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);
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->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3766 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3768 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3769 inode->i_mtime.tv_nsec = 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.tv_sec =
3775 btrfs_timespec_sec(leaf, &inode_item->otime);
3776 BTRFS_I(inode)->i_otime.tv_nsec =
3777 btrfs_timespec_nsec(leaf, &inode_item->otime);
3779 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3780 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3781 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3783 inode_set_iversion_queried(inode,
3784 btrfs_inode_sequence(leaf, inode_item));
3785 inode->i_generation = BTRFS_I(inode)->generation;
3787 rdev = btrfs_inode_rdev(leaf, inode_item);
3789 BTRFS_I(inode)->index_cnt = (u64)-1;
3790 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3791 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3795 * If we were modified in the current generation and evicted from memory
3796 * and then re-read we need to do a full sync since we don't have any
3797 * idea about which extents were modified before we were evicted from
3800 * This is required for both inode re-read from disk and delayed inode
3801 * in delayed_nodes_tree.
3803 if (BTRFS_I(inode)->last_trans == btrfs_get_fs_generation(fs_info))
3804 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3805 &BTRFS_I(inode)->runtime_flags);
3808 * We don't persist the id of the transaction where an unlink operation
3809 * against the inode was last made. So here we assume the inode might
3810 * have been evicted, and therefore the exact value of last_unlink_trans
3811 * lost, and set it to last_trans to avoid metadata inconsistencies
3812 * between the inode and its parent if the inode is fsync'ed and the log
3813 * replayed. For example, in the scenario:
3816 * ln mydir/foo mydir/bar
3819 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3820 * xfs_io -c fsync mydir/foo
3822 * mount fs, triggers fsync log replay
3824 * We must make sure that when we fsync our inode foo we also log its
3825 * parent inode, otherwise after log replay the parent still has the
3826 * dentry with the "bar" name but our inode foo has a link count of 1
3827 * and doesn't have an inode ref with the name "bar" anymore.
3829 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3830 * but it guarantees correctness at the expense of occasional full
3831 * transaction commits on fsync if our inode is a directory, or if our
3832 * inode is not a directory, logging its parent unnecessarily.
3834 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3837 * Same logic as for last_unlink_trans. We don't persist the generation
3838 * of the last transaction where this inode was used for a reflink
3839 * operation, so after eviction and reloading the inode we must be
3840 * pessimistic and assume the last transaction that modified the inode.
3842 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3845 if (inode->i_nlink != 1 ||
3846 path->slots[0] >= btrfs_header_nritems(leaf))
3849 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3850 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3853 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3854 if (location.type == BTRFS_INODE_REF_KEY) {
3855 struct btrfs_inode_ref *ref;
3857 ref = (struct btrfs_inode_ref *)ptr;
3858 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3859 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3860 struct btrfs_inode_extref *extref;
3862 extref = (struct btrfs_inode_extref *)ptr;
3863 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3868 * try to precache a NULL acl entry for files that don't have
3869 * any xattrs or acls
3871 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3872 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3873 if (first_xattr_slot != -1) {
3874 path->slots[0] = first_xattr_slot;
3875 ret = btrfs_load_inode_props(inode, path);
3878 "error loading props for ino %llu (root %llu): %d",
3879 btrfs_ino(BTRFS_I(inode)),
3880 root->root_key.objectid, ret);
3882 if (path != in_path)
3883 btrfs_free_path(path);
3886 cache_no_acl(inode);
3888 switch (inode->i_mode & S_IFMT) {
3890 inode->i_mapping->a_ops = &btrfs_aops;
3891 inode->i_fop = &btrfs_file_operations;
3892 inode->i_op = &btrfs_file_inode_operations;
3895 inode->i_fop = &btrfs_dir_file_operations;
3896 inode->i_op = &btrfs_dir_inode_operations;
3899 inode->i_op = &btrfs_symlink_inode_operations;
3900 inode_nohighmem(inode);
3901 inode->i_mapping->a_ops = &btrfs_aops;
3904 inode->i_op = &btrfs_special_inode_operations;
3905 init_special_inode(inode, inode->i_mode, rdev);
3909 btrfs_sync_inode_flags_to_i_flags(inode);
3914 * given a leaf and an inode, copy the inode fields into the leaf
3916 static void fill_inode_item(struct btrfs_trans_handle *trans,
3917 struct extent_buffer *leaf,
3918 struct btrfs_inode_item *item,
3919 struct inode *inode)
3921 struct btrfs_map_token token;
3924 btrfs_init_map_token(&token, leaf);
3926 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3927 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3928 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3929 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3930 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3932 btrfs_set_token_timespec_sec(&token, &item->atime,
3933 inode->i_atime.tv_sec);
3934 btrfs_set_token_timespec_nsec(&token, &item->atime,
3935 inode->i_atime.tv_nsec);
3937 btrfs_set_token_timespec_sec(&token, &item->mtime,
3938 inode->i_mtime.tv_sec);
3939 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3940 inode->i_mtime.tv_nsec);
3942 btrfs_set_token_timespec_sec(&token, &item->ctime,
3943 inode_get_ctime(inode).tv_sec);
3944 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3945 inode_get_ctime(inode).tv_nsec);
3947 btrfs_set_token_timespec_sec(&token, &item->otime,
3948 BTRFS_I(inode)->i_otime.tv_sec);
3949 btrfs_set_token_timespec_nsec(&token, &item->otime,
3950 BTRFS_I(inode)->i_otime.tv_nsec);
3952 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3953 btrfs_set_token_inode_generation(&token, item,
3954 BTRFS_I(inode)->generation);
3955 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3956 btrfs_set_token_inode_transid(&token, item, trans->transid);
3957 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3958 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3959 BTRFS_I(inode)->ro_flags);
3960 btrfs_set_token_inode_flags(&token, item, flags);
3961 btrfs_set_token_inode_block_group(&token, item, 0);
3965 * copy everything in the in-memory inode into the btree.
3967 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3968 struct btrfs_inode *inode)
3970 struct btrfs_inode_item *inode_item;
3971 struct btrfs_path *path;
3972 struct extent_buffer *leaf;
3975 path = btrfs_alloc_path();
3979 ret = btrfs_lookup_inode(trans, inode->root, path, &inode->location, 1);
3986 leaf = path->nodes[0];
3987 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3988 struct btrfs_inode_item);
3990 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
3991 btrfs_mark_buffer_dirty(trans, leaf);
3992 btrfs_set_inode_last_trans(trans, inode);
3995 btrfs_free_path(path);
4000 * copy everything in the in-memory inode into the btree.
4002 int btrfs_update_inode(struct btrfs_trans_handle *trans,
4003 struct btrfs_inode *inode)
4005 struct btrfs_root *root = inode->root;
4006 struct btrfs_fs_info *fs_info = root->fs_info;
4010 * If the inode is a free space inode, we can deadlock during commit
4011 * if we put it into the delayed code.
4013 * The data relocation inode should also be directly updated
4016 if (!btrfs_is_free_space_inode(inode)
4017 && !btrfs_is_data_reloc_root(root)
4018 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4019 btrfs_update_root_times(trans, root);
4021 ret = btrfs_delayed_update_inode(trans, inode);
4023 btrfs_set_inode_last_trans(trans, inode);
4027 return btrfs_update_inode_item(trans, inode);
4030 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4031 struct btrfs_inode *inode)
4035 ret = btrfs_update_inode(trans, inode);
4037 return btrfs_update_inode_item(trans, inode);
4042 * unlink helper that gets used here in inode.c and in the tree logging
4043 * recovery code. It remove a link in a directory with a given name, and
4044 * also drops the back refs in the inode to the directory
4046 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4047 struct btrfs_inode *dir,
4048 struct btrfs_inode *inode,
4049 const struct fscrypt_str *name,
4050 struct btrfs_rename_ctx *rename_ctx)
4052 struct btrfs_root *root = dir->root;
4053 struct btrfs_fs_info *fs_info = root->fs_info;
4054 struct btrfs_path *path;
4056 struct btrfs_dir_item *di;
4058 u64 ino = btrfs_ino(inode);
4059 u64 dir_ino = btrfs_ino(dir);
4061 path = btrfs_alloc_path();
4067 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4068 if (IS_ERR_OR_NULL(di)) {
4069 ret = di ? PTR_ERR(di) : -ENOENT;
4072 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4075 btrfs_release_path(path);
4078 * If we don't have dir index, we have to get it by looking up
4079 * the inode ref, since we get the inode ref, remove it directly,
4080 * it is unnecessary to do delayed deletion.
4082 * But if we have dir index, needn't search inode ref to get it.
4083 * Since the inode ref is close to the inode item, it is better
4084 * that we delay to delete it, and just do this deletion when
4085 * we update the inode item.
4087 if (inode->dir_index) {
4088 ret = btrfs_delayed_delete_inode_ref(inode);
4090 index = inode->dir_index;
4095 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4098 "failed to delete reference to %.*s, inode %llu parent %llu",
4099 name->len, name->name, ino, dir_ino);
4100 btrfs_abort_transaction(trans, ret);
4105 rename_ctx->index = index;
4107 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4109 btrfs_abort_transaction(trans, ret);
4114 * If we are in a rename context, we don't need to update anything in the
4115 * log. That will be done later during the rename by btrfs_log_new_name().
4116 * Besides that, doing it here would only cause extra unnecessary btree
4117 * operations on the log tree, increasing latency for applications.
4120 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4121 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4125 * If we have a pending delayed iput we could end up with the final iput
4126 * being run in btrfs-cleaner context. If we have enough of these built
4127 * up we can end up burning a lot of time in btrfs-cleaner without any
4128 * way to throttle the unlinks. Since we're currently holding a ref on
4129 * the inode we can run the delayed iput here without any issues as the
4130 * final iput won't be done until after we drop the ref we're currently
4133 btrfs_run_delayed_iput(fs_info, inode);
4135 btrfs_free_path(path);
4139 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4140 inode_inc_iversion(&inode->vfs_inode);
4141 inode_inc_iversion(&dir->vfs_inode);
4142 inode_set_ctime_current(&inode->vfs_inode);
4143 dir->vfs_inode.i_mtime = inode_set_ctime_current(&dir->vfs_inode);
4144 ret = btrfs_update_inode(trans, dir);
4149 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4150 struct btrfs_inode *dir, struct btrfs_inode *inode,
4151 const struct fscrypt_str *name)
4155 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4157 drop_nlink(&inode->vfs_inode);
4158 ret = btrfs_update_inode(trans, inode);
4164 * helper to start transaction for unlink and rmdir.
4166 * unlink and rmdir are special in btrfs, they do not always free space, so
4167 * if we cannot make our reservations the normal way try and see if there is
4168 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4169 * allow the unlink to occur.
4171 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4173 struct btrfs_root *root = dir->root;
4175 return btrfs_start_transaction_fallback_global_rsv(root,
4176 BTRFS_UNLINK_METADATA_UNITS);
4179 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4181 struct btrfs_trans_handle *trans;
4182 struct inode *inode = d_inode(dentry);
4184 struct fscrypt_name fname;
4186 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4190 /* This needs to handle no-key deletions later on */
4192 trans = __unlink_start_trans(BTRFS_I(dir));
4193 if (IS_ERR(trans)) {
4194 ret = PTR_ERR(trans);
4198 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4201 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4206 if (inode->i_nlink == 0) {
4207 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4213 btrfs_end_transaction(trans);
4214 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4216 fscrypt_free_filename(&fname);
4220 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4221 struct btrfs_inode *dir, struct dentry *dentry)
4223 struct btrfs_root *root = dir->root;
4224 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4225 struct btrfs_path *path;
4226 struct extent_buffer *leaf;
4227 struct btrfs_dir_item *di;
4228 struct btrfs_key key;
4232 u64 dir_ino = btrfs_ino(dir);
4233 struct fscrypt_name fname;
4235 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4239 /* This needs to handle no-key deletions later on */
4241 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4242 objectid = inode->root->root_key.objectid;
4243 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4244 objectid = inode->location.objectid;
4247 fscrypt_free_filename(&fname);
4251 path = btrfs_alloc_path();
4257 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4258 &fname.disk_name, -1);
4259 if (IS_ERR_OR_NULL(di)) {
4260 ret = di ? PTR_ERR(di) : -ENOENT;
4264 leaf = path->nodes[0];
4265 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4266 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4267 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4269 btrfs_abort_transaction(trans, ret);
4272 btrfs_release_path(path);
4275 * This is a placeholder inode for a subvolume we didn't have a
4276 * reference to at the time of the snapshot creation. In the meantime
4277 * we could have renamed the real subvol link into our snapshot, so
4278 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4279 * Instead simply lookup the dir_index_item for this entry so we can
4280 * remove it. Otherwise we know we have a ref to the root and we can
4281 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4283 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4284 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4285 if (IS_ERR_OR_NULL(di)) {
4290 btrfs_abort_transaction(trans, ret);
4294 leaf = path->nodes[0];
4295 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4297 btrfs_release_path(path);
4299 ret = btrfs_del_root_ref(trans, objectid,
4300 root->root_key.objectid, dir_ino,
4301 &index, &fname.disk_name);
4303 btrfs_abort_transaction(trans, ret);
4308 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4310 btrfs_abort_transaction(trans, ret);
4314 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4315 inode_inc_iversion(&dir->vfs_inode);
4316 dir->vfs_inode.i_mtime = inode_set_ctime_current(&dir->vfs_inode);
4317 ret = btrfs_update_inode_fallback(trans, dir);
4319 btrfs_abort_transaction(trans, ret);
4321 btrfs_free_path(path);
4322 fscrypt_free_filename(&fname);
4327 * Helper to check if the subvolume references other subvolumes or if it's
4330 static noinline int may_destroy_subvol(struct btrfs_root *root)
4332 struct btrfs_fs_info *fs_info = root->fs_info;
4333 struct btrfs_path *path;
4334 struct btrfs_dir_item *di;
4335 struct btrfs_key key;
4336 struct fscrypt_str name = FSTR_INIT("default", 7);
4340 path = btrfs_alloc_path();
4344 /* Make sure this root isn't set as the default subvol */
4345 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4346 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4348 if (di && !IS_ERR(di)) {
4349 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4350 if (key.objectid == root->root_key.objectid) {
4353 "deleting default subvolume %llu is not allowed",
4357 btrfs_release_path(path);
4360 key.objectid = root->root_key.objectid;
4361 key.type = BTRFS_ROOT_REF_KEY;
4362 key.offset = (u64)-1;
4364 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4370 if (path->slots[0] > 0) {
4372 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4373 if (key.objectid == root->root_key.objectid &&
4374 key.type == BTRFS_ROOT_REF_KEY)
4378 btrfs_free_path(path);
4382 /* Delete all dentries for inodes belonging to the root */
4383 static void btrfs_prune_dentries(struct btrfs_root *root)
4385 struct btrfs_fs_info *fs_info = root->fs_info;
4386 struct rb_node *node;
4387 struct rb_node *prev;
4388 struct btrfs_inode *entry;
4389 struct inode *inode;
4392 if (!BTRFS_FS_ERROR(fs_info))
4393 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4395 spin_lock(&root->inode_lock);
4397 node = root->inode_tree.rb_node;
4401 entry = rb_entry(node, struct btrfs_inode, rb_node);
4403 if (objectid < btrfs_ino(entry))
4404 node = node->rb_left;
4405 else if (objectid > btrfs_ino(entry))
4406 node = node->rb_right;
4412 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4413 if (objectid <= btrfs_ino(entry)) {
4417 prev = rb_next(prev);
4421 entry = rb_entry(node, struct btrfs_inode, rb_node);
4422 objectid = btrfs_ino(entry) + 1;
4423 inode = igrab(&entry->vfs_inode);
4425 spin_unlock(&root->inode_lock);
4426 if (atomic_read(&inode->i_count) > 1)
4427 d_prune_aliases(inode);
4429 * btrfs_drop_inode will have it removed from the inode
4430 * cache when its usage count hits zero.
4434 spin_lock(&root->inode_lock);
4438 if (cond_resched_lock(&root->inode_lock))
4441 node = rb_next(node);
4443 spin_unlock(&root->inode_lock);
4446 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4448 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4449 struct btrfs_root *root = dir->root;
4450 struct inode *inode = d_inode(dentry);
4451 struct btrfs_root *dest = BTRFS_I(inode)->root;
4452 struct btrfs_trans_handle *trans;
4453 struct btrfs_block_rsv block_rsv;
4458 * Don't allow to delete a subvolume with send in progress. This is
4459 * inside the inode lock so the error handling that has to drop the bit
4460 * again is not run concurrently.
4462 spin_lock(&dest->root_item_lock);
4463 if (dest->send_in_progress) {
4464 spin_unlock(&dest->root_item_lock);
4466 "attempt to delete subvolume %llu during send",
4467 dest->root_key.objectid);
4470 if (atomic_read(&dest->nr_swapfiles)) {
4471 spin_unlock(&dest->root_item_lock);
4473 "attempt to delete subvolume %llu with active swapfile",
4474 root->root_key.objectid);
4477 root_flags = btrfs_root_flags(&dest->root_item);
4478 btrfs_set_root_flags(&dest->root_item,
4479 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4480 spin_unlock(&dest->root_item_lock);
4482 down_write(&fs_info->subvol_sem);
4484 ret = may_destroy_subvol(dest);
4488 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4490 * One for dir inode,
4491 * two for dir entries,
4492 * two for root ref/backref.
4494 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4498 trans = btrfs_start_transaction(root, 0);
4499 if (IS_ERR(trans)) {
4500 ret = PTR_ERR(trans);
4503 trans->block_rsv = &block_rsv;
4504 trans->bytes_reserved = block_rsv.size;
4506 btrfs_record_snapshot_destroy(trans, dir);
4508 ret = btrfs_unlink_subvol(trans, dir, dentry);
4510 btrfs_abort_transaction(trans, ret);
4514 ret = btrfs_record_root_in_trans(trans, dest);
4516 btrfs_abort_transaction(trans, ret);
4520 memset(&dest->root_item.drop_progress, 0,
4521 sizeof(dest->root_item.drop_progress));
4522 btrfs_set_root_drop_level(&dest->root_item, 0);
4523 btrfs_set_root_refs(&dest->root_item, 0);
4525 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4526 ret = btrfs_insert_orphan_item(trans,
4528 dest->root_key.objectid);
4530 btrfs_abort_transaction(trans, ret);
4535 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4536 BTRFS_UUID_KEY_SUBVOL,
4537 dest->root_key.objectid);
4538 if (ret && ret != -ENOENT) {
4539 btrfs_abort_transaction(trans, ret);
4542 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4543 ret = btrfs_uuid_tree_remove(trans,
4544 dest->root_item.received_uuid,
4545 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4546 dest->root_key.objectid);
4547 if (ret && ret != -ENOENT) {
4548 btrfs_abort_transaction(trans, ret);
4553 free_anon_bdev(dest->anon_dev);
4556 trans->block_rsv = NULL;
4557 trans->bytes_reserved = 0;
4558 ret = btrfs_end_transaction(trans);
4559 inode->i_flags |= S_DEAD;
4561 btrfs_subvolume_release_metadata(root, &block_rsv);
4563 up_write(&fs_info->subvol_sem);
4565 spin_lock(&dest->root_item_lock);
4566 root_flags = btrfs_root_flags(&dest->root_item);
4567 btrfs_set_root_flags(&dest->root_item,
4568 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4569 spin_unlock(&dest->root_item_lock);
4571 d_invalidate(dentry);
4572 btrfs_prune_dentries(dest);
4573 ASSERT(dest->send_in_progress == 0);
4579 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4581 struct inode *inode = d_inode(dentry);
4582 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4584 struct btrfs_trans_handle *trans;
4585 u64 last_unlink_trans;
4586 struct fscrypt_name fname;
4588 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4590 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4591 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4593 "extent tree v2 doesn't support snapshot deletion yet");
4596 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4599 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4603 /* This needs to handle no-key deletions later on */
4605 trans = __unlink_start_trans(BTRFS_I(dir));
4606 if (IS_ERR(trans)) {
4607 err = PTR_ERR(trans);
4611 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4612 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4616 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4620 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4622 /* now the directory is empty */
4623 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4626 btrfs_i_size_write(BTRFS_I(inode), 0);
4628 * Propagate the last_unlink_trans value of the deleted dir to
4629 * its parent directory. This is to prevent an unrecoverable
4630 * log tree in the case we do something like this:
4632 * 2) create snapshot under dir foo
4633 * 3) delete the snapshot
4636 * 6) fsync foo or some file inside foo
4638 if (last_unlink_trans >= trans->transid)
4639 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4642 btrfs_end_transaction(trans);
4644 btrfs_btree_balance_dirty(fs_info);
4645 fscrypt_free_filename(&fname);
4651 * Read, zero a chunk and write a block.
4653 * @inode - inode that we're zeroing
4654 * @from - the offset to start zeroing
4655 * @len - the length to zero, 0 to zero the entire range respective to the
4657 * @front - zero up to the offset instead of from the offset on
4659 * This will find the block for the "from" offset and cow the block and zero the
4660 * part we want to zero. This is used with truncate and hole punching.
4662 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4665 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4666 struct address_space *mapping = inode->vfs_inode.i_mapping;
4667 struct extent_io_tree *io_tree = &inode->io_tree;
4668 struct btrfs_ordered_extent *ordered;
4669 struct extent_state *cached_state = NULL;
4670 struct extent_changeset *data_reserved = NULL;
4671 bool only_release_metadata = false;
4672 u32 blocksize = fs_info->sectorsize;
4673 pgoff_t index = from >> PAGE_SHIFT;
4674 unsigned offset = from & (blocksize - 1);
4676 gfp_t mask = btrfs_alloc_write_mask(mapping);
4677 size_t write_bytes = blocksize;
4682 if (IS_ALIGNED(offset, blocksize) &&
4683 (!len || IS_ALIGNED(len, blocksize)))
4686 block_start = round_down(from, blocksize);
4687 block_end = block_start + blocksize - 1;
4689 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4692 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4693 /* For nocow case, no need to reserve data space */
4694 only_release_metadata = true;
4699 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4701 if (!only_release_metadata)
4702 btrfs_free_reserved_data_space(inode, data_reserved,
4703 block_start, blocksize);
4707 page = find_or_create_page(mapping, index, mask);
4709 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4711 btrfs_delalloc_release_extents(inode, blocksize);
4716 if (!PageUptodate(page)) {
4717 ret = btrfs_read_folio(NULL, page_folio(page));
4719 if (page->mapping != mapping) {
4724 if (!PageUptodate(page)) {
4731 * We unlock the page after the io is completed and then re-lock it
4732 * above. release_folio() could have come in between that and cleared
4733 * PagePrivate(), but left the page in the mapping. Set the page mapped
4734 * here to make sure it's properly set for the subpage stuff.
4736 ret = set_page_extent_mapped(page);
4740 wait_on_page_writeback(page);
4742 lock_extent(io_tree, block_start, block_end, &cached_state);
4744 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4746 unlock_extent(io_tree, block_start, block_end, &cached_state);
4749 btrfs_start_ordered_extent(ordered);
4750 btrfs_put_ordered_extent(ordered);
4754 clear_extent_bit(&inode->io_tree, block_start, block_end,
4755 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4758 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4761 unlock_extent(io_tree, block_start, block_end, &cached_state);
4765 if (offset != blocksize) {
4767 len = blocksize - offset;
4769 memzero_page(page, (block_start - page_offset(page)),
4772 memzero_page(page, (block_start - page_offset(page)) + offset,
4775 btrfs_page_clear_checked(fs_info, page, block_start,
4776 block_end + 1 - block_start);
4777 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4778 unlock_extent(io_tree, block_start, block_end, &cached_state);
4780 if (only_release_metadata)
4781 set_extent_bit(&inode->io_tree, block_start, block_end,
4782 EXTENT_NORESERVE, NULL);
4786 if (only_release_metadata)
4787 btrfs_delalloc_release_metadata(inode, blocksize, true);
4789 btrfs_delalloc_release_space(inode, data_reserved,
4790 block_start, blocksize, true);
4792 btrfs_delalloc_release_extents(inode, blocksize);
4796 if (only_release_metadata)
4797 btrfs_check_nocow_unlock(inode);
4798 extent_changeset_free(data_reserved);
4802 static int maybe_insert_hole(struct btrfs_inode *inode, u64 offset, u64 len)
4804 struct btrfs_root *root = inode->root;
4805 struct btrfs_fs_info *fs_info = root->fs_info;
4806 struct btrfs_trans_handle *trans;
4807 struct btrfs_drop_extents_args drop_args = { 0 };
4811 * If NO_HOLES is enabled, we don't need to do anything.
4812 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4813 * or btrfs_update_inode() will be called, which guarantee that the next
4814 * fsync will know this inode was changed and needs to be logged.
4816 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4820 * 1 - for the one we're dropping
4821 * 1 - for the one we're adding
4822 * 1 - for updating the inode.
4824 trans = btrfs_start_transaction(root, 3);
4826 return PTR_ERR(trans);
4828 drop_args.start = offset;
4829 drop_args.end = offset + len;
4830 drop_args.drop_cache = true;
4832 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4834 btrfs_abort_transaction(trans, ret);
4835 btrfs_end_transaction(trans);
4839 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4841 btrfs_abort_transaction(trans, ret);
4843 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4844 btrfs_update_inode(trans, inode);
4846 btrfs_end_transaction(trans);
4851 * This function puts in dummy file extents for the area we're creating a hole
4852 * for. So if we are truncating this file to a larger size we need to insert
4853 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4854 * the range between oldsize and size
4856 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4858 struct btrfs_root *root = inode->root;
4859 struct btrfs_fs_info *fs_info = root->fs_info;
4860 struct extent_io_tree *io_tree = &inode->io_tree;
4861 struct extent_map *em = NULL;
4862 struct extent_state *cached_state = NULL;
4863 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4864 u64 block_end = ALIGN(size, fs_info->sectorsize);
4871 * If our size started in the middle of a block we need to zero out the
4872 * rest of the block before we expand the i_size, otherwise we could
4873 * expose stale data.
4875 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4879 if (size <= hole_start)
4882 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4884 cur_offset = hole_start;
4886 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4887 block_end - cur_offset);
4893 last_byte = min(extent_map_end(em), block_end);
4894 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4895 hole_size = last_byte - cur_offset;
4897 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4898 struct extent_map *hole_em;
4900 err = maybe_insert_hole(inode, cur_offset, hole_size);
4904 err = btrfs_inode_set_file_extent_range(inode,
4905 cur_offset, hole_size);
4909 hole_em = alloc_extent_map();
4911 btrfs_drop_extent_map_range(inode, cur_offset,
4912 cur_offset + hole_size - 1,
4914 btrfs_set_inode_full_sync(inode);
4917 hole_em->start = cur_offset;
4918 hole_em->len = hole_size;
4919 hole_em->orig_start = cur_offset;
4921 hole_em->block_start = EXTENT_MAP_HOLE;
4922 hole_em->block_len = 0;
4923 hole_em->orig_block_len = 0;
4924 hole_em->ram_bytes = hole_size;
4925 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4926 hole_em->generation = btrfs_get_fs_generation(fs_info);
4928 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4929 free_extent_map(hole_em);
4931 err = btrfs_inode_set_file_extent_range(inode,
4932 cur_offset, hole_size);
4937 free_extent_map(em);
4939 cur_offset = last_byte;
4940 if (cur_offset >= block_end)
4943 free_extent_map(em);
4944 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4948 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4950 struct btrfs_root *root = BTRFS_I(inode)->root;
4951 struct btrfs_trans_handle *trans;
4952 loff_t oldsize = i_size_read(inode);
4953 loff_t newsize = attr->ia_size;
4954 int mask = attr->ia_valid;
4958 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4959 * special case where we need to update the times despite not having
4960 * these flags set. For all other operations the VFS set these flags
4961 * explicitly if it wants a timestamp update.
4963 if (newsize != oldsize) {
4964 inode_inc_iversion(inode);
4965 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
4966 inode->i_mtime = inode_set_ctime_current(inode);
4970 if (newsize > oldsize) {
4972 * Don't do an expanding truncate while snapshotting is ongoing.
4973 * This is to ensure the snapshot captures a fully consistent
4974 * state of this file - if the snapshot captures this expanding
4975 * truncation, it must capture all writes that happened before
4978 btrfs_drew_write_lock(&root->snapshot_lock);
4979 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4981 btrfs_drew_write_unlock(&root->snapshot_lock);
4985 trans = btrfs_start_transaction(root, 1);
4986 if (IS_ERR(trans)) {
4987 btrfs_drew_write_unlock(&root->snapshot_lock);
4988 return PTR_ERR(trans);
4991 i_size_write(inode, newsize);
4992 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
4993 pagecache_isize_extended(inode, oldsize, newsize);
4994 ret = btrfs_update_inode(trans, BTRFS_I(inode));
4995 btrfs_drew_write_unlock(&root->snapshot_lock);
4996 btrfs_end_transaction(trans);
4998 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5000 if (btrfs_is_zoned(fs_info)) {
5001 ret = btrfs_wait_ordered_range(inode,
5002 ALIGN(newsize, fs_info->sectorsize),
5009 * We're truncating a file that used to have good data down to
5010 * zero. Make sure any new writes to the file get on disk
5014 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5015 &BTRFS_I(inode)->runtime_flags);
5017 truncate_setsize(inode, newsize);
5019 inode_dio_wait(inode);
5021 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5022 if (ret && inode->i_nlink) {
5026 * Truncate failed, so fix up the in-memory size. We
5027 * adjusted disk_i_size down as we removed extents, so
5028 * wait for disk_i_size to be stable and then update the
5029 * in-memory size to match.
5031 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5034 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5041 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5044 struct inode *inode = d_inode(dentry);
5045 struct btrfs_root *root = BTRFS_I(inode)->root;
5048 if (btrfs_root_readonly(root))
5051 err = setattr_prepare(idmap, dentry, attr);
5055 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5056 err = btrfs_setsize(inode, attr);
5061 if (attr->ia_valid) {
5062 setattr_copy(idmap, inode, attr);
5063 inode_inc_iversion(inode);
5064 err = btrfs_dirty_inode(BTRFS_I(inode));
5066 if (!err && attr->ia_valid & ATTR_MODE)
5067 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5074 * While truncating the inode pages during eviction, we get the VFS
5075 * calling btrfs_invalidate_folio() against each folio of the inode. This
5076 * is slow because the calls to btrfs_invalidate_folio() result in a
5077 * huge amount of calls to lock_extent() and clear_extent_bit(),
5078 * which keep merging and splitting extent_state structures over and over,
5079 * wasting lots of time.
5081 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5082 * skip all those expensive operations on a per folio basis and do only
5083 * the ordered io finishing, while we release here the extent_map and
5084 * extent_state structures, without the excessive merging and splitting.
5086 static void evict_inode_truncate_pages(struct inode *inode)
5088 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5089 struct rb_node *node;
5091 ASSERT(inode->i_state & I_FREEING);
5092 truncate_inode_pages_final(&inode->i_data);
5094 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5097 * Keep looping until we have no more ranges in the io tree.
5098 * We can have ongoing bios started by readahead that have
5099 * their endio callback (extent_io.c:end_bio_extent_readpage)
5100 * still in progress (unlocked the pages in the bio but did not yet
5101 * unlocked the ranges in the io tree). Therefore this means some
5102 * ranges can still be locked and eviction started because before
5103 * submitting those bios, which are executed by a separate task (work
5104 * queue kthread), inode references (inode->i_count) were not taken
5105 * (which would be dropped in the end io callback of each bio).
5106 * Therefore here we effectively end up waiting for those bios and
5107 * anyone else holding locked ranges without having bumped the inode's
5108 * reference count - if we don't do it, when they access the inode's
5109 * io_tree to unlock a range it may be too late, leading to an
5110 * use-after-free issue.
5112 spin_lock(&io_tree->lock);
5113 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5114 struct extent_state *state;
5115 struct extent_state *cached_state = NULL;
5118 unsigned state_flags;
5120 node = rb_first(&io_tree->state);
5121 state = rb_entry(node, struct extent_state, rb_node);
5122 start = state->start;
5124 state_flags = state->state;
5125 spin_unlock(&io_tree->lock);
5127 lock_extent(io_tree, start, end, &cached_state);
5130 * If still has DELALLOC flag, the extent didn't reach disk,
5131 * and its reserved space won't be freed by delayed_ref.
5132 * So we need to free its reserved space here.
5133 * (Refer to comment in btrfs_invalidate_folio, case 2)
5135 * Note, end is the bytenr of last byte, so we need + 1 here.
5137 if (state_flags & EXTENT_DELALLOC)
5138 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5141 clear_extent_bit(io_tree, start, end,
5142 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5146 spin_lock(&io_tree->lock);
5148 spin_unlock(&io_tree->lock);
5151 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5152 struct btrfs_block_rsv *rsv)
5154 struct btrfs_fs_info *fs_info = root->fs_info;
5155 struct btrfs_trans_handle *trans;
5156 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5160 * Eviction should be taking place at some place safe because of our
5161 * delayed iputs. However the normal flushing code will run delayed
5162 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5164 * We reserve the delayed_refs_extra here again because we can't use
5165 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5166 * above. We reserve our extra bit here because we generate a ton of
5167 * delayed refs activity by truncating.
5169 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5170 * if we fail to make this reservation we can re-try without the
5171 * delayed_refs_extra so we can make some forward progress.
5173 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5174 BTRFS_RESERVE_FLUSH_EVICT);
5176 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5177 BTRFS_RESERVE_FLUSH_EVICT);
5180 "could not allocate space for delete; will truncate on mount");
5181 return ERR_PTR(-ENOSPC);
5183 delayed_refs_extra = 0;
5186 trans = btrfs_join_transaction(root);
5190 if (delayed_refs_extra) {
5191 trans->block_rsv = &fs_info->trans_block_rsv;
5192 trans->bytes_reserved = delayed_refs_extra;
5193 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5194 delayed_refs_extra, true);
5199 void btrfs_evict_inode(struct inode *inode)
5201 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5202 struct btrfs_trans_handle *trans;
5203 struct btrfs_root *root = BTRFS_I(inode)->root;
5204 struct btrfs_block_rsv *rsv = NULL;
5207 trace_btrfs_inode_evict(inode);
5210 fsverity_cleanup_inode(inode);
5215 evict_inode_truncate_pages(inode);
5217 if (inode->i_nlink &&
5218 ((btrfs_root_refs(&root->root_item) != 0 &&
5219 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5220 btrfs_is_free_space_inode(BTRFS_I(inode))))
5223 if (is_bad_inode(inode))
5226 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5229 if (inode->i_nlink > 0) {
5230 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5231 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5236 * This makes sure the inode item in tree is uptodate and the space for
5237 * the inode update is released.
5239 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5244 * This drops any pending insert or delete operations we have for this
5245 * inode. We could have a delayed dir index deletion queued up, but
5246 * we're removing the inode completely so that'll be taken care of in
5249 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5251 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5254 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5255 rsv->failfast = true;
5257 btrfs_i_size_write(BTRFS_I(inode), 0);
5260 struct btrfs_truncate_control control = {
5261 .inode = BTRFS_I(inode),
5262 .ino = btrfs_ino(BTRFS_I(inode)),
5267 trans = evict_refill_and_join(root, rsv);
5271 trans->block_rsv = rsv;
5273 ret = btrfs_truncate_inode_items(trans, root, &control);
5274 trans->block_rsv = &fs_info->trans_block_rsv;
5275 btrfs_end_transaction(trans);
5277 * We have not added new delayed items for our inode after we
5278 * have flushed its delayed items, so no need to throttle on
5279 * delayed items. However we have modified extent buffers.
5281 btrfs_btree_balance_dirty_nodelay(fs_info);
5282 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5289 * Errors here aren't a big deal, it just means we leave orphan items in
5290 * the tree. They will be cleaned up on the next mount. If the inode
5291 * number gets reused, cleanup deletes the orphan item without doing
5292 * anything, and unlink reuses the existing orphan item.
5294 * If it turns out that we are dropping too many of these, we might want
5295 * to add a mechanism for retrying these after a commit.
5297 trans = evict_refill_and_join(root, rsv);
5298 if (!IS_ERR(trans)) {
5299 trans->block_rsv = rsv;
5300 btrfs_orphan_del(trans, BTRFS_I(inode));
5301 trans->block_rsv = &fs_info->trans_block_rsv;
5302 btrfs_end_transaction(trans);
5306 btrfs_free_block_rsv(fs_info, rsv);
5308 * If we didn't successfully delete, the orphan item will still be in
5309 * the tree and we'll retry on the next mount. Again, we might also want
5310 * to retry these periodically in the future.
5312 btrfs_remove_delayed_node(BTRFS_I(inode));
5313 fsverity_cleanup_inode(inode);
5318 * Return the key found in the dir entry in the location pointer, fill @type
5319 * with BTRFS_FT_*, and return 0.
5321 * If no dir entries were found, returns -ENOENT.
5322 * If found a corrupted location in dir entry, returns -EUCLEAN.
5324 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5325 struct btrfs_key *location, u8 *type)
5327 struct btrfs_dir_item *di;
5328 struct btrfs_path *path;
5329 struct btrfs_root *root = dir->root;
5331 struct fscrypt_name fname;
5333 path = btrfs_alloc_path();
5337 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5341 * fscrypt_setup_filename() should never return a positive value, but
5342 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5346 /* This needs to handle no-key deletions later on */
5348 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5349 &fname.disk_name, 0);
5350 if (IS_ERR_OR_NULL(di)) {
5351 ret = di ? PTR_ERR(di) : -ENOENT;
5355 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5356 if (location->type != BTRFS_INODE_ITEM_KEY &&
5357 location->type != BTRFS_ROOT_ITEM_KEY) {
5359 btrfs_warn(root->fs_info,
5360 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5361 __func__, fname.disk_name.name, btrfs_ino(dir),
5362 location->objectid, location->type, location->offset);
5365 *type = btrfs_dir_ftype(path->nodes[0], di);
5367 fscrypt_free_filename(&fname);
5368 btrfs_free_path(path);
5373 * when we hit a tree root in a directory, the btrfs part of the inode
5374 * needs to be changed to reflect the root directory of the tree root. This
5375 * is kind of like crossing a mount point.
5377 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5378 struct btrfs_inode *dir,
5379 struct dentry *dentry,
5380 struct btrfs_key *location,
5381 struct btrfs_root **sub_root)
5383 struct btrfs_path *path;
5384 struct btrfs_root *new_root;
5385 struct btrfs_root_ref *ref;
5386 struct extent_buffer *leaf;
5387 struct btrfs_key key;
5390 struct fscrypt_name fname;
5392 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5396 path = btrfs_alloc_path();
5403 key.objectid = dir->root->root_key.objectid;
5404 key.type = BTRFS_ROOT_REF_KEY;
5405 key.offset = location->objectid;
5407 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5414 leaf = path->nodes[0];
5415 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5416 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5417 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5420 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5421 (unsigned long)(ref + 1), fname.disk_name.len);
5425 btrfs_release_path(path);
5427 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5428 if (IS_ERR(new_root)) {
5429 err = PTR_ERR(new_root);
5433 *sub_root = new_root;
5434 location->objectid = btrfs_root_dirid(&new_root->root_item);
5435 location->type = BTRFS_INODE_ITEM_KEY;
5436 location->offset = 0;
5439 btrfs_free_path(path);
5440 fscrypt_free_filename(&fname);
5444 static void inode_tree_add(struct btrfs_inode *inode)
5446 struct btrfs_root *root = inode->root;
5447 struct btrfs_inode *entry;
5449 struct rb_node *parent;
5450 struct rb_node *new = &inode->rb_node;
5451 u64 ino = btrfs_ino(inode);
5453 if (inode_unhashed(&inode->vfs_inode))
5456 spin_lock(&root->inode_lock);
5457 p = &root->inode_tree.rb_node;
5460 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5462 if (ino < btrfs_ino(entry))
5463 p = &parent->rb_left;
5464 else if (ino > btrfs_ino(entry))
5465 p = &parent->rb_right;
5467 WARN_ON(!(entry->vfs_inode.i_state &
5468 (I_WILL_FREE | I_FREEING)));
5469 rb_replace_node(parent, new, &root->inode_tree);
5470 RB_CLEAR_NODE(parent);
5471 spin_unlock(&root->inode_lock);
5475 rb_link_node(new, parent, p);
5476 rb_insert_color(new, &root->inode_tree);
5477 spin_unlock(&root->inode_lock);
5480 static void inode_tree_del(struct btrfs_inode *inode)
5482 struct btrfs_root *root = inode->root;
5485 spin_lock(&root->inode_lock);
5486 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5487 rb_erase(&inode->rb_node, &root->inode_tree);
5488 RB_CLEAR_NODE(&inode->rb_node);
5489 empty = RB_EMPTY_ROOT(&root->inode_tree);
5491 spin_unlock(&root->inode_lock);
5493 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5494 spin_lock(&root->inode_lock);
5495 empty = RB_EMPTY_ROOT(&root->inode_tree);
5496 spin_unlock(&root->inode_lock);
5498 btrfs_add_dead_root(root);
5503 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5505 struct btrfs_iget_args *args = p;
5507 inode->i_ino = args->ino;
5508 BTRFS_I(inode)->location.objectid = args->ino;
5509 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5510 BTRFS_I(inode)->location.offset = 0;
5511 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5512 BUG_ON(args->root && !BTRFS_I(inode)->root);
5514 if (args->root && args->root == args->root->fs_info->tree_root &&
5515 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5516 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5517 &BTRFS_I(inode)->runtime_flags);
5521 static int btrfs_find_actor(struct inode *inode, void *opaque)
5523 struct btrfs_iget_args *args = opaque;
5525 return args->ino == BTRFS_I(inode)->location.objectid &&
5526 args->root == BTRFS_I(inode)->root;
5529 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5530 struct btrfs_root *root)
5532 struct inode *inode;
5533 struct btrfs_iget_args args;
5534 unsigned long hashval = btrfs_inode_hash(ino, root);
5539 inode = iget5_locked(s, hashval, btrfs_find_actor,
5540 btrfs_init_locked_inode,
5546 * Get an inode object given its inode number and corresponding root.
5547 * Path can be preallocated to prevent recursing back to iget through
5548 * allocator. NULL is also valid but may require an additional allocation
5551 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5552 struct btrfs_root *root, struct btrfs_path *path)
5554 struct inode *inode;
5556 inode = btrfs_iget_locked(s, ino, root);
5558 return ERR_PTR(-ENOMEM);
5560 if (inode->i_state & I_NEW) {
5563 ret = btrfs_read_locked_inode(inode, path);
5565 inode_tree_add(BTRFS_I(inode));
5566 unlock_new_inode(inode);
5570 * ret > 0 can come from btrfs_search_slot called by
5571 * btrfs_read_locked_inode, this means the inode item
5576 inode = ERR_PTR(ret);
5583 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5585 return btrfs_iget_path(s, ino, root, NULL);
5588 static struct inode *new_simple_dir(struct inode *dir,
5589 struct btrfs_key *key,
5590 struct btrfs_root *root)
5592 struct inode *inode = new_inode(dir->i_sb);
5595 return ERR_PTR(-ENOMEM);
5597 BTRFS_I(inode)->root = btrfs_grab_root(root);
5598 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5599 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5601 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5603 * We only need lookup, the rest is read-only and there's no inode
5604 * associated with the dentry
5606 inode->i_op = &simple_dir_inode_operations;
5607 inode->i_opflags &= ~IOP_XATTR;
5608 inode->i_fop = &simple_dir_operations;
5609 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5610 inode->i_mtime = inode_set_ctime_current(inode);
5611 inode->i_atime = dir->i_atime;
5612 BTRFS_I(inode)->i_otime = inode->i_mtime;
5613 inode->i_uid = dir->i_uid;
5614 inode->i_gid = dir->i_gid;
5619 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5620 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5621 static_assert(BTRFS_FT_DIR == FT_DIR);
5622 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5623 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5624 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5625 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5626 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5628 static inline u8 btrfs_inode_type(struct inode *inode)
5630 return fs_umode_to_ftype(inode->i_mode);
5633 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5635 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5636 struct inode *inode;
5637 struct btrfs_root *root = BTRFS_I(dir)->root;
5638 struct btrfs_root *sub_root = root;
5639 struct btrfs_key location;
5643 if (dentry->d_name.len > BTRFS_NAME_LEN)
5644 return ERR_PTR(-ENAMETOOLONG);
5646 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5648 return ERR_PTR(ret);
5650 if (location.type == BTRFS_INODE_ITEM_KEY) {
5651 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5655 /* Do extra check against inode mode with di_type */
5656 if (btrfs_inode_type(inode) != di_type) {
5658 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5659 inode->i_mode, btrfs_inode_type(inode),
5662 return ERR_PTR(-EUCLEAN);
5667 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5668 &location, &sub_root);
5671 inode = ERR_PTR(ret);
5673 inode = new_simple_dir(dir, &location, root);
5675 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5676 btrfs_put_root(sub_root);
5681 down_read(&fs_info->cleanup_work_sem);
5682 if (!sb_rdonly(inode->i_sb))
5683 ret = btrfs_orphan_cleanup(sub_root);
5684 up_read(&fs_info->cleanup_work_sem);
5687 inode = ERR_PTR(ret);
5694 static int btrfs_dentry_delete(const struct dentry *dentry)
5696 struct btrfs_root *root;
5697 struct inode *inode = d_inode(dentry);
5699 if (!inode && !IS_ROOT(dentry))
5700 inode = d_inode(dentry->d_parent);
5703 root = BTRFS_I(inode)->root;
5704 if (btrfs_root_refs(&root->root_item) == 0)
5707 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5713 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5716 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5718 if (inode == ERR_PTR(-ENOENT))
5720 return d_splice_alias(inode, dentry);
5724 * Find the highest existing sequence number in a directory and then set the
5725 * in-memory index_cnt variable to the first free sequence number.
5727 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5729 struct btrfs_root *root = inode->root;
5730 struct btrfs_key key, found_key;
5731 struct btrfs_path *path;
5732 struct extent_buffer *leaf;
5735 key.objectid = btrfs_ino(inode);
5736 key.type = BTRFS_DIR_INDEX_KEY;
5737 key.offset = (u64)-1;
5739 path = btrfs_alloc_path();
5743 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5746 /* FIXME: we should be able to handle this */
5751 if (path->slots[0] == 0) {
5752 inode->index_cnt = BTRFS_DIR_START_INDEX;
5758 leaf = path->nodes[0];
5759 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5761 if (found_key.objectid != btrfs_ino(inode) ||
5762 found_key.type != BTRFS_DIR_INDEX_KEY) {
5763 inode->index_cnt = BTRFS_DIR_START_INDEX;
5767 inode->index_cnt = found_key.offset + 1;
5769 btrfs_free_path(path);
5773 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5777 btrfs_inode_lock(dir, 0);
5778 if (dir->index_cnt == (u64)-1) {
5779 ret = btrfs_inode_delayed_dir_index_count(dir);
5781 ret = btrfs_set_inode_index_count(dir);
5787 /* index_cnt is the index number of next new entry, so decrement it. */
5788 *index = dir->index_cnt - 1;
5790 btrfs_inode_unlock(dir, 0);
5796 * All this infrastructure exists because dir_emit can fault, and we are holding
5797 * the tree lock when doing readdir. For now just allocate a buffer and copy
5798 * our information into that, and then dir_emit from the buffer. This is
5799 * similar to what NFS does, only we don't keep the buffer around in pagecache
5800 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5801 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5804 static int btrfs_opendir(struct inode *inode, struct file *file)
5806 struct btrfs_file_private *private;
5810 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5814 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5817 private->last_index = last_index;
5818 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5819 if (!private->filldir_buf) {
5823 file->private_data = private;
5827 static loff_t btrfs_dir_llseek(struct file *file, loff_t offset, int whence)
5829 struct btrfs_file_private *private = file->private_data;
5832 ret = btrfs_get_dir_last_index(BTRFS_I(file_inode(file)),
5833 &private->last_index);
5837 return generic_file_llseek(file, offset, whence);
5847 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5850 struct dir_entry *entry = addr;
5851 char *name = (char *)(entry + 1);
5853 ctx->pos = get_unaligned(&entry->offset);
5854 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5855 get_unaligned(&entry->ino),
5856 get_unaligned(&entry->type)))
5858 addr += sizeof(struct dir_entry) +
5859 get_unaligned(&entry->name_len);
5865 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5867 struct inode *inode = file_inode(file);
5868 struct btrfs_root *root = BTRFS_I(inode)->root;
5869 struct btrfs_file_private *private = file->private_data;
5870 struct btrfs_dir_item *di;
5871 struct btrfs_key key;
5872 struct btrfs_key found_key;
5873 struct btrfs_path *path;
5875 LIST_HEAD(ins_list);
5876 LIST_HEAD(del_list);
5883 struct btrfs_key location;
5885 if (!dir_emit_dots(file, ctx))
5888 path = btrfs_alloc_path();
5892 addr = private->filldir_buf;
5893 path->reada = READA_FORWARD;
5895 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5896 &ins_list, &del_list);
5899 key.type = BTRFS_DIR_INDEX_KEY;
5900 key.offset = ctx->pos;
5901 key.objectid = btrfs_ino(BTRFS_I(inode));
5903 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5904 struct dir_entry *entry;
5905 struct extent_buffer *leaf = path->nodes[0];
5908 if (found_key.objectid != key.objectid)
5910 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5912 if (found_key.offset < ctx->pos)
5914 if (found_key.offset > private->last_index)
5916 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5918 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5919 name_len = btrfs_dir_name_len(leaf, di);
5920 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5922 btrfs_release_path(path);
5923 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5926 addr = private->filldir_buf;
5932 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5934 name_ptr = (char *)(entry + 1);
5935 read_extent_buffer(leaf, name_ptr,
5936 (unsigned long)(di + 1), name_len);
5937 put_unaligned(name_len, &entry->name_len);
5938 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5939 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5940 put_unaligned(location.objectid, &entry->ino);
5941 put_unaligned(found_key.offset, &entry->offset);
5943 addr += sizeof(struct dir_entry) + name_len;
5944 total_len += sizeof(struct dir_entry) + name_len;
5946 /* Catch error encountered during iteration */
5950 btrfs_release_path(path);
5952 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5956 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5961 * Stop new entries from being returned after we return the last
5964 * New directory entries are assigned a strictly increasing
5965 * offset. This means that new entries created during readdir
5966 * are *guaranteed* to be seen in the future by that readdir.
5967 * This has broken buggy programs which operate on names as
5968 * they're returned by readdir. Until we re-use freed offsets
5969 * we have this hack to stop new entries from being returned
5970 * under the assumption that they'll never reach this huge
5973 * This is being careful not to overflow 32bit loff_t unless the
5974 * last entry requires it because doing so has broken 32bit apps
5977 if (ctx->pos >= INT_MAX)
5978 ctx->pos = LLONG_MAX;
5985 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5986 btrfs_free_path(path);
5991 * This is somewhat expensive, updating the tree every time the
5992 * inode changes. But, it is most likely to find the inode in cache.
5993 * FIXME, needs more benchmarking...there are no reasons other than performance
5994 * to keep or drop this code.
5996 static int btrfs_dirty_inode(struct btrfs_inode *inode)
5998 struct btrfs_root *root = inode->root;
5999 struct btrfs_fs_info *fs_info = root->fs_info;
6000 struct btrfs_trans_handle *trans;
6003 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6006 trans = btrfs_join_transaction(root);
6008 return PTR_ERR(trans);
6010 ret = btrfs_update_inode(trans, inode);
6011 if (ret == -ENOSPC || ret == -EDQUOT) {
6012 /* whoops, lets try again with the full transaction */
6013 btrfs_end_transaction(trans);
6014 trans = btrfs_start_transaction(root, 1);
6016 return PTR_ERR(trans);
6018 ret = btrfs_update_inode(trans, inode);
6020 btrfs_end_transaction(trans);
6021 if (inode->delayed_node)
6022 btrfs_balance_delayed_items(fs_info);
6028 * This is a copy of file_update_time. We need this so we can return error on
6029 * ENOSPC for updating the inode in the case of file write and mmap writes.
6031 static int btrfs_update_time(struct inode *inode, int flags)
6033 struct btrfs_root *root = BTRFS_I(inode)->root;
6034 bool dirty = flags & ~S_VERSION;
6036 if (btrfs_root_readonly(root))
6039 dirty = inode_update_timestamps(inode, flags);
6040 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6044 * helper to find a free sequence number in a given directory. This current
6045 * code is very simple, later versions will do smarter things in the btree
6047 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6051 if (dir->index_cnt == (u64)-1) {
6052 ret = btrfs_inode_delayed_dir_index_count(dir);
6054 ret = btrfs_set_inode_index_count(dir);
6060 *index = dir->index_cnt;
6066 static int btrfs_insert_inode_locked(struct inode *inode)
6068 struct btrfs_iget_args args;
6070 args.ino = BTRFS_I(inode)->location.objectid;
6071 args.root = BTRFS_I(inode)->root;
6073 return insert_inode_locked4(inode,
6074 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6075 btrfs_find_actor, &args);
6078 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6079 unsigned int *trans_num_items)
6081 struct inode *dir = args->dir;
6082 struct inode *inode = args->inode;
6085 if (!args->orphan) {
6086 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6092 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6094 fscrypt_free_filename(&args->fname);
6098 /* 1 to add inode item */
6099 *trans_num_items = 1;
6100 /* 1 to add compression property */
6101 if (BTRFS_I(dir)->prop_compress)
6102 (*trans_num_items)++;
6103 /* 1 to add default ACL xattr */
6104 if (args->default_acl)
6105 (*trans_num_items)++;
6106 /* 1 to add access ACL xattr */
6108 (*trans_num_items)++;
6109 #ifdef CONFIG_SECURITY
6110 /* 1 to add LSM xattr */
6111 if (dir->i_security)
6112 (*trans_num_items)++;
6115 /* 1 to add orphan item */
6116 (*trans_num_items)++;
6120 * 1 to add dir index
6121 * 1 to update parent inode item
6123 * No need for 1 unit for the inode ref item because it is
6124 * inserted in a batch together with the inode item at
6125 * btrfs_create_new_inode().
6127 *trans_num_items += 3;
6132 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6134 posix_acl_release(args->acl);
6135 posix_acl_release(args->default_acl);
6136 fscrypt_free_filename(&args->fname);
6140 * Inherit flags from the parent inode.
6142 * Currently only the compression flags and the cow flags are inherited.
6144 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6150 if (flags & BTRFS_INODE_NOCOMPRESS) {
6151 inode->flags &= ~BTRFS_INODE_COMPRESS;
6152 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6153 } else if (flags & BTRFS_INODE_COMPRESS) {
6154 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6155 inode->flags |= BTRFS_INODE_COMPRESS;
6158 if (flags & BTRFS_INODE_NODATACOW) {
6159 inode->flags |= BTRFS_INODE_NODATACOW;
6160 if (S_ISREG(inode->vfs_inode.i_mode))
6161 inode->flags |= BTRFS_INODE_NODATASUM;
6164 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6167 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6168 struct btrfs_new_inode_args *args)
6170 struct inode *dir = args->dir;
6171 struct inode *inode = args->inode;
6172 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6173 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6174 struct btrfs_root *root;
6175 struct btrfs_inode_item *inode_item;
6176 struct btrfs_key *location;
6177 struct btrfs_path *path;
6179 struct btrfs_inode_ref *ref;
6180 struct btrfs_key key[2];
6182 struct btrfs_item_batch batch;
6186 path = btrfs_alloc_path();
6191 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6192 root = BTRFS_I(inode)->root;
6194 ret = btrfs_get_free_objectid(root, &objectid);
6197 inode->i_ino = objectid;
6201 * O_TMPFILE, set link count to 0, so that after this point, we
6202 * fill in an inode item with the correct link count.
6204 set_nlink(inode, 0);
6206 trace_btrfs_inode_request(dir);
6208 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6212 /* index_cnt is ignored for everything but a dir. */
6213 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6214 BTRFS_I(inode)->generation = trans->transid;
6215 inode->i_generation = BTRFS_I(inode)->generation;
6218 * Subvolumes don't inherit flags from their parent directory.
6219 * Originally this was probably by accident, but we probably can't
6220 * change it now without compatibility issues.
6223 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6225 if (S_ISREG(inode->i_mode)) {
6226 if (btrfs_test_opt(fs_info, NODATASUM))
6227 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6228 if (btrfs_test_opt(fs_info, NODATACOW))
6229 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6230 BTRFS_INODE_NODATASUM;
6233 location = &BTRFS_I(inode)->location;
6234 location->objectid = objectid;
6235 location->offset = 0;
6236 location->type = BTRFS_INODE_ITEM_KEY;
6238 ret = btrfs_insert_inode_locked(inode);
6241 BTRFS_I(dir)->index_cnt--;
6246 * We could have gotten an inode number from somebody who was fsynced
6247 * and then removed in this same transaction, so let's just set full
6248 * sync since it will be a full sync anyway and this will blow away the
6249 * old info in the log.
6251 btrfs_set_inode_full_sync(BTRFS_I(inode));
6253 key[0].objectid = objectid;
6254 key[0].type = BTRFS_INODE_ITEM_KEY;
6257 sizes[0] = sizeof(struct btrfs_inode_item);
6259 if (!args->orphan) {
6261 * Start new inodes with an inode_ref. This is slightly more
6262 * efficient for small numbers of hard links since they will
6263 * be packed into one item. Extended refs will kick in if we
6264 * add more hard links than can fit in the ref item.
6266 key[1].objectid = objectid;
6267 key[1].type = BTRFS_INODE_REF_KEY;
6269 key[1].offset = objectid;
6270 sizes[1] = 2 + sizeof(*ref);
6272 key[1].offset = btrfs_ino(BTRFS_I(dir));
6273 sizes[1] = name->len + sizeof(*ref);
6277 batch.keys = &key[0];
6278 batch.data_sizes = &sizes[0];
6279 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6280 batch.nr = args->orphan ? 1 : 2;
6281 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6283 btrfs_abort_transaction(trans, ret);
6287 inode->i_mtime = inode_set_ctime_current(inode);
6288 inode->i_atime = inode->i_mtime;
6289 BTRFS_I(inode)->i_otime = inode->i_mtime;
6292 * We're going to fill the inode item now, so at this point the inode
6293 * must be fully initialized.
6296 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6297 struct btrfs_inode_item);
6298 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6299 sizeof(*inode_item));
6300 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6302 if (!args->orphan) {
6303 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6304 struct btrfs_inode_ref);
6305 ptr = (unsigned long)(ref + 1);
6307 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6308 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6309 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6311 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6313 btrfs_set_inode_ref_index(path->nodes[0], ref,
6314 BTRFS_I(inode)->dir_index);
6315 write_extent_buffer(path->nodes[0], name->name, ptr,
6320 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
6322 * We don't need the path anymore, plus inheriting properties, adding
6323 * ACLs, security xattrs, orphan item or adding the link, will result in
6324 * allocating yet another path. So just free our path.
6326 btrfs_free_path(path);
6330 struct inode *parent;
6333 * Subvolumes inherit properties from their parent subvolume,
6334 * not the directory they were created in.
6336 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6337 BTRFS_I(dir)->root);
6338 if (IS_ERR(parent)) {
6339 ret = PTR_ERR(parent);
6341 ret = btrfs_inode_inherit_props(trans, inode, parent);
6345 ret = btrfs_inode_inherit_props(trans, inode, dir);
6349 "error inheriting props for ino %llu (root %llu): %d",
6350 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6355 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6358 if (!args->subvol) {
6359 ret = btrfs_init_inode_security(trans, args);
6361 btrfs_abort_transaction(trans, ret);
6366 inode_tree_add(BTRFS_I(inode));
6368 trace_btrfs_inode_new(inode);
6369 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6371 btrfs_update_root_times(trans, root);
6374 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6376 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6377 0, BTRFS_I(inode)->dir_index);
6380 btrfs_abort_transaction(trans, ret);
6388 * discard_new_inode() calls iput(), but the caller owns the reference
6392 discard_new_inode(inode);
6394 btrfs_free_path(path);
6399 * utility function to add 'inode' into 'parent_inode' with
6400 * a give name and a given sequence number.
6401 * if 'add_backref' is true, also insert a backref from the
6402 * inode to the parent directory.
6404 int btrfs_add_link(struct btrfs_trans_handle *trans,
6405 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6406 const struct fscrypt_str *name, int add_backref, u64 index)
6409 struct btrfs_key key;
6410 struct btrfs_root *root = parent_inode->root;
6411 u64 ino = btrfs_ino(inode);
6412 u64 parent_ino = btrfs_ino(parent_inode);
6414 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6415 memcpy(&key, &inode->root->root_key, sizeof(key));
6418 key.type = BTRFS_INODE_ITEM_KEY;
6422 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6423 ret = btrfs_add_root_ref(trans, key.objectid,
6424 root->root_key.objectid, parent_ino,
6426 } else if (add_backref) {
6427 ret = btrfs_insert_inode_ref(trans, root, name,
6428 ino, parent_ino, index);
6431 /* Nothing to clean up yet */
6435 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6436 btrfs_inode_type(&inode->vfs_inode), index);
6437 if (ret == -EEXIST || ret == -EOVERFLOW)
6440 btrfs_abort_transaction(trans, ret);
6444 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6446 inode_inc_iversion(&parent_inode->vfs_inode);
6448 * If we are replaying a log tree, we do not want to update the mtime
6449 * and ctime of the parent directory with the current time, since the
6450 * log replay procedure is responsible for setting them to their correct
6451 * values (the ones it had when the fsync was done).
6453 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags))
6454 parent_inode->vfs_inode.i_mtime =
6455 inode_set_ctime_current(&parent_inode->vfs_inode);
6457 ret = btrfs_update_inode(trans, parent_inode);
6459 btrfs_abort_transaction(trans, ret);
6463 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6466 err = btrfs_del_root_ref(trans, key.objectid,
6467 root->root_key.objectid, parent_ino,
6468 &local_index, name);
6470 btrfs_abort_transaction(trans, err);
6471 } else if (add_backref) {
6475 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6478 btrfs_abort_transaction(trans, err);
6481 /* Return the original error code */
6485 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6486 struct inode *inode)
6488 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6489 struct btrfs_root *root = BTRFS_I(dir)->root;
6490 struct btrfs_new_inode_args new_inode_args = {
6495 unsigned int trans_num_items;
6496 struct btrfs_trans_handle *trans;
6499 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6503 trans = btrfs_start_transaction(root, trans_num_items);
6504 if (IS_ERR(trans)) {
6505 err = PTR_ERR(trans);
6506 goto out_new_inode_args;
6509 err = btrfs_create_new_inode(trans, &new_inode_args);
6511 d_instantiate_new(dentry, inode);
6513 btrfs_end_transaction(trans);
6514 btrfs_btree_balance_dirty(fs_info);
6516 btrfs_new_inode_args_destroy(&new_inode_args);
6523 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6524 struct dentry *dentry, umode_t mode, dev_t rdev)
6526 struct inode *inode;
6528 inode = new_inode(dir->i_sb);
6531 inode_init_owner(idmap, inode, dir, mode);
6532 inode->i_op = &btrfs_special_inode_operations;
6533 init_special_inode(inode, inode->i_mode, rdev);
6534 return btrfs_create_common(dir, dentry, inode);
6537 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6538 struct dentry *dentry, umode_t mode, bool excl)
6540 struct inode *inode;
6542 inode = new_inode(dir->i_sb);
6545 inode_init_owner(idmap, inode, dir, mode);
6546 inode->i_fop = &btrfs_file_operations;
6547 inode->i_op = &btrfs_file_inode_operations;
6548 inode->i_mapping->a_ops = &btrfs_aops;
6549 return btrfs_create_common(dir, dentry, inode);
6552 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6553 struct dentry *dentry)
6555 struct btrfs_trans_handle *trans = NULL;
6556 struct btrfs_root *root = BTRFS_I(dir)->root;
6557 struct inode *inode = d_inode(old_dentry);
6558 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6559 struct fscrypt_name fname;
6564 /* do not allow sys_link's with other subvols of the same device */
6565 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6568 if (inode->i_nlink >= BTRFS_LINK_MAX)
6571 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6575 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6580 * 2 items for inode and inode ref
6581 * 2 items for dir items
6582 * 1 item for parent inode
6583 * 1 item for orphan item deletion if O_TMPFILE
6585 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6586 if (IS_ERR(trans)) {
6587 err = PTR_ERR(trans);
6592 /* There are several dir indexes for this inode, clear the cache. */
6593 BTRFS_I(inode)->dir_index = 0ULL;
6595 inode_inc_iversion(inode);
6596 inode_set_ctime_current(inode);
6598 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6600 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6601 &fname.disk_name, 1, index);
6606 struct dentry *parent = dentry->d_parent;
6608 err = btrfs_update_inode(trans, BTRFS_I(inode));
6611 if (inode->i_nlink == 1) {
6613 * If new hard link count is 1, it's a file created
6614 * with open(2) O_TMPFILE flag.
6616 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6620 d_instantiate(dentry, inode);
6621 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6625 fscrypt_free_filename(&fname);
6627 btrfs_end_transaction(trans);
6629 inode_dec_link_count(inode);
6632 btrfs_btree_balance_dirty(fs_info);
6636 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6637 struct dentry *dentry, umode_t mode)
6639 struct inode *inode;
6641 inode = new_inode(dir->i_sb);
6644 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6645 inode->i_op = &btrfs_dir_inode_operations;
6646 inode->i_fop = &btrfs_dir_file_operations;
6647 return btrfs_create_common(dir, dentry, inode);
6650 static noinline int uncompress_inline(struct btrfs_path *path,
6652 struct btrfs_file_extent_item *item)
6655 struct extent_buffer *leaf = path->nodes[0];
6658 unsigned long inline_size;
6662 compress_type = btrfs_file_extent_compression(leaf, item);
6663 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6664 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6665 tmp = kmalloc(inline_size, GFP_NOFS);
6668 ptr = btrfs_file_extent_inline_start(item);
6670 read_extent_buffer(leaf, tmp, ptr, inline_size);
6672 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6673 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6676 * decompression code contains a memset to fill in any space between the end
6677 * of the uncompressed data and the end of max_size in case the decompressed
6678 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6679 * the end of an inline extent and the beginning of the next block, so we
6680 * cover that region here.
6683 if (max_size < PAGE_SIZE)
6684 memzero_page(page, max_size, PAGE_SIZE - max_size);
6689 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6692 struct btrfs_file_extent_item *fi;
6696 if (!page || PageUptodate(page))
6699 ASSERT(page_offset(page) == 0);
6701 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6702 struct btrfs_file_extent_item);
6703 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6704 return uncompress_inline(path, page, fi);
6706 copy_size = min_t(u64, PAGE_SIZE,
6707 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6708 kaddr = kmap_local_page(page);
6709 read_extent_buffer(path->nodes[0], kaddr,
6710 btrfs_file_extent_inline_start(fi), copy_size);
6711 kunmap_local(kaddr);
6712 if (copy_size < PAGE_SIZE)
6713 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6718 * Lookup the first extent overlapping a range in a file.
6720 * @inode: file to search in
6721 * @page: page to read extent data into if the extent is inline
6722 * @pg_offset: offset into @page to copy to
6723 * @start: file offset
6724 * @len: length of range starting at @start
6726 * Return the first &struct extent_map which overlaps the given range, reading
6727 * it from the B-tree and caching it if necessary. Note that there may be more
6728 * extents which overlap the given range after the returned extent_map.
6730 * If @page is not NULL and the extent is inline, this also reads the extent
6731 * data directly into the page and marks the extent up to date in the io_tree.
6733 * Return: ERR_PTR on error, non-NULL extent_map on success.
6735 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6736 struct page *page, size_t pg_offset,
6739 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6741 u64 extent_start = 0;
6743 u64 objectid = btrfs_ino(inode);
6744 int extent_type = -1;
6745 struct btrfs_path *path = NULL;
6746 struct btrfs_root *root = inode->root;
6747 struct btrfs_file_extent_item *item;
6748 struct extent_buffer *leaf;
6749 struct btrfs_key found_key;
6750 struct extent_map *em = NULL;
6751 struct extent_map_tree *em_tree = &inode->extent_tree;
6753 read_lock(&em_tree->lock);
6754 em = lookup_extent_mapping(em_tree, start, len);
6755 read_unlock(&em_tree->lock);
6758 if (em->start > start || em->start + em->len <= start)
6759 free_extent_map(em);
6760 else if (em->block_start == EXTENT_MAP_INLINE && page)
6761 free_extent_map(em);
6765 em = alloc_extent_map();
6770 em->start = EXTENT_MAP_HOLE;
6771 em->orig_start = EXTENT_MAP_HOLE;
6773 em->block_len = (u64)-1;
6775 path = btrfs_alloc_path();
6781 /* Chances are we'll be called again, so go ahead and do readahead */
6782 path->reada = READA_FORWARD;
6785 * The same explanation in load_free_space_cache applies here as well,
6786 * we only read when we're loading the free space cache, and at that
6787 * point the commit_root has everything we need.
6789 if (btrfs_is_free_space_inode(inode)) {
6790 path->search_commit_root = 1;
6791 path->skip_locking = 1;
6794 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6797 } else if (ret > 0) {
6798 if (path->slots[0] == 0)
6804 leaf = path->nodes[0];
6805 item = btrfs_item_ptr(leaf, path->slots[0],
6806 struct btrfs_file_extent_item);
6807 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6808 if (found_key.objectid != objectid ||
6809 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6811 * If we backup past the first extent we want to move forward
6812 * and see if there is an extent in front of us, otherwise we'll
6813 * say there is a hole for our whole search range which can
6820 extent_type = btrfs_file_extent_type(leaf, item);
6821 extent_start = found_key.offset;
6822 extent_end = btrfs_file_extent_end(path);
6823 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6824 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6825 /* Only regular file could have regular/prealloc extent */
6826 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6829 "regular/prealloc extent found for non-regular inode %llu",
6833 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6835 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6836 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6841 if (start >= extent_end) {
6843 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6844 ret = btrfs_next_leaf(root, path);
6850 leaf = path->nodes[0];
6852 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6853 if (found_key.objectid != objectid ||
6854 found_key.type != BTRFS_EXTENT_DATA_KEY)
6856 if (start + len <= found_key.offset)
6858 if (start > found_key.offset)
6861 /* New extent overlaps with existing one */
6863 em->orig_start = start;
6864 em->len = found_key.offset - start;
6865 em->block_start = EXTENT_MAP_HOLE;
6869 btrfs_extent_item_to_extent_map(inode, path, item, em);
6871 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6872 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6874 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6876 * Inline extent can only exist at file offset 0. This is
6877 * ensured by tree-checker and inline extent creation path.
6878 * Thus all members representing file offsets should be zero.
6880 ASSERT(pg_offset == 0);
6881 ASSERT(extent_start == 0);
6882 ASSERT(em->start == 0);
6885 * btrfs_extent_item_to_extent_map() should have properly
6886 * initialized em members already.
6888 * Other members are not utilized for inline extents.
6890 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6891 ASSERT(em->len == fs_info->sectorsize);
6893 ret = read_inline_extent(inode, path, page);
6900 em->orig_start = start;
6902 em->block_start = EXTENT_MAP_HOLE;
6905 btrfs_release_path(path);
6906 if (em->start > start || extent_map_end(em) <= start) {
6908 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6909 em->start, em->len, start, len);
6914 write_lock(&em_tree->lock);
6915 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6916 write_unlock(&em_tree->lock);
6918 btrfs_free_path(path);
6920 trace_btrfs_get_extent(root, inode, em);
6923 free_extent_map(em);
6924 return ERR_PTR(ret);
6929 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6930 struct btrfs_dio_data *dio_data,
6933 const u64 orig_start,
6934 const u64 block_start,
6935 const u64 block_len,
6936 const u64 orig_block_len,
6937 const u64 ram_bytes,
6940 struct extent_map *em = NULL;
6941 struct btrfs_ordered_extent *ordered;
6943 if (type != BTRFS_ORDERED_NOCOW) {
6944 em = create_io_em(inode, start, len, orig_start, block_start,
6945 block_len, orig_block_len, ram_bytes,
6946 BTRFS_COMPRESS_NONE, /* compress_type */
6951 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6952 block_start, block_len, 0,
6954 (1 << BTRFS_ORDERED_DIRECT),
6955 BTRFS_COMPRESS_NONE);
6956 if (IS_ERR(ordered)) {
6958 free_extent_map(em);
6959 btrfs_drop_extent_map_range(inode, start,
6960 start + len - 1, false);
6962 em = ERR_CAST(ordered);
6964 ASSERT(!dio_data->ordered);
6965 dio_data->ordered = ordered;
6972 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6973 struct btrfs_dio_data *dio_data,
6976 struct btrfs_root *root = inode->root;
6977 struct btrfs_fs_info *fs_info = root->fs_info;
6978 struct extent_map *em;
6979 struct btrfs_key ins;
6983 alloc_hint = get_extent_allocation_hint(inode, start, len);
6984 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6985 0, alloc_hint, &ins, 1, 1);
6987 return ERR_PTR(ret);
6989 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
6990 ins.objectid, ins.offset, ins.offset,
6991 ins.offset, BTRFS_ORDERED_REGULAR);
6992 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
6994 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7000 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7002 struct btrfs_block_group *block_group;
7003 bool readonly = false;
7005 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7006 if (!block_group || block_group->ro)
7009 btrfs_put_block_group(block_group);
7014 * Check if we can do nocow write into the range [@offset, @offset + @len)
7016 * @offset: File offset
7017 * @len: The length to write, will be updated to the nocow writeable
7019 * @orig_start: (optional) Return the original file offset of the file extent
7020 * @orig_len: (optional) Return the original on-disk length of the file extent
7021 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7022 * @strict: if true, omit optimizations that might force us into unnecessary
7023 * cow. e.g., don't trust generation number.
7026 * >0 and update @len if we can do nocow write
7027 * 0 if we can't do nocow write
7028 * <0 if error happened
7030 * NOTE: This only checks the file extents, caller is responsible to wait for
7031 * any ordered extents.
7033 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7034 u64 *orig_start, u64 *orig_block_len,
7035 u64 *ram_bytes, bool nowait, bool strict)
7037 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7038 struct can_nocow_file_extent_args nocow_args = { 0 };
7039 struct btrfs_path *path;
7041 struct extent_buffer *leaf;
7042 struct btrfs_root *root = BTRFS_I(inode)->root;
7043 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7044 struct btrfs_file_extent_item *fi;
7045 struct btrfs_key key;
7048 path = btrfs_alloc_path();
7051 path->nowait = nowait;
7053 ret = btrfs_lookup_file_extent(NULL, root, path,
7054 btrfs_ino(BTRFS_I(inode)), offset, 0);
7059 if (path->slots[0] == 0) {
7060 /* can't find the item, must cow */
7067 leaf = path->nodes[0];
7068 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7069 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7070 key.type != BTRFS_EXTENT_DATA_KEY) {
7071 /* not our file or wrong item type, must cow */
7075 if (key.offset > offset) {
7076 /* Wrong offset, must cow */
7080 if (btrfs_file_extent_end(path) <= offset)
7083 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7084 found_type = btrfs_file_extent_type(leaf, fi);
7086 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7088 nocow_args.start = offset;
7089 nocow_args.end = offset + *len - 1;
7090 nocow_args.strict = strict;
7091 nocow_args.free_path = true;
7093 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7094 /* can_nocow_file_extent() has freed the path. */
7098 /* Treat errors as not being able to NOCOW. */
7104 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7107 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7108 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7111 range_end = round_up(offset + nocow_args.num_bytes,
7112 root->fs_info->sectorsize) - 1;
7113 ret = test_range_bit_exists(io_tree, offset, range_end, EXTENT_DELALLOC);
7121 *orig_start = key.offset - nocow_args.extent_offset;
7123 *orig_block_len = nocow_args.disk_num_bytes;
7125 *len = nocow_args.num_bytes;
7128 btrfs_free_path(path);
7132 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7133 struct extent_state **cached_state,
7134 unsigned int iomap_flags)
7136 const bool writing = (iomap_flags & IOMAP_WRITE);
7137 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7138 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7139 struct btrfs_ordered_extent *ordered;
7144 if (!try_lock_extent(io_tree, lockstart, lockend,
7148 lock_extent(io_tree, lockstart, lockend, cached_state);
7151 * We're concerned with the entire range that we're going to be
7152 * doing DIO to, so we need to make sure there's no ordered
7153 * extents in this range.
7155 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7156 lockend - lockstart + 1);
7159 * We need to make sure there are no buffered pages in this
7160 * range either, we could have raced between the invalidate in
7161 * generic_file_direct_write and locking the extent. The
7162 * invalidate needs to happen so that reads after a write do not
7166 (!writing || !filemap_range_has_page(inode->i_mapping,
7167 lockstart, lockend)))
7170 unlock_extent(io_tree, lockstart, lockend, cached_state);
7174 btrfs_put_ordered_extent(ordered);
7179 * If we are doing a DIO read and the ordered extent we
7180 * found is for a buffered write, we can not wait for it
7181 * to complete and retry, because if we do so we can
7182 * deadlock with concurrent buffered writes on page
7183 * locks. This happens only if our DIO read covers more
7184 * than one extent map, if at this point has already
7185 * created an ordered extent for a previous extent map
7186 * and locked its range in the inode's io tree, and a
7187 * concurrent write against that previous extent map's
7188 * range and this range started (we unlock the ranges
7189 * in the io tree only when the bios complete and
7190 * buffered writes always lock pages before attempting
7191 * to lock range in the io tree).
7194 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7195 btrfs_start_ordered_extent(ordered);
7197 ret = nowait ? -EAGAIN : -ENOTBLK;
7198 btrfs_put_ordered_extent(ordered);
7201 * We could trigger writeback for this range (and wait
7202 * for it to complete) and then invalidate the pages for
7203 * this range (through invalidate_inode_pages2_range()),
7204 * but that can lead us to a deadlock with a concurrent
7205 * call to readahead (a buffered read or a defrag call
7206 * triggered a readahead) on a page lock due to an
7207 * ordered dio extent we created before but did not have
7208 * yet a corresponding bio submitted (whence it can not
7209 * complete), which makes readahead wait for that
7210 * ordered extent to complete while holding a lock on
7213 ret = nowait ? -EAGAIN : -ENOTBLK;
7225 /* The callers of this must take lock_extent() */
7226 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7227 u64 len, u64 orig_start, u64 block_start,
7228 u64 block_len, u64 orig_block_len,
7229 u64 ram_bytes, int compress_type,
7232 struct extent_map *em;
7235 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7236 type == BTRFS_ORDERED_COMPRESSED ||
7237 type == BTRFS_ORDERED_NOCOW ||
7238 type == BTRFS_ORDERED_REGULAR);
7240 em = alloc_extent_map();
7242 return ERR_PTR(-ENOMEM);
7245 em->orig_start = orig_start;
7247 em->block_len = block_len;
7248 em->block_start = block_start;
7249 em->orig_block_len = orig_block_len;
7250 em->ram_bytes = ram_bytes;
7251 em->generation = -1;
7252 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7253 if (type == BTRFS_ORDERED_PREALLOC) {
7254 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7255 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7256 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7257 em->compress_type = compress_type;
7260 ret = btrfs_replace_extent_map_range(inode, em, true);
7262 free_extent_map(em);
7263 return ERR_PTR(ret);
7266 /* em got 2 refs now, callers needs to do free_extent_map once. */
7271 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7272 struct inode *inode,
7273 struct btrfs_dio_data *dio_data,
7274 u64 start, u64 *lenp,
7275 unsigned int iomap_flags)
7277 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7278 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7279 struct extent_map *em = *map;
7281 u64 block_start, orig_start, orig_block_len, ram_bytes;
7282 struct btrfs_block_group *bg;
7283 bool can_nocow = false;
7284 bool space_reserved = false;
7290 * We don't allocate a new extent in the following cases
7292 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7294 * 2) The extent is marked as PREALLOC. We're good to go here and can
7295 * just use the extent.
7298 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7299 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7300 em->block_start != EXTENT_MAP_HOLE)) {
7301 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7302 type = BTRFS_ORDERED_PREALLOC;
7304 type = BTRFS_ORDERED_NOCOW;
7305 len = min(len, em->len - (start - em->start));
7306 block_start = em->block_start + (start - em->start);
7308 if (can_nocow_extent(inode, start, &len, &orig_start,
7309 &orig_block_len, &ram_bytes, false, false) == 1) {
7310 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7318 struct extent_map *em2;
7320 /* We can NOCOW, so only need to reserve metadata space. */
7321 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7324 /* Our caller expects us to free the input extent map. */
7325 free_extent_map(em);
7327 btrfs_dec_nocow_writers(bg);
7328 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7332 space_reserved = true;
7334 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7335 orig_start, block_start,
7336 len, orig_block_len,
7338 btrfs_dec_nocow_writers(bg);
7339 if (type == BTRFS_ORDERED_PREALLOC) {
7340 free_extent_map(em);
7350 dio_data->nocow_done = true;
7352 /* Our caller expects us to free the input extent map. */
7353 free_extent_map(em);
7362 * If we could not allocate data space before locking the file
7363 * range and we can't do a NOCOW write, then we have to fail.
7365 if (!dio_data->data_space_reserved) {
7371 * We have to COW and we have already reserved data space before,
7372 * so now we reserve only metadata.
7374 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7378 space_reserved = true;
7380 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7386 len = min(len, em->len - (start - em->start));
7388 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7389 prev_len - len, true);
7393 * We have created our ordered extent, so we can now release our reservation
7394 * for an outstanding extent.
7396 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7399 * Need to update the i_size under the extent lock so buffered
7400 * readers will get the updated i_size when we unlock.
7402 if (start + len > i_size_read(inode))
7403 i_size_write(inode, start + len);
7405 if (ret && space_reserved) {
7406 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7407 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7413 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7414 loff_t length, unsigned int flags, struct iomap *iomap,
7415 struct iomap *srcmap)
7417 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7418 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7419 struct extent_map *em;
7420 struct extent_state *cached_state = NULL;
7421 struct btrfs_dio_data *dio_data = iter->private;
7422 u64 lockstart, lockend;
7423 const bool write = !!(flags & IOMAP_WRITE);
7426 const u64 data_alloc_len = length;
7427 bool unlock_extents = false;
7430 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7431 * we're NOWAIT we may submit a bio for a partial range and return
7432 * EIOCBQUEUED, which would result in an errant short read.
7434 * The best way to handle this would be to allow for partial completions
7435 * of iocb's, so we could submit the partial bio, return and fault in
7436 * the rest of the pages, and then submit the io for the rest of the
7437 * range. However we don't have that currently, so simply return
7438 * -EAGAIN at this point so that the normal path is used.
7440 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7444 * Cap the size of reads to that usually seen in buffered I/O as we need
7445 * to allocate a contiguous array for the checksums.
7448 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7451 lockend = start + len - 1;
7454 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7455 * enough if we've written compressed pages to this area, so we need to
7456 * flush the dirty pages again to make absolutely sure that any
7457 * outstanding dirty pages are on disk - the first flush only starts
7458 * compression on the data, while keeping the pages locked, so by the
7459 * time the second flush returns we know bios for the compressed pages
7460 * were submitted and finished, and the pages no longer under writeback.
7462 * If we have a NOWAIT request and we have any pages in the range that
7463 * are locked, likely due to compression still in progress, we don't want
7464 * to block on page locks. We also don't want to block on pages marked as
7465 * dirty or under writeback (same as for the non-compression case).
7466 * iomap_dio_rw() did the same check, but after that and before we got
7467 * here, mmap'ed writes may have happened or buffered reads started
7468 * (readpage() and readahead(), which lock pages), as we haven't locked
7469 * the file range yet.
7471 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7472 &BTRFS_I(inode)->runtime_flags)) {
7473 if (flags & IOMAP_NOWAIT) {
7474 if (filemap_range_needs_writeback(inode->i_mapping,
7475 lockstart, lockend))
7478 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7479 start + length - 1);
7485 memset(dio_data, 0, sizeof(*dio_data));
7488 * We always try to allocate data space and must do it before locking
7489 * the file range, to avoid deadlocks with concurrent writes to the same
7490 * range if the range has several extents and the writes don't expand the
7491 * current i_size (the inode lock is taken in shared mode). If we fail to
7492 * allocate data space here we continue and later, after locking the
7493 * file range, we fail with ENOSPC only if we figure out we can not do a
7496 if (write && !(flags & IOMAP_NOWAIT)) {
7497 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7498 &dio_data->data_reserved,
7499 start, data_alloc_len, false);
7501 dio_data->data_space_reserved = true;
7502 else if (ret && !(BTRFS_I(inode)->flags &
7503 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7508 * If this errors out it's because we couldn't invalidate pagecache for
7509 * this range and we need to fallback to buffered IO, or we are doing a
7510 * NOWAIT read/write and we need to block.
7512 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7516 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7523 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7524 * io. INLINE is special, and we could probably kludge it in here, but
7525 * it's still buffered so for safety lets just fall back to the generic
7528 * For COMPRESSED we _have_ to read the entire extent in so we can
7529 * decompress it, so there will be buffering required no matter what we
7530 * do, so go ahead and fallback to buffered.
7532 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7533 * to buffered IO. Don't blame me, this is the price we pay for using
7536 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7537 em->block_start == EXTENT_MAP_INLINE) {
7538 free_extent_map(em);
7540 * If we are in a NOWAIT context, return -EAGAIN in order to
7541 * fallback to buffered IO. This is not only because we can
7542 * block with buffered IO (no support for NOWAIT semantics at
7543 * the moment) but also to avoid returning short reads to user
7544 * space - this happens if we were able to read some data from
7545 * previous non-compressed extents and then when we fallback to
7546 * buffered IO, at btrfs_file_read_iter() by calling
7547 * filemap_read(), we fail to fault in pages for the read buffer,
7548 * in which case filemap_read() returns a short read (the number
7549 * of bytes previously read is > 0, so it does not return -EFAULT).
7551 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7555 len = min(len, em->len - (start - em->start));
7558 * If we have a NOWAIT request and the range contains multiple extents
7559 * (or a mix of extents and holes), then we return -EAGAIN to make the
7560 * caller fallback to a context where it can do a blocking (without
7561 * NOWAIT) request. This way we avoid doing partial IO and returning
7562 * success to the caller, which is not optimal for writes and for reads
7563 * it can result in unexpected behaviour for an application.
7565 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7566 * iomap_dio_rw(), we can end up returning less data then what the caller
7567 * asked for, resulting in an unexpected, and incorrect, short read.
7568 * That is, the caller asked to read N bytes and we return less than that,
7569 * which is wrong unless we are crossing EOF. This happens if we get a
7570 * page fault error when trying to fault in pages for the buffer that is
7571 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7572 * have previously submitted bios for other extents in the range, in
7573 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7574 * those bios have completed by the time we get the page fault error,
7575 * which we return back to our caller - we should only return EIOCBQUEUED
7576 * after we have submitted bios for all the extents in the range.
7578 if ((flags & IOMAP_NOWAIT) && len < length) {
7579 free_extent_map(em);
7585 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7586 start, &len, flags);
7589 unlock_extents = true;
7590 /* Recalc len in case the new em is smaller than requested */
7591 len = min(len, em->len - (start - em->start));
7592 if (dio_data->data_space_reserved) {
7594 u64 release_len = 0;
7596 if (dio_data->nocow_done) {
7597 release_offset = start;
7598 release_len = data_alloc_len;
7599 } else if (len < data_alloc_len) {
7600 release_offset = start + len;
7601 release_len = data_alloc_len - len;
7604 if (release_len > 0)
7605 btrfs_free_reserved_data_space(BTRFS_I(inode),
7606 dio_data->data_reserved,
7612 * We need to unlock only the end area that we aren't using.
7613 * The rest is going to be unlocked by the endio routine.
7615 lockstart = start + len;
7616 if (lockstart < lockend)
7617 unlock_extents = true;
7621 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7624 free_extent_state(cached_state);
7627 * Translate extent map information to iomap.
7628 * We trim the extents (and move the addr) even though iomap code does
7629 * that, since we have locked only the parts we are performing I/O in.
7631 if ((em->block_start == EXTENT_MAP_HOLE) ||
7632 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7633 iomap->addr = IOMAP_NULL_ADDR;
7634 iomap->type = IOMAP_HOLE;
7636 iomap->addr = em->block_start + (start - em->start);
7637 iomap->type = IOMAP_MAPPED;
7639 iomap->offset = start;
7640 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7641 iomap->length = len;
7642 free_extent_map(em);
7647 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7650 if (dio_data->data_space_reserved) {
7651 btrfs_free_reserved_data_space(BTRFS_I(inode),
7652 dio_data->data_reserved,
7653 start, data_alloc_len);
7654 extent_changeset_free(dio_data->data_reserved);
7660 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7661 ssize_t written, unsigned int flags, struct iomap *iomap)
7663 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7664 struct btrfs_dio_data *dio_data = iter->private;
7665 size_t submitted = dio_data->submitted;
7666 const bool write = !!(flags & IOMAP_WRITE);
7669 if (!write && (iomap->type == IOMAP_HOLE)) {
7670 /* If reading from a hole, unlock and return */
7671 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7676 if (submitted < length) {
7678 length -= submitted;
7680 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7681 pos, length, false);
7683 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7684 pos + length - 1, NULL);
7688 btrfs_put_ordered_extent(dio_data->ordered);
7689 dio_data->ordered = NULL;
7693 extent_changeset_free(dio_data->data_reserved);
7697 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7699 struct btrfs_dio_private *dip =
7700 container_of(bbio, struct btrfs_dio_private, bbio);
7701 struct btrfs_inode *inode = bbio->inode;
7702 struct bio *bio = &bbio->bio;
7704 if (bio->bi_status) {
7705 btrfs_warn(inode->root->fs_info,
7706 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7707 btrfs_ino(inode), bio->bi_opf,
7708 dip->file_offset, dip->bytes, bio->bi_status);
7711 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7712 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7713 dip->file_offset, dip->bytes,
7716 unlock_extent(&inode->io_tree, dip->file_offset,
7717 dip->file_offset + dip->bytes - 1, NULL);
7720 bbio->bio.bi_private = bbio->private;
7721 iomap_dio_bio_end_io(bio);
7724 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7727 struct btrfs_bio *bbio = btrfs_bio(bio);
7728 struct btrfs_dio_private *dip =
7729 container_of(bbio, struct btrfs_dio_private, bbio);
7730 struct btrfs_dio_data *dio_data = iter->private;
7732 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7733 btrfs_dio_end_io, bio->bi_private);
7734 bbio->inode = BTRFS_I(iter->inode);
7735 bbio->file_offset = file_offset;
7737 dip->file_offset = file_offset;
7738 dip->bytes = bio->bi_iter.bi_size;
7740 dio_data->submitted += bio->bi_iter.bi_size;
7743 * Check if we are doing a partial write. If we are, we need to split
7744 * the ordered extent to match the submitted bio. Hang on to the
7745 * remaining unfinishable ordered_extent in dio_data so that it can be
7746 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7747 * remaining pages is blocked on the outstanding ordered extent.
7749 if (iter->flags & IOMAP_WRITE) {
7752 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7754 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7755 file_offset, dip->bytes,
7757 bio->bi_status = errno_to_blk_status(ret);
7758 iomap_dio_bio_end_io(bio);
7763 btrfs_submit_bio(bbio, 0);
7766 static const struct iomap_ops btrfs_dio_iomap_ops = {
7767 .iomap_begin = btrfs_dio_iomap_begin,
7768 .iomap_end = btrfs_dio_iomap_end,
7771 static const struct iomap_dio_ops btrfs_dio_ops = {
7772 .submit_io = btrfs_dio_submit_io,
7773 .bio_set = &btrfs_dio_bioset,
7776 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7778 struct btrfs_dio_data data = { 0 };
7780 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7781 IOMAP_DIO_PARTIAL, &data, done_before);
7784 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
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 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7798 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7803 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7804 * file range (0 to LLONG_MAX), but that is not enough if we have
7805 * compression enabled. The first filemap_fdatawrite_range() only kicks
7806 * in the compression of data (in an async thread) and will return
7807 * before the compression is done and writeback is started. A second
7808 * filemap_fdatawrite_range() is needed to wait for the compression to
7809 * complete and writeback to start. We also need to wait for ordered
7810 * extents to complete, because our fiemap implementation uses mainly
7811 * file extent items to list the extents, searching for extent maps
7812 * only for file ranges with holes or prealloc extents to figure out
7813 * if we have delalloc in those ranges.
7815 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7816 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7821 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7824 static int btrfs_writepages(struct address_space *mapping,
7825 struct writeback_control *wbc)
7827 return extent_writepages(mapping, wbc);
7830 static void btrfs_readahead(struct readahead_control *rac)
7832 extent_readahead(rac);
7836 * For release_folio() and invalidate_folio() we have a race window where
7837 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7838 * If we continue to release/invalidate the page, we could cause use-after-free
7839 * for subpage spinlock. So this function is to spin and wait for subpage
7842 static void wait_subpage_spinlock(struct page *page)
7844 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7845 struct btrfs_subpage *subpage;
7847 if (!btrfs_is_subpage(fs_info, page))
7850 ASSERT(PagePrivate(page) && page->private);
7851 subpage = (struct btrfs_subpage *)page->private;
7854 * This may look insane as we just acquire the spinlock and release it,
7855 * without doing anything. But we just want to make sure no one is
7856 * still holding the subpage spinlock.
7857 * And since the page is not dirty nor writeback, and we have page
7858 * locked, the only possible way to hold a spinlock is from the endio
7859 * function to clear page writeback.
7861 * Here we just acquire the spinlock so that all existing callers
7862 * should exit and we're safe to release/invalidate the page.
7864 spin_lock_irq(&subpage->lock);
7865 spin_unlock_irq(&subpage->lock);
7868 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7870 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7873 wait_subpage_spinlock(&folio->page);
7874 clear_page_extent_mapped(&folio->page);
7879 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7881 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7883 return __btrfs_release_folio(folio, gfp_flags);
7886 #ifdef CONFIG_MIGRATION
7887 static int btrfs_migrate_folio(struct address_space *mapping,
7888 struct folio *dst, struct folio *src,
7889 enum migrate_mode mode)
7891 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7893 if (ret != MIGRATEPAGE_SUCCESS)
7896 if (folio_test_ordered(src)) {
7897 folio_clear_ordered(src);
7898 folio_set_ordered(dst);
7901 return MIGRATEPAGE_SUCCESS;
7904 #define btrfs_migrate_folio NULL
7907 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7910 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7911 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7912 struct extent_io_tree *tree = &inode->io_tree;
7913 struct extent_state *cached_state = NULL;
7914 u64 page_start = folio_pos(folio);
7915 u64 page_end = page_start + folio_size(folio) - 1;
7917 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7920 * We have folio locked so no new ordered extent can be created on this
7921 * page, nor bio can be submitted for this folio.
7923 * But already submitted bio can still be finished on this folio.
7924 * Furthermore, endio function won't skip folio which has Ordered
7925 * (Private2) already cleared, so it's possible for endio and
7926 * invalidate_folio to do the same ordered extent accounting twice
7929 * So here we wait for any submitted bios to finish, so that we won't
7930 * do double ordered extent accounting on the same folio.
7932 folio_wait_writeback(folio);
7933 wait_subpage_spinlock(&folio->page);
7936 * For subpage case, we have call sites like
7937 * btrfs_punch_hole_lock_range() which passes range not aligned to
7939 * If the range doesn't cover the full folio, we don't need to and
7940 * shouldn't clear page extent mapped, as folio->private can still
7941 * record subpage dirty bits for other part of the range.
7943 * For cases that invalidate the full folio even the range doesn't
7944 * cover the full folio, like invalidating the last folio, we're
7945 * still safe to wait for ordered extent to finish.
7947 if (!(offset == 0 && length == folio_size(folio))) {
7948 btrfs_release_folio(folio, GFP_NOFS);
7952 if (!inode_evicting)
7953 lock_extent(tree, page_start, page_end, &cached_state);
7956 while (cur < page_end) {
7957 struct btrfs_ordered_extent *ordered;
7960 u32 extra_flags = 0;
7962 ordered = btrfs_lookup_first_ordered_range(inode, cur,
7963 page_end + 1 - cur);
7965 range_end = page_end;
7967 * No ordered extent covering this range, we are safe
7968 * to delete all extent states in the range.
7970 extra_flags = EXTENT_CLEAR_ALL_BITS;
7973 if (ordered->file_offset > cur) {
7975 * There is a range between [cur, oe->file_offset) not
7976 * covered by any ordered extent.
7977 * We are safe to delete all extent states, and handle
7978 * the ordered extent in the next iteration.
7980 range_end = ordered->file_offset - 1;
7981 extra_flags = EXTENT_CLEAR_ALL_BITS;
7985 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
7987 ASSERT(range_end + 1 - cur < U32_MAX);
7988 range_len = range_end + 1 - cur;
7989 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
7991 * If Ordered (Private2) is cleared, it means endio has
7992 * already been executed for the range.
7993 * We can't delete the extent states as
7994 * btrfs_finish_ordered_io() may still use some of them.
7998 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8001 * IO on this page will never be started, so we need to account
8002 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8003 * here, must leave that up for the ordered extent completion.
8005 * This will also unlock the range for incoming
8006 * btrfs_finish_ordered_io().
8008 if (!inode_evicting)
8009 clear_extent_bit(tree, cur, range_end,
8011 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8012 EXTENT_DEFRAG, &cached_state);
8014 spin_lock_irq(&inode->ordered_tree_lock);
8015 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8016 ordered->truncated_len = min(ordered->truncated_len,
8017 cur - ordered->file_offset);
8018 spin_unlock_irq(&inode->ordered_tree_lock);
8021 * If the ordered extent has finished, we're safe to delete all
8022 * the extent states of the range, otherwise
8023 * btrfs_finish_ordered_io() will get executed by endio for
8024 * other pages, so we can't delete extent states.
8026 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8027 cur, range_end + 1 - cur)) {
8028 btrfs_finish_ordered_io(ordered);
8030 * The ordered extent has finished, now we're again
8031 * safe to delete all extent states of the range.
8033 extra_flags = EXTENT_CLEAR_ALL_BITS;
8037 btrfs_put_ordered_extent(ordered);
8039 * Qgroup reserved space handler
8040 * Sector(s) here will be either:
8042 * 1) Already written to disk or bio already finished
8043 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8044 * Qgroup will be handled by its qgroup_record then.
8045 * btrfs_qgroup_free_data() call will do nothing here.
8047 * 2) Not written to disk yet
8048 * Then btrfs_qgroup_free_data() call will clear the
8049 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8050 * reserved data space.
8051 * Since the IO will never happen for this page.
8053 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8054 if (!inode_evicting) {
8055 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8056 EXTENT_DELALLOC | EXTENT_UPTODATE |
8057 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8058 extra_flags, &cached_state);
8060 cur = range_end + 1;
8063 * We have iterated through all ordered extents of the page, the page
8064 * should not have Ordered (Private2) anymore, or the above iteration
8065 * did something wrong.
8067 ASSERT(!folio_test_ordered(folio));
8068 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8069 if (!inode_evicting)
8070 __btrfs_release_folio(folio, GFP_NOFS);
8071 clear_page_extent_mapped(&folio->page);
8075 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8076 * called from a page fault handler when a page is first dirtied. Hence we must
8077 * be careful to check for EOF conditions here. We set the page up correctly
8078 * for a written page which means we get ENOSPC checking when writing into
8079 * holes and correct delalloc and unwritten extent mapping on filesystems that
8080 * support these features.
8082 * We are not allowed to take the i_mutex here so we have to play games to
8083 * protect against truncate races as the page could now be beyond EOF. Because
8084 * truncate_setsize() writes the inode size before removing pages, once we have
8085 * the page lock we can determine safely if the page is beyond EOF. If it is not
8086 * beyond EOF, then the page is guaranteed safe against truncation until we
8089 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8091 struct page *page = vmf->page;
8092 struct inode *inode = file_inode(vmf->vma->vm_file);
8093 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8094 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8095 struct btrfs_ordered_extent *ordered;
8096 struct extent_state *cached_state = NULL;
8097 struct extent_changeset *data_reserved = NULL;
8098 unsigned long zero_start;
8108 reserved_space = PAGE_SIZE;
8110 sb_start_pagefault(inode->i_sb);
8111 page_start = page_offset(page);
8112 page_end = page_start + PAGE_SIZE - 1;
8116 * Reserving delalloc space after obtaining the page lock can lead to
8117 * deadlock. For example, if a dirty page is locked by this function
8118 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8119 * dirty page write out, then the btrfs_writepages() function could
8120 * end up waiting indefinitely to get a lock on the page currently
8121 * being processed by btrfs_page_mkwrite() function.
8123 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8124 page_start, reserved_space);
8126 ret2 = file_update_time(vmf->vma->vm_file);
8130 ret = vmf_error(ret2);
8136 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8138 down_read(&BTRFS_I(inode)->i_mmap_lock);
8140 size = i_size_read(inode);
8142 if ((page->mapping != inode->i_mapping) ||
8143 (page_start >= size)) {
8144 /* page got truncated out from underneath us */
8147 wait_on_page_writeback(page);
8149 lock_extent(io_tree, page_start, page_end, &cached_state);
8150 ret2 = set_page_extent_mapped(page);
8152 ret = vmf_error(ret2);
8153 unlock_extent(io_tree, page_start, page_end, &cached_state);
8158 * we can't set the delalloc bits if there are pending ordered
8159 * extents. Drop our locks and wait for them to finish
8161 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8164 unlock_extent(io_tree, page_start, page_end, &cached_state);
8166 up_read(&BTRFS_I(inode)->i_mmap_lock);
8167 btrfs_start_ordered_extent(ordered);
8168 btrfs_put_ordered_extent(ordered);
8172 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8173 reserved_space = round_up(size - page_start,
8174 fs_info->sectorsize);
8175 if (reserved_space < PAGE_SIZE) {
8176 end = page_start + reserved_space - 1;
8177 btrfs_delalloc_release_space(BTRFS_I(inode),
8178 data_reserved, page_start,
8179 PAGE_SIZE - reserved_space, true);
8184 * page_mkwrite gets called when the page is firstly dirtied after it's
8185 * faulted in, but write(2) could also dirty a page and set delalloc
8186 * bits, thus in this case for space account reason, we still need to
8187 * clear any delalloc bits within this page range since we have to
8188 * reserve data&meta space before lock_page() (see above comments).
8190 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8191 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8192 EXTENT_DEFRAG, &cached_state);
8194 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8197 unlock_extent(io_tree, page_start, page_end, &cached_state);
8198 ret = VM_FAULT_SIGBUS;
8202 /* page is wholly or partially inside EOF */
8203 if (page_start + PAGE_SIZE > size)
8204 zero_start = offset_in_page(size);
8206 zero_start = PAGE_SIZE;
8208 if (zero_start != PAGE_SIZE)
8209 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8211 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8212 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8213 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8215 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8217 unlock_extent(io_tree, page_start, page_end, &cached_state);
8218 up_read(&BTRFS_I(inode)->i_mmap_lock);
8220 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8221 sb_end_pagefault(inode->i_sb);
8222 extent_changeset_free(data_reserved);
8223 return VM_FAULT_LOCKED;
8227 up_read(&BTRFS_I(inode)->i_mmap_lock);
8229 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8230 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8231 reserved_space, (ret != 0));
8233 sb_end_pagefault(inode->i_sb);
8234 extent_changeset_free(data_reserved);
8238 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8240 struct btrfs_truncate_control control = {
8242 .ino = btrfs_ino(inode),
8243 .min_type = BTRFS_EXTENT_DATA_KEY,
8244 .clear_extent_range = true,
8246 struct btrfs_root *root = inode->root;
8247 struct btrfs_fs_info *fs_info = root->fs_info;
8248 struct btrfs_block_rsv *rsv;
8250 struct btrfs_trans_handle *trans;
8251 u64 mask = fs_info->sectorsize - 1;
8252 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8254 if (!skip_writeback) {
8255 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8256 inode->vfs_inode.i_size & (~mask),
8263 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8264 * things going on here:
8266 * 1) We need to reserve space to update our inode.
8268 * 2) We need to have something to cache all the space that is going to
8269 * be free'd up by the truncate operation, but also have some slack
8270 * space reserved in case it uses space during the truncate (thank you
8271 * very much snapshotting).
8273 * And we need these to be separate. The fact is we can use a lot of
8274 * space doing the truncate, and we have no earthly idea how much space
8275 * we will use, so we need the truncate reservation to be separate so it
8276 * doesn't end up using space reserved for updating the inode. We also
8277 * need to be able to stop the transaction and start a new one, which
8278 * means we need to be able to update the inode several times, and we
8279 * have no idea of knowing how many times that will be, so we can't just
8280 * reserve 1 item for the entirety of the operation, so that has to be
8281 * done separately as well.
8283 * So that leaves us with
8285 * 1) rsv - for the truncate reservation, which we will steal from the
8286 * transaction reservation.
8287 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8288 * updating the inode.
8290 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8293 rsv->size = min_size;
8294 rsv->failfast = true;
8297 * 1 for the truncate slack space
8298 * 1 for updating the inode.
8300 trans = btrfs_start_transaction(root, 2);
8301 if (IS_ERR(trans)) {
8302 ret = PTR_ERR(trans);
8306 /* Migrate the slack space for the truncate to our reserve */
8307 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8310 * We have reserved 2 metadata units when we started the transaction and
8311 * min_size matches 1 unit, so this should never fail, but if it does,
8312 * it's not critical we just fail truncation.
8315 btrfs_end_transaction(trans);
8319 trans->block_rsv = rsv;
8322 struct extent_state *cached_state = NULL;
8323 const u64 new_size = inode->vfs_inode.i_size;
8324 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8326 control.new_size = new_size;
8327 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8329 * We want to drop from the next block forward in case this new
8330 * size is not block aligned since we will be keeping the last
8331 * block of the extent just the way it is.
8333 btrfs_drop_extent_map_range(inode,
8334 ALIGN(new_size, fs_info->sectorsize),
8337 ret = btrfs_truncate_inode_items(trans, root, &control);
8339 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8340 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8342 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8344 trans->block_rsv = &fs_info->trans_block_rsv;
8345 if (ret != -ENOSPC && ret != -EAGAIN)
8348 ret = btrfs_update_inode(trans, inode);
8352 btrfs_end_transaction(trans);
8353 btrfs_btree_balance_dirty(fs_info);
8355 trans = btrfs_start_transaction(root, 2);
8356 if (IS_ERR(trans)) {
8357 ret = PTR_ERR(trans);
8362 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8363 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8364 rsv, min_size, false);
8366 * We have reserved 2 metadata units when we started the
8367 * transaction and min_size matches 1 unit, so this should never
8368 * fail, but if it does, it's not critical we just fail truncation.
8373 trans->block_rsv = rsv;
8377 * We can't call btrfs_truncate_block inside a trans handle as we could
8378 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8379 * know we've truncated everything except the last little bit, and can
8380 * do btrfs_truncate_block and then update the disk_i_size.
8382 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8383 btrfs_end_transaction(trans);
8384 btrfs_btree_balance_dirty(fs_info);
8386 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8389 trans = btrfs_start_transaction(root, 1);
8390 if (IS_ERR(trans)) {
8391 ret = PTR_ERR(trans);
8394 btrfs_inode_safe_disk_i_size_write(inode, 0);
8400 trans->block_rsv = &fs_info->trans_block_rsv;
8401 ret2 = btrfs_update_inode(trans, inode);
8405 ret2 = btrfs_end_transaction(trans);
8408 btrfs_btree_balance_dirty(fs_info);
8411 btrfs_free_block_rsv(fs_info, rsv);
8413 * So if we truncate and then write and fsync we normally would just
8414 * write the extents that changed, which is a problem if we need to
8415 * first truncate that entire inode. So set this flag so we write out
8416 * all of the extents in the inode to the sync log so we're completely
8419 * If no extents were dropped or trimmed we don't need to force the next
8420 * fsync to truncate all the inode's items from the log and re-log them
8421 * all. This means the truncate operation did not change the file size,
8422 * or changed it to a smaller size but there was only an implicit hole
8423 * between the old i_size and the new i_size, and there were no prealloc
8424 * extents beyond i_size to drop.
8426 if (control.extents_found > 0)
8427 btrfs_set_inode_full_sync(inode);
8432 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8435 struct inode *inode;
8437 inode = new_inode(dir->i_sb);
8440 * Subvolumes don't inherit the sgid bit or the parent's gid if
8441 * the parent's sgid bit is set. This is probably a bug.
8443 inode_init_owner(idmap, inode, NULL,
8444 S_IFDIR | (~current_umask() & S_IRWXUGO));
8445 inode->i_op = &btrfs_dir_inode_operations;
8446 inode->i_fop = &btrfs_dir_file_operations;
8451 struct inode *btrfs_alloc_inode(struct super_block *sb)
8453 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8454 struct btrfs_inode *ei;
8455 struct inode *inode;
8457 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8464 ei->last_sub_trans = 0;
8465 ei->logged_trans = 0;
8466 ei->delalloc_bytes = 0;
8467 ei->new_delalloc_bytes = 0;
8468 ei->defrag_bytes = 0;
8469 ei->disk_i_size = 0;
8473 ei->index_cnt = (u64)-1;
8475 ei->last_unlink_trans = 0;
8476 ei->last_reflink_trans = 0;
8477 ei->last_log_commit = 0;
8479 spin_lock_init(&ei->lock);
8480 ei->outstanding_extents = 0;
8481 if (sb->s_magic != BTRFS_TEST_MAGIC)
8482 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8483 BTRFS_BLOCK_RSV_DELALLOC);
8484 ei->runtime_flags = 0;
8485 ei->prop_compress = BTRFS_COMPRESS_NONE;
8486 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8488 ei->delayed_node = NULL;
8490 ei->i_otime.tv_sec = 0;
8491 ei->i_otime.tv_nsec = 0;
8493 inode = &ei->vfs_inode;
8494 extent_map_tree_init(&ei->extent_tree);
8495 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8496 ei->io_tree.inode = ei;
8497 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8498 IO_TREE_INODE_FILE_EXTENT);
8499 mutex_init(&ei->log_mutex);
8500 spin_lock_init(&ei->ordered_tree_lock);
8501 ei->ordered_tree = RB_ROOT;
8502 ei->ordered_tree_last = NULL;
8503 INIT_LIST_HEAD(&ei->delalloc_inodes);
8504 INIT_LIST_HEAD(&ei->delayed_iput);
8505 RB_CLEAR_NODE(&ei->rb_node);
8506 init_rwsem(&ei->i_mmap_lock);
8511 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8512 void btrfs_test_destroy_inode(struct inode *inode)
8514 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8515 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8519 void btrfs_free_inode(struct inode *inode)
8521 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8524 void btrfs_destroy_inode(struct inode *vfs_inode)
8526 struct btrfs_ordered_extent *ordered;
8527 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8528 struct btrfs_root *root = inode->root;
8529 bool freespace_inode;
8531 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8532 WARN_ON(vfs_inode->i_data.nrpages);
8533 WARN_ON(inode->block_rsv.reserved);
8534 WARN_ON(inode->block_rsv.size);
8535 WARN_ON(inode->outstanding_extents);
8536 if (!S_ISDIR(vfs_inode->i_mode)) {
8537 WARN_ON(inode->delalloc_bytes);
8538 WARN_ON(inode->new_delalloc_bytes);
8540 WARN_ON(inode->csum_bytes);
8541 WARN_ON(inode->defrag_bytes);
8544 * This can happen where we create an inode, but somebody else also
8545 * created the same inode and we need to destroy the one we already
8552 * If this is a free space inode do not take the ordered extents lockdep
8555 freespace_inode = btrfs_is_free_space_inode(inode);
8558 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8562 btrfs_err(root->fs_info,
8563 "found ordered extent %llu %llu on inode cleanup",
8564 ordered->file_offset, ordered->num_bytes);
8566 if (!freespace_inode)
8567 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8569 btrfs_remove_ordered_extent(inode, ordered);
8570 btrfs_put_ordered_extent(ordered);
8571 btrfs_put_ordered_extent(ordered);
8574 btrfs_qgroup_check_reserved_leak(inode);
8575 inode_tree_del(inode);
8576 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8577 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8578 btrfs_put_root(inode->root);
8581 int btrfs_drop_inode(struct inode *inode)
8583 struct btrfs_root *root = BTRFS_I(inode)->root;
8588 /* the snap/subvol tree is on deleting */
8589 if (btrfs_root_refs(&root->root_item) == 0)
8592 return generic_drop_inode(inode);
8595 static void init_once(void *foo)
8597 struct btrfs_inode *ei = foo;
8599 inode_init_once(&ei->vfs_inode);
8602 void __cold btrfs_destroy_cachep(void)
8605 * Make sure all delayed rcu free inodes are flushed before we
8609 bioset_exit(&btrfs_dio_bioset);
8610 kmem_cache_destroy(btrfs_inode_cachep);
8613 int __init btrfs_init_cachep(void)
8615 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8616 sizeof(struct btrfs_inode), 0,
8617 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8619 if (!btrfs_inode_cachep)
8622 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8623 offsetof(struct btrfs_dio_private, bbio.bio),
8629 btrfs_destroy_cachep();
8633 static int btrfs_getattr(struct mnt_idmap *idmap,
8634 const struct path *path, struct kstat *stat,
8635 u32 request_mask, unsigned int flags)
8639 struct inode *inode = d_inode(path->dentry);
8640 u32 blocksize = inode->i_sb->s_blocksize;
8641 u32 bi_flags = BTRFS_I(inode)->flags;
8642 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8644 stat->result_mask |= STATX_BTIME;
8645 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8646 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8647 if (bi_flags & BTRFS_INODE_APPEND)
8648 stat->attributes |= STATX_ATTR_APPEND;
8649 if (bi_flags & BTRFS_INODE_COMPRESS)
8650 stat->attributes |= STATX_ATTR_COMPRESSED;
8651 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8652 stat->attributes |= STATX_ATTR_IMMUTABLE;
8653 if (bi_flags & BTRFS_INODE_NODUMP)
8654 stat->attributes |= STATX_ATTR_NODUMP;
8655 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8656 stat->attributes |= STATX_ATTR_VERITY;
8658 stat->attributes_mask |= (STATX_ATTR_APPEND |
8659 STATX_ATTR_COMPRESSED |
8660 STATX_ATTR_IMMUTABLE |
8663 generic_fillattr(idmap, request_mask, inode, stat);
8664 stat->dev = BTRFS_I(inode)->root->anon_dev;
8666 spin_lock(&BTRFS_I(inode)->lock);
8667 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8668 inode_bytes = inode_get_bytes(inode);
8669 spin_unlock(&BTRFS_I(inode)->lock);
8670 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8671 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8675 static int btrfs_rename_exchange(struct inode *old_dir,
8676 struct dentry *old_dentry,
8677 struct inode *new_dir,
8678 struct dentry *new_dentry)
8680 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8681 struct btrfs_trans_handle *trans;
8682 unsigned int trans_num_items;
8683 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8684 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8685 struct inode *new_inode = new_dentry->d_inode;
8686 struct inode *old_inode = old_dentry->d_inode;
8687 struct btrfs_rename_ctx old_rename_ctx;
8688 struct btrfs_rename_ctx new_rename_ctx;
8689 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8690 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8695 bool need_abort = false;
8696 struct fscrypt_name old_fname, new_fname;
8697 struct fscrypt_str *old_name, *new_name;
8700 * For non-subvolumes allow exchange only within one subvolume, in the
8701 * same inode namespace. Two subvolumes (represented as directory) can
8702 * be exchanged as they're a logical link and have a fixed inode number.
8705 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8706 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8709 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8713 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8715 fscrypt_free_filename(&old_fname);
8719 old_name = &old_fname.disk_name;
8720 new_name = &new_fname.disk_name;
8722 /* close the race window with snapshot create/destroy ioctl */
8723 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8724 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8725 down_read(&fs_info->subvol_sem);
8729 * 1 to remove old dir item
8730 * 1 to remove old dir index
8731 * 1 to add new dir item
8732 * 1 to add new dir index
8733 * 1 to update parent inode
8735 * If the parents are the same, we only need to account for one
8737 trans_num_items = (old_dir == new_dir ? 9 : 10);
8738 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8740 * 1 to remove old root ref
8741 * 1 to remove old root backref
8742 * 1 to add new root ref
8743 * 1 to add new root backref
8745 trans_num_items += 4;
8748 * 1 to update inode item
8749 * 1 to remove old inode ref
8750 * 1 to add new inode ref
8752 trans_num_items += 3;
8754 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8755 trans_num_items += 4;
8757 trans_num_items += 3;
8758 trans = btrfs_start_transaction(root, trans_num_items);
8759 if (IS_ERR(trans)) {
8760 ret = PTR_ERR(trans);
8765 ret = btrfs_record_root_in_trans(trans, dest);
8771 * We need to find a free sequence number both in the source and
8772 * in the destination directory for the exchange.
8774 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8777 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8781 BTRFS_I(old_inode)->dir_index = 0ULL;
8782 BTRFS_I(new_inode)->dir_index = 0ULL;
8784 /* Reference for the source. */
8785 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8786 /* force full log commit if subvolume involved. */
8787 btrfs_set_log_full_commit(trans);
8789 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8790 btrfs_ino(BTRFS_I(new_dir)),
8797 /* And now for the dest. */
8798 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8799 /* force full log commit if subvolume involved. */
8800 btrfs_set_log_full_commit(trans);
8802 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8803 btrfs_ino(BTRFS_I(old_dir)),
8807 btrfs_abort_transaction(trans, ret);
8812 /* Update inode version and ctime/mtime. */
8813 inode_inc_iversion(old_dir);
8814 inode_inc_iversion(new_dir);
8815 inode_inc_iversion(old_inode);
8816 inode_inc_iversion(new_inode);
8817 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
8819 if (old_dentry->d_parent != new_dentry->d_parent) {
8820 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8821 BTRFS_I(old_inode), true);
8822 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8823 BTRFS_I(new_inode), true);
8826 /* src is a subvolume */
8827 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8828 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8829 } else { /* src is an inode */
8830 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8831 BTRFS_I(old_dentry->d_inode),
8832 old_name, &old_rename_ctx);
8834 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
8837 btrfs_abort_transaction(trans, ret);
8841 /* dest is a subvolume */
8842 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8843 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8844 } else { /* dest is an inode */
8845 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8846 BTRFS_I(new_dentry->d_inode),
8847 new_name, &new_rename_ctx);
8849 ret = btrfs_update_inode(trans, BTRFS_I(new_inode));
8852 btrfs_abort_transaction(trans, ret);
8856 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8857 new_name, 0, old_idx);
8859 btrfs_abort_transaction(trans, ret);
8863 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8864 old_name, 0, new_idx);
8866 btrfs_abort_transaction(trans, ret);
8870 if (old_inode->i_nlink == 1)
8871 BTRFS_I(old_inode)->dir_index = old_idx;
8872 if (new_inode->i_nlink == 1)
8873 BTRFS_I(new_inode)->dir_index = new_idx;
8876 * Now pin the logs of the roots. We do it to ensure that no other task
8877 * can sync the logs while we are in progress with the rename, because
8878 * that could result in an inconsistency in case any of the inodes that
8879 * are part of this rename operation were logged before.
8881 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8882 btrfs_pin_log_trans(root);
8883 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8884 btrfs_pin_log_trans(dest);
8886 /* Do the log updates for all inodes. */
8887 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8888 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8889 old_rename_ctx.index, new_dentry->d_parent);
8890 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8891 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8892 new_rename_ctx.index, old_dentry->d_parent);
8894 /* Now unpin the logs. */
8895 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8896 btrfs_end_log_trans(root);
8897 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8898 btrfs_end_log_trans(dest);
8900 ret2 = btrfs_end_transaction(trans);
8901 ret = ret ? ret : ret2;
8903 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8904 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8905 up_read(&fs_info->subvol_sem);
8907 fscrypt_free_filename(&new_fname);
8908 fscrypt_free_filename(&old_fname);
8912 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8915 struct inode *inode;
8917 inode = new_inode(dir->i_sb);
8919 inode_init_owner(idmap, inode, dir,
8920 S_IFCHR | WHITEOUT_MODE);
8921 inode->i_op = &btrfs_special_inode_operations;
8922 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8927 static int btrfs_rename(struct mnt_idmap *idmap,
8928 struct inode *old_dir, struct dentry *old_dentry,
8929 struct inode *new_dir, struct dentry *new_dentry,
8932 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8933 struct btrfs_new_inode_args whiteout_args = {
8935 .dentry = old_dentry,
8937 struct btrfs_trans_handle *trans;
8938 unsigned int trans_num_items;
8939 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8940 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8941 struct inode *new_inode = d_inode(new_dentry);
8942 struct inode *old_inode = d_inode(old_dentry);
8943 struct btrfs_rename_ctx rename_ctx;
8947 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8948 struct fscrypt_name old_fname, new_fname;
8950 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8953 /* we only allow rename subvolume link between subvolumes */
8954 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8957 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8958 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
8961 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8962 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8965 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8969 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8971 fscrypt_free_filename(&old_fname);
8975 /* check for collisions, even if the name isn't there */
8976 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
8978 if (ret == -EEXIST) {
8980 * eexist without a new_inode */
8981 if (WARN_ON(!new_inode)) {
8982 goto out_fscrypt_names;
8985 /* maybe -EOVERFLOW */
8986 goto out_fscrypt_names;
8992 * we're using rename to replace one file with another. Start IO on it
8993 * now so we don't add too much work to the end of the transaction
8995 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8996 filemap_flush(old_inode->i_mapping);
8998 if (flags & RENAME_WHITEOUT) {
8999 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9000 if (!whiteout_args.inode) {
9002 goto out_fscrypt_names;
9004 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9006 goto out_whiteout_inode;
9008 /* 1 to update the old parent inode. */
9009 trans_num_items = 1;
9012 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9013 /* Close the race window with snapshot create/destroy ioctl */
9014 down_read(&fs_info->subvol_sem);
9016 * 1 to remove old root ref
9017 * 1 to remove old root backref
9018 * 1 to add new root ref
9019 * 1 to add new root backref
9021 trans_num_items += 4;
9025 * 1 to remove old inode ref
9026 * 1 to add new inode ref
9028 trans_num_items += 3;
9031 * 1 to remove old dir item
9032 * 1 to remove old dir index
9033 * 1 to add new dir item
9034 * 1 to add new dir index
9036 trans_num_items += 4;
9037 /* 1 to update new parent inode if it's not the same as the old parent */
9038 if (new_dir != old_dir)
9043 * 1 to remove inode ref
9044 * 1 to remove dir item
9045 * 1 to remove dir index
9046 * 1 to possibly add orphan item
9048 trans_num_items += 5;
9050 trans = btrfs_start_transaction(root, trans_num_items);
9051 if (IS_ERR(trans)) {
9052 ret = PTR_ERR(trans);
9057 ret = btrfs_record_root_in_trans(trans, dest);
9062 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9066 BTRFS_I(old_inode)->dir_index = 0ULL;
9067 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9068 /* force full log commit if subvolume involved. */
9069 btrfs_set_log_full_commit(trans);
9071 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9072 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9078 inode_inc_iversion(old_dir);
9079 inode_inc_iversion(new_dir);
9080 inode_inc_iversion(old_inode);
9081 simple_rename_timestamp(old_dir, old_dentry, new_dir, new_dentry);
9083 if (old_dentry->d_parent != new_dentry->d_parent)
9084 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9085 BTRFS_I(old_inode), true);
9087 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9088 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9090 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9091 BTRFS_I(d_inode(old_dentry)),
9092 &old_fname.disk_name, &rename_ctx);
9094 ret = btrfs_update_inode(trans, BTRFS_I(old_inode));
9097 btrfs_abort_transaction(trans, ret);
9102 inode_inc_iversion(new_inode);
9103 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9104 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9105 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9106 BUG_ON(new_inode->i_nlink == 0);
9108 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9109 BTRFS_I(d_inode(new_dentry)),
9110 &new_fname.disk_name);
9112 if (!ret && new_inode->i_nlink == 0)
9113 ret = btrfs_orphan_add(trans,
9114 BTRFS_I(d_inode(new_dentry)));
9116 btrfs_abort_transaction(trans, ret);
9121 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9122 &new_fname.disk_name, 0, index);
9124 btrfs_abort_transaction(trans, ret);
9128 if (old_inode->i_nlink == 1)
9129 BTRFS_I(old_inode)->dir_index = index;
9131 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9132 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9133 rename_ctx.index, new_dentry->d_parent);
9135 if (flags & RENAME_WHITEOUT) {
9136 ret = btrfs_create_new_inode(trans, &whiteout_args);
9138 btrfs_abort_transaction(trans, ret);
9141 unlock_new_inode(whiteout_args.inode);
9142 iput(whiteout_args.inode);
9143 whiteout_args.inode = NULL;
9147 ret2 = btrfs_end_transaction(trans);
9148 ret = ret ? ret : ret2;
9150 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9151 up_read(&fs_info->subvol_sem);
9152 if (flags & RENAME_WHITEOUT)
9153 btrfs_new_inode_args_destroy(&whiteout_args);
9155 if (flags & RENAME_WHITEOUT)
9156 iput(whiteout_args.inode);
9158 fscrypt_free_filename(&old_fname);
9159 fscrypt_free_filename(&new_fname);
9163 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9164 struct dentry *old_dentry, struct inode *new_dir,
9165 struct dentry *new_dentry, unsigned int flags)
9169 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9172 if (flags & RENAME_EXCHANGE)
9173 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9176 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9179 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9184 struct btrfs_delalloc_work {
9185 struct inode *inode;
9186 struct completion completion;
9187 struct list_head list;
9188 struct btrfs_work work;
9191 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9193 struct btrfs_delalloc_work *delalloc_work;
9194 struct inode *inode;
9196 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9198 inode = delalloc_work->inode;
9199 filemap_flush(inode->i_mapping);
9200 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9201 &BTRFS_I(inode)->runtime_flags))
9202 filemap_flush(inode->i_mapping);
9205 complete(&delalloc_work->completion);
9208 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9210 struct btrfs_delalloc_work *work;
9212 work = kmalloc(sizeof(*work), GFP_NOFS);
9216 init_completion(&work->completion);
9217 INIT_LIST_HEAD(&work->list);
9218 work->inode = inode;
9219 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL);
9225 * some fairly slow code that needs optimization. This walks the list
9226 * of all the inodes with pending delalloc and forces them to disk.
9228 static int start_delalloc_inodes(struct btrfs_root *root,
9229 struct writeback_control *wbc, bool snapshot,
9230 bool in_reclaim_context)
9232 struct btrfs_inode *binode;
9233 struct inode *inode;
9234 struct btrfs_delalloc_work *work, *next;
9238 bool full_flush = wbc->nr_to_write == LONG_MAX;
9240 mutex_lock(&root->delalloc_mutex);
9241 spin_lock(&root->delalloc_lock);
9242 list_splice_init(&root->delalloc_inodes, &splice);
9243 while (!list_empty(&splice)) {
9244 binode = list_entry(splice.next, struct btrfs_inode,
9247 list_move_tail(&binode->delalloc_inodes,
9248 &root->delalloc_inodes);
9250 if (in_reclaim_context &&
9251 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9254 inode = igrab(&binode->vfs_inode);
9256 cond_resched_lock(&root->delalloc_lock);
9259 spin_unlock(&root->delalloc_lock);
9262 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9263 &binode->runtime_flags);
9265 work = btrfs_alloc_delalloc_work(inode);
9271 list_add_tail(&work->list, &works);
9272 btrfs_queue_work(root->fs_info->flush_workers,
9275 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9276 btrfs_add_delayed_iput(BTRFS_I(inode));
9277 if (ret || wbc->nr_to_write <= 0)
9281 spin_lock(&root->delalloc_lock);
9283 spin_unlock(&root->delalloc_lock);
9286 list_for_each_entry_safe(work, next, &works, list) {
9287 list_del_init(&work->list);
9288 wait_for_completion(&work->completion);
9292 if (!list_empty(&splice)) {
9293 spin_lock(&root->delalloc_lock);
9294 list_splice_tail(&splice, &root->delalloc_inodes);
9295 spin_unlock(&root->delalloc_lock);
9297 mutex_unlock(&root->delalloc_mutex);
9301 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9303 struct writeback_control wbc = {
9304 .nr_to_write = LONG_MAX,
9305 .sync_mode = WB_SYNC_NONE,
9307 .range_end = LLONG_MAX,
9309 struct btrfs_fs_info *fs_info = root->fs_info;
9311 if (BTRFS_FS_ERROR(fs_info))
9314 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9317 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9318 bool in_reclaim_context)
9320 struct writeback_control wbc = {
9322 .sync_mode = WB_SYNC_NONE,
9324 .range_end = LLONG_MAX,
9326 struct btrfs_root *root;
9330 if (BTRFS_FS_ERROR(fs_info))
9333 mutex_lock(&fs_info->delalloc_root_mutex);
9334 spin_lock(&fs_info->delalloc_root_lock);
9335 list_splice_init(&fs_info->delalloc_roots, &splice);
9336 while (!list_empty(&splice)) {
9338 * Reset nr_to_write here so we know that we're doing a full
9342 wbc.nr_to_write = LONG_MAX;
9344 root = list_first_entry(&splice, struct btrfs_root,
9346 root = btrfs_grab_root(root);
9348 list_move_tail(&root->delalloc_root,
9349 &fs_info->delalloc_roots);
9350 spin_unlock(&fs_info->delalloc_root_lock);
9352 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9353 btrfs_put_root(root);
9354 if (ret < 0 || wbc.nr_to_write <= 0)
9356 spin_lock(&fs_info->delalloc_root_lock);
9358 spin_unlock(&fs_info->delalloc_root_lock);
9362 if (!list_empty(&splice)) {
9363 spin_lock(&fs_info->delalloc_root_lock);
9364 list_splice_tail(&splice, &fs_info->delalloc_roots);
9365 spin_unlock(&fs_info->delalloc_root_lock);
9367 mutex_unlock(&fs_info->delalloc_root_mutex);
9371 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9372 struct dentry *dentry, const char *symname)
9374 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9375 struct btrfs_trans_handle *trans;
9376 struct btrfs_root *root = BTRFS_I(dir)->root;
9377 struct btrfs_path *path;
9378 struct btrfs_key key;
9379 struct inode *inode;
9380 struct btrfs_new_inode_args new_inode_args = {
9384 unsigned int trans_num_items;
9389 struct btrfs_file_extent_item *ei;
9390 struct extent_buffer *leaf;
9392 name_len = strlen(symname);
9393 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9394 return -ENAMETOOLONG;
9396 inode = new_inode(dir->i_sb);
9399 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9400 inode->i_op = &btrfs_symlink_inode_operations;
9401 inode_nohighmem(inode);
9402 inode->i_mapping->a_ops = &btrfs_aops;
9403 btrfs_i_size_write(BTRFS_I(inode), name_len);
9404 inode_set_bytes(inode, name_len);
9406 new_inode_args.inode = inode;
9407 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9410 /* 1 additional item for the inline extent */
9413 trans = btrfs_start_transaction(root, trans_num_items);
9414 if (IS_ERR(trans)) {
9415 err = PTR_ERR(trans);
9416 goto out_new_inode_args;
9419 err = btrfs_create_new_inode(trans, &new_inode_args);
9423 path = btrfs_alloc_path();
9426 btrfs_abort_transaction(trans, err);
9427 discard_new_inode(inode);
9431 key.objectid = btrfs_ino(BTRFS_I(inode));
9433 key.type = BTRFS_EXTENT_DATA_KEY;
9434 datasize = btrfs_file_extent_calc_inline_size(name_len);
9435 err = btrfs_insert_empty_item(trans, root, path, &key,
9438 btrfs_abort_transaction(trans, err);
9439 btrfs_free_path(path);
9440 discard_new_inode(inode);
9444 leaf = path->nodes[0];
9445 ei = btrfs_item_ptr(leaf, path->slots[0],
9446 struct btrfs_file_extent_item);
9447 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9448 btrfs_set_file_extent_type(leaf, ei,
9449 BTRFS_FILE_EXTENT_INLINE);
9450 btrfs_set_file_extent_encryption(leaf, ei, 0);
9451 btrfs_set_file_extent_compression(leaf, ei, 0);
9452 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9453 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9455 ptr = btrfs_file_extent_inline_start(ei);
9456 write_extent_buffer(leaf, symname, ptr, name_len);
9457 btrfs_mark_buffer_dirty(trans, leaf);
9458 btrfs_free_path(path);
9460 d_instantiate_new(dentry, inode);
9463 btrfs_end_transaction(trans);
9464 btrfs_btree_balance_dirty(fs_info);
9466 btrfs_new_inode_args_destroy(&new_inode_args);
9473 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9474 struct btrfs_trans_handle *trans_in,
9475 struct btrfs_inode *inode,
9476 struct btrfs_key *ins,
9479 struct btrfs_file_extent_item stack_fi;
9480 struct btrfs_replace_extent_info extent_info;
9481 struct btrfs_trans_handle *trans = trans_in;
9482 struct btrfs_path *path;
9483 u64 start = ins->objectid;
9484 u64 len = ins->offset;
9485 int qgroup_released;
9488 memset(&stack_fi, 0, sizeof(stack_fi));
9490 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9491 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9492 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9493 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9494 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9495 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9496 /* Encryption and other encoding is reserved and all 0 */
9498 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9499 if (qgroup_released < 0)
9500 return ERR_PTR(qgroup_released);
9503 ret = insert_reserved_file_extent(trans, inode,
9504 file_offset, &stack_fi,
9505 true, qgroup_released);
9511 extent_info.disk_offset = start;
9512 extent_info.disk_len = len;
9513 extent_info.data_offset = 0;
9514 extent_info.data_len = len;
9515 extent_info.file_offset = file_offset;
9516 extent_info.extent_buf = (char *)&stack_fi;
9517 extent_info.is_new_extent = true;
9518 extent_info.update_times = true;
9519 extent_info.qgroup_reserved = qgroup_released;
9520 extent_info.insertions = 0;
9522 path = btrfs_alloc_path();
9528 ret = btrfs_replace_file_extents(inode, path, file_offset,
9529 file_offset + len - 1, &extent_info,
9531 btrfs_free_path(path);
9538 * We have released qgroup data range at the beginning of the function,
9539 * and normally qgroup_released bytes will be freed when committing
9541 * But if we error out early, we have to free what we have released
9542 * or we leak qgroup data reservation.
9544 btrfs_qgroup_free_refroot(inode->root->fs_info,
9545 inode->root->root_key.objectid, qgroup_released,
9546 BTRFS_QGROUP_RSV_DATA);
9547 return ERR_PTR(ret);
9550 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9551 u64 start, u64 num_bytes, u64 min_size,
9552 loff_t actual_len, u64 *alloc_hint,
9553 struct btrfs_trans_handle *trans)
9555 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9556 struct extent_map *em;
9557 struct btrfs_root *root = BTRFS_I(inode)->root;
9558 struct btrfs_key ins;
9559 u64 cur_offset = start;
9560 u64 clear_offset = start;
9563 u64 last_alloc = (u64)-1;
9565 bool own_trans = true;
9566 u64 end = start + num_bytes - 1;
9570 while (num_bytes > 0) {
9571 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9572 cur_bytes = max(cur_bytes, min_size);
9574 * If we are severely fragmented we could end up with really
9575 * small allocations, so if the allocator is returning small
9576 * chunks lets make its job easier by only searching for those
9579 cur_bytes = min(cur_bytes, last_alloc);
9580 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9581 min_size, 0, *alloc_hint, &ins, 1, 0);
9586 * We've reserved this space, and thus converted it from
9587 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9588 * from here on out we will only need to clear our reservation
9589 * for the remaining unreserved area, so advance our
9590 * clear_offset by our extent size.
9592 clear_offset += ins.offset;
9594 last_alloc = ins.offset;
9595 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9598 * Now that we inserted the prealloc extent we can finally
9599 * decrement the number of reservations in the block group.
9600 * If we did it before, we could race with relocation and have
9601 * relocation miss the reserved extent, making it fail later.
9603 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9604 if (IS_ERR(trans)) {
9605 ret = PTR_ERR(trans);
9606 btrfs_free_reserved_extent(fs_info, ins.objectid,
9611 em = alloc_extent_map();
9613 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9614 cur_offset + ins.offset - 1, false);
9615 btrfs_set_inode_full_sync(BTRFS_I(inode));
9619 em->start = cur_offset;
9620 em->orig_start = cur_offset;
9621 em->len = ins.offset;
9622 em->block_start = ins.objectid;
9623 em->block_len = ins.offset;
9624 em->orig_block_len = ins.offset;
9625 em->ram_bytes = ins.offset;
9626 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9627 em->generation = trans->transid;
9629 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9630 free_extent_map(em);
9632 num_bytes -= ins.offset;
9633 cur_offset += ins.offset;
9634 *alloc_hint = ins.objectid + ins.offset;
9636 inode_inc_iversion(inode);
9637 inode_set_ctime_current(inode);
9638 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9639 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9640 (actual_len > inode->i_size) &&
9641 (cur_offset > inode->i_size)) {
9642 if (cur_offset > actual_len)
9643 i_size = actual_len;
9645 i_size = cur_offset;
9646 i_size_write(inode, i_size);
9647 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9650 ret = btrfs_update_inode(trans, BTRFS_I(inode));
9653 btrfs_abort_transaction(trans, ret);
9655 btrfs_end_transaction(trans);
9660 btrfs_end_transaction(trans);
9664 if (clear_offset < end)
9665 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9666 end - clear_offset + 1);
9670 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9671 u64 start, u64 num_bytes, u64 min_size,
9672 loff_t actual_len, u64 *alloc_hint)
9674 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9675 min_size, actual_len, alloc_hint,
9679 int btrfs_prealloc_file_range_trans(struct inode *inode,
9680 struct btrfs_trans_handle *trans, int mode,
9681 u64 start, u64 num_bytes, u64 min_size,
9682 loff_t actual_len, u64 *alloc_hint)
9684 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9685 min_size, actual_len, alloc_hint, trans);
9688 static int btrfs_permission(struct mnt_idmap *idmap,
9689 struct inode *inode, int mask)
9691 struct btrfs_root *root = BTRFS_I(inode)->root;
9692 umode_t mode = inode->i_mode;
9694 if (mask & MAY_WRITE &&
9695 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9696 if (btrfs_root_readonly(root))
9698 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9701 return generic_permission(idmap, inode, mask);
9704 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9705 struct file *file, umode_t mode)
9707 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9708 struct btrfs_trans_handle *trans;
9709 struct btrfs_root *root = BTRFS_I(dir)->root;
9710 struct inode *inode;
9711 struct btrfs_new_inode_args new_inode_args = {
9713 .dentry = file->f_path.dentry,
9716 unsigned int trans_num_items;
9719 inode = new_inode(dir->i_sb);
9722 inode_init_owner(idmap, inode, dir, mode);
9723 inode->i_fop = &btrfs_file_operations;
9724 inode->i_op = &btrfs_file_inode_operations;
9725 inode->i_mapping->a_ops = &btrfs_aops;
9727 new_inode_args.inode = inode;
9728 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9732 trans = btrfs_start_transaction(root, trans_num_items);
9733 if (IS_ERR(trans)) {
9734 ret = PTR_ERR(trans);
9735 goto out_new_inode_args;
9738 ret = btrfs_create_new_inode(trans, &new_inode_args);
9741 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9742 * set it to 1 because d_tmpfile() will issue a warning if the count is
9745 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9747 set_nlink(inode, 1);
9750 d_tmpfile(file, inode);
9751 unlock_new_inode(inode);
9752 mark_inode_dirty(inode);
9755 btrfs_end_transaction(trans);
9756 btrfs_btree_balance_dirty(fs_info);
9758 btrfs_new_inode_args_destroy(&new_inode_args);
9762 return finish_open_simple(file, ret);
9765 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9767 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9768 unsigned long index = start >> PAGE_SHIFT;
9769 unsigned long end_index = end >> PAGE_SHIFT;
9773 ASSERT(end + 1 - start <= U32_MAX);
9774 len = end + 1 - start;
9775 while (index <= end_index) {
9776 page = find_get_page(inode->vfs_inode.i_mapping, index);
9777 ASSERT(page); /* Pages should be in the extent_io_tree */
9779 btrfs_page_set_writeback(fs_info, page, start, len);
9785 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9788 switch (compress_type) {
9789 case BTRFS_COMPRESS_NONE:
9790 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9791 case BTRFS_COMPRESS_ZLIB:
9792 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9793 case BTRFS_COMPRESS_LZO:
9795 * The LZO format depends on the sector size. 64K is the maximum
9796 * sector size that we support.
9798 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9800 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9801 (fs_info->sectorsize_bits - 12);
9802 case BTRFS_COMPRESS_ZSTD:
9803 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9809 static ssize_t btrfs_encoded_read_inline(
9811 struct iov_iter *iter, u64 start,
9813 struct extent_state **cached_state,
9814 u64 extent_start, size_t count,
9815 struct btrfs_ioctl_encoded_io_args *encoded,
9818 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9819 struct btrfs_root *root = inode->root;
9820 struct btrfs_fs_info *fs_info = root->fs_info;
9821 struct extent_io_tree *io_tree = &inode->io_tree;
9822 struct btrfs_path *path;
9823 struct extent_buffer *leaf;
9824 struct btrfs_file_extent_item *item;
9830 path = btrfs_alloc_path();
9835 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9839 /* The extent item disappeared? */
9844 leaf = path->nodes[0];
9845 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9847 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9848 ptr = btrfs_file_extent_inline_start(item);
9850 encoded->len = min_t(u64, extent_start + ram_bytes,
9851 inode->vfs_inode.i_size) - iocb->ki_pos;
9852 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9853 btrfs_file_extent_compression(leaf, item));
9856 encoded->compression = ret;
9857 if (encoded->compression) {
9860 inline_size = btrfs_file_extent_inline_item_len(leaf,
9862 if (inline_size > count) {
9866 count = inline_size;
9867 encoded->unencoded_len = ram_bytes;
9868 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9870 count = min_t(u64, count, encoded->len);
9871 encoded->len = count;
9872 encoded->unencoded_len = count;
9873 ptr += iocb->ki_pos - extent_start;
9876 tmp = kmalloc(count, GFP_NOFS);
9881 read_extent_buffer(leaf, tmp, ptr, count);
9882 btrfs_release_path(path);
9883 unlock_extent(io_tree, start, lockend, cached_state);
9884 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9887 ret = copy_to_iter(tmp, count, iter);
9892 btrfs_free_path(path);
9896 struct btrfs_encoded_read_private {
9897 wait_queue_head_t wait;
9899 blk_status_t status;
9902 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9904 struct btrfs_encoded_read_private *priv = bbio->private;
9906 if (bbio->bio.bi_status) {
9908 * The memory barrier implied by the atomic_dec_return() here
9909 * pairs with the memory barrier implied by the
9910 * atomic_dec_return() or io_wait_event() in
9911 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9912 * write is observed before the load of status in
9913 * btrfs_encoded_read_regular_fill_pages().
9915 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9917 if (!atomic_dec_return(&priv->pending))
9918 wake_up(&priv->wait);
9919 bio_put(&bbio->bio);
9922 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9923 u64 file_offset, u64 disk_bytenr,
9924 u64 disk_io_size, struct page **pages)
9926 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9927 struct btrfs_encoded_read_private priv = {
9928 .pending = ATOMIC_INIT(1),
9930 unsigned long i = 0;
9931 struct btrfs_bio *bbio;
9933 init_waitqueue_head(&priv.wait);
9935 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9936 btrfs_encoded_read_endio, &priv);
9937 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9938 bbio->inode = inode;
9941 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
9943 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
9944 atomic_inc(&priv.pending);
9945 btrfs_submit_bio(bbio, 0);
9947 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9948 btrfs_encoded_read_endio, &priv);
9949 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9950 bbio->inode = inode;
9955 disk_bytenr += bytes;
9956 disk_io_size -= bytes;
9957 } while (disk_io_size);
9959 atomic_inc(&priv.pending);
9960 btrfs_submit_bio(bbio, 0);
9962 if (atomic_dec_return(&priv.pending))
9963 io_wait_event(priv.wait, !atomic_read(&priv.pending));
9964 /* See btrfs_encoded_read_endio() for ordering. */
9965 return blk_status_to_errno(READ_ONCE(priv.status));
9968 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
9969 struct iov_iter *iter,
9970 u64 start, u64 lockend,
9971 struct extent_state **cached_state,
9972 u64 disk_bytenr, u64 disk_io_size,
9973 size_t count, bool compressed,
9976 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9977 struct extent_io_tree *io_tree = &inode->io_tree;
9978 struct page **pages;
9979 unsigned long nr_pages, i;
9984 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
9985 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
9988 ret = btrfs_alloc_page_array(nr_pages, pages);
9994 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
9995 disk_io_size, pages);
9999 unlock_extent(io_tree, start, lockend, cached_state);
10000 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10007 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10008 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10011 while (cur < count) {
10012 size_t bytes = min_t(size_t, count - cur,
10013 PAGE_SIZE - page_offset);
10015 if (copy_page_to_iter(pages[i], page_offset, bytes,
10026 for (i = 0; i < nr_pages; i++) {
10028 __free_page(pages[i]);
10034 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10035 struct btrfs_ioctl_encoded_io_args *encoded)
10037 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10038 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10039 struct extent_io_tree *io_tree = &inode->io_tree;
10041 size_t count = iov_iter_count(iter);
10042 u64 start, lockend, disk_bytenr, disk_io_size;
10043 struct extent_state *cached_state = NULL;
10044 struct extent_map *em;
10045 bool unlocked = false;
10047 file_accessed(iocb->ki_filp);
10049 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10051 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10052 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10055 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10057 * We don't know how long the extent containing iocb->ki_pos is, but if
10058 * it's compressed we know that it won't be longer than this.
10060 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10063 struct btrfs_ordered_extent *ordered;
10065 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10066 lockend - start + 1);
10068 goto out_unlock_inode;
10069 lock_extent(io_tree, start, lockend, &cached_state);
10070 ordered = btrfs_lookup_ordered_range(inode, start,
10071 lockend - start + 1);
10074 btrfs_put_ordered_extent(ordered);
10075 unlock_extent(io_tree, start, lockend, &cached_state);
10079 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10082 goto out_unlock_extent;
10085 if (em->block_start == EXTENT_MAP_INLINE) {
10086 u64 extent_start = em->start;
10089 * For inline extents we get everything we need out of the
10092 free_extent_map(em);
10094 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10095 &cached_state, extent_start,
10096 count, encoded, &unlocked);
10101 * We only want to return up to EOF even if the extent extends beyond
10104 encoded->len = min_t(u64, extent_map_end(em),
10105 inode->vfs_inode.i_size) - iocb->ki_pos;
10106 if (em->block_start == EXTENT_MAP_HOLE ||
10107 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10108 disk_bytenr = EXTENT_MAP_HOLE;
10109 count = min_t(u64, count, encoded->len);
10110 encoded->len = count;
10111 encoded->unencoded_len = count;
10112 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10113 disk_bytenr = em->block_start;
10115 * Bail if the buffer isn't large enough to return the whole
10116 * compressed extent.
10118 if (em->block_len > count) {
10122 disk_io_size = em->block_len;
10123 count = em->block_len;
10124 encoded->unencoded_len = em->ram_bytes;
10125 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10126 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10127 em->compress_type);
10130 encoded->compression = ret;
10132 disk_bytenr = em->block_start + (start - em->start);
10133 if (encoded->len > count)
10134 encoded->len = count;
10136 * Don't read beyond what we locked. This also limits the page
10137 * allocations that we'll do.
10139 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10140 count = start + disk_io_size - iocb->ki_pos;
10141 encoded->len = count;
10142 encoded->unencoded_len = count;
10143 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10145 free_extent_map(em);
10148 if (disk_bytenr == EXTENT_MAP_HOLE) {
10149 unlock_extent(io_tree, start, lockend, &cached_state);
10150 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10152 ret = iov_iter_zero(count, iter);
10156 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10157 &cached_state, disk_bytenr,
10158 disk_io_size, count,
10159 encoded->compression,
10165 iocb->ki_pos += encoded->len;
10167 free_extent_map(em);
10170 unlock_extent(io_tree, start, lockend, &cached_state);
10173 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10177 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10178 const struct btrfs_ioctl_encoded_io_args *encoded)
10180 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10181 struct btrfs_root *root = inode->root;
10182 struct btrfs_fs_info *fs_info = root->fs_info;
10183 struct extent_io_tree *io_tree = &inode->io_tree;
10184 struct extent_changeset *data_reserved = NULL;
10185 struct extent_state *cached_state = NULL;
10186 struct btrfs_ordered_extent *ordered;
10190 u64 num_bytes, ram_bytes, disk_num_bytes;
10191 unsigned long nr_pages, i;
10192 struct page **pages;
10193 struct btrfs_key ins;
10194 bool extent_reserved = false;
10195 struct extent_map *em;
10198 switch (encoded->compression) {
10199 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10200 compression = BTRFS_COMPRESS_ZLIB;
10202 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10203 compression = BTRFS_COMPRESS_ZSTD;
10205 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10206 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10207 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10208 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10209 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10210 /* The sector size must match for LZO. */
10211 if (encoded->compression -
10212 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10213 fs_info->sectorsize_bits)
10215 compression = BTRFS_COMPRESS_LZO;
10220 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10223 orig_count = iov_iter_count(from);
10225 /* The extent size must be sane. */
10226 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10227 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10231 * The compressed data must be smaller than the decompressed data.
10233 * It's of course possible for data to compress to larger or the same
10234 * size, but the buffered I/O path falls back to no compression for such
10235 * data, and we don't want to break any assumptions by creating these
10238 * Note that this is less strict than the current check we have that the
10239 * compressed data must be at least one sector smaller than the
10240 * decompressed data. We only want to enforce the weaker requirement
10241 * from old kernels that it is at least one byte smaller.
10243 if (orig_count >= encoded->unencoded_len)
10246 /* The extent must start on a sector boundary. */
10247 start = iocb->ki_pos;
10248 if (!IS_ALIGNED(start, fs_info->sectorsize))
10252 * The extent must end on a sector boundary. However, we allow a write
10253 * which ends at or extends i_size to have an unaligned length; we round
10254 * up the extent size and set i_size to the unaligned end.
10256 if (start + encoded->len < inode->vfs_inode.i_size &&
10257 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10260 /* Finally, the offset in the unencoded data must be sector-aligned. */
10261 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10264 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10265 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10266 end = start + num_bytes - 1;
10269 * If the extent cannot be inline, the compressed data on disk must be
10270 * sector-aligned. For convenience, we extend it with zeroes if it
10273 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10274 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10275 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10278 for (i = 0; i < nr_pages; i++) {
10279 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10282 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10287 kaddr = kmap_local_page(pages[i]);
10288 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10289 kunmap_local(kaddr);
10293 if (bytes < PAGE_SIZE)
10294 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10295 kunmap_local(kaddr);
10299 struct btrfs_ordered_extent *ordered;
10301 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10304 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10305 start >> PAGE_SHIFT,
10306 end >> PAGE_SHIFT);
10309 lock_extent(io_tree, start, end, &cached_state);
10310 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10312 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10315 btrfs_put_ordered_extent(ordered);
10316 unlock_extent(io_tree, start, end, &cached_state);
10321 * We don't use the higher-level delalloc space functions because our
10322 * num_bytes and disk_num_bytes are different.
10324 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10327 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10329 goto out_free_data_space;
10330 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10333 goto out_qgroup_free_data;
10335 /* Try an inline extent first. */
10336 if (start == 0 && encoded->unencoded_len == encoded->len &&
10337 encoded->unencoded_offset == 0) {
10338 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10339 compression, pages, true);
10343 goto out_delalloc_release;
10347 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10348 disk_num_bytes, 0, 0, &ins, 1, 1);
10350 goto out_delalloc_release;
10351 extent_reserved = true;
10353 em = create_io_em(inode, start, num_bytes,
10354 start - encoded->unencoded_offset, ins.objectid,
10355 ins.offset, ins.offset, ram_bytes, compression,
10356 BTRFS_ORDERED_COMPRESSED);
10359 goto out_free_reserved;
10361 free_extent_map(em);
10363 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10364 ins.objectid, ins.offset,
10365 encoded->unencoded_offset,
10366 (1 << BTRFS_ORDERED_ENCODED) |
10367 (1 << BTRFS_ORDERED_COMPRESSED),
10369 if (IS_ERR(ordered)) {
10370 btrfs_drop_extent_map_range(inode, start, end, false);
10371 ret = PTR_ERR(ordered);
10372 goto out_free_reserved;
10374 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10376 if (start + encoded->len > inode->vfs_inode.i_size)
10377 i_size_write(&inode->vfs_inode, start + encoded->len);
10379 unlock_extent(io_tree, start, end, &cached_state);
10381 btrfs_delalloc_release_extents(inode, num_bytes);
10383 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10388 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10389 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10390 out_delalloc_release:
10391 btrfs_delalloc_release_extents(inode, num_bytes);
10392 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10393 out_qgroup_free_data:
10395 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10396 out_free_data_space:
10398 * If btrfs_reserve_extent() succeeded, then we already decremented
10401 if (!extent_reserved)
10402 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10404 unlock_extent(io_tree, start, end, &cached_state);
10406 for (i = 0; i < nr_pages; i++) {
10408 __free_page(pages[i]);
10413 iocb->ki_pos += encoded->len;
10419 * Add an entry indicating a block group or device which is pinned by a
10420 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10421 * negative errno on failure.
10423 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10424 bool is_block_group)
10426 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10427 struct btrfs_swapfile_pin *sp, *entry;
10428 struct rb_node **p;
10429 struct rb_node *parent = NULL;
10431 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10436 sp->is_block_group = is_block_group;
10437 sp->bg_extent_count = 1;
10439 spin_lock(&fs_info->swapfile_pins_lock);
10440 p = &fs_info->swapfile_pins.rb_node;
10443 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10444 if (sp->ptr < entry->ptr ||
10445 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10446 p = &(*p)->rb_left;
10447 } else if (sp->ptr > entry->ptr ||
10448 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10449 p = &(*p)->rb_right;
10451 if (is_block_group)
10452 entry->bg_extent_count++;
10453 spin_unlock(&fs_info->swapfile_pins_lock);
10458 rb_link_node(&sp->node, parent, p);
10459 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10460 spin_unlock(&fs_info->swapfile_pins_lock);
10464 /* Free all of the entries pinned by this swapfile. */
10465 static void btrfs_free_swapfile_pins(struct inode *inode)
10467 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10468 struct btrfs_swapfile_pin *sp;
10469 struct rb_node *node, *next;
10471 spin_lock(&fs_info->swapfile_pins_lock);
10472 node = rb_first(&fs_info->swapfile_pins);
10474 next = rb_next(node);
10475 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10476 if (sp->inode == inode) {
10477 rb_erase(&sp->node, &fs_info->swapfile_pins);
10478 if (sp->is_block_group) {
10479 btrfs_dec_block_group_swap_extents(sp->ptr,
10480 sp->bg_extent_count);
10481 btrfs_put_block_group(sp->ptr);
10487 spin_unlock(&fs_info->swapfile_pins_lock);
10490 struct btrfs_swap_info {
10496 unsigned long nr_pages;
10500 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10501 struct btrfs_swap_info *bsi)
10503 unsigned long nr_pages;
10504 unsigned long max_pages;
10505 u64 first_ppage, first_ppage_reported, next_ppage;
10509 * Our swapfile may have had its size extended after the swap header was
10510 * written. In that case activating the swapfile should not go beyond
10511 * the max size set in the swap header.
10513 if (bsi->nr_pages >= sis->max)
10516 max_pages = sis->max - bsi->nr_pages;
10517 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10518 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10520 if (first_ppage >= next_ppage)
10522 nr_pages = next_ppage - first_ppage;
10523 nr_pages = min(nr_pages, max_pages);
10525 first_ppage_reported = first_ppage;
10526 if (bsi->start == 0)
10527 first_ppage_reported++;
10528 if (bsi->lowest_ppage > first_ppage_reported)
10529 bsi->lowest_ppage = first_ppage_reported;
10530 if (bsi->highest_ppage < (next_ppage - 1))
10531 bsi->highest_ppage = next_ppage - 1;
10533 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10536 bsi->nr_extents += ret;
10537 bsi->nr_pages += nr_pages;
10541 static void btrfs_swap_deactivate(struct file *file)
10543 struct inode *inode = file_inode(file);
10545 btrfs_free_swapfile_pins(inode);
10546 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10549 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10552 struct inode *inode = file_inode(file);
10553 struct btrfs_root *root = BTRFS_I(inode)->root;
10554 struct btrfs_fs_info *fs_info = root->fs_info;
10555 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10556 struct extent_state *cached_state = NULL;
10557 struct extent_map *em = NULL;
10558 struct btrfs_device *device = NULL;
10559 struct btrfs_swap_info bsi = {
10560 .lowest_ppage = (sector_t)-1ULL,
10567 * If the swap file was just created, make sure delalloc is done. If the
10568 * file changes again after this, the user is doing something stupid and
10569 * we don't really care.
10571 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10576 * The inode is locked, so these flags won't change after we check them.
10578 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10579 btrfs_warn(fs_info, "swapfile must not be compressed");
10582 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10583 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10586 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10587 btrfs_warn(fs_info, "swapfile must not be checksummed");
10592 * Balance or device remove/replace/resize can move stuff around from
10593 * under us. The exclop protection makes sure they aren't running/won't
10594 * run concurrently while we are mapping the swap extents, and
10595 * fs_info->swapfile_pins prevents them from running while the swap
10596 * file is active and moving the extents. Note that this also prevents
10597 * a concurrent device add which isn't actually necessary, but it's not
10598 * really worth the trouble to allow it.
10600 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10601 btrfs_warn(fs_info,
10602 "cannot activate swapfile while exclusive operation is running");
10607 * Prevent snapshot creation while we are activating the swap file.
10608 * We do not want to race with snapshot creation. If snapshot creation
10609 * already started before we bumped nr_swapfiles from 0 to 1 and
10610 * completes before the first write into the swap file after it is
10611 * activated, than that write would fallback to COW.
10613 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10614 btrfs_exclop_finish(fs_info);
10615 btrfs_warn(fs_info,
10616 "cannot activate swapfile because snapshot creation is in progress");
10620 * Snapshots can create extents which require COW even if NODATACOW is
10621 * set. We use this counter to prevent snapshots. We must increment it
10622 * before walking the extents because we don't want a concurrent
10623 * snapshot to run after we've already checked the extents.
10625 * It is possible that subvolume is marked for deletion but still not
10626 * removed yet. To prevent this race, we check the root status before
10627 * activating the swapfile.
10629 spin_lock(&root->root_item_lock);
10630 if (btrfs_root_dead(root)) {
10631 spin_unlock(&root->root_item_lock);
10633 btrfs_exclop_finish(fs_info);
10634 btrfs_warn(fs_info,
10635 "cannot activate swapfile because subvolume %llu is being deleted",
10636 root->root_key.objectid);
10639 atomic_inc(&root->nr_swapfiles);
10640 spin_unlock(&root->root_item_lock);
10642 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10644 lock_extent(io_tree, 0, isize - 1, &cached_state);
10646 while (start < isize) {
10647 u64 logical_block_start, physical_block_start;
10648 struct btrfs_block_group *bg;
10649 u64 len = isize - start;
10651 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10657 if (em->block_start == EXTENT_MAP_HOLE) {
10658 btrfs_warn(fs_info, "swapfile must not have holes");
10662 if (em->block_start == EXTENT_MAP_INLINE) {
10664 * It's unlikely we'll ever actually find ourselves
10665 * here, as a file small enough to fit inline won't be
10666 * big enough to store more than the swap header, but in
10667 * case something changes in the future, let's catch it
10668 * here rather than later.
10670 btrfs_warn(fs_info, "swapfile must not be inline");
10674 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10675 btrfs_warn(fs_info, "swapfile must not be compressed");
10680 logical_block_start = em->block_start + (start - em->start);
10681 len = min(len, em->len - (start - em->start));
10682 free_extent_map(em);
10685 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10691 btrfs_warn(fs_info,
10692 "swapfile must not be copy-on-write");
10697 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10703 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10704 btrfs_warn(fs_info,
10705 "swapfile must have single data profile");
10710 if (device == NULL) {
10711 device = em->map_lookup->stripes[0].dev;
10712 ret = btrfs_add_swapfile_pin(inode, device, false);
10717 } else if (device != em->map_lookup->stripes[0].dev) {
10718 btrfs_warn(fs_info, "swapfile must be on one device");
10723 physical_block_start = (em->map_lookup->stripes[0].physical +
10724 (logical_block_start - em->start));
10725 len = min(len, em->len - (logical_block_start - em->start));
10726 free_extent_map(em);
10729 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10731 btrfs_warn(fs_info,
10732 "could not find block group containing swapfile");
10737 if (!btrfs_inc_block_group_swap_extents(bg)) {
10738 btrfs_warn(fs_info,
10739 "block group for swapfile at %llu is read-only%s",
10741 atomic_read(&fs_info->scrubs_running) ?
10742 " (scrub running)" : "");
10743 btrfs_put_block_group(bg);
10748 ret = btrfs_add_swapfile_pin(inode, bg, true);
10750 btrfs_put_block_group(bg);
10757 if (bsi.block_len &&
10758 bsi.block_start + bsi.block_len == physical_block_start) {
10759 bsi.block_len += len;
10761 if (bsi.block_len) {
10762 ret = btrfs_add_swap_extent(sis, &bsi);
10767 bsi.block_start = physical_block_start;
10768 bsi.block_len = len;
10775 ret = btrfs_add_swap_extent(sis, &bsi);
10778 if (!IS_ERR_OR_NULL(em))
10779 free_extent_map(em);
10781 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10784 btrfs_swap_deactivate(file);
10786 btrfs_drew_write_unlock(&root->snapshot_lock);
10788 btrfs_exclop_finish(fs_info);
10794 sis->bdev = device->bdev;
10795 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10796 sis->max = bsi.nr_pages;
10797 sis->pages = bsi.nr_pages - 1;
10798 sis->highest_bit = bsi.nr_pages - 1;
10799 return bsi.nr_extents;
10802 static void btrfs_swap_deactivate(struct file *file)
10806 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10809 return -EOPNOTSUPP;
10814 * Update the number of bytes used in the VFS' inode. When we replace extents in
10815 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10816 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10817 * always get a correct value.
10819 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10820 const u64 add_bytes,
10821 const u64 del_bytes)
10823 if (add_bytes == del_bytes)
10826 spin_lock(&inode->lock);
10828 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10830 inode_add_bytes(&inode->vfs_inode, add_bytes);
10831 spin_unlock(&inode->lock);
10835 * Verify that there are no ordered extents for a given file range.
10837 * @inode: The target inode.
10838 * @start: Start offset of the file range, should be sector size aligned.
10839 * @end: End offset (inclusive) of the file range, its value +1 should be
10840 * sector size aligned.
10842 * This should typically be used for cases where we locked an inode's VFS lock in
10843 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10844 * we have flushed all delalloc in the range, we have waited for all ordered
10845 * extents in the range to complete and finally we have locked the file range in
10846 * the inode's io_tree.
10848 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10850 struct btrfs_root *root = inode->root;
10851 struct btrfs_ordered_extent *ordered;
10853 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10856 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10858 btrfs_err(root->fs_info,
10859 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10860 start, end, btrfs_ino(inode), root->root_key.objectid,
10861 ordered->file_offset,
10862 ordered->file_offset + ordered->num_bytes - 1);
10863 btrfs_put_ordered_extent(ordered);
10866 ASSERT(ordered == NULL);
10869 static const struct inode_operations btrfs_dir_inode_operations = {
10870 .getattr = btrfs_getattr,
10871 .lookup = btrfs_lookup,
10872 .create = btrfs_create,
10873 .unlink = btrfs_unlink,
10874 .link = btrfs_link,
10875 .mkdir = btrfs_mkdir,
10876 .rmdir = btrfs_rmdir,
10877 .rename = btrfs_rename2,
10878 .symlink = btrfs_symlink,
10879 .setattr = btrfs_setattr,
10880 .mknod = btrfs_mknod,
10881 .listxattr = btrfs_listxattr,
10882 .permission = btrfs_permission,
10883 .get_inode_acl = btrfs_get_acl,
10884 .set_acl = btrfs_set_acl,
10885 .update_time = btrfs_update_time,
10886 .tmpfile = btrfs_tmpfile,
10887 .fileattr_get = btrfs_fileattr_get,
10888 .fileattr_set = btrfs_fileattr_set,
10891 static const struct file_operations btrfs_dir_file_operations = {
10892 .llseek = btrfs_dir_llseek,
10893 .read = generic_read_dir,
10894 .iterate_shared = btrfs_real_readdir,
10895 .open = btrfs_opendir,
10896 .unlocked_ioctl = btrfs_ioctl,
10897 #ifdef CONFIG_COMPAT
10898 .compat_ioctl = btrfs_compat_ioctl,
10900 .release = btrfs_release_file,
10901 .fsync = btrfs_sync_file,
10905 * btrfs doesn't support the bmap operation because swapfiles
10906 * use bmap to make a mapping of extents in the file. They assume
10907 * these extents won't change over the life of the file and they
10908 * use the bmap result to do IO directly to the drive.
10910 * the btrfs bmap call would return logical addresses that aren't
10911 * suitable for IO and they also will change frequently as COW
10912 * operations happen. So, swapfile + btrfs == corruption.
10914 * For now we're avoiding this by dropping bmap.
10916 static const struct address_space_operations btrfs_aops = {
10917 .read_folio = btrfs_read_folio,
10918 .writepages = btrfs_writepages,
10919 .readahead = btrfs_readahead,
10920 .invalidate_folio = btrfs_invalidate_folio,
10921 .release_folio = btrfs_release_folio,
10922 .migrate_folio = btrfs_migrate_folio,
10923 .dirty_folio = filemap_dirty_folio,
10924 .error_remove_page = generic_error_remove_page,
10925 .swap_activate = btrfs_swap_activate,
10926 .swap_deactivate = btrfs_swap_deactivate,
10929 static const struct inode_operations btrfs_file_inode_operations = {
10930 .getattr = btrfs_getattr,
10931 .setattr = btrfs_setattr,
10932 .listxattr = btrfs_listxattr,
10933 .permission = btrfs_permission,
10934 .fiemap = btrfs_fiemap,
10935 .get_inode_acl = btrfs_get_acl,
10936 .set_acl = btrfs_set_acl,
10937 .update_time = btrfs_update_time,
10938 .fileattr_get = btrfs_fileattr_get,
10939 .fileattr_set = btrfs_fileattr_set,
10941 static const struct inode_operations btrfs_special_inode_operations = {
10942 .getattr = btrfs_getattr,
10943 .setattr = btrfs_setattr,
10944 .permission = btrfs_permission,
10945 .listxattr = btrfs_listxattr,
10946 .get_inode_acl = btrfs_get_acl,
10947 .set_acl = btrfs_set_acl,
10948 .update_time = btrfs_update_time,
10950 static const struct inode_operations btrfs_symlink_inode_operations = {
10951 .get_link = page_get_link,
10952 .getattr = btrfs_getattr,
10953 .setattr = btrfs_setattr,
10954 .permission = btrfs_permission,
10955 .listxattr = btrfs_listxattr,
10956 .update_time = btrfs_update_time,
10959 const struct dentry_operations btrfs_dentry_operations = {
10960 .d_delete = btrfs_dentry_delete,