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 #include "raid-stripe-tree.h"
76 struct btrfs_iget_args {
78 struct btrfs_root *root;
81 struct btrfs_dio_data {
83 struct extent_changeset *data_reserved;
84 struct btrfs_ordered_extent *ordered;
85 bool data_space_reserved;
89 struct btrfs_dio_private {
94 /* This must be last */
95 struct btrfs_bio bbio;
98 static struct bio_set btrfs_dio_bioset;
100 struct btrfs_rename_ctx {
101 /* Output field. Stores the index number of the old directory entry. */
106 * Used by data_reloc_print_warning_inode() to pass needed info for filename
107 * resolution and output of error message.
109 struct data_reloc_warn {
110 struct btrfs_path path;
111 struct btrfs_fs_info *fs_info;
112 u64 extent_item_size;
118 * For the file_extent_tree, we want to hold the inode lock when we lookup and
119 * update the disk_i_size, but lockdep will complain because our io_tree we hold
120 * the tree lock and get the inode lock when setting delalloc. These two things
121 * are unrelated, so make a class for the file_extent_tree so we don't get the
122 * two locking patterns mixed up.
124 static struct lock_class_key file_extent_tree_class;
126 static const struct inode_operations btrfs_dir_inode_operations;
127 static const struct inode_operations btrfs_symlink_inode_operations;
128 static const struct inode_operations btrfs_special_inode_operations;
129 static const struct inode_operations btrfs_file_inode_operations;
130 static const struct address_space_operations btrfs_aops;
131 static const struct file_operations btrfs_dir_file_operations;
133 static struct kmem_cache *btrfs_inode_cachep;
135 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
136 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
138 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
139 struct page *locked_page, u64 start,
140 u64 end, struct writeback_control *wbc,
142 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
143 u64 len, u64 orig_start, u64 block_start,
144 u64 block_len, u64 orig_block_len,
145 u64 ram_bytes, int compress_type,
148 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
149 u64 root, void *warn_ctx)
151 struct data_reloc_warn *warn = warn_ctx;
152 struct btrfs_fs_info *fs_info = warn->fs_info;
153 struct extent_buffer *eb;
154 struct btrfs_inode_item *inode_item;
155 struct inode_fs_paths *ipath = NULL;
156 struct btrfs_root *local_root;
157 struct btrfs_key key;
158 unsigned int nofs_flag;
162 local_root = btrfs_get_fs_root(fs_info, root, true);
163 if (IS_ERR(local_root)) {
164 ret = PTR_ERR(local_root);
168 /* This makes the path point to (inum INODE_ITEM ioff). */
170 key.type = BTRFS_INODE_ITEM_KEY;
173 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
175 btrfs_put_root(local_root);
176 btrfs_release_path(&warn->path);
180 eb = warn->path.nodes[0];
181 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
182 nlink = btrfs_inode_nlink(eb, inode_item);
183 btrfs_release_path(&warn->path);
185 nofs_flag = memalloc_nofs_save();
186 ipath = init_ipath(4096, local_root, &warn->path);
187 memalloc_nofs_restore(nofs_flag);
189 btrfs_put_root(local_root);
190 ret = PTR_ERR(ipath);
193 * -ENOMEM, not a critical error, just output an generic error
197 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
198 warn->logical, warn->mirror_num, root, inum, offset);
201 ret = paths_from_inode(inum, ipath);
206 * We deliberately ignore the bit ipath might have been too small to
207 * hold all of the paths here
209 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
211 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
212 warn->logical, warn->mirror_num, root, inum, offset,
213 fs_info->sectorsize, nlink,
214 (char *)(unsigned long)ipath->fspath->val[i]);
217 btrfs_put_root(local_root);
223 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
224 warn->logical, warn->mirror_num, root, inum, offset, ret);
231 * Do extra user-friendly error output (e.g. lookup all the affected files).
233 * Return true if we succeeded doing the backref lookup.
234 * Return false if such lookup failed, and has to fallback to the old error message.
236 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
237 const u8 *csum, const u8 *csum_expected,
240 struct btrfs_fs_info *fs_info = inode->root->fs_info;
241 struct btrfs_path path = { 0 };
242 struct btrfs_key found_key = { 0 };
243 struct extent_buffer *eb;
244 struct btrfs_extent_item *ei;
245 const u32 csum_size = fs_info->csum_size;
251 mutex_lock(&fs_info->reloc_mutex);
252 logical = btrfs_get_reloc_bg_bytenr(fs_info);
253 mutex_unlock(&fs_info->reloc_mutex);
255 if (logical == U64_MAX) {
256 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
257 btrfs_warn_rl(fs_info,
258 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
259 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
260 CSUM_FMT_VALUE(csum_size, csum),
261 CSUM_FMT_VALUE(csum_size, csum_expected),
267 btrfs_warn_rl(fs_info,
268 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
269 inode->root->root_key.objectid,
270 btrfs_ino(inode), file_off, logical,
271 CSUM_FMT_VALUE(csum_size, csum),
272 CSUM_FMT_VALUE(csum_size, csum_expected),
275 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
277 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
282 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
283 item_size = btrfs_item_size(eb, path.slots[0]);
284 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
285 unsigned long ptr = 0;
290 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
291 item_size, &ref_root,
294 btrfs_warn_rl(fs_info,
295 "failed to resolve tree backref for logical %llu: %d",
302 btrfs_warn_rl(fs_info,
303 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
305 (ref_level ? "node" : "leaf"),
306 ref_level, ref_root);
308 btrfs_release_path(&path);
310 struct btrfs_backref_walk_ctx ctx = { 0 };
311 struct data_reloc_warn reloc_warn = { 0 };
313 btrfs_release_path(&path);
315 ctx.bytenr = found_key.objectid;
316 ctx.extent_item_pos = logical - found_key.objectid;
317 ctx.fs_info = fs_info;
319 reloc_warn.logical = logical;
320 reloc_warn.extent_item_size = found_key.offset;
321 reloc_warn.mirror_num = mirror_num;
322 reloc_warn.fs_info = fs_info;
324 iterate_extent_inodes(&ctx, true,
325 data_reloc_print_warning_inode, &reloc_warn);
329 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
330 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
332 struct btrfs_root *root = inode->root;
333 const u32 csum_size = root->fs_info->csum_size;
335 /* For data reloc tree, it's better to do a backref lookup instead. */
336 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
337 return print_data_reloc_error(inode, logical_start, csum,
338 csum_expected, mirror_num);
340 /* Output without objectid, which is more meaningful */
341 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
342 btrfs_warn_rl(root->fs_info,
343 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
344 root->root_key.objectid, btrfs_ino(inode),
346 CSUM_FMT_VALUE(csum_size, csum),
347 CSUM_FMT_VALUE(csum_size, csum_expected),
350 btrfs_warn_rl(root->fs_info,
351 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
352 root->root_key.objectid, btrfs_ino(inode),
354 CSUM_FMT_VALUE(csum_size, csum),
355 CSUM_FMT_VALUE(csum_size, csum_expected),
361 * Lock inode i_rwsem based on arguments passed.
363 * ilock_flags can have the following bit set:
365 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
366 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
368 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
370 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
372 if (ilock_flags & BTRFS_ILOCK_SHARED) {
373 if (ilock_flags & BTRFS_ILOCK_TRY) {
374 if (!inode_trylock_shared(&inode->vfs_inode))
379 inode_lock_shared(&inode->vfs_inode);
381 if (ilock_flags & BTRFS_ILOCK_TRY) {
382 if (!inode_trylock(&inode->vfs_inode))
387 inode_lock(&inode->vfs_inode);
389 if (ilock_flags & BTRFS_ILOCK_MMAP)
390 down_write(&inode->i_mmap_lock);
395 * Unock inode i_rwsem.
397 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
398 * to decide whether the lock acquired is shared or exclusive.
400 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
402 if (ilock_flags & BTRFS_ILOCK_MMAP)
403 up_write(&inode->i_mmap_lock);
404 if (ilock_flags & BTRFS_ILOCK_SHARED)
405 inode_unlock_shared(&inode->vfs_inode);
407 inode_unlock(&inode->vfs_inode);
411 * Cleanup all submitted ordered extents in specified range to handle errors
412 * from the btrfs_run_delalloc_range() callback.
414 * NOTE: caller must ensure that when an error happens, it can not call
415 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
416 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
417 * to be released, which we want to happen only when finishing the ordered
418 * extent (btrfs_finish_ordered_io()).
420 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
421 struct page *locked_page,
422 u64 offset, u64 bytes)
424 unsigned long index = offset >> PAGE_SHIFT;
425 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
426 u64 page_start = 0, page_end = 0;
430 page_start = page_offset(locked_page);
431 page_end = page_start + PAGE_SIZE - 1;
434 while (index <= end_index) {
436 * For locked page, we will call btrfs_mark_ordered_io_finished
437 * through btrfs_mark_ordered_io_finished() on it
438 * in run_delalloc_range() for the error handling, which will
439 * clear page Ordered and run the ordered extent accounting.
441 * Here we can't just clear the Ordered bit, or
442 * btrfs_mark_ordered_io_finished() would skip the accounting
443 * for the page range, and the ordered extent will never finish.
445 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
449 page = find_get_page(inode->vfs_inode.i_mapping, index);
455 * Here we just clear all Ordered bits for every page in the
456 * range, then btrfs_mark_ordered_io_finished() will handle
457 * the ordered extent accounting for the range.
459 btrfs_folio_clamp_clear_ordered(inode->root->fs_info,
460 page_folio(page), offset, bytes);
465 /* The locked page covers the full range, nothing needs to be done */
466 if (bytes + offset <= page_start + PAGE_SIZE)
469 * In case this page belongs to the delalloc range being
470 * instantiated then skip it, since the first page of a range is
471 * going to be properly cleaned up by the caller of
474 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
475 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
476 offset = page_offset(locked_page) + PAGE_SIZE;
480 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
483 static int btrfs_dirty_inode(struct btrfs_inode *inode);
485 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
486 struct btrfs_new_inode_args *args)
490 if (args->default_acl) {
491 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
497 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
501 if (!args->default_acl && !args->acl)
502 cache_no_acl(args->inode);
503 return btrfs_xattr_security_init(trans, args->inode, args->dir,
504 &args->dentry->d_name);
508 * this does all the hard work for inserting an inline extent into
509 * the btree. The caller should have done a btrfs_drop_extents so that
510 * no overlapping inline items exist in the btree
512 static int insert_inline_extent(struct btrfs_trans_handle *trans,
513 struct btrfs_path *path,
514 struct btrfs_inode *inode, bool extent_inserted,
515 size_t size, size_t compressed_size,
517 struct page **compressed_pages,
520 struct btrfs_root *root = inode->root;
521 struct extent_buffer *leaf;
522 struct page *page = NULL;
525 struct btrfs_file_extent_item *ei;
527 size_t cur_size = size;
530 ASSERT((compressed_size > 0 && compressed_pages) ||
531 (compressed_size == 0 && !compressed_pages));
533 if (compressed_size && compressed_pages)
534 cur_size = compressed_size;
536 if (!extent_inserted) {
537 struct btrfs_key key;
540 key.objectid = btrfs_ino(inode);
542 key.type = BTRFS_EXTENT_DATA_KEY;
544 datasize = btrfs_file_extent_calc_inline_size(cur_size);
545 ret = btrfs_insert_empty_item(trans, root, path, &key,
550 leaf = path->nodes[0];
551 ei = btrfs_item_ptr(leaf, path->slots[0],
552 struct btrfs_file_extent_item);
553 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
554 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
555 btrfs_set_file_extent_encryption(leaf, ei, 0);
556 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
557 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
558 ptr = btrfs_file_extent_inline_start(ei);
560 if (compress_type != BTRFS_COMPRESS_NONE) {
563 while (compressed_size > 0) {
564 cpage = compressed_pages[i];
565 cur_size = min_t(unsigned long, compressed_size,
568 kaddr = kmap_local_page(cpage);
569 write_extent_buffer(leaf, kaddr, ptr, cur_size);
574 compressed_size -= cur_size;
576 btrfs_set_file_extent_compression(leaf, ei,
579 page = find_get_page(inode->vfs_inode.i_mapping, 0);
580 btrfs_set_file_extent_compression(leaf, ei, 0);
581 kaddr = kmap_local_page(page);
582 write_extent_buffer(leaf, kaddr, ptr, size);
586 btrfs_mark_buffer_dirty(trans, leaf);
587 btrfs_release_path(path);
590 * We align size to sectorsize for inline extents just for simplicity
593 ret = btrfs_inode_set_file_extent_range(inode, 0,
594 ALIGN(size, root->fs_info->sectorsize));
599 * We're an inline extent, so nobody can extend the file past i_size
600 * without locking a page we already have locked.
602 * We must do any i_size and inode updates before we unlock the pages.
603 * Otherwise we could end up racing with unlink.
605 i_size = i_size_read(&inode->vfs_inode);
606 if (update_i_size && size > i_size) {
607 i_size_write(&inode->vfs_inode, size);
610 inode->disk_i_size = i_size;
618 * conditionally insert an inline extent into the file. This
619 * does the checks required to make sure the data is small enough
620 * to fit as an inline extent.
622 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
623 size_t compressed_size,
625 struct page **compressed_pages,
628 struct btrfs_drop_extents_args drop_args = { 0 };
629 struct btrfs_root *root = inode->root;
630 struct btrfs_fs_info *fs_info = root->fs_info;
631 struct btrfs_trans_handle *trans;
632 u64 data_len = (compressed_size ?: size);
634 struct btrfs_path *path;
637 * We can create an inline extent if it ends at or beyond the current
638 * i_size, is no larger than a sector (decompressed), and the (possibly
639 * compressed) data fits in a leaf and the configured maximum inline
642 if (size < i_size_read(&inode->vfs_inode) ||
643 size > fs_info->sectorsize ||
644 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
645 data_len > fs_info->max_inline)
648 path = btrfs_alloc_path();
652 trans = btrfs_join_transaction(root);
654 btrfs_free_path(path);
655 return PTR_ERR(trans);
657 trans->block_rsv = &inode->block_rsv;
659 drop_args.path = path;
661 drop_args.end = fs_info->sectorsize;
662 drop_args.drop_cache = true;
663 drop_args.replace_extent = true;
664 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
665 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
667 btrfs_abort_transaction(trans, ret);
671 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
672 size, compressed_size, compress_type,
673 compressed_pages, update_i_size);
674 if (ret && ret != -ENOSPC) {
675 btrfs_abort_transaction(trans, ret);
677 } else if (ret == -ENOSPC) {
682 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
683 ret = btrfs_update_inode(trans, inode);
684 if (ret && ret != -ENOSPC) {
685 btrfs_abort_transaction(trans, ret);
687 } else if (ret == -ENOSPC) {
692 btrfs_set_inode_full_sync(inode);
695 * Don't forget to free the reserved space, as for inlined extent
696 * it won't count as data extent, free them directly here.
697 * And at reserve time, it's always aligned to page size, so
698 * just free one page here.
700 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE, NULL);
701 btrfs_free_path(path);
702 btrfs_end_transaction(trans);
706 struct async_extent {
711 unsigned long nr_pages;
713 struct list_head list;
717 struct btrfs_inode *inode;
718 struct page *locked_page;
721 blk_opf_t write_flags;
722 struct list_head extents;
723 struct cgroup_subsys_state *blkcg_css;
724 struct btrfs_work work;
725 struct async_cow *async_cow;
730 struct async_chunk chunks[];
733 static noinline int add_async_extent(struct async_chunk *cow,
734 u64 start, u64 ram_size,
737 unsigned long nr_pages,
740 struct async_extent *async_extent;
742 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
743 BUG_ON(!async_extent); /* -ENOMEM */
744 async_extent->start = start;
745 async_extent->ram_size = ram_size;
746 async_extent->compressed_size = compressed_size;
747 async_extent->pages = pages;
748 async_extent->nr_pages = nr_pages;
749 async_extent->compress_type = compress_type;
750 list_add_tail(&async_extent->list, &cow->extents);
755 * Check if the inode needs to be submitted to compression, based on mount
756 * options, defragmentation, properties or heuristics.
758 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
761 struct btrfs_fs_info *fs_info = inode->root->fs_info;
763 if (!btrfs_inode_can_compress(inode)) {
764 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
765 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
770 * Special check for subpage.
772 * We lock the full page then run each delalloc range in the page, thus
773 * for the following case, we will hit some subpage specific corner case:
776 * | |///////| |///////|
779 * In above case, both range A and range B will try to unlock the full
780 * page [0, 64K), causing the one finished later will have page
781 * unlocked already, triggering various page lock requirement BUG_ON()s.
783 * So here we add an artificial limit that subpage compression can only
784 * if the range is fully page aligned.
786 * In theory we only need to ensure the first page is fully covered, but
787 * the tailing partial page will be locked until the full compression
788 * finishes, delaying the write of other range.
790 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
791 * first to prevent any submitted async extent to unlock the full page.
792 * By this, we can ensure for subpage case that only the last async_cow
793 * will unlock the full page.
795 if (fs_info->sectorsize < PAGE_SIZE) {
796 if (!PAGE_ALIGNED(start) ||
797 !PAGE_ALIGNED(end + 1))
802 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
805 if (inode->defrag_compress)
807 /* bad compression ratios */
808 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
810 if (btrfs_test_opt(fs_info, COMPRESS) ||
811 inode->flags & BTRFS_INODE_COMPRESS ||
812 inode->prop_compress)
813 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
817 static inline void inode_should_defrag(struct btrfs_inode *inode,
818 u64 start, u64 end, u64 num_bytes, u32 small_write)
820 /* If this is a small write inside eof, kick off a defrag */
821 if (num_bytes < small_write &&
822 (start > 0 || end + 1 < inode->disk_i_size))
823 btrfs_add_inode_defrag(NULL, inode, small_write);
827 * Work queue call back to started compression on a file and pages.
829 * This is done inside an ordered work queue, and the compression is spread
830 * across many cpus. The actual IO submission is step two, and the ordered work
831 * queue takes care of making sure that happens in the same order things were
832 * put onto the queue by writepages and friends.
834 * If this code finds it can't get good compression, it puts an entry onto the
835 * work queue to write the uncompressed bytes. This makes sure that both
836 * compressed inodes and uncompressed inodes are written in the same order that
837 * the flusher thread sent them down.
839 static void compress_file_range(struct btrfs_work *work)
841 struct async_chunk *async_chunk =
842 container_of(work, struct async_chunk, work);
843 struct btrfs_inode *inode = async_chunk->inode;
844 struct btrfs_fs_info *fs_info = inode->root->fs_info;
845 struct address_space *mapping = inode->vfs_inode.i_mapping;
846 u64 blocksize = fs_info->sectorsize;
847 u64 start = async_chunk->start;
848 u64 end = async_chunk->end;
853 unsigned long nr_pages;
854 unsigned long total_compressed = 0;
855 unsigned long total_in = 0;
858 int compress_type = fs_info->compress_type;
860 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
863 * We need to call clear_page_dirty_for_io on each page in the range.
864 * Otherwise applications with the file mmap'd can wander in and change
865 * the page contents while we are compressing them.
867 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
870 * We need to save i_size before now because it could change in between
871 * us evaluating the size and assigning it. This is because we lock and
872 * unlock the page in truncate and fallocate, and then modify the i_size
875 * The barriers are to emulate READ_ONCE, remove that once i_size_read
879 i_size = i_size_read(&inode->vfs_inode);
881 actual_end = min_t(u64, i_size, end + 1);
884 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
885 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
888 * we don't want to send crud past the end of i_size through
889 * compression, that's just a waste of CPU time. So, if the
890 * end of the file is before the start of our current
891 * requested range of bytes, we bail out to the uncompressed
892 * cleanup code that can deal with all of this.
894 * It isn't really the fastest way to fix things, but this is a
895 * very uncommon corner.
897 if (actual_end <= start)
898 goto cleanup_and_bail_uncompressed;
900 total_compressed = actual_end - start;
903 * Skip compression for a small file range(<=blocksize) that
904 * isn't an inline extent, since it doesn't save disk space at all.
906 if (total_compressed <= blocksize &&
907 (start > 0 || end + 1 < inode->disk_i_size))
908 goto cleanup_and_bail_uncompressed;
911 * For subpage case, we require full page alignment for the sector
913 * Thus we must also check against @actual_end, not just @end.
915 if (blocksize < PAGE_SIZE) {
916 if (!PAGE_ALIGNED(start) ||
917 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
918 goto cleanup_and_bail_uncompressed;
921 total_compressed = min_t(unsigned long, total_compressed,
922 BTRFS_MAX_UNCOMPRESSED);
927 * We do compression for mount -o compress and when the inode has not
928 * been flagged as NOCOMPRESS. This flag can change at any time if we
929 * discover bad compression ratios.
931 if (!inode_need_compress(inode, start, end))
932 goto cleanup_and_bail_uncompressed;
934 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
937 * Memory allocation failure is not a fatal error, we can fall
938 * back to uncompressed code.
940 goto cleanup_and_bail_uncompressed;
943 if (inode->defrag_compress)
944 compress_type = inode->defrag_compress;
945 else if (inode->prop_compress)
946 compress_type = inode->prop_compress;
948 /* Compression level is applied here. */
949 ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4),
950 mapping, start, pages, &nr_pages, &total_in,
953 goto mark_incompressible;
956 * Zero the tail end of the last page, as we might be sending it down
959 poff = offset_in_page(total_compressed);
961 memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff);
964 * Try to create an inline extent.
966 * If we didn't compress the entire range, try to create an uncompressed
967 * inline extent, else a compressed one.
969 * Check cow_file_range() for why we don't even try to create inline
970 * extent for the subpage case.
972 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
973 if (total_in < actual_end) {
974 ret = cow_file_range_inline(inode, actual_end, 0,
975 BTRFS_COMPRESS_NONE, NULL,
978 ret = cow_file_range_inline(inode, actual_end,
980 compress_type, pages,
984 unsigned long clear_flags = EXTENT_DELALLOC |
985 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
986 EXTENT_DO_ACCOUNTING;
989 mapping_set_error(mapping, -EIO);
992 * inline extent creation worked or returned error,
993 * we don't need to create any more async work items.
994 * Unlock and free up our temp pages.
996 * We use DO_ACCOUNTING here because we need the
997 * delalloc_release_metadata to be done _after_ we drop
998 * our outstanding extent for clearing delalloc for this
1001 extent_clear_unlock_delalloc(inode, start, end,
1005 PAGE_START_WRITEBACK |
1006 PAGE_END_WRITEBACK);
1012 * We aren't doing an inline extent. Round the compressed size up to a
1013 * block size boundary so the allocator does sane things.
1015 total_compressed = ALIGN(total_compressed, blocksize);
1018 * One last check to make sure the compression is really a win, compare
1019 * the page count read with the blocks on disk, compression must free at
1022 total_in = round_up(total_in, fs_info->sectorsize);
1023 if (total_compressed + blocksize > total_in)
1024 goto mark_incompressible;
1027 * The async work queues will take care of doing actual allocation on
1028 * disk for these compressed pages, and will submit the bios.
1030 add_async_extent(async_chunk, start, total_in, total_compressed, pages,
1031 nr_pages, compress_type);
1032 if (start + total_in < end) {
1039 mark_incompressible:
1040 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1041 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1042 cleanup_and_bail_uncompressed:
1043 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1044 BTRFS_COMPRESS_NONE);
1047 for (i = 0; i < nr_pages; i++) {
1048 WARN_ON(pages[i]->mapping);
1049 btrfs_free_compr_page(pages[i]);
1055 static void free_async_extent_pages(struct async_extent *async_extent)
1059 if (!async_extent->pages)
1062 for (i = 0; i < async_extent->nr_pages; i++) {
1063 WARN_ON(async_extent->pages[i]->mapping);
1064 btrfs_free_compr_page(async_extent->pages[i]);
1066 kfree(async_extent->pages);
1067 async_extent->nr_pages = 0;
1068 async_extent->pages = NULL;
1071 static void submit_uncompressed_range(struct btrfs_inode *inode,
1072 struct async_extent *async_extent,
1073 struct page *locked_page)
1075 u64 start = async_extent->start;
1076 u64 end = async_extent->start + async_extent->ram_size - 1;
1078 struct writeback_control wbc = {
1079 .sync_mode = WB_SYNC_ALL,
1080 .range_start = start,
1082 .no_cgroup_owner = 1,
1085 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1086 ret = run_delalloc_cow(inode, locked_page, start, end, &wbc, false);
1087 wbc_detach_inode(&wbc);
1089 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1091 const u64 page_start = page_offset(locked_page);
1093 set_page_writeback(locked_page);
1094 end_page_writeback(locked_page);
1095 btrfs_mark_ordered_io_finished(inode, locked_page,
1096 page_start, PAGE_SIZE,
1098 mapping_set_error(locked_page->mapping, ret);
1099 unlock_page(locked_page);
1104 static void submit_one_async_extent(struct async_chunk *async_chunk,
1105 struct async_extent *async_extent,
1108 struct btrfs_inode *inode = async_chunk->inode;
1109 struct extent_io_tree *io_tree = &inode->io_tree;
1110 struct btrfs_root *root = inode->root;
1111 struct btrfs_fs_info *fs_info = root->fs_info;
1112 struct btrfs_ordered_extent *ordered;
1113 struct btrfs_key ins;
1114 struct page *locked_page = NULL;
1115 struct extent_map *em;
1117 u64 start = async_extent->start;
1118 u64 end = async_extent->start + async_extent->ram_size - 1;
1120 if (async_chunk->blkcg_css)
1121 kthread_associate_blkcg(async_chunk->blkcg_css);
1124 * If async_chunk->locked_page is in the async_extent range, we need to
1127 if (async_chunk->locked_page) {
1128 u64 locked_page_start = page_offset(async_chunk->locked_page);
1129 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1131 if (!(start >= locked_page_end || end <= locked_page_start))
1132 locked_page = async_chunk->locked_page;
1134 lock_extent(io_tree, start, end, NULL);
1136 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1137 submit_uncompressed_range(inode, async_extent, locked_page);
1141 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1142 async_extent->compressed_size,
1143 async_extent->compressed_size,
1144 0, *alloc_hint, &ins, 1, 1);
1147 * Here we used to try again by going back to non-compressed
1148 * path for ENOSPC. But we can't reserve space even for
1149 * compressed size, how could it work for uncompressed size
1150 * which requires larger size? So here we directly go error
1156 /* Here we're doing allocation and writeback of the compressed pages */
1157 em = create_io_em(inode, start,
1158 async_extent->ram_size, /* len */
1159 start, /* orig_start */
1160 ins.objectid, /* block_start */
1161 ins.offset, /* block_len */
1162 ins.offset, /* orig_block_len */
1163 async_extent->ram_size, /* ram_bytes */
1164 async_extent->compress_type,
1165 BTRFS_ORDERED_COMPRESSED);
1168 goto out_free_reserve;
1170 free_extent_map(em);
1172 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1173 async_extent->ram_size, /* num_bytes */
1174 async_extent->ram_size, /* ram_bytes */
1175 ins.objectid, /* disk_bytenr */
1176 ins.offset, /* disk_num_bytes */
1178 1 << BTRFS_ORDERED_COMPRESSED,
1179 async_extent->compress_type);
1180 if (IS_ERR(ordered)) {
1181 btrfs_drop_extent_map_range(inode, start, end, false);
1182 ret = PTR_ERR(ordered);
1183 goto out_free_reserve;
1185 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1187 /* Clear dirty, set writeback and unlock the pages. */
1188 extent_clear_unlock_delalloc(inode, start, end,
1189 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1190 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1191 btrfs_submit_compressed_write(ordered,
1192 async_extent->pages, /* compressed_pages */
1193 async_extent->nr_pages,
1194 async_chunk->write_flags, true);
1195 *alloc_hint = ins.objectid + ins.offset;
1197 if (async_chunk->blkcg_css)
1198 kthread_associate_blkcg(NULL);
1199 kfree(async_extent);
1203 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1204 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1206 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1207 extent_clear_unlock_delalloc(inode, start, end,
1208 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1209 EXTENT_DELALLOC_NEW |
1210 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1211 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1212 PAGE_END_WRITEBACK);
1213 free_async_extent_pages(async_extent);
1214 if (async_chunk->blkcg_css)
1215 kthread_associate_blkcg(NULL);
1216 btrfs_debug(fs_info,
1217 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1218 root->root_key.objectid, btrfs_ino(inode), start,
1219 async_extent->ram_size, ret);
1220 kfree(async_extent);
1223 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1226 struct extent_map_tree *em_tree = &inode->extent_tree;
1227 struct extent_map *em;
1230 read_lock(&em_tree->lock);
1231 em = search_extent_mapping(em_tree, start, num_bytes);
1234 * if block start isn't an actual block number then find the
1235 * first block in this inode and use that as a hint. If that
1236 * block is also bogus then just don't worry about it.
1238 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1239 free_extent_map(em);
1240 em = search_extent_mapping(em_tree, 0, 0);
1241 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1242 alloc_hint = em->block_start;
1244 free_extent_map(em);
1246 alloc_hint = em->block_start;
1247 free_extent_map(em);
1250 read_unlock(&em_tree->lock);
1256 * when extent_io.c finds a delayed allocation range in the file,
1257 * the call backs end up in this code. The basic idea is to
1258 * allocate extents on disk for the range, and create ordered data structs
1259 * in ram to track those extents.
1261 * locked_page is the page that writepage had locked already. We use
1262 * it to make sure we don't do extra locks or unlocks.
1264 * When this function fails, it unlocks all pages except @locked_page.
1266 * When this function successfully creates an inline extent, it returns 1 and
1267 * unlocks all pages including locked_page and starts I/O on them.
1268 * (In reality inline extents are limited to a single page, so locked_page is
1269 * the only page handled anyway).
1271 * When this function succeed and creates a normal extent, the page locking
1272 * status depends on the passed in flags:
1274 * - If @keep_locked is set, all pages are kept locked.
1275 * - Else all pages except for @locked_page are unlocked.
1277 * When a failure happens in the second or later iteration of the
1278 * while-loop, the ordered extents created in previous iterations are kept
1279 * intact. So, the caller must clean them up by calling
1280 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1283 static noinline int cow_file_range(struct btrfs_inode *inode,
1284 struct page *locked_page, u64 start, u64 end,
1286 bool keep_locked, bool no_inline)
1288 struct btrfs_root *root = inode->root;
1289 struct btrfs_fs_info *fs_info = root->fs_info;
1291 u64 orig_start = start;
1293 unsigned long ram_size;
1294 u64 cur_alloc_size = 0;
1296 u64 blocksize = fs_info->sectorsize;
1297 struct btrfs_key ins;
1298 struct extent_map *em;
1299 unsigned clear_bits;
1300 unsigned long page_ops;
1301 bool extent_reserved = false;
1304 if (btrfs_is_free_space_inode(inode)) {
1309 num_bytes = ALIGN(end - start + 1, blocksize);
1310 num_bytes = max(blocksize, num_bytes);
1311 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1313 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1316 * Due to the page size limit, for subpage we can only trigger the
1317 * writeback for the dirty sectors of page, that means data writeback
1318 * is doing more writeback than what we want.
1320 * This is especially unexpected for some call sites like fallocate,
1321 * where we only increase i_size after everything is done.
1322 * This means we can trigger inline extent even if we didn't want to.
1323 * So here we skip inline extent creation completely.
1325 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1326 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1329 /* lets try to make an inline extent */
1330 ret = cow_file_range_inline(inode, actual_end, 0,
1331 BTRFS_COMPRESS_NONE, NULL, false);
1334 * We use DO_ACCOUNTING here because we need the
1335 * delalloc_release_metadata to be run _after_ we drop
1336 * our outstanding extent for clearing delalloc for this
1339 extent_clear_unlock_delalloc(inode, start, end,
1341 EXTENT_LOCKED | EXTENT_DELALLOC |
1342 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1343 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1344 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1346 * locked_page is locked by the caller of
1347 * writepage_delalloc(), not locked by
1348 * __process_pages_contig().
1350 * We can't let __process_pages_contig() to unlock it,
1351 * as it doesn't have any subpage::writers recorded.
1353 * Here we manually unlock the page, since the caller
1354 * can't determine if it's an inline extent or a
1355 * compressed extent.
1357 unlock_page(locked_page);
1360 } else if (ret < 0) {
1365 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1368 * Relocation relies on the relocated extents to have exactly the same
1369 * size as the original extents. Normally writeback for relocation data
1370 * extents follows a NOCOW path because relocation preallocates the
1371 * extents. However, due to an operation such as scrub turning a block
1372 * group to RO mode, it may fallback to COW mode, so we must make sure
1373 * an extent allocated during COW has exactly the requested size and can
1374 * not be split into smaller extents, otherwise relocation breaks and
1375 * fails during the stage where it updates the bytenr of file extent
1378 if (btrfs_is_data_reloc_root(root))
1379 min_alloc_size = num_bytes;
1381 min_alloc_size = fs_info->sectorsize;
1383 while (num_bytes > 0) {
1384 struct btrfs_ordered_extent *ordered;
1386 cur_alloc_size = num_bytes;
1387 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1388 min_alloc_size, 0, alloc_hint,
1390 if (ret == -EAGAIN) {
1392 * btrfs_reserve_extent only returns -EAGAIN for zoned
1393 * file systems, which is an indication that there are
1394 * no active zones to allocate from at the moment.
1396 * If this is the first loop iteration, wait for at
1397 * least one zone to finish before retrying the
1398 * allocation. Otherwise ask the caller to write out
1399 * the already allocated blocks before coming back to
1400 * us, or return -ENOSPC if it can't handle retries.
1402 ASSERT(btrfs_is_zoned(fs_info));
1403 if (start == orig_start) {
1404 wait_on_bit_io(&inode->root->fs_info->flags,
1405 BTRFS_FS_NEED_ZONE_FINISH,
1406 TASK_UNINTERRUPTIBLE);
1410 *done_offset = start - 1;
1417 cur_alloc_size = ins.offset;
1418 extent_reserved = true;
1420 ram_size = ins.offset;
1421 em = create_io_em(inode, start, ins.offset, /* len */
1422 start, /* orig_start */
1423 ins.objectid, /* block_start */
1424 ins.offset, /* block_len */
1425 ins.offset, /* orig_block_len */
1426 ram_size, /* ram_bytes */
1427 BTRFS_COMPRESS_NONE, /* compress_type */
1428 BTRFS_ORDERED_REGULAR /* type */);
1433 free_extent_map(em);
1435 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1436 ram_size, ins.objectid, cur_alloc_size,
1437 0, 1 << BTRFS_ORDERED_REGULAR,
1438 BTRFS_COMPRESS_NONE);
1439 if (IS_ERR(ordered)) {
1440 ret = PTR_ERR(ordered);
1441 goto out_drop_extent_cache;
1444 if (btrfs_is_data_reloc_root(root)) {
1445 ret = btrfs_reloc_clone_csums(ordered);
1448 * Only drop cache here, and process as normal.
1450 * We must not allow extent_clear_unlock_delalloc()
1451 * at out_unlock label to free meta of this ordered
1452 * extent, as its meta should be freed by
1453 * btrfs_finish_ordered_io().
1455 * So we must continue until @start is increased to
1456 * skip current ordered extent.
1459 btrfs_drop_extent_map_range(inode, start,
1460 start + ram_size - 1,
1463 btrfs_put_ordered_extent(ordered);
1465 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1468 * We're not doing compressed IO, don't unlock the first page
1469 * (which the caller expects to stay locked), don't clear any
1470 * dirty bits and don't set any writeback bits
1472 * Do set the Ordered (Private2) bit so we know this page was
1473 * properly setup for writepage.
1475 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1476 page_ops |= PAGE_SET_ORDERED;
1478 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1480 EXTENT_LOCKED | EXTENT_DELALLOC,
1482 if (num_bytes < cur_alloc_size)
1485 num_bytes -= cur_alloc_size;
1486 alloc_hint = ins.objectid + ins.offset;
1487 start += cur_alloc_size;
1488 extent_reserved = false;
1491 * btrfs_reloc_clone_csums() error, since start is increased
1492 * extent_clear_unlock_delalloc() at out_unlock label won't
1493 * free metadata of current ordered extent, we're OK to exit.
1503 out_drop_extent_cache:
1504 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1506 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1507 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1510 * Now, we have three regions to clean up:
1512 * |-------(1)----|---(2)---|-------------(3)----------|
1513 * `- orig_start `- start `- start + cur_alloc_size `- end
1515 * We process each region below.
1518 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1519 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1520 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1523 * For the range (1). We have already instantiated the ordered extents
1524 * for this region. They are cleaned up by
1525 * btrfs_cleanup_ordered_extents() in e.g,
1526 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1527 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1528 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1531 * However, in case of @keep_locked, we still need to unlock the pages
1532 * (except @locked_page) to ensure all the pages are unlocked.
1534 if (keep_locked && orig_start < start) {
1536 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1537 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1538 locked_page, 0, page_ops);
1542 * For the range (2). If we reserved an extent for our delalloc range
1543 * (or a subrange) and failed to create the respective ordered extent,
1544 * then it means that when we reserved the extent we decremented the
1545 * extent's size from the data space_info's bytes_may_use counter and
1546 * incremented the space_info's bytes_reserved counter by the same
1547 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1548 * to decrement again the data space_info's bytes_may_use counter,
1549 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1551 if (extent_reserved) {
1552 extent_clear_unlock_delalloc(inode, start,
1553 start + cur_alloc_size - 1,
1557 start += cur_alloc_size;
1561 * For the range (3). We never touched the region. In addition to the
1562 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1563 * space_info's bytes_may_use counter, reserved in
1564 * btrfs_check_data_free_space().
1567 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1568 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1569 clear_bits, page_ops);
1575 * Phase two of compressed writeback. This is the ordered portion of the code,
1576 * which only gets called in the order the work was queued. We walk all the
1577 * async extents created by compress_file_range and send them down to the disk.
1579 * If called with @do_free == true then it'll try to finish the work and free
1580 * the work struct eventually.
1582 static noinline void submit_compressed_extents(struct btrfs_work *work, bool do_free)
1584 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1586 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1587 struct async_extent *async_extent;
1588 unsigned long nr_pages;
1592 struct async_chunk *async_chunk;
1593 struct async_cow *async_cow;
1595 async_chunk = container_of(work, struct async_chunk, work);
1596 btrfs_add_delayed_iput(async_chunk->inode);
1597 if (async_chunk->blkcg_css)
1598 css_put(async_chunk->blkcg_css);
1600 async_cow = async_chunk->async_cow;
1601 if (atomic_dec_and_test(&async_cow->num_chunks))
1606 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1609 while (!list_empty(&async_chunk->extents)) {
1610 async_extent = list_entry(async_chunk->extents.next,
1611 struct async_extent, list);
1612 list_del(&async_extent->list);
1613 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1616 /* atomic_sub_return implies a barrier */
1617 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1619 cond_wake_up_nomb(&fs_info->async_submit_wait);
1622 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1623 struct page *locked_page, u64 start,
1624 u64 end, struct writeback_control *wbc)
1626 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1627 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1628 struct async_cow *ctx;
1629 struct async_chunk *async_chunk;
1630 unsigned long nr_pages;
1631 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1634 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1636 nofs_flag = memalloc_nofs_save();
1637 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1638 memalloc_nofs_restore(nofs_flag);
1642 unlock_extent(&inode->io_tree, start, end, NULL);
1643 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1645 async_chunk = ctx->chunks;
1646 atomic_set(&ctx->num_chunks, num_chunks);
1648 for (i = 0; i < num_chunks; i++) {
1649 u64 cur_end = min(end, start + SZ_512K - 1);
1652 * igrab is called higher up in the call chain, take only the
1653 * lightweight reference for the callback lifetime
1655 ihold(&inode->vfs_inode);
1656 async_chunk[i].async_cow = ctx;
1657 async_chunk[i].inode = inode;
1658 async_chunk[i].start = start;
1659 async_chunk[i].end = cur_end;
1660 async_chunk[i].write_flags = write_flags;
1661 INIT_LIST_HEAD(&async_chunk[i].extents);
1664 * The locked_page comes all the way from writepage and its
1665 * the original page we were actually given. As we spread
1666 * this large delalloc region across multiple async_chunk
1667 * structs, only the first struct needs a pointer to locked_page
1669 * This way we don't need racey decisions about who is supposed
1674 * Depending on the compressibility, the pages might or
1675 * might not go through async. We want all of them to
1676 * be accounted against wbc once. Let's do it here
1677 * before the paths diverge. wbc accounting is used
1678 * only for foreign writeback detection and doesn't
1679 * need full accuracy. Just account the whole thing
1680 * against the first page.
1682 wbc_account_cgroup_owner(wbc, locked_page,
1684 async_chunk[i].locked_page = locked_page;
1687 async_chunk[i].locked_page = NULL;
1690 if (blkcg_css != blkcg_root_css) {
1692 async_chunk[i].blkcg_css = blkcg_css;
1693 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1695 async_chunk[i].blkcg_css = NULL;
1698 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1699 submit_compressed_extents);
1701 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1702 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1704 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1706 start = cur_end + 1;
1712 * Run the delalloc range from start to end, and write back any dirty pages
1713 * covered by the range.
1715 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1716 struct page *locked_page, u64 start,
1717 u64 end, struct writeback_control *wbc,
1720 u64 done_offset = end;
1723 while (start <= end) {
1724 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1728 extent_write_locked_range(&inode->vfs_inode, locked_page, start,
1729 done_offset, wbc, pages_dirty);
1730 start = done_offset + 1;
1736 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1737 u64 bytenr, u64 num_bytes, bool nowait)
1739 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1740 struct btrfs_ordered_sum *sums;
1744 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1746 if (ret == 0 && list_empty(&list))
1749 while (!list_empty(&list)) {
1750 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1751 list_del(&sums->list);
1759 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1760 const u64 start, const u64 end)
1762 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1763 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1764 const u64 range_bytes = end + 1 - start;
1765 struct extent_io_tree *io_tree = &inode->io_tree;
1766 u64 range_start = start;
1771 * If EXTENT_NORESERVE is set it means that when the buffered write was
1772 * made we had not enough available data space and therefore we did not
1773 * reserve data space for it, since we though we could do NOCOW for the
1774 * respective file range (either there is prealloc extent or the inode
1775 * has the NOCOW bit set).
1777 * However when we need to fallback to COW mode (because for example the
1778 * block group for the corresponding extent was turned to RO mode by a
1779 * scrub or relocation) we need to do the following:
1781 * 1) We increment the bytes_may_use counter of the data space info.
1782 * If COW succeeds, it allocates a new data extent and after doing
1783 * that it decrements the space info's bytes_may_use counter and
1784 * increments its bytes_reserved counter by the same amount (we do
1785 * this at btrfs_add_reserved_bytes()). So we need to increment the
1786 * bytes_may_use counter to compensate (when space is reserved at
1787 * buffered write time, the bytes_may_use counter is incremented);
1789 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1790 * that if the COW path fails for any reason, it decrements (through
1791 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1792 * data space info, which we incremented in the step above.
1794 * If we need to fallback to cow and the inode corresponds to a free
1795 * space cache inode or an inode of the data relocation tree, we must
1796 * also increment bytes_may_use of the data space_info for the same
1797 * reason. Space caches and relocated data extents always get a prealloc
1798 * extent for them, however scrub or balance may have set the block
1799 * group that contains that extent to RO mode and therefore force COW
1800 * when starting writeback.
1802 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1803 EXTENT_NORESERVE, 0, NULL);
1804 if (count > 0 || is_space_ino || is_reloc_ino) {
1806 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1807 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1809 if (is_space_ino || is_reloc_ino)
1810 bytes = range_bytes;
1812 spin_lock(&sinfo->lock);
1813 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1814 spin_unlock(&sinfo->lock);
1817 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1822 * Don't try to create inline extents, as a mix of inline extent that
1823 * is written out and unlocked directly and a normal NOCOW extent
1826 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1831 struct can_nocow_file_extent_args {
1834 /* Start file offset of the range we want to NOCOW. */
1836 /* End file offset (inclusive) of the range we want to NOCOW. */
1838 bool writeback_path;
1841 * Free the path passed to can_nocow_file_extent() once it's not needed
1846 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1851 /* Number of bytes that can be written to in NOCOW mode. */
1856 * Check if we can NOCOW the file extent that the path points to.
1857 * This function may return with the path released, so the caller should check
1858 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1860 * Returns: < 0 on error
1861 * 0 if we can not NOCOW
1864 static int can_nocow_file_extent(struct btrfs_path *path,
1865 struct btrfs_key *key,
1866 struct btrfs_inode *inode,
1867 struct can_nocow_file_extent_args *args)
1869 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1870 struct extent_buffer *leaf = path->nodes[0];
1871 struct btrfs_root *root = inode->root;
1872 struct btrfs_file_extent_item *fi;
1877 bool nowait = path->nowait;
1879 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1880 extent_type = btrfs_file_extent_type(leaf, fi);
1882 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1885 /* Can't access these fields unless we know it's not an inline extent. */
1886 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1887 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1888 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1890 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1891 extent_type == BTRFS_FILE_EXTENT_REG)
1895 * If the extent was created before the generation where the last snapshot
1896 * for its subvolume was created, then this implies the extent is shared,
1897 * hence we must COW.
1899 if (!args->strict &&
1900 btrfs_file_extent_generation(leaf, fi) <=
1901 btrfs_root_last_snapshot(&root->root_item))
1904 /* An explicit hole, must COW. */
1905 if (args->disk_bytenr == 0)
1908 /* Compressed/encrypted/encoded extents must be COWed. */
1909 if (btrfs_file_extent_compression(leaf, fi) ||
1910 btrfs_file_extent_encryption(leaf, fi) ||
1911 btrfs_file_extent_other_encoding(leaf, fi))
1914 extent_end = btrfs_file_extent_end(path);
1917 * The following checks can be expensive, as they need to take other
1918 * locks and do btree or rbtree searches, so release the path to avoid
1919 * blocking other tasks for too long.
1921 btrfs_release_path(path);
1923 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1924 key->offset - args->extent_offset,
1925 args->disk_bytenr, args->strict, path);
1926 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1930 if (args->free_path) {
1932 * We don't need the path anymore, plus through the
1933 * csum_exist_in_range() call below we will end up allocating
1934 * another path. So free the path to avoid unnecessary extra
1937 btrfs_free_path(path);
1941 /* If there are pending snapshots for this root, we must COW. */
1942 if (args->writeback_path && !is_freespace_inode &&
1943 atomic_read(&root->snapshot_force_cow))
1946 args->disk_bytenr += args->extent_offset;
1947 args->disk_bytenr += args->start - key->offset;
1948 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1951 * Force COW if csums exist in the range. This ensures that csums for a
1952 * given extent are either valid or do not exist.
1954 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1956 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1962 if (args->free_path && path)
1963 btrfs_free_path(path);
1965 return ret < 0 ? ret : can_nocow;
1969 * when nowcow writeback call back. This checks for snapshots or COW copies
1970 * of the extents that exist in the file, and COWs the file as required.
1972 * If no cow copies or snapshots exist, we write directly to the existing
1975 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1976 struct page *locked_page,
1977 const u64 start, const u64 end)
1979 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1980 struct btrfs_root *root = inode->root;
1981 struct btrfs_path *path;
1982 u64 cow_start = (u64)-1;
1983 u64 cur_offset = start;
1985 bool check_prev = true;
1986 u64 ino = btrfs_ino(inode);
1987 struct can_nocow_file_extent_args nocow_args = { 0 };
1990 * Normally on a zoned device we're only doing COW writes, but in case
1991 * of relocation on a zoned filesystem serializes I/O so that we're only
1992 * writing sequentially and can end up here as well.
1994 ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root));
1996 path = btrfs_alloc_path();
2002 nocow_args.end = end;
2003 nocow_args.writeback_path = true;
2006 struct btrfs_block_group *nocow_bg = NULL;
2007 struct btrfs_ordered_extent *ordered;
2008 struct btrfs_key found_key;
2009 struct btrfs_file_extent_item *fi;
2010 struct extent_buffer *leaf;
2017 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2023 * If there is no extent for our range when doing the initial
2024 * search, then go back to the previous slot as it will be the
2025 * one containing the search offset
2027 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2028 leaf = path->nodes[0];
2029 btrfs_item_key_to_cpu(leaf, &found_key,
2030 path->slots[0] - 1);
2031 if (found_key.objectid == ino &&
2032 found_key.type == BTRFS_EXTENT_DATA_KEY)
2037 /* Go to next leaf if we have exhausted the current one */
2038 leaf = path->nodes[0];
2039 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2040 ret = btrfs_next_leaf(root, path);
2045 leaf = path->nodes[0];
2048 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2050 /* Didn't find anything for our INO */
2051 if (found_key.objectid > ino)
2054 * Keep searching until we find an EXTENT_ITEM or there are no
2055 * more extents for this inode
2057 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2058 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2063 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2064 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2065 found_key.offset > end)
2069 * If the found extent starts after requested offset, then
2070 * adjust extent_end to be right before this extent begins
2072 if (found_key.offset > cur_offset) {
2073 extent_end = found_key.offset;
2079 * Found extent which begins before our range and potentially
2082 fi = btrfs_item_ptr(leaf, path->slots[0],
2083 struct btrfs_file_extent_item);
2084 extent_type = btrfs_file_extent_type(leaf, fi);
2085 /* If this is triggered then we have a memory corruption. */
2086 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2087 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2091 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2092 extent_end = btrfs_file_extent_end(path);
2095 * If the extent we got ends before our current offset, skip to
2098 if (extent_end <= cur_offset) {
2103 nocow_args.start = cur_offset;
2104 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2111 nocow_bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2115 * If we can't perform NOCOW writeback for the range,
2116 * then record the beginning of the range that needs to
2117 * be COWed. It will be written out before the next
2118 * NOCOW range if we find one, or when exiting this
2121 if (cow_start == (u64)-1)
2122 cow_start = cur_offset;
2123 cur_offset = extent_end;
2124 if (cur_offset > end)
2126 if (!path->nodes[0])
2133 * COW range from cow_start to found_key.offset - 1. As the key
2134 * will contain the beginning of the first extent that can be
2135 * NOCOW, following one which needs to be COW'ed
2137 if (cow_start != (u64)-1) {
2138 ret = fallback_to_cow(inode, locked_page,
2139 cow_start, found_key.offset - 1);
2140 cow_start = (u64)-1;
2142 btrfs_dec_nocow_writers(nocow_bg);
2147 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2148 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2150 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2151 struct extent_map *em;
2153 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2155 nocow_args.disk_bytenr, /* block_start */
2156 nocow_args.num_bytes, /* block_len */
2157 nocow_args.disk_num_bytes, /* orig_block_len */
2158 ram_bytes, BTRFS_COMPRESS_NONE,
2159 BTRFS_ORDERED_PREALLOC);
2161 btrfs_dec_nocow_writers(nocow_bg);
2165 free_extent_map(em);
2168 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2169 nocow_args.num_bytes, nocow_args.num_bytes,
2170 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2172 ? (1 << BTRFS_ORDERED_PREALLOC)
2173 : (1 << BTRFS_ORDERED_NOCOW),
2174 BTRFS_COMPRESS_NONE);
2175 btrfs_dec_nocow_writers(nocow_bg);
2176 if (IS_ERR(ordered)) {
2178 btrfs_drop_extent_map_range(inode, cur_offset,
2181 ret = PTR_ERR(ordered);
2185 if (btrfs_is_data_reloc_root(root))
2187 * Error handled later, as we must prevent
2188 * extent_clear_unlock_delalloc() in error handler
2189 * from freeing metadata of created ordered extent.
2191 ret = btrfs_reloc_clone_csums(ordered);
2192 btrfs_put_ordered_extent(ordered);
2194 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2195 locked_page, EXTENT_LOCKED |
2197 EXTENT_CLEAR_DATA_RESV,
2198 PAGE_UNLOCK | PAGE_SET_ORDERED);
2200 cur_offset = extent_end;
2203 * btrfs_reloc_clone_csums() error, now we're OK to call error
2204 * handler, as metadata for created ordered extent will only
2205 * be freed by btrfs_finish_ordered_io().
2209 if (cur_offset > end)
2212 btrfs_release_path(path);
2214 if (cur_offset <= end && cow_start == (u64)-1)
2215 cow_start = cur_offset;
2217 if (cow_start != (u64)-1) {
2219 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2220 cow_start = (u64)-1;
2225 btrfs_free_path(path);
2230 * If an error happened while a COW region is outstanding, cur_offset
2231 * needs to be reset to cow_start to ensure the COW region is unlocked
2234 if (cow_start != (u64)-1)
2235 cur_offset = cow_start;
2236 if (cur_offset < end)
2237 extent_clear_unlock_delalloc(inode, cur_offset, end,
2238 locked_page, EXTENT_LOCKED |
2239 EXTENT_DELALLOC | EXTENT_DEFRAG |
2240 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2241 PAGE_START_WRITEBACK |
2242 PAGE_END_WRITEBACK);
2243 btrfs_free_path(path);
2247 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2249 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2250 if (inode->defrag_bytes &&
2251 test_range_bit_exists(&inode->io_tree, start, end, EXTENT_DEFRAG))
2259 * Function to process delayed allocation (create CoW) for ranges which are
2260 * being touched for the first time.
2262 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2263 u64 start, u64 end, struct writeback_control *wbc)
2265 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2269 * The range must cover part of the @locked_page, or a return of 1
2270 * can confuse the caller.
2272 ASSERT(!(end <= page_offset(locked_page) ||
2273 start >= page_offset(locked_page) + PAGE_SIZE));
2275 if (should_nocow(inode, start, end)) {
2276 ret = run_delalloc_nocow(inode, locked_page, start, end);
2280 if (btrfs_inode_can_compress(inode) &&
2281 inode_need_compress(inode, start, end) &&
2282 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2286 ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2289 ret = cow_file_range(inode, locked_page, start, end, NULL,
2294 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2299 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2300 struct extent_state *orig, u64 split)
2302 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2305 /* not delalloc, ignore it */
2306 if (!(orig->state & EXTENT_DELALLOC))
2309 size = orig->end - orig->start + 1;
2310 if (size > fs_info->max_extent_size) {
2315 * See the explanation in btrfs_merge_delalloc_extent, the same
2316 * applies here, just in reverse.
2318 new_size = orig->end - split + 1;
2319 num_extents = count_max_extents(fs_info, new_size);
2320 new_size = split - orig->start;
2321 num_extents += count_max_extents(fs_info, new_size);
2322 if (count_max_extents(fs_info, size) >= num_extents)
2326 spin_lock(&inode->lock);
2327 btrfs_mod_outstanding_extents(inode, 1);
2328 spin_unlock(&inode->lock);
2332 * Handle merged delayed allocation extents so we can keep track of new extents
2333 * that are just merged onto old extents, such as when we are doing sequential
2334 * writes, so we can properly account for the metadata space we'll need.
2336 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2337 struct extent_state *other)
2339 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2340 u64 new_size, old_size;
2343 /* not delalloc, ignore it */
2344 if (!(other->state & EXTENT_DELALLOC))
2347 if (new->start > other->start)
2348 new_size = new->end - other->start + 1;
2350 new_size = other->end - new->start + 1;
2352 /* we're not bigger than the max, unreserve the space and go */
2353 if (new_size <= fs_info->max_extent_size) {
2354 spin_lock(&inode->lock);
2355 btrfs_mod_outstanding_extents(inode, -1);
2356 spin_unlock(&inode->lock);
2361 * We have to add up either side to figure out how many extents were
2362 * accounted for before we merged into one big extent. If the number of
2363 * extents we accounted for is <= the amount we need for the new range
2364 * then we can return, otherwise drop. Think of it like this
2368 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2369 * need 2 outstanding extents, on one side we have 1 and the other side
2370 * we have 1 so they are == and we can return. But in this case
2372 * [MAX_SIZE+4k][MAX_SIZE+4k]
2374 * Each range on their own accounts for 2 extents, but merged together
2375 * they are only 3 extents worth of accounting, so we need to drop in
2378 old_size = other->end - other->start + 1;
2379 num_extents = count_max_extents(fs_info, old_size);
2380 old_size = new->end - new->start + 1;
2381 num_extents += count_max_extents(fs_info, old_size);
2382 if (count_max_extents(fs_info, new_size) >= num_extents)
2385 spin_lock(&inode->lock);
2386 btrfs_mod_outstanding_extents(inode, -1);
2387 spin_unlock(&inode->lock);
2390 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2391 struct btrfs_inode *inode)
2393 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2395 spin_lock(&root->delalloc_lock);
2396 if (list_empty(&inode->delalloc_inodes)) {
2397 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2398 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2399 root->nr_delalloc_inodes++;
2400 if (root->nr_delalloc_inodes == 1) {
2401 spin_lock(&fs_info->delalloc_root_lock);
2402 BUG_ON(!list_empty(&root->delalloc_root));
2403 list_add_tail(&root->delalloc_root,
2404 &fs_info->delalloc_roots);
2405 spin_unlock(&fs_info->delalloc_root_lock);
2408 spin_unlock(&root->delalloc_lock);
2411 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2412 struct btrfs_inode *inode)
2414 struct btrfs_fs_info *fs_info = root->fs_info;
2416 if (!list_empty(&inode->delalloc_inodes)) {
2417 list_del_init(&inode->delalloc_inodes);
2418 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2419 &inode->runtime_flags);
2420 root->nr_delalloc_inodes--;
2421 if (!root->nr_delalloc_inodes) {
2422 ASSERT(list_empty(&root->delalloc_inodes));
2423 spin_lock(&fs_info->delalloc_root_lock);
2424 BUG_ON(list_empty(&root->delalloc_root));
2425 list_del_init(&root->delalloc_root);
2426 spin_unlock(&fs_info->delalloc_root_lock);
2431 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2432 struct btrfs_inode *inode)
2434 spin_lock(&root->delalloc_lock);
2435 __btrfs_del_delalloc_inode(root, inode);
2436 spin_unlock(&root->delalloc_lock);
2440 * Properly track delayed allocation bytes in the inode and to maintain the
2441 * list of inodes that have pending delalloc work to be done.
2443 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2446 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2448 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2451 * set_bit and clear bit hooks normally require _irqsave/restore
2452 * but in this case, we are only testing for the DELALLOC
2453 * bit, which is only set or cleared with irqs on
2455 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2456 struct btrfs_root *root = inode->root;
2457 u64 len = state->end + 1 - state->start;
2458 u32 num_extents = count_max_extents(fs_info, len);
2459 bool do_list = !btrfs_is_free_space_inode(inode);
2461 spin_lock(&inode->lock);
2462 btrfs_mod_outstanding_extents(inode, num_extents);
2463 spin_unlock(&inode->lock);
2465 /* For sanity tests */
2466 if (btrfs_is_testing(fs_info))
2469 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2470 fs_info->delalloc_batch);
2471 spin_lock(&inode->lock);
2472 inode->delalloc_bytes += len;
2473 if (bits & EXTENT_DEFRAG)
2474 inode->defrag_bytes += len;
2475 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2476 &inode->runtime_flags))
2477 btrfs_add_delalloc_inodes(root, inode);
2478 spin_unlock(&inode->lock);
2481 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2482 (bits & EXTENT_DELALLOC_NEW)) {
2483 spin_lock(&inode->lock);
2484 inode->new_delalloc_bytes += state->end + 1 - state->start;
2485 spin_unlock(&inode->lock);
2490 * Once a range is no longer delalloc this function ensures that proper
2491 * accounting happens.
2493 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2494 struct extent_state *state, u32 bits)
2496 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2497 u64 len = state->end + 1 - state->start;
2498 u32 num_extents = count_max_extents(fs_info, len);
2500 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2501 spin_lock(&inode->lock);
2502 inode->defrag_bytes -= len;
2503 spin_unlock(&inode->lock);
2507 * set_bit and clear bit hooks normally require _irqsave/restore
2508 * but in this case, we are only testing for the DELALLOC
2509 * bit, which is only set or cleared with irqs on
2511 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2512 struct btrfs_root *root = inode->root;
2513 bool do_list = !btrfs_is_free_space_inode(inode);
2515 spin_lock(&inode->lock);
2516 btrfs_mod_outstanding_extents(inode, -num_extents);
2517 spin_unlock(&inode->lock);
2520 * We don't reserve metadata space for space cache inodes so we
2521 * don't need to call delalloc_release_metadata if there is an
2524 if (bits & EXTENT_CLEAR_META_RESV &&
2525 root != fs_info->tree_root)
2526 btrfs_delalloc_release_metadata(inode, len, false);
2528 /* For sanity tests. */
2529 if (btrfs_is_testing(fs_info))
2532 if (!btrfs_is_data_reloc_root(root) &&
2533 do_list && !(state->state & EXTENT_NORESERVE) &&
2534 (bits & EXTENT_CLEAR_DATA_RESV))
2535 btrfs_free_reserved_data_space_noquota(fs_info, len);
2537 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2538 fs_info->delalloc_batch);
2539 spin_lock(&inode->lock);
2540 inode->delalloc_bytes -= len;
2541 if (do_list && inode->delalloc_bytes == 0 &&
2542 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2543 &inode->runtime_flags))
2544 btrfs_del_delalloc_inode(root, inode);
2545 spin_unlock(&inode->lock);
2548 if ((state->state & EXTENT_DELALLOC_NEW) &&
2549 (bits & EXTENT_DELALLOC_NEW)) {
2550 spin_lock(&inode->lock);
2551 ASSERT(inode->new_delalloc_bytes >= len);
2552 inode->new_delalloc_bytes -= len;
2553 if (bits & EXTENT_ADD_INODE_BYTES)
2554 inode_add_bytes(&inode->vfs_inode, len);
2555 spin_unlock(&inode->lock);
2559 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2560 struct btrfs_ordered_extent *ordered)
2562 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2563 u64 len = bbio->bio.bi_iter.bi_size;
2564 struct btrfs_ordered_extent *new;
2567 /* Must always be called for the beginning of an ordered extent. */
2568 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2571 /* No need to split if the ordered extent covers the entire bio. */
2572 if (ordered->disk_num_bytes == len) {
2573 refcount_inc(&ordered->refs);
2574 bbio->ordered = ordered;
2579 * Don't split the extent_map for NOCOW extents, as we're writing into
2580 * a pre-existing one.
2582 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2583 ret = split_extent_map(bbio->inode, bbio->file_offset,
2584 ordered->num_bytes, len,
2585 ordered->disk_bytenr);
2590 new = btrfs_split_ordered_extent(ordered, len);
2592 return PTR_ERR(new);
2593 bbio->ordered = new;
2598 * given a list of ordered sums record them in the inode. This happens
2599 * at IO completion time based on sums calculated at bio submission time.
2601 static int add_pending_csums(struct btrfs_trans_handle *trans,
2602 struct list_head *list)
2604 struct btrfs_ordered_sum *sum;
2605 struct btrfs_root *csum_root = NULL;
2608 list_for_each_entry(sum, list, list) {
2609 trans->adding_csums = true;
2611 csum_root = btrfs_csum_root(trans->fs_info,
2613 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2614 trans->adding_csums = false;
2621 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2624 struct extent_state **cached_state)
2626 u64 search_start = start;
2627 const u64 end = start + len - 1;
2629 while (search_start < end) {
2630 const u64 search_len = end - search_start + 1;
2631 struct extent_map *em;
2635 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2639 if (em->block_start != EXTENT_MAP_HOLE)
2643 if (em->start < search_start)
2644 em_len -= search_start - em->start;
2645 if (em_len > search_len)
2646 em_len = search_len;
2648 ret = set_extent_bit(&inode->io_tree, search_start,
2649 search_start + em_len - 1,
2650 EXTENT_DELALLOC_NEW, cached_state);
2652 search_start = extent_map_end(em);
2653 free_extent_map(em);
2660 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2661 unsigned int extra_bits,
2662 struct extent_state **cached_state)
2664 WARN_ON(PAGE_ALIGNED(end));
2666 if (start >= i_size_read(&inode->vfs_inode) &&
2667 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2669 * There can't be any extents following eof in this case so just
2670 * set the delalloc new bit for the range directly.
2672 extra_bits |= EXTENT_DELALLOC_NEW;
2676 ret = btrfs_find_new_delalloc_bytes(inode, start,
2683 return set_extent_bit(&inode->io_tree, start, end,
2684 EXTENT_DELALLOC | extra_bits, cached_state);
2687 /* see btrfs_writepage_start_hook for details on why this is required */
2688 struct btrfs_writepage_fixup {
2690 struct btrfs_inode *inode;
2691 struct btrfs_work work;
2694 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2696 struct btrfs_writepage_fixup *fixup =
2697 container_of(work, struct btrfs_writepage_fixup, work);
2698 struct btrfs_ordered_extent *ordered;
2699 struct extent_state *cached_state = NULL;
2700 struct extent_changeset *data_reserved = NULL;
2701 struct page *page = fixup->page;
2702 struct btrfs_inode *inode = fixup->inode;
2703 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2704 u64 page_start = page_offset(page);
2705 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2707 bool free_delalloc_space = true;
2710 * This is similar to page_mkwrite, we need to reserve the space before
2711 * we take the page lock.
2713 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2719 * Before we queued this fixup, we took a reference on the page.
2720 * page->mapping may go NULL, but it shouldn't be moved to a different
2723 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2725 * Unfortunately this is a little tricky, either
2727 * 1) We got here and our page had already been dealt with and
2728 * we reserved our space, thus ret == 0, so we need to just
2729 * drop our space reservation and bail. This can happen the
2730 * first time we come into the fixup worker, or could happen
2731 * while waiting for the ordered extent.
2732 * 2) Our page was already dealt with, but we happened to get an
2733 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2734 * this case we obviously don't have anything to release, but
2735 * because the page was already dealt with we don't want to
2736 * mark the page with an error, so make sure we're resetting
2737 * ret to 0. This is why we have this check _before_ the ret
2738 * check, because we do not want to have a surprise ENOSPC
2739 * when the page was already properly dealt with.
2742 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2743 btrfs_delalloc_release_space(inode, data_reserved,
2744 page_start, PAGE_SIZE,
2752 * We can't mess with the page state unless it is locked, so now that
2753 * it is locked bail if we failed to make our space reservation.
2758 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2760 /* already ordered? We're done */
2761 if (PageOrdered(page))
2764 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2766 unlock_extent(&inode->io_tree, page_start, page_end,
2769 btrfs_start_ordered_extent(ordered);
2770 btrfs_put_ordered_extent(ordered);
2774 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2780 * Everything went as planned, we're now the owner of a dirty page with
2781 * delayed allocation bits set and space reserved for our COW
2784 * The page was dirty when we started, nothing should have cleaned it.
2786 BUG_ON(!PageDirty(page));
2787 free_delalloc_space = false;
2789 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2790 if (free_delalloc_space)
2791 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2793 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2797 * We hit ENOSPC or other errors. Update the mapping and page
2798 * to reflect the errors and clean the page.
2800 mapping_set_error(page->mapping, ret);
2801 btrfs_mark_ordered_io_finished(inode, page, page_start,
2803 clear_page_dirty_for_io(page);
2805 btrfs_folio_clear_checked(fs_info, page_folio(page), page_start, PAGE_SIZE);
2809 extent_changeset_free(data_reserved);
2811 * As a precaution, do a delayed iput in case it would be the last iput
2812 * that could need flushing space. Recursing back to fixup worker would
2815 btrfs_add_delayed_iput(inode);
2819 * There are a few paths in the higher layers of the kernel that directly
2820 * set the page dirty bit without asking the filesystem if it is a
2821 * good idea. This causes problems because we want to make sure COW
2822 * properly happens and the data=ordered rules are followed.
2824 * In our case any range that doesn't have the ORDERED bit set
2825 * hasn't been properly setup for IO. We kick off an async process
2826 * to fix it up. The async helper will wait for ordered extents, set
2827 * the delalloc bit and make it safe to write the page.
2829 int btrfs_writepage_cow_fixup(struct page *page)
2831 struct inode *inode = page->mapping->host;
2832 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2833 struct btrfs_writepage_fixup *fixup;
2835 /* This page has ordered extent covering it already */
2836 if (PageOrdered(page))
2840 * PageChecked is set below when we create a fixup worker for this page,
2841 * don't try to create another one if we're already PageChecked()
2843 * The extent_io writepage code will redirty the page if we send back
2846 if (PageChecked(page))
2849 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2854 * We are already holding a reference to this inode from
2855 * write_cache_pages. We need to hold it because the space reservation
2856 * takes place outside of the page lock, and we can't trust
2857 * page->mapping outside of the page lock.
2860 btrfs_folio_set_checked(fs_info, page_folio(page), page_offset(page), PAGE_SIZE);
2862 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL);
2864 fixup->inode = BTRFS_I(inode);
2865 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2870 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2871 struct btrfs_inode *inode, u64 file_pos,
2872 struct btrfs_file_extent_item *stack_fi,
2873 const bool update_inode_bytes,
2874 u64 qgroup_reserved)
2876 struct btrfs_root *root = inode->root;
2877 const u64 sectorsize = root->fs_info->sectorsize;
2878 struct btrfs_path *path;
2879 struct extent_buffer *leaf;
2880 struct btrfs_key ins;
2881 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2882 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2883 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2884 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2885 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2886 struct btrfs_drop_extents_args drop_args = { 0 };
2889 path = btrfs_alloc_path();
2894 * we may be replacing one extent in the tree with another.
2895 * The new extent is pinned in the extent map, and we don't want
2896 * to drop it from the cache until it is completely in the btree.
2898 * So, tell btrfs_drop_extents to leave this extent in the cache.
2899 * the caller is expected to unpin it and allow it to be merged
2902 drop_args.path = path;
2903 drop_args.start = file_pos;
2904 drop_args.end = file_pos + num_bytes;
2905 drop_args.replace_extent = true;
2906 drop_args.extent_item_size = sizeof(*stack_fi);
2907 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2911 if (!drop_args.extent_inserted) {
2912 ins.objectid = btrfs_ino(inode);
2913 ins.offset = file_pos;
2914 ins.type = BTRFS_EXTENT_DATA_KEY;
2916 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2921 leaf = path->nodes[0];
2922 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2923 write_extent_buffer(leaf, stack_fi,
2924 btrfs_item_ptr_offset(leaf, path->slots[0]),
2925 sizeof(struct btrfs_file_extent_item));
2927 btrfs_mark_buffer_dirty(trans, leaf);
2928 btrfs_release_path(path);
2931 * If we dropped an inline extent here, we know the range where it is
2932 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2933 * number of bytes only for that range containing the inline extent.
2934 * The remaining of the range will be processed when clearning the
2935 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2937 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2938 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2940 inline_size = drop_args.bytes_found - inline_size;
2941 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2942 drop_args.bytes_found -= inline_size;
2943 num_bytes -= sectorsize;
2946 if (update_inode_bytes)
2947 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2949 ins.objectid = disk_bytenr;
2950 ins.offset = disk_num_bytes;
2951 ins.type = BTRFS_EXTENT_ITEM_KEY;
2953 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2957 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2959 qgroup_reserved, &ins);
2961 btrfs_free_path(path);
2966 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2969 struct btrfs_block_group *cache;
2971 cache = btrfs_lookup_block_group(fs_info, start);
2974 spin_lock(&cache->lock);
2975 cache->delalloc_bytes -= len;
2976 spin_unlock(&cache->lock);
2978 btrfs_put_block_group(cache);
2981 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2982 struct btrfs_ordered_extent *oe)
2984 struct btrfs_file_extent_item stack_fi;
2985 bool update_inode_bytes;
2986 u64 num_bytes = oe->num_bytes;
2987 u64 ram_bytes = oe->ram_bytes;
2989 memset(&stack_fi, 0, sizeof(stack_fi));
2990 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2991 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2992 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2993 oe->disk_num_bytes);
2994 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2995 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
2996 num_bytes = oe->truncated_len;
2997 ram_bytes = num_bytes;
2999 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3000 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3001 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3002 /* Encryption and other encoding is reserved and all 0 */
3005 * For delalloc, when completing an ordered extent we update the inode's
3006 * bytes when clearing the range in the inode's io tree, so pass false
3007 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3008 * except if the ordered extent was truncated.
3010 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3011 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3012 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3014 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3015 oe->file_offset, &stack_fi,
3016 update_inode_bytes, oe->qgroup_rsv);
3020 * As ordered data IO finishes, this gets called so we can finish
3021 * an ordered extent if the range of bytes in the file it covers are
3024 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3026 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3027 struct btrfs_root *root = inode->root;
3028 struct btrfs_fs_info *fs_info = root->fs_info;
3029 struct btrfs_trans_handle *trans = NULL;
3030 struct extent_io_tree *io_tree = &inode->io_tree;
3031 struct extent_state *cached_state = NULL;
3033 int compress_type = 0;
3035 u64 logical_len = ordered_extent->num_bytes;
3036 bool freespace_inode;
3037 bool truncated = false;
3038 bool clear_reserved_extent = true;
3039 unsigned int clear_bits = EXTENT_DEFRAG;
3041 start = ordered_extent->file_offset;
3042 end = start + ordered_extent->num_bytes - 1;
3044 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3045 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3046 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3047 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3048 clear_bits |= EXTENT_DELALLOC_NEW;
3050 freespace_inode = btrfs_is_free_space_inode(inode);
3051 if (!freespace_inode)
3052 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3054 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3059 if (btrfs_is_zoned(fs_info))
3060 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3061 ordered_extent->disk_num_bytes);
3063 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3065 logical_len = ordered_extent->truncated_len;
3066 /* Truncated the entire extent, don't bother adding */
3071 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3072 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3074 btrfs_inode_safe_disk_i_size_write(inode, 0);
3075 if (freespace_inode)
3076 trans = btrfs_join_transaction_spacecache(root);
3078 trans = btrfs_join_transaction(root);
3079 if (IS_ERR(trans)) {
3080 ret = PTR_ERR(trans);
3084 trans->block_rsv = &inode->block_rsv;
3085 ret = btrfs_update_inode_fallback(trans, inode);
3086 if (ret) /* -ENOMEM or corruption */
3087 btrfs_abort_transaction(trans, ret);
3091 clear_bits |= EXTENT_LOCKED;
3092 lock_extent(io_tree, start, end, &cached_state);
3094 if (freespace_inode)
3095 trans = btrfs_join_transaction_spacecache(root);
3097 trans = btrfs_join_transaction(root);
3098 if (IS_ERR(trans)) {
3099 ret = PTR_ERR(trans);
3104 trans->block_rsv = &inode->block_rsv;
3106 ret = btrfs_insert_raid_extent(trans, ordered_extent);
3110 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3111 compress_type = ordered_extent->compress_type;
3112 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3113 BUG_ON(compress_type);
3114 ret = btrfs_mark_extent_written(trans, inode,
3115 ordered_extent->file_offset,
3116 ordered_extent->file_offset +
3118 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3119 ordered_extent->disk_num_bytes);
3121 BUG_ON(root == fs_info->tree_root);
3122 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3124 clear_reserved_extent = false;
3125 btrfs_release_delalloc_bytes(fs_info,
3126 ordered_extent->disk_bytenr,
3127 ordered_extent->disk_num_bytes);
3130 unpin_extent_cache(inode, ordered_extent->file_offset,
3131 ordered_extent->num_bytes, trans->transid);
3133 btrfs_abort_transaction(trans, ret);
3137 ret = add_pending_csums(trans, &ordered_extent->list);
3139 btrfs_abort_transaction(trans, ret);
3144 * If this is a new delalloc range, clear its new delalloc flag to
3145 * update the inode's number of bytes. This needs to be done first
3146 * before updating the inode item.
3148 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3149 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3150 clear_extent_bit(&inode->io_tree, start, end,
3151 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3154 btrfs_inode_safe_disk_i_size_write(inode, 0);
3155 ret = btrfs_update_inode_fallback(trans, inode);
3156 if (ret) { /* -ENOMEM or corruption */
3157 btrfs_abort_transaction(trans, ret);
3162 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3166 btrfs_end_transaction(trans);
3168 if (ret || truncated) {
3169 u64 unwritten_start = start;
3172 * If we failed to finish this ordered extent for any reason we
3173 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3174 * extent, and mark the inode with the error if it wasn't
3175 * already set. Any error during writeback would have already
3176 * set the mapping error, so we need to set it if we're the ones
3177 * marking this ordered extent as failed.
3179 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3180 &ordered_extent->flags))
3181 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3184 unwritten_start += logical_len;
3185 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3187 /* Drop extent maps for the part of the extent we didn't write. */
3188 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3191 * If the ordered extent had an IOERR or something else went
3192 * wrong we need to return the space for this ordered extent
3193 * back to the allocator. We only free the extent in the
3194 * truncated case if we didn't write out the extent at all.
3196 * If we made it past insert_reserved_file_extent before we
3197 * errored out then we don't need to do this as the accounting
3198 * has already been done.
3200 if ((ret || !logical_len) &&
3201 clear_reserved_extent &&
3202 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3203 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3205 * Discard the range before returning it back to the
3208 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3209 btrfs_discard_extent(fs_info,
3210 ordered_extent->disk_bytenr,
3211 ordered_extent->disk_num_bytes,
3213 btrfs_free_reserved_extent(fs_info,
3214 ordered_extent->disk_bytenr,
3215 ordered_extent->disk_num_bytes, 1);
3217 * Actually free the qgroup rsv which was released when
3218 * the ordered extent was created.
3220 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3221 ordered_extent->qgroup_rsv,
3222 BTRFS_QGROUP_RSV_DATA);
3227 * This needs to be done to make sure anybody waiting knows we are done
3228 * updating everything for this ordered extent.
3230 btrfs_remove_ordered_extent(inode, ordered_extent);
3233 btrfs_put_ordered_extent(ordered_extent);
3234 /* once for the tree */
3235 btrfs_put_ordered_extent(ordered_extent);
3240 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3242 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3243 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3244 list_empty(&ordered->bioc_list))
3245 btrfs_finish_ordered_zoned(ordered);
3246 return btrfs_finish_one_ordered(ordered);
3250 * Verify the checksum for a single sector without any extra action that depend
3251 * on the type of I/O.
3253 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3254 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3256 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3259 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3261 shash->tfm = fs_info->csum_shash;
3263 kaddr = kmap_local_page(page) + pgoff;
3264 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3265 kunmap_local(kaddr);
3267 if (memcmp(csum, csum_expected, fs_info->csum_size))
3273 * Verify the checksum of a single data sector.
3275 * @bbio: btrfs_io_bio which contains the csum
3276 * @dev: device the sector is on
3277 * @bio_offset: offset to the beginning of the bio (in bytes)
3278 * @bv: bio_vec to check
3280 * Check if the checksum on a data block is valid. When a checksum mismatch is
3281 * detected, report the error and fill the corrupted range with zero.
3283 * Return %true if the sector is ok or had no checksum to start with, else %false.
3285 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3286 u32 bio_offset, struct bio_vec *bv)
3288 struct btrfs_inode *inode = bbio->inode;
3289 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3290 u64 file_offset = bbio->file_offset + bio_offset;
3291 u64 end = file_offset + bv->bv_len - 1;
3293 u8 csum[BTRFS_CSUM_SIZE];
3295 ASSERT(bv->bv_len == fs_info->sectorsize);
3300 if (btrfs_is_data_reloc_root(inode->root) &&
3301 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3303 /* Skip the range without csum for data reloc inode */
3304 clear_extent_bits(&inode->io_tree, file_offset, end,
3309 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3311 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3317 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3320 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3326 * Perform a delayed iput on @inode.
3328 * @inode: The inode we want to perform iput on
3330 * This function uses the generic vfs_inode::i_count to track whether we should
3331 * just decrement it (in case it's > 1) or if this is the last iput then link
3332 * the inode to the delayed iput machinery. Delayed iputs are processed at
3333 * transaction commit time/superblock commit/cleaner kthread.
3335 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3337 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3338 unsigned long flags;
3340 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3343 atomic_inc(&fs_info->nr_delayed_iputs);
3345 * Need to be irq safe here because we can be called from either an irq
3346 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3349 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3350 ASSERT(list_empty(&inode->delayed_iput));
3351 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3352 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3353 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3354 wake_up_process(fs_info->cleaner_kthread);
3357 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3358 struct btrfs_inode *inode)
3360 list_del_init(&inode->delayed_iput);
3361 spin_unlock_irq(&fs_info->delayed_iput_lock);
3362 iput(&inode->vfs_inode);
3363 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3364 wake_up(&fs_info->delayed_iputs_wait);
3365 spin_lock_irq(&fs_info->delayed_iput_lock);
3368 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3369 struct btrfs_inode *inode)
3371 if (!list_empty(&inode->delayed_iput)) {
3372 spin_lock_irq(&fs_info->delayed_iput_lock);
3373 if (!list_empty(&inode->delayed_iput))
3374 run_delayed_iput_locked(fs_info, inode);
3375 spin_unlock_irq(&fs_info->delayed_iput_lock);
3379 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3382 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3383 * calls btrfs_add_delayed_iput() and that needs to lock
3384 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3385 * prevent a deadlock.
3387 spin_lock_irq(&fs_info->delayed_iput_lock);
3388 while (!list_empty(&fs_info->delayed_iputs)) {
3389 struct btrfs_inode *inode;
3391 inode = list_first_entry(&fs_info->delayed_iputs,
3392 struct btrfs_inode, delayed_iput);
3393 run_delayed_iput_locked(fs_info, inode);
3394 if (need_resched()) {
3395 spin_unlock_irq(&fs_info->delayed_iput_lock);
3397 spin_lock_irq(&fs_info->delayed_iput_lock);
3400 spin_unlock_irq(&fs_info->delayed_iput_lock);
3404 * Wait for flushing all delayed iputs
3406 * @fs_info: the filesystem
3408 * This will wait on any delayed iputs that are currently running with KILLABLE
3409 * set. Once they are all done running we will return, unless we are killed in
3410 * which case we return EINTR. This helps in user operations like fallocate etc
3411 * that might get blocked on the iputs.
3413 * Return EINTR if we were killed, 0 if nothing's pending
3415 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3417 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3418 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3425 * This creates an orphan entry for the given inode in case something goes wrong
3426 * in the middle of an unlink.
3428 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3429 struct btrfs_inode *inode)
3433 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3434 if (ret && ret != -EEXIST) {
3435 btrfs_abort_transaction(trans, ret);
3443 * We have done the delete so we can go ahead and remove the orphan item for
3444 * this particular inode.
3446 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3447 struct btrfs_inode *inode)
3449 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3453 * this cleans up any orphans that may be left on the list from the last use
3456 int btrfs_orphan_cleanup(struct btrfs_root *root)
3458 struct btrfs_fs_info *fs_info = root->fs_info;
3459 struct btrfs_path *path;
3460 struct extent_buffer *leaf;
3461 struct btrfs_key key, found_key;
3462 struct btrfs_trans_handle *trans;
3463 struct inode *inode;
3464 u64 last_objectid = 0;
3465 int ret = 0, nr_unlink = 0;
3467 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3470 path = btrfs_alloc_path();
3475 path->reada = READA_BACK;
3477 key.objectid = BTRFS_ORPHAN_OBJECTID;
3478 key.type = BTRFS_ORPHAN_ITEM_KEY;
3479 key.offset = (u64)-1;
3482 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3487 * if ret == 0 means we found what we were searching for, which
3488 * is weird, but possible, so only screw with path if we didn't
3489 * find the key and see if we have stuff that matches
3493 if (path->slots[0] == 0)
3498 /* pull out the item */
3499 leaf = path->nodes[0];
3500 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3502 /* make sure the item matches what we want */
3503 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3505 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3508 /* release the path since we're done with it */
3509 btrfs_release_path(path);
3512 * this is where we are basically btrfs_lookup, without the
3513 * crossing root thing. we store the inode number in the
3514 * offset of the orphan item.
3517 if (found_key.offset == last_objectid) {
3519 * We found the same inode as before. This means we were
3520 * not able to remove its items via eviction triggered
3521 * by an iput(). A transaction abort may have happened,
3522 * due to -ENOSPC for example, so try to grab the error
3523 * that lead to a transaction abort, if any.
3526 "Error removing orphan entry, stopping orphan cleanup");
3527 ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL;
3531 last_objectid = found_key.offset;
3533 found_key.objectid = found_key.offset;
3534 found_key.type = BTRFS_INODE_ITEM_KEY;
3535 found_key.offset = 0;
3536 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3537 if (IS_ERR(inode)) {
3538 ret = PTR_ERR(inode);
3544 if (!inode && root == fs_info->tree_root) {
3545 struct btrfs_root *dead_root;
3546 int is_dead_root = 0;
3549 * This is an orphan in the tree root. Currently these
3550 * could come from 2 sources:
3551 * a) a root (snapshot/subvolume) deletion in progress
3552 * b) a free space cache inode
3553 * We need to distinguish those two, as the orphan item
3554 * for a root must not get deleted before the deletion
3555 * of the snapshot/subvolume's tree completes.
3557 * btrfs_find_orphan_roots() ran before us, which has
3558 * found all deleted roots and loaded them into
3559 * fs_info->fs_roots_radix. So here we can find if an
3560 * orphan item corresponds to a deleted root by looking
3561 * up the root from that radix tree.
3564 spin_lock(&fs_info->fs_roots_radix_lock);
3565 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3566 (unsigned long)found_key.objectid);
3567 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3569 spin_unlock(&fs_info->fs_roots_radix_lock);
3572 /* prevent this orphan from being found again */
3573 key.offset = found_key.objectid - 1;
3580 * If we have an inode with links, there are a couple of
3583 * 1. We were halfway through creating fsverity metadata for the
3584 * file. In that case, the orphan item represents incomplete
3585 * fsverity metadata which must be cleaned up with
3586 * btrfs_drop_verity_items and deleting the orphan item.
3588 * 2. Old kernels (before v3.12) used to create an
3589 * orphan item for truncate indicating that there were possibly
3590 * extent items past i_size that needed to be deleted. In v3.12,
3591 * truncate was changed to update i_size in sync with the extent
3592 * items, but the (useless) orphan item was still created. Since
3593 * v4.18, we don't create the orphan item for truncate at all.
3595 * So, this item could mean that we need to do a truncate, but
3596 * only if this filesystem was last used on a pre-v3.12 kernel
3597 * and was not cleanly unmounted. The odds of that are quite
3598 * slim, and it's a pain to do the truncate now, so just delete
3601 * It's also possible that this orphan item was supposed to be
3602 * deleted but wasn't. The inode number may have been reused,
3603 * but either way, we can delete the orphan item.
3605 if (!inode || inode->i_nlink) {
3607 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3613 trans = btrfs_start_transaction(root, 1);
3614 if (IS_ERR(trans)) {
3615 ret = PTR_ERR(trans);
3618 btrfs_debug(fs_info, "auto deleting %Lu",
3619 found_key.objectid);
3620 ret = btrfs_del_orphan_item(trans, root,
3621 found_key.objectid);
3622 btrfs_end_transaction(trans);
3630 /* this will do delete_inode and everything for us */
3633 /* release the path since we're done with it */
3634 btrfs_release_path(path);
3636 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3637 trans = btrfs_join_transaction(root);
3639 btrfs_end_transaction(trans);
3643 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3647 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3648 btrfs_free_path(path);
3653 * very simple check to peek ahead in the leaf looking for xattrs. If we
3654 * don't find any xattrs, we know there can't be any acls.
3656 * slot is the slot the inode is in, objectid is the objectid of the inode
3658 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3659 int slot, u64 objectid,
3660 int *first_xattr_slot)
3662 u32 nritems = btrfs_header_nritems(leaf);
3663 struct btrfs_key found_key;
3664 static u64 xattr_access = 0;
3665 static u64 xattr_default = 0;
3668 if (!xattr_access) {
3669 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3670 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3671 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3672 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3676 *first_xattr_slot = -1;
3677 while (slot < nritems) {
3678 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3680 /* we found a different objectid, there must not be acls */
3681 if (found_key.objectid != objectid)
3684 /* we found an xattr, assume we've got an acl */
3685 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3686 if (*first_xattr_slot == -1)
3687 *first_xattr_slot = slot;
3688 if (found_key.offset == xattr_access ||
3689 found_key.offset == xattr_default)
3694 * we found a key greater than an xattr key, there can't
3695 * be any acls later on
3697 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3704 * it goes inode, inode backrefs, xattrs, extents,
3705 * so if there are a ton of hard links to an inode there can
3706 * be a lot of backrefs. Don't waste time searching too hard,
3707 * this is just an optimization
3712 /* we hit the end of the leaf before we found an xattr or
3713 * something larger than an xattr. We have to assume the inode
3716 if (*first_xattr_slot == -1)
3717 *first_xattr_slot = slot;
3722 * read an inode from the btree into the in-memory inode
3724 static int btrfs_read_locked_inode(struct inode *inode,
3725 struct btrfs_path *in_path)
3727 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3728 struct btrfs_path *path = in_path;
3729 struct extent_buffer *leaf;
3730 struct btrfs_inode_item *inode_item;
3731 struct btrfs_root *root = BTRFS_I(inode)->root;
3732 struct btrfs_key location;
3737 bool filled = false;
3738 int first_xattr_slot;
3740 ret = btrfs_fill_inode(inode, &rdev);
3745 path = btrfs_alloc_path();
3750 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3752 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3754 if (path != in_path)
3755 btrfs_free_path(path);
3759 leaf = path->nodes[0];
3764 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3765 struct btrfs_inode_item);
3766 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3767 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3768 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3769 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3770 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3771 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3772 round_up(i_size_read(inode), fs_info->sectorsize));
3774 inode_set_atime(inode, btrfs_timespec_sec(leaf, &inode_item->atime),
3775 btrfs_timespec_nsec(leaf, &inode_item->atime));
3777 inode_set_mtime(inode, btrfs_timespec_sec(leaf, &inode_item->mtime),
3778 btrfs_timespec_nsec(leaf, &inode_item->mtime));
3780 inode_set_ctime(inode, btrfs_timespec_sec(leaf, &inode_item->ctime),
3781 btrfs_timespec_nsec(leaf, &inode_item->ctime));
3783 BTRFS_I(inode)->i_otime_sec = btrfs_timespec_sec(leaf, &inode_item->otime);
3784 BTRFS_I(inode)->i_otime_nsec = btrfs_timespec_nsec(leaf, &inode_item->otime);
3786 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3787 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3788 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3790 inode_set_iversion_queried(inode,
3791 btrfs_inode_sequence(leaf, inode_item));
3792 inode->i_generation = BTRFS_I(inode)->generation;
3794 rdev = btrfs_inode_rdev(leaf, inode_item);
3796 BTRFS_I(inode)->index_cnt = (u64)-1;
3797 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3798 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3802 * If we were modified in the current generation and evicted from memory
3803 * and then re-read we need to do a full sync since we don't have any
3804 * idea about which extents were modified before we were evicted from
3807 * This is required for both inode re-read from disk and delayed inode
3808 * in the delayed_nodes xarray.
3810 if (BTRFS_I(inode)->last_trans == btrfs_get_fs_generation(fs_info))
3811 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3812 &BTRFS_I(inode)->runtime_flags);
3815 * We don't persist the id of the transaction where an unlink operation
3816 * against the inode was last made. So here we assume the inode might
3817 * have been evicted, and therefore the exact value of last_unlink_trans
3818 * lost, and set it to last_trans to avoid metadata inconsistencies
3819 * between the inode and its parent if the inode is fsync'ed and the log
3820 * replayed. For example, in the scenario:
3823 * ln mydir/foo mydir/bar
3826 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3827 * xfs_io -c fsync mydir/foo
3829 * mount fs, triggers fsync log replay
3831 * We must make sure that when we fsync our inode foo we also log its
3832 * parent inode, otherwise after log replay the parent still has the
3833 * dentry with the "bar" name but our inode foo has a link count of 1
3834 * and doesn't have an inode ref with the name "bar" anymore.
3836 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3837 * but it guarantees correctness at the expense of occasional full
3838 * transaction commits on fsync if our inode is a directory, or if our
3839 * inode is not a directory, logging its parent unnecessarily.
3841 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3844 * Same logic as for last_unlink_trans. We don't persist the generation
3845 * of the last transaction where this inode was used for a reflink
3846 * operation, so after eviction and reloading the inode we must be
3847 * pessimistic and assume the last transaction that modified the inode.
3849 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3852 if (inode->i_nlink != 1 ||
3853 path->slots[0] >= btrfs_header_nritems(leaf))
3856 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3857 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3860 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3861 if (location.type == BTRFS_INODE_REF_KEY) {
3862 struct btrfs_inode_ref *ref;
3864 ref = (struct btrfs_inode_ref *)ptr;
3865 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3866 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3867 struct btrfs_inode_extref *extref;
3869 extref = (struct btrfs_inode_extref *)ptr;
3870 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3875 * try to precache a NULL acl entry for files that don't have
3876 * any xattrs or acls
3878 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3879 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3880 if (first_xattr_slot != -1) {
3881 path->slots[0] = first_xattr_slot;
3882 ret = btrfs_load_inode_props(inode, path);
3885 "error loading props for ino %llu (root %llu): %d",
3886 btrfs_ino(BTRFS_I(inode)),
3887 root->root_key.objectid, ret);
3889 if (path != in_path)
3890 btrfs_free_path(path);
3893 cache_no_acl(inode);
3895 switch (inode->i_mode & S_IFMT) {
3897 inode->i_mapping->a_ops = &btrfs_aops;
3898 inode->i_fop = &btrfs_file_operations;
3899 inode->i_op = &btrfs_file_inode_operations;
3902 inode->i_fop = &btrfs_dir_file_operations;
3903 inode->i_op = &btrfs_dir_inode_operations;
3906 inode->i_op = &btrfs_symlink_inode_operations;
3907 inode_nohighmem(inode);
3908 inode->i_mapping->a_ops = &btrfs_aops;
3911 inode->i_op = &btrfs_special_inode_operations;
3912 init_special_inode(inode, inode->i_mode, rdev);
3916 btrfs_sync_inode_flags_to_i_flags(inode);
3921 * given a leaf and an inode, copy the inode fields into the leaf
3923 static void fill_inode_item(struct btrfs_trans_handle *trans,
3924 struct extent_buffer *leaf,
3925 struct btrfs_inode_item *item,
3926 struct inode *inode)
3928 struct btrfs_map_token token;
3931 btrfs_init_map_token(&token, leaf);
3933 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3934 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3935 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3936 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3937 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3939 btrfs_set_token_timespec_sec(&token, &item->atime,
3940 inode_get_atime_sec(inode));
3941 btrfs_set_token_timespec_nsec(&token, &item->atime,
3942 inode_get_atime_nsec(inode));
3944 btrfs_set_token_timespec_sec(&token, &item->mtime,
3945 inode_get_mtime_sec(inode));
3946 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3947 inode_get_mtime_nsec(inode));
3949 btrfs_set_token_timespec_sec(&token, &item->ctime,
3950 inode_get_ctime_sec(inode));
3951 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3952 inode_get_ctime_nsec(inode));
3954 btrfs_set_token_timespec_sec(&token, &item->otime, BTRFS_I(inode)->i_otime_sec);
3955 btrfs_set_token_timespec_nsec(&token, &item->otime, BTRFS_I(inode)->i_otime_nsec);
3957 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3958 btrfs_set_token_inode_generation(&token, item,
3959 BTRFS_I(inode)->generation);
3960 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3961 btrfs_set_token_inode_transid(&token, item, trans->transid);
3962 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3963 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3964 BTRFS_I(inode)->ro_flags);
3965 btrfs_set_token_inode_flags(&token, item, flags);
3966 btrfs_set_token_inode_block_group(&token, item, 0);
3970 * copy everything in the in-memory inode into the btree.
3972 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3973 struct btrfs_inode *inode)
3975 struct btrfs_inode_item *inode_item;
3976 struct btrfs_path *path;
3977 struct extent_buffer *leaf;
3980 path = btrfs_alloc_path();
3984 ret = btrfs_lookup_inode(trans, inode->root, path, &inode->location, 1);
3991 leaf = path->nodes[0];
3992 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3993 struct btrfs_inode_item);
3995 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3996 btrfs_mark_buffer_dirty(trans, leaf);
3997 btrfs_set_inode_last_trans(trans, inode);
4000 btrfs_free_path(path);
4005 * copy everything in the in-memory inode into the btree.
4007 int btrfs_update_inode(struct btrfs_trans_handle *trans,
4008 struct btrfs_inode *inode)
4010 struct btrfs_root *root = inode->root;
4011 struct btrfs_fs_info *fs_info = root->fs_info;
4015 * If the inode is a free space inode, we can deadlock during commit
4016 * if we put it into the delayed code.
4018 * The data relocation inode should also be directly updated
4021 if (!btrfs_is_free_space_inode(inode)
4022 && !btrfs_is_data_reloc_root(root)
4023 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4024 btrfs_update_root_times(trans, root);
4026 ret = btrfs_delayed_update_inode(trans, inode);
4028 btrfs_set_inode_last_trans(trans, inode);
4032 return btrfs_update_inode_item(trans, inode);
4035 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4036 struct btrfs_inode *inode)
4040 ret = btrfs_update_inode(trans, inode);
4042 return btrfs_update_inode_item(trans, inode);
4047 * unlink helper that gets used here in inode.c and in the tree logging
4048 * recovery code. It remove a link in a directory with a given name, and
4049 * also drops the back refs in the inode to the directory
4051 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4052 struct btrfs_inode *dir,
4053 struct btrfs_inode *inode,
4054 const struct fscrypt_str *name,
4055 struct btrfs_rename_ctx *rename_ctx)
4057 struct btrfs_root *root = dir->root;
4058 struct btrfs_fs_info *fs_info = root->fs_info;
4059 struct btrfs_path *path;
4061 struct btrfs_dir_item *di;
4063 u64 ino = btrfs_ino(inode);
4064 u64 dir_ino = btrfs_ino(dir);
4066 path = btrfs_alloc_path();
4072 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4073 if (IS_ERR_OR_NULL(di)) {
4074 ret = di ? PTR_ERR(di) : -ENOENT;
4077 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4080 btrfs_release_path(path);
4083 * If we don't have dir index, we have to get it by looking up
4084 * the inode ref, since we get the inode ref, remove it directly,
4085 * it is unnecessary to do delayed deletion.
4087 * But if we have dir index, needn't search inode ref to get it.
4088 * Since the inode ref is close to the inode item, it is better
4089 * that we delay to delete it, and just do this deletion when
4090 * we update the inode item.
4092 if (inode->dir_index) {
4093 ret = btrfs_delayed_delete_inode_ref(inode);
4095 index = inode->dir_index;
4100 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4103 "failed to delete reference to %.*s, inode %llu parent %llu",
4104 name->len, name->name, ino, dir_ino);
4105 btrfs_abort_transaction(trans, ret);
4110 rename_ctx->index = index;
4112 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4114 btrfs_abort_transaction(trans, ret);
4119 * If we are in a rename context, we don't need to update anything in the
4120 * log. That will be done later during the rename by btrfs_log_new_name().
4121 * Besides that, doing it here would only cause extra unnecessary btree
4122 * operations on the log tree, increasing latency for applications.
4125 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4126 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4130 * If we have a pending delayed iput we could end up with the final iput
4131 * being run in btrfs-cleaner context. If we have enough of these built
4132 * up we can end up burning a lot of time in btrfs-cleaner without any
4133 * way to throttle the unlinks. Since we're currently holding a ref on
4134 * the inode we can run the delayed iput here without any issues as the
4135 * final iput won't be done until after we drop the ref we're currently
4138 btrfs_run_delayed_iput(fs_info, inode);
4140 btrfs_free_path(path);
4144 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4145 inode_inc_iversion(&inode->vfs_inode);
4146 inode_inc_iversion(&dir->vfs_inode);
4147 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4148 ret = btrfs_update_inode(trans, dir);
4153 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4154 struct btrfs_inode *dir, struct btrfs_inode *inode,
4155 const struct fscrypt_str *name)
4159 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4161 drop_nlink(&inode->vfs_inode);
4162 ret = btrfs_update_inode(trans, inode);
4168 * helper to start transaction for unlink and rmdir.
4170 * unlink and rmdir are special in btrfs, they do not always free space, so
4171 * if we cannot make our reservations the normal way try and see if there is
4172 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4173 * allow the unlink to occur.
4175 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4177 struct btrfs_root *root = dir->root;
4179 return btrfs_start_transaction_fallback_global_rsv(root,
4180 BTRFS_UNLINK_METADATA_UNITS);
4183 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4185 struct btrfs_trans_handle *trans;
4186 struct inode *inode = d_inode(dentry);
4188 struct fscrypt_name fname;
4190 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4194 /* This needs to handle no-key deletions later on */
4196 trans = __unlink_start_trans(BTRFS_I(dir));
4197 if (IS_ERR(trans)) {
4198 ret = PTR_ERR(trans);
4202 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4205 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4210 if (inode->i_nlink == 0) {
4211 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4217 btrfs_end_transaction(trans);
4218 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4220 fscrypt_free_filename(&fname);
4224 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4225 struct btrfs_inode *dir, struct dentry *dentry)
4227 struct btrfs_root *root = dir->root;
4228 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4229 struct btrfs_path *path;
4230 struct extent_buffer *leaf;
4231 struct btrfs_dir_item *di;
4232 struct btrfs_key key;
4236 u64 dir_ino = btrfs_ino(dir);
4237 struct fscrypt_name fname;
4239 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4243 /* This needs to handle no-key deletions later on */
4245 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4246 objectid = inode->root->root_key.objectid;
4247 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4248 objectid = inode->location.objectid;
4251 fscrypt_free_filename(&fname);
4255 path = btrfs_alloc_path();
4261 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4262 &fname.disk_name, -1);
4263 if (IS_ERR_OR_NULL(di)) {
4264 ret = di ? PTR_ERR(di) : -ENOENT;
4268 leaf = path->nodes[0];
4269 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4270 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4271 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4273 btrfs_abort_transaction(trans, ret);
4276 btrfs_release_path(path);
4279 * This is a placeholder inode for a subvolume we didn't have a
4280 * reference to at the time of the snapshot creation. In the meantime
4281 * we could have renamed the real subvol link into our snapshot, so
4282 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4283 * Instead simply lookup the dir_index_item for this entry so we can
4284 * remove it. Otherwise we know we have a ref to the root and we can
4285 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4287 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4288 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4289 if (IS_ERR_OR_NULL(di)) {
4294 btrfs_abort_transaction(trans, ret);
4298 leaf = path->nodes[0];
4299 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4301 btrfs_release_path(path);
4303 ret = btrfs_del_root_ref(trans, objectid,
4304 root->root_key.objectid, dir_ino,
4305 &index, &fname.disk_name);
4307 btrfs_abort_transaction(trans, ret);
4312 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4314 btrfs_abort_transaction(trans, ret);
4318 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4319 inode_inc_iversion(&dir->vfs_inode);
4320 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4321 ret = btrfs_update_inode_fallback(trans, dir);
4323 btrfs_abort_transaction(trans, ret);
4325 btrfs_free_path(path);
4326 fscrypt_free_filename(&fname);
4331 * Helper to check if the subvolume references other subvolumes or if it's
4334 static noinline int may_destroy_subvol(struct btrfs_root *root)
4336 struct btrfs_fs_info *fs_info = root->fs_info;
4337 struct btrfs_path *path;
4338 struct btrfs_dir_item *di;
4339 struct btrfs_key key;
4340 struct fscrypt_str name = FSTR_INIT("default", 7);
4344 path = btrfs_alloc_path();
4348 /* Make sure this root isn't set as the default subvol */
4349 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4350 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4352 if (di && !IS_ERR(di)) {
4353 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4354 if (key.objectid == root->root_key.objectid) {
4357 "deleting default subvolume %llu is not allowed",
4361 btrfs_release_path(path);
4364 key.objectid = root->root_key.objectid;
4365 key.type = BTRFS_ROOT_REF_KEY;
4366 key.offset = (u64)-1;
4368 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4374 if (path->slots[0] > 0) {
4376 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4377 if (key.objectid == root->root_key.objectid &&
4378 key.type == BTRFS_ROOT_REF_KEY)
4382 btrfs_free_path(path);
4386 /* Delete all dentries for inodes belonging to the root */
4387 static void btrfs_prune_dentries(struct btrfs_root *root)
4389 struct btrfs_fs_info *fs_info = root->fs_info;
4390 struct rb_node *node;
4391 struct rb_node *prev;
4392 struct btrfs_inode *entry;
4393 struct inode *inode;
4396 if (!BTRFS_FS_ERROR(fs_info))
4397 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4399 spin_lock(&root->inode_lock);
4401 node = root->inode_tree.rb_node;
4405 entry = rb_entry(node, struct btrfs_inode, rb_node);
4407 if (objectid < btrfs_ino(entry))
4408 node = node->rb_left;
4409 else if (objectid > btrfs_ino(entry))
4410 node = node->rb_right;
4416 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4417 if (objectid <= btrfs_ino(entry)) {
4421 prev = rb_next(prev);
4425 entry = rb_entry(node, struct btrfs_inode, rb_node);
4426 objectid = btrfs_ino(entry) + 1;
4427 inode = igrab(&entry->vfs_inode);
4429 spin_unlock(&root->inode_lock);
4430 if (atomic_read(&inode->i_count) > 1)
4431 d_prune_aliases(inode);
4433 * btrfs_drop_inode will have it removed from the inode
4434 * cache when its usage count hits zero.
4438 spin_lock(&root->inode_lock);
4442 if (cond_resched_lock(&root->inode_lock))
4445 node = rb_next(node);
4447 spin_unlock(&root->inode_lock);
4450 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4452 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4453 struct btrfs_root *root = dir->root;
4454 struct inode *inode = d_inode(dentry);
4455 struct btrfs_root *dest = BTRFS_I(inode)->root;
4456 struct btrfs_trans_handle *trans;
4457 struct btrfs_block_rsv block_rsv;
4462 * Don't allow to delete a subvolume with send in progress. This is
4463 * inside the inode lock so the error handling that has to drop the bit
4464 * again is not run concurrently.
4466 spin_lock(&dest->root_item_lock);
4467 if (dest->send_in_progress) {
4468 spin_unlock(&dest->root_item_lock);
4470 "attempt to delete subvolume %llu during send",
4471 dest->root_key.objectid);
4474 if (atomic_read(&dest->nr_swapfiles)) {
4475 spin_unlock(&dest->root_item_lock);
4477 "attempt to delete subvolume %llu with active swapfile",
4478 root->root_key.objectid);
4481 root_flags = btrfs_root_flags(&dest->root_item);
4482 btrfs_set_root_flags(&dest->root_item,
4483 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4484 spin_unlock(&dest->root_item_lock);
4486 down_write(&fs_info->subvol_sem);
4488 ret = may_destroy_subvol(dest);
4492 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4494 * One for dir inode,
4495 * two for dir entries,
4496 * two for root ref/backref.
4498 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4502 trans = btrfs_start_transaction(root, 0);
4503 if (IS_ERR(trans)) {
4504 ret = PTR_ERR(trans);
4507 trans->block_rsv = &block_rsv;
4508 trans->bytes_reserved = block_rsv.size;
4510 btrfs_record_snapshot_destroy(trans, dir);
4512 ret = btrfs_unlink_subvol(trans, dir, dentry);
4514 btrfs_abort_transaction(trans, ret);
4518 ret = btrfs_record_root_in_trans(trans, dest);
4520 btrfs_abort_transaction(trans, ret);
4524 memset(&dest->root_item.drop_progress, 0,
4525 sizeof(dest->root_item.drop_progress));
4526 btrfs_set_root_drop_level(&dest->root_item, 0);
4527 btrfs_set_root_refs(&dest->root_item, 0);
4529 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4530 ret = btrfs_insert_orphan_item(trans,
4532 dest->root_key.objectid);
4534 btrfs_abort_transaction(trans, ret);
4539 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4540 BTRFS_UUID_KEY_SUBVOL,
4541 dest->root_key.objectid);
4542 if (ret && ret != -ENOENT) {
4543 btrfs_abort_transaction(trans, ret);
4546 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4547 ret = btrfs_uuid_tree_remove(trans,
4548 dest->root_item.received_uuid,
4549 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4550 dest->root_key.objectid);
4551 if (ret && ret != -ENOENT) {
4552 btrfs_abort_transaction(trans, ret);
4557 free_anon_bdev(dest->anon_dev);
4560 trans->block_rsv = NULL;
4561 trans->bytes_reserved = 0;
4562 ret = btrfs_end_transaction(trans);
4563 inode->i_flags |= S_DEAD;
4565 btrfs_subvolume_release_metadata(root, &block_rsv);
4567 up_write(&fs_info->subvol_sem);
4569 spin_lock(&dest->root_item_lock);
4570 root_flags = btrfs_root_flags(&dest->root_item);
4571 btrfs_set_root_flags(&dest->root_item,
4572 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4573 spin_unlock(&dest->root_item_lock);
4575 d_invalidate(dentry);
4576 btrfs_prune_dentries(dest);
4577 ASSERT(dest->send_in_progress == 0);
4583 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4585 struct inode *inode = d_inode(dentry);
4586 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4588 struct btrfs_trans_handle *trans;
4589 u64 last_unlink_trans;
4590 struct fscrypt_name fname;
4592 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4594 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4595 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4597 "extent tree v2 doesn't support snapshot deletion yet");
4600 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4603 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4607 /* This needs to handle no-key deletions later on */
4609 trans = __unlink_start_trans(BTRFS_I(dir));
4610 if (IS_ERR(trans)) {
4611 err = PTR_ERR(trans);
4615 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4616 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4620 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4624 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4626 /* now the directory is empty */
4627 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4630 btrfs_i_size_write(BTRFS_I(inode), 0);
4632 * Propagate the last_unlink_trans value of the deleted dir to
4633 * its parent directory. This is to prevent an unrecoverable
4634 * log tree in the case we do something like this:
4636 * 2) create snapshot under dir foo
4637 * 3) delete the snapshot
4640 * 6) fsync foo or some file inside foo
4642 if (last_unlink_trans >= trans->transid)
4643 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4646 btrfs_end_transaction(trans);
4648 btrfs_btree_balance_dirty(fs_info);
4649 fscrypt_free_filename(&fname);
4655 * Read, zero a chunk and write a block.
4657 * @inode - inode that we're zeroing
4658 * @from - the offset to start zeroing
4659 * @len - the length to zero, 0 to zero the entire range respective to the
4661 * @front - zero up to the offset instead of from the offset on
4663 * This will find the block for the "from" offset and cow the block and zero the
4664 * part we want to zero. This is used with truncate and hole punching.
4666 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4669 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4670 struct address_space *mapping = inode->vfs_inode.i_mapping;
4671 struct extent_io_tree *io_tree = &inode->io_tree;
4672 struct btrfs_ordered_extent *ordered;
4673 struct extent_state *cached_state = NULL;
4674 struct extent_changeset *data_reserved = NULL;
4675 bool only_release_metadata = false;
4676 u32 blocksize = fs_info->sectorsize;
4677 pgoff_t index = from >> PAGE_SHIFT;
4678 unsigned offset = from & (blocksize - 1);
4680 gfp_t mask = btrfs_alloc_write_mask(mapping);
4681 size_t write_bytes = blocksize;
4686 if (IS_ALIGNED(offset, blocksize) &&
4687 (!len || IS_ALIGNED(len, blocksize)))
4690 block_start = round_down(from, blocksize);
4691 block_end = block_start + blocksize - 1;
4693 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4696 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4697 /* For nocow case, no need to reserve data space */
4698 only_release_metadata = true;
4703 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4705 if (!only_release_metadata)
4706 btrfs_free_reserved_data_space(inode, data_reserved,
4707 block_start, blocksize);
4711 page = find_or_create_page(mapping, index, mask);
4713 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4715 btrfs_delalloc_release_extents(inode, blocksize);
4720 if (!PageUptodate(page)) {
4721 ret = btrfs_read_folio(NULL, page_folio(page));
4723 if (page->mapping != mapping) {
4728 if (!PageUptodate(page)) {
4735 * We unlock the page after the io is completed and then re-lock it
4736 * above. release_folio() could have come in between that and cleared
4737 * folio private, but left the page in the mapping. Set the page mapped
4738 * here to make sure it's properly set for the subpage stuff.
4740 ret = set_page_extent_mapped(page);
4744 wait_on_page_writeback(page);
4746 lock_extent(io_tree, block_start, block_end, &cached_state);
4748 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4750 unlock_extent(io_tree, block_start, block_end, &cached_state);
4753 btrfs_start_ordered_extent(ordered);
4754 btrfs_put_ordered_extent(ordered);
4758 clear_extent_bit(&inode->io_tree, block_start, block_end,
4759 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4762 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4765 unlock_extent(io_tree, block_start, block_end, &cached_state);
4769 if (offset != blocksize) {
4771 len = blocksize - offset;
4773 memzero_page(page, (block_start - page_offset(page)),
4776 memzero_page(page, (block_start - page_offset(page)) + offset,
4779 btrfs_folio_clear_checked(fs_info, page_folio(page), block_start,
4780 block_end + 1 - block_start);
4781 btrfs_folio_set_dirty(fs_info, page_folio(page), block_start,
4782 block_end + 1 - block_start);
4783 unlock_extent(io_tree, block_start, block_end, &cached_state);
4785 if (only_release_metadata)
4786 set_extent_bit(&inode->io_tree, block_start, block_end,
4787 EXTENT_NORESERVE, NULL);
4791 if (only_release_metadata)
4792 btrfs_delalloc_release_metadata(inode, blocksize, true);
4794 btrfs_delalloc_release_space(inode, data_reserved,
4795 block_start, blocksize, true);
4797 btrfs_delalloc_release_extents(inode, blocksize);
4801 if (only_release_metadata)
4802 btrfs_check_nocow_unlock(inode);
4803 extent_changeset_free(data_reserved);
4807 static int maybe_insert_hole(struct btrfs_inode *inode, u64 offset, u64 len)
4809 struct btrfs_root *root = inode->root;
4810 struct btrfs_fs_info *fs_info = root->fs_info;
4811 struct btrfs_trans_handle *trans;
4812 struct btrfs_drop_extents_args drop_args = { 0 };
4816 * If NO_HOLES is enabled, we don't need to do anything.
4817 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4818 * or btrfs_update_inode() will be called, which guarantee that the next
4819 * fsync will know this inode was changed and needs to be logged.
4821 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4825 * 1 - for the one we're dropping
4826 * 1 - for the one we're adding
4827 * 1 - for updating the inode.
4829 trans = btrfs_start_transaction(root, 3);
4831 return PTR_ERR(trans);
4833 drop_args.start = offset;
4834 drop_args.end = offset + len;
4835 drop_args.drop_cache = true;
4837 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4839 btrfs_abort_transaction(trans, ret);
4840 btrfs_end_transaction(trans);
4844 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4846 btrfs_abort_transaction(trans, ret);
4848 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4849 btrfs_update_inode(trans, inode);
4851 btrfs_end_transaction(trans);
4856 * This function puts in dummy file extents for the area we're creating a hole
4857 * for. So if we are truncating this file to a larger size we need to insert
4858 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4859 * the range between oldsize and size
4861 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4863 struct btrfs_root *root = inode->root;
4864 struct btrfs_fs_info *fs_info = root->fs_info;
4865 struct extent_io_tree *io_tree = &inode->io_tree;
4866 struct extent_map *em = NULL;
4867 struct extent_state *cached_state = NULL;
4868 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4869 u64 block_end = ALIGN(size, fs_info->sectorsize);
4876 * If our size started in the middle of a block we need to zero out the
4877 * rest of the block before we expand the i_size, otherwise we could
4878 * expose stale data.
4880 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4884 if (size <= hole_start)
4887 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4889 cur_offset = hole_start;
4891 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4892 block_end - cur_offset);
4898 last_byte = min(extent_map_end(em), block_end);
4899 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4900 hole_size = last_byte - cur_offset;
4902 if (!(em->flags & EXTENT_FLAG_PREALLOC)) {
4903 struct extent_map *hole_em;
4905 err = maybe_insert_hole(inode, cur_offset, hole_size);
4909 err = btrfs_inode_set_file_extent_range(inode,
4910 cur_offset, hole_size);
4914 hole_em = alloc_extent_map();
4916 btrfs_drop_extent_map_range(inode, cur_offset,
4917 cur_offset + hole_size - 1,
4919 btrfs_set_inode_full_sync(inode);
4922 hole_em->start = cur_offset;
4923 hole_em->len = hole_size;
4924 hole_em->orig_start = cur_offset;
4926 hole_em->block_start = EXTENT_MAP_HOLE;
4927 hole_em->block_len = 0;
4928 hole_em->orig_block_len = 0;
4929 hole_em->ram_bytes = hole_size;
4930 hole_em->generation = btrfs_get_fs_generation(fs_info);
4932 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4933 free_extent_map(hole_em);
4935 err = btrfs_inode_set_file_extent_range(inode,
4936 cur_offset, hole_size);
4941 free_extent_map(em);
4943 cur_offset = last_byte;
4944 if (cur_offset >= block_end)
4947 free_extent_map(em);
4948 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4952 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4954 struct btrfs_root *root = BTRFS_I(inode)->root;
4955 struct btrfs_trans_handle *trans;
4956 loff_t oldsize = i_size_read(inode);
4957 loff_t newsize = attr->ia_size;
4958 int mask = attr->ia_valid;
4962 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4963 * special case where we need to update the times despite not having
4964 * these flags set. For all other operations the VFS set these flags
4965 * explicitly if it wants a timestamp update.
4967 if (newsize != oldsize) {
4968 inode_inc_iversion(inode);
4969 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
4970 inode_set_mtime_to_ts(inode,
4971 inode_set_ctime_current(inode));
4975 if (newsize > oldsize) {
4977 * Don't do an expanding truncate while snapshotting is ongoing.
4978 * This is to ensure the snapshot captures a fully consistent
4979 * state of this file - if the snapshot captures this expanding
4980 * truncation, it must capture all writes that happened before
4983 btrfs_drew_write_lock(&root->snapshot_lock);
4984 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4986 btrfs_drew_write_unlock(&root->snapshot_lock);
4990 trans = btrfs_start_transaction(root, 1);
4991 if (IS_ERR(trans)) {
4992 btrfs_drew_write_unlock(&root->snapshot_lock);
4993 return PTR_ERR(trans);
4996 i_size_write(inode, newsize);
4997 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
4998 pagecache_isize_extended(inode, oldsize, newsize);
4999 ret = btrfs_update_inode(trans, BTRFS_I(inode));
5000 btrfs_drew_write_unlock(&root->snapshot_lock);
5001 btrfs_end_transaction(trans);
5003 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5005 if (btrfs_is_zoned(fs_info)) {
5006 ret = btrfs_wait_ordered_range(inode,
5007 ALIGN(newsize, fs_info->sectorsize),
5014 * We're truncating a file that used to have good data down to
5015 * zero. Make sure any new writes to the file get on disk
5019 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5020 &BTRFS_I(inode)->runtime_flags);
5022 truncate_setsize(inode, newsize);
5024 inode_dio_wait(inode);
5026 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5027 if (ret && inode->i_nlink) {
5031 * Truncate failed, so fix up the in-memory size. We
5032 * adjusted disk_i_size down as we removed extents, so
5033 * wait for disk_i_size to be stable and then update the
5034 * in-memory size to match.
5036 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5039 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5046 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5049 struct inode *inode = d_inode(dentry);
5050 struct btrfs_root *root = BTRFS_I(inode)->root;
5053 if (btrfs_root_readonly(root))
5056 err = setattr_prepare(idmap, dentry, attr);
5060 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5061 err = btrfs_setsize(inode, attr);
5066 if (attr->ia_valid) {
5067 setattr_copy(idmap, inode, attr);
5068 inode_inc_iversion(inode);
5069 err = btrfs_dirty_inode(BTRFS_I(inode));
5071 if (!err && attr->ia_valid & ATTR_MODE)
5072 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5079 * While truncating the inode pages during eviction, we get the VFS
5080 * calling btrfs_invalidate_folio() against each folio of the inode. This
5081 * is slow because the calls to btrfs_invalidate_folio() result in a
5082 * huge amount of calls to lock_extent() and clear_extent_bit(),
5083 * which keep merging and splitting extent_state structures over and over,
5084 * wasting lots of time.
5086 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5087 * skip all those expensive operations on a per folio basis and do only
5088 * the ordered io finishing, while we release here the extent_map and
5089 * extent_state structures, without the excessive merging and splitting.
5091 static void evict_inode_truncate_pages(struct inode *inode)
5093 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5094 struct rb_node *node;
5096 ASSERT(inode->i_state & I_FREEING);
5097 truncate_inode_pages_final(&inode->i_data);
5099 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5102 * Keep looping until we have no more ranges in the io tree.
5103 * We can have ongoing bios started by readahead that have
5104 * their endio callback (extent_io.c:end_bio_extent_readpage)
5105 * still in progress (unlocked the pages in the bio but did not yet
5106 * unlocked the ranges in the io tree). Therefore this means some
5107 * ranges can still be locked and eviction started because before
5108 * submitting those bios, which are executed by a separate task (work
5109 * queue kthread), inode references (inode->i_count) were not taken
5110 * (which would be dropped in the end io callback of each bio).
5111 * Therefore here we effectively end up waiting for those bios and
5112 * anyone else holding locked ranges without having bumped the inode's
5113 * reference count - if we don't do it, when they access the inode's
5114 * io_tree to unlock a range it may be too late, leading to an
5115 * use-after-free issue.
5117 spin_lock(&io_tree->lock);
5118 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5119 struct extent_state *state;
5120 struct extent_state *cached_state = NULL;
5123 unsigned state_flags;
5125 node = rb_first(&io_tree->state);
5126 state = rb_entry(node, struct extent_state, rb_node);
5127 start = state->start;
5129 state_flags = state->state;
5130 spin_unlock(&io_tree->lock);
5132 lock_extent(io_tree, start, end, &cached_state);
5135 * If still has DELALLOC flag, the extent didn't reach disk,
5136 * and its reserved space won't be freed by delayed_ref.
5137 * So we need to free its reserved space here.
5138 * (Refer to comment in btrfs_invalidate_folio, case 2)
5140 * Note, end is the bytenr of last byte, so we need + 1 here.
5142 if (state_flags & EXTENT_DELALLOC)
5143 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5144 end - start + 1, NULL);
5146 clear_extent_bit(io_tree, start, end,
5147 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5151 spin_lock(&io_tree->lock);
5153 spin_unlock(&io_tree->lock);
5156 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5157 struct btrfs_block_rsv *rsv)
5159 struct btrfs_fs_info *fs_info = root->fs_info;
5160 struct btrfs_trans_handle *trans;
5161 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5165 * Eviction should be taking place at some place safe because of our
5166 * delayed iputs. However the normal flushing code will run delayed
5167 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5169 * We reserve the delayed_refs_extra here again because we can't use
5170 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5171 * above. We reserve our extra bit here because we generate a ton of
5172 * delayed refs activity by truncating.
5174 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5175 * if we fail to make this reservation we can re-try without the
5176 * delayed_refs_extra so we can make some forward progress.
5178 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5179 BTRFS_RESERVE_FLUSH_EVICT);
5181 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5182 BTRFS_RESERVE_FLUSH_EVICT);
5185 "could not allocate space for delete; will truncate on mount");
5186 return ERR_PTR(-ENOSPC);
5188 delayed_refs_extra = 0;
5191 trans = btrfs_join_transaction(root);
5195 if (delayed_refs_extra) {
5196 trans->block_rsv = &fs_info->trans_block_rsv;
5197 trans->bytes_reserved = delayed_refs_extra;
5198 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5199 delayed_refs_extra, true);
5204 void btrfs_evict_inode(struct inode *inode)
5206 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5207 struct btrfs_trans_handle *trans;
5208 struct btrfs_root *root = BTRFS_I(inode)->root;
5209 struct btrfs_block_rsv *rsv = NULL;
5212 trace_btrfs_inode_evict(inode);
5215 fsverity_cleanup_inode(inode);
5220 evict_inode_truncate_pages(inode);
5222 if (inode->i_nlink &&
5223 ((btrfs_root_refs(&root->root_item) != 0 &&
5224 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5225 btrfs_is_free_space_inode(BTRFS_I(inode))))
5228 if (is_bad_inode(inode))
5231 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5234 if (inode->i_nlink > 0) {
5235 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5236 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5241 * This makes sure the inode item in tree is uptodate and the space for
5242 * the inode update is released.
5244 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5249 * This drops any pending insert or delete operations we have for this
5250 * inode. We could have a delayed dir index deletion queued up, but
5251 * we're removing the inode completely so that'll be taken care of in
5254 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5256 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5259 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5260 rsv->failfast = true;
5262 btrfs_i_size_write(BTRFS_I(inode), 0);
5265 struct btrfs_truncate_control control = {
5266 .inode = BTRFS_I(inode),
5267 .ino = btrfs_ino(BTRFS_I(inode)),
5272 trans = evict_refill_and_join(root, rsv);
5276 trans->block_rsv = rsv;
5278 ret = btrfs_truncate_inode_items(trans, root, &control);
5279 trans->block_rsv = &fs_info->trans_block_rsv;
5280 btrfs_end_transaction(trans);
5282 * We have not added new delayed items for our inode after we
5283 * have flushed its delayed items, so no need to throttle on
5284 * delayed items. However we have modified extent buffers.
5286 btrfs_btree_balance_dirty_nodelay(fs_info);
5287 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5294 * Errors here aren't a big deal, it just means we leave orphan items in
5295 * the tree. They will be cleaned up on the next mount. If the inode
5296 * number gets reused, cleanup deletes the orphan item without doing
5297 * anything, and unlink reuses the existing orphan item.
5299 * If it turns out that we are dropping too many of these, we might want
5300 * to add a mechanism for retrying these after a commit.
5302 trans = evict_refill_and_join(root, rsv);
5303 if (!IS_ERR(trans)) {
5304 trans->block_rsv = rsv;
5305 btrfs_orphan_del(trans, BTRFS_I(inode));
5306 trans->block_rsv = &fs_info->trans_block_rsv;
5307 btrfs_end_transaction(trans);
5311 btrfs_free_block_rsv(fs_info, rsv);
5313 * If we didn't successfully delete, the orphan item will still be in
5314 * the tree and we'll retry on the next mount. Again, we might also want
5315 * to retry these periodically in the future.
5317 btrfs_remove_delayed_node(BTRFS_I(inode));
5318 fsverity_cleanup_inode(inode);
5323 * Return the key found in the dir entry in the location pointer, fill @type
5324 * with BTRFS_FT_*, and return 0.
5326 * If no dir entries were found, returns -ENOENT.
5327 * If found a corrupted location in dir entry, returns -EUCLEAN.
5329 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5330 struct btrfs_key *location, u8 *type)
5332 struct btrfs_dir_item *di;
5333 struct btrfs_path *path;
5334 struct btrfs_root *root = dir->root;
5336 struct fscrypt_name fname;
5338 path = btrfs_alloc_path();
5342 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5346 * fscrypt_setup_filename() should never return a positive value, but
5347 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5351 /* This needs to handle no-key deletions later on */
5353 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5354 &fname.disk_name, 0);
5355 if (IS_ERR_OR_NULL(di)) {
5356 ret = di ? PTR_ERR(di) : -ENOENT;
5360 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5361 if (location->type != BTRFS_INODE_ITEM_KEY &&
5362 location->type != BTRFS_ROOT_ITEM_KEY) {
5364 btrfs_warn(root->fs_info,
5365 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5366 __func__, fname.disk_name.name, btrfs_ino(dir),
5367 location->objectid, location->type, location->offset);
5370 *type = btrfs_dir_ftype(path->nodes[0], di);
5372 fscrypt_free_filename(&fname);
5373 btrfs_free_path(path);
5378 * when we hit a tree root in a directory, the btrfs part of the inode
5379 * needs to be changed to reflect the root directory of the tree root. This
5380 * is kind of like crossing a mount point.
5382 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5383 struct btrfs_inode *dir,
5384 struct dentry *dentry,
5385 struct btrfs_key *location,
5386 struct btrfs_root **sub_root)
5388 struct btrfs_path *path;
5389 struct btrfs_root *new_root;
5390 struct btrfs_root_ref *ref;
5391 struct extent_buffer *leaf;
5392 struct btrfs_key key;
5395 struct fscrypt_name fname;
5397 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5401 path = btrfs_alloc_path();
5408 key.objectid = dir->root->root_key.objectid;
5409 key.type = BTRFS_ROOT_REF_KEY;
5410 key.offset = location->objectid;
5412 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5419 leaf = path->nodes[0];
5420 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5421 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5422 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5425 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5426 (unsigned long)(ref + 1), fname.disk_name.len);
5430 btrfs_release_path(path);
5432 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5433 if (IS_ERR(new_root)) {
5434 err = PTR_ERR(new_root);
5438 *sub_root = new_root;
5439 location->objectid = btrfs_root_dirid(&new_root->root_item);
5440 location->type = BTRFS_INODE_ITEM_KEY;
5441 location->offset = 0;
5444 btrfs_free_path(path);
5445 fscrypt_free_filename(&fname);
5449 static void inode_tree_add(struct btrfs_inode *inode)
5451 struct btrfs_root *root = inode->root;
5452 struct btrfs_inode *entry;
5454 struct rb_node *parent;
5455 struct rb_node *new = &inode->rb_node;
5456 u64 ino = btrfs_ino(inode);
5458 if (inode_unhashed(&inode->vfs_inode))
5461 spin_lock(&root->inode_lock);
5462 p = &root->inode_tree.rb_node;
5465 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5467 if (ino < btrfs_ino(entry))
5468 p = &parent->rb_left;
5469 else if (ino > btrfs_ino(entry))
5470 p = &parent->rb_right;
5472 WARN_ON(!(entry->vfs_inode.i_state &
5473 (I_WILL_FREE | I_FREEING)));
5474 rb_replace_node(parent, new, &root->inode_tree);
5475 RB_CLEAR_NODE(parent);
5476 spin_unlock(&root->inode_lock);
5480 rb_link_node(new, parent, p);
5481 rb_insert_color(new, &root->inode_tree);
5482 spin_unlock(&root->inode_lock);
5485 static void inode_tree_del(struct btrfs_inode *inode)
5487 struct btrfs_root *root = inode->root;
5490 spin_lock(&root->inode_lock);
5491 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5492 rb_erase(&inode->rb_node, &root->inode_tree);
5493 RB_CLEAR_NODE(&inode->rb_node);
5494 empty = RB_EMPTY_ROOT(&root->inode_tree);
5496 spin_unlock(&root->inode_lock);
5498 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5499 spin_lock(&root->inode_lock);
5500 empty = RB_EMPTY_ROOT(&root->inode_tree);
5501 spin_unlock(&root->inode_lock);
5503 btrfs_add_dead_root(root);
5508 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5510 struct btrfs_iget_args *args = p;
5512 inode->i_ino = args->ino;
5513 BTRFS_I(inode)->location.objectid = args->ino;
5514 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5515 BTRFS_I(inode)->location.offset = 0;
5516 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5517 BUG_ON(args->root && !BTRFS_I(inode)->root);
5519 if (args->root && args->root == args->root->fs_info->tree_root &&
5520 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5521 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5522 &BTRFS_I(inode)->runtime_flags);
5526 static int btrfs_find_actor(struct inode *inode, void *opaque)
5528 struct btrfs_iget_args *args = opaque;
5530 return args->ino == BTRFS_I(inode)->location.objectid &&
5531 args->root == BTRFS_I(inode)->root;
5534 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5535 struct btrfs_root *root)
5537 struct inode *inode;
5538 struct btrfs_iget_args args;
5539 unsigned long hashval = btrfs_inode_hash(ino, root);
5544 inode = iget5_locked(s, hashval, btrfs_find_actor,
5545 btrfs_init_locked_inode,
5551 * Get an inode object given its inode number and corresponding root.
5552 * Path can be preallocated to prevent recursing back to iget through
5553 * allocator. NULL is also valid but may require an additional allocation
5556 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5557 struct btrfs_root *root, struct btrfs_path *path)
5559 struct inode *inode;
5561 inode = btrfs_iget_locked(s, ino, root);
5563 return ERR_PTR(-ENOMEM);
5565 if (inode->i_state & I_NEW) {
5568 ret = btrfs_read_locked_inode(inode, path);
5570 inode_tree_add(BTRFS_I(inode));
5571 unlock_new_inode(inode);
5575 * ret > 0 can come from btrfs_search_slot called by
5576 * btrfs_read_locked_inode, this means the inode item
5581 inode = ERR_PTR(ret);
5588 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5590 return btrfs_iget_path(s, ino, root, NULL);
5593 static struct inode *new_simple_dir(struct inode *dir,
5594 struct btrfs_key *key,
5595 struct btrfs_root *root)
5597 struct timespec64 ts;
5598 struct inode *inode = new_inode(dir->i_sb);
5601 return ERR_PTR(-ENOMEM);
5603 BTRFS_I(inode)->root = btrfs_grab_root(root);
5604 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5605 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5607 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5609 * We only need lookup, the rest is read-only and there's no inode
5610 * associated with the dentry
5612 inode->i_op = &simple_dir_inode_operations;
5613 inode->i_opflags &= ~IOP_XATTR;
5614 inode->i_fop = &simple_dir_operations;
5615 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5617 ts = inode_set_ctime_current(inode);
5618 inode_set_mtime_to_ts(inode, ts);
5619 inode_set_atime_to_ts(inode, inode_get_atime(dir));
5620 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
5621 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
5623 inode->i_uid = dir->i_uid;
5624 inode->i_gid = dir->i_gid;
5629 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5630 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5631 static_assert(BTRFS_FT_DIR == FT_DIR);
5632 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5633 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5634 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5635 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5636 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5638 static inline u8 btrfs_inode_type(struct inode *inode)
5640 return fs_umode_to_ftype(inode->i_mode);
5643 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5645 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5646 struct inode *inode;
5647 struct btrfs_root *root = BTRFS_I(dir)->root;
5648 struct btrfs_root *sub_root = root;
5649 struct btrfs_key location;
5653 if (dentry->d_name.len > BTRFS_NAME_LEN)
5654 return ERR_PTR(-ENAMETOOLONG);
5656 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5658 return ERR_PTR(ret);
5660 if (location.type == BTRFS_INODE_ITEM_KEY) {
5661 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5665 /* Do extra check against inode mode with di_type */
5666 if (btrfs_inode_type(inode) != di_type) {
5668 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5669 inode->i_mode, btrfs_inode_type(inode),
5672 return ERR_PTR(-EUCLEAN);
5677 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5678 &location, &sub_root);
5681 inode = ERR_PTR(ret);
5683 inode = new_simple_dir(dir, &location, root);
5685 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5686 btrfs_put_root(sub_root);
5691 down_read(&fs_info->cleanup_work_sem);
5692 if (!sb_rdonly(inode->i_sb))
5693 ret = btrfs_orphan_cleanup(sub_root);
5694 up_read(&fs_info->cleanup_work_sem);
5697 inode = ERR_PTR(ret);
5704 static int btrfs_dentry_delete(const struct dentry *dentry)
5706 struct btrfs_root *root;
5707 struct inode *inode = d_inode(dentry);
5709 if (!inode && !IS_ROOT(dentry))
5710 inode = d_inode(dentry->d_parent);
5713 root = BTRFS_I(inode)->root;
5714 if (btrfs_root_refs(&root->root_item) == 0)
5717 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5723 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5726 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5728 if (inode == ERR_PTR(-ENOENT))
5730 return d_splice_alias(inode, dentry);
5734 * Find the highest existing sequence number in a directory and then set the
5735 * in-memory index_cnt variable to the first free sequence number.
5737 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5739 struct btrfs_root *root = inode->root;
5740 struct btrfs_key key, found_key;
5741 struct btrfs_path *path;
5742 struct extent_buffer *leaf;
5745 key.objectid = btrfs_ino(inode);
5746 key.type = BTRFS_DIR_INDEX_KEY;
5747 key.offset = (u64)-1;
5749 path = btrfs_alloc_path();
5753 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5756 /* FIXME: we should be able to handle this */
5761 if (path->slots[0] == 0) {
5762 inode->index_cnt = BTRFS_DIR_START_INDEX;
5768 leaf = path->nodes[0];
5769 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5771 if (found_key.objectid != btrfs_ino(inode) ||
5772 found_key.type != BTRFS_DIR_INDEX_KEY) {
5773 inode->index_cnt = BTRFS_DIR_START_INDEX;
5777 inode->index_cnt = found_key.offset + 1;
5779 btrfs_free_path(path);
5783 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5787 btrfs_inode_lock(dir, 0);
5788 if (dir->index_cnt == (u64)-1) {
5789 ret = btrfs_inode_delayed_dir_index_count(dir);
5791 ret = btrfs_set_inode_index_count(dir);
5797 /* index_cnt is the index number of next new entry, so decrement it. */
5798 *index = dir->index_cnt - 1;
5800 btrfs_inode_unlock(dir, 0);
5806 * All this infrastructure exists because dir_emit can fault, and we are holding
5807 * the tree lock when doing readdir. For now just allocate a buffer and copy
5808 * our information into that, and then dir_emit from the buffer. This is
5809 * similar to what NFS does, only we don't keep the buffer around in pagecache
5810 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5811 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5814 static int btrfs_opendir(struct inode *inode, struct file *file)
5816 struct btrfs_file_private *private;
5820 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5824 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5827 private->last_index = last_index;
5828 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5829 if (!private->filldir_buf) {
5833 file->private_data = private;
5837 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5839 struct btrfs_file_private *private = file->private_data;
5842 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
5843 &private->last_index);
5847 return generic_file_llseek(file, offset, whence);
5857 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5860 struct dir_entry *entry = addr;
5861 char *name = (char *)(entry + 1);
5863 ctx->pos = get_unaligned(&entry->offset);
5864 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5865 get_unaligned(&entry->ino),
5866 get_unaligned(&entry->type)))
5868 addr += sizeof(struct dir_entry) +
5869 get_unaligned(&entry->name_len);
5875 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5877 struct inode *inode = file_inode(file);
5878 struct btrfs_root *root = BTRFS_I(inode)->root;
5879 struct btrfs_file_private *private = file->private_data;
5880 struct btrfs_dir_item *di;
5881 struct btrfs_key key;
5882 struct btrfs_key found_key;
5883 struct btrfs_path *path;
5885 LIST_HEAD(ins_list);
5886 LIST_HEAD(del_list);
5893 struct btrfs_key location;
5895 if (!dir_emit_dots(file, ctx))
5898 path = btrfs_alloc_path();
5902 addr = private->filldir_buf;
5903 path->reada = READA_FORWARD;
5905 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5906 &ins_list, &del_list);
5909 key.type = BTRFS_DIR_INDEX_KEY;
5910 key.offset = ctx->pos;
5911 key.objectid = btrfs_ino(BTRFS_I(inode));
5913 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5914 struct dir_entry *entry;
5915 struct extent_buffer *leaf = path->nodes[0];
5918 if (found_key.objectid != key.objectid)
5920 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5922 if (found_key.offset < ctx->pos)
5924 if (found_key.offset > private->last_index)
5926 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5928 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5929 name_len = btrfs_dir_name_len(leaf, di);
5930 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5932 btrfs_release_path(path);
5933 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5936 addr = private->filldir_buf;
5942 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5944 name_ptr = (char *)(entry + 1);
5945 read_extent_buffer(leaf, name_ptr,
5946 (unsigned long)(di + 1), name_len);
5947 put_unaligned(name_len, &entry->name_len);
5948 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5949 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5950 put_unaligned(location.objectid, &entry->ino);
5951 put_unaligned(found_key.offset, &entry->offset);
5953 addr += sizeof(struct dir_entry) + name_len;
5954 total_len += sizeof(struct dir_entry) + name_len;
5956 /* Catch error encountered during iteration */
5960 btrfs_release_path(path);
5962 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5966 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5971 * Stop new entries from being returned after we return the last
5974 * New directory entries are assigned a strictly increasing
5975 * offset. This means that new entries created during readdir
5976 * are *guaranteed* to be seen in the future by that readdir.
5977 * This has broken buggy programs which operate on names as
5978 * they're returned by readdir. Until we re-use freed offsets
5979 * we have this hack to stop new entries from being returned
5980 * under the assumption that they'll never reach this huge
5983 * This is being careful not to overflow 32bit loff_t unless the
5984 * last entry requires it because doing so has broken 32bit apps
5987 if (ctx->pos >= INT_MAX)
5988 ctx->pos = LLONG_MAX;
5995 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5996 btrfs_free_path(path);
6001 * This is somewhat expensive, updating the tree every time the
6002 * inode changes. But, it is most likely to find the inode in cache.
6003 * FIXME, needs more benchmarking...there are no reasons other than performance
6004 * to keep or drop this code.
6006 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6008 struct btrfs_root *root = inode->root;
6009 struct btrfs_fs_info *fs_info = root->fs_info;
6010 struct btrfs_trans_handle *trans;
6013 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6016 trans = btrfs_join_transaction(root);
6018 return PTR_ERR(trans);
6020 ret = btrfs_update_inode(trans, inode);
6021 if (ret == -ENOSPC || ret == -EDQUOT) {
6022 /* whoops, lets try again with the full transaction */
6023 btrfs_end_transaction(trans);
6024 trans = btrfs_start_transaction(root, 1);
6026 return PTR_ERR(trans);
6028 ret = btrfs_update_inode(trans, inode);
6030 btrfs_end_transaction(trans);
6031 if (inode->delayed_node)
6032 btrfs_balance_delayed_items(fs_info);
6038 * This is a copy of file_update_time. We need this so we can return error on
6039 * ENOSPC for updating the inode in the case of file write and mmap writes.
6041 static int btrfs_update_time(struct inode *inode, int flags)
6043 struct btrfs_root *root = BTRFS_I(inode)->root;
6046 if (btrfs_root_readonly(root))
6049 dirty = inode_update_timestamps(inode, flags);
6050 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6054 * helper to find a free sequence number in a given directory. This current
6055 * code is very simple, later versions will do smarter things in the btree
6057 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6061 if (dir->index_cnt == (u64)-1) {
6062 ret = btrfs_inode_delayed_dir_index_count(dir);
6064 ret = btrfs_set_inode_index_count(dir);
6070 *index = dir->index_cnt;
6076 static int btrfs_insert_inode_locked(struct inode *inode)
6078 struct btrfs_iget_args args;
6080 args.ino = BTRFS_I(inode)->location.objectid;
6081 args.root = BTRFS_I(inode)->root;
6083 return insert_inode_locked4(inode,
6084 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6085 btrfs_find_actor, &args);
6088 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6089 unsigned int *trans_num_items)
6091 struct inode *dir = args->dir;
6092 struct inode *inode = args->inode;
6095 if (!args->orphan) {
6096 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6102 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6104 fscrypt_free_filename(&args->fname);
6108 /* 1 to add inode item */
6109 *trans_num_items = 1;
6110 /* 1 to add compression property */
6111 if (BTRFS_I(dir)->prop_compress)
6112 (*trans_num_items)++;
6113 /* 1 to add default ACL xattr */
6114 if (args->default_acl)
6115 (*trans_num_items)++;
6116 /* 1 to add access ACL xattr */
6118 (*trans_num_items)++;
6119 #ifdef CONFIG_SECURITY
6120 /* 1 to add LSM xattr */
6121 if (dir->i_security)
6122 (*trans_num_items)++;
6125 /* 1 to add orphan item */
6126 (*trans_num_items)++;
6130 * 1 to add dir index
6131 * 1 to update parent inode item
6133 * No need for 1 unit for the inode ref item because it is
6134 * inserted in a batch together with the inode item at
6135 * btrfs_create_new_inode().
6137 *trans_num_items += 3;
6142 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6144 posix_acl_release(args->acl);
6145 posix_acl_release(args->default_acl);
6146 fscrypt_free_filename(&args->fname);
6150 * Inherit flags from the parent inode.
6152 * Currently only the compression flags and the cow flags are inherited.
6154 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6160 if (flags & BTRFS_INODE_NOCOMPRESS) {
6161 inode->flags &= ~BTRFS_INODE_COMPRESS;
6162 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6163 } else if (flags & BTRFS_INODE_COMPRESS) {
6164 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6165 inode->flags |= BTRFS_INODE_COMPRESS;
6168 if (flags & BTRFS_INODE_NODATACOW) {
6169 inode->flags |= BTRFS_INODE_NODATACOW;
6170 if (S_ISREG(inode->vfs_inode.i_mode))
6171 inode->flags |= BTRFS_INODE_NODATASUM;
6174 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6177 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6178 struct btrfs_new_inode_args *args)
6180 struct timespec64 ts;
6181 struct inode *dir = args->dir;
6182 struct inode *inode = args->inode;
6183 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6184 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6185 struct btrfs_root *root;
6186 struct btrfs_inode_item *inode_item;
6187 struct btrfs_key *location;
6188 struct btrfs_path *path;
6190 struct btrfs_inode_ref *ref;
6191 struct btrfs_key key[2];
6193 struct btrfs_item_batch batch;
6197 path = btrfs_alloc_path();
6202 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6203 root = BTRFS_I(inode)->root;
6205 ret = btrfs_get_free_objectid(root, &objectid);
6208 inode->i_ino = objectid;
6212 * O_TMPFILE, set link count to 0, so that after this point, we
6213 * fill in an inode item with the correct link count.
6215 set_nlink(inode, 0);
6217 trace_btrfs_inode_request(dir);
6219 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6223 /* index_cnt is ignored for everything but a dir. */
6224 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6225 BTRFS_I(inode)->generation = trans->transid;
6226 inode->i_generation = BTRFS_I(inode)->generation;
6229 * We don't have any capability xattrs set here yet, shortcut any
6230 * queries for the xattrs here. If we add them later via the inode
6231 * security init path or any other path this flag will be cleared.
6233 set_bit(BTRFS_INODE_NO_CAP_XATTR, &BTRFS_I(inode)->runtime_flags);
6236 * Subvolumes don't inherit flags from their parent directory.
6237 * Originally this was probably by accident, but we probably can't
6238 * change it now without compatibility issues.
6241 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6243 if (S_ISREG(inode->i_mode)) {
6244 if (btrfs_test_opt(fs_info, NODATASUM))
6245 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6246 if (btrfs_test_opt(fs_info, NODATACOW))
6247 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6248 BTRFS_INODE_NODATASUM;
6251 location = &BTRFS_I(inode)->location;
6252 location->objectid = objectid;
6253 location->offset = 0;
6254 location->type = BTRFS_INODE_ITEM_KEY;
6256 ret = btrfs_insert_inode_locked(inode);
6259 BTRFS_I(dir)->index_cnt--;
6264 * We could have gotten an inode number from somebody who was fsynced
6265 * and then removed in this same transaction, so let's just set full
6266 * sync since it will be a full sync anyway and this will blow away the
6267 * old info in the log.
6269 btrfs_set_inode_full_sync(BTRFS_I(inode));
6271 key[0].objectid = objectid;
6272 key[0].type = BTRFS_INODE_ITEM_KEY;
6275 sizes[0] = sizeof(struct btrfs_inode_item);
6277 if (!args->orphan) {
6279 * Start new inodes with an inode_ref. This is slightly more
6280 * efficient for small numbers of hard links since they will
6281 * be packed into one item. Extended refs will kick in if we
6282 * add more hard links than can fit in the ref item.
6284 key[1].objectid = objectid;
6285 key[1].type = BTRFS_INODE_REF_KEY;
6287 key[1].offset = objectid;
6288 sizes[1] = 2 + sizeof(*ref);
6290 key[1].offset = btrfs_ino(BTRFS_I(dir));
6291 sizes[1] = name->len + sizeof(*ref);
6295 batch.keys = &key[0];
6296 batch.data_sizes = &sizes[0];
6297 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6298 batch.nr = args->orphan ? 1 : 2;
6299 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6301 btrfs_abort_transaction(trans, ret);
6305 ts = simple_inode_init_ts(inode);
6306 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
6307 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
6310 * We're going to fill the inode item now, so at this point the inode
6311 * must be fully initialized.
6314 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6315 struct btrfs_inode_item);
6316 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6317 sizeof(*inode_item));
6318 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6320 if (!args->orphan) {
6321 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6322 struct btrfs_inode_ref);
6323 ptr = (unsigned long)(ref + 1);
6325 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6326 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6327 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6329 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6331 btrfs_set_inode_ref_index(path->nodes[0], ref,
6332 BTRFS_I(inode)->dir_index);
6333 write_extent_buffer(path->nodes[0], name->name, ptr,
6338 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
6340 * We don't need the path anymore, plus inheriting properties, adding
6341 * ACLs, security xattrs, orphan item or adding the link, will result in
6342 * allocating yet another path. So just free our path.
6344 btrfs_free_path(path);
6348 struct inode *parent;
6351 * Subvolumes inherit properties from their parent subvolume,
6352 * not the directory they were created in.
6354 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6355 BTRFS_I(dir)->root);
6356 if (IS_ERR(parent)) {
6357 ret = PTR_ERR(parent);
6359 ret = btrfs_inode_inherit_props(trans, inode, parent);
6363 ret = btrfs_inode_inherit_props(trans, inode, dir);
6367 "error inheriting props for ino %llu (root %llu): %d",
6368 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6373 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6376 if (!args->subvol) {
6377 ret = btrfs_init_inode_security(trans, args);
6379 btrfs_abort_transaction(trans, ret);
6384 inode_tree_add(BTRFS_I(inode));
6386 trace_btrfs_inode_new(inode);
6387 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6389 btrfs_update_root_times(trans, root);
6392 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6394 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6395 0, BTRFS_I(inode)->dir_index);
6398 btrfs_abort_transaction(trans, ret);
6406 * discard_new_inode() calls iput(), but the caller owns the reference
6410 discard_new_inode(inode);
6412 btrfs_free_path(path);
6417 * utility function to add 'inode' into 'parent_inode' with
6418 * a give name and a given sequence number.
6419 * if 'add_backref' is true, also insert a backref from the
6420 * inode to the parent directory.
6422 int btrfs_add_link(struct btrfs_trans_handle *trans,
6423 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6424 const struct fscrypt_str *name, int add_backref, u64 index)
6427 struct btrfs_key key;
6428 struct btrfs_root *root = parent_inode->root;
6429 u64 ino = btrfs_ino(inode);
6430 u64 parent_ino = btrfs_ino(parent_inode);
6432 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6433 memcpy(&key, &inode->root->root_key, sizeof(key));
6436 key.type = BTRFS_INODE_ITEM_KEY;
6440 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6441 ret = btrfs_add_root_ref(trans, key.objectid,
6442 root->root_key.objectid, parent_ino,
6444 } else if (add_backref) {
6445 ret = btrfs_insert_inode_ref(trans, root, name,
6446 ino, parent_ino, index);
6449 /* Nothing to clean up yet */
6453 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6454 btrfs_inode_type(&inode->vfs_inode), index);
6455 if (ret == -EEXIST || ret == -EOVERFLOW)
6458 btrfs_abort_transaction(trans, ret);
6462 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6464 inode_inc_iversion(&parent_inode->vfs_inode);
6466 * If we are replaying a log tree, we do not want to update the mtime
6467 * and ctime of the parent directory with the current time, since the
6468 * log replay procedure is responsible for setting them to their correct
6469 * values (the ones it had when the fsync was done).
6471 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6472 inode_set_mtime_to_ts(&parent_inode->vfs_inode,
6473 inode_set_ctime_current(&parent_inode->vfs_inode));
6475 ret = btrfs_update_inode(trans, parent_inode);
6477 btrfs_abort_transaction(trans, ret);
6481 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6484 err = btrfs_del_root_ref(trans, key.objectid,
6485 root->root_key.objectid, parent_ino,
6486 &local_index, name);
6488 btrfs_abort_transaction(trans, err);
6489 } else if (add_backref) {
6493 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6496 btrfs_abort_transaction(trans, err);
6499 /* Return the original error code */
6503 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6504 struct inode *inode)
6506 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6507 struct btrfs_root *root = BTRFS_I(dir)->root;
6508 struct btrfs_new_inode_args new_inode_args = {
6513 unsigned int trans_num_items;
6514 struct btrfs_trans_handle *trans;
6517 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6521 trans = btrfs_start_transaction(root, trans_num_items);
6522 if (IS_ERR(trans)) {
6523 err = PTR_ERR(trans);
6524 goto out_new_inode_args;
6527 err = btrfs_create_new_inode(trans, &new_inode_args);
6529 d_instantiate_new(dentry, inode);
6531 btrfs_end_transaction(trans);
6532 btrfs_btree_balance_dirty(fs_info);
6534 btrfs_new_inode_args_destroy(&new_inode_args);
6541 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6542 struct dentry *dentry, umode_t mode, dev_t rdev)
6544 struct inode *inode;
6546 inode = new_inode(dir->i_sb);
6549 inode_init_owner(idmap, inode, dir, mode);
6550 inode->i_op = &btrfs_special_inode_operations;
6551 init_special_inode(inode, inode->i_mode, rdev);
6552 return btrfs_create_common(dir, dentry, inode);
6555 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6556 struct dentry *dentry, umode_t mode, bool excl)
6558 struct inode *inode;
6560 inode = new_inode(dir->i_sb);
6563 inode_init_owner(idmap, inode, dir, mode);
6564 inode->i_fop = &btrfs_file_operations;
6565 inode->i_op = &btrfs_file_inode_operations;
6566 inode->i_mapping->a_ops = &btrfs_aops;
6567 return btrfs_create_common(dir, dentry, inode);
6570 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6571 struct dentry *dentry)
6573 struct btrfs_trans_handle *trans = NULL;
6574 struct btrfs_root *root = BTRFS_I(dir)->root;
6575 struct inode *inode = d_inode(old_dentry);
6576 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6577 struct fscrypt_name fname;
6582 /* do not allow sys_link's with other subvols of the same device */
6583 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6586 if (inode->i_nlink >= BTRFS_LINK_MAX)
6589 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6593 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6598 * 2 items for inode and inode ref
6599 * 2 items for dir items
6600 * 1 item for parent inode
6601 * 1 item for orphan item deletion if O_TMPFILE
6603 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6604 if (IS_ERR(trans)) {
6605 err = PTR_ERR(trans);
6610 /* There are several dir indexes for this inode, clear the cache. */
6611 BTRFS_I(inode)->dir_index = 0ULL;
6613 inode_inc_iversion(inode);
6614 inode_set_ctime_current(inode);
6616 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6618 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6619 &fname.disk_name, 1, index);
6624 struct dentry *parent = dentry->d_parent;
6626 err = btrfs_update_inode(trans, BTRFS_I(inode));
6629 if (inode->i_nlink == 1) {
6631 * If new hard link count is 1, it's a file created
6632 * with open(2) O_TMPFILE flag.
6634 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6638 d_instantiate(dentry, inode);
6639 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6643 fscrypt_free_filename(&fname);
6645 btrfs_end_transaction(trans);
6647 inode_dec_link_count(inode);
6650 btrfs_btree_balance_dirty(fs_info);
6654 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6655 struct dentry *dentry, umode_t mode)
6657 struct inode *inode;
6659 inode = new_inode(dir->i_sb);
6662 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6663 inode->i_op = &btrfs_dir_inode_operations;
6664 inode->i_fop = &btrfs_dir_file_operations;
6665 return btrfs_create_common(dir, dentry, inode);
6668 static noinline int uncompress_inline(struct btrfs_path *path,
6670 struct btrfs_file_extent_item *item)
6673 struct extent_buffer *leaf = path->nodes[0];
6676 unsigned long inline_size;
6680 compress_type = btrfs_file_extent_compression(leaf, item);
6681 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6682 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6683 tmp = kmalloc(inline_size, GFP_NOFS);
6686 ptr = btrfs_file_extent_inline_start(item);
6688 read_extent_buffer(leaf, tmp, ptr, inline_size);
6690 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6691 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6694 * decompression code contains a memset to fill in any space between the end
6695 * of the uncompressed data and the end of max_size in case the decompressed
6696 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6697 * the end of an inline extent and the beginning of the next block, so we
6698 * cover that region here.
6701 if (max_size < PAGE_SIZE)
6702 memzero_page(page, max_size, PAGE_SIZE - max_size);
6707 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6710 struct btrfs_file_extent_item *fi;
6714 if (!page || PageUptodate(page))
6717 ASSERT(page_offset(page) == 0);
6719 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6720 struct btrfs_file_extent_item);
6721 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6722 return uncompress_inline(path, page, fi);
6724 copy_size = min_t(u64, PAGE_SIZE,
6725 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6726 kaddr = kmap_local_page(page);
6727 read_extent_buffer(path->nodes[0], kaddr,
6728 btrfs_file_extent_inline_start(fi), copy_size);
6729 kunmap_local(kaddr);
6730 if (copy_size < PAGE_SIZE)
6731 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6736 * Lookup the first extent overlapping a range in a file.
6738 * @inode: file to search in
6739 * @page: page to read extent data into if the extent is inline
6740 * @pg_offset: offset into @page to copy to
6741 * @start: file offset
6742 * @len: length of range starting at @start
6744 * Return the first &struct extent_map which overlaps the given range, reading
6745 * it from the B-tree and caching it if necessary. Note that there may be more
6746 * extents which overlap the given range after the returned extent_map.
6748 * If @page is not NULL and the extent is inline, this also reads the extent
6749 * data directly into the page and marks the extent up to date in the io_tree.
6751 * Return: ERR_PTR on error, non-NULL extent_map on success.
6753 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6754 struct page *page, size_t pg_offset,
6757 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6759 u64 extent_start = 0;
6761 u64 objectid = btrfs_ino(inode);
6762 int extent_type = -1;
6763 struct btrfs_path *path = NULL;
6764 struct btrfs_root *root = inode->root;
6765 struct btrfs_file_extent_item *item;
6766 struct extent_buffer *leaf;
6767 struct btrfs_key found_key;
6768 struct extent_map *em = NULL;
6769 struct extent_map_tree *em_tree = &inode->extent_tree;
6771 read_lock(&em_tree->lock);
6772 em = lookup_extent_mapping(em_tree, start, len);
6773 read_unlock(&em_tree->lock);
6776 if (em->start > start || em->start + em->len <= start)
6777 free_extent_map(em);
6778 else if (em->block_start == EXTENT_MAP_INLINE && page)
6779 free_extent_map(em);
6783 em = alloc_extent_map();
6788 em->start = EXTENT_MAP_HOLE;
6789 em->orig_start = EXTENT_MAP_HOLE;
6791 em->block_len = (u64)-1;
6793 path = btrfs_alloc_path();
6799 /* Chances are we'll be called again, so go ahead and do readahead */
6800 path->reada = READA_FORWARD;
6803 * The same explanation in load_free_space_cache applies here as well,
6804 * we only read when we're loading the free space cache, and at that
6805 * point the commit_root has everything we need.
6807 if (btrfs_is_free_space_inode(inode)) {
6808 path->search_commit_root = 1;
6809 path->skip_locking = 1;
6812 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6815 } else if (ret > 0) {
6816 if (path->slots[0] == 0)
6822 leaf = path->nodes[0];
6823 item = btrfs_item_ptr(leaf, path->slots[0],
6824 struct btrfs_file_extent_item);
6825 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6826 if (found_key.objectid != objectid ||
6827 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6829 * If we backup past the first extent we want to move forward
6830 * and see if there is an extent in front of us, otherwise we'll
6831 * say there is a hole for our whole search range which can
6838 extent_type = btrfs_file_extent_type(leaf, item);
6839 extent_start = found_key.offset;
6840 extent_end = btrfs_file_extent_end(path);
6841 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6842 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6843 /* Only regular file could have regular/prealloc extent */
6844 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6847 "regular/prealloc extent found for non-regular inode %llu",
6851 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6853 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6854 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6859 if (start >= extent_end) {
6861 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6862 ret = btrfs_next_leaf(root, path);
6868 leaf = path->nodes[0];
6870 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6871 if (found_key.objectid != objectid ||
6872 found_key.type != BTRFS_EXTENT_DATA_KEY)
6874 if (start + len <= found_key.offset)
6876 if (start > found_key.offset)
6879 /* New extent overlaps with existing one */
6881 em->orig_start = start;
6882 em->len = found_key.offset - start;
6883 em->block_start = EXTENT_MAP_HOLE;
6887 btrfs_extent_item_to_extent_map(inode, path, item, em);
6889 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6890 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6892 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6894 * Inline extent can only exist at file offset 0. This is
6895 * ensured by tree-checker and inline extent creation path.
6896 * Thus all members representing file offsets should be zero.
6898 ASSERT(pg_offset == 0);
6899 ASSERT(extent_start == 0);
6900 ASSERT(em->start == 0);
6903 * btrfs_extent_item_to_extent_map() should have properly
6904 * initialized em members already.
6906 * Other members are not utilized for inline extents.
6908 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6909 ASSERT(em->len == fs_info->sectorsize);
6911 ret = read_inline_extent(inode, path, page);
6918 em->orig_start = start;
6920 em->block_start = EXTENT_MAP_HOLE;
6923 btrfs_release_path(path);
6924 if (em->start > start || extent_map_end(em) <= start) {
6926 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6927 em->start, em->len, start, len);
6932 write_lock(&em_tree->lock);
6933 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6934 write_unlock(&em_tree->lock);
6936 btrfs_free_path(path);
6938 trace_btrfs_get_extent(root, inode, em);
6941 free_extent_map(em);
6942 return ERR_PTR(ret);
6947 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6948 struct btrfs_dio_data *dio_data,
6951 const u64 orig_start,
6952 const u64 block_start,
6953 const u64 block_len,
6954 const u64 orig_block_len,
6955 const u64 ram_bytes,
6958 struct extent_map *em = NULL;
6959 struct btrfs_ordered_extent *ordered;
6961 if (type != BTRFS_ORDERED_NOCOW) {
6962 em = create_io_em(inode, start, len, orig_start, block_start,
6963 block_len, orig_block_len, ram_bytes,
6964 BTRFS_COMPRESS_NONE, /* compress_type */
6969 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6970 block_start, block_len, 0,
6972 (1 << BTRFS_ORDERED_DIRECT),
6973 BTRFS_COMPRESS_NONE);
6974 if (IS_ERR(ordered)) {
6976 free_extent_map(em);
6977 btrfs_drop_extent_map_range(inode, start,
6978 start + len - 1, false);
6980 em = ERR_CAST(ordered);
6982 ASSERT(!dio_data->ordered);
6983 dio_data->ordered = ordered;
6990 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6991 struct btrfs_dio_data *dio_data,
6994 struct btrfs_root *root = inode->root;
6995 struct btrfs_fs_info *fs_info = root->fs_info;
6996 struct extent_map *em;
6997 struct btrfs_key ins;
7001 alloc_hint = get_extent_allocation_hint(inode, start, len);
7003 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7004 0, alloc_hint, &ins, 1, 1);
7005 if (ret == -EAGAIN) {
7006 ASSERT(btrfs_is_zoned(fs_info));
7007 wait_on_bit_io(&inode->root->fs_info->flags, BTRFS_FS_NEED_ZONE_FINISH,
7008 TASK_UNINTERRUPTIBLE);
7012 return ERR_PTR(ret);
7014 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7015 ins.objectid, ins.offset, ins.offset,
7016 ins.offset, BTRFS_ORDERED_REGULAR);
7017 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7019 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7025 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7027 struct btrfs_block_group *block_group;
7028 bool readonly = false;
7030 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7031 if (!block_group || block_group->ro)
7034 btrfs_put_block_group(block_group);
7039 * Check if we can do nocow write into the range [@offset, @offset + @len)
7041 * @offset: File offset
7042 * @len: The length to write, will be updated to the nocow writeable
7044 * @orig_start: (optional) Return the original file offset of the file extent
7045 * @orig_len: (optional) Return the original on-disk length of the file extent
7046 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7047 * @strict: if true, omit optimizations that might force us into unnecessary
7048 * cow. e.g., don't trust generation number.
7051 * >0 and update @len if we can do nocow write
7052 * 0 if we can't do nocow write
7053 * <0 if error happened
7055 * NOTE: This only checks the file extents, caller is responsible to wait for
7056 * any ordered extents.
7058 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7059 u64 *orig_start, u64 *orig_block_len,
7060 u64 *ram_bytes, bool nowait, bool strict)
7062 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7063 struct can_nocow_file_extent_args nocow_args = { 0 };
7064 struct btrfs_path *path;
7066 struct extent_buffer *leaf;
7067 struct btrfs_root *root = BTRFS_I(inode)->root;
7068 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7069 struct btrfs_file_extent_item *fi;
7070 struct btrfs_key key;
7073 path = btrfs_alloc_path();
7076 path->nowait = nowait;
7078 ret = btrfs_lookup_file_extent(NULL, root, path,
7079 btrfs_ino(BTRFS_I(inode)), offset, 0);
7084 if (path->slots[0] == 0) {
7085 /* can't find the item, must cow */
7092 leaf = path->nodes[0];
7093 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7094 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7095 key.type != BTRFS_EXTENT_DATA_KEY) {
7096 /* not our file or wrong item type, must cow */
7100 if (key.offset > offset) {
7101 /* Wrong offset, must cow */
7105 if (btrfs_file_extent_end(path) <= offset)
7108 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7109 found_type = btrfs_file_extent_type(leaf, fi);
7111 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7113 nocow_args.start = offset;
7114 nocow_args.end = offset + *len - 1;
7115 nocow_args.strict = strict;
7116 nocow_args.free_path = true;
7118 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7119 /* can_nocow_file_extent() has freed the path. */
7123 /* Treat errors as not being able to NOCOW. */
7129 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7132 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7133 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7136 range_end = round_up(offset + nocow_args.num_bytes,
7137 root->fs_info->sectorsize) - 1;
7138 ret = test_range_bit_exists(io_tree, offset, range_end, EXTENT_DELALLOC);
7146 *orig_start = key.offset - nocow_args.extent_offset;
7148 *orig_block_len = nocow_args.disk_num_bytes;
7150 *len = nocow_args.num_bytes;
7153 btrfs_free_path(path);
7157 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7158 struct extent_state **cached_state,
7159 unsigned int iomap_flags)
7161 const bool writing = (iomap_flags & IOMAP_WRITE);
7162 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7163 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7164 struct btrfs_ordered_extent *ordered;
7169 if (!try_lock_extent(io_tree, lockstart, lockend,
7173 lock_extent(io_tree, lockstart, lockend, cached_state);
7176 * We're concerned with the entire range that we're going to be
7177 * doing DIO to, so we need to make sure there's no ordered
7178 * extents in this range.
7180 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7181 lockend - lockstart + 1);
7184 * We need to make sure there are no buffered pages in this
7185 * range either, we could have raced between the invalidate in
7186 * generic_file_direct_write and locking the extent. The
7187 * invalidate needs to happen so that reads after a write do not
7191 (!writing || !filemap_range_has_page(inode->i_mapping,
7192 lockstart, lockend)))
7195 unlock_extent(io_tree, lockstart, lockend, cached_state);
7199 btrfs_put_ordered_extent(ordered);
7204 * If we are doing a DIO read and the ordered extent we
7205 * found is for a buffered write, we can not wait for it
7206 * to complete and retry, because if we do so we can
7207 * deadlock with concurrent buffered writes on page
7208 * locks. This happens only if our DIO read covers more
7209 * than one extent map, if at this point has already
7210 * created an ordered extent for a previous extent map
7211 * and locked its range in the inode's io tree, and a
7212 * concurrent write against that previous extent map's
7213 * range and this range started (we unlock the ranges
7214 * in the io tree only when the bios complete and
7215 * buffered writes always lock pages before attempting
7216 * to lock range in the io tree).
7219 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7220 btrfs_start_ordered_extent(ordered);
7222 ret = nowait ? -EAGAIN : -ENOTBLK;
7223 btrfs_put_ordered_extent(ordered);
7226 * We could trigger writeback for this range (and wait
7227 * for it to complete) and then invalidate the pages for
7228 * this range (through invalidate_inode_pages2_range()),
7229 * but that can lead us to a deadlock with a concurrent
7230 * call to readahead (a buffered read or a defrag call
7231 * triggered a readahead) on a page lock due to an
7232 * ordered dio extent we created before but did not have
7233 * yet a corresponding bio submitted (whence it can not
7234 * complete), which makes readahead wait for that
7235 * ordered extent to complete while holding a lock on
7238 ret = nowait ? -EAGAIN : -ENOTBLK;
7250 /* The callers of this must take lock_extent() */
7251 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7252 u64 len, u64 orig_start, u64 block_start,
7253 u64 block_len, u64 orig_block_len,
7254 u64 ram_bytes, int compress_type,
7257 struct extent_map *em;
7260 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7261 type == BTRFS_ORDERED_COMPRESSED ||
7262 type == BTRFS_ORDERED_NOCOW ||
7263 type == BTRFS_ORDERED_REGULAR);
7265 em = alloc_extent_map();
7267 return ERR_PTR(-ENOMEM);
7270 em->orig_start = orig_start;
7272 em->block_len = block_len;
7273 em->block_start = block_start;
7274 em->orig_block_len = orig_block_len;
7275 em->ram_bytes = ram_bytes;
7276 em->generation = -1;
7277 em->flags |= EXTENT_FLAG_PINNED;
7278 if (type == BTRFS_ORDERED_PREALLOC)
7279 em->flags |= EXTENT_FLAG_FILLING;
7280 else if (type == BTRFS_ORDERED_COMPRESSED)
7281 extent_map_set_compression(em, compress_type);
7283 ret = btrfs_replace_extent_map_range(inode, em, true);
7285 free_extent_map(em);
7286 return ERR_PTR(ret);
7289 /* em got 2 refs now, callers needs to do free_extent_map once. */
7294 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7295 struct inode *inode,
7296 struct btrfs_dio_data *dio_data,
7297 u64 start, u64 *lenp,
7298 unsigned int iomap_flags)
7300 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7301 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7302 struct extent_map *em = *map;
7304 u64 block_start, orig_start, orig_block_len, ram_bytes;
7305 struct btrfs_block_group *bg;
7306 bool can_nocow = false;
7307 bool space_reserved = false;
7313 * We don't allocate a new extent in the following cases
7315 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7317 * 2) The extent is marked as PREALLOC. We're good to go here and can
7318 * just use the extent.
7321 if ((em->flags & EXTENT_FLAG_PREALLOC) ||
7322 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7323 em->block_start != EXTENT_MAP_HOLE)) {
7324 if (em->flags & EXTENT_FLAG_PREALLOC)
7325 type = BTRFS_ORDERED_PREALLOC;
7327 type = BTRFS_ORDERED_NOCOW;
7328 len = min(len, em->len - (start - em->start));
7329 block_start = em->block_start + (start - em->start);
7331 if (can_nocow_extent(inode, start, &len, &orig_start,
7332 &orig_block_len, &ram_bytes, false, false) == 1) {
7333 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7341 struct extent_map *em2;
7343 /* We can NOCOW, so only need to reserve metadata space. */
7344 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7347 /* Our caller expects us to free the input extent map. */
7348 free_extent_map(em);
7350 btrfs_dec_nocow_writers(bg);
7351 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7355 space_reserved = true;
7357 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7358 orig_start, block_start,
7359 len, orig_block_len,
7361 btrfs_dec_nocow_writers(bg);
7362 if (type == BTRFS_ORDERED_PREALLOC) {
7363 free_extent_map(em);
7373 dio_data->nocow_done = true;
7375 /* Our caller expects us to free the input extent map. */
7376 free_extent_map(em);
7385 * If we could not allocate data space before locking the file
7386 * range and we can't do a NOCOW write, then we have to fail.
7388 if (!dio_data->data_space_reserved) {
7394 * We have to COW and we have already reserved data space before,
7395 * so now we reserve only metadata.
7397 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7401 space_reserved = true;
7403 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7409 len = min(len, em->len - (start - em->start));
7411 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7412 prev_len - len, true);
7416 * We have created our ordered extent, so we can now release our reservation
7417 * for an outstanding extent.
7419 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7422 * Need to update the i_size under the extent lock so buffered
7423 * readers will get the updated i_size when we unlock.
7425 if (start + len > i_size_read(inode))
7426 i_size_write(inode, start + len);
7428 if (ret && space_reserved) {
7429 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7430 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7436 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7437 loff_t length, unsigned int flags, struct iomap *iomap,
7438 struct iomap *srcmap)
7440 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7441 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7442 struct extent_map *em;
7443 struct extent_state *cached_state = NULL;
7444 struct btrfs_dio_data *dio_data = iter->private;
7445 u64 lockstart, lockend;
7446 const bool write = !!(flags & IOMAP_WRITE);
7449 const u64 data_alloc_len = length;
7450 bool unlock_extents = false;
7453 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7454 * we're NOWAIT we may submit a bio for a partial range and return
7455 * EIOCBQUEUED, which would result in an errant short read.
7457 * The best way to handle this would be to allow for partial completions
7458 * of iocb's, so we could submit the partial bio, return and fault in
7459 * the rest of the pages, and then submit the io for the rest of the
7460 * range. However we don't have that currently, so simply return
7461 * -EAGAIN at this point so that the normal path is used.
7463 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7467 * Cap the size of reads to that usually seen in buffered I/O as we need
7468 * to allocate a contiguous array for the checksums.
7471 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7474 lockend = start + len - 1;
7477 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7478 * enough if we've written compressed pages to this area, so we need to
7479 * flush the dirty pages again to make absolutely sure that any
7480 * outstanding dirty pages are on disk - the first flush only starts
7481 * compression on the data, while keeping the pages locked, so by the
7482 * time the second flush returns we know bios for the compressed pages
7483 * were submitted and finished, and the pages no longer under writeback.
7485 * If we have a NOWAIT request and we have any pages in the range that
7486 * are locked, likely due to compression still in progress, we don't want
7487 * to block on page locks. We also don't want to block on pages marked as
7488 * dirty or under writeback (same as for the non-compression case).
7489 * iomap_dio_rw() did the same check, but after that and before we got
7490 * here, mmap'ed writes may have happened or buffered reads started
7491 * (readpage() and readahead(), which lock pages), as we haven't locked
7492 * the file range yet.
7494 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7495 &BTRFS_I(inode)->runtime_flags)) {
7496 if (flags & IOMAP_NOWAIT) {
7497 if (filemap_range_needs_writeback(inode->i_mapping,
7498 lockstart, lockend))
7501 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7502 start + length - 1);
7508 memset(dio_data, 0, sizeof(*dio_data));
7511 * We always try to allocate data space and must do it before locking
7512 * the file range, to avoid deadlocks with concurrent writes to the same
7513 * range if the range has several extents and the writes don't expand the
7514 * current i_size (the inode lock is taken in shared mode). If we fail to
7515 * allocate data space here we continue and later, after locking the
7516 * file range, we fail with ENOSPC only if we figure out we can not do a
7519 if (write && !(flags & IOMAP_NOWAIT)) {
7520 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7521 &dio_data->data_reserved,
7522 start, data_alloc_len, false);
7524 dio_data->data_space_reserved = true;
7525 else if (ret && !(BTRFS_I(inode)->flags &
7526 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7531 * If this errors out it's because we couldn't invalidate pagecache for
7532 * this range and we need to fallback to buffered IO, or we are doing a
7533 * NOWAIT read/write and we need to block.
7535 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7539 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7546 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7547 * io. INLINE is special, and we could probably kludge it in here, but
7548 * it's still buffered so for safety lets just fall back to the generic
7551 * For COMPRESSED we _have_ to read the entire extent in so we can
7552 * decompress it, so there will be buffering required no matter what we
7553 * do, so go ahead and fallback to buffered.
7555 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7556 * to buffered IO. Don't blame me, this is the price we pay for using
7559 if (extent_map_is_compressed(em) ||
7560 em->block_start == EXTENT_MAP_INLINE) {
7561 free_extent_map(em);
7563 * If we are in a NOWAIT context, return -EAGAIN in order to
7564 * fallback to buffered IO. This is not only because we can
7565 * block with buffered IO (no support for NOWAIT semantics at
7566 * the moment) but also to avoid returning short reads to user
7567 * space - this happens if we were able to read some data from
7568 * previous non-compressed extents and then when we fallback to
7569 * buffered IO, at btrfs_file_read_iter() by calling
7570 * filemap_read(), we fail to fault in pages for the read buffer,
7571 * in which case filemap_read() returns a short read (the number
7572 * of bytes previously read is > 0, so it does not return -EFAULT).
7574 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7578 len = min(len, em->len - (start - em->start));
7581 * If we have a NOWAIT request and the range contains multiple extents
7582 * (or a mix of extents and holes), then we return -EAGAIN to make the
7583 * caller fallback to a context where it can do a blocking (without
7584 * NOWAIT) request. This way we avoid doing partial IO and returning
7585 * success to the caller, which is not optimal for writes and for reads
7586 * it can result in unexpected behaviour for an application.
7588 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7589 * iomap_dio_rw(), we can end up returning less data then what the caller
7590 * asked for, resulting in an unexpected, and incorrect, short read.
7591 * That is, the caller asked to read N bytes and we return less than that,
7592 * which is wrong unless we are crossing EOF. This happens if we get a
7593 * page fault error when trying to fault in pages for the buffer that is
7594 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7595 * have previously submitted bios for other extents in the range, in
7596 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7597 * those bios have completed by the time we get the page fault error,
7598 * which we return back to our caller - we should only return EIOCBQUEUED
7599 * after we have submitted bios for all the extents in the range.
7601 if ((flags & IOMAP_NOWAIT) && len < length) {
7602 free_extent_map(em);
7608 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7609 start, &len, flags);
7612 unlock_extents = true;
7613 /* Recalc len in case the new em is smaller than requested */
7614 len = min(len, em->len - (start - em->start));
7615 if (dio_data->data_space_reserved) {
7617 u64 release_len = 0;
7619 if (dio_data->nocow_done) {
7620 release_offset = start;
7621 release_len = data_alloc_len;
7622 } else if (len < data_alloc_len) {
7623 release_offset = start + len;
7624 release_len = data_alloc_len - len;
7627 if (release_len > 0)
7628 btrfs_free_reserved_data_space(BTRFS_I(inode),
7629 dio_data->data_reserved,
7635 * We need to unlock only the end area that we aren't using.
7636 * The rest is going to be unlocked by the endio routine.
7638 lockstart = start + len;
7639 if (lockstart < lockend)
7640 unlock_extents = true;
7644 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7647 free_extent_state(cached_state);
7650 * Translate extent map information to iomap.
7651 * We trim the extents (and move the addr) even though iomap code does
7652 * that, since we have locked only the parts we are performing I/O in.
7654 if ((em->block_start == EXTENT_MAP_HOLE) ||
7655 ((em->flags & EXTENT_FLAG_PREALLOC) && !write)) {
7656 iomap->addr = IOMAP_NULL_ADDR;
7657 iomap->type = IOMAP_HOLE;
7659 iomap->addr = em->block_start + (start - em->start);
7660 iomap->type = IOMAP_MAPPED;
7662 iomap->offset = start;
7663 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7664 iomap->length = len;
7665 free_extent_map(em);
7670 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7673 if (dio_data->data_space_reserved) {
7674 btrfs_free_reserved_data_space(BTRFS_I(inode),
7675 dio_data->data_reserved,
7676 start, data_alloc_len);
7677 extent_changeset_free(dio_data->data_reserved);
7683 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7684 ssize_t written, unsigned int flags, struct iomap *iomap)
7686 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7687 struct btrfs_dio_data *dio_data = iter->private;
7688 size_t submitted = dio_data->submitted;
7689 const bool write = !!(flags & IOMAP_WRITE);
7692 if (!write && (iomap->type == IOMAP_HOLE)) {
7693 /* If reading from a hole, unlock and return */
7694 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7699 if (submitted < length) {
7701 length -= submitted;
7703 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7704 pos, length, false);
7706 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7707 pos + length - 1, NULL);
7711 btrfs_put_ordered_extent(dio_data->ordered);
7712 dio_data->ordered = NULL;
7716 extent_changeset_free(dio_data->data_reserved);
7720 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7722 struct btrfs_dio_private *dip =
7723 container_of(bbio, struct btrfs_dio_private, bbio);
7724 struct btrfs_inode *inode = bbio->inode;
7725 struct bio *bio = &bbio->bio;
7727 if (bio->bi_status) {
7728 btrfs_warn(inode->root->fs_info,
7729 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7730 btrfs_ino(inode), bio->bi_opf,
7731 dip->file_offset, dip->bytes, bio->bi_status);
7734 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7735 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7736 dip->file_offset, dip->bytes,
7739 unlock_extent(&inode->io_tree, dip->file_offset,
7740 dip->file_offset + dip->bytes - 1, NULL);
7743 bbio->bio.bi_private = bbio->private;
7744 iomap_dio_bio_end_io(bio);
7747 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7750 struct btrfs_bio *bbio = btrfs_bio(bio);
7751 struct btrfs_dio_private *dip =
7752 container_of(bbio, struct btrfs_dio_private, bbio);
7753 struct btrfs_dio_data *dio_data = iter->private;
7755 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7756 btrfs_dio_end_io, bio->bi_private);
7757 bbio->inode = BTRFS_I(iter->inode);
7758 bbio->file_offset = file_offset;
7760 dip->file_offset = file_offset;
7761 dip->bytes = bio->bi_iter.bi_size;
7763 dio_data->submitted += bio->bi_iter.bi_size;
7766 * Check if we are doing a partial write. If we are, we need to split
7767 * the ordered extent to match the submitted bio. Hang on to the
7768 * remaining unfinishable ordered_extent in dio_data so that it can be
7769 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7770 * remaining pages is blocked on the outstanding ordered extent.
7772 if (iter->flags & IOMAP_WRITE) {
7775 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7777 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7778 file_offset, dip->bytes,
7780 bio->bi_status = errno_to_blk_status(ret);
7781 iomap_dio_bio_end_io(bio);
7786 btrfs_submit_bio(bbio, 0);
7789 static const struct iomap_ops btrfs_dio_iomap_ops = {
7790 .iomap_begin = btrfs_dio_iomap_begin,
7791 .iomap_end = btrfs_dio_iomap_end,
7794 static const struct iomap_dio_ops btrfs_dio_ops = {
7795 .submit_io = btrfs_dio_submit_io,
7796 .bio_set = &btrfs_dio_bioset,
7799 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7801 struct btrfs_dio_data data = { 0 };
7803 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7804 IOMAP_DIO_PARTIAL, &data, done_before);
7807 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7810 struct btrfs_dio_data data = { 0 };
7812 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7813 IOMAP_DIO_PARTIAL, &data, done_before);
7816 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7821 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7826 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7827 * file range (0 to LLONG_MAX), but that is not enough if we have
7828 * compression enabled. The first filemap_fdatawrite_range() only kicks
7829 * in the compression of data (in an async thread) and will return
7830 * before the compression is done and writeback is started. A second
7831 * filemap_fdatawrite_range() is needed to wait for the compression to
7832 * complete and writeback to start. We also need to wait for ordered
7833 * extents to complete, because our fiemap implementation uses mainly
7834 * file extent items to list the extents, searching for extent maps
7835 * only for file ranges with holes or prealloc extents to figure out
7836 * if we have delalloc in those ranges.
7838 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7839 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7844 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7847 static int btrfs_writepages(struct address_space *mapping,
7848 struct writeback_control *wbc)
7850 return extent_writepages(mapping, wbc);
7853 static void btrfs_readahead(struct readahead_control *rac)
7855 extent_readahead(rac);
7859 * For release_folio() and invalidate_folio() we have a race window where
7860 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7861 * If we continue to release/invalidate the page, we could cause use-after-free
7862 * for subpage spinlock. So this function is to spin and wait for subpage
7865 static void wait_subpage_spinlock(struct page *page)
7867 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7868 struct folio *folio = page_folio(page);
7869 struct btrfs_subpage *subpage;
7871 if (!btrfs_is_subpage(fs_info, page->mapping))
7874 ASSERT(folio_test_private(folio) && folio_get_private(folio));
7875 subpage = folio_get_private(folio);
7878 * This may look insane as we just acquire the spinlock and release it,
7879 * without doing anything. But we just want to make sure no one is
7880 * still holding the subpage spinlock.
7881 * And since the page is not dirty nor writeback, and we have page
7882 * locked, the only possible way to hold a spinlock is from the endio
7883 * function to clear page writeback.
7885 * Here we just acquire the spinlock so that all existing callers
7886 * should exit and we're safe to release/invalidate the page.
7888 spin_lock_irq(&subpage->lock);
7889 spin_unlock_irq(&subpage->lock);
7892 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7894 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7897 wait_subpage_spinlock(&folio->page);
7898 clear_page_extent_mapped(&folio->page);
7903 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7905 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7907 return __btrfs_release_folio(folio, gfp_flags);
7910 #ifdef CONFIG_MIGRATION
7911 static int btrfs_migrate_folio(struct address_space *mapping,
7912 struct folio *dst, struct folio *src,
7913 enum migrate_mode mode)
7915 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7917 if (ret != MIGRATEPAGE_SUCCESS)
7920 if (folio_test_ordered(src)) {
7921 folio_clear_ordered(src);
7922 folio_set_ordered(dst);
7925 return MIGRATEPAGE_SUCCESS;
7928 #define btrfs_migrate_folio NULL
7931 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7934 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7935 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7936 struct extent_io_tree *tree = &inode->io_tree;
7937 struct extent_state *cached_state = NULL;
7938 u64 page_start = folio_pos(folio);
7939 u64 page_end = page_start + folio_size(folio) - 1;
7941 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7944 * We have folio locked so no new ordered extent can be created on this
7945 * page, nor bio can be submitted for this folio.
7947 * But already submitted bio can still be finished on this folio.
7948 * Furthermore, endio function won't skip folio which has Ordered
7949 * (Private2) already cleared, so it's possible for endio and
7950 * invalidate_folio to do the same ordered extent accounting twice
7953 * So here we wait for any submitted bios to finish, so that we won't
7954 * do double ordered extent accounting on the same folio.
7956 folio_wait_writeback(folio);
7957 wait_subpage_spinlock(&folio->page);
7960 * For subpage case, we have call sites like
7961 * btrfs_punch_hole_lock_range() which passes range not aligned to
7963 * If the range doesn't cover the full folio, we don't need to and
7964 * shouldn't clear page extent mapped, as folio->private can still
7965 * record subpage dirty bits for other part of the range.
7967 * For cases that invalidate the full folio even the range doesn't
7968 * cover the full folio, like invalidating the last folio, we're
7969 * still safe to wait for ordered extent to finish.
7971 if (!(offset == 0 && length == folio_size(folio))) {
7972 btrfs_release_folio(folio, GFP_NOFS);
7976 if (!inode_evicting)
7977 lock_extent(tree, page_start, page_end, &cached_state);
7980 while (cur < page_end) {
7981 struct btrfs_ordered_extent *ordered;
7984 u32 extra_flags = 0;
7986 ordered = btrfs_lookup_first_ordered_range(inode, cur,
7987 page_end + 1 - cur);
7989 range_end = page_end;
7991 * No ordered extent covering this range, we are safe
7992 * to delete all extent states in the range.
7994 extra_flags = EXTENT_CLEAR_ALL_BITS;
7997 if (ordered->file_offset > cur) {
7999 * There is a range between [cur, oe->file_offset) not
8000 * covered by any ordered extent.
8001 * We are safe to delete all extent states, and handle
8002 * the ordered extent in the next iteration.
8004 range_end = ordered->file_offset - 1;
8005 extra_flags = EXTENT_CLEAR_ALL_BITS;
8009 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8011 ASSERT(range_end + 1 - cur < U32_MAX);
8012 range_len = range_end + 1 - cur;
8013 if (!btrfs_folio_test_ordered(fs_info, folio, cur, range_len)) {
8015 * If Ordered (Private2) is cleared, it means endio has
8016 * already been executed for the range.
8017 * We can't delete the extent states as
8018 * btrfs_finish_ordered_io() may still use some of them.
8022 btrfs_folio_clear_ordered(fs_info, folio, cur, range_len);
8025 * IO on this page will never be started, so we need to account
8026 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8027 * here, must leave that up for the ordered extent completion.
8029 * This will also unlock the range for incoming
8030 * btrfs_finish_ordered_io().
8032 if (!inode_evicting)
8033 clear_extent_bit(tree, cur, range_end,
8035 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8036 EXTENT_DEFRAG, &cached_state);
8038 spin_lock_irq(&inode->ordered_tree_lock);
8039 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8040 ordered->truncated_len = min(ordered->truncated_len,
8041 cur - ordered->file_offset);
8042 spin_unlock_irq(&inode->ordered_tree_lock);
8045 * If the ordered extent has finished, we're safe to delete all
8046 * the extent states of the range, otherwise
8047 * btrfs_finish_ordered_io() will get executed by endio for
8048 * other pages, so we can't delete extent states.
8050 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8051 cur, range_end + 1 - cur)) {
8052 btrfs_finish_ordered_io(ordered);
8054 * The ordered extent has finished, now we're again
8055 * safe to delete all extent states of the range.
8057 extra_flags = EXTENT_CLEAR_ALL_BITS;
8061 btrfs_put_ordered_extent(ordered);
8063 * Qgroup reserved space handler
8064 * Sector(s) here will be either:
8066 * 1) Already written to disk or bio already finished
8067 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8068 * Qgroup will be handled by its qgroup_record then.
8069 * btrfs_qgroup_free_data() call will do nothing here.
8071 * 2) Not written to disk yet
8072 * Then btrfs_qgroup_free_data() call will clear the
8073 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8074 * reserved data space.
8075 * Since the IO will never happen for this page.
8077 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur, NULL);
8078 if (!inode_evicting) {
8079 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8080 EXTENT_DELALLOC | EXTENT_UPTODATE |
8081 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8082 extra_flags, &cached_state);
8084 cur = range_end + 1;
8087 * We have iterated through all ordered extents of the page, the page
8088 * should not have Ordered (Private2) anymore, or the above iteration
8089 * did something wrong.
8091 ASSERT(!folio_test_ordered(folio));
8092 btrfs_folio_clear_checked(fs_info, folio, folio_pos(folio), folio_size(folio));
8093 if (!inode_evicting)
8094 __btrfs_release_folio(folio, GFP_NOFS);
8095 clear_page_extent_mapped(&folio->page);
8099 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8100 * called from a page fault handler when a page is first dirtied. Hence we must
8101 * be careful to check for EOF conditions here. We set the page up correctly
8102 * for a written page which means we get ENOSPC checking when writing into
8103 * holes and correct delalloc and unwritten extent mapping on filesystems that
8104 * support these features.
8106 * We are not allowed to take the i_mutex here so we have to play games to
8107 * protect against truncate races as the page could now be beyond EOF. Because
8108 * truncate_setsize() writes the inode size before removing pages, once we have
8109 * the page lock we can determine safely if the page is beyond EOF. If it is not
8110 * beyond EOF, then the page is guaranteed safe against truncation until we
8113 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8115 struct page *page = vmf->page;
8116 struct folio *folio = page_folio(page);
8117 struct inode *inode = file_inode(vmf->vma->vm_file);
8118 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8119 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8120 struct btrfs_ordered_extent *ordered;
8121 struct extent_state *cached_state = NULL;
8122 struct extent_changeset *data_reserved = NULL;
8123 unsigned long zero_start;
8133 ASSERT(folio_order(folio) == 0);
8135 reserved_space = PAGE_SIZE;
8137 sb_start_pagefault(inode->i_sb);
8138 page_start = page_offset(page);
8139 page_end = page_start + PAGE_SIZE - 1;
8143 * Reserving delalloc space after obtaining the page lock can lead to
8144 * deadlock. For example, if a dirty page is locked by this function
8145 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8146 * dirty page write out, then the btrfs_writepages() function could
8147 * end up waiting indefinitely to get a lock on the page currently
8148 * being processed by btrfs_page_mkwrite() function.
8150 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8151 page_start, reserved_space);
8153 ret2 = file_update_time(vmf->vma->vm_file);
8157 ret = vmf_error(ret2);
8163 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8165 down_read(&BTRFS_I(inode)->i_mmap_lock);
8167 size = i_size_read(inode);
8169 if ((page->mapping != inode->i_mapping) ||
8170 (page_start >= size)) {
8171 /* page got truncated out from underneath us */
8174 wait_on_page_writeback(page);
8176 lock_extent(io_tree, page_start, page_end, &cached_state);
8177 ret2 = set_page_extent_mapped(page);
8179 ret = vmf_error(ret2);
8180 unlock_extent(io_tree, page_start, page_end, &cached_state);
8185 * we can't set the delalloc bits if there are pending ordered
8186 * extents. Drop our locks and wait for them to finish
8188 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8191 unlock_extent(io_tree, page_start, page_end, &cached_state);
8193 up_read(&BTRFS_I(inode)->i_mmap_lock);
8194 btrfs_start_ordered_extent(ordered);
8195 btrfs_put_ordered_extent(ordered);
8199 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8200 reserved_space = round_up(size - page_start,
8201 fs_info->sectorsize);
8202 if (reserved_space < PAGE_SIZE) {
8203 end = page_start + reserved_space - 1;
8204 btrfs_delalloc_release_space(BTRFS_I(inode),
8205 data_reserved, page_start,
8206 PAGE_SIZE - reserved_space, true);
8211 * page_mkwrite gets called when the page is firstly dirtied after it's
8212 * faulted in, but write(2) could also dirty a page and set delalloc
8213 * bits, thus in this case for space account reason, we still need to
8214 * clear any delalloc bits within this page range since we have to
8215 * reserve data&meta space before lock_page() (see above comments).
8217 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8218 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8219 EXTENT_DEFRAG, &cached_state);
8221 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8224 unlock_extent(io_tree, page_start, page_end, &cached_state);
8225 ret = VM_FAULT_SIGBUS;
8229 /* page is wholly or partially inside EOF */
8230 if (page_start + PAGE_SIZE > size)
8231 zero_start = offset_in_page(size);
8233 zero_start = PAGE_SIZE;
8235 if (zero_start != PAGE_SIZE)
8236 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8238 btrfs_folio_clear_checked(fs_info, folio, page_start, PAGE_SIZE);
8239 btrfs_folio_set_dirty(fs_info, folio, page_start, end + 1 - page_start);
8240 btrfs_folio_set_uptodate(fs_info, folio, page_start, end + 1 - page_start);
8242 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8244 unlock_extent(io_tree, page_start, page_end, &cached_state);
8245 up_read(&BTRFS_I(inode)->i_mmap_lock);
8247 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8248 sb_end_pagefault(inode->i_sb);
8249 extent_changeset_free(data_reserved);
8250 return VM_FAULT_LOCKED;
8254 up_read(&BTRFS_I(inode)->i_mmap_lock);
8256 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8257 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8258 reserved_space, (ret != 0));
8260 sb_end_pagefault(inode->i_sb);
8261 extent_changeset_free(data_reserved);
8265 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8267 struct btrfs_truncate_control control = {
8269 .ino = btrfs_ino(inode),
8270 .min_type = BTRFS_EXTENT_DATA_KEY,
8271 .clear_extent_range = true,
8273 struct btrfs_root *root = inode->root;
8274 struct btrfs_fs_info *fs_info = root->fs_info;
8275 struct btrfs_block_rsv *rsv;
8277 struct btrfs_trans_handle *trans;
8278 u64 mask = fs_info->sectorsize - 1;
8279 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8281 if (!skip_writeback) {
8282 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8283 inode->vfs_inode.i_size & (~mask),
8290 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8291 * things going on here:
8293 * 1) We need to reserve space to update our inode.
8295 * 2) We need to have something to cache all the space that is going to
8296 * be free'd up by the truncate operation, but also have some slack
8297 * space reserved in case it uses space during the truncate (thank you
8298 * very much snapshotting).
8300 * And we need these to be separate. The fact is we can use a lot of
8301 * space doing the truncate, and we have no earthly idea how much space
8302 * we will use, so we need the truncate reservation to be separate so it
8303 * doesn't end up using space reserved for updating the inode. We also
8304 * need to be able to stop the transaction and start a new one, which
8305 * means we need to be able to update the inode several times, and we
8306 * have no idea of knowing how many times that will be, so we can't just
8307 * reserve 1 item for the entirety of the operation, so that has to be
8308 * done separately as well.
8310 * So that leaves us with
8312 * 1) rsv - for the truncate reservation, which we will steal from the
8313 * transaction reservation.
8314 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8315 * updating the inode.
8317 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8320 rsv->size = min_size;
8321 rsv->failfast = true;
8324 * 1 for the truncate slack space
8325 * 1 for updating the inode.
8327 trans = btrfs_start_transaction(root, 2);
8328 if (IS_ERR(trans)) {
8329 ret = PTR_ERR(trans);
8333 /* Migrate the slack space for the truncate to our reserve */
8334 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8337 * We have reserved 2 metadata units when we started the transaction and
8338 * min_size matches 1 unit, so this should never fail, but if it does,
8339 * it's not critical we just fail truncation.
8342 btrfs_end_transaction(trans);
8346 trans->block_rsv = rsv;
8349 struct extent_state *cached_state = NULL;
8350 const u64 new_size = inode->vfs_inode.i_size;
8351 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8353 control.new_size = new_size;
8354 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8356 * We want to drop from the next block forward in case this new
8357 * size is not block aligned since we will be keeping the last
8358 * block of the extent just the way it is.
8360 btrfs_drop_extent_map_range(inode,
8361 ALIGN(new_size, fs_info->sectorsize),
8364 ret = btrfs_truncate_inode_items(trans, root, &control);
8366 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8367 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8369 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8371 trans->block_rsv = &fs_info->trans_block_rsv;
8372 if (ret != -ENOSPC && ret != -EAGAIN)
8375 ret = btrfs_update_inode(trans, inode);
8379 btrfs_end_transaction(trans);
8380 btrfs_btree_balance_dirty(fs_info);
8382 trans = btrfs_start_transaction(root, 2);
8383 if (IS_ERR(trans)) {
8384 ret = PTR_ERR(trans);
8389 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8390 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8391 rsv, min_size, false);
8393 * We have reserved 2 metadata units when we started the
8394 * transaction and min_size matches 1 unit, so this should never
8395 * fail, but if it does, it's not critical we just fail truncation.
8400 trans->block_rsv = rsv;
8404 * We can't call btrfs_truncate_block inside a trans handle as we could
8405 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8406 * know we've truncated everything except the last little bit, and can
8407 * do btrfs_truncate_block and then update the disk_i_size.
8409 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8410 btrfs_end_transaction(trans);
8411 btrfs_btree_balance_dirty(fs_info);
8413 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8416 trans = btrfs_start_transaction(root, 1);
8417 if (IS_ERR(trans)) {
8418 ret = PTR_ERR(trans);
8421 btrfs_inode_safe_disk_i_size_write(inode, 0);
8427 trans->block_rsv = &fs_info->trans_block_rsv;
8428 ret2 = btrfs_update_inode(trans, inode);
8432 ret2 = btrfs_end_transaction(trans);
8435 btrfs_btree_balance_dirty(fs_info);
8438 btrfs_free_block_rsv(fs_info, rsv);
8440 * So if we truncate and then write and fsync we normally would just
8441 * write the extents that changed, which is a problem if we need to
8442 * first truncate that entire inode. So set this flag so we write out
8443 * all of the extents in the inode to the sync log so we're completely
8446 * If no extents were dropped or trimmed we don't need to force the next
8447 * fsync to truncate all the inode's items from the log and re-log them
8448 * all. This means the truncate operation did not change the file size,
8449 * or changed it to a smaller size but there was only an implicit hole
8450 * between the old i_size and the new i_size, and there were no prealloc
8451 * extents beyond i_size to drop.
8453 if (control.extents_found > 0)
8454 btrfs_set_inode_full_sync(inode);
8459 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8462 struct inode *inode;
8464 inode = new_inode(dir->i_sb);
8467 * Subvolumes don't inherit the sgid bit or the parent's gid if
8468 * the parent's sgid bit is set. This is probably a bug.
8470 inode_init_owner(idmap, inode, NULL,
8471 S_IFDIR | (~current_umask() & S_IRWXUGO));
8472 inode->i_op = &btrfs_dir_inode_operations;
8473 inode->i_fop = &btrfs_dir_file_operations;
8478 struct inode *btrfs_alloc_inode(struct super_block *sb)
8480 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8481 struct btrfs_inode *ei;
8482 struct inode *inode;
8483 struct extent_io_tree *file_extent_tree = NULL;
8485 /* Self tests may pass a NULL fs_info. */
8486 if (fs_info && !btrfs_fs_incompat(fs_info, NO_HOLES)) {
8487 file_extent_tree = kmalloc(sizeof(struct extent_io_tree), GFP_KERNEL);
8488 if (!file_extent_tree)
8492 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8494 kfree(file_extent_tree);
8501 ei->last_sub_trans = 0;
8502 ei->logged_trans = 0;
8503 ei->delalloc_bytes = 0;
8504 ei->new_delalloc_bytes = 0;
8505 ei->defrag_bytes = 0;
8506 ei->disk_i_size = 0;
8510 ei->index_cnt = (u64)-1;
8512 ei->last_unlink_trans = 0;
8513 ei->last_reflink_trans = 0;
8514 ei->last_log_commit = 0;
8516 spin_lock_init(&ei->lock);
8517 ei->outstanding_extents = 0;
8518 if (sb->s_magic != BTRFS_TEST_MAGIC)
8519 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8520 BTRFS_BLOCK_RSV_DELALLOC);
8521 ei->runtime_flags = 0;
8522 ei->prop_compress = BTRFS_COMPRESS_NONE;
8523 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8525 ei->delayed_node = NULL;
8527 ei->i_otime_sec = 0;
8528 ei->i_otime_nsec = 0;
8530 inode = &ei->vfs_inode;
8531 extent_map_tree_init(&ei->extent_tree);
8533 /* This io tree sets the valid inode. */
8534 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8535 ei->io_tree.inode = ei;
8537 ei->file_extent_tree = file_extent_tree;
8538 if (file_extent_tree) {
8539 extent_io_tree_init(fs_info, ei->file_extent_tree,
8540 IO_TREE_INODE_FILE_EXTENT);
8541 /* Lockdep class is set only for the file extent tree. */
8542 lockdep_set_class(&ei->file_extent_tree->lock, &file_extent_tree_class);
8544 mutex_init(&ei->log_mutex);
8545 spin_lock_init(&ei->ordered_tree_lock);
8546 ei->ordered_tree = RB_ROOT;
8547 ei->ordered_tree_last = NULL;
8548 INIT_LIST_HEAD(&ei->delalloc_inodes);
8549 INIT_LIST_HEAD(&ei->delayed_iput);
8550 RB_CLEAR_NODE(&ei->rb_node);
8551 init_rwsem(&ei->i_mmap_lock);
8556 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8557 void btrfs_test_destroy_inode(struct inode *inode)
8559 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8560 kfree(BTRFS_I(inode)->file_extent_tree);
8561 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8565 void btrfs_free_inode(struct inode *inode)
8567 kfree(BTRFS_I(inode)->file_extent_tree);
8568 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8571 void btrfs_destroy_inode(struct inode *vfs_inode)
8573 struct btrfs_ordered_extent *ordered;
8574 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8575 struct btrfs_root *root = inode->root;
8576 bool freespace_inode;
8578 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8579 WARN_ON(vfs_inode->i_data.nrpages);
8580 WARN_ON(inode->block_rsv.reserved);
8581 WARN_ON(inode->block_rsv.size);
8582 WARN_ON(inode->outstanding_extents);
8583 if (!S_ISDIR(vfs_inode->i_mode)) {
8584 WARN_ON(inode->delalloc_bytes);
8585 WARN_ON(inode->new_delalloc_bytes);
8587 WARN_ON(inode->csum_bytes);
8588 WARN_ON(inode->defrag_bytes);
8591 * This can happen where we create an inode, but somebody else also
8592 * created the same inode and we need to destroy the one we already
8599 * If this is a free space inode do not take the ordered extents lockdep
8602 freespace_inode = btrfs_is_free_space_inode(inode);
8605 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8609 btrfs_err(root->fs_info,
8610 "found ordered extent %llu %llu on inode cleanup",
8611 ordered->file_offset, ordered->num_bytes);
8613 if (!freespace_inode)
8614 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8616 btrfs_remove_ordered_extent(inode, ordered);
8617 btrfs_put_ordered_extent(ordered);
8618 btrfs_put_ordered_extent(ordered);
8621 btrfs_qgroup_check_reserved_leak(inode);
8622 inode_tree_del(inode);
8623 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8624 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8625 btrfs_put_root(inode->root);
8628 int btrfs_drop_inode(struct inode *inode)
8630 struct btrfs_root *root = BTRFS_I(inode)->root;
8635 /* the snap/subvol tree is on deleting */
8636 if (btrfs_root_refs(&root->root_item) == 0)
8639 return generic_drop_inode(inode);
8642 static void init_once(void *foo)
8644 struct btrfs_inode *ei = foo;
8646 inode_init_once(&ei->vfs_inode);
8649 void __cold btrfs_destroy_cachep(void)
8652 * Make sure all delayed rcu free inodes are flushed before we
8656 bioset_exit(&btrfs_dio_bioset);
8657 kmem_cache_destroy(btrfs_inode_cachep);
8660 int __init btrfs_init_cachep(void)
8662 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8663 sizeof(struct btrfs_inode), 0,
8664 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8666 if (!btrfs_inode_cachep)
8669 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8670 offsetof(struct btrfs_dio_private, bbio.bio),
8676 btrfs_destroy_cachep();
8680 static int btrfs_getattr(struct mnt_idmap *idmap,
8681 const struct path *path, struct kstat *stat,
8682 u32 request_mask, unsigned int flags)
8686 struct inode *inode = d_inode(path->dentry);
8687 u32 blocksize = inode->i_sb->s_blocksize;
8688 u32 bi_flags = BTRFS_I(inode)->flags;
8689 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8691 stat->result_mask |= STATX_BTIME;
8692 stat->btime.tv_sec = BTRFS_I(inode)->i_otime_sec;
8693 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime_nsec;
8694 if (bi_flags & BTRFS_INODE_APPEND)
8695 stat->attributes |= STATX_ATTR_APPEND;
8696 if (bi_flags & BTRFS_INODE_COMPRESS)
8697 stat->attributes |= STATX_ATTR_COMPRESSED;
8698 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8699 stat->attributes |= STATX_ATTR_IMMUTABLE;
8700 if (bi_flags & BTRFS_INODE_NODUMP)
8701 stat->attributes |= STATX_ATTR_NODUMP;
8702 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8703 stat->attributes |= STATX_ATTR_VERITY;
8705 stat->attributes_mask |= (STATX_ATTR_APPEND |
8706 STATX_ATTR_COMPRESSED |
8707 STATX_ATTR_IMMUTABLE |
8710 generic_fillattr(idmap, request_mask, inode, stat);
8711 stat->dev = BTRFS_I(inode)->root->anon_dev;
8713 spin_lock(&BTRFS_I(inode)->lock);
8714 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8715 inode_bytes = inode_get_bytes(inode);
8716 spin_unlock(&BTRFS_I(inode)->lock);
8717 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8718 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8722 static int btrfs_rename_exchange(struct inode *old_dir,
8723 struct dentry *old_dentry,
8724 struct inode *new_dir,
8725 struct dentry *new_dentry)
8727 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8728 struct btrfs_trans_handle *trans;
8729 unsigned int trans_num_items;
8730 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8731 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8732 struct inode *new_inode = new_dentry->d_inode;
8733 struct inode *old_inode = old_dentry->d_inode;
8734 struct btrfs_rename_ctx old_rename_ctx;
8735 struct btrfs_rename_ctx new_rename_ctx;
8736 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8737 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8742 bool need_abort = false;
8743 struct fscrypt_name old_fname, new_fname;
8744 struct fscrypt_str *old_name, *new_name;
8747 * For non-subvolumes allow exchange only within one subvolume, in the
8748 * same inode namespace. Two subvolumes (represented as directory) can
8749 * be exchanged as they're a logical link and have a fixed inode number.
8752 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8753 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8756 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8760 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8762 fscrypt_free_filename(&old_fname);
8766 old_name = &old_fname.disk_name;
8767 new_name = &new_fname.disk_name;
8769 /* close the race window with snapshot create/destroy ioctl */
8770 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8771 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8772 down_read(&fs_info->subvol_sem);
8776 * 1 to remove old dir item
8777 * 1 to remove old dir index
8778 * 1 to add new dir item
8779 * 1 to add new dir index
8780 * 1 to update parent inode
8782 * If the parents are the same, we only need to account for one
8784 trans_num_items = (old_dir == new_dir ? 9 : 10);
8785 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8787 * 1 to remove old root ref
8788 * 1 to remove old root backref
8789 * 1 to add new root ref
8790 * 1 to add new root backref
8792 trans_num_items += 4;
8795 * 1 to update inode item
8796 * 1 to remove old inode ref
8797 * 1 to add new inode ref
8799 trans_num_items += 3;
8801 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8802 trans_num_items += 4;
8804 trans_num_items += 3;
8805 trans = btrfs_start_transaction(root, trans_num_items);
8806 if (IS_ERR(trans)) {
8807 ret = PTR_ERR(trans);
8812 ret = btrfs_record_root_in_trans(trans, dest);
8818 * We need to find a free sequence number both in the source and
8819 * in the destination directory for the exchange.
8821 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8824 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8828 BTRFS_I(old_inode)->dir_index = 0ULL;
8829 BTRFS_I(new_inode)->dir_index = 0ULL;
8831 /* Reference for the source. */
8832 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8833 /* force full log commit if subvolume involved. */
8834 btrfs_set_log_full_commit(trans);
8836 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8837 btrfs_ino(BTRFS_I(new_dir)),
8844 /* And now for the dest. */
8845 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8846 /* force full log commit if subvolume involved. */
8847 btrfs_set_log_full_commit(trans);
8849 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8850 btrfs_ino(BTRFS_I(old_dir)),
8854 btrfs_abort_transaction(trans, ret);
8859 /* Update inode version and ctime/mtime. */
8860 inode_inc_iversion(old_dir);
8861 inode_inc_iversion(new_dir);
8862 inode_inc_iversion(old_inode);
8863 inode_inc_iversion(new_inode);
8864 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8866 if (old_dentry->d_parent != new_dentry->d_parent) {
8867 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8868 BTRFS_I(old_inode), true);
8869 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8870 BTRFS_I(new_inode), true);
8873 /* src is a subvolume */
8874 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8875 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8876 } else { /* src is an inode */
8877 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8878 BTRFS_I(old_dentry->d_inode),
8879 old_name, &old_rename_ctx);
8881 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
8884 btrfs_abort_transaction(trans, ret);
8888 /* dest is a subvolume */
8889 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8890 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8891 } else { /* dest is an inode */
8892 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8893 BTRFS_I(new_dentry->d_inode),
8894 new_name, &new_rename_ctx);
8896 ret = btrfs_update_inode(trans, BTRFS_I(new_inode));
8899 btrfs_abort_transaction(trans, ret);
8903 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8904 new_name, 0, old_idx);
8906 btrfs_abort_transaction(trans, ret);
8910 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8911 old_name, 0, new_idx);
8913 btrfs_abort_transaction(trans, ret);
8917 if (old_inode->i_nlink == 1)
8918 BTRFS_I(old_inode)->dir_index = old_idx;
8919 if (new_inode->i_nlink == 1)
8920 BTRFS_I(new_inode)->dir_index = new_idx;
8923 * Now pin the logs of the roots. We do it to ensure that no other task
8924 * can sync the logs while we are in progress with the rename, because
8925 * that could result in an inconsistency in case any of the inodes that
8926 * are part of this rename operation were logged before.
8928 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8929 btrfs_pin_log_trans(root);
8930 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8931 btrfs_pin_log_trans(dest);
8933 /* Do the log updates for all inodes. */
8934 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8935 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8936 old_rename_ctx.index, new_dentry->d_parent);
8937 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8938 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8939 new_rename_ctx.index, old_dentry->d_parent);
8941 /* Now unpin the logs. */
8942 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8943 btrfs_end_log_trans(root);
8944 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8945 btrfs_end_log_trans(dest);
8947 ret2 = btrfs_end_transaction(trans);
8948 ret = ret ? ret : ret2;
8950 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8951 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8952 up_read(&fs_info->subvol_sem);
8954 fscrypt_free_filename(&new_fname);
8955 fscrypt_free_filename(&old_fname);
8959 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8962 struct inode *inode;
8964 inode = new_inode(dir->i_sb);
8966 inode_init_owner(idmap, inode, dir,
8967 S_IFCHR | WHITEOUT_MODE);
8968 inode->i_op = &btrfs_special_inode_operations;
8969 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8974 static int btrfs_rename(struct mnt_idmap *idmap,
8975 struct inode *old_dir, struct dentry *old_dentry,
8976 struct inode *new_dir, struct dentry *new_dentry,
8979 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8980 struct btrfs_new_inode_args whiteout_args = {
8982 .dentry = old_dentry,
8984 struct btrfs_trans_handle *trans;
8985 unsigned int trans_num_items;
8986 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8987 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8988 struct inode *new_inode = d_inode(new_dentry);
8989 struct inode *old_inode = d_inode(old_dentry);
8990 struct btrfs_rename_ctx rename_ctx;
8994 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8995 struct fscrypt_name old_fname, new_fname;
8997 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9000 /* we only allow rename subvolume link between subvolumes */
9001 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9004 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9005 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9008 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9009 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9012 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9016 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9018 fscrypt_free_filename(&old_fname);
9022 /* check for collisions, even if the name isn't there */
9023 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9025 if (ret == -EEXIST) {
9027 * eexist without a new_inode */
9028 if (WARN_ON(!new_inode)) {
9029 goto out_fscrypt_names;
9032 /* maybe -EOVERFLOW */
9033 goto out_fscrypt_names;
9039 * we're using rename to replace one file with another. Start IO on it
9040 * now so we don't add too much work to the end of the transaction
9042 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9043 filemap_flush(old_inode->i_mapping);
9045 if (flags & RENAME_WHITEOUT) {
9046 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9047 if (!whiteout_args.inode) {
9049 goto out_fscrypt_names;
9051 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9053 goto out_whiteout_inode;
9055 /* 1 to update the old parent inode. */
9056 trans_num_items = 1;
9059 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9060 /* Close the race window with snapshot create/destroy ioctl */
9061 down_read(&fs_info->subvol_sem);
9063 * 1 to remove old root ref
9064 * 1 to remove old root backref
9065 * 1 to add new root ref
9066 * 1 to add new root backref
9068 trans_num_items += 4;
9072 * 1 to remove old inode ref
9073 * 1 to add new inode ref
9075 trans_num_items += 3;
9078 * 1 to remove old dir item
9079 * 1 to remove old dir index
9080 * 1 to add new dir item
9081 * 1 to add new dir index
9083 trans_num_items += 4;
9084 /* 1 to update new parent inode if it's not the same as the old parent */
9085 if (new_dir != old_dir)
9090 * 1 to remove inode ref
9091 * 1 to remove dir item
9092 * 1 to remove dir index
9093 * 1 to possibly add orphan item
9095 trans_num_items += 5;
9097 trans = btrfs_start_transaction(root, trans_num_items);
9098 if (IS_ERR(trans)) {
9099 ret = PTR_ERR(trans);
9104 ret = btrfs_record_root_in_trans(trans, dest);
9109 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9113 BTRFS_I(old_inode)->dir_index = 0ULL;
9114 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9115 /* force full log commit if subvolume involved. */
9116 btrfs_set_log_full_commit(trans);
9118 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9119 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9125 inode_inc_iversion(old_dir);
9126 inode_inc_iversion(new_dir);
9127 inode_inc_iversion(old_inode);
9128 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9130 if (old_dentry->d_parent != new_dentry->d_parent)
9131 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9132 BTRFS_I(old_inode), true);
9134 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9135 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9137 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9138 BTRFS_I(d_inode(old_dentry)),
9139 &old_fname.disk_name, &rename_ctx);
9141 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
9144 btrfs_abort_transaction(trans, ret);
9149 inode_inc_iversion(new_inode);
9150 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9151 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9152 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9153 BUG_ON(new_inode->i_nlink == 0);
9155 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9156 BTRFS_I(d_inode(new_dentry)),
9157 &new_fname.disk_name);
9159 if (!ret && new_inode->i_nlink == 0)
9160 ret = btrfs_orphan_add(trans,
9161 BTRFS_I(d_inode(new_dentry)));
9163 btrfs_abort_transaction(trans, ret);
9168 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9169 &new_fname.disk_name, 0, index);
9171 btrfs_abort_transaction(trans, ret);
9175 if (old_inode->i_nlink == 1)
9176 BTRFS_I(old_inode)->dir_index = index;
9178 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9179 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9180 rename_ctx.index, new_dentry->d_parent);
9182 if (flags & RENAME_WHITEOUT) {
9183 ret = btrfs_create_new_inode(trans, &whiteout_args);
9185 btrfs_abort_transaction(trans, ret);
9188 unlock_new_inode(whiteout_args.inode);
9189 iput(whiteout_args.inode);
9190 whiteout_args.inode = NULL;
9194 ret2 = btrfs_end_transaction(trans);
9195 ret = ret ? ret : ret2;
9197 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9198 up_read(&fs_info->subvol_sem);
9199 if (flags & RENAME_WHITEOUT)
9200 btrfs_new_inode_args_destroy(&whiteout_args);
9202 if (flags & RENAME_WHITEOUT)
9203 iput(whiteout_args.inode);
9205 fscrypt_free_filename(&old_fname);
9206 fscrypt_free_filename(&new_fname);
9210 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9211 struct dentry *old_dentry, struct inode *new_dir,
9212 struct dentry *new_dentry, unsigned int flags)
9216 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9219 if (flags & RENAME_EXCHANGE)
9220 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9223 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9226 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9231 struct btrfs_delalloc_work {
9232 struct inode *inode;
9233 struct completion completion;
9234 struct list_head list;
9235 struct btrfs_work work;
9238 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9240 struct btrfs_delalloc_work *delalloc_work;
9241 struct inode *inode;
9243 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9245 inode = delalloc_work->inode;
9246 filemap_flush(inode->i_mapping);
9247 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9248 &BTRFS_I(inode)->runtime_flags))
9249 filemap_flush(inode->i_mapping);
9252 complete(&delalloc_work->completion);
9255 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9257 struct btrfs_delalloc_work *work;
9259 work = kmalloc(sizeof(*work), GFP_NOFS);
9263 init_completion(&work->completion);
9264 INIT_LIST_HEAD(&work->list);
9265 work->inode = inode;
9266 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL);
9272 * some fairly slow code that needs optimization. This walks the list
9273 * of all the inodes with pending delalloc and forces them to disk.
9275 static int start_delalloc_inodes(struct btrfs_root *root,
9276 struct writeback_control *wbc, bool snapshot,
9277 bool in_reclaim_context)
9279 struct btrfs_inode *binode;
9280 struct inode *inode;
9281 struct btrfs_delalloc_work *work, *next;
9285 bool full_flush = wbc->nr_to_write == LONG_MAX;
9287 mutex_lock(&root->delalloc_mutex);
9288 spin_lock(&root->delalloc_lock);
9289 list_splice_init(&root->delalloc_inodes, &splice);
9290 while (!list_empty(&splice)) {
9291 binode = list_entry(splice.next, struct btrfs_inode,
9294 list_move_tail(&binode->delalloc_inodes,
9295 &root->delalloc_inodes);
9297 if (in_reclaim_context &&
9298 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9301 inode = igrab(&binode->vfs_inode);
9303 cond_resched_lock(&root->delalloc_lock);
9306 spin_unlock(&root->delalloc_lock);
9309 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9310 &binode->runtime_flags);
9312 work = btrfs_alloc_delalloc_work(inode);
9318 list_add_tail(&work->list, &works);
9319 btrfs_queue_work(root->fs_info->flush_workers,
9322 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9323 btrfs_add_delayed_iput(BTRFS_I(inode));
9324 if (ret || wbc->nr_to_write <= 0)
9328 spin_lock(&root->delalloc_lock);
9330 spin_unlock(&root->delalloc_lock);
9333 list_for_each_entry_safe(work, next, &works, list) {
9334 list_del_init(&work->list);
9335 wait_for_completion(&work->completion);
9339 if (!list_empty(&splice)) {
9340 spin_lock(&root->delalloc_lock);
9341 list_splice_tail(&splice, &root->delalloc_inodes);
9342 spin_unlock(&root->delalloc_lock);
9344 mutex_unlock(&root->delalloc_mutex);
9348 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9350 struct writeback_control wbc = {
9351 .nr_to_write = LONG_MAX,
9352 .sync_mode = WB_SYNC_NONE,
9354 .range_end = LLONG_MAX,
9356 struct btrfs_fs_info *fs_info = root->fs_info;
9358 if (BTRFS_FS_ERROR(fs_info))
9361 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9364 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9365 bool in_reclaim_context)
9367 struct writeback_control wbc = {
9369 .sync_mode = WB_SYNC_NONE,
9371 .range_end = LLONG_MAX,
9373 struct btrfs_root *root;
9377 if (BTRFS_FS_ERROR(fs_info))
9380 mutex_lock(&fs_info->delalloc_root_mutex);
9381 spin_lock(&fs_info->delalloc_root_lock);
9382 list_splice_init(&fs_info->delalloc_roots, &splice);
9383 while (!list_empty(&splice)) {
9385 * Reset nr_to_write here so we know that we're doing a full
9389 wbc.nr_to_write = LONG_MAX;
9391 root = list_first_entry(&splice, struct btrfs_root,
9393 root = btrfs_grab_root(root);
9395 list_move_tail(&root->delalloc_root,
9396 &fs_info->delalloc_roots);
9397 spin_unlock(&fs_info->delalloc_root_lock);
9399 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9400 btrfs_put_root(root);
9401 if (ret < 0 || wbc.nr_to_write <= 0)
9403 spin_lock(&fs_info->delalloc_root_lock);
9405 spin_unlock(&fs_info->delalloc_root_lock);
9409 if (!list_empty(&splice)) {
9410 spin_lock(&fs_info->delalloc_root_lock);
9411 list_splice_tail(&splice, &fs_info->delalloc_roots);
9412 spin_unlock(&fs_info->delalloc_root_lock);
9414 mutex_unlock(&fs_info->delalloc_root_mutex);
9418 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9419 struct dentry *dentry, const char *symname)
9421 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9422 struct btrfs_trans_handle *trans;
9423 struct btrfs_root *root = BTRFS_I(dir)->root;
9424 struct btrfs_path *path;
9425 struct btrfs_key key;
9426 struct inode *inode;
9427 struct btrfs_new_inode_args new_inode_args = {
9431 unsigned int trans_num_items;
9436 struct btrfs_file_extent_item *ei;
9437 struct extent_buffer *leaf;
9439 name_len = strlen(symname);
9440 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9441 return -ENAMETOOLONG;
9443 inode = new_inode(dir->i_sb);
9446 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9447 inode->i_op = &btrfs_symlink_inode_operations;
9448 inode_nohighmem(inode);
9449 inode->i_mapping->a_ops = &btrfs_aops;
9450 btrfs_i_size_write(BTRFS_I(inode), name_len);
9451 inode_set_bytes(inode, name_len);
9453 new_inode_args.inode = inode;
9454 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9457 /* 1 additional item for the inline extent */
9460 trans = btrfs_start_transaction(root, trans_num_items);
9461 if (IS_ERR(trans)) {
9462 err = PTR_ERR(trans);
9463 goto out_new_inode_args;
9466 err = btrfs_create_new_inode(trans, &new_inode_args);
9470 path = btrfs_alloc_path();
9473 btrfs_abort_transaction(trans, err);
9474 discard_new_inode(inode);
9478 key.objectid = btrfs_ino(BTRFS_I(inode));
9480 key.type = BTRFS_EXTENT_DATA_KEY;
9481 datasize = btrfs_file_extent_calc_inline_size(name_len);
9482 err = btrfs_insert_empty_item(trans, root, path, &key,
9485 btrfs_abort_transaction(trans, err);
9486 btrfs_free_path(path);
9487 discard_new_inode(inode);
9491 leaf = path->nodes[0];
9492 ei = btrfs_item_ptr(leaf, path->slots[0],
9493 struct btrfs_file_extent_item);
9494 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9495 btrfs_set_file_extent_type(leaf, ei,
9496 BTRFS_FILE_EXTENT_INLINE);
9497 btrfs_set_file_extent_encryption(leaf, ei, 0);
9498 btrfs_set_file_extent_compression(leaf, ei, 0);
9499 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9500 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9502 ptr = btrfs_file_extent_inline_start(ei);
9503 write_extent_buffer(leaf, symname, ptr, name_len);
9504 btrfs_mark_buffer_dirty(trans, leaf);
9505 btrfs_free_path(path);
9507 d_instantiate_new(dentry, inode);
9510 btrfs_end_transaction(trans);
9511 btrfs_btree_balance_dirty(fs_info);
9513 btrfs_new_inode_args_destroy(&new_inode_args);
9520 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9521 struct btrfs_trans_handle *trans_in,
9522 struct btrfs_inode *inode,
9523 struct btrfs_key *ins,
9526 struct btrfs_file_extent_item stack_fi;
9527 struct btrfs_replace_extent_info extent_info;
9528 struct btrfs_trans_handle *trans = trans_in;
9529 struct btrfs_path *path;
9530 u64 start = ins->objectid;
9531 u64 len = ins->offset;
9532 u64 qgroup_released = 0;
9535 memset(&stack_fi, 0, sizeof(stack_fi));
9537 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9538 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9539 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9540 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9541 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9542 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9543 /* Encryption and other encoding is reserved and all 0 */
9545 ret = btrfs_qgroup_release_data(inode, file_offset, len, &qgroup_released);
9547 return ERR_PTR(ret);
9550 ret = insert_reserved_file_extent(trans, inode,
9551 file_offset, &stack_fi,
9552 true, qgroup_released);
9558 extent_info.disk_offset = start;
9559 extent_info.disk_len = len;
9560 extent_info.data_offset = 0;
9561 extent_info.data_len = len;
9562 extent_info.file_offset = file_offset;
9563 extent_info.extent_buf = (char *)&stack_fi;
9564 extent_info.is_new_extent = true;
9565 extent_info.update_times = true;
9566 extent_info.qgroup_reserved = qgroup_released;
9567 extent_info.insertions = 0;
9569 path = btrfs_alloc_path();
9575 ret = btrfs_replace_file_extents(inode, path, file_offset,
9576 file_offset + len - 1, &extent_info,
9578 btrfs_free_path(path);
9585 * We have released qgroup data range at the beginning of the function,
9586 * and normally qgroup_released bytes will be freed when committing
9588 * But if we error out early, we have to free what we have released
9589 * or we leak qgroup data reservation.
9591 btrfs_qgroup_free_refroot(inode->root->fs_info,
9592 inode->root->root_key.objectid, qgroup_released,
9593 BTRFS_QGROUP_RSV_DATA);
9594 return ERR_PTR(ret);
9597 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9598 u64 start, u64 num_bytes, u64 min_size,
9599 loff_t actual_len, u64 *alloc_hint,
9600 struct btrfs_trans_handle *trans)
9602 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9603 struct extent_map *em;
9604 struct btrfs_root *root = BTRFS_I(inode)->root;
9605 struct btrfs_key ins;
9606 u64 cur_offset = start;
9607 u64 clear_offset = start;
9610 u64 last_alloc = (u64)-1;
9612 bool own_trans = true;
9613 u64 end = start + num_bytes - 1;
9617 while (num_bytes > 0) {
9618 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9619 cur_bytes = max(cur_bytes, min_size);
9621 * If we are severely fragmented we could end up with really
9622 * small allocations, so if the allocator is returning small
9623 * chunks lets make its job easier by only searching for those
9626 cur_bytes = min(cur_bytes, last_alloc);
9627 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9628 min_size, 0, *alloc_hint, &ins, 1, 0);
9633 * We've reserved this space, and thus converted it from
9634 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9635 * from here on out we will only need to clear our reservation
9636 * for the remaining unreserved area, so advance our
9637 * clear_offset by our extent size.
9639 clear_offset += ins.offset;
9641 last_alloc = ins.offset;
9642 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9645 * Now that we inserted the prealloc extent we can finally
9646 * decrement the number of reservations in the block group.
9647 * If we did it before, we could race with relocation and have
9648 * relocation miss the reserved extent, making it fail later.
9650 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9651 if (IS_ERR(trans)) {
9652 ret = PTR_ERR(trans);
9653 btrfs_free_reserved_extent(fs_info, ins.objectid,
9658 em = alloc_extent_map();
9660 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9661 cur_offset + ins.offset - 1, false);
9662 btrfs_set_inode_full_sync(BTRFS_I(inode));
9666 em->start = cur_offset;
9667 em->orig_start = cur_offset;
9668 em->len = ins.offset;
9669 em->block_start = ins.objectid;
9670 em->block_len = ins.offset;
9671 em->orig_block_len = ins.offset;
9672 em->ram_bytes = ins.offset;
9673 em->flags |= EXTENT_FLAG_PREALLOC;
9674 em->generation = trans->transid;
9676 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9677 free_extent_map(em);
9679 num_bytes -= ins.offset;
9680 cur_offset += ins.offset;
9681 *alloc_hint = ins.objectid + ins.offset;
9683 inode_inc_iversion(inode);
9684 inode_set_ctime_current(inode);
9685 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9686 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9687 (actual_len > inode->i_size) &&
9688 (cur_offset > inode->i_size)) {
9689 if (cur_offset > actual_len)
9690 i_size = actual_len;
9692 i_size = cur_offset;
9693 i_size_write(inode, i_size);
9694 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9697 ret = btrfs_update_inode(trans, BTRFS_I(inode));
9700 btrfs_abort_transaction(trans, ret);
9702 btrfs_end_transaction(trans);
9707 btrfs_end_transaction(trans);
9711 if (clear_offset < end)
9712 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9713 end - clear_offset + 1);
9717 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9718 u64 start, u64 num_bytes, u64 min_size,
9719 loff_t actual_len, u64 *alloc_hint)
9721 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9722 min_size, actual_len, alloc_hint,
9726 int btrfs_prealloc_file_range_trans(struct inode *inode,
9727 struct btrfs_trans_handle *trans, int mode,
9728 u64 start, u64 num_bytes, u64 min_size,
9729 loff_t actual_len, u64 *alloc_hint)
9731 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9732 min_size, actual_len, alloc_hint, trans);
9735 static int btrfs_permission(struct mnt_idmap *idmap,
9736 struct inode *inode, int mask)
9738 struct btrfs_root *root = BTRFS_I(inode)->root;
9739 umode_t mode = inode->i_mode;
9741 if (mask & MAY_WRITE &&
9742 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9743 if (btrfs_root_readonly(root))
9745 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9748 return generic_permission(idmap, inode, mask);
9751 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9752 struct file *file, umode_t mode)
9754 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9755 struct btrfs_trans_handle *trans;
9756 struct btrfs_root *root = BTRFS_I(dir)->root;
9757 struct inode *inode;
9758 struct btrfs_new_inode_args new_inode_args = {
9760 .dentry = file->f_path.dentry,
9763 unsigned int trans_num_items;
9766 inode = new_inode(dir->i_sb);
9769 inode_init_owner(idmap, inode, dir, mode);
9770 inode->i_fop = &btrfs_file_operations;
9771 inode->i_op = &btrfs_file_inode_operations;
9772 inode->i_mapping->a_ops = &btrfs_aops;
9774 new_inode_args.inode = inode;
9775 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9779 trans = btrfs_start_transaction(root, trans_num_items);
9780 if (IS_ERR(trans)) {
9781 ret = PTR_ERR(trans);
9782 goto out_new_inode_args;
9785 ret = btrfs_create_new_inode(trans, &new_inode_args);
9788 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9789 * set it to 1 because d_tmpfile() will issue a warning if the count is
9792 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9794 set_nlink(inode, 1);
9797 d_tmpfile(file, inode);
9798 unlock_new_inode(inode);
9799 mark_inode_dirty(inode);
9802 btrfs_end_transaction(trans);
9803 btrfs_btree_balance_dirty(fs_info);
9805 btrfs_new_inode_args_destroy(&new_inode_args);
9809 return finish_open_simple(file, ret);
9812 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9814 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9815 unsigned long index = start >> PAGE_SHIFT;
9816 unsigned long end_index = end >> PAGE_SHIFT;
9820 ASSERT(end + 1 - start <= U32_MAX);
9821 len = end + 1 - start;
9822 while (index <= end_index) {
9823 page = find_get_page(inode->vfs_inode.i_mapping, index);
9824 ASSERT(page); /* Pages should be in the extent_io_tree */
9826 /* This is for data, which doesn't yet support larger folio. */
9827 ASSERT(folio_order(page_folio(page)) == 0);
9828 btrfs_folio_set_writeback(fs_info, page_folio(page), start, len);
9834 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9837 switch (compress_type) {
9838 case BTRFS_COMPRESS_NONE:
9839 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9840 case BTRFS_COMPRESS_ZLIB:
9841 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9842 case BTRFS_COMPRESS_LZO:
9844 * The LZO format depends on the sector size. 64K is the maximum
9845 * sector size that we support.
9847 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9849 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9850 (fs_info->sectorsize_bits - 12);
9851 case BTRFS_COMPRESS_ZSTD:
9852 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9858 static ssize_t btrfs_encoded_read_inline(
9860 struct iov_iter *iter, u64 start,
9862 struct extent_state **cached_state,
9863 u64 extent_start, size_t count,
9864 struct btrfs_ioctl_encoded_io_args *encoded,
9867 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9868 struct btrfs_root *root = inode->root;
9869 struct btrfs_fs_info *fs_info = root->fs_info;
9870 struct extent_io_tree *io_tree = &inode->io_tree;
9871 struct btrfs_path *path;
9872 struct extent_buffer *leaf;
9873 struct btrfs_file_extent_item *item;
9879 path = btrfs_alloc_path();
9884 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9888 /* The extent item disappeared? */
9893 leaf = path->nodes[0];
9894 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9896 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9897 ptr = btrfs_file_extent_inline_start(item);
9899 encoded->len = min_t(u64, extent_start + ram_bytes,
9900 inode->vfs_inode.i_size) - iocb->ki_pos;
9901 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9902 btrfs_file_extent_compression(leaf, item));
9905 encoded->compression = ret;
9906 if (encoded->compression) {
9909 inline_size = btrfs_file_extent_inline_item_len(leaf,
9911 if (inline_size > count) {
9915 count = inline_size;
9916 encoded->unencoded_len = ram_bytes;
9917 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9919 count = min_t(u64, count, encoded->len);
9920 encoded->len = count;
9921 encoded->unencoded_len = count;
9922 ptr += iocb->ki_pos - extent_start;
9925 tmp = kmalloc(count, GFP_NOFS);
9930 read_extent_buffer(leaf, tmp, ptr, count);
9931 btrfs_release_path(path);
9932 unlock_extent(io_tree, start, lockend, cached_state);
9933 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9936 ret = copy_to_iter(tmp, count, iter);
9941 btrfs_free_path(path);
9945 struct btrfs_encoded_read_private {
9946 wait_queue_head_t wait;
9948 blk_status_t status;
9951 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9953 struct btrfs_encoded_read_private *priv = bbio->private;
9955 if (bbio->bio.bi_status) {
9957 * The memory barrier implied by the atomic_dec_return() here
9958 * pairs with the memory barrier implied by the
9959 * atomic_dec_return() or io_wait_event() in
9960 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9961 * write is observed before the load of status in
9962 * btrfs_encoded_read_regular_fill_pages().
9964 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9966 if (!atomic_dec_return(&priv->pending))
9967 wake_up(&priv->wait);
9968 bio_put(&bbio->bio);
9971 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9972 u64 file_offset, u64 disk_bytenr,
9973 u64 disk_io_size, struct page **pages)
9975 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9976 struct btrfs_encoded_read_private priv = {
9977 .pending = ATOMIC_INIT(1),
9979 unsigned long i = 0;
9980 struct btrfs_bio *bbio;
9982 init_waitqueue_head(&priv.wait);
9984 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9985 btrfs_encoded_read_endio, &priv);
9986 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9987 bbio->inode = inode;
9990 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
9992 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
9993 atomic_inc(&priv.pending);
9994 btrfs_submit_bio(bbio, 0);
9996 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9997 btrfs_encoded_read_endio, &priv);
9998 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9999 bbio->inode = inode;
10004 disk_bytenr += bytes;
10005 disk_io_size -= bytes;
10006 } while (disk_io_size);
10008 atomic_inc(&priv.pending);
10009 btrfs_submit_bio(bbio, 0);
10011 if (atomic_dec_return(&priv.pending))
10012 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10013 /* See btrfs_encoded_read_endio() for ordering. */
10014 return blk_status_to_errno(READ_ONCE(priv.status));
10017 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10018 struct iov_iter *iter,
10019 u64 start, u64 lockend,
10020 struct extent_state **cached_state,
10021 u64 disk_bytenr, u64 disk_io_size,
10022 size_t count, bool compressed,
10025 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10026 struct extent_io_tree *io_tree = &inode->io_tree;
10027 struct page **pages;
10028 unsigned long nr_pages, i;
10030 size_t page_offset;
10033 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10034 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10037 ret = btrfs_alloc_page_array(nr_pages, pages, 0);
10043 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10044 disk_io_size, pages);
10048 unlock_extent(io_tree, start, lockend, cached_state);
10049 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10056 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10057 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10060 while (cur < count) {
10061 size_t bytes = min_t(size_t, count - cur,
10062 PAGE_SIZE - page_offset);
10064 if (copy_page_to_iter(pages[i], page_offset, bytes,
10075 for (i = 0; i < nr_pages; i++) {
10077 __free_page(pages[i]);
10083 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10084 struct btrfs_ioctl_encoded_io_args *encoded)
10086 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10087 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10088 struct extent_io_tree *io_tree = &inode->io_tree;
10090 size_t count = iov_iter_count(iter);
10091 u64 start, lockend, disk_bytenr, disk_io_size;
10092 struct extent_state *cached_state = NULL;
10093 struct extent_map *em;
10094 bool unlocked = false;
10096 file_accessed(iocb->ki_filp);
10098 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10100 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10101 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10104 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10106 * We don't know how long the extent containing iocb->ki_pos is, but if
10107 * it's compressed we know that it won't be longer than this.
10109 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10112 struct btrfs_ordered_extent *ordered;
10114 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10115 lockend - start + 1);
10117 goto out_unlock_inode;
10118 lock_extent(io_tree, start, lockend, &cached_state);
10119 ordered = btrfs_lookup_ordered_range(inode, start,
10120 lockend - start + 1);
10123 btrfs_put_ordered_extent(ordered);
10124 unlock_extent(io_tree, start, lockend, &cached_state);
10128 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10131 goto out_unlock_extent;
10134 if (em->block_start == EXTENT_MAP_INLINE) {
10135 u64 extent_start = em->start;
10138 * For inline extents we get everything we need out of the
10141 free_extent_map(em);
10143 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10144 &cached_state, extent_start,
10145 count, encoded, &unlocked);
10150 * We only want to return up to EOF even if the extent extends beyond
10153 encoded->len = min_t(u64, extent_map_end(em),
10154 inode->vfs_inode.i_size) - iocb->ki_pos;
10155 if (em->block_start == EXTENT_MAP_HOLE ||
10156 (em->flags & EXTENT_FLAG_PREALLOC)) {
10157 disk_bytenr = EXTENT_MAP_HOLE;
10158 count = min_t(u64, count, encoded->len);
10159 encoded->len = count;
10160 encoded->unencoded_len = count;
10161 } else if (extent_map_is_compressed(em)) {
10162 disk_bytenr = em->block_start;
10164 * Bail if the buffer isn't large enough to return the whole
10165 * compressed extent.
10167 if (em->block_len > count) {
10171 disk_io_size = em->block_len;
10172 count = em->block_len;
10173 encoded->unencoded_len = em->ram_bytes;
10174 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10175 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10176 extent_map_compression(em));
10179 encoded->compression = ret;
10181 disk_bytenr = em->block_start + (start - em->start);
10182 if (encoded->len > count)
10183 encoded->len = count;
10185 * Don't read beyond what we locked. This also limits the page
10186 * allocations that we'll do.
10188 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10189 count = start + disk_io_size - iocb->ki_pos;
10190 encoded->len = count;
10191 encoded->unencoded_len = count;
10192 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10194 free_extent_map(em);
10197 if (disk_bytenr == EXTENT_MAP_HOLE) {
10198 unlock_extent(io_tree, start, lockend, &cached_state);
10199 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10201 ret = iov_iter_zero(count, iter);
10205 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10206 &cached_state, disk_bytenr,
10207 disk_io_size, count,
10208 encoded->compression,
10214 iocb->ki_pos += encoded->len;
10216 free_extent_map(em);
10219 unlock_extent(io_tree, start, lockend, &cached_state);
10222 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10226 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10227 const struct btrfs_ioctl_encoded_io_args *encoded)
10229 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10230 struct btrfs_root *root = inode->root;
10231 struct btrfs_fs_info *fs_info = root->fs_info;
10232 struct extent_io_tree *io_tree = &inode->io_tree;
10233 struct extent_changeset *data_reserved = NULL;
10234 struct extent_state *cached_state = NULL;
10235 struct btrfs_ordered_extent *ordered;
10239 u64 num_bytes, ram_bytes, disk_num_bytes;
10240 unsigned long nr_pages, i;
10241 struct page **pages;
10242 struct btrfs_key ins;
10243 bool extent_reserved = false;
10244 struct extent_map *em;
10247 switch (encoded->compression) {
10248 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10249 compression = BTRFS_COMPRESS_ZLIB;
10251 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10252 compression = BTRFS_COMPRESS_ZSTD;
10254 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10255 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10256 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10257 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10258 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10259 /* The sector size must match for LZO. */
10260 if (encoded->compression -
10261 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10262 fs_info->sectorsize_bits)
10264 compression = BTRFS_COMPRESS_LZO;
10269 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10272 orig_count = iov_iter_count(from);
10274 /* The extent size must be sane. */
10275 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10276 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10280 * The compressed data must be smaller than the decompressed data.
10282 * It's of course possible for data to compress to larger or the same
10283 * size, but the buffered I/O path falls back to no compression for such
10284 * data, and we don't want to break any assumptions by creating these
10287 * Note that this is less strict than the current check we have that the
10288 * compressed data must be at least one sector smaller than the
10289 * decompressed data. We only want to enforce the weaker requirement
10290 * from old kernels that it is at least one byte smaller.
10292 if (orig_count >= encoded->unencoded_len)
10295 /* The extent must start on a sector boundary. */
10296 start = iocb->ki_pos;
10297 if (!IS_ALIGNED(start, fs_info->sectorsize))
10301 * The extent must end on a sector boundary. However, we allow a write
10302 * which ends at or extends i_size to have an unaligned length; we round
10303 * up the extent size and set i_size to the unaligned end.
10305 if (start + encoded->len < inode->vfs_inode.i_size &&
10306 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10309 /* Finally, the offset in the unencoded data must be sector-aligned. */
10310 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10313 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10314 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10315 end = start + num_bytes - 1;
10318 * If the extent cannot be inline, the compressed data on disk must be
10319 * sector-aligned. For convenience, we extend it with zeroes if it
10322 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10323 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10324 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10327 for (i = 0; i < nr_pages; i++) {
10328 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10331 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10336 kaddr = kmap_local_page(pages[i]);
10337 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10338 kunmap_local(kaddr);
10342 if (bytes < PAGE_SIZE)
10343 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10344 kunmap_local(kaddr);
10348 struct btrfs_ordered_extent *ordered;
10350 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10353 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10354 start >> PAGE_SHIFT,
10355 end >> PAGE_SHIFT);
10358 lock_extent(io_tree, start, end, &cached_state);
10359 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10361 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10364 btrfs_put_ordered_extent(ordered);
10365 unlock_extent(io_tree, start, end, &cached_state);
10370 * We don't use the higher-level delalloc space functions because our
10371 * num_bytes and disk_num_bytes are different.
10373 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10376 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10378 goto out_free_data_space;
10379 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10382 goto out_qgroup_free_data;
10384 /* Try an inline extent first. */
10385 if (start == 0 && encoded->unencoded_len == encoded->len &&
10386 encoded->unencoded_offset == 0) {
10387 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10388 compression, pages, true);
10392 goto out_delalloc_release;
10396 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10397 disk_num_bytes, 0, 0, &ins, 1, 1);
10399 goto out_delalloc_release;
10400 extent_reserved = true;
10402 em = create_io_em(inode, start, num_bytes,
10403 start - encoded->unencoded_offset, ins.objectid,
10404 ins.offset, ins.offset, ram_bytes, compression,
10405 BTRFS_ORDERED_COMPRESSED);
10408 goto out_free_reserved;
10410 free_extent_map(em);
10412 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10413 ins.objectid, ins.offset,
10414 encoded->unencoded_offset,
10415 (1 << BTRFS_ORDERED_ENCODED) |
10416 (1 << BTRFS_ORDERED_COMPRESSED),
10418 if (IS_ERR(ordered)) {
10419 btrfs_drop_extent_map_range(inode, start, end, false);
10420 ret = PTR_ERR(ordered);
10421 goto out_free_reserved;
10423 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10425 if (start + encoded->len > inode->vfs_inode.i_size)
10426 i_size_write(&inode->vfs_inode, start + encoded->len);
10428 unlock_extent(io_tree, start, end, &cached_state);
10430 btrfs_delalloc_release_extents(inode, num_bytes);
10432 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10437 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10438 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10439 out_delalloc_release:
10440 btrfs_delalloc_release_extents(inode, num_bytes);
10441 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10442 out_qgroup_free_data:
10444 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes, NULL);
10445 out_free_data_space:
10447 * If btrfs_reserve_extent() succeeded, then we already decremented
10450 if (!extent_reserved)
10451 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10453 unlock_extent(io_tree, start, end, &cached_state);
10455 for (i = 0; i < nr_pages; i++) {
10457 __free_page(pages[i]);
10462 iocb->ki_pos += encoded->len;
10468 * Add an entry indicating a block group or device which is pinned by a
10469 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10470 * negative errno on failure.
10472 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10473 bool is_block_group)
10475 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10476 struct btrfs_swapfile_pin *sp, *entry;
10477 struct rb_node **p;
10478 struct rb_node *parent = NULL;
10480 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10485 sp->is_block_group = is_block_group;
10486 sp->bg_extent_count = 1;
10488 spin_lock(&fs_info->swapfile_pins_lock);
10489 p = &fs_info->swapfile_pins.rb_node;
10492 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10493 if (sp->ptr < entry->ptr ||
10494 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10495 p = &(*p)->rb_left;
10496 } else if (sp->ptr > entry->ptr ||
10497 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10498 p = &(*p)->rb_right;
10500 if (is_block_group)
10501 entry->bg_extent_count++;
10502 spin_unlock(&fs_info->swapfile_pins_lock);
10507 rb_link_node(&sp->node, parent, p);
10508 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10509 spin_unlock(&fs_info->swapfile_pins_lock);
10513 /* Free all of the entries pinned by this swapfile. */
10514 static void btrfs_free_swapfile_pins(struct inode *inode)
10516 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10517 struct btrfs_swapfile_pin *sp;
10518 struct rb_node *node, *next;
10520 spin_lock(&fs_info->swapfile_pins_lock);
10521 node = rb_first(&fs_info->swapfile_pins);
10523 next = rb_next(node);
10524 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10525 if (sp->inode == inode) {
10526 rb_erase(&sp->node, &fs_info->swapfile_pins);
10527 if (sp->is_block_group) {
10528 btrfs_dec_block_group_swap_extents(sp->ptr,
10529 sp->bg_extent_count);
10530 btrfs_put_block_group(sp->ptr);
10536 spin_unlock(&fs_info->swapfile_pins_lock);
10539 struct btrfs_swap_info {
10545 unsigned long nr_pages;
10549 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10550 struct btrfs_swap_info *bsi)
10552 unsigned long nr_pages;
10553 unsigned long max_pages;
10554 u64 first_ppage, first_ppage_reported, next_ppage;
10558 * Our swapfile may have had its size extended after the swap header was
10559 * written. In that case activating the swapfile should not go beyond
10560 * the max size set in the swap header.
10562 if (bsi->nr_pages >= sis->max)
10565 max_pages = sis->max - bsi->nr_pages;
10566 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10567 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10569 if (first_ppage >= next_ppage)
10571 nr_pages = next_ppage - first_ppage;
10572 nr_pages = min(nr_pages, max_pages);
10574 first_ppage_reported = first_ppage;
10575 if (bsi->start == 0)
10576 first_ppage_reported++;
10577 if (bsi->lowest_ppage > first_ppage_reported)
10578 bsi->lowest_ppage = first_ppage_reported;
10579 if (bsi->highest_ppage < (next_ppage - 1))
10580 bsi->highest_ppage = next_ppage - 1;
10582 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10585 bsi->nr_extents += ret;
10586 bsi->nr_pages += nr_pages;
10590 static void btrfs_swap_deactivate(struct file *file)
10592 struct inode *inode = file_inode(file);
10594 btrfs_free_swapfile_pins(inode);
10595 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10598 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10601 struct inode *inode = file_inode(file);
10602 struct btrfs_root *root = BTRFS_I(inode)->root;
10603 struct btrfs_fs_info *fs_info = root->fs_info;
10604 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10605 struct extent_state *cached_state = NULL;
10606 struct extent_map *em = NULL;
10607 struct btrfs_chunk_map *map = NULL;
10608 struct btrfs_device *device = NULL;
10609 struct btrfs_swap_info bsi = {
10610 .lowest_ppage = (sector_t)-1ULL,
10617 * If the swap file was just created, make sure delalloc is done. If the
10618 * file changes again after this, the user is doing something stupid and
10619 * we don't really care.
10621 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10626 * The inode is locked, so these flags won't change after we check them.
10628 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10629 btrfs_warn(fs_info, "swapfile must not be compressed");
10632 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10633 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10636 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10637 btrfs_warn(fs_info, "swapfile must not be checksummed");
10642 * Balance or device remove/replace/resize can move stuff around from
10643 * under us. The exclop protection makes sure they aren't running/won't
10644 * run concurrently while we are mapping the swap extents, and
10645 * fs_info->swapfile_pins prevents them from running while the swap
10646 * file is active and moving the extents. Note that this also prevents
10647 * a concurrent device add which isn't actually necessary, but it's not
10648 * really worth the trouble to allow it.
10650 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10651 btrfs_warn(fs_info,
10652 "cannot activate swapfile while exclusive operation is running");
10657 * Prevent snapshot creation while we are activating the swap file.
10658 * We do not want to race with snapshot creation. If snapshot creation
10659 * already started before we bumped nr_swapfiles from 0 to 1 and
10660 * completes before the first write into the swap file after it is
10661 * activated, than that write would fallback to COW.
10663 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10664 btrfs_exclop_finish(fs_info);
10665 btrfs_warn(fs_info,
10666 "cannot activate swapfile because snapshot creation is in progress");
10670 * Snapshots can create extents which require COW even if NODATACOW is
10671 * set. We use this counter to prevent snapshots. We must increment it
10672 * before walking the extents because we don't want a concurrent
10673 * snapshot to run after we've already checked the extents.
10675 * It is possible that subvolume is marked for deletion but still not
10676 * removed yet. To prevent this race, we check the root status before
10677 * activating the swapfile.
10679 spin_lock(&root->root_item_lock);
10680 if (btrfs_root_dead(root)) {
10681 spin_unlock(&root->root_item_lock);
10683 btrfs_exclop_finish(fs_info);
10684 btrfs_warn(fs_info,
10685 "cannot activate swapfile because subvolume %llu is being deleted",
10686 root->root_key.objectid);
10689 atomic_inc(&root->nr_swapfiles);
10690 spin_unlock(&root->root_item_lock);
10692 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10694 lock_extent(io_tree, 0, isize - 1, &cached_state);
10696 while (start < isize) {
10697 u64 logical_block_start, physical_block_start;
10698 struct btrfs_block_group *bg;
10699 u64 len = isize - start;
10701 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10707 if (em->block_start == EXTENT_MAP_HOLE) {
10708 btrfs_warn(fs_info, "swapfile must not have holes");
10712 if (em->block_start == EXTENT_MAP_INLINE) {
10714 * It's unlikely we'll ever actually find ourselves
10715 * here, as a file small enough to fit inline won't be
10716 * big enough to store more than the swap header, but in
10717 * case something changes in the future, let's catch it
10718 * here rather than later.
10720 btrfs_warn(fs_info, "swapfile must not be inline");
10724 if (extent_map_is_compressed(em)) {
10725 btrfs_warn(fs_info, "swapfile must not be compressed");
10730 logical_block_start = em->block_start + (start - em->start);
10731 len = min(len, em->len - (start - em->start));
10732 free_extent_map(em);
10735 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10741 btrfs_warn(fs_info,
10742 "swapfile must not be copy-on-write");
10747 map = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10749 ret = PTR_ERR(map);
10753 if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10754 btrfs_warn(fs_info,
10755 "swapfile must have single data profile");
10760 if (device == NULL) {
10761 device = map->stripes[0].dev;
10762 ret = btrfs_add_swapfile_pin(inode, device, false);
10767 } else if (device != map->stripes[0].dev) {
10768 btrfs_warn(fs_info, "swapfile must be on one device");
10773 physical_block_start = (map->stripes[0].physical +
10774 (logical_block_start - map->start));
10775 len = min(len, map->chunk_len - (logical_block_start - map->start));
10776 btrfs_free_chunk_map(map);
10779 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10781 btrfs_warn(fs_info,
10782 "could not find block group containing swapfile");
10787 if (!btrfs_inc_block_group_swap_extents(bg)) {
10788 btrfs_warn(fs_info,
10789 "block group for swapfile at %llu is read-only%s",
10791 atomic_read(&fs_info->scrubs_running) ?
10792 " (scrub running)" : "");
10793 btrfs_put_block_group(bg);
10798 ret = btrfs_add_swapfile_pin(inode, bg, true);
10800 btrfs_put_block_group(bg);
10807 if (bsi.block_len &&
10808 bsi.block_start + bsi.block_len == physical_block_start) {
10809 bsi.block_len += len;
10811 if (bsi.block_len) {
10812 ret = btrfs_add_swap_extent(sis, &bsi);
10817 bsi.block_start = physical_block_start;
10818 bsi.block_len = len;
10825 ret = btrfs_add_swap_extent(sis, &bsi);
10828 if (!IS_ERR_OR_NULL(em))
10829 free_extent_map(em);
10830 if (!IS_ERR_OR_NULL(map))
10831 btrfs_free_chunk_map(map);
10833 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10836 btrfs_swap_deactivate(file);
10838 btrfs_drew_write_unlock(&root->snapshot_lock);
10840 btrfs_exclop_finish(fs_info);
10846 sis->bdev = device->bdev;
10847 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10848 sis->max = bsi.nr_pages;
10849 sis->pages = bsi.nr_pages - 1;
10850 sis->highest_bit = bsi.nr_pages - 1;
10851 return bsi.nr_extents;
10854 static void btrfs_swap_deactivate(struct file *file)
10858 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10861 return -EOPNOTSUPP;
10866 * Update the number of bytes used in the VFS' inode. When we replace extents in
10867 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10868 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10869 * always get a correct value.
10871 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10872 const u64 add_bytes,
10873 const u64 del_bytes)
10875 if (add_bytes == del_bytes)
10878 spin_lock(&inode->lock);
10880 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10882 inode_add_bytes(&inode->vfs_inode, add_bytes);
10883 spin_unlock(&inode->lock);
10887 * Verify that there are no ordered extents for a given file range.
10889 * @inode: The target inode.
10890 * @start: Start offset of the file range, should be sector size aligned.
10891 * @end: End offset (inclusive) of the file range, its value +1 should be
10892 * sector size aligned.
10894 * This should typically be used for cases where we locked an inode's VFS lock in
10895 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10896 * we have flushed all delalloc in the range, we have waited for all ordered
10897 * extents in the range to complete and finally we have locked the file range in
10898 * the inode's io_tree.
10900 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10902 struct btrfs_root *root = inode->root;
10903 struct btrfs_ordered_extent *ordered;
10905 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10908 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10910 btrfs_err(root->fs_info,
10911 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10912 start, end, btrfs_ino(inode), root->root_key.objectid,
10913 ordered->file_offset,
10914 ordered->file_offset + ordered->num_bytes - 1);
10915 btrfs_put_ordered_extent(ordered);
10918 ASSERT(ordered == NULL);
10921 static const struct inode_operations btrfs_dir_inode_operations = {
10922 .getattr = btrfs_getattr,
10923 .lookup = btrfs_lookup,
10924 .create = btrfs_create,
10925 .unlink = btrfs_unlink,
10926 .link = btrfs_link,
10927 .mkdir = btrfs_mkdir,
10928 .rmdir = btrfs_rmdir,
10929 .rename = btrfs_rename2,
10930 .symlink = btrfs_symlink,
10931 .setattr = btrfs_setattr,
10932 .mknod = btrfs_mknod,
10933 .listxattr = btrfs_listxattr,
10934 .permission = btrfs_permission,
10935 .get_inode_acl = btrfs_get_acl,
10936 .set_acl = btrfs_set_acl,
10937 .update_time = btrfs_update_time,
10938 .tmpfile = btrfs_tmpfile,
10939 .fileattr_get = btrfs_fileattr_get,
10940 .fileattr_set = btrfs_fileattr_set,
10943 static const struct file_operations btrfs_dir_file_operations = {
10944 .llseek = btrfs_dir_llseek,
10945 .read = generic_read_dir,
10946 .iterate_shared = btrfs_real_readdir,
10947 .open = btrfs_opendir,
10948 .unlocked_ioctl = btrfs_ioctl,
10949 #ifdef CONFIG_COMPAT
10950 .compat_ioctl = btrfs_compat_ioctl,
10952 .release = btrfs_release_file,
10953 .fsync = btrfs_sync_file,
10957 * btrfs doesn't support the bmap operation because swapfiles
10958 * use bmap to make a mapping of extents in the file. They assume
10959 * these extents won't change over the life of the file and they
10960 * use the bmap result to do IO directly to the drive.
10962 * the btrfs bmap call would return logical addresses that aren't
10963 * suitable for IO and they also will change frequently as COW
10964 * operations happen. So, swapfile + btrfs == corruption.
10966 * For now we're avoiding this by dropping bmap.
10968 static const struct address_space_operations btrfs_aops = {
10969 .read_folio = btrfs_read_folio,
10970 .writepages = btrfs_writepages,
10971 .readahead = btrfs_readahead,
10972 .invalidate_folio = btrfs_invalidate_folio,
10973 .release_folio = btrfs_release_folio,
10974 .migrate_folio = btrfs_migrate_folio,
10975 .dirty_folio = filemap_dirty_folio,
10976 .error_remove_folio = generic_error_remove_folio,
10977 .swap_activate = btrfs_swap_activate,
10978 .swap_deactivate = btrfs_swap_deactivate,
10981 static const struct inode_operations btrfs_file_inode_operations = {
10982 .getattr = btrfs_getattr,
10983 .setattr = btrfs_setattr,
10984 .listxattr = btrfs_listxattr,
10985 .permission = btrfs_permission,
10986 .fiemap = btrfs_fiemap,
10987 .get_inode_acl = btrfs_get_acl,
10988 .set_acl = btrfs_set_acl,
10989 .update_time = btrfs_update_time,
10990 .fileattr_get = btrfs_fileattr_get,
10991 .fileattr_set = btrfs_fileattr_set,
10993 static const struct inode_operations btrfs_special_inode_operations = {
10994 .getattr = btrfs_getattr,
10995 .setattr = btrfs_setattr,
10996 .permission = btrfs_permission,
10997 .listxattr = btrfs_listxattr,
10998 .get_inode_acl = btrfs_get_acl,
10999 .set_acl = btrfs_set_acl,
11000 .update_time = btrfs_update_time,
11002 static const struct inode_operations btrfs_symlink_inode_operations = {
11003 .get_link = page_get_link,
11004 .getattr = btrfs_getattr,
11005 .setattr = btrfs_setattr,
11006 .permission = btrfs_permission,
11007 .listxattr = btrfs_listxattr,
11008 .update_time = btrfs_update_time,
11011 const struct dentry_operations btrfs_dentry_operations = {
11012 .d_delete = btrfs_dentry_delete,