436 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
437 data_len > fs_info->max_inline)
440 path = btrfs_alloc_path();
444 trans = btrfs_join_transaction(root);
446 btrfs_free_path(path);
447 return PTR_ERR(trans);
449 trans->block_rsv = &inode->block_rsv;
451 drop_args.path = path;
453 drop_args.end = fs_info->sectorsize;
454 drop_args.drop_cache = true;
455 drop_args.replace_extent = true;
456 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
457 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
459 btrfs_abort_transaction(trans, ret);
463 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
464 size, compressed_size, compress_type,
465 compressed_pages, update_i_size);
466 if (ret && ret != -ENOSPC) {
467 btrfs_abort_transaction(trans, ret);
469 } else if (ret == -ENOSPC) {
474 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
475 ret = btrfs_update_inode(trans, root, inode);
476 if (ret && ret != -ENOSPC) {
477 btrfs_abort_transaction(trans, ret);
479 } else if (ret == -ENOSPC) {
484 btrfs_set_inode_full_sync(inode);
487 * Don't forget to free the reserved space, as for inlined extent
488 * it won't count as data extent, free them directly here.
489 * And at reserve time, it's always aligned to page size, so
490 * just free one page here.
492 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
493 btrfs_free_path(path);
494 btrfs_end_transaction(trans);
498 struct async_extent {
503 unsigned long nr_pages;
505 struct list_head list;
509 struct btrfs_inode *inode;
510 struct page *locked_page;
513 blk_opf_t write_flags;
514 struct list_head extents;
515 struct cgroup_subsys_state *blkcg_css;
516 struct btrfs_work work;
517 struct async_cow *async_cow;
522 struct async_chunk chunks[];
525 static noinline int add_async_extent(struct async_chunk *cow,
526 u64 start, u64 ram_size,
529 unsigned long nr_pages,
532 struct async_extent *async_extent;
534 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
535 BUG_ON(!async_extent); /* -ENOMEM */
536 async_extent->start = start;
537 async_extent->ram_size = ram_size;
538 async_extent->compressed_size = compressed_size;
539 async_extent->pages = pages;
540 async_extent->nr_pages = nr_pages;
541 async_extent->compress_type = compress_type;
542 list_add_tail(&async_extent->list, &cow->extents);
547 * Check if the inode needs to be submitted to compression, based on mount
548 * options, defragmentation, properties or heuristics.
550 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
553 struct btrfs_fs_info *fs_info = inode->root->fs_info;
555 if (!btrfs_inode_can_compress(inode)) {
556 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
557 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
562 * Special check for subpage.
564 * We lock the full page then run each delalloc range in the page, thus
565 * for the following case, we will hit some subpage specific corner case:
568 * | |///////| |///////|
571 * In above case, both range A and range B will try to unlock the full
572 * page [0, 64K), causing the one finished later will have page
573 * unlocked already, triggering various page lock requirement BUG_ON()s.
575 * So here we add an artificial limit that subpage compression can only
576 * if the range is fully page aligned.
578 * In theory we only need to ensure the first page is fully covered, but
579 * the tailing partial page will be locked until the full compression
580 * finishes, delaying the write of other range.
582 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
583 * first to prevent any submitted async extent to unlock the full page.
584 * By this, we can ensure for subpage case that only the last async_cow
585 * will unlock the full page.
587 if (fs_info->sectorsize < PAGE_SIZE) {
588 if (!PAGE_ALIGNED(start) ||
589 !PAGE_ALIGNED(end + 1))
594 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
597 if (inode->defrag_compress)
599 /* bad compression ratios */
600 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
602 if (btrfs_test_opt(fs_info, COMPRESS) ||
603 inode->flags & BTRFS_INODE_COMPRESS ||
604 inode->prop_compress)
605 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
609 static inline void inode_should_defrag(struct btrfs_inode *inode,
610 u64 start, u64 end, u64 num_bytes, u32 small_write)
612 /* If this is a small write inside eof, kick off a defrag */
613 if (num_bytes < small_write &&
614 (start > 0 || end + 1 < inode->disk_i_size))
615 btrfs_add_inode_defrag(NULL, inode, small_write);
619 * we create compressed extents in two phases. The first
620 * phase compresses a range of pages that have already been
621 * locked (both pages and state bits are locked).
623 * This is done inside an ordered work queue, and the compression
624 * is spread across many cpus. The actual IO submission is step
625 * two, and the ordered work queue takes care of making sure that
626 * happens in the same order things were put onto the queue by
627 * writepages and friends.
629 * If this code finds it can't get good compression, it puts an
630 * entry onto the work queue to write the uncompressed bytes. This
631 * makes sure that both compressed inodes and uncompressed inodes
632 * are written in the same order that the flusher thread sent them
635 static noinline int compress_file_range(struct async_chunk *async_chunk)
637 struct btrfs_inode *inode = async_chunk->inode;
638 struct btrfs_fs_info *fs_info = inode->root->fs_info;
639 u64 blocksize = fs_info->sectorsize;
640 u64 start = async_chunk->start;
641 u64 end = async_chunk->end;
645 struct page **pages = NULL;
646 unsigned long nr_pages;
647 unsigned long total_compressed = 0;
648 unsigned long total_in = 0;
651 int compress_type = fs_info->compress_type;
652 int compressed_extents = 0;
655 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
658 * We need to save i_size before now because it could change in between
659 * us evaluating the size and assigning it. This is because we lock and
660 * unlock the page in truncate and fallocate, and then modify the i_size
663 * The barriers are to emulate READ_ONCE, remove that once i_size_read
667 i_size = i_size_read(&inode->vfs_inode);
669 actual_end = min_t(u64, i_size, end + 1);
672 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
673 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
676 * we don't want to send crud past the end of i_size through
677 * compression, that's just a waste of CPU time. So, if the
678 * end of the file is before the start of our current
679 * requested range of bytes, we bail out to the uncompressed
680 * cleanup code that can deal with all of this.
682 * It isn't really the fastest way to fix things, but this is a
683 * very uncommon corner.
685 if (actual_end <= start)
686 goto cleanup_and_bail_uncompressed;
688 total_compressed = actual_end - start;
691 * Skip compression for a small file range(<=blocksize) that
692 * isn't an inline extent, since it doesn't save disk space at all.
694 if (total_compressed <= blocksize &&
695 (start > 0 || end + 1 < inode->disk_i_size))
696 goto cleanup_and_bail_uncompressed;
699 * For subpage case, we require full page alignment for the sector
701 * Thus we must also check against @actual_end, not just @end.
703 if (blocksize < PAGE_SIZE) {
704 if (!PAGE_ALIGNED(start) ||
705 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
706 goto cleanup_and_bail_uncompressed;
709 total_compressed = min_t(unsigned long, total_compressed,
710 BTRFS_MAX_UNCOMPRESSED);
715 * we do compression for mount -o compress and when the
716 * inode has not been flagged as nocompress. This flag can
717 * change at any time if we discover bad compression ratios.
719 if (inode_need_compress(inode, start, end)) {
721 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
723 /* just bail out to the uncompressed code */
728 if (inode->defrag_compress)
729 compress_type = inode->defrag_compress;
730 else if (inode->prop_compress)
731 compress_type = inode->prop_compress;
734 * we need to call clear_page_dirty_for_io on each
735 * page in the range. Otherwise applications with the file
736 * mmap'd can wander in and change the page contents while
737 * we are compressing them.
739 * If the compression fails for any reason, we set the pages
740 * dirty again later on.
742 * Note that the remaining part is redirtied, the start pointer
743 * has moved, the end is the original one.
746 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
750 /* Compression level is applied here and only here */
751 ret = btrfs_compress_pages(
752 compress_type | (fs_info->compress_level << 4),
753 inode->vfs_inode.i_mapping, start,
760 unsigned long offset = offset_in_page(total_compressed);
761 struct page *page = pages[nr_pages - 1];
763 /* zero the tail end of the last page, we might be
764 * sending it down to disk
767 memzero_page(page, offset, PAGE_SIZE - offset);
773 * Check cow_file_range() for why we don't even try to create inline
774 * extent for subpage case.
776 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
777 /* lets try to make an inline extent */
778 if (ret || total_in < actual_end) {
779 /* we didn't compress the entire range, try
780 * to make an uncompressed inline extent.
782 ret = cow_file_range_inline(inode, actual_end,
783 0, BTRFS_COMPRESS_NONE,
786 /* try making a compressed inline extent */
787 ret = cow_file_range_inline(inode, actual_end,
789 compress_type, pages,
793 unsigned long clear_flags = EXTENT_DELALLOC |
794 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
795 EXTENT_DO_ACCOUNTING;
796 unsigned long page_error_op;
798 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
801 * inline extent creation worked or returned error,
802 * we don't need to create any more async work items.
803 * Unlock and free up our temp pages.
805 * We use DO_ACCOUNTING here because we need the
806 * delalloc_release_metadata to be done _after_ we drop
807 * our outstanding extent for clearing delalloc for this
810 extent_clear_unlock_delalloc(inode, start, end,
814 PAGE_START_WRITEBACK |
819 * Ensure we only free the compressed pages if we have
820 * them allocated, as we can still reach here with
821 * inode_need_compress() == false.
824 for (i = 0; i < nr_pages; i++) {
825 WARN_ON(pages[i]->mapping);
836 * we aren't doing an inline extent round the compressed size
837 * up to a block size boundary so the allocator does sane
840 total_compressed = ALIGN(total_compressed, blocksize);
843 * one last check to make sure the compression is really a
844 * win, compare the page count read with the blocks on disk,
845 * compression must free at least one sector size
847 total_in = round_up(total_in, fs_info->sectorsize);
848 if (total_compressed + blocksize <= total_in) {
849 compressed_extents++;
852 * The async work queues will take care of doing actual
853 * allocation on disk for these compressed pages, and
854 * will submit them to the elevator.
856 add_async_extent(async_chunk, start, total_in,
857 total_compressed, pages, nr_pages,
860 if (start + total_in < end) {
866 return compressed_extents;
871 * the compression code ran but failed to make things smaller,
872 * free any pages it allocated and our page pointer array
874 for (i = 0; i < nr_pages; i++) {
875 WARN_ON(pages[i]->mapping);
880 total_compressed = 0;
883 /* flag the file so we don't compress in the future */
884 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
885 !(inode->prop_compress)) {
886 inode->flags |= BTRFS_INODE_NOCOMPRESS;
889 cleanup_and_bail_uncompressed:
891 * No compression, but we still need to write the pages in the file
892 * we've been given so far. redirty the locked page if it corresponds
893 * to our extent and set things up for the async work queue to run
894 * cow_file_range to do the normal delalloc dance.
896 if (async_chunk->locked_page &&
897 (page_offset(async_chunk->locked_page) >= start &&
898 page_offset(async_chunk->locked_page)) <= end) {
899 __set_page_dirty_nobuffers(async_chunk->locked_page);
900 /* unlocked later on in the async handlers */
904 extent_range_redirty_for_io(&inode->vfs_inode, start, end);
905 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
906 BTRFS_COMPRESS_NONE);
907 compressed_extents++;
909 return compressed_extents;
912 static void free_async_extent_pages(struct async_extent *async_extent)
916 if (!async_extent->pages)
919 for (i = 0; i < async_extent->nr_pages; i++) {
920 WARN_ON(async_extent->pages[i]->mapping);
921 put_page(async_extent->pages[i]);
923 kfree(async_extent->pages);
924 async_extent->nr_pages = 0;
925 async_extent->pages = NULL;
928 static int submit_uncompressed_range(struct btrfs_inode *inode,
929 struct async_extent *async_extent,
930 struct page *locked_page)
932 u64 start = async_extent->start;
933 u64 end = async_extent->start + async_extent->ram_size - 1;
934 unsigned long nr_written = 0;
935 int page_started = 0;
939 * Call cow_file_range() to run the delalloc range directly, since we
940 * won't go to NOCOW or async path again.
942 * Also we call cow_file_range() with @unlock_page == 0, so that we
943 * can directly submit them without interruption.
945 ret = cow_file_range(inode, locked_page, start, end, &page_started,
946 &nr_written, 0, NULL);
947 /* Inline extent inserted, page gets unlocked and everything is done */
952 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
954 const u64 page_start = page_offset(locked_page);
955 const u64 page_end = page_start + PAGE_SIZE - 1;
957 btrfs_page_set_error(inode->root->fs_info, locked_page,
958 page_start, PAGE_SIZE);
959 set_page_writeback(locked_page);
960 end_page_writeback(locked_page);
961 end_extent_writepage(locked_page, ret, page_start, page_end);
962 unlock_page(locked_page);
967 /* All pages will be unlocked, including @locked_page */
968 return extent_write_locked_range(&inode->vfs_inode, start, end);
971 static int submit_one_async_extent(struct btrfs_inode *inode,
972 struct async_chunk *async_chunk,
973 struct async_extent *async_extent,
976 struct extent_io_tree *io_tree = &inode->io_tree;
977 struct btrfs_root *root = inode->root;
978 struct btrfs_fs_info *fs_info = root->fs_info;
979 struct btrfs_key ins;
980 struct page *locked_page = NULL;
981 struct extent_map *em;
983 u64 start = async_extent->start;
984 u64 end = async_extent->start + async_extent->ram_size - 1;
986 if (async_chunk->blkcg_css)
987 kthread_associate_blkcg(async_chunk->blkcg_css);
990 * If async_chunk->locked_page is in the async_extent range, we need to
993 if (async_chunk->locked_page) {
994 u64 locked_page_start = page_offset(async_chunk->locked_page);
995 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
997 if (!(start >= locked_page_end || end <= locked_page_start))
998 locked_page = async_chunk->locked_page;
1000 lock_extent(io_tree, start, end, NULL);
1002 /* We have fall back to uncompressed write */
1003 if (!async_extent->pages) {
1004 ret = submit_uncompressed_range(inode, async_extent, locked_page);
1008 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1009 async_extent->compressed_size,
1010 async_extent->compressed_size,
1011 0, *alloc_hint, &ins, 1, 1);
1013 free_async_extent_pages(async_extent);
1015 * Here we used to try again by going back to non-compressed
1016 * path for ENOSPC. But we can't reserve space even for
1017 * compressed size, how could it work for uncompressed size
1018 * which requires larger size? So here we directly go error
1024 /* Here we're doing allocation and writeback of the compressed pages */
1025 em = create_io_em(inode, start,
1026 async_extent->ram_size, /* len */
1027 start, /* orig_start */
1028 ins.objectid, /* block_start */
1029 ins.offset, /* block_len */
1030 ins.offset, /* orig_block_len */
1031 async_extent->ram_size, /* ram_bytes */
1032 async_extent->compress_type,
1033 BTRFS_ORDERED_COMPRESSED);
1036 goto out_free_reserve;
1038 free_extent_map(em);
1040 ret = btrfs_add_ordered_extent(inode, start, /* file_offset */
1041 async_extent->ram_size, /* num_bytes */
1042 async_extent->ram_size, /* ram_bytes */
1043 ins.objectid, /* disk_bytenr */
1044 ins.offset, /* disk_num_bytes */
1046 1 << BTRFS_ORDERED_COMPRESSED,
1047 async_extent->compress_type);
1049 btrfs_drop_extent_map_range(inode, start, end, false);
1050 goto out_free_reserve;
1052 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1054 /* Clear dirty, set writeback and unlock the pages. */
1055 extent_clear_unlock_delalloc(inode, start, end,
1056 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1057 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1059 btrfs_submit_compressed_write(inode, start, /* file_offset */
1060 async_extent->ram_size, /* num_bytes */
1061 ins.objectid, /* disk_bytenr */
1062 ins.offset, /* compressed_len */
1063 async_extent->pages, /* compressed_pages */
1064 async_extent->nr_pages,
1065 async_chunk->write_flags, true);
1066 *alloc_hint = ins.objectid + ins.offset;
1068 if (async_chunk->blkcg_css)
1069 kthread_associate_blkcg(NULL);
1070 kfree(async_extent);
1074 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1075 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1077 extent_clear_unlock_delalloc(inode, start, end,
1078 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1079 EXTENT_DELALLOC_NEW |
1080 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1081 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1082 PAGE_END_WRITEBACK | PAGE_SET_ERROR);
1083 free_async_extent_pages(async_extent);
1088 * Phase two of compressed writeback. This is the ordered portion of the code,
1089 * which only gets called in the order the work was queued. We walk all the
1090 * async extents created by compress_file_range and send them down to the disk.
1092 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1094 struct btrfs_inode *inode = async_chunk->inode;
1095 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1096 struct async_extent *async_extent;
1100 while (!list_empty(&async_chunk->extents)) {
1104 async_extent = list_entry(async_chunk->extents.next,
1105 struct async_extent, list);
1106 list_del(&async_extent->list);
1107 extent_start = async_extent->start;
1108 ram_size = async_extent->ram_size;
1110 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1112 btrfs_debug(fs_info,
1113 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1114 inode->root->root_key.objectid,
1115 btrfs_ino(inode), extent_start, ram_size, ret);
1119 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1122 struct extent_map_tree *em_tree = &inode->extent_tree;
1123 struct extent_map *em;
1126 read_lock(&em_tree->lock);
1127 em = search_extent_mapping(em_tree, start, num_bytes);
1130 * if block start isn't an actual block number then find the
1131 * first block in this inode and use that as a hint. If that
1132 * block is also bogus then just don't worry about it.
1134 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1135 free_extent_map(em);
1136 em = search_extent_mapping(em_tree, 0, 0);
1137 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1138 alloc_hint = em->block_start;
1140 free_extent_map(em);
1142 alloc_hint = em->block_start;
1143 free_extent_map(em);
1146 read_unlock(&em_tree->lock);
1152 * when extent_io.c finds a delayed allocation range in the file,
1153 * the call backs end up in this code. The basic idea is to
1154 * allocate extents on disk for the range, and create ordered data structs
1155 * in ram to track those extents.
1157 * locked_page is the page that writepage had locked already. We use
1158 * it to make sure we don't do extra locks or unlocks.
1160 * *page_started is set to one if we unlock locked_page and do everything
1161 * required to start IO on it. It may be clean and already done with
1162 * IO when we return.
1164 * When unlock == 1, we unlock the pages in successfully allocated regions.
1165 * When unlock == 0, we leave them locked for writing them out.
1167 * However, we unlock all the pages except @locked_page in case of failure.
1169 * In summary, page locking state will be as follow:
1171 * - page_started == 1 (return value)
1172 * - All the pages are unlocked. IO is started.
1173 * - Note that this can happen only on success
1175 * - All the pages except @locked_page are unlocked in any case
1177 * - On success, all the pages are locked for writing out them
1178 * - On failure, all the pages except @locked_page are unlocked
1180 * When a failure happens in the second or later iteration of the
1181 * while-loop, the ordered extents created in previous iterations are kept
1182 * intact. So, the caller must clean them up by calling
1183 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1186 static noinline int cow_file_range(struct btrfs_inode *inode,
1187 struct page *locked_page,
1188 u64 start, u64 end, int *page_started,
1189 unsigned long *nr_written, int unlock,
1192 struct btrfs_root *root = inode->root;
1193 struct btrfs_fs_info *fs_info = root->fs_info;
1195 u64 orig_start = start;
1197 unsigned long ram_size;
1198 u64 cur_alloc_size = 0;
1200 u64 blocksize = fs_info->sectorsize;
1201 struct btrfs_key ins;
1202 struct extent_map *em;
1203 unsigned clear_bits;
1204 unsigned long page_ops;
1205 bool extent_reserved = false;
1208 if (btrfs_is_free_space_inode(inode)) {
1213 num_bytes = ALIGN(end - start + 1, blocksize);
1214 num_bytes = max(blocksize, num_bytes);
1215 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1217 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1220 * Due to the page size limit, for subpage we can only trigger the
1221 * writeback for the dirty sectors of page, that means data writeback
1222 * is doing more writeback than what we want.
1224 * This is especially unexpected for some call sites like fallocate,
1225 * where we only increase i_size after everything is done.
1226 * This means we can trigger inline extent even if we didn't want to.
1227 * So here we skip inline extent creation completely.
1229 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
1230 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1233 /* lets try to make an inline extent */
1234 ret = cow_file_range_inline(inode, actual_end, 0,
1235 BTRFS_COMPRESS_NONE, NULL, false);
1238 * We use DO_ACCOUNTING here because we need the
1239 * delalloc_release_metadata to be run _after_ we drop
1240 * our outstanding extent for clearing delalloc for this
1243 extent_clear_unlock_delalloc(inode, start, end,
1245 EXTENT_LOCKED | EXTENT_DELALLOC |
1246 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1247 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1248 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1249 *nr_written = *nr_written +
1250 (end - start + PAGE_SIZE) / PAGE_SIZE;
1253 * locked_page is locked by the caller of
1254 * writepage_delalloc(), not locked by
1255 * __process_pages_contig().
1257 * We can't let __process_pages_contig() to unlock it,
1258 * as it doesn't have any subpage::writers recorded.
1260 * Here we manually unlock the page, since the caller
1261 * can't use page_started to determine if it's an
1262 * inline extent or a compressed extent.
1264 unlock_page(locked_page);
1266 } else if (ret < 0) {
1271 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1274 * Relocation relies on the relocated extents to have exactly the same
1275 * size as the original extents. Normally writeback for relocation data
1276 * extents follows a NOCOW path because relocation preallocates the
1277 * extents. However, due to an operation such as scrub turning a block
1278 * group to RO mode, it may fallback to COW mode, so we must make sure
1279 * an extent allocated during COW has exactly the requested size and can
1280 * not be split into smaller extents, otherwise relocation breaks and
1281 * fails during the stage where it updates the bytenr of file extent
1284 if (btrfs_is_data_reloc_root(root))
1285 min_alloc_size = num_bytes;
1287 min_alloc_size = fs_info->sectorsize;
1289 while (num_bytes > 0) {
1290 cur_alloc_size = num_bytes;
1291 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1292 min_alloc_size, 0, alloc_hint,
1296 cur_alloc_size = ins.offset;
1297 extent_reserved = true;
1299 ram_size = ins.offset;
1300 em = create_io_em(inode, start, ins.offset, /* len */
1301 start, /* orig_start */
1302 ins.objectid, /* block_start */
1303 ins.offset, /* block_len */
1304 ins.offset, /* orig_block_len */
1305 ram_size, /* ram_bytes */
1306 BTRFS_COMPRESS_NONE, /* compress_type */
1307 BTRFS_ORDERED_REGULAR /* type */);
1312 free_extent_map(em);
1314 ret = btrfs_add_ordered_extent(inode, start, ram_size, ram_size,
1315 ins.objectid, cur_alloc_size, 0,
1316 1 << BTRFS_ORDERED_REGULAR,
1317 BTRFS_COMPRESS_NONE);
1319 goto out_drop_extent_cache;
1321 if (btrfs_is_data_reloc_root(root)) {
1322 ret = btrfs_reloc_clone_csums(inode, start,
1325 * Only drop cache here, and process as normal.
1327 * We must not allow extent_clear_unlock_delalloc()
1328 * at out_unlock label to free meta of this ordered
1329 * extent, as its meta should be freed by
1330 * btrfs_finish_ordered_io().
1332 * So we must continue until @start is increased to
1333 * skip current ordered extent.
1336 btrfs_drop_extent_map_range(inode, start,
1337 start + ram_size - 1,
1341 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1344 * We're not doing compressed IO, don't unlock the first page
1345 * (which the caller expects to stay locked), don't clear any
1346 * dirty bits and don't set any writeback bits
1348 * Do set the Ordered (Private2) bit so we know this page was
1349 * properly setup for writepage.
1351 page_ops = unlock ? PAGE_UNLOCK : 0;
1352 page_ops |= PAGE_SET_ORDERED;
1354 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1356 EXTENT_LOCKED | EXTENT_DELALLOC,
1358 if (num_bytes < cur_alloc_size)
1361 num_bytes -= cur_alloc_size;
1362 alloc_hint = ins.objectid + ins.offset;
1363 start += cur_alloc_size;
1364 extent_reserved = false;
1367 * btrfs_reloc_clone_csums() error, since start is increased
1368 * extent_clear_unlock_delalloc() at out_unlock label won't
1369 * free metadata of current ordered extent, we're OK to exit.
1377 out_drop_extent_cache:
1378 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1380 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1381 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1384 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1385 * caller to write out the successfully allocated region and retry.
1387 if (done_offset && ret == -EAGAIN) {
1388 if (orig_start < start)
1389 *done_offset = start - 1;
1391 *done_offset = start;
1393 } else if (ret == -EAGAIN) {
1394 /* Convert to -ENOSPC since the caller cannot retry. */
1399 * Now, we have three regions to clean up:
1401 * |-------(1)----|---(2)---|-------------(3)----------|
1402 * `- orig_start `- start `- start + cur_alloc_size `- end
1404 * We process each region below.
1407 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1408 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1409 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1412 * For the range (1). We have already instantiated the ordered extents
1413 * for this region. They are cleaned up by
1414 * btrfs_cleanup_ordered_extents() in e.g,
1415 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1416 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1417 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1420 * However, in case of unlock == 0, we still need to unlock the pages
1421 * (except @locked_page) to ensure all the pages are unlocked.
1423 if (!unlock && orig_start < start) {
1425 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1426 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1427 locked_page, 0, page_ops);
1431 * For the range (2). If we reserved an extent for our delalloc range
1432 * (or a subrange) and failed to create the respective ordered extent,
1433 * then it means that when we reserved the extent we decremented the
1434 * extent's size from the data space_info's bytes_may_use counter and
1435 * incremented the space_info's bytes_reserved counter by the same
1436 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1437 * to decrement again the data space_info's bytes_may_use counter,
1438 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1440 if (extent_reserved) {
1441 extent_clear_unlock_delalloc(inode, start,
1442 start + cur_alloc_size - 1,
1446 start += cur_alloc_size;
1452 * For the range (3). We never touched the region. In addition to the
1453 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1454 * space_info's bytes_may_use counter, reserved in
1455 * btrfs_check_data_free_space().
1457 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1458 clear_bits | EXTENT_CLEAR_DATA_RESV,
1464 * work queue call back to started compression on a file and pages
1466 static noinline void async_cow_start(struct btrfs_work *work)
1468 struct async_chunk *async_chunk;
1469 int compressed_extents;
1471 async_chunk = container_of(work, struct async_chunk, work);
1473 compressed_extents = compress_file_range(async_chunk);
1474 if (compressed_extents == 0) {
1475 btrfs_add_delayed_iput(async_chunk->inode);
1476 async_chunk->inode = NULL;
1481 * work queue call back to submit previously compressed pages
1483 static noinline void async_cow_submit(struct btrfs_work *work)
1485 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1487 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1488 unsigned long nr_pages;
1490 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1494 * ->inode could be NULL if async_chunk_start has failed to compress,
1495 * in which case we don't have anything to submit, yet we need to
1496 * always adjust ->async_delalloc_pages as its paired with the init
1497 * happening in cow_file_range_async
1499 if (async_chunk->inode)
1500 submit_compressed_extents(async_chunk);
1502 /* atomic_sub_return implies a barrier */
1503 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1505 cond_wake_up_nomb(&fs_info->async_submit_wait);
1508 static noinline void async_cow_free(struct btrfs_work *work)
1510 struct async_chunk *async_chunk;
1511 struct async_cow *async_cow;
1513 async_chunk = container_of(work, struct async_chunk, work);
1514 if (async_chunk->inode)
1515 btrfs_add_delayed_iput(async_chunk->inode);
1516 if (async_chunk->blkcg_css)
1517 css_put(async_chunk->blkcg_css);
1519 async_cow = async_chunk->async_cow;
1520 if (atomic_dec_and_test(&async_cow->num_chunks))
1524 static int cow_file_range_async(struct btrfs_inode *inode,
1525 struct writeback_control *wbc,
1526 struct page *locked_page,
1527 u64 start, u64 end, int *page_started,
1528 unsigned long *nr_written)
1530 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1531 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1532 struct async_cow *ctx;
1533 struct async_chunk *async_chunk;
1534 unsigned long nr_pages;
1536 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1538 bool should_compress;
1540 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1542 unlock_extent(&inode->io_tree, start, end, NULL);
1544 if (inode->flags & BTRFS_INODE_NOCOMPRESS &&
1545 !btrfs_test_opt(fs_info, FORCE_COMPRESS)) {
1547 should_compress = false;
1549 should_compress = true;
1552 nofs_flag = memalloc_nofs_save();
1553 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1554 memalloc_nofs_restore(nofs_flag);
1557 unsigned clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC |
1558 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1559 EXTENT_DO_ACCOUNTING;
1560 unsigned long page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK |
1561 PAGE_END_WRITEBACK | PAGE_SET_ERROR;
1563 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1564 clear_bits, page_ops);
1568 async_chunk = ctx->chunks;
1569 atomic_set(&ctx->num_chunks, num_chunks);
1571 for (i = 0; i < num_chunks; i++) {
1572 if (should_compress)
1573 cur_end = min(end, start + SZ_512K - 1);
1578 * igrab is called higher up in the call chain, take only the
1579 * lightweight reference for the callback lifetime
1581 ihold(&inode->vfs_inode);
1582 async_chunk[i].async_cow = ctx;
1583 async_chunk[i].inode = inode;
1584 async_chunk[i].start = start;
1585 async_chunk[i].end = cur_end;
1586 async_chunk[i].write_flags = write_flags;
1587 INIT_LIST_HEAD(&async_chunk[i].extents);
1590 * The locked_page comes all the way from writepage and its
1591 * the original page we were actually given. As we spread
1592 * this large delalloc region across multiple async_chunk
1593 * structs, only the first struct needs a pointer to locked_page
1595 * This way we don't need racey decisions about who is supposed
1600 * Depending on the compressibility, the pages might or
1601 * might not go through async. We want all of them to
1602 * be accounted against wbc once. Let's do it here
1603 * before the paths diverge. wbc accounting is used
1604 * only for foreign writeback detection and doesn't
1605 * need full accuracy. Just account the whole thing
1606 * against the first page.
1608 wbc_account_cgroup_owner(wbc, locked_page,
1610 async_chunk[i].locked_page = locked_page;
1613 async_chunk[i].locked_page = NULL;
1616 if (blkcg_css != blkcg_root_css) {
1618 async_chunk[i].blkcg_css = blkcg_css;
1619 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1621 async_chunk[i].blkcg_css = NULL;
1624 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1625 async_cow_submit, async_cow_free);
1627 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1628 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1630 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1632 *nr_written += nr_pages;
1633 start = cur_end + 1;
1639 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1640 struct page *locked_page, u64 start,
1641 u64 end, int *page_started,
1642 unsigned long *nr_written)
1644 u64 done_offset = end;
1646 bool locked_page_done = false;
1648 while (start <= end) {
1649 ret = cow_file_range(inode, locked_page, start, end, page_started,
1650 nr_written, 0, &done_offset);
1651 if (ret && ret != -EAGAIN)
1654 if (*page_started) {
1662 if (done_offset == start) {
1663 wait_on_bit_io(&inode->root->fs_info->flags,
1664 BTRFS_FS_NEED_ZONE_FINISH,
1665 TASK_UNINTERRUPTIBLE);
1669 if (!locked_page_done) {
1670 __set_page_dirty_nobuffers(locked_page);
1671 account_page_redirty(locked_page);
1673 locked_page_done = true;
1674 extent_write_locked_range(&inode->vfs_inode, start, done_offset);
1676 start = done_offset + 1;
1684 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1685 u64 bytenr, u64 num_bytes, bool nowait)
1687 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1688 struct btrfs_ordered_sum *sums;
1692 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1694 if (ret == 0 && list_empty(&list))
1697 while (!list_empty(&list)) {
1698 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1699 list_del(&sums->list);
1707 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1708 const u64 start, const u64 end,
1709 int *page_started, unsigned long *nr_written)
1711 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1712 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1713 const u64 range_bytes = end + 1 - start;
1714 struct extent_io_tree *io_tree = &inode->io_tree;
1715 u64 range_start = start;
1719 * If EXTENT_NORESERVE is set it means that when the buffered write was
1720 * made we had not enough available data space and therefore we did not
1721 * reserve data space for it, since we though we could do NOCOW for the
1722 * respective file range (either there is prealloc extent or the inode
1723 * has the NOCOW bit set).
1725 * However when we need to fallback to COW mode (because for example the
1726 * block group for the corresponding extent was turned to RO mode by a
1727 * scrub or relocation) we need to do the following:
1729 * 1) We increment the bytes_may_use counter of the data space info.
1730 * If COW succeeds, it allocates a new data extent and after doing
1731 * that it decrements the space info's bytes_may_use counter and
1732 * increments its bytes_reserved counter by the same amount (we do
1733 * this at btrfs_add_reserved_bytes()). So we need to increment the
1734 * bytes_may_use counter to compensate (when space is reserved at
1735 * buffered write time, the bytes_may_use counter is incremented);
1737 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1738 * that if the COW path fails for any reason, it decrements (through
1739 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1740 * data space info, which we incremented in the step above.
1742 * If we need to fallback to cow and the inode corresponds to a free
1743 * space cache inode or an inode of the data relocation tree, we must
1744 * also increment bytes_may_use of the data space_info for the same
1745 * reason. Space caches and relocated data extents always get a prealloc
1746 * extent for them, however scrub or balance may have set the block
1747 * group that contains that extent to RO mode and therefore force COW
1748 * when starting writeback.
1750 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1751 EXTENT_NORESERVE, 0, NULL);
1752 if (count > 0 || is_space_ino || is_reloc_ino) {
1754 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1755 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1757 if (is_space_ino || is_reloc_ino)
1758 bytes = range_bytes;
1760 spin_lock(&sinfo->lock);
1761 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1762 spin_unlock(&sinfo->lock);
1765 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1769 return cow_file_range(inode, locked_page, start, end, page_started,
1770 nr_written, 1, NULL);
1773 struct can_nocow_file_extent_args {
1776 /* Start file offset of the range we want to NOCOW. */
1778 /* End file offset (inclusive) of the range we want to NOCOW. */
1780 bool writeback_path;
1783 * Free the path passed to can_nocow_file_extent() once it's not needed
1788 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1793 /* Number of bytes that can be written to in NOCOW mode. */
1798 * Check if we can NOCOW the file extent that the path points to.
1799 * This function may return with the path released, so the caller should check
1800 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1802 * Returns: < 0 on error
1803 * 0 if we can not NOCOW
1806 static int can_nocow_file_extent(struct btrfs_path *path,
1807 struct btrfs_key *key,
1808 struct btrfs_inode *inode,
1809 struct can_nocow_file_extent_args *args)
1811 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1812 struct extent_buffer *leaf = path->nodes[0];
1813 struct btrfs_root *root = inode->root;
1814 struct btrfs_file_extent_item *fi;
1819 bool nowait = path->nowait;
1821 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1822 extent_type = btrfs_file_extent_type(leaf, fi);
1824 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1827 /* Can't access these fields unless we know it's not an inline extent. */
1828 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1829 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1830 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1832 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1833 extent_type == BTRFS_FILE_EXTENT_REG)
1837 * If the extent was created before the generation where the last snapshot
1838 * for its subvolume was created, then this implies the extent is shared,
1839 * hence we must COW.
1841 if (!args->strict &&
1842 btrfs_file_extent_generation(leaf, fi) <=
1843 btrfs_root_last_snapshot(&root->root_item))
1846 /* An explicit hole, must COW. */
1847 if (args->disk_bytenr == 0)
1850 /* Compressed/encrypted/encoded extents must be COWed. */
1851 if (btrfs_file_extent_compression(leaf, fi) ||
1852 btrfs_file_extent_encryption(leaf, fi) ||
1853 btrfs_file_extent_other_encoding(leaf, fi))
1856 extent_end = btrfs_file_extent_end(path);
1859 * The following checks can be expensive, as they need to take other
1860 * locks and do btree or rbtree searches, so release the path to avoid
1861 * blocking other tasks for too long.
1863 btrfs_release_path(path);
1865 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1866 key->offset - args->extent_offset,
1867 args->disk_bytenr, false, path);
1868 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1872 if (args->free_path) {
1874 * We don't need the path anymore, plus through the
1875 * csum_exist_in_range() call below we will end up allocating
1876 * another path. So free the path to avoid unnecessary extra
1879 btrfs_free_path(path);
1883 /* If there are pending snapshots for this root, we must COW. */
1884 if (args->writeback_path && !is_freespace_inode &&
1885 atomic_read(&root->snapshot_force_cow))
1888 args->disk_bytenr += args->extent_offset;
1889 args->disk_bytenr += args->start - key->offset;
1890 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1893 * Force COW if csums exist in the range. This ensures that csums for a
1894 * given extent are either valid or do not exist.
1896 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1898 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1904 if (args->free_path && path)
1905 btrfs_free_path(path);
1907 return ret < 0 ? ret : can_nocow;
1911 * when nowcow writeback call back. This checks for snapshots or COW copies
1912 * of the extents that exist in the file, and COWs the file as required.
1914 * If no cow copies or snapshots exist, we write directly to the existing
1917 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1918 struct page *locked_page,
1919 const u64 start, const u64 end,
1921 unsigned long *nr_written)
1923 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1924 struct btrfs_root *root = inode->root;
1925 struct btrfs_path *path;
1926 u64 cow_start = (u64)-1;
1927 u64 cur_offset = start;
1929 bool check_prev = true;
1930 u64 ino = btrfs_ino(inode);
1931 struct btrfs_block_group *bg;
1933 struct can_nocow_file_extent_args nocow_args = { 0 };
1935 path = btrfs_alloc_path();
1937 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1938 EXTENT_LOCKED | EXTENT_DELALLOC |
1939 EXTENT_DO_ACCOUNTING |
1940 EXTENT_DEFRAG, PAGE_UNLOCK |
1941 PAGE_START_WRITEBACK |
1942 PAGE_END_WRITEBACK);
1946 nocow_args.end = end;
1947 nocow_args.writeback_path = true;
1950 struct btrfs_key found_key;
1951 struct btrfs_file_extent_item *fi;
1952 struct extent_buffer *leaf;
1960 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1966 * If there is no extent for our range when doing the initial
1967 * search, then go back to the previous slot as it will be the
1968 * one containing the search offset
1970 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1971 leaf = path->nodes[0];
1972 btrfs_item_key_to_cpu(leaf, &found_key,
1973 path->slots[0] - 1);
1974 if (found_key.objectid == ino &&
1975 found_key.type == BTRFS_EXTENT_DATA_KEY)
1980 /* Go to next leaf if we have exhausted the current one */
1981 leaf = path->nodes[0];
1982 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1983 ret = btrfs_next_leaf(root, path);
1985 if (cow_start != (u64)-1)
1986 cur_offset = cow_start;
1991 leaf = path->nodes[0];
1994 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1996 /* Didn't find anything for our INO */
1997 if (found_key.objectid > ino)
2000 * Keep searching until we find an EXTENT_ITEM or there are no
2001 * more extents for this inode
2003 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2004 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2009 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2010 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2011 found_key.offset > end)
2015 * If the found extent starts after requested offset, then
2016 * adjust extent_end to be right before this extent begins
2018 if (found_key.offset > cur_offset) {
2019 extent_end = found_key.offset;
2025 * Found extent which begins before our range and potentially
2028 fi = btrfs_item_ptr(leaf, path->slots[0],
2029 struct btrfs_file_extent_item);
2030 extent_type = btrfs_file_extent_type(leaf, fi);
2031 /* If this is triggered then we have a memory corruption. */
2032 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2033 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2037 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2038 extent_end = btrfs_file_extent_end(path);
2041 * If the extent we got ends before our current offset, skip to
2044 if (extent_end <= cur_offset) {
2049 nocow_args.start = cur_offset;
2050 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2052 if (cow_start != (u64)-1)
2053 cur_offset = cow_start;
2055 } else if (ret == 0) {
2060 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2065 * If nocow is false then record the beginning of the range
2066 * that needs to be COWed
2069 if (cow_start == (u64)-1)
2070 cow_start = cur_offset;
2071 cur_offset = extent_end;
2072 if (cur_offset > end)
2074 if (!path->nodes[0])
2081 * COW range from cow_start to found_key.offset - 1. As the key
2082 * will contain the beginning of the first extent that can be
2083 * NOCOW, following one which needs to be COW'ed
2085 if (cow_start != (u64)-1) {
2086 ret = fallback_to_cow(inode, locked_page,
2087 cow_start, found_key.offset - 1,
2088 page_started, nr_written);
2091 cow_start = (u64)-1;
2094 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2096 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
2097 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2098 struct extent_map *em;
2100 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2102 nocow_args.disk_bytenr, /* block_start */
2103 nocow_args.num_bytes, /* block_len */
2104 nocow_args.disk_num_bytes, /* orig_block_len */
2105 ram_bytes, BTRFS_COMPRESS_NONE,
2106 BTRFS_ORDERED_PREALLOC);
2111 free_extent_map(em);
2112 ret = btrfs_add_ordered_extent(inode,
2113 cur_offset, nocow_args.num_bytes,
2114 nocow_args.num_bytes,
2115 nocow_args.disk_bytenr,
2116 nocow_args.num_bytes, 0,
2117 1 << BTRFS_ORDERED_PREALLOC,
2118 BTRFS_COMPRESS_NONE);
2120 btrfs_drop_extent_map_range(inode, cur_offset,
2125 ret = btrfs_add_ordered_extent(inode, cur_offset,
2126 nocow_args.num_bytes,
2127 nocow_args.num_bytes,
2128 nocow_args.disk_bytenr,
2129 nocow_args.num_bytes,
2131 1 << BTRFS_ORDERED_NOCOW,
2132 BTRFS_COMPRESS_NONE);
2138 btrfs_dec_nocow_writers(bg);
2142 if (btrfs_is_data_reloc_root(root))
2144 * Error handled later, as we must prevent
2145 * extent_clear_unlock_delalloc() in error handler
2146 * from freeing metadata of created ordered extent.
2148 ret = btrfs_reloc_clone_csums(inode, cur_offset,
2149 nocow_args.num_bytes);
2151 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2152 locked_page, EXTENT_LOCKED |
2154 EXTENT_CLEAR_DATA_RESV,
2155 PAGE_UNLOCK | PAGE_SET_ORDERED);
2157 cur_offset = extent_end;
2160 * btrfs_reloc_clone_csums() error, now we're OK to call error
2161 * handler, as metadata for created ordered extent will only
2162 * be freed by btrfs_finish_ordered_io().
2166 if (cur_offset > end)
2169 btrfs_release_path(path);
2171 if (cur_offset <= end && cow_start == (u64)-1)
2172 cow_start = cur_offset;
2174 if (cow_start != (u64)-1) {
2176 ret = fallback_to_cow(inode, locked_page, cow_start, end,
2177 page_started, nr_written);
2184 btrfs_dec_nocow_writers(bg);
2186 if (ret && cur_offset < end)
2187 extent_clear_unlock_delalloc(inode, cur_offset, end,
2188 locked_page, EXTENT_LOCKED |
2189 EXTENT_DELALLOC | EXTENT_DEFRAG |
2190 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2191 PAGE_START_WRITEBACK |
2192 PAGE_END_WRITEBACK);
2193 btrfs_free_path(path);
2197 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2199 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2200 if (inode->defrag_bytes &&
2201 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2210 * Function to process delayed allocation (create CoW) for ranges which are
2211 * being touched for the first time.
2213 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2214 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2215 struct writeback_control *wbc)
2218 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2221 * The range must cover part of the @locked_page, or the returned
2222 * @page_started can confuse the caller.
2224 ASSERT(!(end <= page_offset(locked_page) ||
2225 start >= page_offset(locked_page) + PAGE_SIZE));
2227 if (should_nocow(inode, start, end)) {
2229 * Normally on a zoned device we're only doing COW writes, but
2230 * in case of relocation on a zoned filesystem we have taken
2231 * precaution, that we're only writing sequentially. It's safe
2232 * to use run_delalloc_nocow() here, like for regular
2233 * preallocated inodes.
2235 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2236 ret = run_delalloc_nocow(inode, locked_page, start, end,
2237 page_started, nr_written);
2238 } else if (!btrfs_inode_can_compress(inode) ||
2239 !inode_need_compress(inode, start, end)) {
2241 ret = run_delalloc_zoned(inode, locked_page, start, end,
2242 page_started, nr_written);
2244 ret = cow_file_range(inode, locked_page, start, end,
2245 page_started, nr_written, 1, NULL);
2247 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
2248 ret = cow_file_range_async(inode, wbc, locked_page, start, end,
2249 page_started, nr_written);
2253 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2258 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2259 struct extent_state *orig, u64 split)
2261 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2264 /* not delalloc, ignore it */
2265 if (!(orig->state & EXTENT_DELALLOC))
2268 size = orig->end - orig->start + 1;
2269 if (size > fs_info->max_extent_size) {
2274 * See the explanation in btrfs_merge_delalloc_extent, the same
2275 * applies here, just in reverse.
2277 new_size = orig->end - split + 1;
2278 num_extents = count_max_extents(fs_info, new_size);
2279 new_size = split - orig->start;
2280 num_extents += count_max_extents(fs_info, new_size);
2281 if (count_max_extents(fs_info, size) >= num_extents)
2285 spin_lock(&inode->lock);
2286 btrfs_mod_outstanding_extents(inode, 1);
2287 spin_unlock(&inode->lock);
2291 * Handle merged delayed allocation extents so we can keep track of new extents
2292 * that are just merged onto old extents, such as when we are doing sequential
2293 * writes, so we can properly account for the metadata space we'll need.
2295 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2296 struct extent_state *other)
2298 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2299 u64 new_size, old_size;
2302 /* not delalloc, ignore it */
2303 if (!(other->state & EXTENT_DELALLOC))
2306 if (new->start > other->start)
2307 new_size = new->end - other->start + 1;
2309 new_size = other->end - new->start + 1;
2311 /* we're not bigger than the max, unreserve the space and go */
2312 if (new_size <= fs_info->max_extent_size) {
2313 spin_lock(&inode->lock);
2314 btrfs_mod_outstanding_extents(inode, -1);
2315 spin_unlock(&inode->lock);
2320 * We have to add up either side to figure out how many extents were
2321 * accounted for before we merged into one big extent. If the number of
2322 * extents we accounted for is <= the amount we need for the new range
2323 * then we can return, otherwise drop. Think of it like this
2327 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2328 * need 2 outstanding extents, on one side we have 1 and the other side
2329 * we have 1 so they are == and we can return. But in this case
2331 * [MAX_SIZE+4k][MAX_SIZE+4k]
2333 * Each range on their own accounts for 2 extents, but merged together
2334 * they are only 3 extents worth of accounting, so we need to drop in
2337 old_size = other->end - other->start + 1;
2338 num_extents = count_max_extents(fs_info, old_size);
2339 old_size = new->end - new->start + 1;
2340 num_extents += count_max_extents(fs_info, old_size);
2341 if (count_max_extents(fs_info, new_size) >= num_extents)
2344 spin_lock(&inode->lock);
2345 btrfs_mod_outstanding_extents(inode, -1);
2346 spin_unlock(&inode->lock);
2349 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2350 struct btrfs_inode *inode)
2352 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2354 spin_lock(&root->delalloc_lock);
2355 if (list_empty(&inode->delalloc_inodes)) {
2356 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2357 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2358 root->nr_delalloc_inodes++;
2359 if (root->nr_delalloc_inodes == 1) {
2360 spin_lock(&fs_info->delalloc_root_lock);
2361 BUG_ON(!list_empty(&root->delalloc_root));
2362 list_add_tail(&root->delalloc_root,
2363 &fs_info->delalloc_roots);
2364 spin_unlock(&fs_info->delalloc_root_lock);
2367 spin_unlock(&root->delalloc_lock);
2370 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2371 struct btrfs_inode *inode)
2373 struct btrfs_fs_info *fs_info = root->fs_info;
2375 if (!list_empty(&inode->delalloc_inodes)) {
2376 list_del_init(&inode->delalloc_inodes);
2377 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2378 &inode->runtime_flags);
2379 root->nr_delalloc_inodes--;
2380 if (!root->nr_delalloc_inodes) {
2381 ASSERT(list_empty(&root->delalloc_inodes));
2382 spin_lock(&fs_info->delalloc_root_lock);
2383 BUG_ON(list_empty(&root->delalloc_root));
2384 list_del_init(&root->delalloc_root);
2385 spin_unlock(&fs_info->delalloc_root_lock);
2390 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2391 struct btrfs_inode *inode)
2393 spin_lock(&root->delalloc_lock);
2394 __btrfs_del_delalloc_inode(root, inode);
2395 spin_unlock(&root->delalloc_lock);
2399 * Properly track delayed allocation bytes in the inode and to maintain the
2400 * list of inodes that have pending delalloc work to be done.
2402 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2405 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2407 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2410 * set_bit and clear bit hooks normally require _irqsave/restore
2411 * but in this case, we are only testing for the DELALLOC
2412 * bit, which is only set or cleared with irqs on
2414 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2415 struct btrfs_root *root = inode->root;
2416 u64 len = state->end + 1 - state->start;
2417 u32 num_extents = count_max_extents(fs_info, len);
2418 bool do_list = !btrfs_is_free_space_inode(inode);
2420 spin_lock(&inode->lock);
2421 btrfs_mod_outstanding_extents(inode, num_extents);
2422 spin_unlock(&inode->lock);
2424 /* For sanity tests */
2425 if (btrfs_is_testing(fs_info))
2428 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2429 fs_info->delalloc_batch);
2430 spin_lock(&inode->lock);
2431 inode->delalloc_bytes += len;
2432 if (bits & EXTENT_DEFRAG)
2433 inode->defrag_bytes += len;
2434 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2435 &inode->runtime_flags))
2436 btrfs_add_delalloc_inodes(root, inode);
2437 spin_unlock(&inode->lock);
2440 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2441 (bits & EXTENT_DELALLOC_NEW)) {
2442 spin_lock(&inode->lock);
2443 inode->new_delalloc_bytes += state->end + 1 - state->start;
2444 spin_unlock(&inode->lock);
2449 * Once a range is no longer delalloc this function ensures that proper
2450 * accounting happens.
2452 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2453 struct extent_state *state, u32 bits)
2455 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2456 u64 len = state->end + 1 - state->start;
2457 u32 num_extents = count_max_extents(fs_info, len);
2459 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2460 spin_lock(&inode->lock);
2461 inode->defrag_bytes -= len;
2462 spin_unlock(&inode->lock);
2466 * set_bit and clear bit hooks normally require _irqsave/restore
2467 * but in this case, we are only testing for the DELALLOC
2468 * bit, which is only set or cleared with irqs on
2470 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2471 struct btrfs_root *root = inode->root;
2472 bool do_list = !btrfs_is_free_space_inode(inode);
2474 spin_lock(&inode->lock);
2475 btrfs_mod_outstanding_extents(inode, -num_extents);
2476 spin_unlock(&inode->lock);
2479 * We don't reserve metadata space for space cache inodes so we
2480 * don't need to call delalloc_release_metadata if there is an
2483 if (bits & EXTENT_CLEAR_META_RESV &&
2484 root != fs_info->tree_root)
2485 btrfs_delalloc_release_metadata(inode, len, false);
2487 /* For sanity tests. */
2488 if (btrfs_is_testing(fs_info))
2491 if (!btrfs_is_data_reloc_root(root) &&
2492 do_list && !(state->state & EXTENT_NORESERVE) &&
2493 (bits & EXTENT_CLEAR_DATA_RESV))
2494 btrfs_free_reserved_data_space_noquota(fs_info, len);
2496 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2497 fs_info->delalloc_batch);
2498 spin_lock(&inode->lock);
2499 inode->delalloc_bytes -= len;
2500 if (do_list && inode->delalloc_bytes == 0 &&
2501 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2502 &inode->runtime_flags))
2503 btrfs_del_delalloc_inode(root, inode);
2504 spin_unlock(&inode->lock);
2507 if ((state->state & EXTENT_DELALLOC_NEW) &&
2508 (bits & EXTENT_DELALLOC_NEW)) {
2509 spin_lock(&inode->lock);
2510 ASSERT(inode->new_delalloc_bytes >= len);
2511 inode->new_delalloc_bytes -= len;
2512 if (bits & EXTENT_ADD_INODE_BYTES)
2513 inode_add_bytes(&inode->vfs_inode, len);
2514 spin_unlock(&inode->lock);
2519 * Split off the first pre bytes from the extent_map at [start, start + len]
2521 * This function is intended to be used only for extract_ordered_extent().
2523 static int split_extent_map(struct btrfs_inode *inode, u64 start, u64 len, u64 pre)
2525 struct extent_map_tree *em_tree = &inode->extent_tree;
2526 struct extent_map *em;
2527 struct extent_map *split_pre = NULL;
2528 struct extent_map *split_mid = NULL;
2530 unsigned long flags;
2535 split_pre = alloc_extent_map();
2538 split_mid = alloc_extent_map();
2544 lock_extent(&inode->io_tree, start, start + len - 1, NULL);
2545 write_lock(&em_tree->lock);
2546 em = lookup_extent_mapping(em_tree, start, len);
2552 ASSERT(em->len == len);
2553 ASSERT(!test_bit(EXTENT_FLAG_COMPRESSED, &em->flags));
2554 ASSERT(em->block_start < EXTENT_MAP_LAST_BYTE);
2555 ASSERT(test_bit(EXTENT_FLAG_PINNED, &em->flags));
2556 ASSERT(!test_bit(EXTENT_FLAG_LOGGING, &em->flags));
2557 ASSERT(!list_empty(&em->list));
2560 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
2562 /* First, replace the em with a new extent_map starting from * em->start */
2563 split_pre->start = em->start;
2564 split_pre->len = pre;
2565 split_pre->orig_start = split_pre->start;
2566 split_pre->block_start = em->block_start;
2567 split_pre->block_len = split_pre->len;
2568 split_pre->orig_block_len = split_pre->block_len;
2569 split_pre->ram_bytes = split_pre->len;
2570 split_pre->flags = flags;
2571 split_pre->compress_type = em->compress_type;
2572 split_pre->generation = em->generation;
2574 replace_extent_mapping(em_tree, em, split_pre, 1);
2577 * Now we only have an extent_map at:
2578 * [em->start, em->start + pre]
2581 /* Insert the middle extent_map. */
2582 split_mid->start = em->start + pre;
2583 split_mid->len = em->len - pre;
2584 split_mid->orig_start = split_mid->start;
2585 split_mid->block_start = em->block_start + pre;
2586 split_mid->block_len = split_mid->len;
2587 split_mid->orig_block_len = split_mid->block_len;
2588 split_mid->ram_bytes = split_mid->len;
2589 split_mid->flags = flags;
2590 split_mid->compress_type = em->compress_type;
2591 split_mid->generation = em->generation;
2592 add_extent_mapping(em_tree, split_mid, 1);
2595 free_extent_map(em);
2596 /* Once for the tree */
2597 free_extent_map(em);
2600 write_unlock(&em_tree->lock);
2601 unlock_extent(&inode->io_tree, start, start + len - 1, NULL);
2602 free_extent_map(split_mid);
2604 free_extent_map(split_pre);
2608 int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2609 struct btrfs_ordered_extent *ordered)
2611 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2612 u64 len = bbio->bio.bi_iter.bi_size;
2613 struct btrfs_inode *inode = bbio->inode;
2614 u64 ordered_len = ordered->num_bytes;
2617 /* Must always be called for the beginning of an ordered extent. */
2618 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2621 /* No need to split if the ordered extent covers the entire bio. */
2622 if (ordered->disk_num_bytes == len)
2625 ret = btrfs_split_ordered_extent(ordered, len);
2630 * Don't split the extent_map for NOCOW extents, as we're writing into
2631 * a pre-existing one.
2633 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
2636 return split_extent_map(inode, bbio->file_offset, ordered_len, len);
2640 * given a list of ordered sums record them in the inode. This happens
2641 * at IO completion time based on sums calculated at bio submission time.
2643 static int add_pending_csums(struct btrfs_trans_handle *trans,
2644 struct list_head *list)
2646 struct btrfs_ordered_sum *sum;
2647 struct btrfs_root *csum_root = NULL;
2650 list_for_each_entry(sum, list, list) {
2651 trans->adding_csums = true;
2653 csum_root = btrfs_csum_root(trans->fs_info,
2655 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2656 trans->adding_csums = false;
2663 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2666 struct extent_state **cached_state)
2668 u64 search_start = start;
2669 const u64 end = start + len - 1;
2671 while (search_start < end) {
2672 const u64 search_len = end - search_start + 1;
2673 struct extent_map *em;
2677 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2681 if (em->block_start != EXTENT_MAP_HOLE)
2685 if (em->start < search_start)
2686 em_len -= search_start - em->start;
2687 if (em_len > search_len)
2688 em_len = search_len;
2690 ret = set_extent_bit(&inode->io_tree, search_start,
2691 search_start + em_len - 1,
2692 EXTENT_DELALLOC_NEW, cached_state,
2695 search_start = extent_map_end(em);
2696 free_extent_map(em);
2703 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2704 unsigned int extra_bits,
2705 struct extent_state **cached_state)
2707 WARN_ON(PAGE_ALIGNED(end));
2709 if (start >= i_size_read(&inode->vfs_inode) &&
2710 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2712 * There can't be any extents following eof in this case so just
2713 * set the delalloc new bit for the range directly.
2715 extra_bits |= EXTENT_DELALLOC_NEW;
2719 ret = btrfs_find_new_delalloc_bytes(inode, start,
2726 return set_extent_delalloc(&inode->io_tree, start, end, extra_bits,
2730 /* see btrfs_writepage_start_hook for details on why this is required */
2731 struct btrfs_writepage_fixup {
2733 struct btrfs_inode *inode;
2734 struct btrfs_work work;
2737 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2739 struct btrfs_writepage_fixup *fixup;
2740 struct btrfs_ordered_extent *ordered;
2741 struct extent_state *cached_state = NULL;
2742 struct extent_changeset *data_reserved = NULL;
2744 struct btrfs_inode *inode;
2748 bool free_delalloc_space = true;
2750 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2752 inode = fixup->inode;
2753 page_start = page_offset(page);
2754 page_end = page_offset(page) + PAGE_SIZE - 1;
2757 * This is similar to page_mkwrite, we need to reserve the space before
2758 * we take the page lock.
2760 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2766 * Before we queued this fixup, we took a reference on the page.
2767 * page->mapping may go NULL, but it shouldn't be moved to a different
2770 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2772 * Unfortunately this is a little tricky, either
2774 * 1) We got here and our page had already been dealt with and
2775 * we reserved our space, thus ret == 0, so we need to just
2776 * drop our space reservation and bail. This can happen the
2777 * first time we come into the fixup worker, or could happen
2778 * while waiting for the ordered extent.
2779 * 2) Our page was already dealt with, but we happened to get an
2780 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2781 * this case we obviously don't have anything to release, but
2782 * because the page was already dealt with we don't want to
2783 * mark the page with an error, so make sure we're resetting
2784 * ret to 0. This is why we have this check _before_ the ret
2785 * check, because we do not want to have a surprise ENOSPC
2786 * when the page was already properly dealt with.
2789 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2790 btrfs_delalloc_release_space(inode, data_reserved,
2791 page_start, PAGE_SIZE,
2799 * We can't mess with the page state unless it is locked, so now that
2800 * it is locked bail if we failed to make our space reservation.
2805 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2807 /* already ordered? We're done */
2808 if (PageOrdered(page))
2811 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2813 unlock_extent(&inode->io_tree, page_start, page_end,
2816 btrfs_start_ordered_extent(ordered);
2817 btrfs_put_ordered_extent(ordered);
2821 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2827 * Everything went as planned, we're now the owner of a dirty page with
2828 * delayed allocation bits set and space reserved for our COW
2831 * The page was dirty when we started, nothing should have cleaned it.
2833 BUG_ON(!PageDirty(page));
2834 free_delalloc_space = false;
2836 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2837 if (free_delalloc_space)
2838 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2840 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2844 * We hit ENOSPC or other errors. Update the mapping and page
2845 * to reflect the errors and clean the page.
2847 mapping_set_error(page->mapping, ret);
2848 end_extent_writepage(page, ret, page_start, page_end);
2849 clear_page_dirty_for_io(page);
2852 btrfs_page_clear_checked(inode->root->fs_info, page, page_start, PAGE_SIZE);
2856 extent_changeset_free(data_reserved);
2858 * As a precaution, do a delayed iput in case it would be the last iput
2859 * that could need flushing space. Recursing back to fixup worker would
2862 btrfs_add_delayed_iput(inode);
2866 * There are a few paths in the higher layers of the kernel that directly
2867 * set the page dirty bit without asking the filesystem if it is a
2868 * good idea. This causes problems because we want to make sure COW
2869 * properly happens and the data=ordered rules are followed.
2871 * In our case any range that doesn't have the ORDERED bit set
2872 * hasn't been properly setup for IO. We kick off an async process
2873 * to fix it up. The async helper will wait for ordered extents, set
2874 * the delalloc bit and make it safe to write the page.
2876 int btrfs_writepage_cow_fixup(struct page *page)
2878 struct inode *inode = page->mapping->host;
2879 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2880 struct btrfs_writepage_fixup *fixup;
2882 /* This page has ordered extent covering it already */
2883 if (PageOrdered(page))
2887 * PageChecked is set below when we create a fixup worker for this page,
2888 * don't try to create another one if we're already PageChecked()
2890 * The extent_io writepage code will redirty the page if we send back
2893 if (PageChecked(page))
2896 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2901 * We are already holding a reference to this inode from
2902 * write_cache_pages. We need to hold it because the space reservation
2903 * takes place outside of the page lock, and we can't trust
2904 * page->mapping outside of the page lock.
2907 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2909 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2911 fixup->inode = BTRFS_I(inode);
2912 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2917 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2918 struct btrfs_inode *inode, u64 file_pos,
2919 struct btrfs_file_extent_item *stack_fi,
2920 const bool update_inode_bytes,
2921 u64 qgroup_reserved)
2923 struct btrfs_root *root = inode->root;
2924 const u64 sectorsize = root->fs_info->sectorsize;
2925 struct btrfs_path *path;
2926 struct extent_buffer *leaf;
2927 struct btrfs_key ins;
2928 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2929 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2930 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2931 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2932 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2933 struct btrfs_drop_extents_args drop_args = { 0 };
2936 path = btrfs_alloc_path();
2941 * we may be replacing one extent in the tree with another.
2942 * The new extent is pinned in the extent map, and we don't want
2943 * to drop it from the cache until it is completely in the btree.
2945 * So, tell btrfs_drop_extents to leave this extent in the cache.
2946 * the caller is expected to unpin it and allow it to be merged
2949 drop_args.path = path;
2950 drop_args.start = file_pos;
2951 drop_args.end = file_pos + num_bytes;
2952 drop_args.replace_extent = true;
2953 drop_args.extent_item_size = sizeof(*stack_fi);
2954 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2958 if (!drop_args.extent_inserted) {
2959 ins.objectid = btrfs_ino(inode);
2960 ins.offset = file_pos;
2961 ins.type = BTRFS_EXTENT_DATA_KEY;
2963 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2968 leaf = path->nodes[0];
2969 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2970 write_extent_buffer(leaf, stack_fi,
2971 btrfs_item_ptr_offset(leaf, path->slots[0]),
2972 sizeof(struct btrfs_file_extent_item));
2974 btrfs_mark_buffer_dirty(leaf);
2975 btrfs_release_path(path);
2978 * If we dropped an inline extent here, we know the range where it is
2979 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2980 * number of bytes only for that range containing the inline extent.
2981 * The remaining of the range will be processed when clearning the
2982 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2984 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2985 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2987 inline_size = drop_args.bytes_found - inline_size;
2988 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2989 drop_args.bytes_found -= inline_size;
2990 num_bytes -= sectorsize;
2993 if (update_inode_bytes)
2994 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2996 ins.objectid = disk_bytenr;
2997 ins.offset = disk_num_bytes;
2998 ins.type = BTRFS_EXTENT_ITEM_KEY;
3000 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3004 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3006 qgroup_reserved, &ins);
3008 btrfs_free_path(path);
3013 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3016 struct btrfs_block_group *cache;
3018 cache = btrfs_lookup_block_group(fs_info, start);
3021 spin_lock(&cache->lock);
3022 cache->delalloc_bytes -= len;
3023 spin_unlock(&cache->lock);
3025 btrfs_put_block_group(cache);
3028 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3029 struct btrfs_ordered_extent *oe)
3031 struct btrfs_file_extent_item stack_fi;
3032 bool update_inode_bytes;
3033 u64 num_bytes = oe->num_bytes;
3034 u64 ram_bytes = oe->ram_bytes;
3036 memset(&stack_fi, 0, sizeof(stack_fi));
3037 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3038 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3039 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3040 oe->disk_num_bytes);
3041 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3042 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3043 num_bytes = oe->truncated_len;
3044 ram_bytes = num_bytes;
3046 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3047 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3048 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3049 /* Encryption and other encoding is reserved and all 0 */
3052 * For delalloc, when completing an ordered extent we update the inode's
3053 * bytes when clearing the range in the inode's io tree, so pass false
3054 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3055 * except if the ordered extent was truncated.
3057 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3058 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3059 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3061 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3062 oe->file_offset, &stack_fi,
3063 update_inode_bytes, oe->qgroup_rsv);
3067 * As ordered data IO finishes, this gets called so we can finish
3068 * an ordered extent if the range of bytes in the file it covers are
3071 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
3073 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3074 struct btrfs_root *root = inode->root;
3075 struct btrfs_fs_info *fs_info = root->fs_info;
3076 struct btrfs_trans_handle *trans = NULL;
3077 struct extent_io_tree *io_tree = &inode->io_tree;
3078 struct extent_state *cached_state = NULL;
3080 int compress_type = 0;
3082 u64 logical_len = ordered_extent->num_bytes;
3083 bool freespace_inode;
3084 bool truncated = false;
3085 bool clear_reserved_extent = true;
3086 unsigned int clear_bits = EXTENT_DEFRAG;
3088 start = ordered_extent->file_offset;
3089 end = start + ordered_extent->num_bytes - 1;
3091 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3092 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3093 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3094 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3095 clear_bits |= EXTENT_DELALLOC_NEW;
3097 freespace_inode = btrfs_is_free_space_inode(inode);
3098 if (!freespace_inode)
3099 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3101 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3106 /* A valid ->physical implies a write on a sequential zone. */
3107 if (ordered_extent->physical != (u64)-1) {
3108 btrfs_rewrite_logical_zoned(ordered_extent);
3109 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3110 ordered_extent->disk_num_bytes);
3111 } else if (btrfs_is_data_reloc_root(inode->root)) {
3112 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3113 ordered_extent->disk_num_bytes);
3116 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3118 logical_len = ordered_extent->truncated_len;
3119 /* Truncated the entire extent, don't bother adding */
3124 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3125 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3127 btrfs_inode_safe_disk_i_size_write(inode, 0);
3128 if (freespace_inode)
3129 trans = btrfs_join_transaction_spacecache(root);
3131 trans = btrfs_join_transaction(root);
3132 if (IS_ERR(trans)) {
3133 ret = PTR_ERR(trans);
3137 trans->block_rsv = &inode->block_rsv;
3138 ret = btrfs_update_inode_fallback(trans, root, inode);
3139 if (ret) /* -ENOMEM or corruption */
3140 btrfs_abort_transaction(trans, ret);
3144 clear_bits |= EXTENT_LOCKED;
3145 lock_extent(io_tree, start, end, &cached_state);
3147 if (freespace_inode)
3148 trans = btrfs_join_transaction_spacecache(root);
3150 trans = btrfs_join_transaction(root);
3151 if (IS_ERR(trans)) {
3152 ret = PTR_ERR(trans);
3157 trans->block_rsv = &inode->block_rsv;
3159 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3160 compress_type = ordered_extent->compress_type;
3161 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3162 BUG_ON(compress_type);
3163 ret = btrfs_mark_extent_written(trans, inode,
3164 ordered_extent->file_offset,
3165 ordered_extent->file_offset +
3167 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3168 ordered_extent->disk_num_bytes);
3170 BUG_ON(root == fs_info->tree_root);
3171 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3173 clear_reserved_extent = false;
3174 btrfs_release_delalloc_bytes(fs_info,
3175 ordered_extent->disk_bytenr,
3176 ordered_extent->disk_num_bytes);
3179 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3180 ordered_extent->num_bytes, trans->transid);
3182 btrfs_abort_transaction(trans, ret);
3186 ret = add_pending_csums(trans, &ordered_extent->list);
3188 btrfs_abort_transaction(trans, ret);
3193 * If this is a new delalloc range, clear its new delalloc flag to
3194 * update the inode's number of bytes. This needs to be done first
3195 * before updating the inode item.
3197 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3198 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3199 clear_extent_bit(&inode->io_tree, start, end,
3200 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3203 btrfs_inode_safe_disk_i_size_write(inode, 0);
3204 ret = btrfs_update_inode_fallback(trans, root, inode);
3205 if (ret) { /* -ENOMEM or corruption */
3206 btrfs_abort_transaction(trans, ret);
3211 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3215 btrfs_end_transaction(trans);
3217 if (ret || truncated) {
3218 u64 unwritten_start = start;
3221 * If we failed to finish this ordered extent for any reason we
3222 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3223 * extent, and mark the inode with the error if it wasn't
3224 * already set. Any error during writeback would have already
3225 * set the mapping error, so we need to set it if we're the ones
3226 * marking this ordered extent as failed.
3228 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3229 &ordered_extent->flags))
3230 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3233 unwritten_start += logical_len;
3234 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3236 /* Drop extent maps for the part of the extent we didn't write. */
3237 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3240 * If the ordered extent had an IOERR or something else went
3241 * wrong we need to return the space for this ordered extent
3242 * back to the allocator. We only free the extent in the
3243 * truncated case if we didn't write out the extent at all.
3245 * If we made it past insert_reserved_file_extent before we
3246 * errored out then we don't need to do this as the accounting
3247 * has already been done.
3249 if ((ret || !logical_len) &&
3250 clear_reserved_extent &&
3251 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3252 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3254 * Discard the range before returning it back to the
3257 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3258 btrfs_discard_extent(fs_info,
3259 ordered_extent->disk_bytenr,
3260 ordered_extent->disk_num_bytes,
3262 btrfs_free_reserved_extent(fs_info,
3263 ordered_extent->disk_bytenr,
3264 ordered_extent->disk_num_bytes, 1);
3269 * This needs to be done to make sure anybody waiting knows we are done
3270 * updating everything for this ordered extent.
3272 btrfs_remove_ordered_extent(inode, ordered_extent);
3275 btrfs_put_ordered_extent(ordered_extent);
3276 /* once for the tree */
3277 btrfs_put_ordered_extent(ordered_extent);
3282 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode *inode,
3283 struct page *page, u64 start,
3284 u64 end, bool uptodate)
3286 trace_btrfs_writepage_end_io_hook(inode, start, end, uptodate);
3288 btrfs_mark_ordered_io_finished(inode, page, start, end + 1 - start, uptodate);
3292 * Verify the checksum for a single sector without any extra action that depend
3293 * on the type of I/O.
3295 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3296 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3298 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3301 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3303 shash->tfm = fs_info->csum_shash;
3305 kaddr = kmap_local_page(page) + pgoff;
3306 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3307 kunmap_local(kaddr);
3309 if (memcmp(csum, csum_expected, fs_info->csum_size))
3315 * Verify the checksum of a single data sector.
3317 * @bbio: btrfs_io_bio which contains the csum
3318 * @dev: device the sector is on
3319 * @bio_offset: offset to the beginning of the bio (in bytes)
3320 * @bv: bio_vec to check
3322 * Check if the checksum on a data block is valid. When a checksum mismatch is
3323 * detected, report the error and fill the corrupted range with zero.
3325 * Return %true if the sector is ok or had no checksum to start with, else %false.
3327 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3328 u32 bio_offset, struct bio_vec *bv)
3330 struct btrfs_inode *inode = bbio->inode;
3331 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3332 u64 file_offset = bbio->file_offset + bio_offset;
3333 u64 end = file_offset + bv->bv_len - 1;
3335 u8 csum[BTRFS_CSUM_SIZE];
3337 ASSERT(bv->bv_len == fs_info->sectorsize);
3342 if (btrfs_is_data_reloc_root(inode->root) &&
3343 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3345 /* Skip the range without csum for data reloc inode */
3346 clear_extent_bits(&inode->io_tree, file_offset, end,
3351 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3353 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3359 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3362 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3368 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3370 * @inode: The inode we want to perform iput on
3372 * This function uses the generic vfs_inode::i_count to track whether we should
3373 * just decrement it (in case it's > 1) or if this is the last iput then link
3374 * the inode to the delayed iput machinery. Delayed iputs are processed at
3375 * transaction commit time/superblock commit/cleaner kthread.
3377 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3379 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3381 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3384 atomic_inc(&fs_info->nr_delayed_iputs);
3385 spin_lock(&fs_info->delayed_iput_lock);
3386 ASSERT(list_empty(&inode->delayed_iput));
3387 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3388 spin_unlock(&fs_info->delayed_iput_lock);
3389 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3390 wake_up_process(fs_info->cleaner_kthread);
3393 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3394 struct btrfs_inode *inode)
3396 list_del_init(&inode->delayed_iput);
3397 spin_unlock(&fs_info->delayed_iput_lock);
3398 iput(&inode->vfs_inode);
3399 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3400 wake_up(&fs_info->delayed_iputs_wait);
3401 spin_lock(&fs_info->delayed_iput_lock);
3404 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3405 struct btrfs_inode *inode)
3407 if (!list_empty(&inode->delayed_iput)) {
3408 spin_lock(&fs_info->delayed_iput_lock);
3409 if (!list_empty(&inode->delayed_iput))
3410 run_delayed_iput_locked(fs_info, inode);
3411 spin_unlock(&fs_info->delayed_iput_lock);
3415 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3418 spin_lock(&fs_info->delayed_iput_lock);
3419 while (!list_empty(&fs_info->delayed_iputs)) {
3420 struct btrfs_inode *inode;
3422 inode = list_first_entry(&fs_info->delayed_iputs,
3423 struct btrfs_inode, delayed_iput);
3424 run_delayed_iput_locked(fs_info, inode);
3425 cond_resched_lock(&fs_info->delayed_iput_lock);
3427 spin_unlock(&fs_info->delayed_iput_lock);
3431 * Wait for flushing all delayed iputs
3433 * @fs_info: the filesystem
3435 * This will wait on any delayed iputs that are currently running with KILLABLE
3436 * set. Once they are all done running we will return, unless we are killed in
3437 * which case we return EINTR. This helps in user operations like fallocate etc
3438 * that might get blocked on the iputs.
3440 * Return EINTR if we were killed, 0 if nothing's pending
3442 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3444 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3445 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3452 * This creates an orphan entry for the given inode in case something goes wrong
3453 * in the middle of an unlink.
3455 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3456 struct btrfs_inode *inode)
3460 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3461 if (ret && ret != -EEXIST) {
3462 btrfs_abort_transaction(trans, ret);
3470 * We have done the delete so we can go ahead and remove the orphan item for
3471 * this particular inode.
3473 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3474 struct btrfs_inode *inode)
3476 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3480 * this cleans up any orphans that may be left on the list from the last use
3483 int btrfs_orphan_cleanup(struct btrfs_root *root)
3485 struct btrfs_fs_info *fs_info = root->fs_info;
3486 struct btrfs_path *path;
3487 struct extent_buffer *leaf;
3488 struct btrfs_key key, found_key;
3489 struct btrfs_trans_handle *trans;
3490 struct inode *inode;
3491 u64 last_objectid = 0;
3492 int ret = 0, nr_unlink = 0;
3494 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3497 path = btrfs_alloc_path();
3502 path->reada = READA_BACK;
3504 key.objectid = BTRFS_ORPHAN_OBJECTID;
3505 key.type = BTRFS_ORPHAN_ITEM_KEY;
3506 key.offset = (u64)-1;
3509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3514 * if ret == 0 means we found what we were searching for, which
3515 * is weird, but possible, so only screw with path if we didn't
3516 * find the key and see if we have stuff that matches
3520 if (path->slots[0] == 0)
3525 /* pull out the item */
3526 leaf = path->nodes[0];
3527 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3529 /* make sure the item matches what we want */
3530 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3532 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3535 /* release the path since we're done with it */
3536 btrfs_release_path(path);
3539 * this is where we are basically btrfs_lookup, without the
3540 * crossing root thing. we store the inode number in the
3541 * offset of the orphan item.
3544 if (found_key.offset == last_objectid) {
3546 "Error removing orphan entry, stopping orphan cleanup");
3551 last_objectid = found_key.offset;
3553 found_key.objectid = found_key.offset;
3554 found_key.type = BTRFS_INODE_ITEM_KEY;
3555 found_key.offset = 0;
3556 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3557 ret = PTR_ERR_OR_ZERO(inode);
3558 if (ret && ret != -ENOENT)
3561 if (ret == -ENOENT && root == fs_info->tree_root) {
3562 struct btrfs_root *dead_root;
3563 int is_dead_root = 0;
3566 * This is an orphan in the tree root. Currently these
3567 * could come from 2 sources:
3568 * a) a root (snapshot/subvolume) deletion in progress
3569 * b) a free space cache inode
3570 * We need to distinguish those two, as the orphan item
3571 * for a root must not get deleted before the deletion
3572 * of the snapshot/subvolume's tree completes.
3574 * btrfs_find_orphan_roots() ran before us, which has
3575 * found all deleted roots and loaded them into
3576 * fs_info->fs_roots_radix. So here we can find if an
3577 * orphan item corresponds to a deleted root by looking
3578 * up the root from that radix tree.
3581 spin_lock(&fs_info->fs_roots_radix_lock);
3582 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3583 (unsigned long)found_key.objectid);
3584 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3586 spin_unlock(&fs_info->fs_roots_radix_lock);
3589 /* prevent this orphan from being found again */
3590 key.offset = found_key.objectid - 1;
3597 * If we have an inode with links, there are a couple of
3600 * 1. We were halfway through creating fsverity metadata for the
3601 * file. In that case, the orphan item represents incomplete
3602 * fsverity metadata which must be cleaned up with
3603 * btrfs_drop_verity_items and deleting the orphan item.
3605 * 2. Old kernels (before v3.12) used to create an
3606 * orphan item for truncate indicating that there were possibly
3607 * extent items past i_size that needed to be deleted. In v3.12,
3608 * truncate was changed to update i_size in sync with the extent
3609 * items, but the (useless) orphan item was still created. Since
3610 * v4.18, we don't create the orphan item for truncate at all.
3612 * So, this item could mean that we need to do a truncate, but
3613 * only if this filesystem was last used on a pre-v3.12 kernel
3614 * and was not cleanly unmounted. The odds of that are quite
3615 * slim, and it's a pain to do the truncate now, so just delete
3618 * It's also possible that this orphan item was supposed to be
3619 * deleted but wasn't. The inode number may have been reused,
3620 * but either way, we can delete the orphan item.
3622 if (ret == -ENOENT || inode->i_nlink) {
3624 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3629 trans = btrfs_start_transaction(root, 1);
3630 if (IS_ERR(trans)) {
3631 ret = PTR_ERR(trans);
3635 btrfs_debug(fs_info, "auto deleting %Lu",
3636 found_key.objectid);
3637 ret = btrfs_del_orphan_item(trans, root,
3638 found_key.objectid);
3639 btrfs_end_transaction(trans);
3649 /* this will do delete_inode and everything for us */
3652 /* release the path since we're done with it */
3653 btrfs_release_path(path);
3655 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3656 trans = btrfs_join_transaction(root);
3658 btrfs_end_transaction(trans);
3662 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3666 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3667 btrfs_free_path(path);
3672 * very simple check to peek ahead in the leaf looking for xattrs. If we
3673 * don't find any xattrs, we know there can't be any acls.
3675 * slot is the slot the inode is in, objectid is the objectid of the inode
3677 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3678 int slot, u64 objectid,
3679 int *first_xattr_slot)
3681 u32 nritems = btrfs_header_nritems(leaf);
3682 struct btrfs_key found_key;
3683 static u64 xattr_access = 0;
3684 static u64 xattr_default = 0;
3687 if (!xattr_access) {
3688 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3689 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3690 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3691 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3695 *first_xattr_slot = -1;
3696 while (slot < nritems) {
3697 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3699 /* we found a different objectid, there must not be acls */
3700 if (found_key.objectid != objectid)
3703 /* we found an xattr, assume we've got an acl */
3704 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3705 if (*first_xattr_slot == -1)
3706 *first_xattr_slot = slot;
3707 if (found_key.offset == xattr_access ||
3708 found_key.offset == xattr_default)
3713 * we found a key greater than an xattr key, there can't
3714 * be any acls later on
3716 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3723 * it goes inode, inode backrefs, xattrs, extents,
3724 * so if there are a ton of hard links to an inode there can
3725 * be a lot of backrefs. Don't waste time searching too hard,
3726 * this is just an optimization
3731 /* we hit the end of the leaf before we found an xattr or
3732 * something larger than an xattr. We have to assume the inode
3735 if (*first_xattr_slot == -1)
3736 *first_xattr_slot = slot;
3741 * read an inode from the btree into the in-memory inode
3743 static int btrfs_read_locked_inode(struct inode *inode,
3744 struct btrfs_path *in_path)
3746 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3747 struct btrfs_path *path = in_path;
3748 struct extent_buffer *leaf;
3749 struct btrfs_inode_item *inode_item;
3750 struct btrfs_root *root = BTRFS_I(inode)->root;
3751 struct btrfs_key location;
3756 bool filled = false;
3757 int first_xattr_slot;
3759 ret = btrfs_fill_inode(inode, &rdev);
3764 path = btrfs_alloc_path();
3769 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3771 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3773 if (path != in_path)
3774 btrfs_free_path(path);
3778 leaf = path->nodes[0];
3783 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3784 struct btrfs_inode_item);
3785 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3786 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3787 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3788 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3789 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3790 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3791 round_up(i_size_read(inode), fs_info->sectorsize));
3793 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3794 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3796 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3797 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3799 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3800 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3802 BTRFS_I(inode)->i_otime.tv_sec =
3803 btrfs_timespec_sec(leaf, &inode_item->otime);
3804 BTRFS_I(inode)->i_otime.tv_nsec =
3805 btrfs_timespec_nsec(leaf, &inode_item->otime);
3807 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3808 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3809 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3811 inode_set_iversion_queried(inode,
3812 btrfs_inode_sequence(leaf, inode_item));
3813 inode->i_generation = BTRFS_I(inode)->generation;
3815 rdev = btrfs_inode_rdev(leaf, inode_item);
3817 BTRFS_I(inode)->index_cnt = (u64)-1;
3818 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3819 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3823 * If we were modified in the current generation and evicted from memory
3824 * and then re-read we need to do a full sync since we don't have any
3825 * idea about which extents were modified before we were evicted from
3828 * This is required for both inode re-read from disk and delayed inode
3829 * in delayed_nodes_tree.
3831 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3832 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3833 &BTRFS_I(inode)->runtime_flags);
3836 * We don't persist the id of the transaction where an unlink operation
3837 * against the inode was last made. So here we assume the inode might
3838 * have been evicted, and therefore the exact value of last_unlink_trans
3839 * lost, and set it to last_trans to avoid metadata inconsistencies
3840 * between the inode and its parent if the inode is fsync'ed and the log
3841 * replayed. For example, in the scenario:
3844 * ln mydir/foo mydir/bar
3847 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3848 * xfs_io -c fsync mydir/foo
3850 * mount fs, triggers fsync log replay
3852 * We must make sure that when we fsync our inode foo we also log its
3853 * parent inode, otherwise after log replay the parent still has the
3854 * dentry with the "bar" name but our inode foo has a link count of 1
3855 * and doesn't have an inode ref with the name "bar" anymore.
3857 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3858 * but it guarantees correctness at the expense of occasional full
3859 * transaction commits on fsync if our inode is a directory, or if our
3860 * inode is not a directory, logging its parent unnecessarily.
3862 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3865 * Same logic as for last_unlink_trans. We don't persist the generation
3866 * of the last transaction where this inode was used for a reflink
3867 * operation, so after eviction and reloading the inode we must be
3868 * pessimistic and assume the last transaction that modified the inode.
3870 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3873 if (inode->i_nlink != 1 ||
3874 path->slots[0] >= btrfs_header_nritems(leaf))
3877 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3878 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3881 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3882 if (location.type == BTRFS_INODE_REF_KEY) {
3883 struct btrfs_inode_ref *ref;
3885 ref = (struct btrfs_inode_ref *)ptr;
3886 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3887 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3888 struct btrfs_inode_extref *extref;
3890 extref = (struct btrfs_inode_extref *)ptr;
3891 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3896 * try to precache a NULL acl entry for files that don't have
3897 * any xattrs or acls
3899 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3900 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3901 if (first_xattr_slot != -1) {
3902 path->slots[0] = first_xattr_slot;
3903 ret = btrfs_load_inode_props(inode, path);
3906 "error loading props for ino %llu (root %llu): %d",
3907 btrfs_ino(BTRFS_I(inode)),
3908 root->root_key.objectid, ret);
3910 if (path != in_path)
3911 btrfs_free_path(path);
3914 cache_no_acl(inode);
3916 switch (inode->i_mode & S_IFMT) {
3918 inode->i_mapping->a_ops = &btrfs_aops;
3919 inode->i_fop = &btrfs_file_operations;
3920 inode->i_op = &btrfs_file_inode_operations;
3923 inode->i_fop = &btrfs_dir_file_operations;
3924 inode->i_op = &btrfs_dir_inode_operations;
3927 inode->i_op = &btrfs_symlink_inode_operations;
3928 inode_nohighmem(inode);
3929 inode->i_mapping->a_ops = &btrfs_aops;
3932 inode->i_op = &btrfs_special_inode_operations;
3933 init_special_inode(inode, inode->i_mode, rdev);
3937 btrfs_sync_inode_flags_to_i_flags(inode);
3942 * given a leaf and an inode, copy the inode fields into the leaf
3944 static void fill_inode_item(struct btrfs_trans_handle *trans,
3945 struct extent_buffer *leaf,
3946 struct btrfs_inode_item *item,
3947 struct inode *inode)
3949 struct btrfs_map_token token;
3952 btrfs_init_map_token(&token, leaf);
3954 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3955 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3956 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3957 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3958 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3960 btrfs_set_token_timespec_sec(&token, &item->atime,
3961 inode->i_atime.tv_sec);
3962 btrfs_set_token_timespec_nsec(&token, &item->atime,
3963 inode->i_atime.tv_nsec);
3965 btrfs_set_token_timespec_sec(&token, &item->mtime,
3966 inode->i_mtime.tv_sec);
3967 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3968 inode->i_mtime.tv_nsec);
3970 btrfs_set_token_timespec_sec(&token, &item->ctime,
3971 inode->i_ctime.tv_sec);
3972 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3973 inode->i_ctime.tv_nsec);
3975 btrfs_set_token_timespec_sec(&token, &item->otime,
3976 BTRFS_I(inode)->i_otime.tv_sec);
3977 btrfs_set_token_timespec_nsec(&token, &item->otime,
3978 BTRFS_I(inode)->i_otime.tv_nsec);
3980 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3981 btrfs_set_token_inode_generation(&token, item,
3982 BTRFS_I(inode)->generation);
3983 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3984 btrfs_set_token_inode_transid(&token, item, trans->transid);
3985 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3986 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3987 BTRFS_I(inode)->ro_flags);
3988 btrfs_set_token_inode_flags(&token, item, flags);
3989 btrfs_set_token_inode_block_group(&token, item, 0);
3993 * copy everything in the in-memory inode into the btree.
3995 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3996 struct btrfs_root *root,
3997 struct btrfs_inode *inode)
3999 struct btrfs_inode_item *inode_item;
4000 struct btrfs_path *path;
4001 struct extent_buffer *leaf;
4004 path = btrfs_alloc_path();
4008 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4015 leaf = path->nodes[0];
4016 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4017 struct btrfs_inode_item);
4019 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4020 btrfs_mark_buffer_dirty(leaf);
4021 btrfs_set_inode_last_trans(trans, inode);
4024 btrfs_free_path(path);
4029 * copy everything in the in-memory inode into the btree.
4031 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4032 struct btrfs_root *root,
4033 struct btrfs_inode *inode)
4035 struct btrfs_fs_info *fs_info = root->fs_info;
4039 * If the inode is a free space inode, we can deadlock during commit
4040 * if we put it into the delayed code.
4042 * The data relocation inode should also be directly updated
4045 if (!btrfs_is_free_space_inode(inode)
4046 && !btrfs_is_data_reloc_root(root)
4047 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4048 btrfs_update_root_times(trans, root);
4050 ret = btrfs_delayed_update_inode(trans, root, inode);
4052 btrfs_set_inode_last_trans(trans, inode);
4056 return btrfs_update_inode_item(trans, root, inode);
4059 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4060 struct btrfs_root *root, struct btrfs_inode *inode)
4064 ret = btrfs_update_inode(trans, root, inode);
4066 return btrfs_update_inode_item(trans, root, inode);
4071 * unlink helper that gets used here in inode.c and in the tree logging
4072 * recovery code. It remove a link in a directory with a given name, and
4073 * also drops the back refs in the inode to the directory
4075 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4076 struct btrfs_inode *dir,
4077 struct btrfs_inode *inode,
4078 const struct fscrypt_str *name,
4079 struct btrfs_rename_ctx *rename_ctx)
4081 struct btrfs_root *root = dir->root;
4082 struct btrfs_fs_info *fs_info = root->fs_info;
4083 struct btrfs_path *path;
4085 struct btrfs_dir_item *di;
4087 u64 ino = btrfs_ino(inode);
4088 u64 dir_ino = btrfs_ino(dir);
4090 path = btrfs_alloc_path();
4096 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4097 if (IS_ERR_OR_NULL(di)) {
4098 ret = di ? PTR_ERR(di) : -ENOENT;
4101 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4104 btrfs_release_path(path);
4107 * If we don't have dir index, we have to get it by looking up
4108 * the inode ref, since we get the inode ref, remove it directly,
4109 * it is unnecessary to do delayed deletion.
4111 * But if we have dir index, needn't search inode ref to get it.
4112 * Since the inode ref is close to the inode item, it is better
4113 * that we delay to delete it, and just do this deletion when
4114 * we update the inode item.
4116 if (inode->dir_index) {
4117 ret = btrfs_delayed_delete_inode_ref(inode);
4119 index = inode->dir_index;
4124 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4127 "failed to delete reference to %.*s, inode %llu parent %llu",
4128 name->len, name->name, ino, dir_ino);
4129 btrfs_abort_transaction(trans, ret);
4134 rename_ctx->index = index;
4136 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4138 btrfs_abort_transaction(trans, ret);
4143 * If we are in a rename context, we don't need to update anything in the
4144 * log. That will be done later during the rename by btrfs_log_new_name().
4145 * Besides that, doing it here would only cause extra unnecessary btree
4146 * operations on the log tree, increasing latency for applications.
4149 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4150 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4154 * If we have a pending delayed iput we could end up with the final iput
4155 * being run in btrfs-cleaner context. If we have enough of these built
4156 * up we can end up burning a lot of time in btrfs-cleaner without any
4157 * way to throttle the unlinks. Since we're currently holding a ref on
4158 * the inode we can run the delayed iput here without any issues as the
4159 * final iput won't be done until after we drop the ref we're currently
4162 btrfs_run_delayed_iput(fs_info, inode);
4164 btrfs_free_path(path);
4168 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4169 inode_inc_iversion(&inode->vfs_inode);
4170 inode_inc_iversion(&dir->vfs_inode);
4171 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4172 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4173 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4174 ret = btrfs_update_inode(trans, root, dir);
4179 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4180 struct btrfs_inode *dir, struct btrfs_inode *inode,
4181 const struct fscrypt_str *name)
4185 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4187 drop_nlink(&inode->vfs_inode);
4188 ret = btrfs_update_inode(trans, inode->root, inode);
4194 * helper to start transaction for unlink and rmdir.
4196 * unlink and rmdir are special in btrfs, they do not always free space, so
4197 * if we cannot make our reservations the normal way try and see if there is
4198 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4199 * allow the unlink to occur.
4201 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4203 struct btrfs_root *root = dir->root;
4205 return btrfs_start_transaction_fallback_global_rsv(root,
4206 BTRFS_UNLINK_METADATA_UNITS);
4209 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4211 struct btrfs_trans_handle *trans;
4212 struct inode *inode = d_inode(dentry);
4214 struct fscrypt_name fname;
4216 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4220 /* This needs to handle no-key deletions later on */
4222 trans = __unlink_start_trans(BTRFS_I(dir));
4223 if (IS_ERR(trans)) {
4224 ret = PTR_ERR(trans);
4228 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4231 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4236 if (inode->i_nlink == 0) {
4237 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4243 btrfs_end_transaction(trans);
4244 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4246 fscrypt_free_filename(&fname);
4250 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4251 struct btrfs_inode *dir, struct dentry *dentry)
4253 struct btrfs_root *root = dir->root;
4254 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4255 struct btrfs_path *path;
4256 struct extent_buffer *leaf;
4257 struct btrfs_dir_item *di;
4258 struct btrfs_key key;
4262 u64 dir_ino = btrfs_ino(dir);
4263 struct fscrypt_name fname;
4265 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4269 /* This needs to handle no-key deletions later on */
4271 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4272 objectid = inode->root->root_key.objectid;
4273 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4274 objectid = inode->location.objectid;
4277 fscrypt_free_filename(&fname);
4281 path = btrfs_alloc_path();
4287 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4288 &fname.disk_name, -1);
4289 if (IS_ERR_OR_NULL(di)) {
4290 ret = di ? PTR_ERR(di) : -ENOENT;
4294 leaf = path->nodes[0];
4295 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4296 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4297 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4299 btrfs_abort_transaction(trans, ret);
4302 btrfs_release_path(path);
4305 * This is a placeholder inode for a subvolume we didn't have a
4306 * reference to at the time of the snapshot creation. In the meantime
4307 * we could have renamed the real subvol link into our snapshot, so
4308 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4309 * Instead simply lookup the dir_index_item for this entry so we can
4310 * remove it. Otherwise we know we have a ref to the root and we can
4311 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4313 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4314 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4315 if (IS_ERR_OR_NULL(di)) {
4320 btrfs_abort_transaction(trans, ret);
4324 leaf = path->nodes[0];
4325 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4327 btrfs_release_path(path);
4329 ret = btrfs_del_root_ref(trans, objectid,
4330 root->root_key.objectid, dir_ino,
4331 &index, &fname.disk_name);
4333 btrfs_abort_transaction(trans, ret);
4338 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4340 btrfs_abort_transaction(trans, ret);
4344 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4345 inode_inc_iversion(&dir->vfs_inode);
4346 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4347 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4348 ret = btrfs_update_inode_fallback(trans, root, dir);
4350 btrfs_abort_transaction(trans, ret);
4352 btrfs_free_path(path);
4353 fscrypt_free_filename(&fname);
4358 * Helper to check if the subvolume references other subvolumes or if it's
4361 static noinline int may_destroy_subvol(struct btrfs_root *root)
4363 struct btrfs_fs_info *fs_info = root->fs_info;
4364 struct btrfs_path *path;
4365 struct btrfs_dir_item *di;
4366 struct btrfs_key key;
4367 struct fscrypt_str name = FSTR_INIT("default", 7);
4371 path = btrfs_alloc_path();
4375 /* Make sure this root isn't set as the default subvol */
4376 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4377 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4379 if (di && !IS_ERR(di)) {
4380 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4381 if (key.objectid == root->root_key.objectid) {
4384 "deleting default subvolume %llu is not allowed",
4388 btrfs_release_path(path);
4391 key.objectid = root->root_key.objectid;
4392 key.type = BTRFS_ROOT_REF_KEY;
4393 key.offset = (u64)-1;
4395 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4401 if (path->slots[0] > 0) {
4403 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4404 if (key.objectid == root->root_key.objectid &&
4405 key.type == BTRFS_ROOT_REF_KEY)
4409 btrfs_free_path(path);
4413 /* Delete all dentries for inodes belonging to the root */
4414 static void btrfs_prune_dentries(struct btrfs_root *root)
4416 struct btrfs_fs_info *fs_info = root->fs_info;
4417 struct rb_node *node;
4418 struct rb_node *prev;
4419 struct btrfs_inode *entry;
4420 struct inode *inode;
4423 if (!BTRFS_FS_ERROR(fs_info))
4424 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4426 spin_lock(&root->inode_lock);
4428 node = root->inode_tree.rb_node;
4432 entry = rb_entry(node, struct btrfs_inode, rb_node);
4434 if (objectid < btrfs_ino(entry))
4435 node = node->rb_left;
4436 else if (objectid > btrfs_ino(entry))
4437 node = node->rb_right;
4443 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4444 if (objectid <= btrfs_ino(entry)) {
4448 prev = rb_next(prev);
4452 entry = rb_entry(node, struct btrfs_inode, rb_node);
4453 objectid = btrfs_ino(entry) + 1;
4454 inode = igrab(&entry->vfs_inode);
4456 spin_unlock(&root->inode_lock);
4457 if (atomic_read(&inode->i_count) > 1)
4458 d_prune_aliases(inode);
4460 * btrfs_drop_inode will have it removed from the inode
4461 * cache when its usage count hits zero.
4465 spin_lock(&root->inode_lock);
4469 if (cond_resched_lock(&root->inode_lock))
4472 node = rb_next(node);
4474 spin_unlock(&root->inode_lock);
4477 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4479 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4480 struct btrfs_root *root = dir->root;
4481 struct inode *inode = d_inode(dentry);
4482 struct btrfs_root *dest = BTRFS_I(inode)->root;
4483 struct btrfs_trans_handle *trans;
4484 struct btrfs_block_rsv block_rsv;
4489 * Don't allow to delete a subvolume with send in progress. This is
4490 * inside the inode lock so the error handling that has to drop the bit
4491 * again is not run concurrently.
4493 spin_lock(&dest->root_item_lock);
4494 if (dest->send_in_progress) {
4495 spin_unlock(&dest->root_item_lock);
4497 "attempt to delete subvolume %llu during send",
4498 dest->root_key.objectid);
4501 if (atomic_read(&dest->nr_swapfiles)) {
4502 spin_unlock(&dest->root_item_lock);
4504 "attempt to delete subvolume %llu with active swapfile",
4505 root->root_key.objectid);
4508 root_flags = btrfs_root_flags(&dest->root_item);
4509 btrfs_set_root_flags(&dest->root_item,
4510 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4511 spin_unlock(&dest->root_item_lock);
4513 down_write(&fs_info->subvol_sem);
4515 ret = may_destroy_subvol(dest);
4519 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4521 * One for dir inode,
4522 * two for dir entries,
4523 * two for root ref/backref.
4525 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4529 trans = btrfs_start_transaction(root, 0);
4530 if (IS_ERR(trans)) {
4531 ret = PTR_ERR(trans);
4534 trans->block_rsv = &block_rsv;
4535 trans->bytes_reserved = block_rsv.size;
4537 btrfs_record_snapshot_destroy(trans, dir);
4539 ret = btrfs_unlink_subvol(trans, dir, dentry);
4541 btrfs_abort_transaction(trans, ret);
4545 ret = btrfs_record_root_in_trans(trans, dest);
4547 btrfs_abort_transaction(trans, ret);
4551 memset(&dest->root_item.drop_progress, 0,
4552 sizeof(dest->root_item.drop_progress));
4553 btrfs_set_root_drop_level(&dest->root_item, 0);
4554 btrfs_set_root_refs(&dest->root_item, 0);
4556 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4557 ret = btrfs_insert_orphan_item(trans,
4559 dest->root_key.objectid);
4561 btrfs_abort_transaction(trans, ret);
4566 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4567 BTRFS_UUID_KEY_SUBVOL,
4568 dest->root_key.objectid);
4569 if (ret && ret != -ENOENT) {
4570 btrfs_abort_transaction(trans, ret);
4573 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4574 ret = btrfs_uuid_tree_remove(trans,
4575 dest->root_item.received_uuid,
4576 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4577 dest->root_key.objectid);
4578 if (ret && ret != -ENOENT) {
4579 btrfs_abort_transaction(trans, ret);
4584 free_anon_bdev(dest->anon_dev);
4587 trans->block_rsv = NULL;
4588 trans->bytes_reserved = 0;
4589 ret = btrfs_end_transaction(trans);
4590 inode->i_flags |= S_DEAD;
4592 btrfs_subvolume_release_metadata(root, &block_rsv);
4594 up_write(&fs_info->subvol_sem);
4596 spin_lock(&dest->root_item_lock);
4597 root_flags = btrfs_root_flags(&dest->root_item);
4598 btrfs_set_root_flags(&dest->root_item,
4599 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4600 spin_unlock(&dest->root_item_lock);
4602 d_invalidate(dentry);
4603 btrfs_prune_dentries(dest);
4604 ASSERT(dest->send_in_progress == 0);
4610 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4612 struct inode *inode = d_inode(dentry);
4613 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4615 struct btrfs_trans_handle *trans;
4616 u64 last_unlink_trans;
4617 struct fscrypt_name fname;
4619 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4621 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4622 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4624 "extent tree v2 doesn't support snapshot deletion yet");
4627 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4630 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4634 /* This needs to handle no-key deletions later on */
4636 trans = __unlink_start_trans(BTRFS_I(dir));
4637 if (IS_ERR(trans)) {
4638 err = PTR_ERR(trans);
4642 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4643 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4647 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4651 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4653 /* now the directory is empty */
4654 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4657 btrfs_i_size_write(BTRFS_I(inode), 0);
4659 * Propagate the last_unlink_trans value of the deleted dir to
4660 * its parent directory. This is to prevent an unrecoverable
4661 * log tree in the case we do something like this:
4663 * 2) create snapshot under dir foo
4664 * 3) delete the snapshot
4667 * 6) fsync foo or some file inside foo
4669 if (last_unlink_trans >= trans->transid)
4670 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4673 btrfs_end_transaction(trans);
4675 btrfs_btree_balance_dirty(fs_info);
4676 fscrypt_free_filename(&fname);
4682 * btrfs_truncate_block - read, zero a chunk and write a block
4683 * @inode - inode that we're zeroing
4684 * @from - the offset to start zeroing
4685 * @len - the length to zero, 0 to zero the entire range respective to the
4687 * @front - zero up to the offset instead of from the offset on
4689 * This will find the block for the "from" offset and cow the block and zero the
4690 * part we want to zero. This is used with truncate and hole punching.
4692 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4695 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4696 struct address_space *mapping = inode->vfs_inode.i_mapping;
4697 struct extent_io_tree *io_tree = &inode->io_tree;
4698 struct btrfs_ordered_extent *ordered;
4699 struct extent_state *cached_state = NULL;
4700 struct extent_changeset *data_reserved = NULL;
4701 bool only_release_metadata = false;
4702 u32 blocksize = fs_info->sectorsize;
4703 pgoff_t index = from >> PAGE_SHIFT;
4704 unsigned offset = from & (blocksize - 1);
4706 gfp_t mask = btrfs_alloc_write_mask(mapping);
4707 size_t write_bytes = blocksize;
4712 if (IS_ALIGNED(offset, blocksize) &&
4713 (!len || IS_ALIGNED(len, blocksize)))
4716 block_start = round_down(from, blocksize);
4717 block_end = block_start + blocksize - 1;
4719 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4722 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4723 /* For nocow case, no need to reserve data space */
4724 only_release_metadata = true;
4729 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4731 if (!only_release_metadata)
4732 btrfs_free_reserved_data_space(inode, data_reserved,
4733 block_start, blocksize);
4737 page = find_or_create_page(mapping, index, mask);
4739 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4741 btrfs_delalloc_release_extents(inode, blocksize);
4745 ret = set_page_extent_mapped(page);
4749 if (!PageUptodate(page)) {
4750 ret = btrfs_read_folio(NULL, page_folio(page));
4752 if (page->mapping != mapping) {
4757 if (!PageUptodate(page)) {
4762 wait_on_page_writeback(page);
4764 lock_extent(io_tree, block_start, block_end, &cached_state);
4766 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4768 unlock_extent(io_tree, block_start, block_end, &cached_state);
4771 btrfs_start_ordered_extent(ordered);
4772 btrfs_put_ordered_extent(ordered);
4776 clear_extent_bit(&inode->io_tree, block_start, block_end,
4777 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4780 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4783 unlock_extent(io_tree, block_start, block_end, &cached_state);
4787 if (offset != blocksize) {
4789 len = blocksize - offset;
4791 memzero_page(page, (block_start - page_offset(page)),
4794 memzero_page(page, (block_start - page_offset(page)) + offset,
4797 btrfs_page_clear_checked(fs_info, page, block_start,
4798 block_end + 1 - block_start);
4799 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4800 unlock_extent(io_tree, block_start, block_end, &cached_state);
4802 if (only_release_metadata)
4803 set_extent_bit(&inode->io_tree, block_start, block_end,
4804 EXTENT_NORESERVE, NULL, GFP_NOFS);
4808 if (only_release_metadata)
4809 btrfs_delalloc_release_metadata(inode, blocksize, true);
4811 btrfs_delalloc_release_space(inode, data_reserved,
4812 block_start, blocksize, true);
4814 btrfs_delalloc_release_extents(inode, blocksize);
4818 if (only_release_metadata)
4819 btrfs_check_nocow_unlock(inode);
4820 extent_changeset_free(data_reserved);
4824 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4825 u64 offset, u64 len)
4827 struct btrfs_fs_info *fs_info = root->fs_info;
4828 struct btrfs_trans_handle *trans;
4829 struct btrfs_drop_extents_args drop_args = { 0 };
4833 * If NO_HOLES is enabled, we don't need to do anything.
4834 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4835 * or btrfs_update_inode() will be called, which guarantee that the next
4836 * fsync will know this inode was changed and needs to be logged.
4838 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4842 * 1 - for the one we're dropping
4843 * 1 - for the one we're adding
4844 * 1 - for updating the inode.
4846 trans = btrfs_start_transaction(root, 3);
4848 return PTR_ERR(trans);
4850 drop_args.start = offset;
4851 drop_args.end = offset + len;
4852 drop_args.drop_cache = true;
4854 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4856 btrfs_abort_transaction(trans, ret);
4857 btrfs_end_transaction(trans);
4861 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4863 btrfs_abort_transaction(trans, ret);
4865 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4866 btrfs_update_inode(trans, root, inode);
4868 btrfs_end_transaction(trans);
4873 * This function puts in dummy file extents for the area we're creating a hole
4874 * for. So if we are truncating this file to a larger size we need to insert
4875 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4876 * the range between oldsize and size
4878 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4880 struct btrfs_root *root = inode->root;
4881 struct btrfs_fs_info *fs_info = root->fs_info;
4882 struct extent_io_tree *io_tree = &inode->io_tree;
4883 struct extent_map *em = NULL;
4884 struct extent_state *cached_state = NULL;
4885 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4886 u64 block_end = ALIGN(size, fs_info->sectorsize);
4893 * If our size started in the middle of a block we need to zero out the
4894 * rest of the block before we expand the i_size, otherwise we could
4895 * expose stale data.
4897 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4901 if (size <= hole_start)
4904 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4906 cur_offset = hole_start;
4908 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4909 block_end - cur_offset);
4915 last_byte = min(extent_map_end(em), block_end);
4916 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4917 hole_size = last_byte - cur_offset;
4919 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4920 struct extent_map *hole_em;
4922 err = maybe_insert_hole(root, inode, cur_offset,
4927 err = btrfs_inode_set_file_extent_range(inode,
4928 cur_offset, hole_size);
4932 hole_em = alloc_extent_map();
4934 btrfs_drop_extent_map_range(inode, cur_offset,
4935 cur_offset + hole_size - 1,
4937 btrfs_set_inode_full_sync(inode);
4940 hole_em->start = cur_offset;
4941 hole_em->len = hole_size;
4942 hole_em->orig_start = cur_offset;
4944 hole_em->block_start = EXTENT_MAP_HOLE;
4945 hole_em->block_len = 0;
4946 hole_em->orig_block_len = 0;
4947 hole_em->ram_bytes = hole_size;
4948 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4949 hole_em->generation = fs_info->generation;
4951 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4952 free_extent_map(hole_em);
4954 err = btrfs_inode_set_file_extent_range(inode,
4955 cur_offset, hole_size);
4960 free_extent_map(em);
4962 cur_offset = last_byte;
4963 if (cur_offset >= block_end)
4966 free_extent_map(em);
4967 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4971 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4973 struct btrfs_root *root = BTRFS_I(inode)->root;
4974 struct btrfs_trans_handle *trans;
4975 loff_t oldsize = i_size_read(inode);
4976 loff_t newsize = attr->ia_size;
4977 int mask = attr->ia_valid;
4981 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4982 * special case where we need to update the times despite not having
4983 * these flags set. For all other operations the VFS set these flags
4984 * explicitly if it wants a timestamp update.
4986 if (newsize != oldsize) {
4987 inode_inc_iversion(inode);
4988 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
4989 inode->i_mtime = current_time(inode);
4990 inode->i_ctime = inode->i_mtime;
4994 if (newsize > oldsize) {
4996 * Don't do an expanding truncate while snapshotting is ongoing.
4997 * This is to ensure the snapshot captures a fully consistent
4998 * state of this file - if the snapshot captures this expanding
4999 * truncation, it must capture all writes that happened before
5002 btrfs_drew_write_lock(&root->snapshot_lock);
5003 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5005 btrfs_drew_write_unlock(&root->snapshot_lock);
5009 trans = btrfs_start_transaction(root, 1);
5010 if (IS_ERR(trans)) {
5011 btrfs_drew_write_unlock(&root->snapshot_lock);
5012 return PTR_ERR(trans);
5015 i_size_write(inode, newsize);
5016 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5017 pagecache_isize_extended(inode, oldsize, newsize);
5018 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5019 btrfs_drew_write_unlock(&root->snapshot_lock);
5020 btrfs_end_transaction(trans);
5022 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5024 if (btrfs_is_zoned(fs_info)) {
5025 ret = btrfs_wait_ordered_range(inode,
5026 ALIGN(newsize, fs_info->sectorsize),
5033 * We're truncating a file that used to have good data down to
5034 * zero. Make sure any new writes to the file get on disk
5038 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5039 &BTRFS_I(inode)->runtime_flags);
5041 truncate_setsize(inode, newsize);
5043 inode_dio_wait(inode);
5045 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5046 if (ret && inode->i_nlink) {
5050 * Truncate failed, so fix up the in-memory size. We
5051 * adjusted disk_i_size down as we removed extents, so
5052 * wait for disk_i_size to be stable and then update the
5053 * in-memory size to match.
5055 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5058 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5065 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5068 struct inode *inode = d_inode(dentry);
5069 struct btrfs_root *root = BTRFS_I(inode)->root;
5072 if (btrfs_root_readonly(root))
5075 err = setattr_prepare(idmap, dentry, attr);
5079 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5080 err = btrfs_setsize(inode, attr);
5085 if (attr->ia_valid) {
5086 setattr_copy(idmap, inode, attr);
5087 inode_inc_iversion(inode);
5088 err = btrfs_dirty_inode(BTRFS_I(inode));
5090 if (!err && attr->ia_valid & ATTR_MODE)
5091 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5098 * While truncating the inode pages during eviction, we get the VFS
5099 * calling btrfs_invalidate_folio() against each folio of the inode. This
5100 * is slow because the calls to btrfs_invalidate_folio() result in a
5101 * huge amount of calls to lock_extent() and clear_extent_bit(),
5102 * which keep merging and splitting extent_state structures over and over,
5103 * wasting lots of time.
5105 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5106 * skip all those expensive operations on a per folio basis and do only
5107 * the ordered io finishing, while we release here the extent_map and
5108 * extent_state structures, without the excessive merging and splitting.
5110 static void evict_inode_truncate_pages(struct inode *inode)
5112 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5113 struct rb_node *node;
5115 ASSERT(inode->i_state & I_FREEING);
5116 truncate_inode_pages_final(&inode->i_data);
5118 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5121 * Keep looping until we have no more ranges in the io tree.
5122 * We can have ongoing bios started by readahead that have
5123 * their endio callback (extent_io.c:end_bio_extent_readpage)
5124 * still in progress (unlocked the pages in the bio but did not yet
5125 * unlocked the ranges in the io tree). Therefore this means some
5126 * ranges can still be locked and eviction started because before
5127 * submitting those bios, which are executed by a separate task (work
5128 * queue kthread), inode references (inode->i_count) were not taken
5129 * (which would be dropped in the end io callback of each bio).
5130 * Therefore here we effectively end up waiting for those bios and
5131 * anyone else holding locked ranges without having bumped the inode's
5132 * reference count - if we don't do it, when they access the inode's
5133 * io_tree to unlock a range it may be too late, leading to an
5134 * use-after-free issue.
5136 spin_lock(&io_tree->lock);
5137 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5138 struct extent_state *state;
5139 struct extent_state *cached_state = NULL;
5142 unsigned state_flags;
5144 node = rb_first(&io_tree->state);
5145 state = rb_entry(node, struct extent_state, rb_node);
5146 start = state->start;
5148 state_flags = state->state;
5149 spin_unlock(&io_tree->lock);
5151 lock_extent(io_tree, start, end, &cached_state);
5154 * If still has DELALLOC flag, the extent didn't reach disk,
5155 * and its reserved space won't be freed by delayed_ref.
5156 * So we need to free its reserved space here.
5157 * (Refer to comment in btrfs_invalidate_folio, case 2)
5159 * Note, end is the bytenr of last byte, so we need + 1 here.
5161 if (state_flags & EXTENT_DELALLOC)
5162 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5165 clear_extent_bit(io_tree, start, end,
5166 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5170 spin_lock(&io_tree->lock);
5172 spin_unlock(&io_tree->lock);
5175 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5176 struct btrfs_block_rsv *rsv)
5178 struct btrfs_fs_info *fs_info = root->fs_info;
5179 struct btrfs_trans_handle *trans;
5180 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(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 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5194 * if we fail to make this reservation we can re-try without the
5195 * delayed_refs_extra so we can make some forward progress.
5197 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5198 BTRFS_RESERVE_FLUSH_EVICT);
5200 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5201 BTRFS_RESERVE_FLUSH_EVICT);
5204 "could not allocate space for delete; will truncate on mount");
5205 return ERR_PTR(-ENOSPC);
5207 delayed_refs_extra = 0;
5210 trans = btrfs_join_transaction(root);
5214 if (delayed_refs_extra) {
5215 trans->block_rsv = &fs_info->trans_block_rsv;
5216 trans->bytes_reserved = delayed_refs_extra;
5217 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5218 delayed_refs_extra, true);
5223 void btrfs_evict_inode(struct inode *inode)
5225 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5226 struct btrfs_trans_handle *trans;
5227 struct btrfs_root *root = BTRFS_I(inode)->root;
5228 struct btrfs_block_rsv *rsv = NULL;
5231 trace_btrfs_inode_evict(inode);
5234 fsverity_cleanup_inode(inode);
5239 evict_inode_truncate_pages(inode);
5241 if (inode->i_nlink &&
5242 ((btrfs_root_refs(&root->root_item) != 0 &&
5243 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5244 btrfs_is_free_space_inode(BTRFS_I(inode))))
5247 if (is_bad_inode(inode))
5250 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5253 if (inode->i_nlink > 0) {
5254 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5255 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5260 * This makes sure the inode item in tree is uptodate and the space for
5261 * the inode update is released.
5263 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5268 * This drops any pending insert or delete operations we have for this
5269 * inode. We could have a delayed dir index deletion queued up, but
5270 * we're removing the inode completely so that'll be taken care of in
5273 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5275 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5278 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5279 rsv->failfast = true;
5281 btrfs_i_size_write(BTRFS_I(inode), 0);
5284 struct btrfs_truncate_control control = {
5285 .inode = BTRFS_I(inode),
5286 .ino = btrfs_ino(BTRFS_I(inode)),
5291 trans = evict_refill_and_join(root, rsv);
5295 trans->block_rsv = rsv;
5297 ret = btrfs_truncate_inode_items(trans, root, &control);
5298 trans->block_rsv = &fs_info->trans_block_rsv;
5299 btrfs_end_transaction(trans);
5301 * We have not added new delayed items for our inode after we
5302 * have flushed its delayed items, so no need to throttle on
5303 * delayed items. However we have modified extent buffers.
5305 btrfs_btree_balance_dirty_nodelay(fs_info);
5306 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5313 * Errors here aren't a big deal, it just means we leave orphan items in
5314 * the tree. They will be cleaned up on the next mount. If the inode
5315 * number gets reused, cleanup deletes the orphan item without doing
5316 * anything, and unlink reuses the existing orphan item.
5318 * If it turns out that we are dropping too many of these, we might want
5319 * to add a mechanism for retrying these after a commit.
5321 trans = evict_refill_and_join(root, rsv);
5322 if (!IS_ERR(trans)) {
5323 trans->block_rsv = rsv;
5324 btrfs_orphan_del(trans, BTRFS_I(inode));
5325 trans->block_rsv = &fs_info->trans_block_rsv;
5326 btrfs_end_transaction(trans);
5330 btrfs_free_block_rsv(fs_info, rsv);
5332 * If we didn't successfully delete, the orphan item will still be in
5333 * the tree and we'll retry on the next mount. Again, we might also want
5334 * to retry these periodically in the future.
5336 btrfs_remove_delayed_node(BTRFS_I(inode));
5337 fsverity_cleanup_inode(inode);
5342 * Return the key found in the dir entry in the location pointer, fill @type
5343 * with BTRFS_FT_*, and return 0.
5345 * If no dir entries were found, returns -ENOENT.
5346 * If found a corrupted location in dir entry, returns -EUCLEAN.
5348 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5349 struct btrfs_key *location, u8 *type)
5351 struct btrfs_dir_item *di;
5352 struct btrfs_path *path;
5353 struct btrfs_root *root = dir->root;
5355 struct fscrypt_name fname;
5357 path = btrfs_alloc_path();
5361 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5365 * fscrypt_setup_filename() should never return a positive value, but
5366 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5370 /* This needs to handle no-key deletions later on */
5372 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5373 &fname.disk_name, 0);
5374 if (IS_ERR_OR_NULL(di)) {
5375 ret = di ? PTR_ERR(di) : -ENOENT;
5379 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5380 if (location->type != BTRFS_INODE_ITEM_KEY &&
5381 location->type != BTRFS_ROOT_ITEM_KEY) {
5383 btrfs_warn(root->fs_info,
5384 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5385 __func__, fname.disk_name.name, btrfs_ino(dir),
5386 location->objectid, location->type, location->offset);
5389 *type = btrfs_dir_ftype(path->nodes[0], di);
5391 fscrypt_free_filename(&fname);
5392 btrfs_free_path(path);
5397 * when we hit a tree root in a directory, the btrfs part of the inode
5398 * needs to be changed to reflect the root directory of the tree root. This
5399 * is kind of like crossing a mount point.
5401 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5402 struct btrfs_inode *dir,
5403 struct dentry *dentry,
5404 struct btrfs_key *location,
5405 struct btrfs_root **sub_root)
5407 struct btrfs_path *path;
5408 struct btrfs_root *new_root;
5409 struct btrfs_root_ref *ref;
5410 struct extent_buffer *leaf;
5411 struct btrfs_key key;
5414 struct fscrypt_name fname;
5416 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5420 path = btrfs_alloc_path();
5427 key.objectid = dir->root->root_key.objectid;
5428 key.type = BTRFS_ROOT_REF_KEY;
5429 key.offset = location->objectid;
5431 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5438 leaf = path->nodes[0];
5439 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5440 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5441 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5444 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5445 (unsigned long)(ref + 1), fname.disk_name.len);
5449 btrfs_release_path(path);
5451 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5452 if (IS_ERR(new_root)) {
5453 err = PTR_ERR(new_root);
5457 *sub_root = new_root;
5458 location->objectid = btrfs_root_dirid(&new_root->root_item);
5459 location->type = BTRFS_INODE_ITEM_KEY;
5460 location->offset = 0;
5463 btrfs_free_path(path);
5464 fscrypt_free_filename(&fname);
5468 static void inode_tree_add(struct btrfs_inode *inode)
5470 struct btrfs_root *root = inode->root;
5471 struct btrfs_inode *entry;
5473 struct rb_node *parent;
5474 struct rb_node *new = &inode->rb_node;
5475 u64 ino = btrfs_ino(inode);
5477 if (inode_unhashed(&inode->vfs_inode))
5480 spin_lock(&root->inode_lock);
5481 p = &root->inode_tree.rb_node;
5484 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5486 if (ino < btrfs_ino(entry))
5487 p = &parent->rb_left;
5488 else if (ino > btrfs_ino(entry))
5489 p = &parent->rb_right;
5491 WARN_ON(!(entry->vfs_inode.i_state &
5492 (I_WILL_FREE | I_FREEING)));
5493 rb_replace_node(parent, new, &root->inode_tree);
5494 RB_CLEAR_NODE(parent);
5495 spin_unlock(&root->inode_lock);
5499 rb_link_node(new, parent, p);
5500 rb_insert_color(new, &root->inode_tree);
5501 spin_unlock(&root->inode_lock);
5504 static void inode_tree_del(struct btrfs_inode *inode)
5506 struct btrfs_root *root = inode->root;
5509 spin_lock(&root->inode_lock);
5510 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5511 rb_erase(&inode->rb_node, &root->inode_tree);
5512 RB_CLEAR_NODE(&inode->rb_node);
5513 empty = RB_EMPTY_ROOT(&root->inode_tree);
5515 spin_unlock(&root->inode_lock);
5517 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5518 spin_lock(&root->inode_lock);
5519 empty = RB_EMPTY_ROOT(&root->inode_tree);
5520 spin_unlock(&root->inode_lock);
5522 btrfs_add_dead_root(root);
5527 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5529 struct btrfs_iget_args *args = p;
5531 inode->i_ino = args->ino;
5532 BTRFS_I(inode)->location.objectid = args->ino;
5533 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5534 BTRFS_I(inode)->location.offset = 0;
5535 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5536 BUG_ON(args->root && !BTRFS_I(inode)->root);
5538 if (args->root && args->root == args->root->fs_info->tree_root &&
5539 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5540 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5541 &BTRFS_I(inode)->runtime_flags);
5545 static int btrfs_find_actor(struct inode *inode, void *opaque)
5547 struct btrfs_iget_args *args = opaque;
5549 return args->ino == BTRFS_I(inode)->location.objectid &&
5550 args->root == BTRFS_I(inode)->root;
5553 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5554 struct btrfs_root *root)
5556 struct inode *inode;
5557 struct btrfs_iget_args args;
5558 unsigned long hashval = btrfs_inode_hash(ino, root);
5563 inode = iget5_locked(s, hashval, btrfs_find_actor,
5564 btrfs_init_locked_inode,
5570 * Get an inode object given its inode number and corresponding root.
5571 * Path can be preallocated to prevent recursing back to iget through
5572 * allocator. NULL is also valid but may require an additional allocation
5575 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5576 struct btrfs_root *root, struct btrfs_path *path)
5578 struct inode *inode;
5580 inode = btrfs_iget_locked(s, ino, root);
5582 return ERR_PTR(-ENOMEM);
5584 if (inode->i_state & I_NEW) {
5587 ret = btrfs_read_locked_inode(inode, path);
5589 inode_tree_add(BTRFS_I(inode));
5590 unlock_new_inode(inode);
5594 * ret > 0 can come from btrfs_search_slot called by
5595 * btrfs_read_locked_inode, this means the inode item
5600 inode = ERR_PTR(ret);
5607 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5609 return btrfs_iget_path(s, ino, root, NULL);
5612 static struct inode *new_simple_dir(struct super_block *s,
5613 struct btrfs_key *key,
5614 struct btrfs_root *root)
5616 struct inode *inode = new_inode(s);
5619 return ERR_PTR(-ENOMEM);
5621 BTRFS_I(inode)->root = btrfs_grab_root(root);
5622 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5623 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5625 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5627 * We only need lookup, the rest is read-only and there's no inode
5628 * associated with the dentry
5630 inode->i_op = &simple_dir_inode_operations;
5631 inode->i_opflags &= ~IOP_XATTR;
5632 inode->i_fop = &simple_dir_operations;
5633 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5634 inode->i_mtime = current_time(inode);
5635 inode->i_atime = inode->i_mtime;
5636 inode->i_ctime = inode->i_mtime;
5637 BTRFS_I(inode)->i_otime = inode->i_mtime;
5642 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5643 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5644 static_assert(BTRFS_FT_DIR == FT_DIR);
5645 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5646 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5647 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5648 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5649 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5651 static inline u8 btrfs_inode_type(struct inode *inode)
5653 return fs_umode_to_ftype(inode->i_mode);
5656 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5658 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5659 struct inode *inode;
5660 struct btrfs_root *root = BTRFS_I(dir)->root;
5661 struct btrfs_root *sub_root = root;
5662 struct btrfs_key location;
5666 if (dentry->d_name.len > BTRFS_NAME_LEN)
5667 return ERR_PTR(-ENAMETOOLONG);
5669 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5671 return ERR_PTR(ret);
5673 if (location.type == BTRFS_INODE_ITEM_KEY) {
5674 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5678 /* Do extra check against inode mode with di_type */
5679 if (btrfs_inode_type(inode) != di_type) {
5681 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5682 inode->i_mode, btrfs_inode_type(inode),
5685 return ERR_PTR(-EUCLEAN);
5690 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5691 &location, &sub_root);
5694 inode = ERR_PTR(ret);
5696 inode = new_simple_dir(dir->i_sb, &location, root);
5698 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5699 btrfs_put_root(sub_root);
5704 down_read(&fs_info->cleanup_work_sem);
5705 if (!sb_rdonly(inode->i_sb))
5706 ret = btrfs_orphan_cleanup(sub_root);
5707 up_read(&fs_info->cleanup_work_sem);
5710 inode = ERR_PTR(ret);
5717 static int btrfs_dentry_delete(const struct dentry *dentry)
5719 struct btrfs_root *root;
5720 struct inode *inode = d_inode(dentry);
5722 if (!inode && !IS_ROOT(dentry))
5723 inode = d_inode(dentry->d_parent);
5726 root = BTRFS_I(inode)->root;
5727 if (btrfs_root_refs(&root->root_item) == 0)
5730 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5736 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5739 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5741 if (inode == ERR_PTR(-ENOENT))
5743 return d_splice_alias(inode, dentry);
5747 * All this infrastructure exists because dir_emit can fault, and we are holding
5748 * the tree lock when doing readdir. For now just allocate a buffer and copy
5749 * our information into that, and then dir_emit from the buffer. This is
5750 * similar to what NFS does, only we don't keep the buffer around in pagecache
5751 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5752 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5755 static int btrfs_opendir(struct inode *inode, struct file *file)
5757 struct btrfs_file_private *private;
5759 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5762 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5763 if (!private->filldir_buf) {
5767 file->private_data = private;
5778 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5781 struct dir_entry *entry = addr;
5782 char *name = (char *)(entry + 1);
5784 ctx->pos = get_unaligned(&entry->offset);
5785 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5786 get_unaligned(&entry->ino),
5787 get_unaligned(&entry->type)))
5789 addr += sizeof(struct dir_entry) +
5790 get_unaligned(&entry->name_len);
5796 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5798 struct inode *inode = file_inode(file);
5799 struct btrfs_root *root = BTRFS_I(inode)->root;
5800 struct btrfs_file_private *private = file->private_data;
5801 struct btrfs_dir_item *di;
5802 struct btrfs_key key;
5803 struct btrfs_key found_key;
5804 struct btrfs_path *path;
5806 struct list_head ins_list;
5807 struct list_head del_list;
5814 struct btrfs_key location;
5816 if (!dir_emit_dots(file, ctx))
5819 path = btrfs_alloc_path();
5823 addr = private->filldir_buf;
5824 path->reada = READA_FORWARD;
5826 INIT_LIST_HEAD(&ins_list);
5827 INIT_LIST_HEAD(&del_list);
5828 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5831 key.type = BTRFS_DIR_INDEX_KEY;
5832 key.offset = ctx->pos;
5833 key.objectid = btrfs_ino(BTRFS_I(inode));
5835 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5836 struct dir_entry *entry;
5837 struct extent_buffer *leaf = path->nodes[0];
5840 if (found_key.objectid != key.objectid)
5842 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5844 if (found_key.offset < ctx->pos)
5846 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5848 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5849 name_len = btrfs_dir_name_len(leaf, di);
5850 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5852 btrfs_release_path(path);
5853 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5856 addr = private->filldir_buf;
5862 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5864 name_ptr = (char *)(entry + 1);
5865 read_extent_buffer(leaf, name_ptr,
5866 (unsigned long)(di + 1), name_len);
5867 put_unaligned(name_len, &entry->name_len);
5868 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5869 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5870 put_unaligned(location.objectid, &entry->ino);
5871 put_unaligned(found_key.offset, &entry->offset);
5873 addr += sizeof(struct dir_entry) + name_len;
5874 total_len += sizeof(struct dir_entry) + name_len;
5876 /* Catch error encountered during iteration */
5880 btrfs_release_path(path);
5882 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5886 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5891 * Stop new entries from being returned after we return the last
5894 * New directory entries are assigned a strictly increasing
5895 * offset. This means that new entries created during readdir
5896 * are *guaranteed* to be seen in the future by that readdir.
5897 * This has broken buggy programs which operate on names as
5898 * they're returned by readdir. Until we re-use freed offsets
5899 * we have this hack to stop new entries from being returned
5900 * under the assumption that they'll never reach this huge
5903 * This is being careful not to overflow 32bit loff_t unless the
5904 * last entry requires it because doing so has broken 32bit apps
5907 if (ctx->pos >= INT_MAX)
5908 ctx->pos = LLONG_MAX;
5915 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5916 btrfs_free_path(path);
5921 * This is somewhat expensive, updating the tree every time the
5922 * inode changes. But, it is most likely to find the inode in cache.
5923 * FIXME, needs more benchmarking...there are no reasons other than performance
5924 * to keep or drop this code.
5926 static int btrfs_dirty_inode(struct btrfs_inode *inode)
5928 struct btrfs_root *root = inode->root;
5929 struct btrfs_fs_info *fs_info = root->fs_info;
5930 struct btrfs_trans_handle *trans;
5933 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
5936 trans = btrfs_join_transaction(root);
5938 return PTR_ERR(trans);
5940 ret = btrfs_update_inode(trans, root, inode);
5941 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
5942 /* whoops, lets try again with the full transaction */
5943 btrfs_end_transaction(trans);
5944 trans = btrfs_start_transaction(root, 1);
5946 return PTR_ERR(trans);
5948 ret = btrfs_update_inode(trans, root, inode);
5950 btrfs_end_transaction(trans);
5951 if (inode->delayed_node)
5952 btrfs_balance_delayed_items(fs_info);
5958 * This is a copy of file_update_time. We need this so we can return error on
5959 * ENOSPC for updating the inode in the case of file write and mmap writes.
5961 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
5964 struct btrfs_root *root = BTRFS_I(inode)->root;
5965 bool dirty = flags & ~S_VERSION;
5967 if (btrfs_root_readonly(root))
5970 if (flags & S_VERSION)
5971 dirty |= inode_maybe_inc_iversion(inode, dirty);
5972 if (flags & S_CTIME)
5973 inode->i_ctime = *now;
5974 if (flags & S_MTIME)
5975 inode->i_mtime = *now;
5976 if (flags & S_ATIME)
5977 inode->i_atime = *now;
5978 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
5982 * find the highest existing sequence number in a directory
5983 * and then set the in-memory index_cnt variable to reflect
5984 * free sequence numbers
5986 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5988 struct btrfs_root *root = inode->root;
5989 struct btrfs_key key, found_key;
5990 struct btrfs_path *path;
5991 struct extent_buffer *leaf;
5994 key.objectid = btrfs_ino(inode);
5995 key.type = BTRFS_DIR_INDEX_KEY;
5996 key.offset = (u64)-1;
5998 path = btrfs_alloc_path();
6002 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6005 /* FIXME: we should be able to handle this */
6010 if (path->slots[0] == 0) {
6011 inode->index_cnt = BTRFS_DIR_START_INDEX;
6017 leaf = path->nodes[0];
6018 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6020 if (found_key.objectid != btrfs_ino(inode) ||
6021 found_key.type != BTRFS_DIR_INDEX_KEY) {
6022 inode->index_cnt = BTRFS_DIR_START_INDEX;
6026 inode->index_cnt = found_key.offset + 1;
6028 btrfs_free_path(path);
6033 * helper to find a free sequence number in a given directory. This current
6034 * code is very simple, later versions will do smarter things in the btree
6036 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6040 if (dir->index_cnt == (u64)-1) {
6041 ret = btrfs_inode_delayed_dir_index_count(dir);
6043 ret = btrfs_set_inode_index_count(dir);
6049 *index = dir->index_cnt;
6055 static int btrfs_insert_inode_locked(struct inode *inode)
6057 struct btrfs_iget_args args;
6059 args.ino = BTRFS_I(inode)->location.objectid;
6060 args.root = BTRFS_I(inode)->root;
6062 return insert_inode_locked4(inode,
6063 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6064 btrfs_find_actor, &args);
6067 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6068 unsigned int *trans_num_items)
6070 struct inode *dir = args->dir;
6071 struct inode *inode = args->inode;
6074 if (!args->orphan) {
6075 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6081 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6083 fscrypt_free_filename(&args->fname);
6087 /* 1 to add inode item */
6088 *trans_num_items = 1;
6089 /* 1 to add compression property */
6090 if (BTRFS_I(dir)->prop_compress)
6091 (*trans_num_items)++;
6092 /* 1 to add default ACL xattr */
6093 if (args->default_acl)
6094 (*trans_num_items)++;
6095 /* 1 to add access ACL xattr */
6097 (*trans_num_items)++;
6098 #ifdef CONFIG_SECURITY
6099 /* 1 to add LSM xattr */
6100 if (dir->i_security)
6101 (*trans_num_items)++;
6104 /* 1 to add orphan item */
6105 (*trans_num_items)++;
6109 * 1 to add dir index
6110 * 1 to update parent inode item
6112 * No need for 1 unit for the inode ref item because it is
6113 * inserted in a batch together with the inode item at
6114 * btrfs_create_new_inode().
6116 *trans_num_items += 3;
6121 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6123 posix_acl_release(args->acl);
6124 posix_acl_release(args->default_acl);
6125 fscrypt_free_filename(&args->fname);
6129 * Inherit flags from the parent inode.
6131 * Currently only the compression flags and the cow flags are inherited.
6133 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6139 if (flags & BTRFS_INODE_NOCOMPRESS) {
6140 inode->flags &= ~BTRFS_INODE_COMPRESS;
6141 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6142 } else if (flags & BTRFS_INODE_COMPRESS) {
6143 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6144 inode->flags |= BTRFS_INODE_COMPRESS;
6147 if (flags & BTRFS_INODE_NODATACOW) {
6148 inode->flags |= BTRFS_INODE_NODATACOW;
6149 if (S_ISREG(inode->vfs_inode.i_mode))
6150 inode->flags |= BTRFS_INODE_NODATASUM;
6153 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6156 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6157 struct btrfs_new_inode_args *args)
6159 struct inode *dir = args->dir;
6160 struct inode *inode = args->inode;
6161 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6162 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6163 struct btrfs_root *root;
6164 struct btrfs_inode_item *inode_item;
6165 struct btrfs_key *location;
6166 struct btrfs_path *path;
6168 struct btrfs_inode_ref *ref;
6169 struct btrfs_key key[2];
6171 struct btrfs_item_batch batch;
6175 path = btrfs_alloc_path();
6180 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6181 root = BTRFS_I(inode)->root;
6183 ret = btrfs_get_free_objectid(root, &objectid);
6186 inode->i_ino = objectid;
6190 * O_TMPFILE, set link count to 0, so that after this point, we
6191 * fill in an inode item with the correct link count.
6193 set_nlink(inode, 0);
6195 trace_btrfs_inode_request(dir);
6197 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6201 /* index_cnt is ignored for everything but a dir. */
6202 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6203 BTRFS_I(inode)->generation = trans->transid;
6204 inode->i_generation = BTRFS_I(inode)->generation;
6207 * Subvolumes don't inherit flags from their parent directory.
6208 * Originally this was probably by accident, but we probably can't
6209 * change it now without compatibility issues.
6212 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6214 if (S_ISREG(inode->i_mode)) {
6215 if (btrfs_test_opt(fs_info, NODATASUM))
6216 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6217 if (btrfs_test_opt(fs_info, NODATACOW))
6218 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6219 BTRFS_INODE_NODATASUM;
6222 location = &BTRFS_I(inode)->location;
6223 location->objectid = objectid;
6224 location->offset = 0;
6225 location->type = BTRFS_INODE_ITEM_KEY;
6227 ret = btrfs_insert_inode_locked(inode);
6230 BTRFS_I(dir)->index_cnt--;
6235 * We could have gotten an inode number from somebody who was fsynced
6236 * and then removed in this same transaction, so let's just set full
6237 * sync since it will be a full sync anyway and this will blow away the
6238 * old info in the log.
6240 btrfs_set_inode_full_sync(BTRFS_I(inode));
6242 key[0].objectid = objectid;
6243 key[0].type = BTRFS_INODE_ITEM_KEY;
6246 sizes[0] = sizeof(struct btrfs_inode_item);
6248 if (!args->orphan) {
6250 * Start new inodes with an inode_ref. This is slightly more
6251 * efficient for small numbers of hard links since they will
6252 * be packed into one item. Extended refs will kick in if we
6253 * add more hard links than can fit in the ref item.
6255 key[1].objectid = objectid;
6256 key[1].type = BTRFS_INODE_REF_KEY;
6258 key[1].offset = objectid;
6259 sizes[1] = 2 + sizeof(*ref);
6261 key[1].offset = btrfs_ino(BTRFS_I(dir));
6262 sizes[1] = name->len + sizeof(*ref);
6266 batch.keys = &key[0];
6267 batch.data_sizes = &sizes[0];
6268 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6269 batch.nr = args->orphan ? 1 : 2;
6270 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6272 btrfs_abort_transaction(trans, ret);
6276 inode->i_mtime = current_time(inode);
6277 inode->i_atime = inode->i_mtime;
6278 inode->i_ctime = inode->i_mtime;
6279 BTRFS_I(inode)->i_otime = inode->i_mtime;
6282 * We're going to fill the inode item now, so at this point the inode
6283 * must be fully initialized.
6286 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6287 struct btrfs_inode_item);
6288 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6289 sizeof(*inode_item));
6290 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6292 if (!args->orphan) {
6293 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6294 struct btrfs_inode_ref);
6295 ptr = (unsigned long)(ref + 1);
6297 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6298 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6299 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6301 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6303 btrfs_set_inode_ref_index(path->nodes[0], ref,
6304 BTRFS_I(inode)->dir_index);
6305 write_extent_buffer(path->nodes[0], name->name, ptr,
6310 btrfs_mark_buffer_dirty(path->nodes[0]);
6312 * We don't need the path anymore, plus inheriting properties, adding
6313 * ACLs, security xattrs, orphan item or adding the link, will result in
6314 * allocating yet another path. So just free our path.
6316 btrfs_free_path(path);
6320 struct inode *parent;
6323 * Subvolumes inherit properties from their parent subvolume,
6324 * not the directory they were created in.
6326 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6327 BTRFS_I(dir)->root);
6328 if (IS_ERR(parent)) {
6329 ret = PTR_ERR(parent);
6331 ret = btrfs_inode_inherit_props(trans, inode, parent);
6335 ret = btrfs_inode_inherit_props(trans, inode, dir);
6339 "error inheriting props for ino %llu (root %llu): %d",
6340 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6345 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6348 if (!args->subvol) {
6349 ret = btrfs_init_inode_security(trans, args);
6351 btrfs_abort_transaction(trans, ret);
6356 inode_tree_add(BTRFS_I(inode));
6358 trace_btrfs_inode_new(inode);
6359 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6361 btrfs_update_root_times(trans, root);
6364 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6366 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6367 0, BTRFS_I(inode)->dir_index);
6370 btrfs_abort_transaction(trans, ret);
6378 * discard_new_inode() calls iput(), but the caller owns the reference
6382 discard_new_inode(inode);
6384 btrfs_free_path(path);
6389 * utility function to add 'inode' into 'parent_inode' with
6390 * a give name and a given sequence number.
6391 * if 'add_backref' is true, also insert a backref from the
6392 * inode to the parent directory.
6394 int btrfs_add_link(struct btrfs_trans_handle *trans,
6395 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6396 const struct fscrypt_str *name, int add_backref, u64 index)
6399 struct btrfs_key key;
6400 struct btrfs_root *root = parent_inode->root;
6401 u64 ino = btrfs_ino(inode);
6402 u64 parent_ino = btrfs_ino(parent_inode);
6404 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6405 memcpy(&key, &inode->root->root_key, sizeof(key));
6408 key.type = BTRFS_INODE_ITEM_KEY;
6412 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6413 ret = btrfs_add_root_ref(trans, key.objectid,
6414 root->root_key.objectid, parent_ino,
6416 } else if (add_backref) {
6417 ret = btrfs_insert_inode_ref(trans, root, name,
6418 ino, parent_ino, index);
6421 /* Nothing to clean up yet */
6425 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6426 btrfs_inode_type(&inode->vfs_inode), index);
6427 if (ret == -EEXIST || ret == -EOVERFLOW)
6430 btrfs_abort_transaction(trans, ret);
6434 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6436 inode_inc_iversion(&parent_inode->vfs_inode);
6438 * If we are replaying a log tree, we do not want to update the mtime
6439 * and ctime of the parent directory with the current time, since the
6440 * log replay procedure is responsible for setting them to their correct
6441 * values (the ones it had when the fsync was done).
6443 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6444 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6446 parent_inode->vfs_inode.i_mtime = now;
6447 parent_inode->vfs_inode.i_ctime = now;
6449 ret = btrfs_update_inode(trans, root, parent_inode);
6451 btrfs_abort_transaction(trans, ret);
6455 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6458 err = btrfs_del_root_ref(trans, key.objectid,
6459 root->root_key.objectid, parent_ino,
6460 &local_index, name);
6462 btrfs_abort_transaction(trans, err);
6463 } else if (add_backref) {
6467 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6470 btrfs_abort_transaction(trans, err);
6473 /* Return the original error code */
6477 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6478 struct inode *inode)
6480 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6481 struct btrfs_root *root = BTRFS_I(dir)->root;
6482 struct btrfs_new_inode_args new_inode_args = {
6487 unsigned int trans_num_items;
6488 struct btrfs_trans_handle *trans;
6491 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6495 trans = btrfs_start_transaction(root, trans_num_items);
6496 if (IS_ERR(trans)) {
6497 err = PTR_ERR(trans);
6498 goto out_new_inode_args;
6501 err = btrfs_create_new_inode(trans, &new_inode_args);
6503 d_instantiate_new(dentry, inode);
6505 btrfs_end_transaction(trans);
6506 btrfs_btree_balance_dirty(fs_info);
6508 btrfs_new_inode_args_destroy(&new_inode_args);
6515 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6516 struct dentry *dentry, umode_t mode, dev_t rdev)
6518 struct inode *inode;
6520 inode = new_inode(dir->i_sb);
6523 inode_init_owner(idmap, inode, dir, mode);
6524 inode->i_op = &btrfs_special_inode_operations;
6525 init_special_inode(inode, inode->i_mode, rdev);
6526 return btrfs_create_common(dir, dentry, inode);
6529 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6530 struct dentry *dentry, umode_t mode, bool excl)
6532 struct inode *inode;
6534 inode = new_inode(dir->i_sb);
6537 inode_init_owner(idmap, inode, dir, mode);
6538 inode->i_fop = &btrfs_file_operations;
6539 inode->i_op = &btrfs_file_inode_operations;
6540 inode->i_mapping->a_ops = &btrfs_aops;
6541 return btrfs_create_common(dir, dentry, inode);
6544 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6545 struct dentry *dentry)
6547 struct btrfs_trans_handle *trans = NULL;
6548 struct btrfs_root *root = BTRFS_I(dir)->root;
6549 struct inode *inode = d_inode(old_dentry);
6550 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6551 struct fscrypt_name fname;
6556 /* do not allow sys_link's with other subvols of the same device */
6557 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6560 if (inode->i_nlink >= BTRFS_LINK_MAX)
6563 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6567 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6572 * 2 items for inode and inode ref
6573 * 2 items for dir items
6574 * 1 item for parent inode
6575 * 1 item for orphan item deletion if O_TMPFILE
6577 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6578 if (IS_ERR(trans)) {
6579 err = PTR_ERR(trans);
6584 /* There are several dir indexes for this inode, clear the cache. */
6585 BTRFS_I(inode)->dir_index = 0ULL;
6587 inode_inc_iversion(inode);
6588 inode->i_ctime = current_time(inode);
6590 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6592 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6593 &fname.disk_name, 1, index);
6598 struct dentry *parent = dentry->d_parent;
6600 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6603 if (inode->i_nlink == 1) {
6605 * If new hard link count is 1, it's a file created
6606 * with open(2) O_TMPFILE flag.
6608 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6612 d_instantiate(dentry, inode);
6613 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6617 fscrypt_free_filename(&fname);
6619 btrfs_end_transaction(trans);
6621 inode_dec_link_count(inode);
6624 btrfs_btree_balance_dirty(fs_info);
6628 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6629 struct dentry *dentry, umode_t mode)
6631 struct inode *inode;
6633 inode = new_inode(dir->i_sb);
6636 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6637 inode->i_op = &btrfs_dir_inode_operations;
6638 inode->i_fop = &btrfs_dir_file_operations;
6639 return btrfs_create_common(dir, dentry, inode);
6642 static noinline int uncompress_inline(struct btrfs_path *path,
6644 struct btrfs_file_extent_item *item)
6647 struct extent_buffer *leaf = path->nodes[0];
6650 unsigned long inline_size;
6654 compress_type = btrfs_file_extent_compression(leaf, item);
6655 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6656 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6657 tmp = kmalloc(inline_size, GFP_NOFS);
6660 ptr = btrfs_file_extent_inline_start(item);
6662 read_extent_buffer(leaf, tmp, ptr, inline_size);
6664 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6665 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6668 * decompression code contains a memset to fill in any space between the end
6669 * of the uncompressed data and the end of max_size in case the decompressed
6670 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6671 * the end of an inline extent and the beginning of the next block, so we
6672 * cover that region here.
6675 if (max_size < PAGE_SIZE)
6676 memzero_page(page, max_size, PAGE_SIZE - max_size);
6681 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6684 struct btrfs_file_extent_item *fi;
6688 if (!page || PageUptodate(page))
6691 ASSERT(page_offset(page) == 0);
6693 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6694 struct btrfs_file_extent_item);
6695 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6696 return uncompress_inline(path, page, fi);
6698 copy_size = min_t(u64, PAGE_SIZE,
6699 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6700 kaddr = kmap_local_page(page);
6701 read_extent_buffer(path->nodes[0], kaddr,
6702 btrfs_file_extent_inline_start(fi), copy_size);
6703 kunmap_local(kaddr);
6704 if (copy_size < PAGE_SIZE)
6705 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6710 * Lookup the first extent overlapping a range in a file.
6712 * @inode: file to search in
6713 * @page: page to read extent data into if the extent is inline
6714 * @pg_offset: offset into @page to copy to
6715 * @start: file offset
6716 * @len: length of range starting at @start
6718 * Return the first &struct extent_map which overlaps the given range, reading
6719 * it from the B-tree and caching it if necessary. Note that there may be more
6720 * extents which overlap the given range after the returned extent_map.
6722 * If @page is not NULL and the extent is inline, this also reads the extent
6723 * data directly into the page and marks the extent up to date in the io_tree.
6725 * Return: ERR_PTR on error, non-NULL extent_map on success.
6727 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6728 struct page *page, size_t pg_offset,
6731 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6733 u64 extent_start = 0;
6735 u64 objectid = btrfs_ino(inode);
6736 int extent_type = -1;
6737 struct btrfs_path *path = NULL;
6738 struct btrfs_root *root = inode->root;
6739 struct btrfs_file_extent_item *item;
6740 struct extent_buffer *leaf;
6741 struct btrfs_key found_key;
6742 struct extent_map *em = NULL;
6743 struct extent_map_tree *em_tree = &inode->extent_tree;
6745 read_lock(&em_tree->lock);
6746 em = lookup_extent_mapping(em_tree, start, len);
6747 read_unlock(&em_tree->lock);
6750 if (em->start > start || em->start + em->len <= start)
6751 free_extent_map(em);
6752 else if (em->block_start == EXTENT_MAP_INLINE && page)
6753 free_extent_map(em);
6757 em = alloc_extent_map();
6762 em->start = EXTENT_MAP_HOLE;
6763 em->orig_start = EXTENT_MAP_HOLE;
6765 em->block_len = (u64)-1;
6767 path = btrfs_alloc_path();
6773 /* Chances are we'll be called again, so go ahead and do readahead */
6774 path->reada = READA_FORWARD;
6777 * The same explanation in load_free_space_cache applies here as well,
6778 * we only read when we're loading the free space cache, and at that
6779 * point the commit_root has everything we need.
6781 if (btrfs_is_free_space_inode(inode)) {
6782 path->search_commit_root = 1;
6783 path->skip_locking = 1;
6786 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6789 } else if (ret > 0) {
6790 if (path->slots[0] == 0)
6796 leaf = path->nodes[0];
6797 item = btrfs_item_ptr(leaf, path->slots[0],
6798 struct btrfs_file_extent_item);
6799 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6800 if (found_key.objectid != objectid ||
6801 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6803 * If we backup past the first extent we want to move forward
6804 * and see if there is an extent in front of us, otherwise we'll
6805 * say there is a hole for our whole search range which can
6812 extent_type = btrfs_file_extent_type(leaf, item);
6813 extent_start = found_key.offset;
6814 extent_end = btrfs_file_extent_end(path);
6815 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6816 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6817 /* Only regular file could have regular/prealloc extent */
6818 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6821 "regular/prealloc extent found for non-regular inode %llu",
6825 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6827 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6828 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6833 if (start >= extent_end) {
6835 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6836 ret = btrfs_next_leaf(root, path);
6842 leaf = path->nodes[0];
6844 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6845 if (found_key.objectid != objectid ||
6846 found_key.type != BTRFS_EXTENT_DATA_KEY)
6848 if (start + len <= found_key.offset)
6850 if (start > found_key.offset)
6853 /* New extent overlaps with existing one */
6855 em->orig_start = start;
6856 em->len = found_key.offset - start;
6857 em->block_start = EXTENT_MAP_HOLE;
6861 btrfs_extent_item_to_extent_map(inode, path, item, em);
6863 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6864 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6866 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6868 * Inline extent can only exist at file offset 0. This is
6869 * ensured by tree-checker and inline extent creation path.
6870 * Thus all members representing file offsets should be zero.
6872 ASSERT(pg_offset == 0);
6873 ASSERT(extent_start == 0);
6874 ASSERT(em->start == 0);
6877 * btrfs_extent_item_to_extent_map() should have properly
6878 * initialized em members already.
6880 * Other members are not utilized for inline extents.
6882 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6883 ASSERT(em->len == fs_info->sectorsize);
6885 ret = read_inline_extent(inode, path, page);
6892 em->orig_start = start;
6894 em->block_start = EXTENT_MAP_HOLE;
6897 btrfs_release_path(path);
6898 if (em->start > start || extent_map_end(em) <= start) {
6900 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6901 em->start, em->len, start, len);
6906 write_lock(&em_tree->lock);
6907 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6908 write_unlock(&em_tree->lock);
6910 btrfs_free_path(path);
6912 trace_btrfs_get_extent(root, inode, em);
6915 free_extent_map(em);
6916 return ERR_PTR(ret);
6921 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6922 struct btrfs_dio_data *dio_data,
6925 const u64 orig_start,
6926 const u64 block_start,
6927 const u64 block_len,
6928 const u64 orig_block_len,
6929 const u64 ram_bytes,
6932 struct extent_map *em = NULL;
6933 struct btrfs_ordered_extent *ordered;
6935 if (type != BTRFS_ORDERED_NOCOW) {
6936 em = create_io_em(inode, start, len, orig_start, block_start,
6937 block_len, orig_block_len, ram_bytes,
6938 BTRFS_COMPRESS_NONE, /* compress_type */
6943 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6944 block_start, block_len, 0,
6946 (1 << BTRFS_ORDERED_DIRECT),
6947 BTRFS_COMPRESS_NONE);
6948 if (IS_ERR(ordered)) {
6950 free_extent_map(em);
6951 btrfs_drop_extent_map_range(inode, start,
6952 start + len - 1, false);
6954 em = ERR_CAST(ordered);
6956 ASSERT(!dio_data->ordered);
6957 dio_data->ordered = ordered;
6964 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6965 struct btrfs_dio_data *dio_data,
6968 struct btrfs_root *root = inode->root;
6969 struct btrfs_fs_info *fs_info = root->fs_info;
6970 struct extent_map *em;
6971 struct btrfs_key ins;
6975 alloc_hint = get_extent_allocation_hint(inode, start, len);
6976 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6977 0, alloc_hint, &ins, 1, 1);
6979 return ERR_PTR(ret);
6981 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
6982 ins.objectid, ins.offset, ins.offset,
6983 ins.offset, BTRFS_ORDERED_REGULAR);
6984 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6986 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
6992 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
6994 struct btrfs_block_group *block_group;
6995 bool readonly = false;
6997 block_group = btrfs_lookup_block_group(fs_info, bytenr);
6998 if (!block_group || block_group->ro)
7001 btrfs_put_block_group(block_group);
7006 * Check if we can do nocow write into the range [@offset, @offset + @len)
7008 * @offset: File offset
7009 * @len: The length to write, will be updated to the nocow writeable
7011 * @orig_start: (optional) Return the original file offset of the file extent
7012 * @orig_len: (optional) Return the original on-disk length of the file extent
7013 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7014 * @strict: if true, omit optimizations that might force us into unnecessary
7015 * cow. e.g., don't trust generation number.
7018 * >0 and update @len if we can do nocow write
7019 * 0 if we can't do nocow write
7020 * <0 if error happened
7022 * NOTE: This only checks the file extents, caller is responsible to wait for
7023 * any ordered extents.
7025 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7026 u64 *orig_start, u64 *orig_block_len,
7027 u64 *ram_bytes, bool nowait, bool strict)
7029 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7030 struct can_nocow_file_extent_args nocow_args = { 0 };
7031 struct btrfs_path *path;
7033 struct extent_buffer *leaf;
7034 struct btrfs_root *root = BTRFS_I(inode)->root;
7035 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7036 struct btrfs_file_extent_item *fi;
7037 struct btrfs_key key;
7040 path = btrfs_alloc_path();
7043 path->nowait = nowait;
7045 ret = btrfs_lookup_file_extent(NULL, root, path,
7046 btrfs_ino(BTRFS_I(inode)), offset, 0);
7051 if (path->slots[0] == 0) {
7052 /* can't find the item, must cow */
7059 leaf = path->nodes[0];
7060 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7061 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7062 key.type != BTRFS_EXTENT_DATA_KEY) {
7063 /* not our file or wrong item type, must cow */
7067 if (key.offset > offset) {
7068 /* Wrong offset, must cow */
7072 if (btrfs_file_extent_end(path) <= offset)
7075 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7076 found_type = btrfs_file_extent_type(leaf, fi);
7078 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7080 nocow_args.start = offset;
7081 nocow_args.end = offset + *len - 1;
7082 nocow_args.strict = strict;
7083 nocow_args.free_path = true;
7085 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7086 /* can_nocow_file_extent() has freed the path. */
7090 /* Treat errors as not being able to NOCOW. */
7096 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7099 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7100 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7103 range_end = round_up(offset + nocow_args.num_bytes,
7104 root->fs_info->sectorsize) - 1;
7105 ret = test_range_bit(io_tree, offset, range_end,
7106 EXTENT_DELALLOC, 0, NULL);
7114 *orig_start = key.offset - nocow_args.extent_offset;
7116 *orig_block_len = nocow_args.disk_num_bytes;
7118 *len = nocow_args.num_bytes;
7121 btrfs_free_path(path);
7125 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7126 struct extent_state **cached_state,
7127 unsigned int iomap_flags)
7129 const bool writing = (iomap_flags & IOMAP_WRITE);
7130 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7131 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7132 struct btrfs_ordered_extent *ordered;
7137 if (!try_lock_extent(io_tree, lockstart, lockend,
7141 lock_extent(io_tree, lockstart, lockend, cached_state);
7144 * We're concerned with the entire range that we're going to be
7145 * doing DIO to, so we need to make sure there's no ordered
7146 * extents in this range.
7148 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7149 lockend - lockstart + 1);
7152 * We need to make sure there are no buffered pages in this
7153 * range either, we could have raced between the invalidate in
7154 * generic_file_direct_write and locking the extent. The
7155 * invalidate needs to happen so that reads after a write do not
7159 (!writing || !filemap_range_has_page(inode->i_mapping,
7160 lockstart, lockend)))
7163 unlock_extent(io_tree, lockstart, lockend, cached_state);
7167 btrfs_put_ordered_extent(ordered);
7172 * If we are doing a DIO read and the ordered extent we
7173 * found is for a buffered write, we can not wait for it
7174 * to complete and retry, because if we do so we can
7175 * deadlock with concurrent buffered writes on page
7176 * locks. This happens only if our DIO read covers more
7177 * than one extent map, if at this point has already
7178 * created an ordered extent for a previous extent map
7179 * and locked its range in the inode's io tree, and a
7180 * concurrent write against that previous extent map's
7181 * range and this range started (we unlock the ranges
7182 * in the io tree only when the bios complete and
7183 * buffered writes always lock pages before attempting
7184 * to lock range in the io tree).
7187 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7188 btrfs_start_ordered_extent(ordered);
7190 ret = nowait ? -EAGAIN : -ENOTBLK;
7191 btrfs_put_ordered_extent(ordered);
7194 * We could trigger writeback for this range (and wait
7195 * for it to complete) and then invalidate the pages for
7196 * this range (through invalidate_inode_pages2_range()),
7197 * but that can lead us to a deadlock with a concurrent
7198 * call to readahead (a buffered read or a defrag call
7199 * triggered a readahead) on a page lock due to an
7200 * ordered dio extent we created before but did not have
7201 * yet a corresponding bio submitted (whence it can not
7202 * complete), which makes readahead wait for that
7203 * ordered extent to complete while holding a lock on
7206 ret = nowait ? -EAGAIN : -ENOTBLK;
7218 /* The callers of this must take lock_extent() */
7219 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7220 u64 len, u64 orig_start, u64 block_start,
7221 u64 block_len, u64 orig_block_len,
7222 u64 ram_bytes, int compress_type,
7225 struct extent_map *em;
7228 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7229 type == BTRFS_ORDERED_COMPRESSED ||
7230 type == BTRFS_ORDERED_NOCOW ||
7231 type == BTRFS_ORDERED_REGULAR);
7233 em = alloc_extent_map();
7235 return ERR_PTR(-ENOMEM);
7238 em->orig_start = orig_start;
7240 em->block_len = block_len;
7241 em->block_start = block_start;
7242 em->orig_block_len = orig_block_len;
7243 em->ram_bytes = ram_bytes;
7244 em->generation = -1;
7245 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7246 if (type == BTRFS_ORDERED_PREALLOC) {
7247 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7248 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7249 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7250 em->compress_type = compress_type;
7253 ret = btrfs_replace_extent_map_range(inode, em, true);
7255 free_extent_map(em);
7256 return ERR_PTR(ret);
7259 /* em got 2 refs now, callers needs to do free_extent_map once. */
7264 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7265 struct inode *inode,
7266 struct btrfs_dio_data *dio_data,
7268 unsigned int iomap_flags)
7270 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7271 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7272 struct extent_map *em = *map;
7274 u64 block_start, orig_start, orig_block_len, ram_bytes;
7275 struct btrfs_block_group *bg;
7276 bool can_nocow = false;
7277 bool space_reserved = false;
7282 * We don't allocate a new extent in the following cases
7284 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7286 * 2) The extent is marked as PREALLOC. We're good to go here and can
7287 * just use the extent.
7290 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7291 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7292 em->block_start != EXTENT_MAP_HOLE)) {
7293 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7294 type = BTRFS_ORDERED_PREALLOC;
7296 type = BTRFS_ORDERED_NOCOW;
7297 len = min(len, em->len - (start - em->start));
7298 block_start = em->block_start + (start - em->start);
7300 if (can_nocow_extent(inode, start, &len, &orig_start,
7301 &orig_block_len, &ram_bytes, false, false) == 1) {
7302 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7310 struct extent_map *em2;
7312 /* We can NOCOW, so only need to reserve metadata space. */
7313 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7316 /* Our caller expects us to free the input extent map. */
7317 free_extent_map(em);
7319 btrfs_dec_nocow_writers(bg);
7320 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7324 space_reserved = true;
7326 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7327 orig_start, block_start,
7328 len, orig_block_len,
7330 btrfs_dec_nocow_writers(bg);
7331 if (type == BTRFS_ORDERED_PREALLOC) {
7332 free_extent_map(em);
7342 dio_data->nocow_done = true;
7344 /* Our caller expects us to free the input extent map. */
7345 free_extent_map(em);
7352 * If we could not allocate data space before locking the file
7353 * range and we can't do a NOCOW write, then we have to fail.
7355 if (!dio_data->data_space_reserved)
7359 * We have to COW and we have already reserved data space before,
7360 * so now we reserve only metadata.
7362 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7366 space_reserved = true;
7368 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7374 len = min(len, em->len - (start - em->start));
7376 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7377 prev_len - len, true);
7381 * We have created our ordered extent, so we can now release our reservation
7382 * for an outstanding extent.
7384 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7387 * Need to update the i_size under the extent lock so buffered
7388 * readers will get the updated i_size when we unlock.
7390 if (start + len > i_size_read(inode))
7391 i_size_write(inode, start + len);
7393 if (ret && space_reserved) {
7394 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7395 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7400 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7401 loff_t length, unsigned int flags, struct iomap *iomap,
7402 struct iomap *srcmap)
7404 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7405 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7406 struct extent_map *em;
7407 struct extent_state *cached_state = NULL;
7408 struct btrfs_dio_data *dio_data = iter->private;
7409 u64 lockstart, lockend;
7410 const bool write = !!(flags & IOMAP_WRITE);
7413 const u64 data_alloc_len = length;
7414 bool unlock_extents = false;
7417 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7418 * we're NOWAIT we may submit a bio for a partial range and return
7419 * EIOCBQUEUED, which would result in an errant short read.
7421 * The best way to handle this would be to allow for partial completions
7422 * of iocb's, so we could submit the partial bio, return and fault in
7423 * the rest of the pages, and then submit the io for the rest of the
7424 * range. However we don't have that currently, so simply return
7425 * -EAGAIN at this point so that the normal path is used.
7427 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7431 * Cap the size of reads to that usually seen in buffered I/O as we need
7432 * to allocate a contiguous array for the checksums.
7435 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7438 lockend = start + len - 1;
7441 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7442 * enough if we've written compressed pages to this area, so we need to
7443 * flush the dirty pages again to make absolutely sure that any
7444 * outstanding dirty pages are on disk - the first flush only starts
7445 * compression on the data, while keeping the pages locked, so by the
7446 * time the second flush returns we know bios for the compressed pages
7447 * were submitted and finished, and the pages no longer under writeback.
7449 * If we have a NOWAIT request and we have any pages in the range that
7450 * are locked, likely due to compression still in progress, we don't want
7451 * to block on page locks. We also don't want to block on pages marked as
7452 * dirty or under writeback (same as for the non-compression case).
7453 * iomap_dio_rw() did the same check, but after that and before we got
7454 * here, mmap'ed writes may have happened or buffered reads started
7455 * (readpage() and readahead(), which lock pages), as we haven't locked
7456 * the file range yet.
7458 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7459 &BTRFS_I(inode)->runtime_flags)) {
7460 if (flags & IOMAP_NOWAIT) {
7461 if (filemap_range_needs_writeback(inode->i_mapping,
7462 lockstart, lockend))
7465 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7466 start + length - 1);
7472 memset(dio_data, 0, sizeof(*dio_data));
7475 * We always try to allocate data space and must do it before locking
7476 * the file range, to avoid deadlocks with concurrent writes to the same
7477 * range if the range has several extents and the writes don't expand the
7478 * current i_size (the inode lock is taken in shared mode). If we fail to
7479 * allocate data space here we continue and later, after locking the
7480 * file range, we fail with ENOSPC only if we figure out we can not do a
7483 if (write && !(flags & IOMAP_NOWAIT)) {
7484 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7485 &dio_data->data_reserved,
7486 start, data_alloc_len, false);
7488 dio_data->data_space_reserved = true;
7489 else if (ret && !(BTRFS_I(inode)->flags &
7490 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7495 * If this errors out it's because we couldn't invalidate pagecache for
7496 * this range and we need to fallback to buffered IO, or we are doing a
7497 * NOWAIT read/write and we need to block.
7499 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7503 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7510 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7511 * io. INLINE is special, and we could probably kludge it in here, but
7512 * it's still buffered so for safety lets just fall back to the generic
7515 * For COMPRESSED we _have_ to read the entire extent in so we can
7516 * decompress it, so there will be buffering required no matter what we
7517 * do, so go ahead and fallback to buffered.
7519 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7520 * to buffered IO. Don't blame me, this is the price we pay for using
7523 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7524 em->block_start == EXTENT_MAP_INLINE) {
7525 free_extent_map(em);
7527 * If we are in a NOWAIT context, return -EAGAIN in order to
7528 * fallback to buffered IO. This is not only because we can
7529 * block with buffered IO (no support for NOWAIT semantics at
7530 * the moment) but also to avoid returning short reads to user
7531 * space - this happens if we were able to read some data from
7532 * previous non-compressed extents and then when we fallback to
7533 * buffered IO, at btrfs_file_read_iter() by calling
7534 * filemap_read(), we fail to fault in pages for the read buffer,
7535 * in which case filemap_read() returns a short read (the number
7536 * of bytes previously read is > 0, so it does not return -EFAULT).
7538 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7542 len = min(len, em->len - (start - em->start));
7545 * If we have a NOWAIT request and the range contains multiple extents
7546 * (or a mix of extents and holes), then we return -EAGAIN to make the
7547 * caller fallback to a context where it can do a blocking (without
7548 * NOWAIT) request. This way we avoid doing partial IO and returning
7549 * success to the caller, which is not optimal for writes and for reads
7550 * it can result in unexpected behaviour for an application.
7552 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7553 * iomap_dio_rw(), we can end up returning less data then what the caller
7554 * asked for, resulting in an unexpected, and incorrect, short read.
7555 * That is, the caller asked to read N bytes and we return less than that,
7556 * which is wrong unless we are crossing EOF. This happens if we get a
7557 * page fault error when trying to fault in pages for the buffer that is
7558 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7559 * have previously submitted bios for other extents in the range, in
7560 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7561 * those bios have completed by the time we get the page fault error,
7562 * which we return back to our caller - we should only return EIOCBQUEUED
7563 * after we have submitted bios for all the extents in the range.
7565 if ((flags & IOMAP_NOWAIT) && len < length) {
7566 free_extent_map(em);
7572 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7576 unlock_extents = true;
7577 /* Recalc len in case the new em is smaller than requested */
7578 len = min(len, em->len - (start - em->start));
7579 if (dio_data->data_space_reserved) {
7581 u64 release_len = 0;
7583 if (dio_data->nocow_done) {
7584 release_offset = start;
7585 release_len = data_alloc_len;
7586 } else if (len < data_alloc_len) {
7587 release_offset = start + len;
7588 release_len = data_alloc_len - len;
7591 if (release_len > 0)
7592 btrfs_free_reserved_data_space(BTRFS_I(inode),
7593 dio_data->data_reserved,
7599 * We need to unlock only the end area that we aren't using.
7600 * The rest is going to be unlocked by the endio routine.
7602 lockstart = start + len;
7603 if (lockstart < lockend)
7604 unlock_extents = true;
7608 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7611 free_extent_state(cached_state);
7614 * Translate extent map information to iomap.
7615 * We trim the extents (and move the addr) even though iomap code does
7616 * that, since we have locked only the parts we are performing I/O in.
7618 if ((em->block_start == EXTENT_MAP_HOLE) ||
7619 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7620 iomap->addr = IOMAP_NULL_ADDR;
7621 iomap->type = IOMAP_HOLE;
7623 iomap->addr = em->block_start + (start - em->start);
7624 iomap->type = IOMAP_MAPPED;
7626 iomap->offset = start;
7627 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7628 iomap->length = len;
7629 free_extent_map(em);
7634 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7637 if (dio_data->data_space_reserved) {
7638 btrfs_free_reserved_data_space(BTRFS_I(inode),
7639 dio_data->data_reserved,
7640 start, data_alloc_len);
7641 extent_changeset_free(dio_data->data_reserved);
7647 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7648 ssize_t written, unsigned int flags, struct iomap *iomap)
7650 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7651 struct btrfs_dio_data *dio_data = iter->private;
7652 size_t submitted = dio_data->submitted;
7653 const bool write = !!(flags & IOMAP_WRITE);
7656 if (!write && (iomap->type == IOMAP_HOLE)) {
7657 /* If reading from a hole, unlock and return */
7658 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7663 if (submitted < length) {
7665 length -= submitted;
7667 btrfs_mark_ordered_io_finished(BTRFS_I(inode), NULL,
7668 pos, length, false);
7670 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7671 pos + length - 1, NULL);
7675 btrfs_put_ordered_extent(dio_data->ordered);
7676 dio_data->ordered = NULL;
7680 extent_changeset_free(dio_data->data_reserved);
7684 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7686 struct btrfs_dio_private *dip =
7687 container_of(bbio, struct btrfs_dio_private, bbio);
7688 struct btrfs_inode *inode = bbio->inode;
7689 struct bio *bio = &bbio->bio;
7691 if (bio->bi_status) {
7692 btrfs_warn(inode->root->fs_info,
7693 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7694 btrfs_ino(inode), bio->bi_opf,
7695 dip->file_offset, dip->bytes, bio->bi_status);
7698 if (btrfs_op(bio) == BTRFS_MAP_WRITE)
7699 btrfs_mark_ordered_io_finished(inode, NULL, dip->file_offset,
7700 dip->bytes, !bio->bi_status);
7702 unlock_extent(&inode->io_tree, dip->file_offset,
7703 dip->file_offset + dip->bytes - 1, NULL);
7705 bbio->bio.bi_private = bbio->private;
7706 iomap_dio_bio_end_io(bio);
7709 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7712 struct btrfs_bio *bbio = btrfs_bio(bio);
7713 struct btrfs_dio_private *dip =
7714 container_of(bbio, struct btrfs_dio_private, bbio);
7715 struct btrfs_dio_data *dio_data = iter->private;
7717 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7718 btrfs_dio_end_io, bio->bi_private);
7719 bbio->inode = BTRFS_I(iter->inode);
7720 bbio->file_offset = file_offset;
7722 dip->file_offset = file_offset;
7723 dip->bytes = bio->bi_iter.bi_size;
7725 dio_data->submitted += bio->bi_iter.bi_size;
7728 * Check if we are doing a partial write. If we are, we need to split
7729 * the ordered extent to match the submitted bio. Hang on to the
7730 * remaining unfinishable ordered_extent in dio_data so that it can be
7731 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7732 * remaining pages is blocked on the outstanding ordered extent.
7734 if (iter->flags & IOMAP_WRITE) {
7737 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7739 btrfs_bio_end_io(bbio, errno_to_blk_status(ret));
7744 btrfs_submit_bio(bbio, 0);
7747 static const struct iomap_ops btrfs_dio_iomap_ops = {
7748 .iomap_begin = btrfs_dio_iomap_begin,
7749 .iomap_end = btrfs_dio_iomap_end,
7752 static const struct iomap_dio_ops btrfs_dio_ops = {
7753 .submit_io = btrfs_dio_submit_io,
7754 .bio_set = &btrfs_dio_bioset,
7757 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7759 struct btrfs_dio_data data = { 0 };
7761 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7762 IOMAP_DIO_PARTIAL, &data, done_before);
7765 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7768 struct btrfs_dio_data data = { 0 };
7770 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7771 IOMAP_DIO_PARTIAL, &data, done_before);
7774 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7779 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7784 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7785 * file range (0 to LLONG_MAX), but that is not enough if we have
7786 * compression enabled. The first filemap_fdatawrite_range() only kicks
7787 * in the compression of data (in an async thread) and will return
7788 * before the compression is done and writeback is started. A second
7789 * filemap_fdatawrite_range() is needed to wait for the compression to
7790 * complete and writeback to start. We also need to wait for ordered
7791 * extents to complete, because our fiemap implementation uses mainly
7792 * file extent items to list the extents, searching for extent maps
7793 * only for file ranges with holes or prealloc extents to figure out
7794 * if we have delalloc in those ranges.
7796 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7797 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7802 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7805 static int btrfs_writepages(struct address_space *mapping,
7806 struct writeback_control *wbc)
7808 return extent_writepages(mapping, wbc);
7811 static void btrfs_readahead(struct readahead_control *rac)
7813 extent_readahead(rac);
7817 * For release_folio() and invalidate_folio() we have a race window where
7818 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7819 * If we continue to release/invalidate the page, we could cause use-after-free
7820 * for subpage spinlock. So this function is to spin and wait for subpage
7823 static void wait_subpage_spinlock(struct page *page)
7825 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7826 struct btrfs_subpage *subpage;
7828 if (!btrfs_is_subpage(fs_info, page))
7831 ASSERT(PagePrivate(page) && page->private);
7832 subpage = (struct btrfs_subpage *)page->private;
7835 * This may look insane as we just acquire the spinlock and release it,
7836 * without doing anything. But we just want to make sure no one is
7837 * still holding the subpage spinlock.
7838 * And since the page is not dirty nor writeback, and we have page
7839 * locked, the only possible way to hold a spinlock is from the endio
7840 * function to clear page writeback.
7842 * Here we just acquire the spinlock so that all existing callers
7843 * should exit and we're safe to release/invalidate the page.
7845 spin_lock_irq(&subpage->lock);
7846 spin_unlock_irq(&subpage->lock);
7849 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7851 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7854 wait_subpage_spinlock(&folio->page);
7855 clear_page_extent_mapped(&folio->page);
7860 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7862 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7864 return __btrfs_release_folio(folio, gfp_flags);
7867 #ifdef CONFIG_MIGRATION
7868 static int btrfs_migrate_folio(struct address_space *mapping,
7869 struct folio *dst, struct folio *src,
7870 enum migrate_mode mode)
7872 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7874 if (ret != MIGRATEPAGE_SUCCESS)
7877 if (folio_test_ordered(src)) {
7878 folio_clear_ordered(src);
7879 folio_set_ordered(dst);
7882 return MIGRATEPAGE_SUCCESS;
7885 #define btrfs_migrate_folio NULL
7888 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7891 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7892 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7893 struct extent_io_tree *tree = &inode->io_tree;
7894 struct extent_state *cached_state = NULL;
7895 u64 page_start = folio_pos(folio);
7896 u64 page_end = page_start + folio_size(folio) - 1;
7898 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7901 * We have folio locked so no new ordered extent can be created on this
7902 * page, nor bio can be submitted for this folio.
7904 * But already submitted bio can still be finished on this folio.
7905 * Furthermore, endio function won't skip folio which has Ordered
7906 * (Private2) already cleared, so it's possible for endio and
7907 * invalidate_folio to do the same ordered extent accounting twice
7910 * So here we wait for any submitted bios to finish, so that we won't
7911 * do double ordered extent accounting on the same folio.
7913 folio_wait_writeback(folio);
7914 wait_subpage_spinlock(&folio->page);
7917 * For subpage case, we have call sites like
7918 * btrfs_punch_hole_lock_range() which passes range not aligned to
7920 * If the range doesn't cover the full folio, we don't need to and
7921 * shouldn't clear page extent mapped, as folio->private can still
7922 * record subpage dirty bits for other part of the range.
7924 * For cases that invalidate the full folio even the range doesn't
7925 * cover the full folio, like invalidating the last folio, we're
7926 * still safe to wait for ordered extent to finish.
7928 if (!(offset == 0 && length == folio_size(folio))) {
7929 btrfs_release_folio(folio, GFP_NOFS);
7933 if (!inode_evicting)
7934 lock_extent(tree, page_start, page_end, &cached_state);
7937 while (cur < page_end) {
7938 struct btrfs_ordered_extent *ordered;
7941 u32 extra_flags = 0;
7943 ordered = btrfs_lookup_first_ordered_range(inode, cur,
7944 page_end + 1 - cur);
7946 range_end = page_end;
7948 * No ordered extent covering this range, we are safe
7949 * to delete all extent states in the range.
7951 extra_flags = EXTENT_CLEAR_ALL_BITS;
7954 if (ordered->file_offset > cur) {
7956 * There is a range between [cur, oe->file_offset) not
7957 * covered by any ordered extent.
7958 * We are safe to delete all extent states, and handle
7959 * the ordered extent in the next iteration.
7961 range_end = ordered->file_offset - 1;
7962 extra_flags = EXTENT_CLEAR_ALL_BITS;
7966 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
7968 ASSERT(range_end + 1 - cur < U32_MAX);
7969 range_len = range_end + 1 - cur;
7970 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
7972 * If Ordered (Private2) is cleared, it means endio has
7973 * already been executed for the range.
7974 * We can't delete the extent states as
7975 * btrfs_finish_ordered_io() may still use some of them.
7979 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
7982 * IO on this page will never be started, so we need to account
7983 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
7984 * here, must leave that up for the ordered extent completion.
7986 * This will also unlock the range for incoming
7987 * btrfs_finish_ordered_io().
7989 if (!inode_evicting)
7990 clear_extent_bit(tree, cur, range_end,
7992 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7993 EXTENT_DEFRAG, &cached_state);
7995 spin_lock_irq(&inode->ordered_tree.lock);
7996 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7997 ordered->truncated_len = min(ordered->truncated_len,
7998 cur - ordered->file_offset);
7999 spin_unlock_irq(&inode->ordered_tree.lock);
8002 * If the ordered extent has finished, we're safe to delete all
8003 * the extent states of the range, otherwise
8004 * btrfs_finish_ordered_io() will get executed by endio for
8005 * other pages, so we can't delete extent states.
8007 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8008 cur, range_end + 1 - cur)) {
8009 btrfs_finish_ordered_io(ordered);
8011 * The ordered extent has finished, now we're again
8012 * safe to delete all extent states of the range.
8014 extra_flags = EXTENT_CLEAR_ALL_BITS;
8018 btrfs_put_ordered_extent(ordered);
8020 * Qgroup reserved space handler
8021 * Sector(s) here will be either:
8023 * 1) Already written to disk or bio already finished
8024 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8025 * Qgroup will be handled by its qgroup_record then.
8026 * btrfs_qgroup_free_data() call will do nothing here.
8028 * 2) Not written to disk yet
8029 * Then btrfs_qgroup_free_data() call will clear the
8030 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8031 * reserved data space.
8032 * Since the IO will never happen for this page.
8034 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8035 if (!inode_evicting) {
8036 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8037 EXTENT_DELALLOC | EXTENT_UPTODATE |
8038 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8039 extra_flags, &cached_state);
8041 cur = range_end + 1;
8044 * We have iterated through all ordered extents of the page, the page
8045 * should not have Ordered (Private2) anymore, or the above iteration
8046 * did something wrong.
8048 ASSERT(!folio_test_ordered(folio));
8049 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8050 if (!inode_evicting)
8051 __btrfs_release_folio(folio, GFP_NOFS);
8052 clear_page_extent_mapped(&folio->page);
8056 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8057 * called from a page fault handler when a page is first dirtied. Hence we must
8058 * be careful to check for EOF conditions here. We set the page up correctly
8059 * for a written page which means we get ENOSPC checking when writing into
8060 * holes and correct delalloc and unwritten extent mapping on filesystems that
8061 * support these features.
8063 * We are not allowed to take the i_mutex here so we have to play games to
8064 * protect against truncate races as the page could now be beyond EOF. Because
8065 * truncate_setsize() writes the inode size before removing pages, once we have
8066 * the page lock we can determine safely if the page is beyond EOF. If it is not
8067 * beyond EOF, then the page is guaranteed safe against truncation until we
8070 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8072 struct page *page = vmf->page;
8073 struct inode *inode = file_inode(vmf->vma->vm_file);
8074 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8075 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8076 struct btrfs_ordered_extent *ordered;
8077 struct extent_state *cached_state = NULL;
8078 struct extent_changeset *data_reserved = NULL;
8079 unsigned long zero_start;
8089 reserved_space = PAGE_SIZE;
8091 sb_start_pagefault(inode->i_sb);
8092 page_start = page_offset(page);
8093 page_end = page_start + PAGE_SIZE - 1;
8097 * Reserving delalloc space after obtaining the page lock can lead to
8098 * deadlock. For example, if a dirty page is locked by this function
8099 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8100 * dirty page write out, then the btrfs_writepages() function could
8101 * end up waiting indefinitely to get a lock on the page currently
8102 * being processed by btrfs_page_mkwrite() function.
8104 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8105 page_start, reserved_space);
8107 ret2 = file_update_time(vmf->vma->vm_file);
8111 ret = vmf_error(ret2);
8117 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8119 down_read(&BTRFS_I(inode)->i_mmap_lock);
8121 size = i_size_read(inode);
8123 if ((page->mapping != inode->i_mapping) ||
8124 (page_start >= size)) {
8125 /* page got truncated out from underneath us */
8128 wait_on_page_writeback(page);
8130 lock_extent(io_tree, page_start, page_end, &cached_state);
8131 ret2 = set_page_extent_mapped(page);
8133 ret = vmf_error(ret2);
8134 unlock_extent(io_tree, page_start, page_end, &cached_state);
8139 * we can't set the delalloc bits if there are pending ordered
8140 * extents. Drop our locks and wait for them to finish
8142 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8145 unlock_extent(io_tree, page_start, page_end, &cached_state);
8147 up_read(&BTRFS_I(inode)->i_mmap_lock);
8148 btrfs_start_ordered_extent(ordered);
8149 btrfs_put_ordered_extent(ordered);
8153 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8154 reserved_space = round_up(size - page_start,
8155 fs_info->sectorsize);
8156 if (reserved_space < PAGE_SIZE) {
8157 end = page_start + reserved_space - 1;
8158 btrfs_delalloc_release_space(BTRFS_I(inode),
8159 data_reserved, page_start,
8160 PAGE_SIZE - reserved_space, true);
8165 * page_mkwrite gets called when the page is firstly dirtied after it's
8166 * faulted in, but write(2) could also dirty a page and set delalloc
8167 * bits, thus in this case for space account reason, we still need to
8168 * clear any delalloc bits within this page range since we have to
8169 * reserve data&meta space before lock_page() (see above comments).
8171 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8172 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8173 EXTENT_DEFRAG, &cached_state);
8175 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8178 unlock_extent(io_tree, page_start, page_end, &cached_state);
8179 ret = VM_FAULT_SIGBUS;
8183 /* page is wholly or partially inside EOF */
8184 if (page_start + PAGE_SIZE > size)
8185 zero_start = offset_in_page(size);
8187 zero_start = PAGE_SIZE;
8189 if (zero_start != PAGE_SIZE)
8190 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8192 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8193 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8194 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8196 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8198 unlock_extent(io_tree, page_start, page_end, &cached_state);
8199 up_read(&BTRFS_I(inode)->i_mmap_lock);
8201 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8202 sb_end_pagefault(inode->i_sb);
8203 extent_changeset_free(data_reserved);
8204 return VM_FAULT_LOCKED;
8208 up_read(&BTRFS_I(inode)->i_mmap_lock);
8210 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8211 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8212 reserved_space, (ret != 0));
8214 sb_end_pagefault(inode->i_sb);
8215 extent_changeset_free(data_reserved);
8219 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8221 struct btrfs_truncate_control control = {
8223 .ino = btrfs_ino(inode),
8224 .min_type = BTRFS_EXTENT_DATA_KEY,
8225 .clear_extent_range = true,
8227 struct btrfs_root *root = inode->root;
8228 struct btrfs_fs_info *fs_info = root->fs_info;
8229 struct btrfs_block_rsv *rsv;
8231 struct btrfs_trans_handle *trans;
8232 u64 mask = fs_info->sectorsize - 1;
8233 u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8235 if (!skip_writeback) {
8236 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8237 inode->vfs_inode.i_size & (~mask),
8244 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8245 * things going on here:
8247 * 1) We need to reserve space to update our inode.
8249 * 2) We need to have something to cache all the space that is going to
8250 * be free'd up by the truncate operation, but also have some slack
8251 * space reserved in case it uses space during the truncate (thank you
8252 * very much snapshotting).
8254 * And we need these to be separate. The fact is we can use a lot of
8255 * space doing the truncate, and we have no earthly idea how much space
8256 * we will use, so we need the truncate reservation to be separate so it
8257 * doesn't end up using space reserved for updating the inode. We also
8258 * need to be able to stop the transaction and start a new one, which
8259 * means we need to be able to update the inode several times, and we
8260 * have no idea of knowing how many times that will be, so we can't just
8261 * reserve 1 item for the entirety of the operation, so that has to be
8262 * done separately as well.
8264 * So that leaves us with
8266 * 1) rsv - for the truncate reservation, which we will steal from the
8267 * transaction reservation.
8268 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8269 * updating the inode.
8271 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8274 rsv->size = min_size;
8275 rsv->failfast = true;
8278 * 1 for the truncate slack space
8279 * 1 for updating the inode.
8281 trans = btrfs_start_transaction(root, 2);
8282 if (IS_ERR(trans)) {
8283 ret = PTR_ERR(trans);
8287 /* Migrate the slack space for the truncate to our reserve */
8288 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8292 trans->block_rsv = rsv;
8295 struct extent_state *cached_state = NULL;
8296 const u64 new_size = inode->vfs_inode.i_size;
8297 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8299 control.new_size = new_size;
8300 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8302 * We want to drop from the next block forward in case this new
8303 * size is not block aligned since we will be keeping the last
8304 * block of the extent just the way it is.
8306 btrfs_drop_extent_map_range(inode,
8307 ALIGN(new_size, fs_info->sectorsize),
8310 ret = btrfs_truncate_inode_items(trans, root, &control);
8312 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8313 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8315 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8317 trans->block_rsv = &fs_info->trans_block_rsv;
8318 if (ret != -ENOSPC && ret != -EAGAIN)
8321 ret = btrfs_update_inode(trans, root, inode);
8325 btrfs_end_transaction(trans);
8326 btrfs_btree_balance_dirty(fs_info);
8328 trans = btrfs_start_transaction(root, 2);
8329 if (IS_ERR(trans)) {
8330 ret = PTR_ERR(trans);
8335 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8336 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8337 rsv, min_size, false);
8338 BUG_ON(ret); /* shouldn't happen */
8339 trans->block_rsv = rsv;
8343 * We can't call btrfs_truncate_block inside a trans handle as we could
8344 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8345 * know we've truncated everything except the last little bit, and can
8346 * do btrfs_truncate_block and then update the disk_i_size.
8348 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8349 btrfs_end_transaction(trans);
8350 btrfs_btree_balance_dirty(fs_info);
8352 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8355 trans = btrfs_start_transaction(root, 1);
8356 if (IS_ERR(trans)) {
8357 ret = PTR_ERR(trans);
8360 btrfs_inode_safe_disk_i_size_write(inode, 0);
8366 trans->block_rsv = &fs_info->trans_block_rsv;
8367 ret2 = btrfs_update_inode(trans, root, inode);
8371 ret2 = btrfs_end_transaction(trans);
8374 btrfs_btree_balance_dirty(fs_info);
8377 btrfs_free_block_rsv(fs_info, rsv);
8379 * So if we truncate and then write and fsync we normally would just
8380 * write the extents that changed, which is a problem if we need to
8381 * first truncate that entire inode. So set this flag so we write out
8382 * all of the extents in the inode to the sync log so we're completely
8385 * If no extents were dropped or trimmed we don't need to force the next
8386 * fsync to truncate all the inode's items from the log and re-log them
8387 * all. This means the truncate operation did not change the file size,
8388 * or changed it to a smaller size but there was only an implicit hole
8389 * between the old i_size and the new i_size, and there were no prealloc
8390 * extents beyond i_size to drop.
8392 if (control.extents_found > 0)
8393 btrfs_set_inode_full_sync(inode);
8398 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8401 struct inode *inode;
8403 inode = new_inode(dir->i_sb);
8406 * Subvolumes don't inherit the sgid bit or the parent's gid if
8407 * the parent's sgid bit is set. This is probably a bug.
8409 inode_init_owner(idmap, inode, NULL,
8410 S_IFDIR | (~current_umask() & S_IRWXUGO));
8411 inode->i_op = &btrfs_dir_inode_operations;
8412 inode->i_fop = &btrfs_dir_file_operations;
8417 struct inode *btrfs_alloc_inode(struct super_block *sb)
8419 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8420 struct btrfs_inode *ei;
8421 struct inode *inode;
8423 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8430 ei->last_sub_trans = 0;
8431 ei->logged_trans = 0;
8432 ei->delalloc_bytes = 0;
8433 ei->new_delalloc_bytes = 0;
8434 ei->defrag_bytes = 0;
8435 ei->disk_i_size = 0;
8439 ei->index_cnt = (u64)-1;
8441 ei->last_unlink_trans = 0;
8442 ei->last_reflink_trans = 0;
8443 ei->last_log_commit = 0;
8445 spin_lock_init(&ei->lock);
8446 ei->outstanding_extents = 0;
8447 if (sb->s_magic != BTRFS_TEST_MAGIC)
8448 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8449 BTRFS_BLOCK_RSV_DELALLOC);
8450 ei->runtime_flags = 0;
8451 ei->prop_compress = BTRFS_COMPRESS_NONE;
8452 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8454 ei->delayed_node = NULL;
8456 ei->i_otime.tv_sec = 0;
8457 ei->i_otime.tv_nsec = 0;
8459 inode = &ei->vfs_inode;
8460 extent_map_tree_init(&ei->extent_tree);
8461 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8462 ei->io_tree.inode = ei;
8463 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8464 IO_TREE_INODE_FILE_EXTENT);
8465 atomic_set(&ei->sync_writers, 0);
8466 mutex_init(&ei->log_mutex);
8467 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8468 INIT_LIST_HEAD(&ei->delalloc_inodes);
8469 INIT_LIST_HEAD(&ei->delayed_iput);
8470 RB_CLEAR_NODE(&ei->rb_node);
8471 init_rwsem(&ei->i_mmap_lock);
8476 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8477 void btrfs_test_destroy_inode(struct inode *inode)
8479 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8480 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8484 void btrfs_free_inode(struct inode *inode)
8486 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8489 void btrfs_destroy_inode(struct inode *vfs_inode)
8491 struct btrfs_ordered_extent *ordered;
8492 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8493 struct btrfs_root *root = inode->root;
8494 bool freespace_inode;
8496 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8497 WARN_ON(vfs_inode->i_data.nrpages);
8498 WARN_ON(inode->block_rsv.reserved);
8499 WARN_ON(inode->block_rsv.size);
8500 WARN_ON(inode->outstanding_extents);
8501 if (!S_ISDIR(vfs_inode->i_mode)) {
8502 WARN_ON(inode->delalloc_bytes);
8503 WARN_ON(inode->new_delalloc_bytes);
8505 WARN_ON(inode->csum_bytes);
8506 WARN_ON(inode->defrag_bytes);
8509 * This can happen where we create an inode, but somebody else also
8510 * created the same inode and we need to destroy the one we already
8517 * If this is a free space inode do not take the ordered extents lockdep
8520 freespace_inode = btrfs_is_free_space_inode(inode);
8523 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8527 btrfs_err(root->fs_info,
8528 "found ordered extent %llu %llu on inode cleanup",
8529 ordered->file_offset, ordered->num_bytes);
8531 if (!freespace_inode)
8532 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8534 btrfs_remove_ordered_extent(inode, ordered);
8535 btrfs_put_ordered_extent(ordered);
8536 btrfs_put_ordered_extent(ordered);
8539 btrfs_qgroup_check_reserved_leak(inode);
8540 inode_tree_del(inode);
8541 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8542 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8543 btrfs_put_root(inode->root);
8546 int btrfs_drop_inode(struct inode *inode)
8548 struct btrfs_root *root = BTRFS_I(inode)->root;
8553 /* the snap/subvol tree is on deleting */
8554 if (btrfs_root_refs(&root->root_item) == 0)
8557 return generic_drop_inode(inode);
8560 static void init_once(void *foo)
8562 struct btrfs_inode *ei = foo;
8564 inode_init_once(&ei->vfs_inode);
8567 void __cold btrfs_destroy_cachep(void)
8570 * Make sure all delayed rcu free inodes are flushed before we
8574 bioset_exit(&btrfs_dio_bioset);
8575 kmem_cache_destroy(btrfs_inode_cachep);
8578 int __init btrfs_init_cachep(void)
8580 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8581 sizeof(struct btrfs_inode), 0,
8582 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8584 if (!btrfs_inode_cachep)
8587 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8588 offsetof(struct btrfs_dio_private, bbio.bio),
8594 btrfs_destroy_cachep();
8598 static int btrfs_getattr(struct mnt_idmap *idmap,
8599 const struct path *path, struct kstat *stat,
8600 u32 request_mask, unsigned int flags)
8604 struct inode *inode = d_inode(path->dentry);
8605 u32 blocksize = inode->i_sb->s_blocksize;
8606 u32 bi_flags = BTRFS_I(inode)->flags;
8607 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8609 stat->result_mask |= STATX_BTIME;
8610 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8611 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8612 if (bi_flags & BTRFS_INODE_APPEND)
8613 stat->attributes |= STATX_ATTR_APPEND;
8614 if (bi_flags & BTRFS_INODE_COMPRESS)
8615 stat->attributes |= STATX_ATTR_COMPRESSED;
8616 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8617 stat->attributes |= STATX_ATTR_IMMUTABLE;
8618 if (bi_flags & BTRFS_INODE_NODUMP)
8619 stat->attributes |= STATX_ATTR_NODUMP;
8620 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8621 stat->attributes |= STATX_ATTR_VERITY;
8623 stat->attributes_mask |= (STATX_ATTR_APPEND |
8624 STATX_ATTR_COMPRESSED |
8625 STATX_ATTR_IMMUTABLE |
8628 generic_fillattr(idmap, inode, stat);
8629 stat->dev = BTRFS_I(inode)->root->anon_dev;
8631 spin_lock(&BTRFS_I(inode)->lock);
8632 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8633 inode_bytes = inode_get_bytes(inode);
8634 spin_unlock(&BTRFS_I(inode)->lock);
8635 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8636 ALIGN(delalloc_bytes, blocksize)) >> 9;
8640 static int btrfs_rename_exchange(struct inode *old_dir,
8641 struct dentry *old_dentry,
8642 struct inode *new_dir,
8643 struct dentry *new_dentry)
8645 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8646 struct btrfs_trans_handle *trans;
8647 unsigned int trans_num_items;
8648 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8649 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8650 struct inode *new_inode = new_dentry->d_inode;
8651 struct inode *old_inode = old_dentry->d_inode;
8652 struct timespec64 ctime = current_time(old_inode);
8653 struct btrfs_rename_ctx old_rename_ctx;
8654 struct btrfs_rename_ctx new_rename_ctx;
8655 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8656 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8661 bool need_abort = false;
8662 struct fscrypt_name old_fname, new_fname;
8663 struct fscrypt_str *old_name, *new_name;
8666 * For non-subvolumes allow exchange only within one subvolume, in the
8667 * same inode namespace. Two subvolumes (represented as directory) can
8668 * be exchanged as they're a logical link and have a fixed inode number.
8671 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8672 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8675 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8679 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8681 fscrypt_free_filename(&old_fname);
8685 old_name = &old_fname.disk_name;
8686 new_name = &new_fname.disk_name;
8688 /* close the race window with snapshot create/destroy ioctl */
8689 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8690 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8691 down_read(&fs_info->subvol_sem);
8695 * 1 to remove old dir item
8696 * 1 to remove old dir index
8697 * 1 to add new dir item
8698 * 1 to add new dir index
8699 * 1 to update parent inode
8701 * If the parents are the same, we only need to account for one
8703 trans_num_items = (old_dir == new_dir ? 9 : 10);
8704 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8706 * 1 to remove old root ref
8707 * 1 to remove old root backref
8708 * 1 to add new root ref
8709 * 1 to add new root backref
8711 trans_num_items += 4;
8714 * 1 to update inode item
8715 * 1 to remove old inode ref
8716 * 1 to add new inode ref
8718 trans_num_items += 3;
8720 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8721 trans_num_items += 4;
8723 trans_num_items += 3;
8724 trans = btrfs_start_transaction(root, trans_num_items);
8725 if (IS_ERR(trans)) {
8726 ret = PTR_ERR(trans);
8731 ret = btrfs_record_root_in_trans(trans, dest);
8737 * We need to find a free sequence number both in the source and
8738 * in the destination directory for the exchange.
8740 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8743 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8747 BTRFS_I(old_inode)->dir_index = 0ULL;
8748 BTRFS_I(new_inode)->dir_index = 0ULL;
8750 /* Reference for the source. */
8751 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8752 /* force full log commit if subvolume involved. */
8753 btrfs_set_log_full_commit(trans);
8755 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8756 btrfs_ino(BTRFS_I(new_dir)),
8763 /* And now for the dest. */
8764 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8765 /* force full log commit if subvolume involved. */
8766 btrfs_set_log_full_commit(trans);
8768 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8769 btrfs_ino(BTRFS_I(old_dir)),
8773 btrfs_abort_transaction(trans, ret);
8778 /* Update inode version and ctime/mtime. */
8779 inode_inc_iversion(old_dir);
8780 inode_inc_iversion(new_dir);
8781 inode_inc_iversion(old_inode);
8782 inode_inc_iversion(new_inode);
8783 old_dir->i_mtime = ctime;
8784 old_dir->i_ctime = ctime;
8785 new_dir->i_mtime = ctime;
8786 new_dir->i_ctime = ctime;
8787 old_inode->i_ctime = ctime;
8788 new_inode->i_ctime = ctime;
8790 if (old_dentry->d_parent != new_dentry->d_parent) {
8791 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8792 BTRFS_I(old_inode), 1);
8793 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8794 BTRFS_I(new_inode), 1);
8797 /* src is a subvolume */
8798 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8799 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8800 } else { /* src is an inode */
8801 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8802 BTRFS_I(old_dentry->d_inode),
8803 old_name, &old_rename_ctx);
8805 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8808 btrfs_abort_transaction(trans, ret);
8812 /* dest is a subvolume */
8813 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8814 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8815 } else { /* dest is an inode */
8816 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8817 BTRFS_I(new_dentry->d_inode),
8818 new_name, &new_rename_ctx);
8820 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8823 btrfs_abort_transaction(trans, ret);
8827 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8828 new_name, 0, old_idx);
8830 btrfs_abort_transaction(trans, ret);
8834 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8835 old_name, 0, new_idx);
8837 btrfs_abort_transaction(trans, ret);
8841 if (old_inode->i_nlink == 1)
8842 BTRFS_I(old_inode)->dir_index = old_idx;
8843 if (new_inode->i_nlink == 1)
8844 BTRFS_I(new_inode)->dir_index = new_idx;
8847 * Now pin the logs of the roots. We do it to ensure that no other task
8848 * can sync the logs while we are in progress with the rename, because
8849 * that could result in an inconsistency in case any of the inodes that
8850 * are part of this rename operation were logged before.
8852 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8853 btrfs_pin_log_trans(root);
8854 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8855 btrfs_pin_log_trans(dest);
8857 /* Do the log updates for all inodes. */
8858 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8859 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8860 old_rename_ctx.index, new_dentry->d_parent);
8861 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8862 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8863 new_rename_ctx.index, old_dentry->d_parent);
8865 /* Now unpin the logs. */
8866 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8867 btrfs_end_log_trans(root);
8868 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8869 btrfs_end_log_trans(dest);
8871 ret2 = btrfs_end_transaction(trans);
8872 ret = ret ? ret : ret2;
8874 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8875 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8876 up_read(&fs_info->subvol_sem);
8878 fscrypt_free_filename(&new_fname);
8879 fscrypt_free_filename(&old_fname);
8883 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8886 struct inode *inode;
8888 inode = new_inode(dir->i_sb);
8890 inode_init_owner(idmap, inode, dir,
8891 S_IFCHR | WHITEOUT_MODE);
8892 inode->i_op = &btrfs_special_inode_operations;
8893 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8898 static int btrfs_rename(struct mnt_idmap *idmap,
8899 struct inode *old_dir, struct dentry *old_dentry,
8900 struct inode *new_dir, struct dentry *new_dentry,
8903 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8904 struct btrfs_new_inode_args whiteout_args = {
8906 .dentry = old_dentry,
8908 struct btrfs_trans_handle *trans;
8909 unsigned int trans_num_items;
8910 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8911 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8912 struct inode *new_inode = d_inode(new_dentry);
8913 struct inode *old_inode = d_inode(old_dentry);
8914 struct btrfs_rename_ctx rename_ctx;
8918 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8919 struct fscrypt_name old_fname, new_fname;
8921 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8924 /* we only allow rename subvolume link between subvolumes */
8925 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8928 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8929 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
8932 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8933 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8936 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8940 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8942 fscrypt_free_filename(&old_fname);
8946 /* check for collisions, even if the name isn't there */
8947 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
8949 if (ret == -EEXIST) {
8951 * eexist without a new_inode */
8952 if (WARN_ON(!new_inode)) {
8953 goto out_fscrypt_names;
8956 /* maybe -EOVERFLOW */
8957 goto out_fscrypt_names;
8963 * we're using rename to replace one file with another. Start IO on it
8964 * now so we don't add too much work to the end of the transaction
8966 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8967 filemap_flush(old_inode->i_mapping);
8969 if (flags & RENAME_WHITEOUT) {
8970 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
8971 if (!whiteout_args.inode) {
8973 goto out_fscrypt_names;
8975 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
8977 goto out_whiteout_inode;
8979 /* 1 to update the old parent inode. */
8980 trans_num_items = 1;
8983 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8984 /* Close the race window with snapshot create/destroy ioctl */
8985 down_read(&fs_info->subvol_sem);
8987 * 1 to remove old root ref
8988 * 1 to remove old root backref
8989 * 1 to add new root ref
8990 * 1 to add new root backref
8992 trans_num_items += 4;
8996 * 1 to remove old inode ref
8997 * 1 to add new inode ref
8999 trans_num_items += 3;
9002 * 1 to remove old dir item
9003 * 1 to remove old dir index
9004 * 1 to add new dir item
9005 * 1 to add new dir index
9007 trans_num_items += 4;
9008 /* 1 to update new parent inode if it's not the same as the old parent */
9009 if (new_dir != old_dir)
9014 * 1 to remove inode ref
9015 * 1 to remove dir item
9016 * 1 to remove dir index
9017 * 1 to possibly add orphan item
9019 trans_num_items += 5;
9021 trans = btrfs_start_transaction(root, trans_num_items);
9022 if (IS_ERR(trans)) {
9023 ret = PTR_ERR(trans);
9028 ret = btrfs_record_root_in_trans(trans, dest);
9033 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9037 BTRFS_I(old_inode)->dir_index = 0ULL;
9038 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9039 /* force full log commit if subvolume involved. */
9040 btrfs_set_log_full_commit(trans);
9042 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9043 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9049 inode_inc_iversion(old_dir);
9050 inode_inc_iversion(new_dir);
9051 inode_inc_iversion(old_inode);
9052 old_dir->i_mtime = current_time(old_dir);
9053 old_dir->i_ctime = old_dir->i_mtime;
9054 new_dir->i_mtime = old_dir->i_mtime;
9055 new_dir->i_ctime = old_dir->i_mtime;
9056 old_inode->i_ctime = old_dir->i_mtime;
9058 if (old_dentry->d_parent != new_dentry->d_parent)
9059 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9060 BTRFS_I(old_inode), 1);
9062 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9063 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9065 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9066 BTRFS_I(d_inode(old_dentry)),
9067 &old_fname.disk_name, &rename_ctx);
9069 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9072 btrfs_abort_transaction(trans, ret);
9077 inode_inc_iversion(new_inode);
9078 new_inode->i_ctime = current_time(new_inode);
9079 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9080 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9081 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9082 BUG_ON(new_inode->i_nlink == 0);
9084 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9085 BTRFS_I(d_inode(new_dentry)),
9086 &new_fname.disk_name);
9088 if (!ret && new_inode->i_nlink == 0)
9089 ret = btrfs_orphan_add(trans,
9090 BTRFS_I(d_inode(new_dentry)));
9092 btrfs_abort_transaction(trans, ret);
9097 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9098 &new_fname.disk_name, 0, index);
9100 btrfs_abort_transaction(trans, ret);
9104 if (old_inode->i_nlink == 1)
9105 BTRFS_I(old_inode)->dir_index = index;
9107 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9108 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9109 rename_ctx.index, new_dentry->d_parent);
9111 if (flags & RENAME_WHITEOUT) {
9112 ret = btrfs_create_new_inode(trans, &whiteout_args);
9114 btrfs_abort_transaction(trans, ret);
9117 unlock_new_inode(whiteout_args.inode);
9118 iput(whiteout_args.inode);
9119 whiteout_args.inode = NULL;
9123 ret2 = btrfs_end_transaction(trans);
9124 ret = ret ? ret : ret2;
9126 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9127 up_read(&fs_info->subvol_sem);
9128 if (flags & RENAME_WHITEOUT)
9129 btrfs_new_inode_args_destroy(&whiteout_args);
9131 if (flags & RENAME_WHITEOUT)
9132 iput(whiteout_args.inode);
9134 fscrypt_free_filename(&old_fname);
9135 fscrypt_free_filename(&new_fname);
9139 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9140 struct dentry *old_dentry, struct inode *new_dir,
9141 struct dentry *new_dentry, unsigned int flags)
9145 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9148 if (flags & RENAME_EXCHANGE)
9149 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9152 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9155 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9160 struct btrfs_delalloc_work {
9161 struct inode *inode;
9162 struct completion completion;
9163 struct list_head list;
9164 struct btrfs_work work;
9167 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9169 struct btrfs_delalloc_work *delalloc_work;
9170 struct inode *inode;
9172 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9174 inode = delalloc_work->inode;
9175 filemap_flush(inode->i_mapping);
9176 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9177 &BTRFS_I(inode)->runtime_flags))
9178 filemap_flush(inode->i_mapping);
9181 complete(&delalloc_work->completion);
9184 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9186 struct btrfs_delalloc_work *work;
9188 work = kmalloc(sizeof(*work), GFP_NOFS);
9192 init_completion(&work->completion);
9193 INIT_LIST_HEAD(&work->list);
9194 work->inode = inode;
9195 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9201 * some fairly slow code that needs optimization. This walks the list
9202 * of all the inodes with pending delalloc and forces them to disk.
9204 static int start_delalloc_inodes(struct btrfs_root *root,
9205 struct writeback_control *wbc, bool snapshot,
9206 bool in_reclaim_context)
9208 struct btrfs_inode *binode;
9209 struct inode *inode;
9210 struct btrfs_delalloc_work *work, *next;
9211 struct list_head works;
9212 struct list_head splice;
9214 bool full_flush = wbc->nr_to_write == LONG_MAX;
9216 INIT_LIST_HEAD(&works);
9217 INIT_LIST_HEAD(&splice);
9219 mutex_lock(&root->delalloc_mutex);
9220 spin_lock(&root->delalloc_lock);
9221 list_splice_init(&root->delalloc_inodes, &splice);
9222 while (!list_empty(&splice)) {
9223 binode = list_entry(splice.next, struct btrfs_inode,
9226 list_move_tail(&binode->delalloc_inodes,
9227 &root->delalloc_inodes);
9229 if (in_reclaim_context &&
9230 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9233 inode = igrab(&binode->vfs_inode);
9235 cond_resched_lock(&root->delalloc_lock);
9238 spin_unlock(&root->delalloc_lock);
9241 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9242 &binode->runtime_flags);
9244 work = btrfs_alloc_delalloc_work(inode);
9250 list_add_tail(&work->list, &works);
9251 btrfs_queue_work(root->fs_info->flush_workers,
9254 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9255 btrfs_add_delayed_iput(BTRFS_I(inode));
9256 if (ret || wbc->nr_to_write <= 0)
9260 spin_lock(&root->delalloc_lock);
9262 spin_unlock(&root->delalloc_lock);
9265 list_for_each_entry_safe(work, next, &works, list) {
9266 list_del_init(&work->list);
9267 wait_for_completion(&work->completion);
9271 if (!list_empty(&splice)) {
9272 spin_lock(&root->delalloc_lock);
9273 list_splice_tail(&splice, &root->delalloc_inodes);
9274 spin_unlock(&root->delalloc_lock);
9276 mutex_unlock(&root->delalloc_mutex);
9280 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9282 struct writeback_control wbc = {
9283 .nr_to_write = LONG_MAX,
9284 .sync_mode = WB_SYNC_NONE,
9286 .range_end = LLONG_MAX,
9288 struct btrfs_fs_info *fs_info = root->fs_info;
9290 if (BTRFS_FS_ERROR(fs_info))
9293 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9296 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9297 bool in_reclaim_context)
9299 struct writeback_control wbc = {
9301 .sync_mode = WB_SYNC_NONE,
9303 .range_end = LLONG_MAX,
9305 struct btrfs_root *root;
9306 struct list_head splice;
9309 if (BTRFS_FS_ERROR(fs_info))
9312 INIT_LIST_HEAD(&splice);
9314 mutex_lock(&fs_info->delalloc_root_mutex);
9315 spin_lock(&fs_info->delalloc_root_lock);
9316 list_splice_init(&fs_info->delalloc_roots, &splice);
9317 while (!list_empty(&splice)) {
9319 * Reset nr_to_write here so we know that we're doing a full
9323 wbc.nr_to_write = LONG_MAX;
9325 root = list_first_entry(&splice, struct btrfs_root,
9327 root = btrfs_grab_root(root);
9329 list_move_tail(&root->delalloc_root,
9330 &fs_info->delalloc_roots);
9331 spin_unlock(&fs_info->delalloc_root_lock);
9333 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9334 btrfs_put_root(root);
9335 if (ret < 0 || wbc.nr_to_write <= 0)
9337 spin_lock(&fs_info->delalloc_root_lock);
9339 spin_unlock(&fs_info->delalloc_root_lock);
9343 if (!list_empty(&splice)) {
9344 spin_lock(&fs_info->delalloc_root_lock);
9345 list_splice_tail(&splice, &fs_info->delalloc_roots);
9346 spin_unlock(&fs_info->delalloc_root_lock);
9348 mutex_unlock(&fs_info->delalloc_root_mutex);
9352 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9353 struct dentry *dentry, const char *symname)
9355 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9356 struct btrfs_trans_handle *trans;
9357 struct btrfs_root *root = BTRFS_I(dir)->root;
9358 struct btrfs_path *path;
9359 struct btrfs_key key;
9360 struct inode *inode;
9361 struct btrfs_new_inode_args new_inode_args = {
9365 unsigned int trans_num_items;
9370 struct btrfs_file_extent_item *ei;
9371 struct extent_buffer *leaf;
9373 name_len = strlen(symname);
9374 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9375 return -ENAMETOOLONG;
9377 inode = new_inode(dir->i_sb);
9380 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9381 inode->i_op = &btrfs_symlink_inode_operations;
9382 inode_nohighmem(inode);
9383 inode->i_mapping->a_ops = &btrfs_aops;
9384 btrfs_i_size_write(BTRFS_I(inode), name_len);
9385 inode_set_bytes(inode, name_len);
9387 new_inode_args.inode = inode;
9388 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9391 /* 1 additional item for the inline extent */
9394 trans = btrfs_start_transaction(root, trans_num_items);
9395 if (IS_ERR(trans)) {
9396 err = PTR_ERR(trans);
9397 goto out_new_inode_args;
9400 err = btrfs_create_new_inode(trans, &new_inode_args);
9404 path = btrfs_alloc_path();
9407 btrfs_abort_transaction(trans, err);
9408 discard_new_inode(inode);
9412 key.objectid = btrfs_ino(BTRFS_I(inode));
9414 key.type = BTRFS_EXTENT_DATA_KEY;
9415 datasize = btrfs_file_extent_calc_inline_size(name_len);
9416 err = btrfs_insert_empty_item(trans, root, path, &key,
9419 btrfs_abort_transaction(trans, err);
9420 btrfs_free_path(path);
9421 discard_new_inode(inode);
9425 leaf = path->nodes[0];
9426 ei = btrfs_item_ptr(leaf, path->slots[0],
9427 struct btrfs_file_extent_item);
9428 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9429 btrfs_set_file_extent_type(leaf, ei,
9430 BTRFS_FILE_EXTENT_INLINE);
9431 btrfs_set_file_extent_encryption(leaf, ei, 0);
9432 btrfs_set_file_extent_compression(leaf, ei, 0);
9433 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9434 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9436 ptr = btrfs_file_extent_inline_start(ei);
9437 write_extent_buffer(leaf, symname, ptr, name_len);
9438 btrfs_mark_buffer_dirty(leaf);
9439 btrfs_free_path(path);
9441 d_instantiate_new(dentry, inode);
9444 btrfs_end_transaction(trans);
9445 btrfs_btree_balance_dirty(fs_info);
9447 btrfs_new_inode_args_destroy(&new_inode_args);
9454 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9455 struct btrfs_trans_handle *trans_in,
9456 struct btrfs_inode *inode,
9457 struct btrfs_key *ins,
9460 struct btrfs_file_extent_item stack_fi;
9461 struct btrfs_replace_extent_info extent_info;
9462 struct btrfs_trans_handle *trans = trans_in;
9463 struct btrfs_path *path;
9464 u64 start = ins->objectid;
9465 u64 len = ins->offset;
9466 int qgroup_released;
9469 memset(&stack_fi, 0, sizeof(stack_fi));
9471 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9472 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9473 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9474 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9475 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9476 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9477 /* Encryption and other encoding is reserved and all 0 */
9479 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9480 if (qgroup_released < 0)
9481 return ERR_PTR(qgroup_released);
9484 ret = insert_reserved_file_extent(trans, inode,
9485 file_offset, &stack_fi,
9486 true, qgroup_released);
9492 extent_info.disk_offset = start;
9493 extent_info.disk_len = len;
9494 extent_info.data_offset = 0;
9495 extent_info.data_len = len;
9496 extent_info.file_offset = file_offset;
9497 extent_info.extent_buf = (char *)&stack_fi;
9498 extent_info.is_new_extent = true;
9499 extent_info.update_times = true;
9500 extent_info.qgroup_reserved = qgroup_released;
9501 extent_info.insertions = 0;
9503 path = btrfs_alloc_path();
9509 ret = btrfs_replace_file_extents(inode, path, file_offset,
9510 file_offset + len - 1, &extent_info,
9512 btrfs_free_path(path);
9519 * We have released qgroup data range at the beginning of the function,
9520 * and normally qgroup_released bytes will be freed when committing
9522 * But if we error out early, we have to free what we have released
9523 * or we leak qgroup data reservation.
9525 btrfs_qgroup_free_refroot(inode->root->fs_info,
9526 inode->root->root_key.objectid, qgroup_released,
9527 BTRFS_QGROUP_RSV_DATA);
9528 return ERR_PTR(ret);
9531 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9532 u64 start, u64 num_bytes, u64 min_size,
9533 loff_t actual_len, u64 *alloc_hint,
9534 struct btrfs_trans_handle *trans)
9536 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9537 struct extent_map *em;
9538 struct btrfs_root *root = BTRFS_I(inode)->root;
9539 struct btrfs_key ins;
9540 u64 cur_offset = start;
9541 u64 clear_offset = start;
9544 u64 last_alloc = (u64)-1;
9546 bool own_trans = true;
9547 u64 end = start + num_bytes - 1;
9551 while (num_bytes > 0) {
9552 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9553 cur_bytes = max(cur_bytes, min_size);
9555 * If we are severely fragmented we could end up with really
9556 * small allocations, so if the allocator is returning small
9557 * chunks lets make its job easier by only searching for those
9560 cur_bytes = min(cur_bytes, last_alloc);
9561 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9562 min_size, 0, *alloc_hint, &ins, 1, 0);
9567 * We've reserved this space, and thus converted it from
9568 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9569 * from here on out we will only need to clear our reservation
9570 * for the remaining unreserved area, so advance our
9571 * clear_offset by our extent size.
9573 clear_offset += ins.offset;
9575 last_alloc = ins.offset;
9576 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9579 * Now that we inserted the prealloc extent we can finally
9580 * decrement the number of reservations in the block group.
9581 * If we did it before, we could race with relocation and have
9582 * relocation miss the reserved extent, making it fail later.
9584 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9585 if (IS_ERR(trans)) {
9586 ret = PTR_ERR(trans);
9587 btrfs_free_reserved_extent(fs_info, ins.objectid,
9592 em = alloc_extent_map();
9594 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9595 cur_offset + ins.offset - 1, false);
9596 btrfs_set_inode_full_sync(BTRFS_I(inode));
9600 em->start = cur_offset;
9601 em->orig_start = cur_offset;
9602 em->len = ins.offset;
9603 em->block_start = ins.objectid;
9604 em->block_len = ins.offset;
9605 em->orig_block_len = ins.offset;
9606 em->ram_bytes = ins.offset;
9607 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9608 em->generation = trans->transid;
9610 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9611 free_extent_map(em);
9613 num_bytes -= ins.offset;
9614 cur_offset += ins.offset;
9615 *alloc_hint = ins.objectid + ins.offset;
9617 inode_inc_iversion(inode);
9618 inode->i_ctime = current_time(inode);
9619 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9620 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9621 (actual_len > inode->i_size) &&
9622 (cur_offset > inode->i_size)) {
9623 if (cur_offset > actual_len)
9624 i_size = actual_len;
9626 i_size = cur_offset;
9627 i_size_write(inode, i_size);
9628 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9631 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9634 btrfs_abort_transaction(trans, ret);
9636 btrfs_end_transaction(trans);
9641 btrfs_end_transaction(trans);
9645 if (clear_offset < end)
9646 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9647 end - clear_offset + 1);
9651 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9652 u64 start, u64 num_bytes, u64 min_size,
9653 loff_t actual_len, u64 *alloc_hint)
9655 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9656 min_size, actual_len, alloc_hint,
9660 int btrfs_prealloc_file_range_trans(struct inode *inode,
9661 struct btrfs_trans_handle *trans, int mode,
9662 u64 start, u64 num_bytes, u64 min_size,
9663 loff_t actual_len, u64 *alloc_hint)
9665 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9666 min_size, actual_len, alloc_hint, trans);
9669 static int btrfs_permission(struct mnt_idmap *idmap,
9670 struct inode *inode, int mask)
9672 struct btrfs_root *root = BTRFS_I(inode)->root;
9673 umode_t mode = inode->i_mode;
9675 if (mask & MAY_WRITE &&
9676 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9677 if (btrfs_root_readonly(root))
9679 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9682 return generic_permission(idmap, inode, mask);
9685 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9686 struct file *file, umode_t mode)
9688 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9689 struct btrfs_trans_handle *trans;
9690 struct btrfs_root *root = BTRFS_I(dir)->root;
9691 struct inode *inode;
9692 struct btrfs_new_inode_args new_inode_args = {
9694 .dentry = file->f_path.dentry,
9697 unsigned int trans_num_items;
9700 inode = new_inode(dir->i_sb);
9703 inode_init_owner(idmap, inode, dir, mode);
9704 inode->i_fop = &btrfs_file_operations;
9705 inode->i_op = &btrfs_file_inode_operations;
9706 inode->i_mapping->a_ops = &btrfs_aops;
9708 new_inode_args.inode = inode;
9709 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9713 trans = btrfs_start_transaction(root, trans_num_items);
9714 if (IS_ERR(trans)) {
9715 ret = PTR_ERR(trans);
9716 goto out_new_inode_args;
9719 ret = btrfs_create_new_inode(trans, &new_inode_args);
9722 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9723 * set it to 1 because d_tmpfile() will issue a warning if the count is
9726 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9728 set_nlink(inode, 1);
9731 d_tmpfile(file, inode);
9732 unlock_new_inode(inode);
9733 mark_inode_dirty(inode);
9736 btrfs_end_transaction(trans);
9737 btrfs_btree_balance_dirty(fs_info);
9739 btrfs_new_inode_args_destroy(&new_inode_args);
9743 return finish_open_simple(file, ret);
9746 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9748 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9749 unsigned long index = start >> PAGE_SHIFT;
9750 unsigned long end_index = end >> PAGE_SHIFT;
9754 ASSERT(end + 1 - start <= U32_MAX);
9755 len = end + 1 - start;
9756 while (index <= end_index) {
9757 page = find_get_page(inode->vfs_inode.i_mapping, index);
9758 ASSERT(page); /* Pages should be in the extent_io_tree */
9760 btrfs_page_set_writeback(fs_info, page, start, len);
9766 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9769 switch (compress_type) {
9770 case BTRFS_COMPRESS_NONE:
9771 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9772 case BTRFS_COMPRESS_ZLIB:
9773 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9774 case BTRFS_COMPRESS_LZO:
9776 * The LZO format depends on the sector size. 64K is the maximum
9777 * sector size that we support.
9779 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9781 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9782 (fs_info->sectorsize_bits - 12);
9783 case BTRFS_COMPRESS_ZSTD:
9784 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9790 static ssize_t btrfs_encoded_read_inline(
9792 struct iov_iter *iter, u64 start,
9794 struct extent_state **cached_state,
9795 u64 extent_start, size_t count,
9796 struct btrfs_ioctl_encoded_io_args *encoded,
9799 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9800 struct btrfs_root *root = inode->root;
9801 struct btrfs_fs_info *fs_info = root->fs_info;
9802 struct extent_io_tree *io_tree = &inode->io_tree;
9803 struct btrfs_path *path;
9804 struct extent_buffer *leaf;
9805 struct btrfs_file_extent_item *item;
9811 path = btrfs_alloc_path();
9816 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9820 /* The extent item disappeared? */
9825 leaf = path->nodes[0];
9826 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9828 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9829 ptr = btrfs_file_extent_inline_start(item);
9831 encoded->len = min_t(u64, extent_start + ram_bytes,
9832 inode->vfs_inode.i_size) - iocb->ki_pos;
9833 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9834 btrfs_file_extent_compression(leaf, item));
9837 encoded->compression = ret;
9838 if (encoded->compression) {
9841 inline_size = btrfs_file_extent_inline_item_len(leaf,
9843 if (inline_size > count) {
9847 count = inline_size;
9848 encoded->unencoded_len = ram_bytes;
9849 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9851 count = min_t(u64, count, encoded->len);
9852 encoded->len = count;
9853 encoded->unencoded_len = count;
9854 ptr += iocb->ki_pos - extent_start;
9857 tmp = kmalloc(count, GFP_NOFS);
9862 read_extent_buffer(leaf, tmp, ptr, count);
9863 btrfs_release_path(path);
9864 unlock_extent(io_tree, start, lockend, cached_state);
9865 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9868 ret = copy_to_iter(tmp, count, iter);
9873 btrfs_free_path(path);
9877 struct btrfs_encoded_read_private {
9878 wait_queue_head_t wait;
9880 blk_status_t status;
9883 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9885 struct btrfs_encoded_read_private *priv = bbio->private;
9887 if (bbio->bio.bi_status) {
9889 * The memory barrier implied by the atomic_dec_return() here
9890 * pairs with the memory barrier implied by the
9891 * atomic_dec_return() or io_wait_event() in
9892 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9893 * write is observed before the load of status in
9894 * btrfs_encoded_read_regular_fill_pages().
9896 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9898 if (!atomic_dec_return(&priv->pending))
9899 wake_up(&priv->wait);
9900 bio_put(&bbio->bio);
9903 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9904 u64 file_offset, u64 disk_bytenr,
9905 u64 disk_io_size, struct page **pages)
9907 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9908 struct btrfs_encoded_read_private priv = {
9909 .pending = ATOMIC_INIT(1),
9911 unsigned long i = 0;
9912 struct btrfs_bio *bbio;
9914 init_waitqueue_head(&priv.wait);
9916 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9917 btrfs_encoded_read_endio, &priv);
9918 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9919 bbio->inode = inode;
9922 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
9924 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
9925 atomic_inc(&priv.pending);
9926 btrfs_submit_bio(bbio, 0);
9928 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9929 btrfs_encoded_read_endio, &priv);
9930 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9931 bbio->inode = inode;
9936 disk_bytenr += bytes;
9937 disk_io_size -= bytes;
9938 } while (disk_io_size);
9940 atomic_inc(&priv.pending);
9941 btrfs_submit_bio(bbio, 0);
9943 if (atomic_dec_return(&priv.pending))
9944 io_wait_event(priv.wait, !atomic_read(&priv.pending));
9945 /* See btrfs_encoded_read_endio() for ordering. */
9946 return blk_status_to_errno(READ_ONCE(priv.status));
9949 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
9950 struct iov_iter *iter,
9951 u64 start, u64 lockend,
9952 struct extent_state **cached_state,
9953 u64 disk_bytenr, u64 disk_io_size,
9954 size_t count, bool compressed,
9957 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9958 struct extent_io_tree *io_tree = &inode->io_tree;
9959 struct page **pages;
9960 unsigned long nr_pages, i;
9965 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
9966 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
9969 ret = btrfs_alloc_page_array(nr_pages, pages);
9975 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
9976 disk_io_size, pages);
9980 unlock_extent(io_tree, start, lockend, cached_state);
9981 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9988 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
9989 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
9992 while (cur < count) {
9993 size_t bytes = min_t(size_t, count - cur,
9994 PAGE_SIZE - page_offset);
9996 if (copy_page_to_iter(pages[i], page_offset, bytes,
10007 for (i = 0; i < nr_pages; i++) {
10009 __free_page(pages[i]);
10015 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10016 struct btrfs_ioctl_encoded_io_args *encoded)
10018 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10019 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10020 struct extent_io_tree *io_tree = &inode->io_tree;
10022 size_t count = iov_iter_count(iter);
10023 u64 start, lockend, disk_bytenr, disk_io_size;
10024 struct extent_state *cached_state = NULL;
10025 struct extent_map *em;
10026 bool unlocked = false;
10028 file_accessed(iocb->ki_filp);
10030 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10032 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10033 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10036 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10038 * We don't know how long the extent containing iocb->ki_pos is, but if
10039 * it's compressed we know that it won't be longer than this.
10041 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10044 struct btrfs_ordered_extent *ordered;
10046 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10047 lockend - start + 1);
10049 goto out_unlock_inode;
10050 lock_extent(io_tree, start, lockend, &cached_state);
10051 ordered = btrfs_lookup_ordered_range(inode, start,
10052 lockend - start + 1);
10055 btrfs_put_ordered_extent(ordered);
10056 unlock_extent(io_tree, start, lockend, &cached_state);
10060 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10063 goto out_unlock_extent;
10066 if (em->block_start == EXTENT_MAP_INLINE) {
10067 u64 extent_start = em->start;
10070 * For inline extents we get everything we need out of the
10073 free_extent_map(em);
10075 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10076 &cached_state, extent_start,
10077 count, encoded, &unlocked);
10082 * We only want to return up to EOF even if the extent extends beyond
10085 encoded->len = min_t(u64, extent_map_end(em),
10086 inode->vfs_inode.i_size) - iocb->ki_pos;
10087 if (em->block_start == EXTENT_MAP_HOLE ||
10088 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10089 disk_bytenr = EXTENT_MAP_HOLE;
10090 count = min_t(u64, count, encoded->len);
10091 encoded->len = count;
10092 encoded->unencoded_len = count;
10093 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10094 disk_bytenr = em->block_start;
10096 * Bail if the buffer isn't large enough to return the whole
10097 * compressed extent.
10099 if (em->block_len > count) {
10103 disk_io_size = em->block_len;
10104 count = em->block_len;
10105 encoded->unencoded_len = em->ram_bytes;
10106 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10107 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10108 em->compress_type);
10111 encoded->compression = ret;
10113 disk_bytenr = em->block_start + (start - em->start);
10114 if (encoded->len > count)
10115 encoded->len = count;
10117 * Don't read beyond what we locked. This also limits the page
10118 * allocations that we'll do.
10120 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10121 count = start + disk_io_size - iocb->ki_pos;
10122 encoded->len = count;
10123 encoded->unencoded_len = count;
10124 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10126 free_extent_map(em);
10129 if (disk_bytenr == EXTENT_MAP_HOLE) {
10130 unlock_extent(io_tree, start, lockend, &cached_state);
10131 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10133 ret = iov_iter_zero(count, iter);
10137 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10138 &cached_state, disk_bytenr,
10139 disk_io_size, count,
10140 encoded->compression,
10146 iocb->ki_pos += encoded->len;
10148 free_extent_map(em);
10151 unlock_extent(io_tree, start, lockend, &cached_state);
10154 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10158 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10159 const struct btrfs_ioctl_encoded_io_args *encoded)
10161 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10162 struct btrfs_root *root = inode->root;
10163 struct btrfs_fs_info *fs_info = root->fs_info;
10164 struct extent_io_tree *io_tree = &inode->io_tree;
10165 struct extent_changeset *data_reserved = NULL;
10166 struct extent_state *cached_state = NULL;
10170 u64 num_bytes, ram_bytes, disk_num_bytes;
10171 unsigned long nr_pages, i;
10172 struct page **pages;
10173 struct btrfs_key ins;
10174 bool extent_reserved = false;
10175 struct extent_map *em;
10178 switch (encoded->compression) {
10179 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10180 compression = BTRFS_COMPRESS_ZLIB;
10182 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10183 compression = BTRFS_COMPRESS_ZSTD;
10185 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10186 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10187 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10188 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10189 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10190 /* The sector size must match for LZO. */
10191 if (encoded->compression -
10192 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10193 fs_info->sectorsize_bits)
10195 compression = BTRFS_COMPRESS_LZO;
10200 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10203 orig_count = iov_iter_count(from);
10205 /* The extent size must be sane. */
10206 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10207 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10211 * The compressed data must be smaller than the decompressed data.
10213 * It's of course possible for data to compress to larger or the same
10214 * size, but the buffered I/O path falls back to no compression for such
10215 * data, and we don't want to break any assumptions by creating these
10218 * Note that this is less strict than the current check we have that the
10219 * compressed data must be at least one sector smaller than the
10220 * decompressed data. We only want to enforce the weaker requirement
10221 * from old kernels that it is at least one byte smaller.
10223 if (orig_count >= encoded->unencoded_len)
10226 /* The extent must start on a sector boundary. */
10227 start = iocb->ki_pos;
10228 if (!IS_ALIGNED(start, fs_info->sectorsize))
10232 * The extent must end on a sector boundary. However, we allow a write
10233 * which ends at or extends i_size to have an unaligned length; we round
10234 * up the extent size and set i_size to the unaligned end.
10236 if (start + encoded->len < inode->vfs_inode.i_size &&
10237 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10240 /* Finally, the offset in the unencoded data must be sector-aligned. */
10241 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10244 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10245 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10246 end = start + num_bytes - 1;
10249 * If the extent cannot be inline, the compressed data on disk must be
10250 * sector-aligned. For convenience, we extend it with zeroes if it
10253 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10254 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10255 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10258 for (i = 0; i < nr_pages; i++) {
10259 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10262 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10267 kaddr = kmap_local_page(pages[i]);
10268 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10269 kunmap_local(kaddr);
10273 if (bytes < PAGE_SIZE)
10274 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10275 kunmap_local(kaddr);
10279 struct btrfs_ordered_extent *ordered;
10281 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10284 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10285 start >> PAGE_SHIFT,
10286 end >> PAGE_SHIFT);
10289 lock_extent(io_tree, start, end, &cached_state);
10290 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10292 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10295 btrfs_put_ordered_extent(ordered);
10296 unlock_extent(io_tree, start, end, &cached_state);
10301 * We don't use the higher-level delalloc space functions because our
10302 * num_bytes and disk_num_bytes are different.
10304 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10307 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10309 goto out_free_data_space;
10310 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10313 goto out_qgroup_free_data;
10315 /* Try an inline extent first. */
10316 if (start == 0 && encoded->unencoded_len == encoded->len &&
10317 encoded->unencoded_offset == 0) {
10318 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10319 compression, pages, true);
10323 goto out_delalloc_release;
10327 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10328 disk_num_bytes, 0, 0, &ins, 1, 1);
10330 goto out_delalloc_release;
10331 extent_reserved = true;
10333 em = create_io_em(inode, start, num_bytes,
10334 start - encoded->unencoded_offset, ins.objectid,
10335 ins.offset, ins.offset, ram_bytes, compression,
10336 BTRFS_ORDERED_COMPRESSED);
10339 goto out_free_reserved;
10341 free_extent_map(em);
10343 ret = btrfs_add_ordered_extent(inode, start, num_bytes, ram_bytes,
10344 ins.objectid, ins.offset,
10345 encoded->unencoded_offset,
10346 (1 << BTRFS_ORDERED_ENCODED) |
10347 (1 << BTRFS_ORDERED_COMPRESSED),
10350 btrfs_drop_extent_map_range(inode, start, end, false);
10351 goto out_free_reserved;
10353 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10355 if (start + encoded->len > inode->vfs_inode.i_size)
10356 i_size_write(&inode->vfs_inode, start + encoded->len);
10358 unlock_extent(io_tree, start, end, &cached_state);
10360 btrfs_delalloc_release_extents(inode, num_bytes);
10362 btrfs_submit_compressed_write(inode, start, num_bytes, ins.objectid,
10363 ins.offset, pages, nr_pages, 0, false);
10368 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10369 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10370 out_delalloc_release:
10371 btrfs_delalloc_release_extents(inode, num_bytes);
10372 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10373 out_qgroup_free_data:
10375 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10376 out_free_data_space:
10378 * If btrfs_reserve_extent() succeeded, then we already decremented
10381 if (!extent_reserved)
10382 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10384 unlock_extent(io_tree, start, end, &cached_state);
10386 for (i = 0; i < nr_pages; i++) {
10388 __free_page(pages[i]);
10393 iocb->ki_pos += encoded->len;
10399 * Add an entry indicating a block group or device which is pinned by a
10400 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10401 * negative errno on failure.
10403 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10404 bool is_block_group)
10406 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10407 struct btrfs_swapfile_pin *sp, *entry;
10408 struct rb_node **p;
10409 struct rb_node *parent = NULL;
10411 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10416 sp->is_block_group = is_block_group;
10417 sp->bg_extent_count = 1;
10419 spin_lock(&fs_info->swapfile_pins_lock);
10420 p = &fs_info->swapfile_pins.rb_node;
10423 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10424 if (sp->ptr < entry->ptr ||
10425 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10426 p = &(*p)->rb_left;
10427 } else if (sp->ptr > entry->ptr ||
10428 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10429 p = &(*p)->rb_right;
10431 if (is_block_group)
10432 entry->bg_extent_count++;
10433 spin_unlock(&fs_info->swapfile_pins_lock);
10438 rb_link_node(&sp->node, parent, p);
10439 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10440 spin_unlock(&fs_info->swapfile_pins_lock);
10444 /* Free all of the entries pinned by this swapfile. */
10445 static void btrfs_free_swapfile_pins(struct inode *inode)
10447 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10448 struct btrfs_swapfile_pin *sp;
10449 struct rb_node *node, *next;
10451 spin_lock(&fs_info->swapfile_pins_lock);
10452 node = rb_first(&fs_info->swapfile_pins);
10454 next = rb_next(node);
10455 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10456 if (sp->inode == inode) {
10457 rb_erase(&sp->node, &fs_info->swapfile_pins);
10458 if (sp->is_block_group) {
10459 btrfs_dec_block_group_swap_extents(sp->ptr,
10460 sp->bg_extent_count);
10461 btrfs_put_block_group(sp->ptr);
10467 spin_unlock(&fs_info->swapfile_pins_lock);
10470 struct btrfs_swap_info {
10476 unsigned long nr_pages;
10480 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10481 struct btrfs_swap_info *bsi)
10483 unsigned long nr_pages;
10484 unsigned long max_pages;
10485 u64 first_ppage, first_ppage_reported, next_ppage;
10489 * Our swapfile may have had its size extended after the swap header was
10490 * written. In that case activating the swapfile should not go beyond
10491 * the max size set in the swap header.
10493 if (bsi->nr_pages >= sis->max)
10496 max_pages = sis->max - bsi->nr_pages;
10497 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10498 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10500 if (first_ppage >= next_ppage)
10502 nr_pages = next_ppage - first_ppage;
10503 nr_pages = min(nr_pages, max_pages);
10505 first_ppage_reported = first_ppage;
10506 if (bsi->start == 0)
10507 first_ppage_reported++;
10508 if (bsi->lowest_ppage > first_ppage_reported)
10509 bsi->lowest_ppage = first_ppage_reported;
10510 if (bsi->highest_ppage < (next_ppage - 1))
10511 bsi->highest_ppage = next_ppage - 1;
10513 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10516 bsi->nr_extents += ret;
10517 bsi->nr_pages += nr_pages;
10521 static void btrfs_swap_deactivate(struct file *file)
10523 struct inode *inode = file_inode(file);
10525 btrfs_free_swapfile_pins(inode);
10526 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10529 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10532 struct inode *inode = file_inode(file);
10533 struct btrfs_root *root = BTRFS_I(inode)->root;
10534 struct btrfs_fs_info *fs_info = root->fs_info;
10535 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10536 struct extent_state *cached_state = NULL;
10537 struct extent_map *em = NULL;
10538 struct btrfs_device *device = NULL;
10539 struct btrfs_swap_info bsi = {
10540 .lowest_ppage = (sector_t)-1ULL,
10547 * If the swap file was just created, make sure delalloc is done. If the
10548 * file changes again after this, the user is doing something stupid and
10549 * we don't really care.
10551 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10556 * The inode is locked, so these flags won't change after we check them.
10558 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10559 btrfs_warn(fs_info, "swapfile must not be compressed");
10562 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10563 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10566 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10567 btrfs_warn(fs_info, "swapfile must not be checksummed");
10572 * Balance or device remove/replace/resize can move stuff around from
10573 * under us. The exclop protection makes sure they aren't running/won't
10574 * run concurrently while we are mapping the swap extents, and
10575 * fs_info->swapfile_pins prevents them from running while the swap
10576 * file is active and moving the extents. Note that this also prevents
10577 * a concurrent device add which isn't actually necessary, but it's not
10578 * really worth the trouble to allow it.
10580 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10581 btrfs_warn(fs_info,
10582 "cannot activate swapfile while exclusive operation is running");
10587 * Prevent snapshot creation while we are activating the swap file.
10588 * We do not want to race with snapshot creation. If snapshot creation
10589 * already started before we bumped nr_swapfiles from 0 to 1 and
10590 * completes before the first write into the swap file after it is
10591 * activated, than that write would fallback to COW.
10593 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10594 btrfs_exclop_finish(fs_info);
10595 btrfs_warn(fs_info,
10596 "cannot activate swapfile because snapshot creation is in progress");
10600 * Snapshots can create extents which require COW even if NODATACOW is
10601 * set. We use this counter to prevent snapshots. We must increment it
10602 * before walking the extents because we don't want a concurrent
10603 * snapshot to run after we've already checked the extents.
10605 * It is possible that subvolume is marked for deletion but still not
10606 * removed yet. To prevent this race, we check the root status before
10607 * activating the swapfile.
10609 spin_lock(&root->root_item_lock);
10610 if (btrfs_root_dead(root)) {
10611 spin_unlock(&root->root_item_lock);
10613 btrfs_exclop_finish(fs_info);
10614 btrfs_warn(fs_info,
10615 "cannot activate swapfile because subvolume %llu is being deleted",
10616 root->root_key.objectid);
10619 atomic_inc(&root->nr_swapfiles);
10620 spin_unlock(&root->root_item_lock);
10622 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10624 lock_extent(io_tree, 0, isize - 1, &cached_state);
10626 while (start < isize) {
10627 u64 logical_block_start, physical_block_start;
10628 struct btrfs_block_group *bg;
10629 u64 len = isize - start;
10631 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10637 if (em->block_start == EXTENT_MAP_HOLE) {
10638 btrfs_warn(fs_info, "swapfile must not have holes");
10642 if (em->block_start == EXTENT_MAP_INLINE) {
10644 * It's unlikely we'll ever actually find ourselves
10645 * here, as a file small enough to fit inline won't be
10646 * big enough to store more than the swap header, but in
10647 * case something changes in the future, let's catch it
10648 * here rather than later.
10650 btrfs_warn(fs_info, "swapfile must not be inline");
10654 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10655 btrfs_warn(fs_info, "swapfile must not be compressed");
10660 logical_block_start = em->block_start + (start - em->start);
10661 len = min(len, em->len - (start - em->start));
10662 free_extent_map(em);
10665 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10671 btrfs_warn(fs_info,
10672 "swapfile must not be copy-on-write");
10677 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10683 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10684 btrfs_warn(fs_info,
10685 "swapfile must have single data profile");
10690 if (device == NULL) {
10691 device = em->map_lookup->stripes[0].dev;
10692 ret = btrfs_add_swapfile_pin(inode, device, false);
10697 } else if (device != em->map_lookup->stripes[0].dev) {
10698 btrfs_warn(fs_info, "swapfile must be on one device");
10703 physical_block_start = (em->map_lookup->stripes[0].physical +
10704 (logical_block_start - em->start));
10705 len = min(len, em->len - (logical_block_start - em->start));
10706 free_extent_map(em);
10709 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10711 btrfs_warn(fs_info,
10712 "could not find block group containing swapfile");
10717 if (!btrfs_inc_block_group_swap_extents(bg)) {
10718 btrfs_warn(fs_info,
10719 "block group for swapfile at %llu is read-only%s",
10721 atomic_read(&fs_info->scrubs_running) ?
10722 " (scrub running)" : "");
10723 btrfs_put_block_group(bg);
10728 ret = btrfs_add_swapfile_pin(inode, bg, true);
10730 btrfs_put_block_group(bg);
10737 if (bsi.block_len &&
10738 bsi.block_start + bsi.block_len == physical_block_start) {
10739 bsi.block_len += len;
10741 if (bsi.block_len) {
10742 ret = btrfs_add_swap_extent(sis, &bsi);
10747 bsi.block_start = physical_block_start;
10748 bsi.block_len = len;
10755 ret = btrfs_add_swap_extent(sis, &bsi);
10758 if (!IS_ERR_OR_NULL(em))
10759 free_extent_map(em);
10761 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10764 btrfs_swap_deactivate(file);
10766 btrfs_drew_write_unlock(&root->snapshot_lock);
10768 btrfs_exclop_finish(fs_info);
10774 sis->bdev = device->bdev;
10775 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10776 sis->max = bsi.nr_pages;
10777 sis->pages = bsi.nr_pages - 1;
10778 sis->highest_bit = bsi.nr_pages - 1;
10779 return bsi.nr_extents;
10782 static void btrfs_swap_deactivate(struct file *file)
10786 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10789 return -EOPNOTSUPP;
10794 * Update the number of bytes used in the VFS' inode. When we replace extents in
10795 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10796 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10797 * always get a correct value.
10799 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10800 const u64 add_bytes,
10801 const u64 del_bytes)
10803 if (add_bytes == del_bytes)
10806 spin_lock(&inode->lock);
10808 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10810 inode_add_bytes(&inode->vfs_inode, add_bytes);
10811 spin_unlock(&inode->lock);
10815 * Verify that there are no ordered extents for a given file range.
10817 * @inode: The target inode.
10818 * @start: Start offset of the file range, should be sector size aligned.
10819 * @end: End offset (inclusive) of the file range, its value +1 should be
10820 * sector size aligned.
10822 * This should typically be used for cases where we locked an inode's VFS lock in
10823 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10824 * we have flushed all delalloc in the range, we have waited for all ordered
10825 * extents in the range to complete and finally we have locked the file range in
10826 * the inode's io_tree.
10828 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10830 struct btrfs_root *root = inode->root;
10831 struct btrfs_ordered_extent *ordered;
10833 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10836 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10838 btrfs_err(root->fs_info,
10839 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10840 start, end, btrfs_ino(inode), root->root_key.objectid,
10841 ordered->file_offset,
10842 ordered->file_offset + ordered->num_bytes - 1);
10843 btrfs_put_ordered_extent(ordered);
10846 ASSERT(ordered == NULL);
10849 static const struct inode_operations btrfs_dir_inode_operations = {
10850 .getattr = btrfs_getattr,
10851 .lookup = btrfs_lookup,
10852 .create = btrfs_create,
10853 .unlink = btrfs_unlink,
10854 .link = btrfs_link,
10855 .mkdir = btrfs_mkdir,
10856 .rmdir = btrfs_rmdir,
10857 .rename = btrfs_rename2,
10858 .symlink = btrfs_symlink,
10859 .setattr = btrfs_setattr,
10860 .mknod = btrfs_mknod,
10861 .listxattr = btrfs_listxattr,
10862 .permission = btrfs_permission,
10863 .get_inode_acl = btrfs_get_acl,
10864 .set_acl = btrfs_set_acl,
10865 .update_time = btrfs_update_time,
10866 .tmpfile = btrfs_tmpfile,
10867 .fileattr_get = btrfs_fileattr_get,
10868 .fileattr_set = btrfs_fileattr_set,
10871 static const struct file_operations btrfs_dir_file_operations = {
10872 .llseek = generic_file_llseek,
10873 .read = generic_read_dir,
10874 .iterate_shared = btrfs_real_readdir,
10875 .open = btrfs_opendir,
10876 .unlocked_ioctl = btrfs_ioctl,
10877 #ifdef CONFIG_COMPAT
10878 .compat_ioctl = btrfs_compat_ioctl,
10880 .release = btrfs_release_file,
10881 .fsync = btrfs_sync_file,
10885 * btrfs doesn't support the bmap operation because swapfiles
10886 * use bmap to make a mapping of extents in the file. They assume
10887 * these extents won't change over the life of the file and they
10888 * use the bmap result to do IO directly to the drive.
10890 * the btrfs bmap call would return logical addresses that aren't
10891 * suitable for IO and they also will change frequently as COW
10892 * operations happen. So, swapfile + btrfs == corruption.
10894 * For now we're avoiding this by dropping bmap.
10896 static const struct address_space_operations btrfs_aops = {
10897 .read_folio = btrfs_read_folio,
10898 .writepages = btrfs_writepages,
10899 .readahead = btrfs_readahead,
10900 .direct_IO = noop_direct_IO,
10901 .invalidate_folio = btrfs_invalidate_folio,
10902 .release_folio = btrfs_release_folio,
10903 .migrate_folio = btrfs_migrate_folio,
10904 .dirty_folio = filemap_dirty_folio,
10905 .error_remove_page = generic_error_remove_page,
10906 .swap_activate = btrfs_swap_activate,
10907 .swap_deactivate = btrfs_swap_deactivate,
10910 static const struct inode_operations btrfs_file_inode_operations = {
10911 .getattr = btrfs_getattr,
10912 .setattr = btrfs_setattr,
10913 .listxattr = btrfs_listxattr,
10914 .permission = btrfs_permission,
10915 .fiemap = btrfs_fiemap,
10916 .get_inode_acl = btrfs_get_acl,
10917 .set_acl = btrfs_set_acl,
10918 .update_time = btrfs_update_time,
10919 .fileattr_get = btrfs_fileattr_get,
10920 .fileattr_set = btrfs_fileattr_set,
10922 static const struct inode_operations btrfs_special_inode_operations = {
10923 .getattr = btrfs_getattr,
10924 .setattr = btrfs_setattr,
10925 .permission = btrfs_permission,
10926 .listxattr = btrfs_listxattr,
10927 .get_inode_acl = btrfs_get_acl,
10928 .set_acl = btrfs_set_acl,
10929 .update_time = btrfs_update_time,
10931 static const struct inode_operations btrfs_symlink_inode_operations = {
10932 .get_link = page_get_link,
10933 .getattr = btrfs_getattr,
10934 .setattr = btrfs_setattr,
10935 .permission = btrfs_permission,
10936 .listxattr = btrfs_listxattr,
10937 .update_time = btrfs_update_time,
10940 const struct dentry_operations btrfs_dentry_operations = {
10941 .d_delete = btrfs_dentry_delete,