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;
117 static const struct inode_operations btrfs_dir_inode_operations;
118 static const struct inode_operations btrfs_symlink_inode_operations;
119 static const struct inode_operations btrfs_special_inode_operations;
120 static const struct inode_operations btrfs_file_inode_operations;
121 static const struct address_space_operations btrfs_aops;
122 static const struct file_operations btrfs_dir_file_operations;
124 static struct kmem_cache *btrfs_inode_cachep;
126 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
127 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
129 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
130 struct page *locked_page, u64 start,
131 u64 end, struct writeback_control *wbc,
133 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
134 u64 len, u64 orig_start, u64 block_start,
135 u64 block_len, u64 orig_block_len,
136 u64 ram_bytes, int compress_type,
139 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
140 u64 root, void *warn_ctx)
142 struct data_reloc_warn *warn = warn_ctx;
143 struct btrfs_fs_info *fs_info = warn->fs_info;
144 struct extent_buffer *eb;
145 struct btrfs_inode_item *inode_item;
146 struct inode_fs_paths *ipath = NULL;
147 struct btrfs_root *local_root;
148 struct btrfs_key key;
149 unsigned int nofs_flag;
153 local_root = btrfs_get_fs_root(fs_info, root, true);
154 if (IS_ERR(local_root)) {
155 ret = PTR_ERR(local_root);
159 /* This makes the path point to (inum INODE_ITEM ioff). */
161 key.type = BTRFS_INODE_ITEM_KEY;
164 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
166 btrfs_put_root(local_root);
167 btrfs_release_path(&warn->path);
171 eb = warn->path.nodes[0];
172 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
173 nlink = btrfs_inode_nlink(eb, inode_item);
174 btrfs_release_path(&warn->path);
176 nofs_flag = memalloc_nofs_save();
177 ipath = init_ipath(4096, local_root, &warn->path);
178 memalloc_nofs_restore(nofs_flag);
180 btrfs_put_root(local_root);
181 ret = PTR_ERR(ipath);
184 * -ENOMEM, not a critical error, just output an generic error
188 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
189 warn->logical, warn->mirror_num, root, inum, offset);
192 ret = paths_from_inode(inum, ipath);
197 * We deliberately ignore the bit ipath might have been too small to
198 * hold all of the paths here
200 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
202 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
203 warn->logical, warn->mirror_num, root, inum, offset,
204 fs_info->sectorsize, nlink,
205 (char *)(unsigned long)ipath->fspath->val[i]);
208 btrfs_put_root(local_root);
214 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
215 warn->logical, warn->mirror_num, root, inum, offset, ret);
222 * Do extra user-friendly error output (e.g. lookup all the affected files).
224 * Return true if we succeeded doing the backref lookup.
225 * Return false if such lookup failed, and has to fallback to the old error message.
227 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
228 const u8 *csum, const u8 *csum_expected,
231 struct btrfs_fs_info *fs_info = inode->root->fs_info;
232 struct btrfs_path path = { 0 };
233 struct btrfs_key found_key = { 0 };
234 struct extent_buffer *eb;
235 struct btrfs_extent_item *ei;
236 const u32 csum_size = fs_info->csum_size;
242 mutex_lock(&fs_info->reloc_mutex);
243 logical = btrfs_get_reloc_bg_bytenr(fs_info);
244 mutex_unlock(&fs_info->reloc_mutex);
246 if (logical == U64_MAX) {
247 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
248 btrfs_warn_rl(fs_info,
249 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
250 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
251 CSUM_FMT_VALUE(csum_size, csum),
252 CSUM_FMT_VALUE(csum_size, csum_expected),
258 btrfs_warn_rl(fs_info,
259 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
260 inode->root->root_key.objectid,
261 btrfs_ino(inode), file_off, logical,
262 CSUM_FMT_VALUE(csum_size, csum),
263 CSUM_FMT_VALUE(csum_size, csum_expected),
266 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
268 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
273 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
274 item_size = btrfs_item_size(eb, path.slots[0]);
275 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
276 unsigned long ptr = 0;
281 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
282 item_size, &ref_root,
285 btrfs_warn_rl(fs_info,
286 "failed to resolve tree backref for logical %llu: %d",
293 btrfs_warn_rl(fs_info,
294 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
296 (ref_level ? "node" : "leaf"),
297 ref_level, ref_root);
299 btrfs_release_path(&path);
301 struct btrfs_backref_walk_ctx ctx = { 0 };
302 struct data_reloc_warn reloc_warn = { 0 };
304 btrfs_release_path(&path);
306 ctx.bytenr = found_key.objectid;
307 ctx.extent_item_pos = logical - found_key.objectid;
308 ctx.fs_info = fs_info;
310 reloc_warn.logical = logical;
311 reloc_warn.extent_item_size = found_key.offset;
312 reloc_warn.mirror_num = mirror_num;
313 reloc_warn.fs_info = fs_info;
315 iterate_extent_inodes(&ctx, true,
316 data_reloc_print_warning_inode, &reloc_warn);
320 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
321 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
323 struct btrfs_root *root = inode->root;
324 const u32 csum_size = root->fs_info->csum_size;
326 /* For data reloc tree, it's better to do a backref lookup instead. */
327 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
328 return print_data_reloc_error(inode, logical_start, csum,
329 csum_expected, mirror_num);
331 /* Output without objectid, which is more meaningful */
332 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
333 btrfs_warn_rl(root->fs_info,
334 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
335 root->root_key.objectid, btrfs_ino(inode),
337 CSUM_FMT_VALUE(csum_size, csum),
338 CSUM_FMT_VALUE(csum_size, csum_expected),
341 btrfs_warn_rl(root->fs_info,
342 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
343 root->root_key.objectid, btrfs_ino(inode),
345 CSUM_FMT_VALUE(csum_size, csum),
346 CSUM_FMT_VALUE(csum_size, csum_expected),
352 * Lock inode i_rwsem based on arguments passed.
354 * ilock_flags can have the following bit set:
356 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
357 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
359 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
361 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
363 if (ilock_flags & BTRFS_ILOCK_SHARED) {
364 if (ilock_flags & BTRFS_ILOCK_TRY) {
365 if (!inode_trylock_shared(&inode->vfs_inode))
370 inode_lock_shared(&inode->vfs_inode);
372 if (ilock_flags & BTRFS_ILOCK_TRY) {
373 if (!inode_trylock(&inode->vfs_inode))
378 inode_lock(&inode->vfs_inode);
380 if (ilock_flags & BTRFS_ILOCK_MMAP)
381 down_write(&inode->i_mmap_lock);
386 * Unock inode i_rwsem.
388 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
389 * to decide whether the lock acquired is shared or exclusive.
391 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
393 if (ilock_flags & BTRFS_ILOCK_MMAP)
394 up_write(&inode->i_mmap_lock);
395 if (ilock_flags & BTRFS_ILOCK_SHARED)
396 inode_unlock_shared(&inode->vfs_inode);
398 inode_unlock(&inode->vfs_inode);
402 * Cleanup all submitted ordered extents in specified range to handle errors
403 * from the btrfs_run_delalloc_range() callback.
405 * NOTE: caller must ensure that when an error happens, it can not call
406 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
407 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
408 * to be released, which we want to happen only when finishing the ordered
409 * extent (btrfs_finish_ordered_io()).
411 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
412 struct page *locked_page,
413 u64 offset, u64 bytes)
415 unsigned long index = offset >> PAGE_SHIFT;
416 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
417 u64 page_start = 0, page_end = 0;
421 page_start = page_offset(locked_page);
422 page_end = page_start + PAGE_SIZE - 1;
425 while (index <= end_index) {
427 * For locked page, we will call btrfs_mark_ordered_io_finished
428 * through btrfs_mark_ordered_io_finished() on it
429 * in run_delalloc_range() for the error handling, which will
430 * clear page Ordered and run the ordered extent accounting.
432 * Here we can't just clear the Ordered bit, or
433 * btrfs_mark_ordered_io_finished() would skip the accounting
434 * for the page range, and the ordered extent will never finish.
436 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
440 page = find_get_page(inode->vfs_inode.i_mapping, index);
446 * Here we just clear all Ordered bits for every page in the
447 * range, then btrfs_mark_ordered_io_finished() will handle
448 * the ordered extent accounting for the range.
450 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
456 /* The locked page covers the full range, nothing needs to be done */
457 if (bytes + offset <= page_start + PAGE_SIZE)
460 * In case this page belongs to the delalloc range being
461 * instantiated then skip it, since the first page of a range is
462 * going to be properly cleaned up by the caller of
465 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
466 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
467 offset = page_offset(locked_page) + PAGE_SIZE;
471 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
474 static int btrfs_dirty_inode(struct btrfs_inode *inode);
476 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
477 struct btrfs_new_inode_args *args)
481 if (args->default_acl) {
482 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
488 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
492 if (!args->default_acl && !args->acl)
493 cache_no_acl(args->inode);
494 return btrfs_xattr_security_init(trans, args->inode, args->dir,
495 &args->dentry->d_name);
499 * this does all the hard work for inserting an inline extent into
500 * the btree. The caller should have done a btrfs_drop_extents so that
501 * no overlapping inline items exist in the btree
503 static int insert_inline_extent(struct btrfs_trans_handle *trans,
504 struct btrfs_path *path,
505 struct btrfs_inode *inode, bool extent_inserted,
506 size_t size, size_t compressed_size,
508 struct page **compressed_pages,
511 struct btrfs_root *root = inode->root;
512 struct extent_buffer *leaf;
513 struct page *page = NULL;
516 struct btrfs_file_extent_item *ei;
518 size_t cur_size = size;
521 ASSERT((compressed_size > 0 && compressed_pages) ||
522 (compressed_size == 0 && !compressed_pages));
524 if (compressed_size && compressed_pages)
525 cur_size = compressed_size;
527 if (!extent_inserted) {
528 struct btrfs_key key;
531 key.objectid = btrfs_ino(inode);
533 key.type = BTRFS_EXTENT_DATA_KEY;
535 datasize = btrfs_file_extent_calc_inline_size(cur_size);
536 ret = btrfs_insert_empty_item(trans, root, path, &key,
541 leaf = path->nodes[0];
542 ei = btrfs_item_ptr(leaf, path->slots[0],
543 struct btrfs_file_extent_item);
544 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
545 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
546 btrfs_set_file_extent_encryption(leaf, ei, 0);
547 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
548 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
549 ptr = btrfs_file_extent_inline_start(ei);
551 if (compress_type != BTRFS_COMPRESS_NONE) {
554 while (compressed_size > 0) {
555 cpage = compressed_pages[i];
556 cur_size = min_t(unsigned long, compressed_size,
559 kaddr = kmap_local_page(cpage);
560 write_extent_buffer(leaf, kaddr, ptr, cur_size);
565 compressed_size -= cur_size;
567 btrfs_set_file_extent_compression(leaf, ei,
570 page = find_get_page(inode->vfs_inode.i_mapping, 0);
571 btrfs_set_file_extent_compression(leaf, ei, 0);
572 kaddr = kmap_local_page(page);
573 write_extent_buffer(leaf, kaddr, ptr, size);
577 btrfs_mark_buffer_dirty(trans, leaf);
578 btrfs_release_path(path);
581 * We align size to sectorsize for inline extents just for simplicity
584 ret = btrfs_inode_set_file_extent_range(inode, 0,
585 ALIGN(size, root->fs_info->sectorsize));
590 * We're an inline extent, so nobody can extend the file past i_size
591 * without locking a page we already have locked.
593 * We must do any i_size and inode updates before we unlock the pages.
594 * Otherwise we could end up racing with unlink.
596 i_size = i_size_read(&inode->vfs_inode);
597 if (update_i_size && size > i_size) {
598 i_size_write(&inode->vfs_inode, size);
601 inode->disk_i_size = i_size;
609 * conditionally insert an inline extent into the file. This
610 * does the checks required to make sure the data is small enough
611 * to fit as an inline extent.
613 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
614 size_t compressed_size,
616 struct page **compressed_pages,
619 struct btrfs_drop_extents_args drop_args = { 0 };
620 struct btrfs_root *root = inode->root;
621 struct btrfs_fs_info *fs_info = root->fs_info;
622 struct btrfs_trans_handle *trans;
623 u64 data_len = (compressed_size ?: size);
625 struct btrfs_path *path;
628 * We can create an inline extent if it ends at or beyond the current
629 * i_size, is no larger than a sector (decompressed), and the (possibly
630 * compressed) data fits in a leaf and the configured maximum inline
633 if (size < i_size_read(&inode->vfs_inode) ||
634 size > fs_info->sectorsize ||
635 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
636 data_len > fs_info->max_inline)
639 path = btrfs_alloc_path();
643 trans = btrfs_join_transaction(root);
645 btrfs_free_path(path);
646 return PTR_ERR(trans);
648 trans->block_rsv = &inode->block_rsv;
650 drop_args.path = path;
652 drop_args.end = fs_info->sectorsize;
653 drop_args.drop_cache = true;
654 drop_args.replace_extent = true;
655 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
656 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
658 btrfs_abort_transaction(trans, ret);
662 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
663 size, compressed_size, compress_type,
664 compressed_pages, update_i_size);
665 if (ret && ret != -ENOSPC) {
666 btrfs_abort_transaction(trans, ret);
668 } else if (ret == -ENOSPC) {
673 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
674 ret = btrfs_update_inode(trans, inode);
675 if (ret && ret != -ENOSPC) {
676 btrfs_abort_transaction(trans, ret);
678 } else if (ret == -ENOSPC) {
683 btrfs_set_inode_full_sync(inode);
686 * Don't forget to free the reserved space, as for inlined extent
687 * it won't count as data extent, free them directly here.
688 * And at reserve time, it's always aligned to page size, so
689 * just free one page here.
691 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE, NULL);
692 btrfs_free_path(path);
693 btrfs_end_transaction(trans);
697 struct async_extent {
702 unsigned long nr_pages;
704 struct list_head list;
708 struct btrfs_inode *inode;
709 struct page *locked_page;
712 blk_opf_t write_flags;
713 struct list_head extents;
714 struct cgroup_subsys_state *blkcg_css;
715 struct btrfs_work work;
716 struct async_cow *async_cow;
721 struct async_chunk chunks[];
724 static noinline int add_async_extent(struct async_chunk *cow,
725 u64 start, u64 ram_size,
728 unsigned long nr_pages,
731 struct async_extent *async_extent;
733 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
734 BUG_ON(!async_extent); /* -ENOMEM */
735 async_extent->start = start;
736 async_extent->ram_size = ram_size;
737 async_extent->compressed_size = compressed_size;
738 async_extent->pages = pages;
739 async_extent->nr_pages = nr_pages;
740 async_extent->compress_type = compress_type;
741 list_add_tail(&async_extent->list, &cow->extents);
746 * Check if the inode needs to be submitted to compression, based on mount
747 * options, defragmentation, properties or heuristics.
749 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
752 struct btrfs_fs_info *fs_info = inode->root->fs_info;
754 if (!btrfs_inode_can_compress(inode)) {
755 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
756 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
761 * Special check for subpage.
763 * We lock the full page then run each delalloc range in the page, thus
764 * for the following case, we will hit some subpage specific corner case:
767 * | |///////| |///////|
770 * In above case, both range A and range B will try to unlock the full
771 * page [0, 64K), causing the one finished later will have page
772 * unlocked already, triggering various page lock requirement BUG_ON()s.
774 * So here we add an artificial limit that subpage compression can only
775 * if the range is fully page aligned.
777 * In theory we only need to ensure the first page is fully covered, but
778 * the tailing partial page will be locked until the full compression
779 * finishes, delaying the write of other range.
781 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
782 * first to prevent any submitted async extent to unlock the full page.
783 * By this, we can ensure for subpage case that only the last async_cow
784 * will unlock the full page.
786 if (fs_info->sectorsize < PAGE_SIZE) {
787 if (!PAGE_ALIGNED(start) ||
788 !PAGE_ALIGNED(end + 1))
793 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
796 if (inode->defrag_compress)
798 /* bad compression ratios */
799 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
801 if (btrfs_test_opt(fs_info, COMPRESS) ||
802 inode->flags & BTRFS_INODE_COMPRESS ||
803 inode->prop_compress)
804 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
808 static inline void inode_should_defrag(struct btrfs_inode *inode,
809 u64 start, u64 end, u64 num_bytes, u32 small_write)
811 /* If this is a small write inside eof, kick off a defrag */
812 if (num_bytes < small_write &&
813 (start > 0 || end + 1 < inode->disk_i_size))
814 btrfs_add_inode_defrag(NULL, inode, small_write);
818 * Work queue call back to started compression on a file and pages.
820 * This is done inside an ordered work queue, and the compression is spread
821 * across many cpus. The actual IO submission is step two, and the ordered work
822 * queue takes care of making sure that happens in the same order things were
823 * put onto the queue by writepages and friends.
825 * If this code finds it can't get good compression, it puts an entry onto the
826 * work queue to write the uncompressed bytes. This makes sure that both
827 * compressed inodes and uncompressed inodes are written in the same order that
828 * the flusher thread sent them down.
830 static void compress_file_range(struct btrfs_work *work)
832 struct async_chunk *async_chunk =
833 container_of(work, struct async_chunk, work);
834 struct btrfs_inode *inode = async_chunk->inode;
835 struct btrfs_fs_info *fs_info = inode->root->fs_info;
836 struct address_space *mapping = inode->vfs_inode.i_mapping;
837 u64 blocksize = fs_info->sectorsize;
838 u64 start = async_chunk->start;
839 u64 end = async_chunk->end;
844 unsigned long nr_pages;
845 unsigned long total_compressed = 0;
846 unsigned long total_in = 0;
849 int compress_type = fs_info->compress_type;
851 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
854 * We need to call clear_page_dirty_for_io on each page in the range.
855 * Otherwise applications with the file mmap'd can wander in and change
856 * the page contents while we are compressing them.
858 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
861 * We need to save i_size before now because it could change in between
862 * us evaluating the size and assigning it. This is because we lock and
863 * unlock the page in truncate and fallocate, and then modify the i_size
866 * The barriers are to emulate READ_ONCE, remove that once i_size_read
870 i_size = i_size_read(&inode->vfs_inode);
872 actual_end = min_t(u64, i_size, end + 1);
875 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
876 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
879 * we don't want to send crud past the end of i_size through
880 * compression, that's just a waste of CPU time. So, if the
881 * end of the file is before the start of our current
882 * requested range of bytes, we bail out to the uncompressed
883 * cleanup code that can deal with all of this.
885 * It isn't really the fastest way to fix things, but this is a
886 * very uncommon corner.
888 if (actual_end <= start)
889 goto cleanup_and_bail_uncompressed;
891 total_compressed = actual_end - start;
894 * Skip compression for a small file range(<=blocksize) that
895 * isn't an inline extent, since it doesn't save disk space at all.
897 if (total_compressed <= blocksize &&
898 (start > 0 || end + 1 < inode->disk_i_size))
899 goto cleanup_and_bail_uncompressed;
902 * For subpage case, we require full page alignment for the sector
904 * Thus we must also check against @actual_end, not just @end.
906 if (blocksize < PAGE_SIZE) {
907 if (!PAGE_ALIGNED(start) ||
908 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
909 goto cleanup_and_bail_uncompressed;
912 total_compressed = min_t(unsigned long, total_compressed,
913 BTRFS_MAX_UNCOMPRESSED);
918 * We do compression for mount -o compress and when the inode has not
919 * been flagged as NOCOMPRESS. This flag can change at any time if we
920 * discover bad compression ratios.
922 if (!inode_need_compress(inode, start, end))
923 goto cleanup_and_bail_uncompressed;
925 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
928 * Memory allocation failure is not a fatal error, we can fall
929 * back to uncompressed code.
931 goto cleanup_and_bail_uncompressed;
934 if (inode->defrag_compress)
935 compress_type = inode->defrag_compress;
936 else if (inode->prop_compress)
937 compress_type = inode->prop_compress;
939 /* Compression level is applied here. */
940 ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4),
941 mapping, start, pages, &nr_pages, &total_in,
944 goto mark_incompressible;
947 * Zero the tail end of the last page, as we might be sending it down
950 poff = offset_in_page(total_compressed);
952 memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff);
955 * Try to create an inline extent.
957 * If we didn't compress the entire range, try to create an uncompressed
958 * inline extent, else a compressed one.
960 * Check cow_file_range() for why we don't even try to create inline
961 * extent for the subpage case.
963 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
964 if (total_in < actual_end) {
965 ret = cow_file_range_inline(inode, actual_end, 0,
966 BTRFS_COMPRESS_NONE, NULL,
969 ret = cow_file_range_inline(inode, actual_end,
971 compress_type, pages,
975 unsigned long clear_flags = EXTENT_DELALLOC |
976 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
977 EXTENT_DO_ACCOUNTING;
980 mapping_set_error(mapping, -EIO);
983 * inline extent creation worked or returned error,
984 * we don't need to create any more async work items.
985 * Unlock and free up our temp pages.
987 * We use DO_ACCOUNTING here because we need the
988 * delalloc_release_metadata to be done _after_ we drop
989 * our outstanding extent for clearing delalloc for this
992 extent_clear_unlock_delalloc(inode, start, end,
996 PAGE_START_WRITEBACK |
1003 * We aren't doing an inline extent. Round the compressed size up to a
1004 * block size boundary so the allocator does sane things.
1006 total_compressed = ALIGN(total_compressed, blocksize);
1009 * One last check to make sure the compression is really a win, compare
1010 * the page count read with the blocks on disk, compression must free at
1013 total_in = round_up(total_in, fs_info->sectorsize);
1014 if (total_compressed + blocksize > total_in)
1015 goto mark_incompressible;
1018 * The async work queues will take care of doing actual allocation on
1019 * disk for these compressed pages, and will submit the bios.
1021 add_async_extent(async_chunk, start, total_in, total_compressed, pages,
1022 nr_pages, compress_type);
1023 if (start + total_in < end) {
1030 mark_incompressible:
1031 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1032 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1033 cleanup_and_bail_uncompressed:
1034 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1035 BTRFS_COMPRESS_NONE);
1038 for (i = 0; i < nr_pages; i++) {
1039 WARN_ON(pages[i]->mapping);
1046 static void free_async_extent_pages(struct async_extent *async_extent)
1050 if (!async_extent->pages)
1053 for (i = 0; i < async_extent->nr_pages; i++) {
1054 WARN_ON(async_extent->pages[i]->mapping);
1055 put_page(async_extent->pages[i]);
1057 kfree(async_extent->pages);
1058 async_extent->nr_pages = 0;
1059 async_extent->pages = NULL;
1062 static void submit_uncompressed_range(struct btrfs_inode *inode,
1063 struct async_extent *async_extent,
1064 struct page *locked_page)
1066 u64 start = async_extent->start;
1067 u64 end = async_extent->start + async_extent->ram_size - 1;
1069 struct writeback_control wbc = {
1070 .sync_mode = WB_SYNC_ALL,
1071 .range_start = start,
1073 .no_cgroup_owner = 1,
1076 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1077 ret = run_delalloc_cow(inode, locked_page, start, end, &wbc, false);
1078 wbc_detach_inode(&wbc);
1080 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1082 const u64 page_start = page_offset(locked_page);
1084 set_page_writeback(locked_page);
1085 end_page_writeback(locked_page);
1086 btrfs_mark_ordered_io_finished(inode, locked_page,
1087 page_start, PAGE_SIZE,
1089 mapping_set_error(locked_page->mapping, ret);
1090 unlock_page(locked_page);
1095 static void submit_one_async_extent(struct async_chunk *async_chunk,
1096 struct async_extent *async_extent,
1099 struct btrfs_inode *inode = async_chunk->inode;
1100 struct extent_io_tree *io_tree = &inode->io_tree;
1101 struct btrfs_root *root = inode->root;
1102 struct btrfs_fs_info *fs_info = root->fs_info;
1103 struct btrfs_ordered_extent *ordered;
1104 struct btrfs_key ins;
1105 struct page *locked_page = NULL;
1106 struct extent_map *em;
1108 u64 start = async_extent->start;
1109 u64 end = async_extent->start + async_extent->ram_size - 1;
1111 if (async_chunk->blkcg_css)
1112 kthread_associate_blkcg(async_chunk->blkcg_css);
1115 * If async_chunk->locked_page is in the async_extent range, we need to
1118 if (async_chunk->locked_page) {
1119 u64 locked_page_start = page_offset(async_chunk->locked_page);
1120 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1122 if (!(start >= locked_page_end || end <= locked_page_start))
1123 locked_page = async_chunk->locked_page;
1125 lock_extent(io_tree, start, end, NULL);
1127 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1128 submit_uncompressed_range(inode, async_extent, locked_page);
1132 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1133 async_extent->compressed_size,
1134 async_extent->compressed_size,
1135 0, *alloc_hint, &ins, 1, 1);
1138 * Here we used to try again by going back to non-compressed
1139 * path for ENOSPC. But we can't reserve space even for
1140 * compressed size, how could it work for uncompressed size
1141 * which requires larger size? So here we directly go error
1147 /* Here we're doing allocation and writeback of the compressed pages */
1148 em = create_io_em(inode, start,
1149 async_extent->ram_size, /* len */
1150 start, /* orig_start */
1151 ins.objectid, /* block_start */
1152 ins.offset, /* block_len */
1153 ins.offset, /* orig_block_len */
1154 async_extent->ram_size, /* ram_bytes */
1155 async_extent->compress_type,
1156 BTRFS_ORDERED_COMPRESSED);
1159 goto out_free_reserve;
1161 free_extent_map(em);
1163 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1164 async_extent->ram_size, /* num_bytes */
1165 async_extent->ram_size, /* ram_bytes */
1166 ins.objectid, /* disk_bytenr */
1167 ins.offset, /* disk_num_bytes */
1169 1 << BTRFS_ORDERED_COMPRESSED,
1170 async_extent->compress_type);
1171 if (IS_ERR(ordered)) {
1172 btrfs_drop_extent_map_range(inode, start, end, false);
1173 ret = PTR_ERR(ordered);
1174 goto out_free_reserve;
1176 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1178 /* Clear dirty, set writeback and unlock the pages. */
1179 extent_clear_unlock_delalloc(inode, start, end,
1180 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1181 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1182 btrfs_submit_compressed_write(ordered,
1183 async_extent->pages, /* compressed_pages */
1184 async_extent->nr_pages,
1185 async_chunk->write_flags, true);
1186 *alloc_hint = ins.objectid + ins.offset;
1188 if (async_chunk->blkcg_css)
1189 kthread_associate_blkcg(NULL);
1190 kfree(async_extent);
1194 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1195 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1197 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1198 extent_clear_unlock_delalloc(inode, start, end,
1199 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1200 EXTENT_DELALLOC_NEW |
1201 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1202 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1203 PAGE_END_WRITEBACK);
1204 free_async_extent_pages(async_extent);
1205 if (async_chunk->blkcg_css)
1206 kthread_associate_blkcg(NULL);
1207 btrfs_debug(fs_info,
1208 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1209 root->root_key.objectid, btrfs_ino(inode), start,
1210 async_extent->ram_size, ret);
1211 kfree(async_extent);
1214 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1217 struct extent_map_tree *em_tree = &inode->extent_tree;
1218 struct extent_map *em;
1221 read_lock(&em_tree->lock);
1222 em = search_extent_mapping(em_tree, start, num_bytes);
1225 * if block start isn't an actual block number then find the
1226 * first block in this inode and use that as a hint. If that
1227 * block is also bogus then just don't worry about it.
1229 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1230 free_extent_map(em);
1231 em = search_extent_mapping(em_tree, 0, 0);
1232 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1233 alloc_hint = em->block_start;
1235 free_extent_map(em);
1237 alloc_hint = em->block_start;
1238 free_extent_map(em);
1241 read_unlock(&em_tree->lock);
1247 * when extent_io.c finds a delayed allocation range in the file,
1248 * the call backs end up in this code. The basic idea is to
1249 * allocate extents on disk for the range, and create ordered data structs
1250 * in ram to track those extents.
1252 * locked_page is the page that writepage had locked already. We use
1253 * it to make sure we don't do extra locks or unlocks.
1255 * When this function fails, it unlocks all pages except @locked_page.
1257 * When this function successfully creates an inline extent, it returns 1 and
1258 * unlocks all pages including locked_page and starts I/O on them.
1259 * (In reality inline extents are limited to a single page, so locked_page is
1260 * the only page handled anyway).
1262 * When this function succeed and creates a normal extent, the page locking
1263 * status depends on the passed in flags:
1265 * - If @keep_locked is set, all pages are kept locked.
1266 * - Else all pages except for @locked_page are unlocked.
1268 * When a failure happens in the second or later iteration of the
1269 * while-loop, the ordered extents created in previous iterations are kept
1270 * intact. So, the caller must clean them up by calling
1271 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1274 static noinline int cow_file_range(struct btrfs_inode *inode,
1275 struct page *locked_page, u64 start, u64 end,
1277 bool keep_locked, bool no_inline)
1279 struct btrfs_root *root = inode->root;
1280 struct btrfs_fs_info *fs_info = root->fs_info;
1282 u64 orig_start = start;
1284 unsigned long ram_size;
1285 u64 cur_alloc_size = 0;
1287 u64 blocksize = fs_info->sectorsize;
1288 struct btrfs_key ins;
1289 struct extent_map *em;
1290 unsigned clear_bits;
1291 unsigned long page_ops;
1292 bool extent_reserved = false;
1295 if (btrfs_is_free_space_inode(inode)) {
1300 num_bytes = ALIGN(end - start + 1, blocksize);
1301 num_bytes = max(blocksize, num_bytes);
1302 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1304 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1307 * Due to the page size limit, for subpage we can only trigger the
1308 * writeback for the dirty sectors of page, that means data writeback
1309 * is doing more writeback than what we want.
1311 * This is especially unexpected for some call sites like fallocate,
1312 * where we only increase i_size after everything is done.
1313 * This means we can trigger inline extent even if we didn't want to.
1314 * So here we skip inline extent creation completely.
1316 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1317 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1320 /* lets try to make an inline extent */
1321 ret = cow_file_range_inline(inode, actual_end, 0,
1322 BTRFS_COMPRESS_NONE, NULL, false);
1325 * We use DO_ACCOUNTING here because we need the
1326 * delalloc_release_metadata to be run _after_ we drop
1327 * our outstanding extent for clearing delalloc for this
1330 extent_clear_unlock_delalloc(inode, start, end,
1332 EXTENT_LOCKED | EXTENT_DELALLOC |
1333 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1334 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1335 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1337 * locked_page is locked by the caller of
1338 * writepage_delalloc(), not locked by
1339 * __process_pages_contig().
1341 * We can't let __process_pages_contig() to unlock it,
1342 * as it doesn't have any subpage::writers recorded.
1344 * Here we manually unlock the page, since the caller
1345 * can't determine if it's an inline extent or a
1346 * compressed extent.
1348 unlock_page(locked_page);
1351 } else if (ret < 0) {
1356 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1359 * Relocation relies on the relocated extents to have exactly the same
1360 * size as the original extents. Normally writeback for relocation data
1361 * extents follows a NOCOW path because relocation preallocates the
1362 * extents. However, due to an operation such as scrub turning a block
1363 * group to RO mode, it may fallback to COW mode, so we must make sure
1364 * an extent allocated during COW has exactly the requested size and can
1365 * not be split into smaller extents, otherwise relocation breaks and
1366 * fails during the stage where it updates the bytenr of file extent
1369 if (btrfs_is_data_reloc_root(root))
1370 min_alloc_size = num_bytes;
1372 min_alloc_size = fs_info->sectorsize;
1374 while (num_bytes > 0) {
1375 struct btrfs_ordered_extent *ordered;
1377 cur_alloc_size = num_bytes;
1378 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1379 min_alloc_size, 0, alloc_hint,
1381 if (ret == -EAGAIN) {
1383 * btrfs_reserve_extent only returns -EAGAIN for zoned
1384 * file systems, which is an indication that there are
1385 * no active zones to allocate from at the moment.
1387 * If this is the first loop iteration, wait for at
1388 * least one zone to finish before retrying the
1389 * allocation. Otherwise ask the caller to write out
1390 * the already allocated blocks before coming back to
1391 * us, or return -ENOSPC if it can't handle retries.
1393 ASSERT(btrfs_is_zoned(fs_info));
1394 if (start == orig_start) {
1395 wait_on_bit_io(&inode->root->fs_info->flags,
1396 BTRFS_FS_NEED_ZONE_FINISH,
1397 TASK_UNINTERRUPTIBLE);
1401 *done_offset = start - 1;
1408 cur_alloc_size = ins.offset;
1409 extent_reserved = true;
1411 ram_size = ins.offset;
1412 em = create_io_em(inode, start, ins.offset, /* len */
1413 start, /* orig_start */
1414 ins.objectid, /* block_start */
1415 ins.offset, /* block_len */
1416 ins.offset, /* orig_block_len */
1417 ram_size, /* ram_bytes */
1418 BTRFS_COMPRESS_NONE, /* compress_type */
1419 BTRFS_ORDERED_REGULAR /* type */);
1424 free_extent_map(em);
1426 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1427 ram_size, ins.objectid, cur_alloc_size,
1428 0, 1 << BTRFS_ORDERED_REGULAR,
1429 BTRFS_COMPRESS_NONE);
1430 if (IS_ERR(ordered)) {
1431 ret = PTR_ERR(ordered);
1432 goto out_drop_extent_cache;
1435 if (btrfs_is_data_reloc_root(root)) {
1436 ret = btrfs_reloc_clone_csums(ordered);
1439 * Only drop cache here, and process as normal.
1441 * We must not allow extent_clear_unlock_delalloc()
1442 * at out_unlock label to free meta of this ordered
1443 * extent, as its meta should be freed by
1444 * btrfs_finish_ordered_io().
1446 * So we must continue until @start is increased to
1447 * skip current ordered extent.
1450 btrfs_drop_extent_map_range(inode, start,
1451 start + ram_size - 1,
1454 btrfs_put_ordered_extent(ordered);
1456 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1459 * We're not doing compressed IO, don't unlock the first page
1460 * (which the caller expects to stay locked), don't clear any
1461 * dirty bits and don't set any writeback bits
1463 * Do set the Ordered (Private2) bit so we know this page was
1464 * properly setup for writepage.
1466 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1467 page_ops |= PAGE_SET_ORDERED;
1469 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1471 EXTENT_LOCKED | EXTENT_DELALLOC,
1473 if (num_bytes < cur_alloc_size)
1476 num_bytes -= cur_alloc_size;
1477 alloc_hint = ins.objectid + ins.offset;
1478 start += cur_alloc_size;
1479 extent_reserved = false;
1482 * btrfs_reloc_clone_csums() error, since start is increased
1483 * extent_clear_unlock_delalloc() at out_unlock label won't
1484 * free metadata of current ordered extent, we're OK to exit.
1494 out_drop_extent_cache:
1495 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1497 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1498 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1501 * Now, we have three regions to clean up:
1503 * |-------(1)----|---(2)---|-------------(3)----------|
1504 * `- orig_start `- start `- start + cur_alloc_size `- end
1506 * We process each region below.
1509 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1510 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1511 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1514 * For the range (1). We have already instantiated the ordered extents
1515 * for this region. They are cleaned up by
1516 * btrfs_cleanup_ordered_extents() in e.g,
1517 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1518 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1519 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1522 * However, in case of @keep_locked, we still need to unlock the pages
1523 * (except @locked_page) to ensure all the pages are unlocked.
1525 if (keep_locked && orig_start < start) {
1527 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1528 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1529 locked_page, 0, page_ops);
1533 * For the range (2). If we reserved an extent for our delalloc range
1534 * (or a subrange) and failed to create the respective ordered extent,
1535 * then it means that when we reserved the extent we decremented the
1536 * extent's size from the data space_info's bytes_may_use counter and
1537 * incremented the space_info's bytes_reserved counter by the same
1538 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1539 * to decrement again the data space_info's bytes_may_use counter,
1540 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1542 if (extent_reserved) {
1543 extent_clear_unlock_delalloc(inode, start,
1544 start + cur_alloc_size - 1,
1548 start += cur_alloc_size;
1552 * For the range (3). We never touched the region. In addition to the
1553 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1554 * space_info's bytes_may_use counter, reserved in
1555 * btrfs_check_data_free_space().
1558 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1559 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1560 clear_bits, page_ops);
1566 * Phase two of compressed writeback. This is the ordered portion of the code,
1567 * which only gets called in the order the work was queued. We walk all the
1568 * async extents created by compress_file_range and send them down to the disk.
1570 * If called with @do_free == true then it'll try to finish the work and free
1571 * the work struct eventually.
1573 static noinline void submit_compressed_extents(struct btrfs_work *work, bool do_free)
1575 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1577 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1578 struct async_extent *async_extent;
1579 unsigned long nr_pages;
1583 struct async_chunk *async_chunk;
1584 struct async_cow *async_cow;
1586 async_chunk = container_of(work, struct async_chunk, work);
1587 btrfs_add_delayed_iput(async_chunk->inode);
1588 if (async_chunk->blkcg_css)
1589 css_put(async_chunk->blkcg_css);
1591 async_cow = async_chunk->async_cow;
1592 if (atomic_dec_and_test(&async_cow->num_chunks))
1597 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1600 while (!list_empty(&async_chunk->extents)) {
1601 async_extent = list_entry(async_chunk->extents.next,
1602 struct async_extent, list);
1603 list_del(&async_extent->list);
1604 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1607 /* atomic_sub_return implies a barrier */
1608 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1610 cond_wake_up_nomb(&fs_info->async_submit_wait);
1613 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1614 struct page *locked_page, u64 start,
1615 u64 end, struct writeback_control *wbc)
1617 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1618 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1619 struct async_cow *ctx;
1620 struct async_chunk *async_chunk;
1621 unsigned long nr_pages;
1622 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1625 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1627 nofs_flag = memalloc_nofs_save();
1628 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1629 memalloc_nofs_restore(nofs_flag);
1633 unlock_extent(&inode->io_tree, start, end, NULL);
1634 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1636 async_chunk = ctx->chunks;
1637 atomic_set(&ctx->num_chunks, num_chunks);
1639 for (i = 0; i < num_chunks; i++) {
1640 u64 cur_end = min(end, start + SZ_512K - 1);
1643 * igrab is called higher up in the call chain, take only the
1644 * lightweight reference for the callback lifetime
1646 ihold(&inode->vfs_inode);
1647 async_chunk[i].async_cow = ctx;
1648 async_chunk[i].inode = inode;
1649 async_chunk[i].start = start;
1650 async_chunk[i].end = cur_end;
1651 async_chunk[i].write_flags = write_flags;
1652 INIT_LIST_HEAD(&async_chunk[i].extents);
1655 * The locked_page comes all the way from writepage and its
1656 * the original page we were actually given. As we spread
1657 * this large delalloc region across multiple async_chunk
1658 * structs, only the first struct needs a pointer to locked_page
1660 * This way we don't need racey decisions about who is supposed
1665 * Depending on the compressibility, the pages might or
1666 * might not go through async. We want all of them to
1667 * be accounted against wbc once. Let's do it here
1668 * before the paths diverge. wbc accounting is used
1669 * only for foreign writeback detection and doesn't
1670 * need full accuracy. Just account the whole thing
1671 * against the first page.
1673 wbc_account_cgroup_owner(wbc, locked_page,
1675 async_chunk[i].locked_page = locked_page;
1678 async_chunk[i].locked_page = NULL;
1681 if (blkcg_css != blkcg_root_css) {
1683 async_chunk[i].blkcg_css = blkcg_css;
1684 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1686 async_chunk[i].blkcg_css = NULL;
1689 btrfs_init_work(&async_chunk[i].work, compress_file_range,
1690 submit_compressed_extents);
1692 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1693 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1695 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1697 start = cur_end + 1;
1703 * Run the delalloc range from start to end, and write back any dirty pages
1704 * covered by the range.
1706 static noinline int run_delalloc_cow(struct btrfs_inode *inode,
1707 struct page *locked_page, u64 start,
1708 u64 end, struct writeback_control *wbc,
1711 u64 done_offset = end;
1714 while (start <= end) {
1715 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1719 extent_write_locked_range(&inode->vfs_inode, locked_page, start,
1720 done_offset, wbc, pages_dirty);
1721 start = done_offset + 1;
1727 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1728 u64 bytenr, u64 num_bytes, bool nowait)
1730 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1731 struct btrfs_ordered_sum *sums;
1735 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1737 if (ret == 0 && list_empty(&list))
1740 while (!list_empty(&list)) {
1741 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1742 list_del(&sums->list);
1750 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1751 const u64 start, const u64 end)
1753 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1754 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1755 const u64 range_bytes = end + 1 - start;
1756 struct extent_io_tree *io_tree = &inode->io_tree;
1757 u64 range_start = start;
1762 * If EXTENT_NORESERVE is set it means that when the buffered write was
1763 * made we had not enough available data space and therefore we did not
1764 * reserve data space for it, since we though we could do NOCOW for the
1765 * respective file range (either there is prealloc extent or the inode
1766 * has the NOCOW bit set).
1768 * However when we need to fallback to COW mode (because for example the
1769 * block group for the corresponding extent was turned to RO mode by a
1770 * scrub or relocation) we need to do the following:
1772 * 1) We increment the bytes_may_use counter of the data space info.
1773 * If COW succeeds, it allocates a new data extent and after doing
1774 * that it decrements the space info's bytes_may_use counter and
1775 * increments its bytes_reserved counter by the same amount (we do
1776 * this at btrfs_add_reserved_bytes()). So we need to increment the
1777 * bytes_may_use counter to compensate (when space is reserved at
1778 * buffered write time, the bytes_may_use counter is incremented);
1780 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1781 * that if the COW path fails for any reason, it decrements (through
1782 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1783 * data space info, which we incremented in the step above.
1785 * If we need to fallback to cow and the inode corresponds to a free
1786 * space cache inode or an inode of the data relocation tree, we must
1787 * also increment bytes_may_use of the data space_info for the same
1788 * reason. Space caches and relocated data extents always get a prealloc
1789 * extent for them, however scrub or balance may have set the block
1790 * group that contains that extent to RO mode and therefore force COW
1791 * when starting writeback.
1793 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1794 EXTENT_NORESERVE, 0, NULL);
1795 if (count > 0 || is_space_ino || is_reloc_ino) {
1797 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1798 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1800 if (is_space_ino || is_reloc_ino)
1801 bytes = range_bytes;
1803 spin_lock(&sinfo->lock);
1804 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1805 spin_unlock(&sinfo->lock);
1808 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1813 * Don't try to create inline extents, as a mix of inline extent that
1814 * is written out and unlocked directly and a normal NOCOW extent
1817 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1822 struct can_nocow_file_extent_args {
1825 /* Start file offset of the range we want to NOCOW. */
1827 /* End file offset (inclusive) of the range we want to NOCOW. */
1829 bool writeback_path;
1832 * Free the path passed to can_nocow_file_extent() once it's not needed
1837 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1842 /* Number of bytes that can be written to in NOCOW mode. */
1847 * Check if we can NOCOW the file extent that the path points to.
1848 * This function may return with the path released, so the caller should check
1849 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1851 * Returns: < 0 on error
1852 * 0 if we can not NOCOW
1855 static int can_nocow_file_extent(struct btrfs_path *path,
1856 struct btrfs_key *key,
1857 struct btrfs_inode *inode,
1858 struct can_nocow_file_extent_args *args)
1860 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1861 struct extent_buffer *leaf = path->nodes[0];
1862 struct btrfs_root *root = inode->root;
1863 struct btrfs_file_extent_item *fi;
1868 bool nowait = path->nowait;
1870 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1871 extent_type = btrfs_file_extent_type(leaf, fi);
1873 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1876 /* Can't access these fields unless we know it's not an inline extent. */
1877 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1878 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1879 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1881 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1882 extent_type == BTRFS_FILE_EXTENT_REG)
1886 * If the extent was created before the generation where the last snapshot
1887 * for its subvolume was created, then this implies the extent is shared,
1888 * hence we must COW.
1890 if (!args->strict &&
1891 btrfs_file_extent_generation(leaf, fi) <=
1892 btrfs_root_last_snapshot(&root->root_item))
1895 /* An explicit hole, must COW. */
1896 if (args->disk_bytenr == 0)
1899 /* Compressed/encrypted/encoded extents must be COWed. */
1900 if (btrfs_file_extent_compression(leaf, fi) ||
1901 btrfs_file_extent_encryption(leaf, fi) ||
1902 btrfs_file_extent_other_encoding(leaf, fi))
1905 extent_end = btrfs_file_extent_end(path);
1908 * The following checks can be expensive, as they need to take other
1909 * locks and do btree or rbtree searches, so release the path to avoid
1910 * blocking other tasks for too long.
1912 btrfs_release_path(path);
1914 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1915 key->offset - args->extent_offset,
1916 args->disk_bytenr, args->strict, path);
1917 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1921 if (args->free_path) {
1923 * We don't need the path anymore, plus through the
1924 * csum_exist_in_range() call below we will end up allocating
1925 * another path. So free the path to avoid unnecessary extra
1928 btrfs_free_path(path);
1932 /* If there are pending snapshots for this root, we must COW. */
1933 if (args->writeback_path && !is_freespace_inode &&
1934 atomic_read(&root->snapshot_force_cow))
1937 args->disk_bytenr += args->extent_offset;
1938 args->disk_bytenr += args->start - key->offset;
1939 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1942 * Force COW if csums exist in the range. This ensures that csums for a
1943 * given extent are either valid or do not exist.
1945 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1947 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1953 if (args->free_path && path)
1954 btrfs_free_path(path);
1956 return ret < 0 ? ret : can_nocow;
1960 * when nowcow writeback call back. This checks for snapshots or COW copies
1961 * of the extents that exist in the file, and COWs the file as required.
1963 * If no cow copies or snapshots exist, we write directly to the existing
1966 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1967 struct page *locked_page,
1968 const u64 start, const u64 end)
1970 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1971 struct btrfs_root *root = inode->root;
1972 struct btrfs_path *path;
1973 u64 cow_start = (u64)-1;
1974 u64 cur_offset = start;
1976 bool check_prev = true;
1977 u64 ino = btrfs_ino(inode);
1978 struct can_nocow_file_extent_args nocow_args = { 0 };
1981 * Normally on a zoned device we're only doing COW writes, but in case
1982 * of relocation on a zoned filesystem serializes I/O so that we're only
1983 * writing sequentially and can end up here as well.
1985 ASSERT(!btrfs_is_zoned(fs_info) || btrfs_is_data_reloc_root(root));
1987 path = btrfs_alloc_path();
1993 nocow_args.end = end;
1994 nocow_args.writeback_path = true;
1997 struct btrfs_block_group *nocow_bg = NULL;
1998 struct btrfs_ordered_extent *ordered;
1999 struct btrfs_key found_key;
2000 struct btrfs_file_extent_item *fi;
2001 struct extent_buffer *leaf;
2008 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2014 * If there is no extent for our range when doing the initial
2015 * search, then go back to the previous slot as it will be the
2016 * one containing the search offset
2018 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2019 leaf = path->nodes[0];
2020 btrfs_item_key_to_cpu(leaf, &found_key,
2021 path->slots[0] - 1);
2022 if (found_key.objectid == ino &&
2023 found_key.type == BTRFS_EXTENT_DATA_KEY)
2028 /* Go to next leaf if we have exhausted the current one */
2029 leaf = path->nodes[0];
2030 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2031 ret = btrfs_next_leaf(root, path);
2036 leaf = path->nodes[0];
2039 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2041 /* Didn't find anything for our INO */
2042 if (found_key.objectid > ino)
2045 * Keep searching until we find an EXTENT_ITEM or there are no
2046 * more extents for this inode
2048 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2049 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2054 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2055 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2056 found_key.offset > end)
2060 * If the found extent starts after requested offset, then
2061 * adjust extent_end to be right before this extent begins
2063 if (found_key.offset > cur_offset) {
2064 extent_end = found_key.offset;
2070 * Found extent which begins before our range and potentially
2073 fi = btrfs_item_ptr(leaf, path->slots[0],
2074 struct btrfs_file_extent_item);
2075 extent_type = btrfs_file_extent_type(leaf, fi);
2076 /* If this is triggered then we have a memory corruption. */
2077 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2078 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2082 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2083 extent_end = btrfs_file_extent_end(path);
2086 * If the extent we got ends before our current offset, skip to
2089 if (extent_end <= cur_offset) {
2094 nocow_args.start = cur_offset;
2095 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2102 nocow_bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2106 * If we can't perform NOCOW writeback for the range,
2107 * then record the beginning of the range that needs to
2108 * be COWed. It will be written out before the next
2109 * NOCOW range if we find one, or when exiting this
2112 if (cow_start == (u64)-1)
2113 cow_start = cur_offset;
2114 cur_offset = extent_end;
2115 if (cur_offset > end)
2117 if (!path->nodes[0])
2124 * COW range from cow_start to found_key.offset - 1. As the key
2125 * will contain the beginning of the first extent that can be
2126 * NOCOW, following one which needs to be COW'ed
2128 if (cow_start != (u64)-1) {
2129 ret = fallback_to_cow(inode, locked_page,
2130 cow_start, found_key.offset - 1);
2131 cow_start = (u64)-1;
2133 btrfs_dec_nocow_writers(nocow_bg);
2138 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2139 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2141 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2142 struct extent_map *em;
2144 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2146 nocow_args.disk_bytenr, /* block_start */
2147 nocow_args.num_bytes, /* block_len */
2148 nocow_args.disk_num_bytes, /* orig_block_len */
2149 ram_bytes, BTRFS_COMPRESS_NONE,
2150 BTRFS_ORDERED_PREALLOC);
2152 btrfs_dec_nocow_writers(nocow_bg);
2156 free_extent_map(em);
2159 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2160 nocow_args.num_bytes, nocow_args.num_bytes,
2161 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2163 ? (1 << BTRFS_ORDERED_PREALLOC)
2164 : (1 << BTRFS_ORDERED_NOCOW),
2165 BTRFS_COMPRESS_NONE);
2166 btrfs_dec_nocow_writers(nocow_bg);
2167 if (IS_ERR(ordered)) {
2169 btrfs_drop_extent_map_range(inode, cur_offset,
2172 ret = PTR_ERR(ordered);
2176 if (btrfs_is_data_reloc_root(root))
2178 * Error handled later, as we must prevent
2179 * extent_clear_unlock_delalloc() in error handler
2180 * from freeing metadata of created ordered extent.
2182 ret = btrfs_reloc_clone_csums(ordered);
2183 btrfs_put_ordered_extent(ordered);
2185 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2186 locked_page, EXTENT_LOCKED |
2188 EXTENT_CLEAR_DATA_RESV,
2189 PAGE_UNLOCK | PAGE_SET_ORDERED);
2191 cur_offset = extent_end;
2194 * btrfs_reloc_clone_csums() error, now we're OK to call error
2195 * handler, as metadata for created ordered extent will only
2196 * be freed by btrfs_finish_ordered_io().
2200 if (cur_offset > end)
2203 btrfs_release_path(path);
2205 if (cur_offset <= end && cow_start == (u64)-1)
2206 cow_start = cur_offset;
2208 if (cow_start != (u64)-1) {
2210 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2211 cow_start = (u64)-1;
2216 btrfs_free_path(path);
2221 * If an error happened while a COW region is outstanding, cur_offset
2222 * needs to be reset to cow_start to ensure the COW region is unlocked
2225 if (cow_start != (u64)-1)
2226 cur_offset = cow_start;
2227 if (cur_offset < end)
2228 extent_clear_unlock_delalloc(inode, cur_offset, end,
2229 locked_page, EXTENT_LOCKED |
2230 EXTENT_DELALLOC | EXTENT_DEFRAG |
2231 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2232 PAGE_START_WRITEBACK |
2233 PAGE_END_WRITEBACK);
2234 btrfs_free_path(path);
2238 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2240 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2241 if (inode->defrag_bytes &&
2242 test_range_bit_exists(&inode->io_tree, start, end, EXTENT_DEFRAG))
2250 * Function to process delayed allocation (create CoW) for ranges which are
2251 * being touched for the first time.
2253 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2254 u64 start, u64 end, struct writeback_control *wbc)
2256 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2260 * The range must cover part of the @locked_page, or a return of 1
2261 * can confuse the caller.
2263 ASSERT(!(end <= page_offset(locked_page) ||
2264 start >= page_offset(locked_page) + PAGE_SIZE));
2266 if (should_nocow(inode, start, end)) {
2267 ret = run_delalloc_nocow(inode, locked_page, start, end);
2271 if (btrfs_inode_can_compress(inode) &&
2272 inode_need_compress(inode, start, end) &&
2273 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2277 ret = run_delalloc_cow(inode, locked_page, start, end, wbc,
2280 ret = cow_file_range(inode, locked_page, start, end, NULL,
2285 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2290 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2291 struct extent_state *orig, u64 split)
2293 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2296 /* not delalloc, ignore it */
2297 if (!(orig->state & EXTENT_DELALLOC))
2300 size = orig->end - orig->start + 1;
2301 if (size > fs_info->max_extent_size) {
2306 * See the explanation in btrfs_merge_delalloc_extent, the same
2307 * applies here, just in reverse.
2309 new_size = orig->end - split + 1;
2310 num_extents = count_max_extents(fs_info, new_size);
2311 new_size = split - orig->start;
2312 num_extents += count_max_extents(fs_info, new_size);
2313 if (count_max_extents(fs_info, size) >= num_extents)
2317 spin_lock(&inode->lock);
2318 btrfs_mod_outstanding_extents(inode, 1);
2319 spin_unlock(&inode->lock);
2323 * Handle merged delayed allocation extents so we can keep track of new extents
2324 * that are just merged onto old extents, such as when we are doing sequential
2325 * writes, so we can properly account for the metadata space we'll need.
2327 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2328 struct extent_state *other)
2330 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2331 u64 new_size, old_size;
2334 /* not delalloc, ignore it */
2335 if (!(other->state & EXTENT_DELALLOC))
2338 if (new->start > other->start)
2339 new_size = new->end - other->start + 1;
2341 new_size = other->end - new->start + 1;
2343 /* we're not bigger than the max, unreserve the space and go */
2344 if (new_size <= fs_info->max_extent_size) {
2345 spin_lock(&inode->lock);
2346 btrfs_mod_outstanding_extents(inode, -1);
2347 spin_unlock(&inode->lock);
2352 * We have to add up either side to figure out how many extents were
2353 * accounted for before we merged into one big extent. If the number of
2354 * extents we accounted for is <= the amount we need for the new range
2355 * then we can return, otherwise drop. Think of it like this
2359 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2360 * need 2 outstanding extents, on one side we have 1 and the other side
2361 * we have 1 so they are == and we can return. But in this case
2363 * [MAX_SIZE+4k][MAX_SIZE+4k]
2365 * Each range on their own accounts for 2 extents, but merged together
2366 * they are only 3 extents worth of accounting, so we need to drop in
2369 old_size = other->end - other->start + 1;
2370 num_extents = count_max_extents(fs_info, old_size);
2371 old_size = new->end - new->start + 1;
2372 num_extents += count_max_extents(fs_info, old_size);
2373 if (count_max_extents(fs_info, new_size) >= num_extents)
2376 spin_lock(&inode->lock);
2377 btrfs_mod_outstanding_extents(inode, -1);
2378 spin_unlock(&inode->lock);
2381 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2382 struct btrfs_inode *inode)
2384 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2386 spin_lock(&root->delalloc_lock);
2387 if (list_empty(&inode->delalloc_inodes)) {
2388 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2389 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2390 root->nr_delalloc_inodes++;
2391 if (root->nr_delalloc_inodes == 1) {
2392 spin_lock(&fs_info->delalloc_root_lock);
2393 BUG_ON(!list_empty(&root->delalloc_root));
2394 list_add_tail(&root->delalloc_root,
2395 &fs_info->delalloc_roots);
2396 spin_unlock(&fs_info->delalloc_root_lock);
2399 spin_unlock(&root->delalloc_lock);
2402 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2403 struct btrfs_inode *inode)
2405 struct btrfs_fs_info *fs_info = root->fs_info;
2407 if (!list_empty(&inode->delalloc_inodes)) {
2408 list_del_init(&inode->delalloc_inodes);
2409 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2410 &inode->runtime_flags);
2411 root->nr_delalloc_inodes--;
2412 if (!root->nr_delalloc_inodes) {
2413 ASSERT(list_empty(&root->delalloc_inodes));
2414 spin_lock(&fs_info->delalloc_root_lock);
2415 BUG_ON(list_empty(&root->delalloc_root));
2416 list_del_init(&root->delalloc_root);
2417 spin_unlock(&fs_info->delalloc_root_lock);
2422 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2423 struct btrfs_inode *inode)
2425 spin_lock(&root->delalloc_lock);
2426 __btrfs_del_delalloc_inode(root, inode);
2427 spin_unlock(&root->delalloc_lock);
2431 * Properly track delayed allocation bytes in the inode and to maintain the
2432 * list of inodes that have pending delalloc work to be done.
2434 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2437 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2439 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2442 * set_bit and clear bit hooks normally require _irqsave/restore
2443 * but in this case, we are only testing for the DELALLOC
2444 * bit, which is only set or cleared with irqs on
2446 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2447 struct btrfs_root *root = inode->root;
2448 u64 len = state->end + 1 - state->start;
2449 u32 num_extents = count_max_extents(fs_info, len);
2450 bool do_list = !btrfs_is_free_space_inode(inode);
2452 spin_lock(&inode->lock);
2453 btrfs_mod_outstanding_extents(inode, num_extents);
2454 spin_unlock(&inode->lock);
2456 /* For sanity tests */
2457 if (btrfs_is_testing(fs_info))
2460 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2461 fs_info->delalloc_batch);
2462 spin_lock(&inode->lock);
2463 inode->delalloc_bytes += len;
2464 if (bits & EXTENT_DEFRAG)
2465 inode->defrag_bytes += len;
2466 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2467 &inode->runtime_flags))
2468 btrfs_add_delalloc_inodes(root, inode);
2469 spin_unlock(&inode->lock);
2472 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2473 (bits & EXTENT_DELALLOC_NEW)) {
2474 spin_lock(&inode->lock);
2475 inode->new_delalloc_bytes += state->end + 1 - state->start;
2476 spin_unlock(&inode->lock);
2481 * Once a range is no longer delalloc this function ensures that proper
2482 * accounting happens.
2484 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2485 struct extent_state *state, u32 bits)
2487 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2488 u64 len = state->end + 1 - state->start;
2489 u32 num_extents = count_max_extents(fs_info, len);
2491 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2492 spin_lock(&inode->lock);
2493 inode->defrag_bytes -= len;
2494 spin_unlock(&inode->lock);
2498 * set_bit and clear bit hooks normally require _irqsave/restore
2499 * but in this case, we are only testing for the DELALLOC
2500 * bit, which is only set or cleared with irqs on
2502 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2503 struct btrfs_root *root = inode->root;
2504 bool do_list = !btrfs_is_free_space_inode(inode);
2506 spin_lock(&inode->lock);
2507 btrfs_mod_outstanding_extents(inode, -num_extents);
2508 spin_unlock(&inode->lock);
2511 * We don't reserve metadata space for space cache inodes so we
2512 * don't need to call delalloc_release_metadata if there is an
2515 if (bits & EXTENT_CLEAR_META_RESV &&
2516 root != fs_info->tree_root)
2517 btrfs_delalloc_release_metadata(inode, len, false);
2519 /* For sanity tests. */
2520 if (btrfs_is_testing(fs_info))
2523 if (!btrfs_is_data_reloc_root(root) &&
2524 do_list && !(state->state & EXTENT_NORESERVE) &&
2525 (bits & EXTENT_CLEAR_DATA_RESV))
2526 btrfs_free_reserved_data_space_noquota(fs_info, len);
2528 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2529 fs_info->delalloc_batch);
2530 spin_lock(&inode->lock);
2531 inode->delalloc_bytes -= len;
2532 if (do_list && inode->delalloc_bytes == 0 &&
2533 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2534 &inode->runtime_flags))
2535 btrfs_del_delalloc_inode(root, inode);
2536 spin_unlock(&inode->lock);
2539 if ((state->state & EXTENT_DELALLOC_NEW) &&
2540 (bits & EXTENT_DELALLOC_NEW)) {
2541 spin_lock(&inode->lock);
2542 ASSERT(inode->new_delalloc_bytes >= len);
2543 inode->new_delalloc_bytes -= len;
2544 if (bits & EXTENT_ADD_INODE_BYTES)
2545 inode_add_bytes(&inode->vfs_inode, len);
2546 spin_unlock(&inode->lock);
2550 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2551 struct btrfs_ordered_extent *ordered)
2553 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2554 u64 len = bbio->bio.bi_iter.bi_size;
2555 struct btrfs_ordered_extent *new;
2558 /* Must always be called for the beginning of an ordered extent. */
2559 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2562 /* No need to split if the ordered extent covers the entire bio. */
2563 if (ordered->disk_num_bytes == len) {
2564 refcount_inc(&ordered->refs);
2565 bbio->ordered = ordered;
2570 * Don't split the extent_map for NOCOW extents, as we're writing into
2571 * a pre-existing one.
2573 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2574 ret = split_extent_map(bbio->inode, bbio->file_offset,
2575 ordered->num_bytes, len,
2576 ordered->disk_bytenr);
2581 new = btrfs_split_ordered_extent(ordered, len);
2583 return PTR_ERR(new);
2584 bbio->ordered = new;
2589 * given a list of ordered sums record them in the inode. This happens
2590 * at IO completion time based on sums calculated at bio submission time.
2592 static int add_pending_csums(struct btrfs_trans_handle *trans,
2593 struct list_head *list)
2595 struct btrfs_ordered_sum *sum;
2596 struct btrfs_root *csum_root = NULL;
2599 list_for_each_entry(sum, list, list) {
2600 trans->adding_csums = true;
2602 csum_root = btrfs_csum_root(trans->fs_info,
2604 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2605 trans->adding_csums = false;
2612 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2615 struct extent_state **cached_state)
2617 u64 search_start = start;
2618 const u64 end = start + len - 1;
2620 while (search_start < end) {
2621 const u64 search_len = end - search_start + 1;
2622 struct extent_map *em;
2626 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2630 if (em->block_start != EXTENT_MAP_HOLE)
2634 if (em->start < search_start)
2635 em_len -= search_start - em->start;
2636 if (em_len > search_len)
2637 em_len = search_len;
2639 ret = set_extent_bit(&inode->io_tree, search_start,
2640 search_start + em_len - 1,
2641 EXTENT_DELALLOC_NEW, cached_state);
2643 search_start = extent_map_end(em);
2644 free_extent_map(em);
2651 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2652 unsigned int extra_bits,
2653 struct extent_state **cached_state)
2655 WARN_ON(PAGE_ALIGNED(end));
2657 if (start >= i_size_read(&inode->vfs_inode) &&
2658 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2660 * There can't be any extents following eof in this case so just
2661 * set the delalloc new bit for the range directly.
2663 extra_bits |= EXTENT_DELALLOC_NEW;
2667 ret = btrfs_find_new_delalloc_bytes(inode, start,
2674 return set_extent_bit(&inode->io_tree, start, end,
2675 EXTENT_DELALLOC | extra_bits, cached_state);
2678 /* see btrfs_writepage_start_hook for details on why this is required */
2679 struct btrfs_writepage_fixup {
2681 struct btrfs_inode *inode;
2682 struct btrfs_work work;
2685 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2687 struct btrfs_writepage_fixup *fixup =
2688 container_of(work, struct btrfs_writepage_fixup, work);
2689 struct btrfs_ordered_extent *ordered;
2690 struct extent_state *cached_state = NULL;
2691 struct extent_changeset *data_reserved = NULL;
2692 struct page *page = fixup->page;
2693 struct btrfs_inode *inode = fixup->inode;
2694 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2695 u64 page_start = page_offset(page);
2696 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2698 bool free_delalloc_space = true;
2701 * This is similar to page_mkwrite, we need to reserve the space before
2702 * we take the page lock.
2704 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2710 * Before we queued this fixup, we took a reference on the page.
2711 * page->mapping may go NULL, but it shouldn't be moved to a different
2714 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2716 * Unfortunately this is a little tricky, either
2718 * 1) We got here and our page had already been dealt with and
2719 * we reserved our space, thus ret == 0, so we need to just
2720 * drop our space reservation and bail. This can happen the
2721 * first time we come into the fixup worker, or could happen
2722 * while waiting for the ordered extent.
2723 * 2) Our page was already dealt with, but we happened to get an
2724 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2725 * this case we obviously don't have anything to release, but
2726 * because the page was already dealt with we don't want to
2727 * mark the page with an error, so make sure we're resetting
2728 * ret to 0. This is why we have this check _before_ the ret
2729 * check, because we do not want to have a surprise ENOSPC
2730 * when the page was already properly dealt with.
2733 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2734 btrfs_delalloc_release_space(inode, data_reserved,
2735 page_start, PAGE_SIZE,
2743 * We can't mess with the page state unless it is locked, so now that
2744 * it is locked bail if we failed to make our space reservation.
2749 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2751 /* already ordered? We're done */
2752 if (PageOrdered(page))
2755 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2757 unlock_extent(&inode->io_tree, page_start, page_end,
2760 btrfs_start_ordered_extent(ordered);
2761 btrfs_put_ordered_extent(ordered);
2765 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2771 * Everything went as planned, we're now the owner of a dirty page with
2772 * delayed allocation bits set and space reserved for our COW
2775 * The page was dirty when we started, nothing should have cleaned it.
2777 BUG_ON(!PageDirty(page));
2778 free_delalloc_space = false;
2780 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2781 if (free_delalloc_space)
2782 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2784 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2788 * We hit ENOSPC or other errors. Update the mapping and page
2789 * to reflect the errors and clean the page.
2791 mapping_set_error(page->mapping, ret);
2792 btrfs_mark_ordered_io_finished(inode, page, page_start,
2794 clear_page_dirty_for_io(page);
2796 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2800 extent_changeset_free(data_reserved);
2802 * As a precaution, do a delayed iput in case it would be the last iput
2803 * that could need flushing space. Recursing back to fixup worker would
2806 btrfs_add_delayed_iput(inode);
2810 * There are a few paths in the higher layers of the kernel that directly
2811 * set the page dirty bit without asking the filesystem if it is a
2812 * good idea. This causes problems because we want to make sure COW
2813 * properly happens and the data=ordered rules are followed.
2815 * In our case any range that doesn't have the ORDERED bit set
2816 * hasn't been properly setup for IO. We kick off an async process
2817 * to fix it up. The async helper will wait for ordered extents, set
2818 * the delalloc bit and make it safe to write the page.
2820 int btrfs_writepage_cow_fixup(struct page *page)
2822 struct inode *inode = page->mapping->host;
2823 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2824 struct btrfs_writepage_fixup *fixup;
2826 /* This page has ordered extent covering it already */
2827 if (PageOrdered(page))
2831 * PageChecked is set below when we create a fixup worker for this page,
2832 * don't try to create another one if we're already PageChecked()
2834 * The extent_io writepage code will redirty the page if we send back
2837 if (PageChecked(page))
2840 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2845 * We are already holding a reference to this inode from
2846 * write_cache_pages. We need to hold it because the space reservation
2847 * takes place outside of the page lock, and we can't trust
2848 * page->mapping outside of the page lock.
2851 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2853 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL);
2855 fixup->inode = BTRFS_I(inode);
2856 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2861 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2862 struct btrfs_inode *inode, u64 file_pos,
2863 struct btrfs_file_extent_item *stack_fi,
2864 const bool update_inode_bytes,
2865 u64 qgroup_reserved)
2867 struct btrfs_root *root = inode->root;
2868 const u64 sectorsize = root->fs_info->sectorsize;
2869 struct btrfs_path *path;
2870 struct extent_buffer *leaf;
2871 struct btrfs_key ins;
2872 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2873 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2874 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2875 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2876 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2877 struct btrfs_drop_extents_args drop_args = { 0 };
2880 path = btrfs_alloc_path();
2885 * we may be replacing one extent in the tree with another.
2886 * The new extent is pinned in the extent map, and we don't want
2887 * to drop it from the cache until it is completely in the btree.
2889 * So, tell btrfs_drop_extents to leave this extent in the cache.
2890 * the caller is expected to unpin it and allow it to be merged
2893 drop_args.path = path;
2894 drop_args.start = file_pos;
2895 drop_args.end = file_pos + num_bytes;
2896 drop_args.replace_extent = true;
2897 drop_args.extent_item_size = sizeof(*stack_fi);
2898 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2902 if (!drop_args.extent_inserted) {
2903 ins.objectid = btrfs_ino(inode);
2904 ins.offset = file_pos;
2905 ins.type = BTRFS_EXTENT_DATA_KEY;
2907 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2912 leaf = path->nodes[0];
2913 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2914 write_extent_buffer(leaf, stack_fi,
2915 btrfs_item_ptr_offset(leaf, path->slots[0]),
2916 sizeof(struct btrfs_file_extent_item));
2918 btrfs_mark_buffer_dirty(trans, leaf);
2919 btrfs_release_path(path);
2922 * If we dropped an inline extent here, we know the range where it is
2923 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2924 * number of bytes only for that range containing the inline extent.
2925 * The remaining of the range will be processed when clearning the
2926 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2928 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2929 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2931 inline_size = drop_args.bytes_found - inline_size;
2932 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2933 drop_args.bytes_found -= inline_size;
2934 num_bytes -= sectorsize;
2937 if (update_inode_bytes)
2938 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2940 ins.objectid = disk_bytenr;
2941 ins.offset = disk_num_bytes;
2942 ins.type = BTRFS_EXTENT_ITEM_KEY;
2944 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2948 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2950 qgroup_reserved, &ins);
2952 btrfs_free_path(path);
2957 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2960 struct btrfs_block_group *cache;
2962 cache = btrfs_lookup_block_group(fs_info, start);
2965 spin_lock(&cache->lock);
2966 cache->delalloc_bytes -= len;
2967 spin_unlock(&cache->lock);
2969 btrfs_put_block_group(cache);
2972 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2973 struct btrfs_ordered_extent *oe)
2975 struct btrfs_file_extent_item stack_fi;
2976 bool update_inode_bytes;
2977 u64 num_bytes = oe->num_bytes;
2978 u64 ram_bytes = oe->ram_bytes;
2980 memset(&stack_fi, 0, sizeof(stack_fi));
2981 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
2982 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
2983 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
2984 oe->disk_num_bytes);
2985 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
2986 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
2987 num_bytes = oe->truncated_len;
2988 ram_bytes = num_bytes;
2990 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
2991 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
2992 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
2993 /* Encryption and other encoding is reserved and all 0 */
2996 * For delalloc, when completing an ordered extent we update the inode's
2997 * bytes when clearing the range in the inode's io tree, so pass false
2998 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
2999 * except if the ordered extent was truncated.
3001 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3002 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3003 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3005 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3006 oe->file_offset, &stack_fi,
3007 update_inode_bytes, oe->qgroup_rsv);
3011 * As ordered data IO finishes, this gets called so we can finish
3012 * an ordered extent if the range of bytes in the file it covers are
3015 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3017 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3018 struct btrfs_root *root = inode->root;
3019 struct btrfs_fs_info *fs_info = root->fs_info;
3020 struct btrfs_trans_handle *trans = NULL;
3021 struct extent_io_tree *io_tree = &inode->io_tree;
3022 struct extent_state *cached_state = NULL;
3024 int compress_type = 0;
3026 u64 logical_len = ordered_extent->num_bytes;
3027 bool freespace_inode;
3028 bool truncated = false;
3029 bool clear_reserved_extent = true;
3030 unsigned int clear_bits = EXTENT_DEFRAG;
3032 start = ordered_extent->file_offset;
3033 end = start + ordered_extent->num_bytes - 1;
3035 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3036 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3037 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3038 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3039 clear_bits |= EXTENT_DELALLOC_NEW;
3041 freespace_inode = btrfs_is_free_space_inode(inode);
3042 if (!freespace_inode)
3043 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3045 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3050 if (btrfs_is_zoned(fs_info))
3051 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3052 ordered_extent->disk_num_bytes);
3054 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3056 logical_len = ordered_extent->truncated_len;
3057 /* Truncated the entire extent, don't bother adding */
3062 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3063 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3065 btrfs_inode_safe_disk_i_size_write(inode, 0);
3066 if (freespace_inode)
3067 trans = btrfs_join_transaction_spacecache(root);
3069 trans = btrfs_join_transaction(root);
3070 if (IS_ERR(trans)) {
3071 ret = PTR_ERR(trans);
3075 trans->block_rsv = &inode->block_rsv;
3076 ret = btrfs_update_inode_fallback(trans, inode);
3077 if (ret) /* -ENOMEM or corruption */
3078 btrfs_abort_transaction(trans, ret);
3082 clear_bits |= EXTENT_LOCKED;
3083 lock_extent(io_tree, start, end, &cached_state);
3085 if (freespace_inode)
3086 trans = btrfs_join_transaction_spacecache(root);
3088 trans = btrfs_join_transaction(root);
3089 if (IS_ERR(trans)) {
3090 ret = PTR_ERR(trans);
3095 trans->block_rsv = &inode->block_rsv;
3097 ret = btrfs_insert_raid_extent(trans, ordered_extent);
3101 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3102 compress_type = ordered_extent->compress_type;
3103 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3104 BUG_ON(compress_type);
3105 ret = btrfs_mark_extent_written(trans, inode,
3106 ordered_extent->file_offset,
3107 ordered_extent->file_offset +
3109 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3110 ordered_extent->disk_num_bytes);
3112 BUG_ON(root == fs_info->tree_root);
3113 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3115 clear_reserved_extent = false;
3116 btrfs_release_delalloc_bytes(fs_info,
3117 ordered_extent->disk_bytenr,
3118 ordered_extent->disk_num_bytes);
3121 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3122 ordered_extent->num_bytes, trans->transid);
3124 btrfs_abort_transaction(trans, ret);
3128 ret = add_pending_csums(trans, &ordered_extent->list);
3130 btrfs_abort_transaction(trans, ret);
3135 * If this is a new delalloc range, clear its new delalloc flag to
3136 * update the inode's number of bytes. This needs to be done first
3137 * before updating the inode item.
3139 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3140 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3141 clear_extent_bit(&inode->io_tree, start, end,
3142 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3145 btrfs_inode_safe_disk_i_size_write(inode, 0);
3146 ret = btrfs_update_inode_fallback(trans, inode);
3147 if (ret) { /* -ENOMEM or corruption */
3148 btrfs_abort_transaction(trans, ret);
3153 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3157 btrfs_end_transaction(trans);
3159 if (ret || truncated) {
3160 u64 unwritten_start = start;
3163 * If we failed to finish this ordered extent for any reason we
3164 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3165 * extent, and mark the inode with the error if it wasn't
3166 * already set. Any error during writeback would have already
3167 * set the mapping error, so we need to set it if we're the ones
3168 * marking this ordered extent as failed.
3170 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3171 &ordered_extent->flags))
3172 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3175 unwritten_start += logical_len;
3176 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3178 /* Drop extent maps for the part of the extent we didn't write. */
3179 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3182 * If the ordered extent had an IOERR or something else went
3183 * wrong we need to return the space for this ordered extent
3184 * back to the allocator. We only free the extent in the
3185 * truncated case if we didn't write out the extent at all.
3187 * If we made it past insert_reserved_file_extent before we
3188 * errored out then we don't need to do this as the accounting
3189 * has already been done.
3191 if ((ret || !logical_len) &&
3192 clear_reserved_extent &&
3193 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3194 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3196 * Discard the range before returning it back to the
3199 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3200 btrfs_discard_extent(fs_info,
3201 ordered_extent->disk_bytenr,
3202 ordered_extent->disk_num_bytes,
3204 btrfs_free_reserved_extent(fs_info,
3205 ordered_extent->disk_bytenr,
3206 ordered_extent->disk_num_bytes, 1);
3208 * Actually free the qgroup rsv which was released when
3209 * the ordered extent was created.
3211 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3212 ordered_extent->qgroup_rsv,
3213 BTRFS_QGROUP_RSV_DATA);
3218 * This needs to be done to make sure anybody waiting knows we are done
3219 * updating everything for this ordered extent.
3221 btrfs_remove_ordered_extent(inode, ordered_extent);
3224 btrfs_put_ordered_extent(ordered_extent);
3225 /* once for the tree */
3226 btrfs_put_ordered_extent(ordered_extent);
3231 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3233 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3234 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags) &&
3235 list_empty(&ordered->bioc_list))
3236 btrfs_finish_ordered_zoned(ordered);
3237 return btrfs_finish_one_ordered(ordered);
3241 * Verify the checksum for a single sector without any extra action that depend
3242 * on the type of I/O.
3244 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3245 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3247 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3250 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3252 shash->tfm = fs_info->csum_shash;
3254 kaddr = kmap_local_page(page) + pgoff;
3255 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3256 kunmap_local(kaddr);
3258 if (memcmp(csum, csum_expected, fs_info->csum_size))
3264 * Verify the checksum of a single data sector.
3266 * @bbio: btrfs_io_bio which contains the csum
3267 * @dev: device the sector is on
3268 * @bio_offset: offset to the beginning of the bio (in bytes)
3269 * @bv: bio_vec to check
3271 * Check if the checksum on a data block is valid. When a checksum mismatch is
3272 * detected, report the error and fill the corrupted range with zero.
3274 * Return %true if the sector is ok or had no checksum to start with, else %false.
3276 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3277 u32 bio_offset, struct bio_vec *bv)
3279 struct btrfs_inode *inode = bbio->inode;
3280 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3281 u64 file_offset = bbio->file_offset + bio_offset;
3282 u64 end = file_offset + bv->bv_len - 1;
3284 u8 csum[BTRFS_CSUM_SIZE];
3286 ASSERT(bv->bv_len == fs_info->sectorsize);
3291 if (btrfs_is_data_reloc_root(inode->root) &&
3292 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3294 /* Skip the range without csum for data reloc inode */
3295 clear_extent_bits(&inode->io_tree, file_offset, end,
3300 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3302 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3308 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3311 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3317 * Perform a delayed iput on @inode.
3319 * @inode: The inode we want to perform iput on
3321 * This function uses the generic vfs_inode::i_count to track whether we should
3322 * just decrement it (in case it's > 1) or if this is the last iput then link
3323 * the inode to the delayed iput machinery. Delayed iputs are processed at
3324 * transaction commit time/superblock commit/cleaner kthread.
3326 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3328 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3329 unsigned long flags;
3331 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3334 atomic_inc(&fs_info->nr_delayed_iputs);
3336 * Need to be irq safe here because we can be called from either an irq
3337 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3340 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3341 ASSERT(list_empty(&inode->delayed_iput));
3342 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3343 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3344 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3345 wake_up_process(fs_info->cleaner_kthread);
3348 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3349 struct btrfs_inode *inode)
3351 list_del_init(&inode->delayed_iput);
3352 spin_unlock_irq(&fs_info->delayed_iput_lock);
3353 iput(&inode->vfs_inode);
3354 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3355 wake_up(&fs_info->delayed_iputs_wait);
3356 spin_lock_irq(&fs_info->delayed_iput_lock);
3359 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3360 struct btrfs_inode *inode)
3362 if (!list_empty(&inode->delayed_iput)) {
3363 spin_lock_irq(&fs_info->delayed_iput_lock);
3364 if (!list_empty(&inode->delayed_iput))
3365 run_delayed_iput_locked(fs_info, inode);
3366 spin_unlock_irq(&fs_info->delayed_iput_lock);
3370 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3373 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3374 * calls btrfs_add_delayed_iput() and that needs to lock
3375 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3376 * prevent a deadlock.
3378 spin_lock_irq(&fs_info->delayed_iput_lock);
3379 while (!list_empty(&fs_info->delayed_iputs)) {
3380 struct btrfs_inode *inode;
3382 inode = list_first_entry(&fs_info->delayed_iputs,
3383 struct btrfs_inode, delayed_iput);
3384 run_delayed_iput_locked(fs_info, inode);
3385 if (need_resched()) {
3386 spin_unlock_irq(&fs_info->delayed_iput_lock);
3388 spin_lock_irq(&fs_info->delayed_iput_lock);
3391 spin_unlock_irq(&fs_info->delayed_iput_lock);
3395 * Wait for flushing all delayed iputs
3397 * @fs_info: the filesystem
3399 * This will wait on any delayed iputs that are currently running with KILLABLE
3400 * set. Once they are all done running we will return, unless we are killed in
3401 * which case we return EINTR. This helps in user operations like fallocate etc
3402 * that might get blocked on the iputs.
3404 * Return EINTR if we were killed, 0 if nothing's pending
3406 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3408 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3409 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3416 * This creates an orphan entry for the given inode in case something goes wrong
3417 * in the middle of an unlink.
3419 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3420 struct btrfs_inode *inode)
3424 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3425 if (ret && ret != -EEXIST) {
3426 btrfs_abort_transaction(trans, ret);
3434 * We have done the delete so we can go ahead and remove the orphan item for
3435 * this particular inode.
3437 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3438 struct btrfs_inode *inode)
3440 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3444 * this cleans up any orphans that may be left on the list from the last use
3447 int btrfs_orphan_cleanup(struct btrfs_root *root)
3449 struct btrfs_fs_info *fs_info = root->fs_info;
3450 struct btrfs_path *path;
3451 struct extent_buffer *leaf;
3452 struct btrfs_key key, found_key;
3453 struct btrfs_trans_handle *trans;
3454 struct inode *inode;
3455 u64 last_objectid = 0;
3456 int ret = 0, nr_unlink = 0;
3458 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3461 path = btrfs_alloc_path();
3466 path->reada = READA_BACK;
3468 key.objectid = BTRFS_ORPHAN_OBJECTID;
3469 key.type = BTRFS_ORPHAN_ITEM_KEY;
3470 key.offset = (u64)-1;
3473 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3478 * if ret == 0 means we found what we were searching for, which
3479 * is weird, but possible, so only screw with path if we didn't
3480 * find the key and see if we have stuff that matches
3484 if (path->slots[0] == 0)
3489 /* pull out the item */
3490 leaf = path->nodes[0];
3491 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3493 /* make sure the item matches what we want */
3494 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3496 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3499 /* release the path since we're done with it */
3500 btrfs_release_path(path);
3503 * this is where we are basically btrfs_lookup, without the
3504 * crossing root thing. we store the inode number in the
3505 * offset of the orphan item.
3508 if (found_key.offset == last_objectid) {
3510 * We found the same inode as before. This means we were
3511 * not able to remove its items via eviction triggered
3512 * by an iput(). A transaction abort may have happened,
3513 * due to -ENOSPC for example, so try to grab the error
3514 * that lead to a transaction abort, if any.
3517 "Error removing orphan entry, stopping orphan cleanup");
3518 ret = BTRFS_FS_ERROR(fs_info) ?: -EINVAL;
3522 last_objectid = found_key.offset;
3524 found_key.objectid = found_key.offset;
3525 found_key.type = BTRFS_INODE_ITEM_KEY;
3526 found_key.offset = 0;
3527 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3528 if (IS_ERR(inode)) {
3529 ret = PTR_ERR(inode);
3535 if (!inode && root == fs_info->tree_root) {
3536 struct btrfs_root *dead_root;
3537 int is_dead_root = 0;
3540 * This is an orphan in the tree root. Currently these
3541 * could come from 2 sources:
3542 * a) a root (snapshot/subvolume) deletion in progress
3543 * b) a free space cache inode
3544 * We need to distinguish those two, as the orphan item
3545 * for a root must not get deleted before the deletion
3546 * of the snapshot/subvolume's tree completes.
3548 * btrfs_find_orphan_roots() ran before us, which has
3549 * found all deleted roots and loaded them into
3550 * fs_info->fs_roots_radix. So here we can find if an
3551 * orphan item corresponds to a deleted root by looking
3552 * up the root from that radix tree.
3555 spin_lock(&fs_info->fs_roots_radix_lock);
3556 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3557 (unsigned long)found_key.objectid);
3558 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3560 spin_unlock(&fs_info->fs_roots_radix_lock);
3563 /* prevent this orphan from being found again */
3564 key.offset = found_key.objectid - 1;
3571 * If we have an inode with links, there are a couple of
3574 * 1. We were halfway through creating fsverity metadata for the
3575 * file. In that case, the orphan item represents incomplete
3576 * fsverity metadata which must be cleaned up with
3577 * btrfs_drop_verity_items and deleting the orphan item.
3579 * 2. Old kernels (before v3.12) used to create an
3580 * orphan item for truncate indicating that there were possibly
3581 * extent items past i_size that needed to be deleted. In v3.12,
3582 * truncate was changed to update i_size in sync with the extent
3583 * items, but the (useless) orphan item was still created. Since
3584 * v4.18, we don't create the orphan item for truncate at all.
3586 * So, this item could mean that we need to do a truncate, but
3587 * only if this filesystem was last used on a pre-v3.12 kernel
3588 * and was not cleanly unmounted. The odds of that are quite
3589 * slim, and it's a pain to do the truncate now, so just delete
3592 * It's also possible that this orphan item was supposed to be
3593 * deleted but wasn't. The inode number may have been reused,
3594 * but either way, we can delete the orphan item.
3596 if (!inode || inode->i_nlink) {
3598 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3604 trans = btrfs_start_transaction(root, 1);
3605 if (IS_ERR(trans)) {
3606 ret = PTR_ERR(trans);
3609 btrfs_debug(fs_info, "auto deleting %Lu",
3610 found_key.objectid);
3611 ret = btrfs_del_orphan_item(trans, root,
3612 found_key.objectid);
3613 btrfs_end_transaction(trans);
3621 /* this will do delete_inode and everything for us */
3624 /* release the path since we're done with it */
3625 btrfs_release_path(path);
3627 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3628 trans = btrfs_join_transaction(root);
3630 btrfs_end_transaction(trans);
3634 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3638 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3639 btrfs_free_path(path);
3644 * very simple check to peek ahead in the leaf looking for xattrs. If we
3645 * don't find any xattrs, we know there can't be any acls.
3647 * slot is the slot the inode is in, objectid is the objectid of the inode
3649 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3650 int slot, u64 objectid,
3651 int *first_xattr_slot)
3653 u32 nritems = btrfs_header_nritems(leaf);
3654 struct btrfs_key found_key;
3655 static u64 xattr_access = 0;
3656 static u64 xattr_default = 0;
3659 if (!xattr_access) {
3660 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3661 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3662 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3663 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3667 *first_xattr_slot = -1;
3668 while (slot < nritems) {
3669 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3671 /* we found a different objectid, there must not be acls */
3672 if (found_key.objectid != objectid)
3675 /* we found an xattr, assume we've got an acl */
3676 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3677 if (*first_xattr_slot == -1)
3678 *first_xattr_slot = slot;
3679 if (found_key.offset == xattr_access ||
3680 found_key.offset == xattr_default)
3685 * we found a key greater than an xattr key, there can't
3686 * be any acls later on
3688 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3695 * it goes inode, inode backrefs, xattrs, extents,
3696 * so if there are a ton of hard links to an inode there can
3697 * be a lot of backrefs. Don't waste time searching too hard,
3698 * this is just an optimization
3703 /* we hit the end of the leaf before we found an xattr or
3704 * something larger than an xattr. We have to assume the inode
3707 if (*first_xattr_slot == -1)
3708 *first_xattr_slot = slot;
3713 * read an inode from the btree into the in-memory inode
3715 static int btrfs_read_locked_inode(struct inode *inode,
3716 struct btrfs_path *in_path)
3718 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3719 struct btrfs_path *path = in_path;
3720 struct extent_buffer *leaf;
3721 struct btrfs_inode_item *inode_item;
3722 struct btrfs_root *root = BTRFS_I(inode)->root;
3723 struct btrfs_key location;
3728 bool filled = false;
3729 int first_xattr_slot;
3731 ret = btrfs_fill_inode(inode, &rdev);
3736 path = btrfs_alloc_path();
3741 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3743 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3745 if (path != in_path)
3746 btrfs_free_path(path);
3750 leaf = path->nodes[0];
3755 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3756 struct btrfs_inode_item);
3757 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3758 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3759 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3760 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3761 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3762 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3763 round_up(i_size_read(inode), fs_info->sectorsize));
3765 inode_set_atime(inode, btrfs_timespec_sec(leaf, &inode_item->atime),
3766 btrfs_timespec_nsec(leaf, &inode_item->atime));
3768 inode_set_mtime(inode, btrfs_timespec_sec(leaf, &inode_item->mtime),
3769 btrfs_timespec_nsec(leaf, &inode_item->mtime));
3771 inode_set_ctime(inode, btrfs_timespec_sec(leaf, &inode_item->ctime),
3772 btrfs_timespec_nsec(leaf, &inode_item->ctime));
3774 BTRFS_I(inode)->i_otime_sec = btrfs_timespec_sec(leaf, &inode_item->otime);
3775 BTRFS_I(inode)->i_otime_nsec = btrfs_timespec_nsec(leaf, &inode_item->otime);
3777 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3778 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3779 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3781 inode_set_iversion_queried(inode,
3782 btrfs_inode_sequence(leaf, inode_item));
3783 inode->i_generation = BTRFS_I(inode)->generation;
3785 rdev = btrfs_inode_rdev(leaf, inode_item);
3787 BTRFS_I(inode)->index_cnt = (u64)-1;
3788 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3789 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3793 * If we were modified in the current generation and evicted from memory
3794 * and then re-read we need to do a full sync since we don't have any
3795 * idea about which extents were modified before we were evicted from
3798 * This is required for both inode re-read from disk and delayed inode
3799 * in delayed_nodes_tree.
3801 if (BTRFS_I(inode)->last_trans == btrfs_get_fs_generation(fs_info))
3802 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3803 &BTRFS_I(inode)->runtime_flags);
3806 * We don't persist the id of the transaction where an unlink operation
3807 * against the inode was last made. So here we assume the inode might
3808 * have been evicted, and therefore the exact value of last_unlink_trans
3809 * lost, and set it to last_trans to avoid metadata inconsistencies
3810 * between the inode and its parent if the inode is fsync'ed and the log
3811 * replayed. For example, in the scenario:
3814 * ln mydir/foo mydir/bar
3817 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3818 * xfs_io -c fsync mydir/foo
3820 * mount fs, triggers fsync log replay
3822 * We must make sure that when we fsync our inode foo we also log its
3823 * parent inode, otherwise after log replay the parent still has the
3824 * dentry with the "bar" name but our inode foo has a link count of 1
3825 * and doesn't have an inode ref with the name "bar" anymore.
3827 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3828 * but it guarantees correctness at the expense of occasional full
3829 * transaction commits on fsync if our inode is a directory, or if our
3830 * inode is not a directory, logging its parent unnecessarily.
3832 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3835 * Same logic as for last_unlink_trans. We don't persist the generation
3836 * of the last transaction where this inode was used for a reflink
3837 * operation, so after eviction and reloading the inode we must be
3838 * pessimistic and assume the last transaction that modified the inode.
3840 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3843 if (inode->i_nlink != 1 ||
3844 path->slots[0] >= btrfs_header_nritems(leaf))
3847 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3848 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3851 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3852 if (location.type == BTRFS_INODE_REF_KEY) {
3853 struct btrfs_inode_ref *ref;
3855 ref = (struct btrfs_inode_ref *)ptr;
3856 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3857 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3858 struct btrfs_inode_extref *extref;
3860 extref = (struct btrfs_inode_extref *)ptr;
3861 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3866 * try to precache a NULL acl entry for files that don't have
3867 * any xattrs or acls
3869 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3870 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3871 if (first_xattr_slot != -1) {
3872 path->slots[0] = first_xattr_slot;
3873 ret = btrfs_load_inode_props(inode, path);
3876 "error loading props for ino %llu (root %llu): %d",
3877 btrfs_ino(BTRFS_I(inode)),
3878 root->root_key.objectid, ret);
3880 if (path != in_path)
3881 btrfs_free_path(path);
3884 cache_no_acl(inode);
3886 switch (inode->i_mode & S_IFMT) {
3888 inode->i_mapping->a_ops = &btrfs_aops;
3889 inode->i_fop = &btrfs_file_operations;
3890 inode->i_op = &btrfs_file_inode_operations;
3893 inode->i_fop = &btrfs_dir_file_operations;
3894 inode->i_op = &btrfs_dir_inode_operations;
3897 inode->i_op = &btrfs_symlink_inode_operations;
3898 inode_nohighmem(inode);
3899 inode->i_mapping->a_ops = &btrfs_aops;
3902 inode->i_op = &btrfs_special_inode_operations;
3903 init_special_inode(inode, inode->i_mode, rdev);
3907 btrfs_sync_inode_flags_to_i_flags(inode);
3912 * given a leaf and an inode, copy the inode fields into the leaf
3914 static void fill_inode_item(struct btrfs_trans_handle *trans,
3915 struct extent_buffer *leaf,
3916 struct btrfs_inode_item *item,
3917 struct inode *inode)
3919 struct btrfs_map_token token;
3922 btrfs_init_map_token(&token, leaf);
3924 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3925 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3926 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3927 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3928 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3930 btrfs_set_token_timespec_sec(&token, &item->atime,
3931 inode_get_atime_sec(inode));
3932 btrfs_set_token_timespec_nsec(&token, &item->atime,
3933 inode_get_atime_nsec(inode));
3935 btrfs_set_token_timespec_sec(&token, &item->mtime,
3936 inode_get_mtime_sec(inode));
3937 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3938 inode_get_mtime_nsec(inode));
3940 btrfs_set_token_timespec_sec(&token, &item->ctime,
3941 inode_get_ctime_sec(inode));
3942 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3943 inode_get_ctime_nsec(inode));
3945 btrfs_set_token_timespec_sec(&token, &item->otime, BTRFS_I(inode)->i_otime_sec);
3946 btrfs_set_token_timespec_nsec(&token, &item->otime, BTRFS_I(inode)->i_otime_nsec);
3948 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3949 btrfs_set_token_inode_generation(&token, item,
3950 BTRFS_I(inode)->generation);
3951 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3952 btrfs_set_token_inode_transid(&token, item, trans->transid);
3953 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3954 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3955 BTRFS_I(inode)->ro_flags);
3956 btrfs_set_token_inode_flags(&token, item, flags);
3957 btrfs_set_token_inode_block_group(&token, item, 0);
3961 * copy everything in the in-memory inode into the btree.
3963 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3964 struct btrfs_inode *inode)
3966 struct btrfs_inode_item *inode_item;
3967 struct btrfs_path *path;
3968 struct extent_buffer *leaf;
3971 path = btrfs_alloc_path();
3975 ret = btrfs_lookup_inode(trans, inode->root, path, &inode->location, 1);
3982 leaf = path->nodes[0];
3983 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3984 struct btrfs_inode_item);
3986 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3987 btrfs_mark_buffer_dirty(trans, leaf);
3988 btrfs_set_inode_last_trans(trans, inode);
3991 btrfs_free_path(path);
3996 * copy everything in the in-memory inode into the btree.
3998 int btrfs_update_inode(struct btrfs_trans_handle *trans,
3999 struct btrfs_inode *inode)
4001 struct btrfs_root *root = inode->root;
4002 struct btrfs_fs_info *fs_info = root->fs_info;
4006 * If the inode is a free space inode, we can deadlock during commit
4007 * if we put it into the delayed code.
4009 * The data relocation inode should also be directly updated
4012 if (!btrfs_is_free_space_inode(inode)
4013 && !btrfs_is_data_reloc_root(root)
4014 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4015 btrfs_update_root_times(trans, root);
4017 ret = btrfs_delayed_update_inode(trans, inode);
4019 btrfs_set_inode_last_trans(trans, inode);
4023 return btrfs_update_inode_item(trans, inode);
4026 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4027 struct btrfs_inode *inode)
4031 ret = btrfs_update_inode(trans, inode);
4033 return btrfs_update_inode_item(trans, inode);
4038 * unlink helper that gets used here in inode.c and in the tree logging
4039 * recovery code. It remove a link in a directory with a given name, and
4040 * also drops the back refs in the inode to the directory
4042 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4043 struct btrfs_inode *dir,
4044 struct btrfs_inode *inode,
4045 const struct fscrypt_str *name,
4046 struct btrfs_rename_ctx *rename_ctx)
4048 struct btrfs_root *root = dir->root;
4049 struct btrfs_fs_info *fs_info = root->fs_info;
4050 struct btrfs_path *path;
4052 struct btrfs_dir_item *di;
4054 u64 ino = btrfs_ino(inode);
4055 u64 dir_ino = btrfs_ino(dir);
4057 path = btrfs_alloc_path();
4063 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4064 if (IS_ERR_OR_NULL(di)) {
4065 ret = di ? PTR_ERR(di) : -ENOENT;
4068 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4071 btrfs_release_path(path);
4074 * If we don't have dir index, we have to get it by looking up
4075 * the inode ref, since we get the inode ref, remove it directly,
4076 * it is unnecessary to do delayed deletion.
4078 * But if we have dir index, needn't search inode ref to get it.
4079 * Since the inode ref is close to the inode item, it is better
4080 * that we delay to delete it, and just do this deletion when
4081 * we update the inode item.
4083 if (inode->dir_index) {
4084 ret = btrfs_delayed_delete_inode_ref(inode);
4086 index = inode->dir_index;
4091 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4094 "failed to delete reference to %.*s, inode %llu parent %llu",
4095 name->len, name->name, ino, dir_ino);
4096 btrfs_abort_transaction(trans, ret);
4101 rename_ctx->index = index;
4103 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4105 btrfs_abort_transaction(trans, ret);
4110 * If we are in a rename context, we don't need to update anything in the
4111 * log. That will be done later during the rename by btrfs_log_new_name().
4112 * Besides that, doing it here would only cause extra unnecessary btree
4113 * operations on the log tree, increasing latency for applications.
4116 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4117 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4121 * If we have a pending delayed iput we could end up with the final iput
4122 * being run in btrfs-cleaner context. If we have enough of these built
4123 * up we can end up burning a lot of time in btrfs-cleaner without any
4124 * way to throttle the unlinks. Since we're currently holding a ref on
4125 * the inode we can run the delayed iput here without any issues as the
4126 * final iput won't be done until after we drop the ref we're currently
4129 btrfs_run_delayed_iput(fs_info, inode);
4131 btrfs_free_path(path);
4135 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4136 inode_inc_iversion(&inode->vfs_inode);
4137 inode_inc_iversion(&dir->vfs_inode);
4138 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4139 ret = btrfs_update_inode(trans, dir);
4144 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4145 struct btrfs_inode *dir, struct btrfs_inode *inode,
4146 const struct fscrypt_str *name)
4150 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4152 drop_nlink(&inode->vfs_inode);
4153 ret = btrfs_update_inode(trans, inode);
4159 * helper to start transaction for unlink and rmdir.
4161 * unlink and rmdir are special in btrfs, they do not always free space, so
4162 * if we cannot make our reservations the normal way try and see if there is
4163 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4164 * allow the unlink to occur.
4166 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4168 struct btrfs_root *root = dir->root;
4170 return btrfs_start_transaction_fallback_global_rsv(root,
4171 BTRFS_UNLINK_METADATA_UNITS);
4174 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4176 struct btrfs_trans_handle *trans;
4177 struct inode *inode = d_inode(dentry);
4179 struct fscrypt_name fname;
4181 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4185 /* This needs to handle no-key deletions later on */
4187 trans = __unlink_start_trans(BTRFS_I(dir));
4188 if (IS_ERR(trans)) {
4189 ret = PTR_ERR(trans);
4193 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4196 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4201 if (inode->i_nlink == 0) {
4202 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4208 btrfs_end_transaction(trans);
4209 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4211 fscrypt_free_filename(&fname);
4215 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4216 struct btrfs_inode *dir, struct dentry *dentry)
4218 struct btrfs_root *root = dir->root;
4219 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4220 struct btrfs_path *path;
4221 struct extent_buffer *leaf;
4222 struct btrfs_dir_item *di;
4223 struct btrfs_key key;
4227 u64 dir_ino = btrfs_ino(dir);
4228 struct fscrypt_name fname;
4230 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4234 /* This needs to handle no-key deletions later on */
4236 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4237 objectid = inode->root->root_key.objectid;
4238 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4239 objectid = inode->location.objectid;
4242 fscrypt_free_filename(&fname);
4246 path = btrfs_alloc_path();
4252 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4253 &fname.disk_name, -1);
4254 if (IS_ERR_OR_NULL(di)) {
4255 ret = di ? PTR_ERR(di) : -ENOENT;
4259 leaf = path->nodes[0];
4260 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4261 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4262 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4264 btrfs_abort_transaction(trans, ret);
4267 btrfs_release_path(path);
4270 * This is a placeholder inode for a subvolume we didn't have a
4271 * reference to at the time of the snapshot creation. In the meantime
4272 * we could have renamed the real subvol link into our snapshot, so
4273 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4274 * Instead simply lookup the dir_index_item for this entry so we can
4275 * remove it. Otherwise we know we have a ref to the root and we can
4276 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4278 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4279 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4280 if (IS_ERR_OR_NULL(di)) {
4285 btrfs_abort_transaction(trans, ret);
4289 leaf = path->nodes[0];
4290 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4292 btrfs_release_path(path);
4294 ret = btrfs_del_root_ref(trans, objectid,
4295 root->root_key.objectid, dir_ino,
4296 &index, &fname.disk_name);
4298 btrfs_abort_transaction(trans, ret);
4303 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4305 btrfs_abort_transaction(trans, ret);
4309 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4310 inode_inc_iversion(&dir->vfs_inode);
4311 inode_set_mtime_to_ts(&dir->vfs_inode, inode_set_ctime_current(&dir->vfs_inode));
4312 ret = btrfs_update_inode_fallback(trans, dir);
4314 btrfs_abort_transaction(trans, ret);
4316 btrfs_free_path(path);
4317 fscrypt_free_filename(&fname);
4322 * Helper to check if the subvolume references other subvolumes or if it's
4325 static noinline int may_destroy_subvol(struct btrfs_root *root)
4327 struct btrfs_fs_info *fs_info = root->fs_info;
4328 struct btrfs_path *path;
4329 struct btrfs_dir_item *di;
4330 struct btrfs_key key;
4331 struct fscrypt_str name = FSTR_INIT("default", 7);
4335 path = btrfs_alloc_path();
4339 /* Make sure this root isn't set as the default subvol */
4340 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4341 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4343 if (di && !IS_ERR(di)) {
4344 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4345 if (key.objectid == root->root_key.objectid) {
4348 "deleting default subvolume %llu is not allowed",
4352 btrfs_release_path(path);
4355 key.objectid = root->root_key.objectid;
4356 key.type = BTRFS_ROOT_REF_KEY;
4357 key.offset = (u64)-1;
4359 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4365 if (path->slots[0] > 0) {
4367 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4368 if (key.objectid == root->root_key.objectid &&
4369 key.type == BTRFS_ROOT_REF_KEY)
4373 btrfs_free_path(path);
4377 /* Delete all dentries for inodes belonging to the root */
4378 static void btrfs_prune_dentries(struct btrfs_root *root)
4380 struct btrfs_fs_info *fs_info = root->fs_info;
4381 struct rb_node *node;
4382 struct rb_node *prev;
4383 struct btrfs_inode *entry;
4384 struct inode *inode;
4387 if (!BTRFS_FS_ERROR(fs_info))
4388 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4390 spin_lock(&root->inode_lock);
4392 node = root->inode_tree.rb_node;
4396 entry = rb_entry(node, struct btrfs_inode, rb_node);
4398 if (objectid < btrfs_ino(entry))
4399 node = node->rb_left;
4400 else if (objectid > btrfs_ino(entry))
4401 node = node->rb_right;
4407 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4408 if (objectid <= btrfs_ino(entry)) {
4412 prev = rb_next(prev);
4416 entry = rb_entry(node, struct btrfs_inode, rb_node);
4417 objectid = btrfs_ino(entry) + 1;
4418 inode = igrab(&entry->vfs_inode);
4420 spin_unlock(&root->inode_lock);
4421 if (atomic_read(&inode->i_count) > 1)
4422 d_prune_aliases(inode);
4424 * btrfs_drop_inode will have it removed from the inode
4425 * cache when its usage count hits zero.
4429 spin_lock(&root->inode_lock);
4433 if (cond_resched_lock(&root->inode_lock))
4436 node = rb_next(node);
4438 spin_unlock(&root->inode_lock);
4441 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4443 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4444 struct btrfs_root *root = dir->root;
4445 struct inode *inode = d_inode(dentry);
4446 struct btrfs_root *dest = BTRFS_I(inode)->root;
4447 struct btrfs_trans_handle *trans;
4448 struct btrfs_block_rsv block_rsv;
4453 * Don't allow to delete a subvolume with send in progress. This is
4454 * inside the inode lock so the error handling that has to drop the bit
4455 * again is not run concurrently.
4457 spin_lock(&dest->root_item_lock);
4458 if (dest->send_in_progress) {
4459 spin_unlock(&dest->root_item_lock);
4461 "attempt to delete subvolume %llu during send",
4462 dest->root_key.objectid);
4465 if (atomic_read(&dest->nr_swapfiles)) {
4466 spin_unlock(&dest->root_item_lock);
4468 "attempt to delete subvolume %llu with active swapfile",
4469 root->root_key.objectid);
4472 root_flags = btrfs_root_flags(&dest->root_item);
4473 btrfs_set_root_flags(&dest->root_item,
4474 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4475 spin_unlock(&dest->root_item_lock);
4477 down_write(&fs_info->subvol_sem);
4479 ret = may_destroy_subvol(dest);
4483 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4485 * One for dir inode,
4486 * two for dir entries,
4487 * two for root ref/backref.
4489 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4493 trans = btrfs_start_transaction(root, 0);
4494 if (IS_ERR(trans)) {
4495 ret = PTR_ERR(trans);
4498 trans->block_rsv = &block_rsv;
4499 trans->bytes_reserved = block_rsv.size;
4501 btrfs_record_snapshot_destroy(trans, dir);
4503 ret = btrfs_unlink_subvol(trans, dir, dentry);
4505 btrfs_abort_transaction(trans, ret);
4509 ret = btrfs_record_root_in_trans(trans, dest);
4511 btrfs_abort_transaction(trans, ret);
4515 memset(&dest->root_item.drop_progress, 0,
4516 sizeof(dest->root_item.drop_progress));
4517 btrfs_set_root_drop_level(&dest->root_item, 0);
4518 btrfs_set_root_refs(&dest->root_item, 0);
4520 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4521 ret = btrfs_insert_orphan_item(trans,
4523 dest->root_key.objectid);
4525 btrfs_abort_transaction(trans, ret);
4530 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4531 BTRFS_UUID_KEY_SUBVOL,
4532 dest->root_key.objectid);
4533 if (ret && ret != -ENOENT) {
4534 btrfs_abort_transaction(trans, ret);
4537 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4538 ret = btrfs_uuid_tree_remove(trans,
4539 dest->root_item.received_uuid,
4540 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4541 dest->root_key.objectid);
4542 if (ret && ret != -ENOENT) {
4543 btrfs_abort_transaction(trans, ret);
4548 free_anon_bdev(dest->anon_dev);
4551 trans->block_rsv = NULL;
4552 trans->bytes_reserved = 0;
4553 ret = btrfs_end_transaction(trans);
4554 inode->i_flags |= S_DEAD;
4556 btrfs_subvolume_release_metadata(root, &block_rsv);
4558 up_write(&fs_info->subvol_sem);
4560 spin_lock(&dest->root_item_lock);
4561 root_flags = btrfs_root_flags(&dest->root_item);
4562 btrfs_set_root_flags(&dest->root_item,
4563 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4564 spin_unlock(&dest->root_item_lock);
4566 d_invalidate(dentry);
4567 btrfs_prune_dentries(dest);
4568 ASSERT(dest->send_in_progress == 0);
4574 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4576 struct inode *inode = d_inode(dentry);
4577 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4579 struct btrfs_trans_handle *trans;
4580 u64 last_unlink_trans;
4581 struct fscrypt_name fname;
4583 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4585 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4586 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4588 "extent tree v2 doesn't support snapshot deletion yet");
4591 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4594 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4598 /* This needs to handle no-key deletions later on */
4600 trans = __unlink_start_trans(BTRFS_I(dir));
4601 if (IS_ERR(trans)) {
4602 err = PTR_ERR(trans);
4606 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4607 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4611 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4615 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4617 /* now the directory is empty */
4618 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4621 btrfs_i_size_write(BTRFS_I(inode), 0);
4623 * Propagate the last_unlink_trans value of the deleted dir to
4624 * its parent directory. This is to prevent an unrecoverable
4625 * log tree in the case we do something like this:
4627 * 2) create snapshot under dir foo
4628 * 3) delete the snapshot
4631 * 6) fsync foo or some file inside foo
4633 if (last_unlink_trans >= trans->transid)
4634 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4637 btrfs_end_transaction(trans);
4639 btrfs_btree_balance_dirty(fs_info);
4640 fscrypt_free_filename(&fname);
4646 * Read, zero a chunk and write a block.
4648 * @inode - inode that we're zeroing
4649 * @from - the offset to start zeroing
4650 * @len - the length to zero, 0 to zero the entire range respective to the
4652 * @front - zero up to the offset instead of from the offset on
4654 * This will find the block for the "from" offset and cow the block and zero the
4655 * part we want to zero. This is used with truncate and hole punching.
4657 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4660 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4661 struct address_space *mapping = inode->vfs_inode.i_mapping;
4662 struct extent_io_tree *io_tree = &inode->io_tree;
4663 struct btrfs_ordered_extent *ordered;
4664 struct extent_state *cached_state = NULL;
4665 struct extent_changeset *data_reserved = NULL;
4666 bool only_release_metadata = false;
4667 u32 blocksize = fs_info->sectorsize;
4668 pgoff_t index = from >> PAGE_SHIFT;
4669 unsigned offset = from & (blocksize - 1);
4671 gfp_t mask = btrfs_alloc_write_mask(mapping);
4672 size_t write_bytes = blocksize;
4677 if (IS_ALIGNED(offset, blocksize) &&
4678 (!len || IS_ALIGNED(len, blocksize)))
4681 block_start = round_down(from, blocksize);
4682 block_end = block_start + blocksize - 1;
4684 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4687 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4688 /* For nocow case, no need to reserve data space */
4689 only_release_metadata = true;
4694 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4696 if (!only_release_metadata)
4697 btrfs_free_reserved_data_space(inode, data_reserved,
4698 block_start, blocksize);
4702 page = find_or_create_page(mapping, index, mask);
4704 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4706 btrfs_delalloc_release_extents(inode, blocksize);
4711 if (!PageUptodate(page)) {
4712 ret = btrfs_read_folio(NULL, page_folio(page));
4714 if (page->mapping != mapping) {
4719 if (!PageUptodate(page)) {
4726 * We unlock the page after the io is completed and then re-lock it
4727 * above. release_folio() could have come in between that and cleared
4728 * PagePrivate(), but left the page in the mapping. Set the page mapped
4729 * here to make sure it's properly set for the subpage stuff.
4731 ret = set_page_extent_mapped(page);
4735 wait_on_page_writeback(page);
4737 lock_extent(io_tree, block_start, block_end, &cached_state);
4739 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4741 unlock_extent(io_tree, block_start, block_end, &cached_state);
4744 btrfs_start_ordered_extent(ordered);
4745 btrfs_put_ordered_extent(ordered);
4749 clear_extent_bit(&inode->io_tree, block_start, block_end,
4750 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4753 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4756 unlock_extent(io_tree, block_start, block_end, &cached_state);
4760 if (offset != blocksize) {
4762 len = blocksize - offset;
4764 memzero_page(page, (block_start - page_offset(page)),
4767 memzero_page(page, (block_start - page_offset(page)) + offset,
4770 btrfs_page_clear_checked(fs_info, page, block_start,
4771 block_end + 1 - block_start);
4772 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4773 unlock_extent(io_tree, block_start, block_end, &cached_state);
4775 if (only_release_metadata)
4776 set_extent_bit(&inode->io_tree, block_start, block_end,
4777 EXTENT_NORESERVE, NULL);
4781 if (only_release_metadata)
4782 btrfs_delalloc_release_metadata(inode, blocksize, true);
4784 btrfs_delalloc_release_space(inode, data_reserved,
4785 block_start, blocksize, true);
4787 btrfs_delalloc_release_extents(inode, blocksize);
4791 if (only_release_metadata)
4792 btrfs_check_nocow_unlock(inode);
4793 extent_changeset_free(data_reserved);
4797 static int maybe_insert_hole(struct btrfs_inode *inode, u64 offset, u64 len)
4799 struct btrfs_root *root = inode->root;
4800 struct btrfs_fs_info *fs_info = root->fs_info;
4801 struct btrfs_trans_handle *trans;
4802 struct btrfs_drop_extents_args drop_args = { 0 };
4806 * If NO_HOLES is enabled, we don't need to do anything.
4807 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4808 * or btrfs_update_inode() will be called, which guarantee that the next
4809 * fsync will know this inode was changed and needs to be logged.
4811 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4815 * 1 - for the one we're dropping
4816 * 1 - for the one we're adding
4817 * 1 - for updating the inode.
4819 trans = btrfs_start_transaction(root, 3);
4821 return PTR_ERR(trans);
4823 drop_args.start = offset;
4824 drop_args.end = offset + len;
4825 drop_args.drop_cache = true;
4827 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4829 btrfs_abort_transaction(trans, ret);
4830 btrfs_end_transaction(trans);
4834 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4836 btrfs_abort_transaction(trans, ret);
4838 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4839 btrfs_update_inode(trans, inode);
4841 btrfs_end_transaction(trans);
4846 * This function puts in dummy file extents for the area we're creating a hole
4847 * for. So if we are truncating this file to a larger size we need to insert
4848 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4849 * the range between oldsize and size
4851 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4853 struct btrfs_root *root = inode->root;
4854 struct btrfs_fs_info *fs_info = root->fs_info;
4855 struct extent_io_tree *io_tree = &inode->io_tree;
4856 struct extent_map *em = NULL;
4857 struct extent_state *cached_state = NULL;
4858 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4859 u64 block_end = ALIGN(size, fs_info->sectorsize);
4866 * If our size started in the middle of a block we need to zero out the
4867 * rest of the block before we expand the i_size, otherwise we could
4868 * expose stale data.
4870 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4874 if (size <= hole_start)
4877 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4879 cur_offset = hole_start;
4881 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4882 block_end - cur_offset);
4888 last_byte = min(extent_map_end(em), block_end);
4889 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4890 hole_size = last_byte - cur_offset;
4892 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4893 struct extent_map *hole_em;
4895 err = maybe_insert_hole(inode, cur_offset, hole_size);
4899 err = btrfs_inode_set_file_extent_range(inode,
4900 cur_offset, hole_size);
4904 hole_em = alloc_extent_map();
4906 btrfs_drop_extent_map_range(inode, cur_offset,
4907 cur_offset + hole_size - 1,
4909 btrfs_set_inode_full_sync(inode);
4912 hole_em->start = cur_offset;
4913 hole_em->len = hole_size;
4914 hole_em->orig_start = cur_offset;
4916 hole_em->block_start = EXTENT_MAP_HOLE;
4917 hole_em->block_len = 0;
4918 hole_em->orig_block_len = 0;
4919 hole_em->ram_bytes = hole_size;
4920 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4921 hole_em->generation = btrfs_get_fs_generation(fs_info);
4923 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4924 free_extent_map(hole_em);
4926 err = btrfs_inode_set_file_extent_range(inode,
4927 cur_offset, hole_size);
4932 free_extent_map(em);
4934 cur_offset = last_byte;
4935 if (cur_offset >= block_end)
4938 free_extent_map(em);
4939 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4943 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4945 struct btrfs_root *root = BTRFS_I(inode)->root;
4946 struct btrfs_trans_handle *trans;
4947 loff_t oldsize = i_size_read(inode);
4948 loff_t newsize = attr->ia_size;
4949 int mask = attr->ia_valid;
4953 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4954 * special case where we need to update the times despite not having
4955 * these flags set. For all other operations the VFS set these flags
4956 * explicitly if it wants a timestamp update.
4958 if (newsize != oldsize) {
4959 inode_inc_iversion(inode);
4960 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
4961 inode_set_mtime_to_ts(inode,
4962 inode_set_ctime_current(inode));
4966 if (newsize > oldsize) {
4968 * Don't do an expanding truncate while snapshotting is ongoing.
4969 * This is to ensure the snapshot captures a fully consistent
4970 * state of this file - if the snapshot captures this expanding
4971 * truncation, it must capture all writes that happened before
4974 btrfs_drew_write_lock(&root->snapshot_lock);
4975 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4977 btrfs_drew_write_unlock(&root->snapshot_lock);
4981 trans = btrfs_start_transaction(root, 1);
4982 if (IS_ERR(trans)) {
4983 btrfs_drew_write_unlock(&root->snapshot_lock);
4984 return PTR_ERR(trans);
4987 i_size_write(inode, newsize);
4988 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
4989 pagecache_isize_extended(inode, oldsize, newsize);
4990 ret = btrfs_update_inode(trans, BTRFS_I(inode));
4991 btrfs_drew_write_unlock(&root->snapshot_lock);
4992 btrfs_end_transaction(trans);
4994 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4996 if (btrfs_is_zoned(fs_info)) {
4997 ret = btrfs_wait_ordered_range(inode,
4998 ALIGN(newsize, fs_info->sectorsize),
5005 * We're truncating a file that used to have good data down to
5006 * zero. Make sure any new writes to the file get on disk
5010 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5011 &BTRFS_I(inode)->runtime_flags);
5013 truncate_setsize(inode, newsize);
5015 inode_dio_wait(inode);
5017 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5018 if (ret && inode->i_nlink) {
5022 * Truncate failed, so fix up the in-memory size. We
5023 * adjusted disk_i_size down as we removed extents, so
5024 * wait for disk_i_size to be stable and then update the
5025 * in-memory size to match.
5027 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5030 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5037 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5040 struct inode *inode = d_inode(dentry);
5041 struct btrfs_root *root = BTRFS_I(inode)->root;
5044 if (btrfs_root_readonly(root))
5047 err = setattr_prepare(idmap, dentry, attr);
5051 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5052 err = btrfs_setsize(inode, attr);
5057 if (attr->ia_valid) {
5058 setattr_copy(idmap, inode, attr);
5059 inode_inc_iversion(inode);
5060 err = btrfs_dirty_inode(BTRFS_I(inode));
5062 if (!err && attr->ia_valid & ATTR_MODE)
5063 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5070 * While truncating the inode pages during eviction, we get the VFS
5071 * calling btrfs_invalidate_folio() against each folio of the inode. This
5072 * is slow because the calls to btrfs_invalidate_folio() result in a
5073 * huge amount of calls to lock_extent() and clear_extent_bit(),
5074 * which keep merging and splitting extent_state structures over and over,
5075 * wasting lots of time.
5077 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5078 * skip all those expensive operations on a per folio basis and do only
5079 * the ordered io finishing, while we release here the extent_map and
5080 * extent_state structures, without the excessive merging and splitting.
5082 static void evict_inode_truncate_pages(struct inode *inode)
5084 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5085 struct rb_node *node;
5087 ASSERT(inode->i_state & I_FREEING);
5088 truncate_inode_pages_final(&inode->i_data);
5090 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5093 * Keep looping until we have no more ranges in the io tree.
5094 * We can have ongoing bios started by readahead that have
5095 * their endio callback (extent_io.c:end_bio_extent_readpage)
5096 * still in progress (unlocked the pages in the bio but did not yet
5097 * unlocked the ranges in the io tree). Therefore this means some
5098 * ranges can still be locked and eviction started because before
5099 * submitting those bios, which are executed by a separate task (work
5100 * queue kthread), inode references (inode->i_count) were not taken
5101 * (which would be dropped in the end io callback of each bio).
5102 * Therefore here we effectively end up waiting for those bios and
5103 * anyone else holding locked ranges without having bumped the inode's
5104 * reference count - if we don't do it, when they access the inode's
5105 * io_tree to unlock a range it may be too late, leading to an
5106 * use-after-free issue.
5108 spin_lock(&io_tree->lock);
5109 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5110 struct extent_state *state;
5111 struct extent_state *cached_state = NULL;
5114 unsigned state_flags;
5116 node = rb_first(&io_tree->state);
5117 state = rb_entry(node, struct extent_state, rb_node);
5118 start = state->start;
5120 state_flags = state->state;
5121 spin_unlock(&io_tree->lock);
5123 lock_extent(io_tree, start, end, &cached_state);
5126 * If still has DELALLOC flag, the extent didn't reach disk,
5127 * and its reserved space won't be freed by delayed_ref.
5128 * So we need to free its reserved space here.
5129 * (Refer to comment in btrfs_invalidate_folio, case 2)
5131 * Note, end is the bytenr of last byte, so we need + 1 here.
5133 if (state_flags & EXTENT_DELALLOC)
5134 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5135 end - start + 1, NULL);
5137 clear_extent_bit(io_tree, start, end,
5138 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5142 spin_lock(&io_tree->lock);
5144 spin_unlock(&io_tree->lock);
5147 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5148 struct btrfs_block_rsv *rsv)
5150 struct btrfs_fs_info *fs_info = root->fs_info;
5151 struct btrfs_trans_handle *trans;
5152 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5156 * Eviction should be taking place at some place safe because of our
5157 * delayed iputs. However the normal flushing code will run delayed
5158 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5160 * We reserve the delayed_refs_extra here again because we can't use
5161 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5162 * above. We reserve our extra bit here because we generate a ton of
5163 * delayed refs activity by truncating.
5165 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5166 * if we fail to make this reservation we can re-try without the
5167 * delayed_refs_extra so we can make some forward progress.
5169 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5170 BTRFS_RESERVE_FLUSH_EVICT);
5172 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5173 BTRFS_RESERVE_FLUSH_EVICT);
5176 "could not allocate space for delete; will truncate on mount");
5177 return ERR_PTR(-ENOSPC);
5179 delayed_refs_extra = 0;
5182 trans = btrfs_join_transaction(root);
5186 if (delayed_refs_extra) {
5187 trans->block_rsv = &fs_info->trans_block_rsv;
5188 trans->bytes_reserved = delayed_refs_extra;
5189 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5190 delayed_refs_extra, true);
5195 void btrfs_evict_inode(struct inode *inode)
5197 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5198 struct btrfs_trans_handle *trans;
5199 struct btrfs_root *root = BTRFS_I(inode)->root;
5200 struct btrfs_block_rsv *rsv = NULL;
5203 trace_btrfs_inode_evict(inode);
5206 fsverity_cleanup_inode(inode);
5211 evict_inode_truncate_pages(inode);
5213 if (inode->i_nlink &&
5214 ((btrfs_root_refs(&root->root_item) != 0 &&
5215 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5216 btrfs_is_free_space_inode(BTRFS_I(inode))))
5219 if (is_bad_inode(inode))
5222 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5225 if (inode->i_nlink > 0) {
5226 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5227 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5232 * This makes sure the inode item in tree is uptodate and the space for
5233 * the inode update is released.
5235 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5240 * This drops any pending insert or delete operations we have for this
5241 * inode. We could have a delayed dir index deletion queued up, but
5242 * we're removing the inode completely so that'll be taken care of in
5245 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5247 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5250 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5251 rsv->failfast = true;
5253 btrfs_i_size_write(BTRFS_I(inode), 0);
5256 struct btrfs_truncate_control control = {
5257 .inode = BTRFS_I(inode),
5258 .ino = btrfs_ino(BTRFS_I(inode)),
5263 trans = evict_refill_and_join(root, rsv);
5267 trans->block_rsv = rsv;
5269 ret = btrfs_truncate_inode_items(trans, root, &control);
5270 trans->block_rsv = &fs_info->trans_block_rsv;
5271 btrfs_end_transaction(trans);
5273 * We have not added new delayed items for our inode after we
5274 * have flushed its delayed items, so no need to throttle on
5275 * delayed items. However we have modified extent buffers.
5277 btrfs_btree_balance_dirty_nodelay(fs_info);
5278 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5285 * Errors here aren't a big deal, it just means we leave orphan items in
5286 * the tree. They will be cleaned up on the next mount. If the inode
5287 * number gets reused, cleanup deletes the orphan item without doing
5288 * anything, and unlink reuses the existing orphan item.
5290 * If it turns out that we are dropping too many of these, we might want
5291 * to add a mechanism for retrying these after a commit.
5293 trans = evict_refill_and_join(root, rsv);
5294 if (!IS_ERR(trans)) {
5295 trans->block_rsv = rsv;
5296 btrfs_orphan_del(trans, BTRFS_I(inode));
5297 trans->block_rsv = &fs_info->trans_block_rsv;
5298 btrfs_end_transaction(trans);
5302 btrfs_free_block_rsv(fs_info, rsv);
5304 * If we didn't successfully delete, the orphan item will still be in
5305 * the tree and we'll retry on the next mount. Again, we might also want
5306 * to retry these periodically in the future.
5308 btrfs_remove_delayed_node(BTRFS_I(inode));
5309 fsverity_cleanup_inode(inode);
5314 * Return the key found in the dir entry in the location pointer, fill @type
5315 * with BTRFS_FT_*, and return 0.
5317 * If no dir entries were found, returns -ENOENT.
5318 * If found a corrupted location in dir entry, returns -EUCLEAN.
5320 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5321 struct btrfs_key *location, u8 *type)
5323 struct btrfs_dir_item *di;
5324 struct btrfs_path *path;
5325 struct btrfs_root *root = dir->root;
5327 struct fscrypt_name fname;
5329 path = btrfs_alloc_path();
5333 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5337 * fscrypt_setup_filename() should never return a positive value, but
5338 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5342 /* This needs to handle no-key deletions later on */
5344 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5345 &fname.disk_name, 0);
5346 if (IS_ERR_OR_NULL(di)) {
5347 ret = di ? PTR_ERR(di) : -ENOENT;
5351 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5352 if (location->type != BTRFS_INODE_ITEM_KEY &&
5353 location->type != BTRFS_ROOT_ITEM_KEY) {
5355 btrfs_warn(root->fs_info,
5356 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5357 __func__, fname.disk_name.name, btrfs_ino(dir),
5358 location->objectid, location->type, location->offset);
5361 *type = btrfs_dir_ftype(path->nodes[0], di);
5363 fscrypt_free_filename(&fname);
5364 btrfs_free_path(path);
5369 * when we hit a tree root in a directory, the btrfs part of the inode
5370 * needs to be changed to reflect the root directory of the tree root. This
5371 * is kind of like crossing a mount point.
5373 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5374 struct btrfs_inode *dir,
5375 struct dentry *dentry,
5376 struct btrfs_key *location,
5377 struct btrfs_root **sub_root)
5379 struct btrfs_path *path;
5380 struct btrfs_root *new_root;
5381 struct btrfs_root_ref *ref;
5382 struct extent_buffer *leaf;
5383 struct btrfs_key key;
5386 struct fscrypt_name fname;
5388 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5392 path = btrfs_alloc_path();
5399 key.objectid = dir->root->root_key.objectid;
5400 key.type = BTRFS_ROOT_REF_KEY;
5401 key.offset = location->objectid;
5403 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5410 leaf = path->nodes[0];
5411 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5412 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5413 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5416 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5417 (unsigned long)(ref + 1), fname.disk_name.len);
5421 btrfs_release_path(path);
5423 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5424 if (IS_ERR(new_root)) {
5425 err = PTR_ERR(new_root);
5429 *sub_root = new_root;
5430 location->objectid = btrfs_root_dirid(&new_root->root_item);
5431 location->type = BTRFS_INODE_ITEM_KEY;
5432 location->offset = 0;
5435 btrfs_free_path(path);
5436 fscrypt_free_filename(&fname);
5440 static void inode_tree_add(struct btrfs_inode *inode)
5442 struct btrfs_root *root = inode->root;
5443 struct btrfs_inode *entry;
5445 struct rb_node *parent;
5446 struct rb_node *new = &inode->rb_node;
5447 u64 ino = btrfs_ino(inode);
5449 if (inode_unhashed(&inode->vfs_inode))
5452 spin_lock(&root->inode_lock);
5453 p = &root->inode_tree.rb_node;
5456 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5458 if (ino < btrfs_ino(entry))
5459 p = &parent->rb_left;
5460 else if (ino > btrfs_ino(entry))
5461 p = &parent->rb_right;
5463 WARN_ON(!(entry->vfs_inode.i_state &
5464 (I_WILL_FREE | I_FREEING)));
5465 rb_replace_node(parent, new, &root->inode_tree);
5466 RB_CLEAR_NODE(parent);
5467 spin_unlock(&root->inode_lock);
5471 rb_link_node(new, parent, p);
5472 rb_insert_color(new, &root->inode_tree);
5473 spin_unlock(&root->inode_lock);
5476 static void inode_tree_del(struct btrfs_inode *inode)
5478 struct btrfs_root *root = inode->root;
5481 spin_lock(&root->inode_lock);
5482 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5483 rb_erase(&inode->rb_node, &root->inode_tree);
5484 RB_CLEAR_NODE(&inode->rb_node);
5485 empty = RB_EMPTY_ROOT(&root->inode_tree);
5487 spin_unlock(&root->inode_lock);
5489 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5490 spin_lock(&root->inode_lock);
5491 empty = RB_EMPTY_ROOT(&root->inode_tree);
5492 spin_unlock(&root->inode_lock);
5494 btrfs_add_dead_root(root);
5499 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5501 struct btrfs_iget_args *args = p;
5503 inode->i_ino = args->ino;
5504 BTRFS_I(inode)->location.objectid = args->ino;
5505 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5506 BTRFS_I(inode)->location.offset = 0;
5507 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5508 BUG_ON(args->root && !BTRFS_I(inode)->root);
5510 if (args->root && args->root == args->root->fs_info->tree_root &&
5511 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5512 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5513 &BTRFS_I(inode)->runtime_flags);
5517 static int btrfs_find_actor(struct inode *inode, void *opaque)
5519 struct btrfs_iget_args *args = opaque;
5521 return args->ino == BTRFS_I(inode)->location.objectid &&
5522 args->root == BTRFS_I(inode)->root;
5525 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5526 struct btrfs_root *root)
5528 struct inode *inode;
5529 struct btrfs_iget_args args;
5530 unsigned long hashval = btrfs_inode_hash(ino, root);
5535 inode = iget5_locked(s, hashval, btrfs_find_actor,
5536 btrfs_init_locked_inode,
5542 * Get an inode object given its inode number and corresponding root.
5543 * Path can be preallocated to prevent recursing back to iget through
5544 * allocator. NULL is also valid but may require an additional allocation
5547 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5548 struct btrfs_root *root, struct btrfs_path *path)
5550 struct inode *inode;
5552 inode = btrfs_iget_locked(s, ino, root);
5554 return ERR_PTR(-ENOMEM);
5556 if (inode->i_state & I_NEW) {
5559 ret = btrfs_read_locked_inode(inode, path);
5561 inode_tree_add(BTRFS_I(inode));
5562 unlock_new_inode(inode);
5566 * ret > 0 can come from btrfs_search_slot called by
5567 * btrfs_read_locked_inode, this means the inode item
5572 inode = ERR_PTR(ret);
5579 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5581 return btrfs_iget_path(s, ino, root, NULL);
5584 static struct inode *new_simple_dir(struct inode *dir,
5585 struct btrfs_key *key,
5586 struct btrfs_root *root)
5588 struct timespec64 ts;
5589 struct inode *inode = new_inode(dir->i_sb);
5592 return ERR_PTR(-ENOMEM);
5594 BTRFS_I(inode)->root = btrfs_grab_root(root);
5595 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5596 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5598 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5600 * We only need lookup, the rest is read-only and there's no inode
5601 * associated with the dentry
5603 inode->i_op = &simple_dir_inode_operations;
5604 inode->i_opflags &= ~IOP_XATTR;
5605 inode->i_fop = &simple_dir_operations;
5606 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5608 ts = inode_set_ctime_current(inode);
5609 inode_set_mtime_to_ts(inode, ts);
5610 inode_set_atime_to_ts(inode, inode_get_atime(dir));
5611 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
5612 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
5614 inode->i_uid = dir->i_uid;
5615 inode->i_gid = dir->i_gid;
5620 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5621 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5622 static_assert(BTRFS_FT_DIR == FT_DIR);
5623 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5624 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5625 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5626 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5627 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5629 static inline u8 btrfs_inode_type(struct inode *inode)
5631 return fs_umode_to_ftype(inode->i_mode);
5634 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5636 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5637 struct inode *inode;
5638 struct btrfs_root *root = BTRFS_I(dir)->root;
5639 struct btrfs_root *sub_root = root;
5640 struct btrfs_key location;
5644 if (dentry->d_name.len > BTRFS_NAME_LEN)
5645 return ERR_PTR(-ENAMETOOLONG);
5647 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5649 return ERR_PTR(ret);
5651 if (location.type == BTRFS_INODE_ITEM_KEY) {
5652 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5656 /* Do extra check against inode mode with di_type */
5657 if (btrfs_inode_type(inode) != di_type) {
5659 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5660 inode->i_mode, btrfs_inode_type(inode),
5663 return ERR_PTR(-EUCLEAN);
5668 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5669 &location, &sub_root);
5672 inode = ERR_PTR(ret);
5674 inode = new_simple_dir(dir, &location, root);
5676 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5677 btrfs_put_root(sub_root);
5682 down_read(&fs_info->cleanup_work_sem);
5683 if (!sb_rdonly(inode->i_sb))
5684 ret = btrfs_orphan_cleanup(sub_root);
5685 up_read(&fs_info->cleanup_work_sem);
5688 inode = ERR_PTR(ret);
5695 static int btrfs_dentry_delete(const struct dentry *dentry)
5697 struct btrfs_root *root;
5698 struct inode *inode = d_inode(dentry);
5700 if (!inode && !IS_ROOT(dentry))
5701 inode = d_inode(dentry->d_parent);
5704 root = BTRFS_I(inode)->root;
5705 if (btrfs_root_refs(&root->root_item) == 0)
5708 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5714 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5717 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5719 if (inode == ERR_PTR(-ENOENT))
5721 return d_splice_alias(inode, dentry);
5725 * Find the highest existing sequence number in a directory and then set the
5726 * in-memory index_cnt variable to the first free sequence number.
5728 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5730 struct btrfs_root *root = inode->root;
5731 struct btrfs_key key, found_key;
5732 struct btrfs_path *path;
5733 struct extent_buffer *leaf;
5736 key.objectid = btrfs_ino(inode);
5737 key.type = BTRFS_DIR_INDEX_KEY;
5738 key.offset = (u64)-1;
5740 path = btrfs_alloc_path();
5744 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5747 /* FIXME: we should be able to handle this */
5752 if (path->slots[0] == 0) {
5753 inode->index_cnt = BTRFS_DIR_START_INDEX;
5759 leaf = path->nodes[0];
5760 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5762 if (found_key.objectid != btrfs_ino(inode) ||
5763 found_key.type != BTRFS_DIR_INDEX_KEY) {
5764 inode->index_cnt = BTRFS_DIR_START_INDEX;
5768 inode->index_cnt = found_key.offset + 1;
5770 btrfs_free_path(path);
5774 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5778 btrfs_inode_lock(dir, 0);
5779 if (dir->index_cnt == (u64)-1) {
5780 ret = btrfs_inode_delayed_dir_index_count(dir);
5782 ret = btrfs_set_inode_index_count(dir);
5788 /* index_cnt is the index number of next new entry, so decrement it. */
5789 *index = dir->index_cnt - 1;
5791 btrfs_inode_unlock(dir, 0);
5797 * All this infrastructure exists because dir_emit can fault, and we are holding
5798 * the tree lock when doing readdir. For now just allocate a buffer and copy
5799 * our information into that, and then dir_emit from the buffer. This is
5800 * similar to what NFS does, only we don't keep the buffer around in pagecache
5801 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5802 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5805 static int btrfs_opendir(struct inode *inode, struct file *file)
5807 struct btrfs_file_private *private;
5811 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5815 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5818 private->last_index = last_index;
5819 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5820 if (!private->filldir_buf) {
5824 file->private_data = private;
5828 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5830 struct btrfs_file_private *private = file->private_data;
5833 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
5834 &private->last_index);
5838 return generic_file_llseek(file, offset, whence);
5848 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5851 struct dir_entry *entry = addr;
5852 char *name = (char *)(entry + 1);
5854 ctx->pos = get_unaligned(&entry->offset);
5855 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5856 get_unaligned(&entry->ino),
5857 get_unaligned(&entry->type)))
5859 addr += sizeof(struct dir_entry) +
5860 get_unaligned(&entry->name_len);
5866 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5868 struct inode *inode = file_inode(file);
5869 struct btrfs_root *root = BTRFS_I(inode)->root;
5870 struct btrfs_file_private *private = file->private_data;
5871 struct btrfs_dir_item *di;
5872 struct btrfs_key key;
5873 struct btrfs_key found_key;
5874 struct btrfs_path *path;
5876 LIST_HEAD(ins_list);
5877 LIST_HEAD(del_list);
5884 struct btrfs_key location;
5886 if (!dir_emit_dots(file, ctx))
5889 path = btrfs_alloc_path();
5893 addr = private->filldir_buf;
5894 path->reada = READA_FORWARD;
5896 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5897 &ins_list, &del_list);
5900 key.type = BTRFS_DIR_INDEX_KEY;
5901 key.offset = ctx->pos;
5902 key.objectid = btrfs_ino(BTRFS_I(inode));
5904 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5905 struct dir_entry *entry;
5906 struct extent_buffer *leaf = path->nodes[0];
5909 if (found_key.objectid != key.objectid)
5911 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5913 if (found_key.offset < ctx->pos)
5915 if (found_key.offset > private->last_index)
5917 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5919 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5920 name_len = btrfs_dir_name_len(leaf, di);
5921 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5923 btrfs_release_path(path);
5924 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5927 addr = private->filldir_buf;
5933 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5935 name_ptr = (char *)(entry + 1);
5936 read_extent_buffer(leaf, name_ptr,
5937 (unsigned long)(di + 1), name_len);
5938 put_unaligned(name_len, &entry->name_len);
5939 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5940 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5941 put_unaligned(location.objectid, &entry->ino);
5942 put_unaligned(found_key.offset, &entry->offset);
5944 addr += sizeof(struct dir_entry) + name_len;
5945 total_len += sizeof(struct dir_entry) + name_len;
5947 /* Catch error encountered during iteration */
5951 btrfs_release_path(path);
5953 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5957 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5962 * Stop new entries from being returned after we return the last
5965 * New directory entries are assigned a strictly increasing
5966 * offset. This means that new entries created during readdir
5967 * are *guaranteed* to be seen in the future by that readdir.
5968 * This has broken buggy programs which operate on names as
5969 * they're returned by readdir. Until we re-use freed offsets
5970 * we have this hack to stop new entries from being returned
5971 * under the assumption that they'll never reach this huge
5974 * This is being careful not to overflow 32bit loff_t unless the
5975 * last entry requires it because doing so has broken 32bit apps
5978 if (ctx->pos >= INT_MAX)
5979 ctx->pos = LLONG_MAX;
5986 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5987 btrfs_free_path(path);
5992 * This is somewhat expensive, updating the tree every time the
5993 * inode changes. But, it is most likely to find the inode in cache.
5994 * FIXME, needs more benchmarking...there are no reasons other than performance
5995 * to keep or drop this code.
5997 static int btrfs_dirty_inode(struct btrfs_inode *inode)
5999 struct btrfs_root *root = inode->root;
6000 struct btrfs_fs_info *fs_info = root->fs_info;
6001 struct btrfs_trans_handle *trans;
6004 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6007 trans = btrfs_join_transaction(root);
6009 return PTR_ERR(trans);
6011 ret = btrfs_update_inode(trans, inode);
6012 if (ret == -ENOSPC || ret == -EDQUOT) {
6013 /* whoops, lets try again with the full transaction */
6014 btrfs_end_transaction(trans);
6015 trans = btrfs_start_transaction(root, 1);
6017 return PTR_ERR(trans);
6019 ret = btrfs_update_inode(trans, inode);
6021 btrfs_end_transaction(trans);
6022 if (inode->delayed_node)
6023 btrfs_balance_delayed_items(fs_info);
6029 * This is a copy of file_update_time. We need this so we can return error on
6030 * ENOSPC for updating the inode in the case of file write and mmap writes.
6032 static int btrfs_update_time(struct inode *inode, int flags)
6034 struct btrfs_root *root = BTRFS_I(inode)->root;
6037 if (btrfs_root_readonly(root))
6040 dirty = inode_update_timestamps(inode, flags);
6041 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6045 * helper to find a free sequence number in a given directory. This current
6046 * code is very simple, later versions will do smarter things in the btree
6048 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6052 if (dir->index_cnt == (u64)-1) {
6053 ret = btrfs_inode_delayed_dir_index_count(dir);
6055 ret = btrfs_set_inode_index_count(dir);
6061 *index = dir->index_cnt;
6067 static int btrfs_insert_inode_locked(struct inode *inode)
6069 struct btrfs_iget_args args;
6071 args.ino = BTRFS_I(inode)->location.objectid;
6072 args.root = BTRFS_I(inode)->root;
6074 return insert_inode_locked4(inode,
6075 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6076 btrfs_find_actor, &args);
6079 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6080 unsigned int *trans_num_items)
6082 struct inode *dir = args->dir;
6083 struct inode *inode = args->inode;
6086 if (!args->orphan) {
6087 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6093 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6095 fscrypt_free_filename(&args->fname);
6099 /* 1 to add inode item */
6100 *trans_num_items = 1;
6101 /* 1 to add compression property */
6102 if (BTRFS_I(dir)->prop_compress)
6103 (*trans_num_items)++;
6104 /* 1 to add default ACL xattr */
6105 if (args->default_acl)
6106 (*trans_num_items)++;
6107 /* 1 to add access ACL xattr */
6109 (*trans_num_items)++;
6110 #ifdef CONFIG_SECURITY
6111 /* 1 to add LSM xattr */
6112 if (dir->i_security)
6113 (*trans_num_items)++;
6116 /* 1 to add orphan item */
6117 (*trans_num_items)++;
6121 * 1 to add dir index
6122 * 1 to update parent inode item
6124 * No need for 1 unit for the inode ref item because it is
6125 * inserted in a batch together with the inode item at
6126 * btrfs_create_new_inode().
6128 *trans_num_items += 3;
6133 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6135 posix_acl_release(args->acl);
6136 posix_acl_release(args->default_acl);
6137 fscrypt_free_filename(&args->fname);
6141 * Inherit flags from the parent inode.
6143 * Currently only the compression flags and the cow flags are inherited.
6145 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6151 if (flags & BTRFS_INODE_NOCOMPRESS) {
6152 inode->flags &= ~BTRFS_INODE_COMPRESS;
6153 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6154 } else if (flags & BTRFS_INODE_COMPRESS) {
6155 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6156 inode->flags |= BTRFS_INODE_COMPRESS;
6159 if (flags & BTRFS_INODE_NODATACOW) {
6160 inode->flags |= BTRFS_INODE_NODATACOW;
6161 if (S_ISREG(inode->vfs_inode.i_mode))
6162 inode->flags |= BTRFS_INODE_NODATASUM;
6165 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6168 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6169 struct btrfs_new_inode_args *args)
6171 struct timespec64 ts;
6172 struct inode *dir = args->dir;
6173 struct inode *inode = args->inode;
6174 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6175 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6176 struct btrfs_root *root;
6177 struct btrfs_inode_item *inode_item;
6178 struct btrfs_key *location;
6179 struct btrfs_path *path;
6181 struct btrfs_inode_ref *ref;
6182 struct btrfs_key key[2];
6184 struct btrfs_item_batch batch;
6188 path = btrfs_alloc_path();
6193 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6194 root = BTRFS_I(inode)->root;
6196 ret = btrfs_get_free_objectid(root, &objectid);
6199 inode->i_ino = objectid;
6203 * O_TMPFILE, set link count to 0, so that after this point, we
6204 * fill in an inode item with the correct link count.
6206 set_nlink(inode, 0);
6208 trace_btrfs_inode_request(dir);
6210 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6214 /* index_cnt is ignored for everything but a dir. */
6215 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6216 BTRFS_I(inode)->generation = trans->transid;
6217 inode->i_generation = BTRFS_I(inode)->generation;
6220 * Subvolumes don't inherit flags from their parent directory.
6221 * Originally this was probably by accident, but we probably can't
6222 * change it now without compatibility issues.
6225 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6227 if (S_ISREG(inode->i_mode)) {
6228 if (btrfs_test_opt(fs_info, NODATASUM))
6229 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6230 if (btrfs_test_opt(fs_info, NODATACOW))
6231 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6232 BTRFS_INODE_NODATASUM;
6235 location = &BTRFS_I(inode)->location;
6236 location->objectid = objectid;
6237 location->offset = 0;
6238 location->type = BTRFS_INODE_ITEM_KEY;
6240 ret = btrfs_insert_inode_locked(inode);
6243 BTRFS_I(dir)->index_cnt--;
6248 * We could have gotten an inode number from somebody who was fsynced
6249 * and then removed in this same transaction, so let's just set full
6250 * sync since it will be a full sync anyway and this will blow away the
6251 * old info in the log.
6253 btrfs_set_inode_full_sync(BTRFS_I(inode));
6255 key[0].objectid = objectid;
6256 key[0].type = BTRFS_INODE_ITEM_KEY;
6259 sizes[0] = sizeof(struct btrfs_inode_item);
6261 if (!args->orphan) {
6263 * Start new inodes with an inode_ref. This is slightly more
6264 * efficient for small numbers of hard links since they will
6265 * be packed into one item. Extended refs will kick in if we
6266 * add more hard links than can fit in the ref item.
6268 key[1].objectid = objectid;
6269 key[1].type = BTRFS_INODE_REF_KEY;
6271 key[1].offset = objectid;
6272 sizes[1] = 2 + sizeof(*ref);
6274 key[1].offset = btrfs_ino(BTRFS_I(dir));
6275 sizes[1] = name->len + sizeof(*ref);
6279 batch.keys = &key[0];
6280 batch.data_sizes = &sizes[0];
6281 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6282 batch.nr = args->orphan ? 1 : 2;
6283 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6285 btrfs_abort_transaction(trans, ret);
6289 ts = simple_inode_init_ts(inode);
6290 BTRFS_I(inode)->i_otime_sec = ts.tv_sec;
6291 BTRFS_I(inode)->i_otime_nsec = ts.tv_nsec;
6294 * We're going to fill the inode item now, so at this point the inode
6295 * must be fully initialized.
6298 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6299 struct btrfs_inode_item);
6300 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6301 sizeof(*inode_item));
6302 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6304 if (!args->orphan) {
6305 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6306 struct btrfs_inode_ref);
6307 ptr = (unsigned long)(ref + 1);
6309 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6310 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6311 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6313 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6315 btrfs_set_inode_ref_index(path->nodes[0], ref,
6316 BTRFS_I(inode)->dir_index);
6317 write_extent_buffer(path->nodes[0], name->name, ptr,
6322 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
6324 * We don't need the path anymore, plus inheriting properties, adding
6325 * ACLs, security xattrs, orphan item or adding the link, will result in
6326 * allocating yet another path. So just free our path.
6328 btrfs_free_path(path);
6332 struct inode *parent;
6335 * Subvolumes inherit properties from their parent subvolume,
6336 * not the directory they were created in.
6338 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6339 BTRFS_I(dir)->root);
6340 if (IS_ERR(parent)) {
6341 ret = PTR_ERR(parent);
6343 ret = btrfs_inode_inherit_props(trans, inode, parent);
6347 ret = btrfs_inode_inherit_props(trans, inode, dir);
6351 "error inheriting props for ino %llu (root %llu): %d",
6352 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6357 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6360 if (!args->subvol) {
6361 ret = btrfs_init_inode_security(trans, args);
6363 btrfs_abort_transaction(trans, ret);
6368 inode_tree_add(BTRFS_I(inode));
6370 trace_btrfs_inode_new(inode);
6371 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6373 btrfs_update_root_times(trans, root);
6376 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6378 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6379 0, BTRFS_I(inode)->dir_index);
6382 btrfs_abort_transaction(trans, ret);
6390 * discard_new_inode() calls iput(), but the caller owns the reference
6394 discard_new_inode(inode);
6396 btrfs_free_path(path);
6401 * utility function to add 'inode' into 'parent_inode' with
6402 * a give name and a given sequence number.
6403 * if 'add_backref' is true, also insert a backref from the
6404 * inode to the parent directory.
6406 int btrfs_add_link(struct btrfs_trans_handle *trans,
6407 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6408 const struct fscrypt_str *name, int add_backref, u64 index)
6411 struct btrfs_key key;
6412 struct btrfs_root *root = parent_inode->root;
6413 u64 ino = btrfs_ino(inode);
6414 u64 parent_ino = btrfs_ino(parent_inode);
6416 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6417 memcpy(&key, &inode->root->root_key, sizeof(key));
6420 key.type = BTRFS_INODE_ITEM_KEY;
6424 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6425 ret = btrfs_add_root_ref(trans, key.objectid,
6426 root->root_key.objectid, parent_ino,
6428 } else if (add_backref) {
6429 ret = btrfs_insert_inode_ref(trans, root, name,
6430 ino, parent_ino, index);
6433 /* Nothing to clean up yet */
6437 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6438 btrfs_inode_type(&inode->vfs_inode), index);
6439 if (ret == -EEXIST || ret == -EOVERFLOW)
6442 btrfs_abort_transaction(trans, ret);
6446 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6448 inode_inc_iversion(&parent_inode->vfs_inode);
6450 * If we are replaying a log tree, we do not want to update the mtime
6451 * and ctime of the parent directory with the current time, since the
6452 * log replay procedure is responsible for setting them to their correct
6453 * values (the ones it had when the fsync was done).
6455 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6456 inode_set_mtime_to_ts(&parent_inode->vfs_inode,
6457 inode_set_ctime_current(&parent_inode->vfs_inode));
6459 ret = btrfs_update_inode(trans, parent_inode);
6461 btrfs_abort_transaction(trans, ret);
6465 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6468 err = btrfs_del_root_ref(trans, key.objectid,
6469 root->root_key.objectid, parent_ino,
6470 &local_index, name);
6472 btrfs_abort_transaction(trans, err);
6473 } else if (add_backref) {
6477 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6480 btrfs_abort_transaction(trans, err);
6483 /* Return the original error code */
6487 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6488 struct inode *inode)
6490 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6491 struct btrfs_root *root = BTRFS_I(dir)->root;
6492 struct btrfs_new_inode_args new_inode_args = {
6497 unsigned int trans_num_items;
6498 struct btrfs_trans_handle *trans;
6501 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6505 trans = btrfs_start_transaction(root, trans_num_items);
6506 if (IS_ERR(trans)) {
6507 err = PTR_ERR(trans);
6508 goto out_new_inode_args;
6511 err = btrfs_create_new_inode(trans, &new_inode_args);
6513 d_instantiate_new(dentry, inode);
6515 btrfs_end_transaction(trans);
6516 btrfs_btree_balance_dirty(fs_info);
6518 btrfs_new_inode_args_destroy(&new_inode_args);
6525 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6526 struct dentry *dentry, umode_t mode, dev_t rdev)
6528 struct inode *inode;
6530 inode = new_inode(dir->i_sb);
6533 inode_init_owner(idmap, inode, dir, mode);
6534 inode->i_op = &btrfs_special_inode_operations;
6535 init_special_inode(inode, inode->i_mode, rdev);
6536 return btrfs_create_common(dir, dentry, inode);
6539 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6540 struct dentry *dentry, umode_t mode, bool excl)
6542 struct inode *inode;
6544 inode = new_inode(dir->i_sb);
6547 inode_init_owner(idmap, inode, dir, mode);
6548 inode->i_fop = &btrfs_file_operations;
6549 inode->i_op = &btrfs_file_inode_operations;
6550 inode->i_mapping->a_ops = &btrfs_aops;
6551 return btrfs_create_common(dir, dentry, inode);
6554 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6555 struct dentry *dentry)
6557 struct btrfs_trans_handle *trans = NULL;
6558 struct btrfs_root *root = BTRFS_I(dir)->root;
6559 struct inode *inode = d_inode(old_dentry);
6560 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6561 struct fscrypt_name fname;
6566 /* do not allow sys_link's with other subvols of the same device */
6567 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6570 if (inode->i_nlink >= BTRFS_LINK_MAX)
6573 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6577 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6582 * 2 items for inode and inode ref
6583 * 2 items for dir items
6584 * 1 item for parent inode
6585 * 1 item for orphan item deletion if O_TMPFILE
6587 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6588 if (IS_ERR(trans)) {
6589 err = PTR_ERR(trans);
6594 /* There are several dir indexes for this inode, clear the cache. */
6595 BTRFS_I(inode)->dir_index = 0ULL;
6597 inode_inc_iversion(inode);
6598 inode_set_ctime_current(inode);
6600 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6602 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6603 &fname.disk_name, 1, index);
6608 struct dentry *parent = dentry->d_parent;
6610 err = btrfs_update_inode(trans, BTRFS_I(inode));
6613 if (inode->i_nlink == 1) {
6615 * If new hard link count is 1, it's a file created
6616 * with open(2) O_TMPFILE flag.
6618 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6622 d_instantiate(dentry, inode);
6623 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6627 fscrypt_free_filename(&fname);
6629 btrfs_end_transaction(trans);
6631 inode_dec_link_count(inode);
6634 btrfs_btree_balance_dirty(fs_info);
6638 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6639 struct dentry *dentry, umode_t mode)
6641 struct inode *inode;
6643 inode = new_inode(dir->i_sb);
6646 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6647 inode->i_op = &btrfs_dir_inode_operations;
6648 inode->i_fop = &btrfs_dir_file_operations;
6649 return btrfs_create_common(dir, dentry, inode);
6652 static noinline int uncompress_inline(struct btrfs_path *path,
6654 struct btrfs_file_extent_item *item)
6657 struct extent_buffer *leaf = path->nodes[0];
6660 unsigned long inline_size;
6664 compress_type = btrfs_file_extent_compression(leaf, item);
6665 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6666 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6667 tmp = kmalloc(inline_size, GFP_NOFS);
6670 ptr = btrfs_file_extent_inline_start(item);
6672 read_extent_buffer(leaf, tmp, ptr, inline_size);
6674 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6675 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6678 * decompression code contains a memset to fill in any space between the end
6679 * of the uncompressed data and the end of max_size in case the decompressed
6680 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6681 * the end of an inline extent and the beginning of the next block, so we
6682 * cover that region here.
6685 if (max_size < PAGE_SIZE)
6686 memzero_page(page, max_size, PAGE_SIZE - max_size);
6691 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6694 struct btrfs_file_extent_item *fi;
6698 if (!page || PageUptodate(page))
6701 ASSERT(page_offset(page) == 0);
6703 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6704 struct btrfs_file_extent_item);
6705 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6706 return uncompress_inline(path, page, fi);
6708 copy_size = min_t(u64, PAGE_SIZE,
6709 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6710 kaddr = kmap_local_page(page);
6711 read_extent_buffer(path->nodes[0], kaddr,
6712 btrfs_file_extent_inline_start(fi), copy_size);
6713 kunmap_local(kaddr);
6714 if (copy_size < PAGE_SIZE)
6715 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6720 * Lookup the first extent overlapping a range in a file.
6722 * @inode: file to search in
6723 * @page: page to read extent data into if the extent is inline
6724 * @pg_offset: offset into @page to copy to
6725 * @start: file offset
6726 * @len: length of range starting at @start
6728 * Return the first &struct extent_map which overlaps the given range, reading
6729 * it from the B-tree and caching it if necessary. Note that there may be more
6730 * extents which overlap the given range after the returned extent_map.
6732 * If @page is not NULL and the extent is inline, this also reads the extent
6733 * data directly into the page and marks the extent up to date in the io_tree.
6735 * Return: ERR_PTR on error, non-NULL extent_map on success.
6737 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6738 struct page *page, size_t pg_offset,
6741 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6743 u64 extent_start = 0;
6745 u64 objectid = btrfs_ino(inode);
6746 int extent_type = -1;
6747 struct btrfs_path *path = NULL;
6748 struct btrfs_root *root = inode->root;
6749 struct btrfs_file_extent_item *item;
6750 struct extent_buffer *leaf;
6751 struct btrfs_key found_key;
6752 struct extent_map *em = NULL;
6753 struct extent_map_tree *em_tree = &inode->extent_tree;
6755 read_lock(&em_tree->lock);
6756 em = lookup_extent_mapping(em_tree, start, len);
6757 read_unlock(&em_tree->lock);
6760 if (em->start > start || em->start + em->len <= start)
6761 free_extent_map(em);
6762 else if (em->block_start == EXTENT_MAP_INLINE && page)
6763 free_extent_map(em);
6767 em = alloc_extent_map();
6772 em->start = EXTENT_MAP_HOLE;
6773 em->orig_start = EXTENT_MAP_HOLE;
6775 em->block_len = (u64)-1;
6777 path = btrfs_alloc_path();
6783 /* Chances are we'll be called again, so go ahead and do readahead */
6784 path->reada = READA_FORWARD;
6787 * The same explanation in load_free_space_cache applies here as well,
6788 * we only read when we're loading the free space cache, and at that
6789 * point the commit_root has everything we need.
6791 if (btrfs_is_free_space_inode(inode)) {
6792 path->search_commit_root = 1;
6793 path->skip_locking = 1;
6796 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6799 } else if (ret > 0) {
6800 if (path->slots[0] == 0)
6806 leaf = path->nodes[0];
6807 item = btrfs_item_ptr(leaf, path->slots[0],
6808 struct btrfs_file_extent_item);
6809 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6810 if (found_key.objectid != objectid ||
6811 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6813 * If we backup past the first extent we want to move forward
6814 * and see if there is an extent in front of us, otherwise we'll
6815 * say there is a hole for our whole search range which can
6822 extent_type = btrfs_file_extent_type(leaf, item);
6823 extent_start = found_key.offset;
6824 extent_end = btrfs_file_extent_end(path);
6825 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6826 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6827 /* Only regular file could have regular/prealloc extent */
6828 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6831 "regular/prealloc extent found for non-regular inode %llu",
6835 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6837 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6838 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6843 if (start >= extent_end) {
6845 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6846 ret = btrfs_next_leaf(root, path);
6852 leaf = path->nodes[0];
6854 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6855 if (found_key.objectid != objectid ||
6856 found_key.type != BTRFS_EXTENT_DATA_KEY)
6858 if (start + len <= found_key.offset)
6860 if (start > found_key.offset)
6863 /* New extent overlaps with existing one */
6865 em->orig_start = start;
6866 em->len = found_key.offset - start;
6867 em->block_start = EXTENT_MAP_HOLE;
6871 btrfs_extent_item_to_extent_map(inode, path, item, em);
6873 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6874 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6876 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6878 * Inline extent can only exist at file offset 0. This is
6879 * ensured by tree-checker and inline extent creation path.
6880 * Thus all members representing file offsets should be zero.
6882 ASSERT(pg_offset == 0);
6883 ASSERT(extent_start == 0);
6884 ASSERT(em->start == 0);
6887 * btrfs_extent_item_to_extent_map() should have properly
6888 * initialized em members already.
6890 * Other members are not utilized for inline extents.
6892 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6893 ASSERT(em->len == fs_info->sectorsize);
6895 ret = read_inline_extent(inode, path, page);
6902 em->orig_start = start;
6904 em->block_start = EXTENT_MAP_HOLE;
6907 btrfs_release_path(path);
6908 if (em->start > start || extent_map_end(em) <= start) {
6910 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6911 em->start, em->len, start, len);
6916 write_lock(&em_tree->lock);
6917 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6918 write_unlock(&em_tree->lock);
6920 btrfs_free_path(path);
6922 trace_btrfs_get_extent(root, inode, em);
6925 free_extent_map(em);
6926 return ERR_PTR(ret);
6931 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6932 struct btrfs_dio_data *dio_data,
6935 const u64 orig_start,
6936 const u64 block_start,
6937 const u64 block_len,
6938 const u64 orig_block_len,
6939 const u64 ram_bytes,
6942 struct extent_map *em = NULL;
6943 struct btrfs_ordered_extent *ordered;
6945 if (type != BTRFS_ORDERED_NOCOW) {
6946 em = create_io_em(inode, start, len, orig_start, block_start,
6947 block_len, orig_block_len, ram_bytes,
6948 BTRFS_COMPRESS_NONE, /* compress_type */
6953 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6954 block_start, block_len, 0,
6956 (1 << BTRFS_ORDERED_DIRECT),
6957 BTRFS_COMPRESS_NONE);
6958 if (IS_ERR(ordered)) {
6960 free_extent_map(em);
6961 btrfs_drop_extent_map_range(inode, start,
6962 start + len - 1, false);
6964 em = ERR_CAST(ordered);
6966 ASSERT(!dio_data->ordered);
6967 dio_data->ordered = ordered;
6974 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6975 struct btrfs_dio_data *dio_data,
6978 struct btrfs_root *root = inode->root;
6979 struct btrfs_fs_info *fs_info = root->fs_info;
6980 struct extent_map *em;
6981 struct btrfs_key ins;
6985 alloc_hint = get_extent_allocation_hint(inode, start, len);
6987 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6988 0, alloc_hint, &ins, 1, 1);
6989 if (ret == -EAGAIN) {
6990 ASSERT(btrfs_is_zoned(fs_info));
6991 wait_on_bit_io(&inode->root->fs_info->flags, BTRFS_FS_NEED_ZONE_FINISH,
6992 TASK_UNINTERRUPTIBLE);
6996 return ERR_PTR(ret);
6998 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
6999 ins.objectid, ins.offset, ins.offset,
7000 ins.offset, BTRFS_ORDERED_REGULAR);
7001 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7003 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7009 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7011 struct btrfs_block_group *block_group;
7012 bool readonly = false;
7014 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7015 if (!block_group || block_group->ro)
7018 btrfs_put_block_group(block_group);
7023 * Check if we can do nocow write into the range [@offset, @offset + @len)
7025 * @offset: File offset
7026 * @len: The length to write, will be updated to the nocow writeable
7028 * @orig_start: (optional) Return the original file offset of the file extent
7029 * @orig_len: (optional) Return the original on-disk length of the file extent
7030 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7031 * @strict: if true, omit optimizations that might force us into unnecessary
7032 * cow. e.g., don't trust generation number.
7035 * >0 and update @len if we can do nocow write
7036 * 0 if we can't do nocow write
7037 * <0 if error happened
7039 * NOTE: This only checks the file extents, caller is responsible to wait for
7040 * any ordered extents.
7042 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7043 u64 *orig_start, u64 *orig_block_len,
7044 u64 *ram_bytes, bool nowait, bool strict)
7046 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7047 struct can_nocow_file_extent_args nocow_args = { 0 };
7048 struct btrfs_path *path;
7050 struct extent_buffer *leaf;
7051 struct btrfs_root *root = BTRFS_I(inode)->root;
7052 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7053 struct btrfs_file_extent_item *fi;
7054 struct btrfs_key key;
7057 path = btrfs_alloc_path();
7060 path->nowait = nowait;
7062 ret = btrfs_lookup_file_extent(NULL, root, path,
7063 btrfs_ino(BTRFS_I(inode)), offset, 0);
7068 if (path->slots[0] == 0) {
7069 /* can't find the item, must cow */
7076 leaf = path->nodes[0];
7077 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7078 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7079 key.type != BTRFS_EXTENT_DATA_KEY) {
7080 /* not our file or wrong item type, must cow */
7084 if (key.offset > offset) {
7085 /* Wrong offset, must cow */
7089 if (btrfs_file_extent_end(path) <= offset)
7092 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7093 found_type = btrfs_file_extent_type(leaf, fi);
7095 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7097 nocow_args.start = offset;
7098 nocow_args.end = offset + *len - 1;
7099 nocow_args.strict = strict;
7100 nocow_args.free_path = true;
7102 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7103 /* can_nocow_file_extent() has freed the path. */
7107 /* Treat errors as not being able to NOCOW. */
7113 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7116 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7117 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7120 range_end = round_up(offset + nocow_args.num_bytes,
7121 root->fs_info->sectorsize) - 1;
7122 ret = test_range_bit_exists(io_tree, offset, range_end, EXTENT_DELALLOC);
7130 *orig_start = key.offset - nocow_args.extent_offset;
7132 *orig_block_len = nocow_args.disk_num_bytes;
7134 *len = nocow_args.num_bytes;
7137 btrfs_free_path(path);
7141 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7142 struct extent_state **cached_state,
7143 unsigned int iomap_flags)
7145 const bool writing = (iomap_flags & IOMAP_WRITE);
7146 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7147 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7148 struct btrfs_ordered_extent *ordered;
7153 if (!try_lock_extent(io_tree, lockstart, lockend,
7157 lock_extent(io_tree, lockstart, lockend, cached_state);
7160 * We're concerned with the entire range that we're going to be
7161 * doing DIO to, so we need to make sure there's no ordered
7162 * extents in this range.
7164 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7165 lockend - lockstart + 1);
7168 * We need to make sure there are no buffered pages in this
7169 * range either, we could have raced between the invalidate in
7170 * generic_file_direct_write and locking the extent. The
7171 * invalidate needs to happen so that reads after a write do not
7175 (!writing || !filemap_range_has_page(inode->i_mapping,
7176 lockstart, lockend)))
7179 unlock_extent(io_tree, lockstart, lockend, cached_state);
7183 btrfs_put_ordered_extent(ordered);
7188 * If we are doing a DIO read and the ordered extent we
7189 * found is for a buffered write, we can not wait for it
7190 * to complete and retry, because if we do so we can
7191 * deadlock with concurrent buffered writes on page
7192 * locks. This happens only if our DIO read covers more
7193 * than one extent map, if at this point has already
7194 * created an ordered extent for a previous extent map
7195 * and locked its range in the inode's io tree, and a
7196 * concurrent write against that previous extent map's
7197 * range and this range started (we unlock the ranges
7198 * in the io tree only when the bios complete and
7199 * buffered writes always lock pages before attempting
7200 * to lock range in the io tree).
7203 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7204 btrfs_start_ordered_extent(ordered);
7206 ret = nowait ? -EAGAIN : -ENOTBLK;
7207 btrfs_put_ordered_extent(ordered);
7210 * We could trigger writeback for this range (and wait
7211 * for it to complete) and then invalidate the pages for
7212 * this range (through invalidate_inode_pages2_range()),
7213 * but that can lead us to a deadlock with a concurrent
7214 * call to readahead (a buffered read or a defrag call
7215 * triggered a readahead) on a page lock due to an
7216 * ordered dio extent we created before but did not have
7217 * yet a corresponding bio submitted (whence it can not
7218 * complete), which makes readahead wait for that
7219 * ordered extent to complete while holding a lock on
7222 ret = nowait ? -EAGAIN : -ENOTBLK;
7234 /* The callers of this must take lock_extent() */
7235 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7236 u64 len, u64 orig_start, u64 block_start,
7237 u64 block_len, u64 orig_block_len,
7238 u64 ram_bytes, int compress_type,
7241 struct extent_map *em;
7244 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7245 type == BTRFS_ORDERED_COMPRESSED ||
7246 type == BTRFS_ORDERED_NOCOW ||
7247 type == BTRFS_ORDERED_REGULAR);
7249 em = alloc_extent_map();
7251 return ERR_PTR(-ENOMEM);
7254 em->orig_start = orig_start;
7256 em->block_len = block_len;
7257 em->block_start = block_start;
7258 em->orig_block_len = orig_block_len;
7259 em->ram_bytes = ram_bytes;
7260 em->generation = -1;
7261 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7262 if (type == BTRFS_ORDERED_PREALLOC) {
7263 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7264 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7265 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7266 em->compress_type = compress_type;
7269 ret = btrfs_replace_extent_map_range(inode, em, true);
7271 free_extent_map(em);
7272 return ERR_PTR(ret);
7275 /* em got 2 refs now, callers needs to do free_extent_map once. */
7280 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7281 struct inode *inode,
7282 struct btrfs_dio_data *dio_data,
7283 u64 start, u64 *lenp,
7284 unsigned int iomap_flags)
7286 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7287 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7288 struct extent_map *em = *map;
7290 u64 block_start, orig_start, orig_block_len, ram_bytes;
7291 struct btrfs_block_group *bg;
7292 bool can_nocow = false;
7293 bool space_reserved = false;
7299 * We don't allocate a new extent in the following cases
7301 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7303 * 2) The extent is marked as PREALLOC. We're good to go here and can
7304 * just use the extent.
7307 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7308 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7309 em->block_start != EXTENT_MAP_HOLE)) {
7310 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7311 type = BTRFS_ORDERED_PREALLOC;
7313 type = BTRFS_ORDERED_NOCOW;
7314 len = min(len, em->len - (start - em->start));
7315 block_start = em->block_start + (start - em->start);
7317 if (can_nocow_extent(inode, start, &len, &orig_start,
7318 &orig_block_len, &ram_bytes, false, false) == 1) {
7319 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7327 struct extent_map *em2;
7329 /* We can NOCOW, so only need to reserve metadata space. */
7330 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7333 /* Our caller expects us to free the input extent map. */
7334 free_extent_map(em);
7336 btrfs_dec_nocow_writers(bg);
7337 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7341 space_reserved = true;
7343 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7344 orig_start, block_start,
7345 len, orig_block_len,
7347 btrfs_dec_nocow_writers(bg);
7348 if (type == BTRFS_ORDERED_PREALLOC) {
7349 free_extent_map(em);
7359 dio_data->nocow_done = true;
7361 /* Our caller expects us to free the input extent map. */
7362 free_extent_map(em);
7371 * If we could not allocate data space before locking the file
7372 * range and we can't do a NOCOW write, then we have to fail.
7374 if (!dio_data->data_space_reserved) {
7380 * We have to COW and we have already reserved data space before,
7381 * so now we reserve only metadata.
7383 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7387 space_reserved = true;
7389 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7395 len = min(len, em->len - (start - em->start));
7397 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7398 prev_len - len, true);
7402 * We have created our ordered extent, so we can now release our reservation
7403 * for an outstanding extent.
7405 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7408 * Need to update the i_size under the extent lock so buffered
7409 * readers will get the updated i_size when we unlock.
7411 if (start + len > i_size_read(inode))
7412 i_size_write(inode, start + len);
7414 if (ret && space_reserved) {
7415 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7416 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7422 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7423 loff_t length, unsigned int flags, struct iomap *iomap,
7424 struct iomap *srcmap)
7426 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7427 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7428 struct extent_map *em;
7429 struct extent_state *cached_state = NULL;
7430 struct btrfs_dio_data *dio_data = iter->private;
7431 u64 lockstart, lockend;
7432 const bool write = !!(flags & IOMAP_WRITE);
7435 const u64 data_alloc_len = length;
7436 bool unlock_extents = false;
7439 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7440 * we're NOWAIT we may submit a bio for a partial range and return
7441 * EIOCBQUEUED, which would result in an errant short read.
7443 * The best way to handle this would be to allow for partial completions
7444 * of iocb's, so we could submit the partial bio, return and fault in
7445 * the rest of the pages, and then submit the io for the rest of the
7446 * range. However we don't have that currently, so simply return
7447 * -EAGAIN at this point so that the normal path is used.
7449 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7453 * Cap the size of reads to that usually seen in buffered I/O as we need
7454 * to allocate a contiguous array for the checksums.
7457 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7460 lockend = start + len - 1;
7463 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7464 * enough if we've written compressed pages to this area, so we need to
7465 * flush the dirty pages again to make absolutely sure that any
7466 * outstanding dirty pages are on disk - the first flush only starts
7467 * compression on the data, while keeping the pages locked, so by the
7468 * time the second flush returns we know bios for the compressed pages
7469 * were submitted and finished, and the pages no longer under writeback.
7471 * If we have a NOWAIT request and we have any pages in the range that
7472 * are locked, likely due to compression still in progress, we don't want
7473 * to block on page locks. We also don't want to block on pages marked as
7474 * dirty or under writeback (same as for the non-compression case).
7475 * iomap_dio_rw() did the same check, but after that and before we got
7476 * here, mmap'ed writes may have happened or buffered reads started
7477 * (readpage() and readahead(), which lock pages), as we haven't locked
7478 * the file range yet.
7480 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7481 &BTRFS_I(inode)->runtime_flags)) {
7482 if (flags & IOMAP_NOWAIT) {
7483 if (filemap_range_needs_writeback(inode->i_mapping,
7484 lockstart, lockend))
7487 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7488 start + length - 1);
7494 memset(dio_data, 0, sizeof(*dio_data));
7497 * We always try to allocate data space and must do it before locking
7498 * the file range, to avoid deadlocks with concurrent writes to the same
7499 * range if the range has several extents and the writes don't expand the
7500 * current i_size (the inode lock is taken in shared mode). If we fail to
7501 * allocate data space here we continue and later, after locking the
7502 * file range, we fail with ENOSPC only if we figure out we can not do a
7505 if (write && !(flags & IOMAP_NOWAIT)) {
7506 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7507 &dio_data->data_reserved,
7508 start, data_alloc_len, false);
7510 dio_data->data_space_reserved = true;
7511 else if (ret && !(BTRFS_I(inode)->flags &
7512 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7517 * If this errors out it's because we couldn't invalidate pagecache for
7518 * this range and we need to fallback to buffered IO, or we are doing a
7519 * NOWAIT read/write and we need to block.
7521 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7525 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7532 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7533 * io. INLINE is special, and we could probably kludge it in here, but
7534 * it's still buffered so for safety lets just fall back to the generic
7537 * For COMPRESSED we _have_ to read the entire extent in so we can
7538 * decompress it, so there will be buffering required no matter what we
7539 * do, so go ahead and fallback to buffered.
7541 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7542 * to buffered IO. Don't blame me, this is the price we pay for using
7545 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7546 em->block_start == EXTENT_MAP_INLINE) {
7547 free_extent_map(em);
7549 * If we are in a NOWAIT context, return -EAGAIN in order to
7550 * fallback to buffered IO. This is not only because we can
7551 * block with buffered IO (no support for NOWAIT semantics at
7552 * the moment) but also to avoid returning short reads to user
7553 * space - this happens if we were able to read some data from
7554 * previous non-compressed extents and then when we fallback to
7555 * buffered IO, at btrfs_file_read_iter() by calling
7556 * filemap_read(), we fail to fault in pages for the read buffer,
7557 * in which case filemap_read() returns a short read (the number
7558 * of bytes previously read is > 0, so it does not return -EFAULT).
7560 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7564 len = min(len, em->len - (start - em->start));
7567 * If we have a NOWAIT request and the range contains multiple extents
7568 * (or a mix of extents and holes), then we return -EAGAIN to make the
7569 * caller fallback to a context where it can do a blocking (without
7570 * NOWAIT) request. This way we avoid doing partial IO and returning
7571 * success to the caller, which is not optimal for writes and for reads
7572 * it can result in unexpected behaviour for an application.
7574 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7575 * iomap_dio_rw(), we can end up returning less data then what the caller
7576 * asked for, resulting in an unexpected, and incorrect, short read.
7577 * That is, the caller asked to read N bytes and we return less than that,
7578 * which is wrong unless we are crossing EOF. This happens if we get a
7579 * page fault error when trying to fault in pages for the buffer that is
7580 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7581 * have previously submitted bios for other extents in the range, in
7582 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7583 * those bios have completed by the time we get the page fault error,
7584 * which we return back to our caller - we should only return EIOCBQUEUED
7585 * after we have submitted bios for all the extents in the range.
7587 if ((flags & IOMAP_NOWAIT) && len < length) {
7588 free_extent_map(em);
7594 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7595 start, &len, flags);
7598 unlock_extents = true;
7599 /* Recalc len in case the new em is smaller than requested */
7600 len = min(len, em->len - (start - em->start));
7601 if (dio_data->data_space_reserved) {
7603 u64 release_len = 0;
7605 if (dio_data->nocow_done) {
7606 release_offset = start;
7607 release_len = data_alloc_len;
7608 } else if (len < data_alloc_len) {
7609 release_offset = start + len;
7610 release_len = data_alloc_len - len;
7613 if (release_len > 0)
7614 btrfs_free_reserved_data_space(BTRFS_I(inode),
7615 dio_data->data_reserved,
7621 * We need to unlock only the end area that we aren't using.
7622 * The rest is going to be unlocked by the endio routine.
7624 lockstart = start + len;
7625 if (lockstart < lockend)
7626 unlock_extents = true;
7630 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7633 free_extent_state(cached_state);
7636 * Translate extent map information to iomap.
7637 * We trim the extents (and move the addr) even though iomap code does
7638 * that, since we have locked only the parts we are performing I/O in.
7640 if ((em->block_start == EXTENT_MAP_HOLE) ||
7641 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7642 iomap->addr = IOMAP_NULL_ADDR;
7643 iomap->type = IOMAP_HOLE;
7645 iomap->addr = em->block_start + (start - em->start);
7646 iomap->type = IOMAP_MAPPED;
7648 iomap->offset = start;
7649 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7650 iomap->length = len;
7651 free_extent_map(em);
7656 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7659 if (dio_data->data_space_reserved) {
7660 btrfs_free_reserved_data_space(BTRFS_I(inode),
7661 dio_data->data_reserved,
7662 start, data_alloc_len);
7663 extent_changeset_free(dio_data->data_reserved);
7669 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7670 ssize_t written, unsigned int flags, struct iomap *iomap)
7672 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7673 struct btrfs_dio_data *dio_data = iter->private;
7674 size_t submitted = dio_data->submitted;
7675 const bool write = !!(flags & IOMAP_WRITE);
7678 if (!write && (iomap->type == IOMAP_HOLE)) {
7679 /* If reading from a hole, unlock and return */
7680 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7685 if (submitted < length) {
7687 length -= submitted;
7689 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7690 pos, length, false);
7692 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7693 pos + length - 1, NULL);
7697 btrfs_put_ordered_extent(dio_data->ordered);
7698 dio_data->ordered = NULL;
7702 extent_changeset_free(dio_data->data_reserved);
7706 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7708 struct btrfs_dio_private *dip =
7709 container_of(bbio, struct btrfs_dio_private, bbio);
7710 struct btrfs_inode *inode = bbio->inode;
7711 struct bio *bio = &bbio->bio;
7713 if (bio->bi_status) {
7714 btrfs_warn(inode->root->fs_info,
7715 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7716 btrfs_ino(inode), bio->bi_opf,
7717 dip->file_offset, dip->bytes, bio->bi_status);
7720 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7721 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7722 dip->file_offset, dip->bytes,
7725 unlock_extent(&inode->io_tree, dip->file_offset,
7726 dip->file_offset + dip->bytes - 1, NULL);
7729 bbio->bio.bi_private = bbio->private;
7730 iomap_dio_bio_end_io(bio);
7733 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7736 struct btrfs_bio *bbio = btrfs_bio(bio);
7737 struct btrfs_dio_private *dip =
7738 container_of(bbio, struct btrfs_dio_private, bbio);
7739 struct btrfs_dio_data *dio_data = iter->private;
7741 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7742 btrfs_dio_end_io, bio->bi_private);
7743 bbio->inode = BTRFS_I(iter->inode);
7744 bbio->file_offset = file_offset;
7746 dip->file_offset = file_offset;
7747 dip->bytes = bio->bi_iter.bi_size;
7749 dio_data->submitted += bio->bi_iter.bi_size;
7752 * Check if we are doing a partial write. If we are, we need to split
7753 * the ordered extent to match the submitted bio. Hang on to the
7754 * remaining unfinishable ordered_extent in dio_data so that it can be
7755 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7756 * remaining pages is blocked on the outstanding ordered extent.
7758 if (iter->flags & IOMAP_WRITE) {
7761 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7763 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7764 file_offset, dip->bytes,
7766 bio->bi_status = errno_to_blk_status(ret);
7767 iomap_dio_bio_end_io(bio);
7772 btrfs_submit_bio(bbio, 0);
7775 static const struct iomap_ops btrfs_dio_iomap_ops = {
7776 .iomap_begin = btrfs_dio_iomap_begin,
7777 .iomap_end = btrfs_dio_iomap_end,
7780 static const struct iomap_dio_ops btrfs_dio_ops = {
7781 .submit_io = btrfs_dio_submit_io,
7782 .bio_set = &btrfs_dio_bioset,
7785 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7787 struct btrfs_dio_data data = { 0 };
7789 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7790 IOMAP_DIO_PARTIAL, &data, done_before);
7793 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7796 struct btrfs_dio_data data = { 0 };
7798 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7799 IOMAP_DIO_PARTIAL, &data, done_before);
7802 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7807 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7812 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7813 * file range (0 to LLONG_MAX), but that is not enough if we have
7814 * compression enabled. The first filemap_fdatawrite_range() only kicks
7815 * in the compression of data (in an async thread) and will return
7816 * before the compression is done and writeback is started. A second
7817 * filemap_fdatawrite_range() is needed to wait for the compression to
7818 * complete and writeback to start. We also need to wait for ordered
7819 * extents to complete, because our fiemap implementation uses mainly
7820 * file extent items to list the extents, searching for extent maps
7821 * only for file ranges with holes or prealloc extents to figure out
7822 * if we have delalloc in those ranges.
7824 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7825 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7830 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7833 static int btrfs_writepages(struct address_space *mapping,
7834 struct writeback_control *wbc)
7836 return extent_writepages(mapping, wbc);
7839 static void btrfs_readahead(struct readahead_control *rac)
7841 extent_readahead(rac);
7845 * For release_folio() and invalidate_folio() we have a race window where
7846 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7847 * If we continue to release/invalidate the page, we could cause use-after-free
7848 * for subpage spinlock. So this function is to spin and wait for subpage
7851 static void wait_subpage_spinlock(struct page *page)
7853 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7854 struct btrfs_subpage *subpage;
7856 if (!btrfs_is_subpage(fs_info, page))
7859 ASSERT(PagePrivate(page) && page->private);
7860 subpage = (struct btrfs_subpage *)page->private;
7863 * This may look insane as we just acquire the spinlock and release it,
7864 * without doing anything. But we just want to make sure no one is
7865 * still holding the subpage spinlock.
7866 * And since the page is not dirty nor writeback, and we have page
7867 * locked, the only possible way to hold a spinlock is from the endio
7868 * function to clear page writeback.
7870 * Here we just acquire the spinlock so that all existing callers
7871 * should exit and we're safe to release/invalidate the page.
7873 spin_lock_irq(&subpage->lock);
7874 spin_unlock_irq(&subpage->lock);
7877 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7879 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7882 wait_subpage_spinlock(&folio->page);
7883 clear_page_extent_mapped(&folio->page);
7888 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7890 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7892 return __btrfs_release_folio(folio, gfp_flags);
7895 #ifdef CONFIG_MIGRATION
7896 static int btrfs_migrate_folio(struct address_space *mapping,
7897 struct folio *dst, struct folio *src,
7898 enum migrate_mode mode)
7900 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7902 if (ret != MIGRATEPAGE_SUCCESS)
7905 if (folio_test_ordered(src)) {
7906 folio_clear_ordered(src);
7907 folio_set_ordered(dst);
7910 return MIGRATEPAGE_SUCCESS;
7913 #define btrfs_migrate_folio NULL
7916 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7919 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7920 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7921 struct extent_io_tree *tree = &inode->io_tree;
7922 struct extent_state *cached_state = NULL;
7923 u64 page_start = folio_pos(folio);
7924 u64 page_end = page_start + folio_size(folio) - 1;
7926 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7929 * We have folio locked so no new ordered extent can be created on this
7930 * page, nor bio can be submitted for this folio.
7932 * But already submitted bio can still be finished on this folio.
7933 * Furthermore, endio function won't skip folio which has Ordered
7934 * (Private2) already cleared, so it's possible for endio and
7935 * invalidate_folio to do the same ordered extent accounting twice
7938 * So here we wait for any submitted bios to finish, so that we won't
7939 * do double ordered extent accounting on the same folio.
7941 folio_wait_writeback(folio);
7942 wait_subpage_spinlock(&folio->page);
7945 * For subpage case, we have call sites like
7946 * btrfs_punch_hole_lock_range() which passes range not aligned to
7948 * If the range doesn't cover the full folio, we don't need to and
7949 * shouldn't clear page extent mapped, as folio->private can still
7950 * record subpage dirty bits for other part of the range.
7952 * For cases that invalidate the full folio even the range doesn't
7953 * cover the full folio, like invalidating the last folio, we're
7954 * still safe to wait for ordered extent to finish.
7956 if (!(offset == 0 && length == folio_size(folio))) {
7957 btrfs_release_folio(folio, GFP_NOFS);
7961 if (!inode_evicting)
7962 lock_extent(tree, page_start, page_end, &cached_state);
7965 while (cur < page_end) {
7966 struct btrfs_ordered_extent *ordered;
7969 u32 extra_flags = 0;
7971 ordered = btrfs_lookup_first_ordered_range(inode, cur,
7972 page_end + 1 - cur);
7974 range_end = page_end;
7976 * No ordered extent covering this range, we are safe
7977 * to delete all extent states in the range.
7979 extra_flags = EXTENT_CLEAR_ALL_BITS;
7982 if (ordered->file_offset > cur) {
7984 * There is a range between [cur, oe->file_offset) not
7985 * covered by any ordered extent.
7986 * We are safe to delete all extent states, and handle
7987 * the ordered extent in the next iteration.
7989 range_end = ordered->file_offset - 1;
7990 extra_flags = EXTENT_CLEAR_ALL_BITS;
7994 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
7996 ASSERT(range_end + 1 - cur < U32_MAX);
7997 range_len = range_end + 1 - cur;
7998 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8000 * If Ordered (Private2) is cleared, it means endio has
8001 * already been executed for the range.
8002 * We can't delete the extent states as
8003 * btrfs_finish_ordered_io() may still use some of them.
8007 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8010 * IO on this page will never be started, so we need to account
8011 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8012 * here, must leave that up for the ordered extent completion.
8014 * This will also unlock the range for incoming
8015 * btrfs_finish_ordered_io().
8017 if (!inode_evicting)
8018 clear_extent_bit(tree, cur, range_end,
8020 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8021 EXTENT_DEFRAG, &cached_state);
8023 spin_lock_irq(&inode->ordered_tree_lock);
8024 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8025 ordered->truncated_len = min(ordered->truncated_len,
8026 cur - ordered->file_offset);
8027 spin_unlock_irq(&inode->ordered_tree_lock);
8030 * If the ordered extent has finished, we're safe to delete all
8031 * the extent states of the range, otherwise
8032 * btrfs_finish_ordered_io() will get executed by endio for
8033 * other pages, so we can't delete extent states.
8035 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8036 cur, range_end + 1 - cur)) {
8037 btrfs_finish_ordered_io(ordered);
8039 * The ordered extent has finished, now we're again
8040 * safe to delete all extent states of the range.
8042 extra_flags = EXTENT_CLEAR_ALL_BITS;
8046 btrfs_put_ordered_extent(ordered);
8048 * Qgroup reserved space handler
8049 * Sector(s) here will be either:
8051 * 1) Already written to disk or bio already finished
8052 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8053 * Qgroup will be handled by its qgroup_record then.
8054 * btrfs_qgroup_free_data() call will do nothing here.
8056 * 2) Not written to disk yet
8057 * Then btrfs_qgroup_free_data() call will clear the
8058 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8059 * reserved data space.
8060 * Since the IO will never happen for this page.
8062 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur, NULL);
8063 if (!inode_evicting) {
8064 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8065 EXTENT_DELALLOC | EXTENT_UPTODATE |
8066 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8067 extra_flags, &cached_state);
8069 cur = range_end + 1;
8072 * We have iterated through all ordered extents of the page, the page
8073 * should not have Ordered (Private2) anymore, or the above iteration
8074 * did something wrong.
8076 ASSERT(!folio_test_ordered(folio));
8077 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8078 if (!inode_evicting)
8079 __btrfs_release_folio(folio, GFP_NOFS);
8080 clear_page_extent_mapped(&folio->page);
8084 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8085 * called from a page fault handler when a page is first dirtied. Hence we must
8086 * be careful to check for EOF conditions here. We set the page up correctly
8087 * for a written page which means we get ENOSPC checking when writing into
8088 * holes and correct delalloc and unwritten extent mapping on filesystems that
8089 * support these features.
8091 * We are not allowed to take the i_mutex here so we have to play games to
8092 * protect against truncate races as the page could now be beyond EOF. Because
8093 * truncate_setsize() writes the inode size before removing pages, once we have
8094 * the page lock we can determine safely if the page is beyond EOF. If it is not
8095 * beyond EOF, then the page is guaranteed safe against truncation until we
8098 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8100 struct page *page = vmf->page;
8101 struct inode *inode = file_inode(vmf->vma->vm_file);
8102 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8103 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8104 struct btrfs_ordered_extent *ordered;
8105 struct extent_state *cached_state = NULL;
8106 struct extent_changeset *data_reserved = NULL;
8107 unsigned long zero_start;
8117 reserved_space = PAGE_SIZE;
8119 sb_start_pagefault(inode->i_sb);
8120 page_start = page_offset(page);
8121 page_end = page_start + PAGE_SIZE - 1;
8125 * Reserving delalloc space after obtaining the page lock can lead to
8126 * deadlock. For example, if a dirty page is locked by this function
8127 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8128 * dirty page write out, then the btrfs_writepages() function could
8129 * end up waiting indefinitely to get a lock on the page currently
8130 * being processed by btrfs_page_mkwrite() function.
8132 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8133 page_start, reserved_space);
8135 ret2 = file_update_time(vmf->vma->vm_file);
8139 ret = vmf_error(ret2);
8145 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8147 down_read(&BTRFS_I(inode)->i_mmap_lock);
8149 size = i_size_read(inode);
8151 if ((page->mapping != inode->i_mapping) ||
8152 (page_start >= size)) {
8153 /* page got truncated out from underneath us */
8156 wait_on_page_writeback(page);
8158 lock_extent(io_tree, page_start, page_end, &cached_state);
8159 ret2 = set_page_extent_mapped(page);
8161 ret = vmf_error(ret2);
8162 unlock_extent(io_tree, page_start, page_end, &cached_state);
8167 * we can't set the delalloc bits if there are pending ordered
8168 * extents. Drop our locks and wait for them to finish
8170 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8173 unlock_extent(io_tree, page_start, page_end, &cached_state);
8175 up_read(&BTRFS_I(inode)->i_mmap_lock);
8176 btrfs_start_ordered_extent(ordered);
8177 btrfs_put_ordered_extent(ordered);
8181 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8182 reserved_space = round_up(size - page_start,
8183 fs_info->sectorsize);
8184 if (reserved_space < PAGE_SIZE) {
8185 end = page_start + reserved_space - 1;
8186 btrfs_delalloc_release_space(BTRFS_I(inode),
8187 data_reserved, page_start,
8188 PAGE_SIZE - reserved_space, true);
8193 * page_mkwrite gets called when the page is firstly dirtied after it's
8194 * faulted in, but write(2) could also dirty a page and set delalloc
8195 * bits, thus in this case for space account reason, we still need to
8196 * clear any delalloc bits within this page range since we have to
8197 * reserve data&meta space before lock_page() (see above comments).
8199 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8200 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8201 EXTENT_DEFRAG, &cached_state);
8203 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8206 unlock_extent(io_tree, page_start, page_end, &cached_state);
8207 ret = VM_FAULT_SIGBUS;
8211 /* page is wholly or partially inside EOF */
8212 if (page_start + PAGE_SIZE > size)
8213 zero_start = offset_in_page(size);
8215 zero_start = PAGE_SIZE;
8217 if (zero_start != PAGE_SIZE)
8218 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8220 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8221 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8222 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8224 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8226 unlock_extent(io_tree, page_start, page_end, &cached_state);
8227 up_read(&BTRFS_I(inode)->i_mmap_lock);
8229 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8230 sb_end_pagefault(inode->i_sb);
8231 extent_changeset_free(data_reserved);
8232 return VM_FAULT_LOCKED;
8236 up_read(&BTRFS_I(inode)->i_mmap_lock);
8238 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8239 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8240 reserved_space, (ret != 0));
8242 sb_end_pagefault(inode->i_sb);
8243 extent_changeset_free(data_reserved);
8247 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8249 struct btrfs_truncate_control control = {
8251 .ino = btrfs_ino(inode),
8252 .min_type = BTRFS_EXTENT_DATA_KEY,
8253 .clear_extent_range = true,
8255 struct btrfs_root *root = inode->root;
8256 struct btrfs_fs_info *fs_info = root->fs_info;
8257 struct btrfs_block_rsv *rsv;
8259 struct btrfs_trans_handle *trans;
8260 u64 mask = fs_info->sectorsize - 1;
8261 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8263 if (!skip_writeback) {
8264 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8265 inode->vfs_inode.i_size & (~mask),
8272 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8273 * things going on here:
8275 * 1) We need to reserve space to update our inode.
8277 * 2) We need to have something to cache all the space that is going to
8278 * be free'd up by the truncate operation, but also have some slack
8279 * space reserved in case it uses space during the truncate (thank you
8280 * very much snapshotting).
8282 * And we need these to be separate. The fact is we can use a lot of
8283 * space doing the truncate, and we have no earthly idea how much space
8284 * we will use, so we need the truncate reservation to be separate so it
8285 * doesn't end up using space reserved for updating the inode. We also
8286 * need to be able to stop the transaction and start a new one, which
8287 * means we need to be able to update the inode several times, and we
8288 * have no idea of knowing how many times that will be, so we can't just
8289 * reserve 1 item for the entirety of the operation, so that has to be
8290 * done separately as well.
8292 * So that leaves us with
8294 * 1) rsv - for the truncate reservation, which we will steal from the
8295 * transaction reservation.
8296 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8297 * updating the inode.
8299 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8302 rsv->size = min_size;
8303 rsv->failfast = true;
8306 * 1 for the truncate slack space
8307 * 1 for updating the inode.
8309 trans = btrfs_start_transaction(root, 2);
8310 if (IS_ERR(trans)) {
8311 ret = PTR_ERR(trans);
8315 /* Migrate the slack space for the truncate to our reserve */
8316 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8319 * We have reserved 2 metadata units when we started the transaction and
8320 * min_size matches 1 unit, so this should never fail, but if it does,
8321 * it's not critical we just fail truncation.
8324 btrfs_end_transaction(trans);
8328 trans->block_rsv = rsv;
8331 struct extent_state *cached_state = NULL;
8332 const u64 new_size = inode->vfs_inode.i_size;
8333 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8335 control.new_size = new_size;
8336 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8338 * We want to drop from the next block forward in case this new
8339 * size is not block aligned since we will be keeping the last
8340 * block of the extent just the way it is.
8342 btrfs_drop_extent_map_range(inode,
8343 ALIGN(new_size, fs_info->sectorsize),
8346 ret = btrfs_truncate_inode_items(trans, root, &control);
8348 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8349 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8351 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8353 trans->block_rsv = &fs_info->trans_block_rsv;
8354 if (ret != -ENOSPC && ret != -EAGAIN)
8357 ret = btrfs_update_inode(trans, inode);
8361 btrfs_end_transaction(trans);
8362 btrfs_btree_balance_dirty(fs_info);
8364 trans = btrfs_start_transaction(root, 2);
8365 if (IS_ERR(trans)) {
8366 ret = PTR_ERR(trans);
8371 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8372 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8373 rsv, min_size, false);
8375 * We have reserved 2 metadata units when we started the
8376 * transaction and min_size matches 1 unit, so this should never
8377 * fail, but if it does, it's not critical we just fail truncation.
8382 trans->block_rsv = rsv;
8386 * We can't call btrfs_truncate_block inside a trans handle as we could
8387 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8388 * know we've truncated everything except the last little bit, and can
8389 * do btrfs_truncate_block and then update the disk_i_size.
8391 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8392 btrfs_end_transaction(trans);
8393 btrfs_btree_balance_dirty(fs_info);
8395 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8398 trans = btrfs_start_transaction(root, 1);
8399 if (IS_ERR(trans)) {
8400 ret = PTR_ERR(trans);
8403 btrfs_inode_safe_disk_i_size_write(inode, 0);
8409 trans->block_rsv = &fs_info->trans_block_rsv;
8410 ret2 = btrfs_update_inode(trans, inode);
8414 ret2 = btrfs_end_transaction(trans);
8417 btrfs_btree_balance_dirty(fs_info);
8420 btrfs_free_block_rsv(fs_info, rsv);
8422 * So if we truncate and then write and fsync we normally would just
8423 * write the extents that changed, which is a problem if we need to
8424 * first truncate that entire inode. So set this flag so we write out
8425 * all of the extents in the inode to the sync log so we're completely
8428 * If no extents were dropped or trimmed we don't need to force the next
8429 * fsync to truncate all the inode's items from the log and re-log them
8430 * all. This means the truncate operation did not change the file size,
8431 * or changed it to a smaller size but there was only an implicit hole
8432 * between the old i_size and the new i_size, and there were no prealloc
8433 * extents beyond i_size to drop.
8435 if (control.extents_found > 0)
8436 btrfs_set_inode_full_sync(inode);
8441 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8444 struct inode *inode;
8446 inode = new_inode(dir->i_sb);
8449 * Subvolumes don't inherit the sgid bit or the parent's gid if
8450 * the parent's sgid bit is set. This is probably a bug.
8452 inode_init_owner(idmap, inode, NULL,
8453 S_IFDIR | (~current_umask() & S_IRWXUGO));
8454 inode->i_op = &btrfs_dir_inode_operations;
8455 inode->i_fop = &btrfs_dir_file_operations;
8460 struct inode *btrfs_alloc_inode(struct super_block *sb)
8462 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8463 struct btrfs_inode *ei;
8464 struct inode *inode;
8466 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8473 ei->last_sub_trans = 0;
8474 ei->logged_trans = 0;
8475 ei->delalloc_bytes = 0;
8476 ei->new_delalloc_bytes = 0;
8477 ei->defrag_bytes = 0;
8478 ei->disk_i_size = 0;
8482 ei->index_cnt = (u64)-1;
8484 ei->last_unlink_trans = 0;
8485 ei->last_reflink_trans = 0;
8486 ei->last_log_commit = 0;
8488 spin_lock_init(&ei->lock);
8489 ei->outstanding_extents = 0;
8490 if (sb->s_magic != BTRFS_TEST_MAGIC)
8491 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8492 BTRFS_BLOCK_RSV_DELALLOC);
8493 ei->runtime_flags = 0;
8494 ei->prop_compress = BTRFS_COMPRESS_NONE;
8495 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8497 ei->delayed_node = NULL;
8499 ei->i_otime_sec = 0;
8500 ei->i_otime_nsec = 0;
8502 inode = &ei->vfs_inode;
8503 extent_map_tree_init(&ei->extent_tree);
8504 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8505 ei->io_tree.inode = ei;
8506 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8507 IO_TREE_INODE_FILE_EXTENT);
8508 mutex_init(&ei->log_mutex);
8509 spin_lock_init(&ei->ordered_tree_lock);
8510 ei->ordered_tree = RB_ROOT;
8511 ei->ordered_tree_last = NULL;
8512 INIT_LIST_HEAD(&ei->delalloc_inodes);
8513 INIT_LIST_HEAD(&ei->delayed_iput);
8514 RB_CLEAR_NODE(&ei->rb_node);
8515 init_rwsem(&ei->i_mmap_lock);
8520 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8521 void btrfs_test_destroy_inode(struct inode *inode)
8523 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8524 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8528 void btrfs_free_inode(struct inode *inode)
8530 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8533 void btrfs_destroy_inode(struct inode *vfs_inode)
8535 struct btrfs_ordered_extent *ordered;
8536 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8537 struct btrfs_root *root = inode->root;
8538 bool freespace_inode;
8540 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8541 WARN_ON(vfs_inode->i_data.nrpages);
8542 WARN_ON(inode->block_rsv.reserved);
8543 WARN_ON(inode->block_rsv.size);
8544 WARN_ON(inode->outstanding_extents);
8545 if (!S_ISDIR(vfs_inode->i_mode)) {
8546 WARN_ON(inode->delalloc_bytes);
8547 WARN_ON(inode->new_delalloc_bytes);
8549 WARN_ON(inode->csum_bytes);
8550 WARN_ON(inode->defrag_bytes);
8553 * This can happen where we create an inode, but somebody else also
8554 * created the same inode and we need to destroy the one we already
8561 * If this is a free space inode do not take the ordered extents lockdep
8564 freespace_inode = btrfs_is_free_space_inode(inode);
8567 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8571 btrfs_err(root->fs_info,
8572 "found ordered extent %llu %llu on inode cleanup",
8573 ordered->file_offset, ordered->num_bytes);
8575 if (!freespace_inode)
8576 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8578 btrfs_remove_ordered_extent(inode, ordered);
8579 btrfs_put_ordered_extent(ordered);
8580 btrfs_put_ordered_extent(ordered);
8583 btrfs_qgroup_check_reserved_leak(inode);
8584 inode_tree_del(inode);
8585 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8586 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8587 btrfs_put_root(inode->root);
8590 int btrfs_drop_inode(struct inode *inode)
8592 struct btrfs_root *root = BTRFS_I(inode)->root;
8597 /* the snap/subvol tree is on deleting */
8598 if (btrfs_root_refs(&root->root_item) == 0)
8601 return generic_drop_inode(inode);
8604 static void init_once(void *foo)
8606 struct btrfs_inode *ei = foo;
8608 inode_init_once(&ei->vfs_inode);
8611 void __cold btrfs_destroy_cachep(void)
8614 * Make sure all delayed rcu free inodes are flushed before we
8618 bioset_exit(&btrfs_dio_bioset);
8619 kmem_cache_destroy(btrfs_inode_cachep);
8622 int __init btrfs_init_cachep(void)
8624 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8625 sizeof(struct btrfs_inode), 0,
8626 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8628 if (!btrfs_inode_cachep)
8631 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8632 offsetof(struct btrfs_dio_private, bbio.bio),
8638 btrfs_destroy_cachep();
8642 static int btrfs_getattr(struct mnt_idmap *idmap,
8643 const struct path *path, struct kstat *stat,
8644 u32 request_mask, unsigned int flags)
8648 struct inode *inode = d_inode(path->dentry);
8649 u32 blocksize = inode->i_sb->s_blocksize;
8650 u32 bi_flags = BTRFS_I(inode)->flags;
8651 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8653 stat->result_mask |= STATX_BTIME;
8654 stat->btime.tv_sec = BTRFS_I(inode)->i_otime_sec;
8655 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime_nsec;
8656 if (bi_flags & BTRFS_INODE_APPEND)
8657 stat->attributes |= STATX_ATTR_APPEND;
8658 if (bi_flags & BTRFS_INODE_COMPRESS)
8659 stat->attributes |= STATX_ATTR_COMPRESSED;
8660 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8661 stat->attributes |= STATX_ATTR_IMMUTABLE;
8662 if (bi_flags & BTRFS_INODE_NODUMP)
8663 stat->attributes |= STATX_ATTR_NODUMP;
8664 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8665 stat->attributes |= STATX_ATTR_VERITY;
8667 stat->attributes_mask |= (STATX_ATTR_APPEND |
8668 STATX_ATTR_COMPRESSED |
8669 STATX_ATTR_IMMUTABLE |
8672 generic_fillattr(idmap, request_mask, inode, stat);
8673 stat->dev = BTRFS_I(inode)->root->anon_dev;
8675 spin_lock(&BTRFS_I(inode)->lock);
8676 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8677 inode_bytes = inode_get_bytes(inode);
8678 spin_unlock(&BTRFS_I(inode)->lock);
8679 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8680 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8684 static int btrfs_rename_exchange(struct inode *old_dir,
8685 struct dentry *old_dentry,
8686 struct inode *new_dir,
8687 struct dentry *new_dentry)
8689 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8690 struct btrfs_trans_handle *trans;
8691 unsigned int trans_num_items;
8692 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8693 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8694 struct inode *new_inode = new_dentry->d_inode;
8695 struct inode *old_inode = old_dentry->d_inode;
8696 struct btrfs_rename_ctx old_rename_ctx;
8697 struct btrfs_rename_ctx new_rename_ctx;
8698 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8699 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8704 bool need_abort = false;
8705 struct fscrypt_name old_fname, new_fname;
8706 struct fscrypt_str *old_name, *new_name;
8709 * For non-subvolumes allow exchange only within one subvolume, in the
8710 * same inode namespace. Two subvolumes (represented as directory) can
8711 * be exchanged as they're a logical link and have a fixed inode number.
8714 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8715 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8718 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8722 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8724 fscrypt_free_filename(&old_fname);
8728 old_name = &old_fname.disk_name;
8729 new_name = &new_fname.disk_name;
8731 /* close the race window with snapshot create/destroy ioctl */
8732 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8733 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8734 down_read(&fs_info->subvol_sem);
8738 * 1 to remove old dir item
8739 * 1 to remove old dir index
8740 * 1 to add new dir item
8741 * 1 to add new dir index
8742 * 1 to update parent inode
8744 * If the parents are the same, we only need to account for one
8746 trans_num_items = (old_dir == new_dir ? 9 : 10);
8747 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8749 * 1 to remove old root ref
8750 * 1 to remove old root backref
8751 * 1 to add new root ref
8752 * 1 to add new root backref
8754 trans_num_items += 4;
8757 * 1 to update inode item
8758 * 1 to remove old inode ref
8759 * 1 to add new inode ref
8761 trans_num_items += 3;
8763 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8764 trans_num_items += 4;
8766 trans_num_items += 3;
8767 trans = btrfs_start_transaction(root, trans_num_items);
8768 if (IS_ERR(trans)) {
8769 ret = PTR_ERR(trans);
8774 ret = btrfs_record_root_in_trans(trans, dest);
8780 * We need to find a free sequence number both in the source and
8781 * in the destination directory for the exchange.
8783 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8786 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8790 BTRFS_I(old_inode)->dir_index = 0ULL;
8791 BTRFS_I(new_inode)->dir_index = 0ULL;
8793 /* Reference for the source. */
8794 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8795 /* force full log commit if subvolume involved. */
8796 btrfs_set_log_full_commit(trans);
8798 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8799 btrfs_ino(BTRFS_I(new_dir)),
8806 /* And now for the dest. */
8807 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8808 /* force full log commit if subvolume involved. */
8809 btrfs_set_log_full_commit(trans);
8811 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8812 btrfs_ino(BTRFS_I(old_dir)),
8816 btrfs_abort_transaction(trans, ret);
8821 /* Update inode version and ctime/mtime. */
8822 inode_inc_iversion(old_dir);
8823 inode_inc_iversion(new_dir);
8824 inode_inc_iversion(old_inode);
8825 inode_inc_iversion(new_inode);
8826 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8828 if (old_dentry->d_parent != new_dentry->d_parent) {
8829 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8830 BTRFS_I(old_inode), true);
8831 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8832 BTRFS_I(new_inode), true);
8835 /* src is a subvolume */
8836 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8837 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8838 } else { /* src is an inode */
8839 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8840 BTRFS_I(old_dentry->d_inode),
8841 old_name, &old_rename_ctx);
8843 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
8846 btrfs_abort_transaction(trans, ret);
8850 /* dest is a subvolume */
8851 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8852 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8853 } else { /* dest is an inode */
8854 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8855 BTRFS_I(new_dentry->d_inode),
8856 new_name, &new_rename_ctx);
8858 ret = btrfs_update_inode(trans, BTRFS_I(new_inode));
8861 btrfs_abort_transaction(trans, ret);
8865 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8866 new_name, 0, old_idx);
8868 btrfs_abort_transaction(trans, ret);
8872 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8873 old_name, 0, new_idx);
8875 btrfs_abort_transaction(trans, ret);
8879 if (old_inode->i_nlink == 1)
8880 BTRFS_I(old_inode)->dir_index = old_idx;
8881 if (new_inode->i_nlink == 1)
8882 BTRFS_I(new_inode)->dir_index = new_idx;
8885 * Now pin the logs of the roots. We do it to ensure that no other task
8886 * can sync the logs while we are in progress with the rename, because
8887 * that could result in an inconsistency in case any of the inodes that
8888 * are part of this rename operation were logged before.
8890 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8891 btrfs_pin_log_trans(root);
8892 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8893 btrfs_pin_log_trans(dest);
8895 /* Do the log updates for all inodes. */
8896 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8897 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8898 old_rename_ctx.index, new_dentry->d_parent);
8899 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8900 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8901 new_rename_ctx.index, old_dentry->d_parent);
8903 /* Now unpin the logs. */
8904 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8905 btrfs_end_log_trans(root);
8906 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8907 btrfs_end_log_trans(dest);
8909 ret2 = btrfs_end_transaction(trans);
8910 ret = ret ? ret : ret2;
8912 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8913 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8914 up_read(&fs_info->subvol_sem);
8916 fscrypt_free_filename(&new_fname);
8917 fscrypt_free_filename(&old_fname);
8921 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8924 struct inode *inode;
8926 inode = new_inode(dir->i_sb);
8928 inode_init_owner(idmap, inode, dir,
8929 S_IFCHR | WHITEOUT_MODE);
8930 inode->i_op = &btrfs_special_inode_operations;
8931 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8936 static int btrfs_rename(struct mnt_idmap *idmap,
8937 struct inode *old_dir, struct dentry *old_dentry,
8938 struct inode *new_dir, struct dentry *new_dentry,
8941 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8942 struct btrfs_new_inode_args whiteout_args = {
8944 .dentry = old_dentry,
8946 struct btrfs_trans_handle *trans;
8947 unsigned int trans_num_items;
8948 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8949 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8950 struct inode *new_inode = d_inode(new_dentry);
8951 struct inode *old_inode = d_inode(old_dentry);
8952 struct btrfs_rename_ctx rename_ctx;
8956 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8957 struct fscrypt_name old_fname, new_fname;
8959 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8962 /* we only allow rename subvolume link between subvolumes */
8963 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8966 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8967 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
8970 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8971 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8974 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8978 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8980 fscrypt_free_filename(&old_fname);
8984 /* check for collisions, even if the name isn't there */
8985 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
8987 if (ret == -EEXIST) {
8989 * eexist without a new_inode */
8990 if (WARN_ON(!new_inode)) {
8991 goto out_fscrypt_names;
8994 /* maybe -EOVERFLOW */
8995 goto out_fscrypt_names;
9001 * we're using rename to replace one file with another. Start IO on it
9002 * now so we don't add too much work to the end of the transaction
9004 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9005 filemap_flush(old_inode->i_mapping);
9007 if (flags & RENAME_WHITEOUT) {
9008 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9009 if (!whiteout_args.inode) {
9011 goto out_fscrypt_names;
9013 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9015 goto out_whiteout_inode;
9017 /* 1 to update the old parent inode. */
9018 trans_num_items = 1;
9021 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9022 /* Close the race window with snapshot create/destroy ioctl */
9023 down_read(&fs_info->subvol_sem);
9025 * 1 to remove old root ref
9026 * 1 to remove old root backref
9027 * 1 to add new root ref
9028 * 1 to add new root backref
9030 trans_num_items += 4;
9034 * 1 to remove old inode ref
9035 * 1 to add new inode ref
9037 trans_num_items += 3;
9040 * 1 to remove old dir item
9041 * 1 to remove old dir index
9042 * 1 to add new dir item
9043 * 1 to add new dir index
9045 trans_num_items += 4;
9046 /* 1 to update new parent inode if it's not the same as the old parent */
9047 if (new_dir != old_dir)
9052 * 1 to remove inode ref
9053 * 1 to remove dir item
9054 * 1 to remove dir index
9055 * 1 to possibly add orphan item
9057 trans_num_items += 5;
9059 trans = btrfs_start_transaction(root, trans_num_items);
9060 if (IS_ERR(trans)) {
9061 ret = PTR_ERR(trans);
9066 ret = btrfs_record_root_in_trans(trans, dest);
9071 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9075 BTRFS_I(old_inode)->dir_index = 0ULL;
9076 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9077 /* force full log commit if subvolume involved. */
9078 btrfs_set_log_full_commit(trans);
9080 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9081 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9087 inode_inc_iversion(old_dir);
9088 inode_inc_iversion(new_dir);
9089 inode_inc_iversion(old_inode);
9090 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9092 if (old_dentry->d_parent != new_dentry->d_parent)
9093 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9094 BTRFS_I(old_inode), true);
9096 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9097 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9099 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9100 BTRFS_I(d_inode(old_dentry)),
9101 &old_fname.disk_name, &rename_ctx);
9103 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
9106 btrfs_abort_transaction(trans, ret);
9111 inode_inc_iversion(new_inode);
9112 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9113 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9114 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9115 BUG_ON(new_inode->i_nlink == 0);
9117 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9118 BTRFS_I(d_inode(new_dentry)),
9119 &new_fname.disk_name);
9121 if (!ret && new_inode->i_nlink == 0)
9122 ret = btrfs_orphan_add(trans,
9123 BTRFS_I(d_inode(new_dentry)));
9125 btrfs_abort_transaction(trans, ret);
9130 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9131 &new_fname.disk_name, 0, index);
9133 btrfs_abort_transaction(trans, ret);
9137 if (old_inode->i_nlink == 1)
9138 BTRFS_I(old_inode)->dir_index = index;
9140 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9141 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9142 rename_ctx.index, new_dentry->d_parent);
9144 if (flags & RENAME_WHITEOUT) {
9145 ret = btrfs_create_new_inode(trans, &whiteout_args);
9147 btrfs_abort_transaction(trans, ret);
9150 unlock_new_inode(whiteout_args.inode);
9151 iput(whiteout_args.inode);
9152 whiteout_args.inode = NULL;
9156 ret2 = btrfs_end_transaction(trans);
9157 ret = ret ? ret : ret2;
9159 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9160 up_read(&fs_info->subvol_sem);
9161 if (flags & RENAME_WHITEOUT)
9162 btrfs_new_inode_args_destroy(&whiteout_args);
9164 if (flags & RENAME_WHITEOUT)
9165 iput(whiteout_args.inode);
9167 fscrypt_free_filename(&old_fname);
9168 fscrypt_free_filename(&new_fname);
9172 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9173 struct dentry *old_dentry, struct inode *new_dir,
9174 struct dentry *new_dentry, unsigned int flags)
9178 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9181 if (flags & RENAME_EXCHANGE)
9182 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9185 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9188 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9193 struct btrfs_delalloc_work {
9194 struct inode *inode;
9195 struct completion completion;
9196 struct list_head list;
9197 struct btrfs_work work;
9200 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9202 struct btrfs_delalloc_work *delalloc_work;
9203 struct inode *inode;
9205 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9207 inode = delalloc_work->inode;
9208 filemap_flush(inode->i_mapping);
9209 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9210 &BTRFS_I(inode)->runtime_flags))
9211 filemap_flush(inode->i_mapping);
9214 complete(&delalloc_work->completion);
9217 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9219 struct btrfs_delalloc_work *work;
9221 work = kmalloc(sizeof(*work), GFP_NOFS);
9225 init_completion(&work->completion);
9226 INIT_LIST_HEAD(&work->list);
9227 work->inode = inode;
9228 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL);
9234 * some fairly slow code that needs optimization. This walks the list
9235 * of all the inodes with pending delalloc and forces them to disk.
9237 static int start_delalloc_inodes(struct btrfs_root *root,
9238 struct writeback_control *wbc, bool snapshot,
9239 bool in_reclaim_context)
9241 struct btrfs_inode *binode;
9242 struct inode *inode;
9243 struct btrfs_delalloc_work *work, *next;
9247 bool full_flush = wbc->nr_to_write == LONG_MAX;
9249 mutex_lock(&root->delalloc_mutex);
9250 spin_lock(&root->delalloc_lock);
9251 list_splice_init(&root->delalloc_inodes, &splice);
9252 while (!list_empty(&splice)) {
9253 binode = list_entry(splice.next, struct btrfs_inode,
9256 list_move_tail(&binode->delalloc_inodes,
9257 &root->delalloc_inodes);
9259 if (in_reclaim_context &&
9260 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9263 inode = igrab(&binode->vfs_inode);
9265 cond_resched_lock(&root->delalloc_lock);
9268 spin_unlock(&root->delalloc_lock);
9271 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9272 &binode->runtime_flags);
9274 work = btrfs_alloc_delalloc_work(inode);
9280 list_add_tail(&work->list, &works);
9281 btrfs_queue_work(root->fs_info->flush_workers,
9284 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9285 btrfs_add_delayed_iput(BTRFS_I(inode));
9286 if (ret || wbc->nr_to_write <= 0)
9290 spin_lock(&root->delalloc_lock);
9292 spin_unlock(&root->delalloc_lock);
9295 list_for_each_entry_safe(work, next, &works, list) {
9296 list_del_init(&work->list);
9297 wait_for_completion(&work->completion);
9301 if (!list_empty(&splice)) {
9302 spin_lock(&root->delalloc_lock);
9303 list_splice_tail(&splice, &root->delalloc_inodes);
9304 spin_unlock(&root->delalloc_lock);
9306 mutex_unlock(&root->delalloc_mutex);
9310 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9312 struct writeback_control wbc = {
9313 .nr_to_write = LONG_MAX,
9314 .sync_mode = WB_SYNC_NONE,
9316 .range_end = LLONG_MAX,
9318 struct btrfs_fs_info *fs_info = root->fs_info;
9320 if (BTRFS_FS_ERROR(fs_info))
9323 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9326 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9327 bool in_reclaim_context)
9329 struct writeback_control wbc = {
9331 .sync_mode = WB_SYNC_NONE,
9333 .range_end = LLONG_MAX,
9335 struct btrfs_root *root;
9339 if (BTRFS_FS_ERROR(fs_info))
9342 mutex_lock(&fs_info->delalloc_root_mutex);
9343 spin_lock(&fs_info->delalloc_root_lock);
9344 list_splice_init(&fs_info->delalloc_roots, &splice);
9345 while (!list_empty(&splice)) {
9347 * Reset nr_to_write here so we know that we're doing a full
9351 wbc.nr_to_write = LONG_MAX;
9353 root = list_first_entry(&splice, struct btrfs_root,
9355 root = btrfs_grab_root(root);
9357 list_move_tail(&root->delalloc_root,
9358 &fs_info->delalloc_roots);
9359 spin_unlock(&fs_info->delalloc_root_lock);
9361 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9362 btrfs_put_root(root);
9363 if (ret < 0 || wbc.nr_to_write <= 0)
9365 spin_lock(&fs_info->delalloc_root_lock);
9367 spin_unlock(&fs_info->delalloc_root_lock);
9371 if (!list_empty(&splice)) {
9372 spin_lock(&fs_info->delalloc_root_lock);
9373 list_splice_tail(&splice, &fs_info->delalloc_roots);
9374 spin_unlock(&fs_info->delalloc_root_lock);
9376 mutex_unlock(&fs_info->delalloc_root_mutex);
9380 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9381 struct dentry *dentry, const char *symname)
9383 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9384 struct btrfs_trans_handle *trans;
9385 struct btrfs_root *root = BTRFS_I(dir)->root;
9386 struct btrfs_path *path;
9387 struct btrfs_key key;
9388 struct inode *inode;
9389 struct btrfs_new_inode_args new_inode_args = {
9393 unsigned int trans_num_items;
9398 struct btrfs_file_extent_item *ei;
9399 struct extent_buffer *leaf;
9401 name_len = strlen(symname);
9402 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9403 return -ENAMETOOLONG;
9405 inode = new_inode(dir->i_sb);
9408 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9409 inode->i_op = &btrfs_symlink_inode_operations;
9410 inode_nohighmem(inode);
9411 inode->i_mapping->a_ops = &btrfs_aops;
9412 btrfs_i_size_write(BTRFS_I(inode), name_len);
9413 inode_set_bytes(inode, name_len);
9415 new_inode_args.inode = inode;
9416 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9419 /* 1 additional item for the inline extent */
9422 trans = btrfs_start_transaction(root, trans_num_items);
9423 if (IS_ERR(trans)) {
9424 err = PTR_ERR(trans);
9425 goto out_new_inode_args;
9428 err = btrfs_create_new_inode(trans, &new_inode_args);
9432 path = btrfs_alloc_path();
9435 btrfs_abort_transaction(trans, err);
9436 discard_new_inode(inode);
9440 key.objectid = btrfs_ino(BTRFS_I(inode));
9442 key.type = BTRFS_EXTENT_DATA_KEY;
9443 datasize = btrfs_file_extent_calc_inline_size(name_len);
9444 err = btrfs_insert_empty_item(trans, root, path, &key,
9447 btrfs_abort_transaction(trans, err);
9448 btrfs_free_path(path);
9449 discard_new_inode(inode);
9453 leaf = path->nodes[0];
9454 ei = btrfs_item_ptr(leaf, path->slots[0],
9455 struct btrfs_file_extent_item);
9456 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9457 btrfs_set_file_extent_type(leaf, ei,
9458 BTRFS_FILE_EXTENT_INLINE);
9459 btrfs_set_file_extent_encryption(leaf, ei, 0);
9460 btrfs_set_file_extent_compression(leaf, ei, 0);
9461 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9462 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9464 ptr = btrfs_file_extent_inline_start(ei);
9465 write_extent_buffer(leaf, symname, ptr, name_len);
9466 btrfs_mark_buffer_dirty(trans, leaf);
9467 btrfs_free_path(path);
9469 d_instantiate_new(dentry, inode);
9472 btrfs_end_transaction(trans);
9473 btrfs_btree_balance_dirty(fs_info);
9475 btrfs_new_inode_args_destroy(&new_inode_args);
9482 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9483 struct btrfs_trans_handle *trans_in,
9484 struct btrfs_inode *inode,
9485 struct btrfs_key *ins,
9488 struct btrfs_file_extent_item stack_fi;
9489 struct btrfs_replace_extent_info extent_info;
9490 struct btrfs_trans_handle *trans = trans_in;
9491 struct btrfs_path *path;
9492 u64 start = ins->objectid;
9493 u64 len = ins->offset;
9494 u64 qgroup_released = 0;
9497 memset(&stack_fi, 0, sizeof(stack_fi));
9499 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9500 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9501 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9502 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9503 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9504 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9505 /* Encryption and other encoding is reserved and all 0 */
9507 ret = btrfs_qgroup_release_data(inode, file_offset, len, &qgroup_released);
9509 return ERR_PTR(ret);
9512 ret = insert_reserved_file_extent(trans, inode,
9513 file_offset, &stack_fi,
9514 true, qgroup_released);
9520 extent_info.disk_offset = start;
9521 extent_info.disk_len = len;
9522 extent_info.data_offset = 0;
9523 extent_info.data_len = len;
9524 extent_info.file_offset = file_offset;
9525 extent_info.extent_buf = (char *)&stack_fi;
9526 extent_info.is_new_extent = true;
9527 extent_info.update_times = true;
9528 extent_info.qgroup_reserved = qgroup_released;
9529 extent_info.insertions = 0;
9531 path = btrfs_alloc_path();
9537 ret = btrfs_replace_file_extents(inode, path, file_offset,
9538 file_offset + len - 1, &extent_info,
9540 btrfs_free_path(path);
9547 * We have released qgroup data range at the beginning of the function,
9548 * and normally qgroup_released bytes will be freed when committing
9550 * But if we error out early, we have to free what we have released
9551 * or we leak qgroup data reservation.
9553 btrfs_qgroup_free_refroot(inode->root->fs_info,
9554 inode->root->root_key.objectid, qgroup_released,
9555 BTRFS_QGROUP_RSV_DATA);
9556 return ERR_PTR(ret);
9559 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9560 u64 start, u64 num_bytes, u64 min_size,
9561 loff_t actual_len, u64 *alloc_hint,
9562 struct btrfs_trans_handle *trans)
9564 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9565 struct extent_map *em;
9566 struct btrfs_root *root = BTRFS_I(inode)->root;
9567 struct btrfs_key ins;
9568 u64 cur_offset = start;
9569 u64 clear_offset = start;
9572 u64 last_alloc = (u64)-1;
9574 bool own_trans = true;
9575 u64 end = start + num_bytes - 1;
9579 while (num_bytes > 0) {
9580 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9581 cur_bytes = max(cur_bytes, min_size);
9583 * If we are severely fragmented we could end up with really
9584 * small allocations, so if the allocator is returning small
9585 * chunks lets make its job easier by only searching for those
9588 cur_bytes = min(cur_bytes, last_alloc);
9589 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9590 min_size, 0, *alloc_hint, &ins, 1, 0);
9595 * We've reserved this space, and thus converted it from
9596 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9597 * from here on out we will only need to clear our reservation
9598 * for the remaining unreserved area, so advance our
9599 * clear_offset by our extent size.
9601 clear_offset += ins.offset;
9603 last_alloc = ins.offset;
9604 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9607 * Now that we inserted the prealloc extent we can finally
9608 * decrement the number of reservations in the block group.
9609 * If we did it before, we could race with relocation and have
9610 * relocation miss the reserved extent, making it fail later.
9612 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9613 if (IS_ERR(trans)) {
9614 ret = PTR_ERR(trans);
9615 btrfs_free_reserved_extent(fs_info, ins.objectid,
9620 em = alloc_extent_map();
9622 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9623 cur_offset + ins.offset - 1, false);
9624 btrfs_set_inode_full_sync(BTRFS_I(inode));
9628 em->start = cur_offset;
9629 em->orig_start = cur_offset;
9630 em->len = ins.offset;
9631 em->block_start = ins.objectid;
9632 em->block_len = ins.offset;
9633 em->orig_block_len = ins.offset;
9634 em->ram_bytes = ins.offset;
9635 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9636 em->generation = trans->transid;
9638 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9639 free_extent_map(em);
9641 num_bytes -= ins.offset;
9642 cur_offset += ins.offset;
9643 *alloc_hint = ins.objectid + ins.offset;
9645 inode_inc_iversion(inode);
9646 inode_set_ctime_current(inode);
9647 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9648 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9649 (actual_len > inode->i_size) &&
9650 (cur_offset > inode->i_size)) {
9651 if (cur_offset > actual_len)
9652 i_size = actual_len;
9654 i_size = cur_offset;
9655 i_size_write(inode, i_size);
9656 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9659 ret = btrfs_update_inode(trans, BTRFS_I(inode));
9662 btrfs_abort_transaction(trans, ret);
9664 btrfs_end_transaction(trans);
9669 btrfs_end_transaction(trans);
9673 if (clear_offset < end)
9674 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9675 end - clear_offset + 1);
9679 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9680 u64 start, u64 num_bytes, u64 min_size,
9681 loff_t actual_len, u64 *alloc_hint)
9683 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9684 min_size, actual_len, alloc_hint,
9688 int btrfs_prealloc_file_range_trans(struct inode *inode,
9689 struct btrfs_trans_handle *trans, int mode,
9690 u64 start, u64 num_bytes, u64 min_size,
9691 loff_t actual_len, u64 *alloc_hint)
9693 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9694 min_size, actual_len, alloc_hint, trans);
9697 static int btrfs_permission(struct mnt_idmap *idmap,
9698 struct inode *inode, int mask)
9700 struct btrfs_root *root = BTRFS_I(inode)->root;
9701 umode_t mode = inode->i_mode;
9703 if (mask & MAY_WRITE &&
9704 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9705 if (btrfs_root_readonly(root))
9707 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9710 return generic_permission(idmap, inode, mask);
9713 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9714 struct file *file, umode_t mode)
9716 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9717 struct btrfs_trans_handle *trans;
9718 struct btrfs_root *root = BTRFS_I(dir)->root;
9719 struct inode *inode;
9720 struct btrfs_new_inode_args new_inode_args = {
9722 .dentry = file->f_path.dentry,
9725 unsigned int trans_num_items;
9728 inode = new_inode(dir->i_sb);
9731 inode_init_owner(idmap, inode, dir, mode);
9732 inode->i_fop = &btrfs_file_operations;
9733 inode->i_op = &btrfs_file_inode_operations;
9734 inode->i_mapping->a_ops = &btrfs_aops;
9736 new_inode_args.inode = inode;
9737 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9741 trans = btrfs_start_transaction(root, trans_num_items);
9742 if (IS_ERR(trans)) {
9743 ret = PTR_ERR(trans);
9744 goto out_new_inode_args;
9747 ret = btrfs_create_new_inode(trans, &new_inode_args);
9750 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9751 * set it to 1 because d_tmpfile() will issue a warning if the count is
9754 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9756 set_nlink(inode, 1);
9759 d_tmpfile(file, inode);
9760 unlock_new_inode(inode);
9761 mark_inode_dirty(inode);
9764 btrfs_end_transaction(trans);
9765 btrfs_btree_balance_dirty(fs_info);
9767 btrfs_new_inode_args_destroy(&new_inode_args);
9771 return finish_open_simple(file, ret);
9774 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9776 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9777 unsigned long index = start >> PAGE_SHIFT;
9778 unsigned long end_index = end >> PAGE_SHIFT;
9782 ASSERT(end + 1 - start <= U32_MAX);
9783 len = end + 1 - start;
9784 while (index <= end_index) {
9785 page = find_get_page(inode->vfs_inode.i_mapping, index);
9786 ASSERT(page); /* Pages should be in the extent_io_tree */
9788 btrfs_page_set_writeback(fs_info, page, start, len);
9794 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9797 switch (compress_type) {
9798 case BTRFS_COMPRESS_NONE:
9799 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9800 case BTRFS_COMPRESS_ZLIB:
9801 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9802 case BTRFS_COMPRESS_LZO:
9804 * The LZO format depends on the sector size. 64K is the maximum
9805 * sector size that we support.
9807 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9809 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9810 (fs_info->sectorsize_bits - 12);
9811 case BTRFS_COMPRESS_ZSTD:
9812 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9818 static ssize_t btrfs_encoded_read_inline(
9820 struct iov_iter *iter, u64 start,
9822 struct extent_state **cached_state,
9823 u64 extent_start, size_t count,
9824 struct btrfs_ioctl_encoded_io_args *encoded,
9827 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9828 struct btrfs_root *root = inode->root;
9829 struct btrfs_fs_info *fs_info = root->fs_info;
9830 struct extent_io_tree *io_tree = &inode->io_tree;
9831 struct btrfs_path *path;
9832 struct extent_buffer *leaf;
9833 struct btrfs_file_extent_item *item;
9839 path = btrfs_alloc_path();
9844 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9848 /* The extent item disappeared? */
9853 leaf = path->nodes[0];
9854 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9856 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9857 ptr = btrfs_file_extent_inline_start(item);
9859 encoded->len = min_t(u64, extent_start + ram_bytes,
9860 inode->vfs_inode.i_size) - iocb->ki_pos;
9861 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9862 btrfs_file_extent_compression(leaf, item));
9865 encoded->compression = ret;
9866 if (encoded->compression) {
9869 inline_size = btrfs_file_extent_inline_item_len(leaf,
9871 if (inline_size > count) {
9875 count = inline_size;
9876 encoded->unencoded_len = ram_bytes;
9877 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9879 count = min_t(u64, count, encoded->len);
9880 encoded->len = count;
9881 encoded->unencoded_len = count;
9882 ptr += iocb->ki_pos - extent_start;
9885 tmp = kmalloc(count, GFP_NOFS);
9890 read_extent_buffer(leaf, tmp, ptr, count);
9891 btrfs_release_path(path);
9892 unlock_extent(io_tree, start, lockend, cached_state);
9893 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9896 ret = copy_to_iter(tmp, count, iter);
9901 btrfs_free_path(path);
9905 struct btrfs_encoded_read_private {
9906 wait_queue_head_t wait;
9908 blk_status_t status;
9911 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9913 struct btrfs_encoded_read_private *priv = bbio->private;
9915 if (bbio->bio.bi_status) {
9917 * The memory barrier implied by the atomic_dec_return() here
9918 * pairs with the memory barrier implied by the
9919 * atomic_dec_return() or io_wait_event() in
9920 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9921 * write is observed before the load of status in
9922 * btrfs_encoded_read_regular_fill_pages().
9924 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9926 if (!atomic_dec_return(&priv->pending))
9927 wake_up(&priv->wait);
9928 bio_put(&bbio->bio);
9931 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9932 u64 file_offset, u64 disk_bytenr,
9933 u64 disk_io_size, struct page **pages)
9935 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9936 struct btrfs_encoded_read_private priv = {
9937 .pending = ATOMIC_INIT(1),
9939 unsigned long i = 0;
9940 struct btrfs_bio *bbio;
9942 init_waitqueue_head(&priv.wait);
9944 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9945 btrfs_encoded_read_endio, &priv);
9946 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9947 bbio->inode = inode;
9950 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
9952 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
9953 atomic_inc(&priv.pending);
9954 btrfs_submit_bio(bbio, 0);
9956 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9957 btrfs_encoded_read_endio, &priv);
9958 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9959 bbio->inode = inode;
9964 disk_bytenr += bytes;
9965 disk_io_size -= bytes;
9966 } while (disk_io_size);
9968 atomic_inc(&priv.pending);
9969 btrfs_submit_bio(bbio, 0);
9971 if (atomic_dec_return(&priv.pending))
9972 io_wait_event(priv.wait, !atomic_read(&priv.pending));
9973 /* See btrfs_encoded_read_endio() for ordering. */
9974 return blk_status_to_errno(READ_ONCE(priv.status));
9977 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
9978 struct iov_iter *iter,
9979 u64 start, u64 lockend,
9980 struct extent_state **cached_state,
9981 u64 disk_bytenr, u64 disk_io_size,
9982 size_t count, bool compressed,
9985 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9986 struct extent_io_tree *io_tree = &inode->io_tree;
9987 struct page **pages;
9988 unsigned long nr_pages, i;
9993 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
9994 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
9997 ret = btrfs_alloc_page_array(nr_pages, pages);
10003 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10004 disk_io_size, pages);
10008 unlock_extent(io_tree, start, lockend, cached_state);
10009 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10016 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10017 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10020 while (cur < count) {
10021 size_t bytes = min_t(size_t, count - cur,
10022 PAGE_SIZE - page_offset);
10024 if (copy_page_to_iter(pages[i], page_offset, bytes,
10035 for (i = 0; i < nr_pages; i++) {
10037 __free_page(pages[i]);
10043 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10044 struct btrfs_ioctl_encoded_io_args *encoded)
10046 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10047 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10048 struct extent_io_tree *io_tree = &inode->io_tree;
10050 size_t count = iov_iter_count(iter);
10051 u64 start, lockend, disk_bytenr, disk_io_size;
10052 struct extent_state *cached_state = NULL;
10053 struct extent_map *em;
10054 bool unlocked = false;
10056 file_accessed(iocb->ki_filp);
10058 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10060 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10061 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10064 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10066 * We don't know how long the extent containing iocb->ki_pos is, but if
10067 * it's compressed we know that it won't be longer than this.
10069 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10072 struct btrfs_ordered_extent *ordered;
10074 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10075 lockend - start + 1);
10077 goto out_unlock_inode;
10078 lock_extent(io_tree, start, lockend, &cached_state);
10079 ordered = btrfs_lookup_ordered_range(inode, start,
10080 lockend - start + 1);
10083 btrfs_put_ordered_extent(ordered);
10084 unlock_extent(io_tree, start, lockend, &cached_state);
10088 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10091 goto out_unlock_extent;
10094 if (em->block_start == EXTENT_MAP_INLINE) {
10095 u64 extent_start = em->start;
10098 * For inline extents we get everything we need out of the
10101 free_extent_map(em);
10103 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10104 &cached_state, extent_start,
10105 count, encoded, &unlocked);
10110 * We only want to return up to EOF even if the extent extends beyond
10113 encoded->len = min_t(u64, extent_map_end(em),
10114 inode->vfs_inode.i_size) - iocb->ki_pos;
10115 if (em->block_start == EXTENT_MAP_HOLE ||
10116 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10117 disk_bytenr = EXTENT_MAP_HOLE;
10118 count = min_t(u64, count, encoded->len);
10119 encoded->len = count;
10120 encoded->unencoded_len = count;
10121 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10122 disk_bytenr = em->block_start;
10124 * Bail if the buffer isn't large enough to return the whole
10125 * compressed extent.
10127 if (em->block_len > count) {
10131 disk_io_size = em->block_len;
10132 count = em->block_len;
10133 encoded->unencoded_len = em->ram_bytes;
10134 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10135 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10136 em->compress_type);
10139 encoded->compression = ret;
10141 disk_bytenr = em->block_start + (start - em->start);
10142 if (encoded->len > count)
10143 encoded->len = count;
10145 * Don't read beyond what we locked. This also limits the page
10146 * allocations that we'll do.
10148 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10149 count = start + disk_io_size - iocb->ki_pos;
10150 encoded->len = count;
10151 encoded->unencoded_len = count;
10152 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10154 free_extent_map(em);
10157 if (disk_bytenr == EXTENT_MAP_HOLE) {
10158 unlock_extent(io_tree, start, lockend, &cached_state);
10159 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10161 ret = iov_iter_zero(count, iter);
10165 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10166 &cached_state, disk_bytenr,
10167 disk_io_size, count,
10168 encoded->compression,
10174 iocb->ki_pos += encoded->len;
10176 free_extent_map(em);
10179 unlock_extent(io_tree, start, lockend, &cached_state);
10182 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10186 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10187 const struct btrfs_ioctl_encoded_io_args *encoded)
10189 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10190 struct btrfs_root *root = inode->root;
10191 struct btrfs_fs_info *fs_info = root->fs_info;
10192 struct extent_io_tree *io_tree = &inode->io_tree;
10193 struct extent_changeset *data_reserved = NULL;
10194 struct extent_state *cached_state = NULL;
10195 struct btrfs_ordered_extent *ordered;
10199 u64 num_bytes, ram_bytes, disk_num_bytes;
10200 unsigned long nr_pages, i;
10201 struct page **pages;
10202 struct btrfs_key ins;
10203 bool extent_reserved = false;
10204 struct extent_map *em;
10207 switch (encoded->compression) {
10208 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10209 compression = BTRFS_COMPRESS_ZLIB;
10211 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10212 compression = BTRFS_COMPRESS_ZSTD;
10214 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10215 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10216 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10217 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10218 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10219 /* The sector size must match for LZO. */
10220 if (encoded->compression -
10221 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10222 fs_info->sectorsize_bits)
10224 compression = BTRFS_COMPRESS_LZO;
10229 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10232 orig_count = iov_iter_count(from);
10234 /* The extent size must be sane. */
10235 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10236 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10240 * The compressed data must be smaller than the decompressed data.
10242 * It's of course possible for data to compress to larger or the same
10243 * size, but the buffered I/O path falls back to no compression for such
10244 * data, and we don't want to break any assumptions by creating these
10247 * Note that this is less strict than the current check we have that the
10248 * compressed data must be at least one sector smaller than the
10249 * decompressed data. We only want to enforce the weaker requirement
10250 * from old kernels that it is at least one byte smaller.
10252 if (orig_count >= encoded->unencoded_len)
10255 /* The extent must start on a sector boundary. */
10256 start = iocb->ki_pos;
10257 if (!IS_ALIGNED(start, fs_info->sectorsize))
10261 * The extent must end on a sector boundary. However, we allow a write
10262 * which ends at or extends i_size to have an unaligned length; we round
10263 * up the extent size and set i_size to the unaligned end.
10265 if (start + encoded->len < inode->vfs_inode.i_size &&
10266 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10269 /* Finally, the offset in the unencoded data must be sector-aligned. */
10270 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10273 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10274 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10275 end = start + num_bytes - 1;
10278 * If the extent cannot be inline, the compressed data on disk must be
10279 * sector-aligned. For convenience, we extend it with zeroes if it
10282 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10283 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10284 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10287 for (i = 0; i < nr_pages; i++) {
10288 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10291 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10296 kaddr = kmap_local_page(pages[i]);
10297 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10298 kunmap_local(kaddr);
10302 if (bytes < PAGE_SIZE)
10303 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10304 kunmap_local(kaddr);
10308 struct btrfs_ordered_extent *ordered;
10310 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10313 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10314 start >> PAGE_SHIFT,
10315 end >> PAGE_SHIFT);
10318 lock_extent(io_tree, start, end, &cached_state);
10319 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10321 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10324 btrfs_put_ordered_extent(ordered);
10325 unlock_extent(io_tree, start, end, &cached_state);
10330 * We don't use the higher-level delalloc space functions because our
10331 * num_bytes and disk_num_bytes are different.
10333 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10336 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10338 goto out_free_data_space;
10339 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10342 goto out_qgroup_free_data;
10344 /* Try an inline extent first. */
10345 if (start == 0 && encoded->unencoded_len == encoded->len &&
10346 encoded->unencoded_offset == 0) {
10347 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10348 compression, pages, true);
10352 goto out_delalloc_release;
10356 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10357 disk_num_bytes, 0, 0, &ins, 1, 1);
10359 goto out_delalloc_release;
10360 extent_reserved = true;
10362 em = create_io_em(inode, start, num_bytes,
10363 start - encoded->unencoded_offset, ins.objectid,
10364 ins.offset, ins.offset, ram_bytes, compression,
10365 BTRFS_ORDERED_COMPRESSED);
10368 goto out_free_reserved;
10370 free_extent_map(em);
10372 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10373 ins.objectid, ins.offset,
10374 encoded->unencoded_offset,
10375 (1 << BTRFS_ORDERED_ENCODED) |
10376 (1 << BTRFS_ORDERED_COMPRESSED),
10378 if (IS_ERR(ordered)) {
10379 btrfs_drop_extent_map_range(inode, start, end, false);
10380 ret = PTR_ERR(ordered);
10381 goto out_free_reserved;
10383 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10385 if (start + encoded->len > inode->vfs_inode.i_size)
10386 i_size_write(&inode->vfs_inode, start + encoded->len);
10388 unlock_extent(io_tree, start, end, &cached_state);
10390 btrfs_delalloc_release_extents(inode, num_bytes);
10392 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10397 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10398 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10399 out_delalloc_release:
10400 btrfs_delalloc_release_extents(inode, num_bytes);
10401 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10402 out_qgroup_free_data:
10404 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes, NULL);
10405 out_free_data_space:
10407 * If btrfs_reserve_extent() succeeded, then we already decremented
10410 if (!extent_reserved)
10411 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10413 unlock_extent(io_tree, start, end, &cached_state);
10415 for (i = 0; i < nr_pages; i++) {
10417 __free_page(pages[i]);
10422 iocb->ki_pos += encoded->len;
10428 * Add an entry indicating a block group or device which is pinned by a
10429 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10430 * negative errno on failure.
10432 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10433 bool is_block_group)
10435 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10436 struct btrfs_swapfile_pin *sp, *entry;
10437 struct rb_node **p;
10438 struct rb_node *parent = NULL;
10440 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10445 sp->is_block_group = is_block_group;
10446 sp->bg_extent_count = 1;
10448 spin_lock(&fs_info->swapfile_pins_lock);
10449 p = &fs_info->swapfile_pins.rb_node;
10452 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10453 if (sp->ptr < entry->ptr ||
10454 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10455 p = &(*p)->rb_left;
10456 } else if (sp->ptr > entry->ptr ||
10457 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10458 p = &(*p)->rb_right;
10460 if (is_block_group)
10461 entry->bg_extent_count++;
10462 spin_unlock(&fs_info->swapfile_pins_lock);
10467 rb_link_node(&sp->node, parent, p);
10468 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10469 spin_unlock(&fs_info->swapfile_pins_lock);
10473 /* Free all of the entries pinned by this swapfile. */
10474 static void btrfs_free_swapfile_pins(struct inode *inode)
10476 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10477 struct btrfs_swapfile_pin *sp;
10478 struct rb_node *node, *next;
10480 spin_lock(&fs_info->swapfile_pins_lock);
10481 node = rb_first(&fs_info->swapfile_pins);
10483 next = rb_next(node);
10484 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10485 if (sp->inode == inode) {
10486 rb_erase(&sp->node, &fs_info->swapfile_pins);
10487 if (sp->is_block_group) {
10488 btrfs_dec_block_group_swap_extents(sp->ptr,
10489 sp->bg_extent_count);
10490 btrfs_put_block_group(sp->ptr);
10496 spin_unlock(&fs_info->swapfile_pins_lock);
10499 struct btrfs_swap_info {
10505 unsigned long nr_pages;
10509 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10510 struct btrfs_swap_info *bsi)
10512 unsigned long nr_pages;
10513 unsigned long max_pages;
10514 u64 first_ppage, first_ppage_reported, next_ppage;
10518 * Our swapfile may have had its size extended after the swap header was
10519 * written. In that case activating the swapfile should not go beyond
10520 * the max size set in the swap header.
10522 if (bsi->nr_pages >= sis->max)
10525 max_pages = sis->max - bsi->nr_pages;
10526 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10527 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10529 if (first_ppage >= next_ppage)
10531 nr_pages = next_ppage - first_ppage;
10532 nr_pages = min(nr_pages, max_pages);
10534 first_ppage_reported = first_ppage;
10535 if (bsi->start == 0)
10536 first_ppage_reported++;
10537 if (bsi->lowest_ppage > first_ppage_reported)
10538 bsi->lowest_ppage = first_ppage_reported;
10539 if (bsi->highest_ppage < (next_ppage - 1))
10540 bsi->highest_ppage = next_ppage - 1;
10542 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10545 bsi->nr_extents += ret;
10546 bsi->nr_pages += nr_pages;
10550 static void btrfs_swap_deactivate(struct file *file)
10552 struct inode *inode = file_inode(file);
10554 btrfs_free_swapfile_pins(inode);
10555 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10558 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10561 struct inode *inode = file_inode(file);
10562 struct btrfs_root *root = BTRFS_I(inode)->root;
10563 struct btrfs_fs_info *fs_info = root->fs_info;
10564 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10565 struct extent_state *cached_state = NULL;
10566 struct extent_map *em = NULL;
10567 struct btrfs_device *device = NULL;
10568 struct btrfs_swap_info bsi = {
10569 .lowest_ppage = (sector_t)-1ULL,
10576 * If the swap file was just created, make sure delalloc is done. If the
10577 * file changes again after this, the user is doing something stupid and
10578 * we don't really care.
10580 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10585 * The inode is locked, so these flags won't change after we check them.
10587 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10588 btrfs_warn(fs_info, "swapfile must not be compressed");
10591 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10592 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10595 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10596 btrfs_warn(fs_info, "swapfile must not be checksummed");
10601 * Balance or device remove/replace/resize can move stuff around from
10602 * under us. The exclop protection makes sure they aren't running/won't
10603 * run concurrently while we are mapping the swap extents, and
10604 * fs_info->swapfile_pins prevents them from running while the swap
10605 * file is active and moving the extents. Note that this also prevents
10606 * a concurrent device add which isn't actually necessary, but it's not
10607 * really worth the trouble to allow it.
10609 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10610 btrfs_warn(fs_info,
10611 "cannot activate swapfile while exclusive operation is running");
10616 * Prevent snapshot creation while we are activating the swap file.
10617 * We do not want to race with snapshot creation. If snapshot creation
10618 * already started before we bumped nr_swapfiles from 0 to 1 and
10619 * completes before the first write into the swap file after it is
10620 * activated, than that write would fallback to COW.
10622 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10623 btrfs_exclop_finish(fs_info);
10624 btrfs_warn(fs_info,
10625 "cannot activate swapfile because snapshot creation is in progress");
10629 * Snapshots can create extents which require COW even if NODATACOW is
10630 * set. We use this counter to prevent snapshots. We must increment it
10631 * before walking the extents because we don't want a concurrent
10632 * snapshot to run after we've already checked the extents.
10634 * It is possible that subvolume is marked for deletion but still not
10635 * removed yet. To prevent this race, we check the root status before
10636 * activating the swapfile.
10638 spin_lock(&root->root_item_lock);
10639 if (btrfs_root_dead(root)) {
10640 spin_unlock(&root->root_item_lock);
10642 btrfs_exclop_finish(fs_info);
10643 btrfs_warn(fs_info,
10644 "cannot activate swapfile because subvolume %llu is being deleted",
10645 root->root_key.objectid);
10648 atomic_inc(&root->nr_swapfiles);
10649 spin_unlock(&root->root_item_lock);
10651 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10653 lock_extent(io_tree, 0, isize - 1, &cached_state);
10655 while (start < isize) {
10656 u64 logical_block_start, physical_block_start;
10657 struct btrfs_block_group *bg;
10658 u64 len = isize - start;
10660 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10666 if (em->block_start == EXTENT_MAP_HOLE) {
10667 btrfs_warn(fs_info, "swapfile must not have holes");
10671 if (em->block_start == EXTENT_MAP_INLINE) {
10673 * It's unlikely we'll ever actually find ourselves
10674 * here, as a file small enough to fit inline won't be
10675 * big enough to store more than the swap header, but in
10676 * case something changes in the future, let's catch it
10677 * here rather than later.
10679 btrfs_warn(fs_info, "swapfile must not be inline");
10683 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10684 btrfs_warn(fs_info, "swapfile must not be compressed");
10689 logical_block_start = em->block_start + (start - em->start);
10690 len = min(len, em->len - (start - em->start));
10691 free_extent_map(em);
10694 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10700 btrfs_warn(fs_info,
10701 "swapfile must not be copy-on-write");
10706 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10712 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10713 btrfs_warn(fs_info,
10714 "swapfile must have single data profile");
10719 if (device == NULL) {
10720 device = em->map_lookup->stripes[0].dev;
10721 ret = btrfs_add_swapfile_pin(inode, device, false);
10726 } else if (device != em->map_lookup->stripes[0].dev) {
10727 btrfs_warn(fs_info, "swapfile must be on one device");
10732 physical_block_start = (em->map_lookup->stripes[0].physical +
10733 (logical_block_start - em->start));
10734 len = min(len, em->len - (logical_block_start - em->start));
10735 free_extent_map(em);
10738 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10740 btrfs_warn(fs_info,
10741 "could not find block group containing swapfile");
10746 if (!btrfs_inc_block_group_swap_extents(bg)) {
10747 btrfs_warn(fs_info,
10748 "block group for swapfile at %llu is read-only%s",
10750 atomic_read(&fs_info->scrubs_running) ?
10751 " (scrub running)" : "");
10752 btrfs_put_block_group(bg);
10757 ret = btrfs_add_swapfile_pin(inode, bg, true);
10759 btrfs_put_block_group(bg);
10766 if (bsi.block_len &&
10767 bsi.block_start + bsi.block_len == physical_block_start) {
10768 bsi.block_len += len;
10770 if (bsi.block_len) {
10771 ret = btrfs_add_swap_extent(sis, &bsi);
10776 bsi.block_start = physical_block_start;
10777 bsi.block_len = len;
10784 ret = btrfs_add_swap_extent(sis, &bsi);
10787 if (!IS_ERR_OR_NULL(em))
10788 free_extent_map(em);
10790 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10793 btrfs_swap_deactivate(file);
10795 btrfs_drew_write_unlock(&root->snapshot_lock);
10797 btrfs_exclop_finish(fs_info);
10803 sis->bdev = device->bdev;
10804 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10805 sis->max = bsi.nr_pages;
10806 sis->pages = bsi.nr_pages - 1;
10807 sis->highest_bit = bsi.nr_pages - 1;
10808 return bsi.nr_extents;
10811 static void btrfs_swap_deactivate(struct file *file)
10815 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10818 return -EOPNOTSUPP;
10823 * Update the number of bytes used in the VFS' inode. When we replace extents in
10824 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10825 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10826 * always get a correct value.
10828 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10829 const u64 add_bytes,
10830 const u64 del_bytes)
10832 if (add_bytes == del_bytes)
10835 spin_lock(&inode->lock);
10837 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10839 inode_add_bytes(&inode->vfs_inode, add_bytes);
10840 spin_unlock(&inode->lock);
10844 * Verify that there are no ordered extents for a given file range.
10846 * @inode: The target inode.
10847 * @start: Start offset of the file range, should be sector size aligned.
10848 * @end: End offset (inclusive) of the file range, its value +1 should be
10849 * sector size aligned.
10851 * This should typically be used for cases where we locked an inode's VFS lock in
10852 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10853 * we have flushed all delalloc in the range, we have waited for all ordered
10854 * extents in the range to complete and finally we have locked the file range in
10855 * the inode's io_tree.
10857 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10859 struct btrfs_root *root = inode->root;
10860 struct btrfs_ordered_extent *ordered;
10862 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10865 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10867 btrfs_err(root->fs_info,
10868 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10869 start, end, btrfs_ino(inode), root->root_key.objectid,
10870 ordered->file_offset,
10871 ordered->file_offset + ordered->num_bytes - 1);
10872 btrfs_put_ordered_extent(ordered);
10875 ASSERT(ordered == NULL);
10878 static const struct inode_operations btrfs_dir_inode_operations = {
10879 .getattr = btrfs_getattr,
10880 .lookup = btrfs_lookup,
10881 .create = btrfs_create,
10882 .unlink = btrfs_unlink,
10883 .link = btrfs_link,
10884 .mkdir = btrfs_mkdir,
10885 .rmdir = btrfs_rmdir,
10886 .rename = btrfs_rename2,
10887 .symlink = btrfs_symlink,
10888 .setattr = btrfs_setattr,
10889 .mknod = btrfs_mknod,
10890 .listxattr = btrfs_listxattr,
10891 .permission = btrfs_permission,
10892 .get_inode_acl = btrfs_get_acl,
10893 .set_acl = btrfs_set_acl,
10894 .update_time = btrfs_update_time,
10895 .tmpfile = btrfs_tmpfile,
10896 .fileattr_get = btrfs_fileattr_get,
10897 .fileattr_set = btrfs_fileattr_set,
10900 static const struct file_operations btrfs_dir_file_operations = {
10901 .llseek = btrfs_dir_llseek,
10902 .read = generic_read_dir,
10903 .iterate_shared = btrfs_real_readdir,
10904 .open = btrfs_opendir,
10905 .unlocked_ioctl = btrfs_ioctl,
10906 #ifdef CONFIG_COMPAT
10907 .compat_ioctl = btrfs_compat_ioctl,
10909 .release = btrfs_release_file,
10910 .fsync = btrfs_sync_file,
10914 * btrfs doesn't support the bmap operation because swapfiles
10915 * use bmap to make a mapping of extents in the file. They assume
10916 * these extents won't change over the life of the file and they
10917 * use the bmap result to do IO directly to the drive.
10919 * the btrfs bmap call would return logical addresses that aren't
10920 * suitable for IO and they also will change frequently as COW
10921 * operations happen. So, swapfile + btrfs == corruption.
10923 * For now we're avoiding this by dropping bmap.
10925 static const struct address_space_operations btrfs_aops = {
10926 .read_folio = btrfs_read_folio,
10927 .writepages = btrfs_writepages,
10928 .readahead = btrfs_readahead,
10929 .invalidate_folio = btrfs_invalidate_folio,
10930 .release_folio = btrfs_release_folio,
10931 .migrate_folio = btrfs_migrate_folio,
10932 .dirty_folio = filemap_dirty_folio,
10933 .error_remove_page = generic_error_remove_page,
10934 .swap_activate = btrfs_swap_activate,
10935 .swap_deactivate = btrfs_swap_deactivate,
10938 static const struct inode_operations btrfs_file_inode_operations = {
10939 .getattr = btrfs_getattr,
10940 .setattr = btrfs_setattr,
10941 .listxattr = btrfs_listxattr,
10942 .permission = btrfs_permission,
10943 .fiemap = btrfs_fiemap,
10944 .get_inode_acl = btrfs_get_acl,
10945 .set_acl = btrfs_set_acl,
10946 .update_time = btrfs_update_time,
10947 .fileattr_get = btrfs_fileattr_get,
10948 .fileattr_set = btrfs_fileattr_set,
10950 static const struct inode_operations btrfs_special_inode_operations = {
10951 .getattr = btrfs_getattr,
10952 .setattr = btrfs_setattr,
10953 .permission = btrfs_permission,
10954 .listxattr = btrfs_listxattr,
10955 .get_inode_acl = btrfs_get_acl,
10956 .set_acl = btrfs_set_acl,
10957 .update_time = btrfs_update_time,
10959 static const struct inode_operations btrfs_symlink_inode_operations = {
10960 .get_link = page_get_link,
10961 .getattr = btrfs_getattr,
10962 .setattr = btrfs_setattr,
10963 .permission = btrfs_permission,
10964 .listxattr = btrfs_listxattr,
10965 .update_time = btrfs_update_time,
10968 const struct dentry_operations btrfs_dentry_operations = {
10969 .d_delete = btrfs_dentry_delete,