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"
75 struct btrfs_iget_args {
77 struct btrfs_root *root;
80 struct btrfs_dio_data {
82 struct extent_changeset *data_reserved;
83 struct btrfs_ordered_extent *ordered;
84 bool data_space_reserved;
88 struct btrfs_dio_private {
93 /* This must be last */
94 struct btrfs_bio bbio;
97 static struct bio_set btrfs_dio_bioset;
99 struct btrfs_rename_ctx {
100 /* Output field. Stores the index number of the old directory entry. */
105 * Used by data_reloc_print_warning_inode() to pass needed info for filename
106 * resolution and output of error message.
108 struct data_reloc_warn {
109 struct btrfs_path path;
110 struct btrfs_fs_info *fs_info;
111 u64 extent_item_size;
116 static const struct inode_operations btrfs_dir_inode_operations;
117 static const struct inode_operations btrfs_symlink_inode_operations;
118 static const struct inode_operations btrfs_special_inode_operations;
119 static const struct inode_operations btrfs_file_inode_operations;
120 static const struct address_space_operations btrfs_aops;
121 static const struct file_operations btrfs_dir_file_operations;
123 static struct kmem_cache *btrfs_inode_cachep;
125 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
126 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback);
128 static noinline int cow_file_range(struct btrfs_inode *inode,
129 struct page *locked_page,
130 u64 start, u64 end, u64 *done_offset,
131 bool keep_locked, bool no_inline);
132 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
133 u64 len, u64 orig_start, u64 block_start,
134 u64 block_len, u64 orig_block_len,
135 u64 ram_bytes, int compress_type,
138 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
139 u64 root, void *warn_ctx)
141 struct data_reloc_warn *warn = warn_ctx;
142 struct btrfs_fs_info *fs_info = warn->fs_info;
143 struct extent_buffer *eb;
144 struct btrfs_inode_item *inode_item;
145 struct inode_fs_paths *ipath = NULL;
146 struct btrfs_root *local_root;
147 struct btrfs_key key;
148 unsigned int nofs_flag;
152 local_root = btrfs_get_fs_root(fs_info, root, true);
153 if (IS_ERR(local_root)) {
154 ret = PTR_ERR(local_root);
158 /* This makes the path point to (inum INODE_ITEM ioff). */
160 key.type = BTRFS_INODE_ITEM_KEY;
163 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
165 btrfs_put_root(local_root);
166 btrfs_release_path(&warn->path);
170 eb = warn->path.nodes[0];
171 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
172 nlink = btrfs_inode_nlink(eb, inode_item);
173 btrfs_release_path(&warn->path);
175 nofs_flag = memalloc_nofs_save();
176 ipath = init_ipath(4096, local_root, &warn->path);
177 memalloc_nofs_restore(nofs_flag);
179 btrfs_put_root(local_root);
180 ret = PTR_ERR(ipath);
183 * -ENOMEM, not a critical error, just output an generic error
187 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
188 warn->logical, warn->mirror_num, root, inum, offset);
191 ret = paths_from_inode(inum, ipath);
196 * We deliberately ignore the bit ipath might have been too small to
197 * hold all of the paths here
199 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
201 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
202 warn->logical, warn->mirror_num, root, inum, offset,
203 fs_info->sectorsize, nlink,
204 (char *)(unsigned long)ipath->fspath->val[i]);
207 btrfs_put_root(local_root);
213 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
214 warn->logical, warn->mirror_num, root, inum, offset, ret);
221 * Do extra user-friendly error output (e.g. lookup all the affected files).
223 * Return true if we succeeded doing the backref lookup.
224 * Return false if such lookup failed, and has to fallback to the old error message.
226 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
227 const u8 *csum, const u8 *csum_expected,
230 struct btrfs_fs_info *fs_info = inode->root->fs_info;
231 struct btrfs_path path = { 0 };
232 struct btrfs_key found_key = { 0 };
233 struct extent_buffer *eb;
234 struct btrfs_extent_item *ei;
235 const u32 csum_size = fs_info->csum_size;
241 mutex_lock(&fs_info->reloc_mutex);
242 logical = btrfs_get_reloc_bg_bytenr(fs_info);
243 mutex_unlock(&fs_info->reloc_mutex);
245 if (logical == U64_MAX) {
246 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
247 btrfs_warn_rl(fs_info,
248 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
249 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
250 CSUM_FMT_VALUE(csum_size, csum),
251 CSUM_FMT_VALUE(csum_size, csum_expected),
257 btrfs_warn_rl(fs_info,
258 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
259 inode->root->root_key.objectid,
260 btrfs_ino(inode), file_off, logical,
261 CSUM_FMT_VALUE(csum_size, csum),
262 CSUM_FMT_VALUE(csum_size, csum_expected),
265 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
267 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
272 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
273 item_size = btrfs_item_size(eb, path.slots[0]);
274 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
275 unsigned long ptr = 0;
280 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
281 item_size, &ref_root,
284 btrfs_warn_rl(fs_info,
285 "failed to resolve tree backref for logical %llu: %d",
292 btrfs_warn_rl(fs_info,
293 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
295 (ref_level ? "node" : "leaf"),
296 ref_level, ref_root);
298 btrfs_release_path(&path);
300 struct btrfs_backref_walk_ctx ctx = { 0 };
301 struct data_reloc_warn reloc_warn = { 0 };
303 btrfs_release_path(&path);
305 ctx.bytenr = found_key.objectid;
306 ctx.extent_item_pos = logical - found_key.objectid;
307 ctx.fs_info = fs_info;
309 reloc_warn.logical = logical;
310 reloc_warn.extent_item_size = found_key.offset;
311 reloc_warn.mirror_num = mirror_num;
312 reloc_warn.fs_info = fs_info;
314 iterate_extent_inodes(&ctx, true,
315 data_reloc_print_warning_inode, &reloc_warn);
319 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
320 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
322 struct btrfs_root *root = inode->root;
323 const u32 csum_size = root->fs_info->csum_size;
325 /* For data reloc tree, it's better to do a backref lookup instead. */
326 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
327 return print_data_reloc_error(inode, logical_start, csum,
328 csum_expected, mirror_num);
330 /* Output without objectid, which is more meaningful */
331 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
332 btrfs_warn_rl(root->fs_info,
333 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
334 root->root_key.objectid, btrfs_ino(inode),
336 CSUM_FMT_VALUE(csum_size, csum),
337 CSUM_FMT_VALUE(csum_size, csum_expected),
340 btrfs_warn_rl(root->fs_info,
341 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
342 root->root_key.objectid, btrfs_ino(inode),
344 CSUM_FMT_VALUE(csum_size, csum),
345 CSUM_FMT_VALUE(csum_size, csum_expected),
351 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
353 * ilock_flags can have the following bit set:
355 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
356 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
358 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
360 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
362 if (ilock_flags & BTRFS_ILOCK_SHARED) {
363 if (ilock_flags & BTRFS_ILOCK_TRY) {
364 if (!inode_trylock_shared(&inode->vfs_inode))
369 inode_lock_shared(&inode->vfs_inode);
371 if (ilock_flags & BTRFS_ILOCK_TRY) {
372 if (!inode_trylock(&inode->vfs_inode))
377 inode_lock(&inode->vfs_inode);
379 if (ilock_flags & BTRFS_ILOCK_MMAP)
380 down_write(&inode->i_mmap_lock);
385 * btrfs_inode_unlock - unock inode i_rwsem
387 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
388 * to decide whether the lock acquired is shared or exclusive.
390 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
392 if (ilock_flags & BTRFS_ILOCK_MMAP)
393 up_write(&inode->i_mmap_lock);
394 if (ilock_flags & BTRFS_ILOCK_SHARED)
395 inode_unlock_shared(&inode->vfs_inode);
397 inode_unlock(&inode->vfs_inode);
401 * Cleanup all submitted ordered extents in specified range to handle errors
402 * from the btrfs_run_delalloc_range() callback.
404 * NOTE: caller must ensure that when an error happens, it can not call
405 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
406 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
407 * to be released, which we want to happen only when finishing the ordered
408 * extent (btrfs_finish_ordered_io()).
410 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
411 struct page *locked_page,
412 u64 offset, u64 bytes)
414 unsigned long index = offset >> PAGE_SHIFT;
415 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
416 u64 page_start = 0, page_end = 0;
420 page_start = page_offset(locked_page);
421 page_end = page_start + PAGE_SIZE - 1;
424 while (index <= end_index) {
426 * For locked page, we will call btrfs_mark_ordered_io_finished
427 * through btrfs_mark_ordered_io_finished() on it
428 * in run_delalloc_range() for the error handling, which will
429 * clear page Ordered and run the ordered extent accounting.
431 * Here we can't just clear the Ordered bit, or
432 * btrfs_mark_ordered_io_finished() would skip the accounting
433 * for the page range, and the ordered extent will never finish.
435 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
439 page = find_get_page(inode->vfs_inode.i_mapping, index);
445 * Here we just clear all Ordered bits for every page in the
446 * range, then btrfs_mark_ordered_io_finished() will handle
447 * the ordered extent accounting for the range.
449 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
455 /* The locked page covers the full range, nothing needs to be done */
456 if (bytes + offset <= page_start + PAGE_SIZE)
459 * In case this page belongs to the delalloc range being
460 * instantiated then skip it, since the first page of a range is
461 * going to be properly cleaned up by the caller of
464 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
465 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
466 offset = page_offset(locked_page) + PAGE_SIZE;
470 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
473 static int btrfs_dirty_inode(struct btrfs_inode *inode);
475 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
476 struct btrfs_new_inode_args *args)
480 if (args->default_acl) {
481 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
487 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
491 if (!args->default_acl && !args->acl)
492 cache_no_acl(args->inode);
493 return btrfs_xattr_security_init(trans, args->inode, args->dir,
494 &args->dentry->d_name);
498 * this does all the hard work for inserting an inline extent into
499 * the btree. The caller should have done a btrfs_drop_extents so that
500 * no overlapping inline items exist in the btree
502 static int insert_inline_extent(struct btrfs_trans_handle *trans,
503 struct btrfs_path *path,
504 struct btrfs_inode *inode, bool extent_inserted,
505 size_t size, size_t compressed_size,
507 struct page **compressed_pages,
510 struct btrfs_root *root = inode->root;
511 struct extent_buffer *leaf;
512 struct page *page = NULL;
515 struct btrfs_file_extent_item *ei;
517 size_t cur_size = size;
520 ASSERT((compressed_size > 0 && compressed_pages) ||
521 (compressed_size == 0 && !compressed_pages));
523 if (compressed_size && compressed_pages)
524 cur_size = compressed_size;
526 if (!extent_inserted) {
527 struct btrfs_key key;
530 key.objectid = btrfs_ino(inode);
532 key.type = BTRFS_EXTENT_DATA_KEY;
534 datasize = btrfs_file_extent_calc_inline_size(cur_size);
535 ret = btrfs_insert_empty_item(trans, root, path, &key,
540 leaf = path->nodes[0];
541 ei = btrfs_item_ptr(leaf, path->slots[0],
542 struct btrfs_file_extent_item);
543 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
544 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
545 btrfs_set_file_extent_encryption(leaf, ei, 0);
546 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
547 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
548 ptr = btrfs_file_extent_inline_start(ei);
550 if (compress_type != BTRFS_COMPRESS_NONE) {
553 while (compressed_size > 0) {
554 cpage = compressed_pages[i];
555 cur_size = min_t(unsigned long, compressed_size,
558 kaddr = kmap_local_page(cpage);
559 write_extent_buffer(leaf, kaddr, ptr, cur_size);
564 compressed_size -= cur_size;
566 btrfs_set_file_extent_compression(leaf, ei,
569 page = find_get_page(inode->vfs_inode.i_mapping, 0);
570 btrfs_set_file_extent_compression(leaf, ei, 0);
571 kaddr = kmap_local_page(page);
572 write_extent_buffer(leaf, kaddr, ptr, size);
576 btrfs_mark_buffer_dirty(leaf);
577 btrfs_release_path(path);
580 * We align size to sectorsize for inline extents just for simplicity
583 ret = btrfs_inode_set_file_extent_range(inode, 0,
584 ALIGN(size, root->fs_info->sectorsize));
589 * We're an inline extent, so nobody can extend the file past i_size
590 * without locking a page we already have locked.
592 * We must do any i_size and inode updates before we unlock the pages.
593 * Otherwise we could end up racing with unlink.
595 i_size = i_size_read(&inode->vfs_inode);
596 if (update_i_size && size > i_size) {
597 i_size_write(&inode->vfs_inode, size);
600 inode->disk_i_size = i_size;
608 * conditionally insert an inline extent into the file. This
609 * does the checks required to make sure the data is small enough
610 * to fit as an inline extent.
612 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
613 size_t compressed_size,
615 struct page **compressed_pages,
618 struct btrfs_drop_extents_args drop_args = { 0 };
619 struct btrfs_root *root = inode->root;
620 struct btrfs_fs_info *fs_info = root->fs_info;
621 struct btrfs_trans_handle *trans;
622 u64 data_len = (compressed_size ?: size);
624 struct btrfs_path *path;
627 * We can create an inline extent if it ends at or beyond the current
628 * i_size, is no larger than a sector (decompressed), and the (possibly
629 * compressed) data fits in a leaf and the configured maximum inline
632 if (size < i_size_read(&inode->vfs_inode) ||
633 size > fs_info->sectorsize ||
634 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
635 data_len > fs_info->max_inline)
638 path = btrfs_alloc_path();
642 trans = btrfs_join_transaction(root);
644 btrfs_free_path(path);
645 return PTR_ERR(trans);
647 trans->block_rsv = &inode->block_rsv;
649 drop_args.path = path;
651 drop_args.end = fs_info->sectorsize;
652 drop_args.drop_cache = true;
653 drop_args.replace_extent = true;
654 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
655 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
657 btrfs_abort_transaction(trans, ret);
661 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
662 size, compressed_size, compress_type,
663 compressed_pages, update_i_size);
664 if (ret && ret != -ENOSPC) {
665 btrfs_abort_transaction(trans, ret);
667 } else if (ret == -ENOSPC) {
672 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
673 ret = btrfs_update_inode(trans, root, inode);
674 if (ret && ret != -ENOSPC) {
675 btrfs_abort_transaction(trans, ret);
677 } else if (ret == -ENOSPC) {
682 btrfs_set_inode_full_sync(inode);
685 * Don't forget to free the reserved space, as for inlined extent
686 * it won't count as data extent, free them directly here.
687 * And at reserve time, it's always aligned to page size, so
688 * just free one page here.
690 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
691 btrfs_free_path(path);
692 btrfs_end_transaction(trans);
696 struct async_extent {
701 unsigned long nr_pages;
703 struct list_head list;
707 struct btrfs_inode *inode;
708 struct page *locked_page;
711 blk_opf_t write_flags;
712 struct list_head extents;
713 struct cgroup_subsys_state *blkcg_css;
714 struct btrfs_work work;
715 struct async_cow *async_cow;
720 struct async_chunk chunks[];
723 static noinline int add_async_extent(struct async_chunk *cow,
724 u64 start, u64 ram_size,
727 unsigned long nr_pages,
730 struct async_extent *async_extent;
732 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
733 BUG_ON(!async_extent); /* -ENOMEM */
734 async_extent->start = start;
735 async_extent->ram_size = ram_size;
736 async_extent->compressed_size = compressed_size;
737 async_extent->pages = pages;
738 async_extent->nr_pages = nr_pages;
739 async_extent->compress_type = compress_type;
740 list_add_tail(&async_extent->list, &cow->extents);
745 * Check if the inode needs to be submitted to compression, based on mount
746 * options, defragmentation, properties or heuristics.
748 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
751 struct btrfs_fs_info *fs_info = inode->root->fs_info;
753 if (!btrfs_inode_can_compress(inode)) {
754 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
755 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
760 * Special check for subpage.
762 * We lock the full page then run each delalloc range in the page, thus
763 * for the following case, we will hit some subpage specific corner case:
766 * | |///////| |///////|
769 * In above case, both range A and range B will try to unlock the full
770 * page [0, 64K), causing the one finished later will have page
771 * unlocked already, triggering various page lock requirement BUG_ON()s.
773 * So here we add an artificial limit that subpage compression can only
774 * if the range is fully page aligned.
776 * In theory we only need to ensure the first page is fully covered, but
777 * the tailing partial page will be locked until the full compression
778 * finishes, delaying the write of other range.
780 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
781 * first to prevent any submitted async extent to unlock the full page.
782 * By this, we can ensure for subpage case that only the last async_cow
783 * will unlock the full page.
785 if (fs_info->sectorsize < PAGE_SIZE) {
786 if (!PAGE_ALIGNED(start) ||
787 !PAGE_ALIGNED(end + 1))
792 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
795 if (inode->defrag_compress)
797 /* bad compression ratios */
798 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
800 if (btrfs_test_opt(fs_info, COMPRESS) ||
801 inode->flags & BTRFS_INODE_COMPRESS ||
802 inode->prop_compress)
803 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
807 static inline void inode_should_defrag(struct btrfs_inode *inode,
808 u64 start, u64 end, u64 num_bytes, u32 small_write)
810 /* If this is a small write inside eof, kick off a defrag */
811 if (num_bytes < small_write &&
812 (start > 0 || end + 1 < inode->disk_i_size))
813 btrfs_add_inode_defrag(NULL, inode, small_write);
817 * Work queue call back to started compression on a file and pages.
819 * This is done inside an ordered work queue, and the compression is spread
820 * across many cpus. The actual IO submission is step two, and the ordered work
821 * queue takes care of making sure that happens in the same order things were
822 * put onto the queue by writepages and friends.
824 * If this code finds it can't get good compression, it puts an entry onto the
825 * work queue to write the uncompressed bytes. This makes sure that both
826 * compressed inodes and uncompressed inodes are written in the same order that
827 * the flusher thread sent them down.
829 static void compress_file_range(struct btrfs_work *work)
831 struct async_chunk *async_chunk =
832 container_of(work, struct async_chunk, work);
833 struct btrfs_inode *inode = async_chunk->inode;
834 struct btrfs_fs_info *fs_info = inode->root->fs_info;
835 struct address_space *mapping = inode->vfs_inode.i_mapping;
836 u64 blocksize = fs_info->sectorsize;
837 u64 start = async_chunk->start;
838 u64 end = async_chunk->end;
843 unsigned long nr_pages;
844 unsigned long total_compressed = 0;
845 unsigned long total_in = 0;
848 int compress_type = fs_info->compress_type;
850 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
853 * We need to call clear_page_dirty_for_io on each page in the range.
854 * Otherwise applications with the file mmap'd can wander in and change
855 * the page contents while we are compressing them.
857 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
860 * We need to save i_size before now because it could change in between
861 * us evaluating the size and assigning it. This is because we lock and
862 * unlock the page in truncate and fallocate, and then modify the i_size
865 * The barriers are to emulate READ_ONCE, remove that once i_size_read
869 i_size = i_size_read(&inode->vfs_inode);
871 actual_end = min_t(u64, i_size, end + 1);
874 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
875 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
878 * we don't want to send crud past the end of i_size through
879 * compression, that's just a waste of CPU time. So, if the
880 * end of the file is before the start of our current
881 * requested range of bytes, we bail out to the uncompressed
882 * cleanup code that can deal with all of this.
884 * It isn't really the fastest way to fix things, but this is a
885 * very uncommon corner.
887 if (actual_end <= start)
888 goto cleanup_and_bail_uncompressed;
890 total_compressed = actual_end - start;
893 * Skip compression for a small file range(<=blocksize) that
894 * isn't an inline extent, since it doesn't save disk space at all.
896 if (total_compressed <= blocksize &&
897 (start > 0 || end + 1 < inode->disk_i_size))
898 goto cleanup_and_bail_uncompressed;
901 * For subpage case, we require full page alignment for the sector
903 * Thus we must also check against @actual_end, not just @end.
905 if (blocksize < PAGE_SIZE) {
906 if (!PAGE_ALIGNED(start) ||
907 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
908 goto cleanup_and_bail_uncompressed;
911 total_compressed = min_t(unsigned long, total_compressed,
912 BTRFS_MAX_UNCOMPRESSED);
917 * We do compression for mount -o compress and when the inode has not
918 * been flagged as NOCOMPRESS. This flag can change at any time if we
919 * discover bad compression ratios.
921 if (!inode_need_compress(inode, start, end))
922 goto cleanup_and_bail_uncompressed;
924 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
927 * Memory allocation failure is not a fatal error, we can fall
928 * back to uncompressed code.
930 goto cleanup_and_bail_uncompressed;
933 if (inode->defrag_compress)
934 compress_type = inode->defrag_compress;
935 else if (inode->prop_compress)
936 compress_type = inode->prop_compress;
938 /* Compression level is applied here. */
939 ret = btrfs_compress_pages(compress_type | (fs_info->compress_level << 4),
940 mapping, start, pages, &nr_pages, &total_in,
943 goto mark_incompressible;
946 * Zero the tail end of the last page, as we might be sending it down
949 poff = offset_in_page(total_compressed);
951 memzero_page(pages[nr_pages - 1], poff, PAGE_SIZE - poff);
954 * Try to create an inline extent.
956 * If we didn't compress the entire range, try to create an uncompressed
957 * inline extent, else a compressed one.
959 * Check cow_file_range() for why we don't even try to create inline
960 * extent for the subpage case.
962 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
963 if (total_in < actual_end) {
964 ret = cow_file_range_inline(inode, actual_end, 0,
965 BTRFS_COMPRESS_NONE, NULL,
968 ret = cow_file_range_inline(inode, actual_end,
970 compress_type, pages,
974 unsigned long clear_flags = EXTENT_DELALLOC |
975 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
976 EXTENT_DO_ACCOUNTING;
979 mapping_set_error(mapping, -EIO);
982 * inline extent creation worked or returned error,
983 * we don't need to create any more async work items.
984 * Unlock and free up our temp pages.
986 * We use DO_ACCOUNTING here because we need the
987 * delalloc_release_metadata to be done _after_ we drop
988 * our outstanding extent for clearing delalloc for this
991 extent_clear_unlock_delalloc(inode, start, end,
995 PAGE_START_WRITEBACK |
1002 * We aren't doing an inline extent. Round the compressed size up to a
1003 * block size boundary so the allocator does sane things.
1005 total_compressed = ALIGN(total_compressed, blocksize);
1008 * One last check to make sure the compression is really a win, compare
1009 * the page count read with the blocks on disk, compression must free at
1012 total_in = round_up(total_in, fs_info->sectorsize);
1013 if (total_compressed + blocksize > total_in)
1014 goto mark_incompressible;
1017 * The async work queues will take care of doing actual allocation on
1018 * disk for these compressed pages, and will submit the bios.
1020 add_async_extent(async_chunk, start, total_in, total_compressed, pages,
1021 nr_pages, compress_type);
1022 if (start + total_in < end) {
1029 mark_incompressible:
1030 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) && !inode->prop_compress)
1031 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1032 cleanup_and_bail_uncompressed:
1033 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1034 BTRFS_COMPRESS_NONE);
1037 for (i = 0; i < nr_pages; i++) {
1038 WARN_ON(pages[i]->mapping);
1045 static void free_async_extent_pages(struct async_extent *async_extent)
1049 if (!async_extent->pages)
1052 for (i = 0; i < async_extent->nr_pages; i++) {
1053 WARN_ON(async_extent->pages[i]->mapping);
1054 put_page(async_extent->pages[i]);
1056 kfree(async_extent->pages);
1057 async_extent->nr_pages = 0;
1058 async_extent->pages = NULL;
1061 static void submit_uncompressed_range(struct btrfs_inode *inode,
1062 struct async_extent *async_extent,
1063 struct page *locked_page)
1065 u64 start = async_extent->start;
1066 u64 end = async_extent->start + async_extent->ram_size - 1;
1068 struct writeback_control wbc = {
1069 .sync_mode = WB_SYNC_ALL,
1070 .range_start = start,
1072 .no_cgroup_owner = 1,
1076 * Call cow_file_range() to run the delalloc range directly, since we
1077 * won't go to NOCOW or async path again.
1079 * Also we call cow_file_range() with @unlock_page == 0, so that we
1080 * can directly submit them without interruption.
1082 ret = cow_file_range(inode, locked_page, start, end, NULL, true, false);
1083 /* Inline extent inserted, page gets unlocked and everything is done */
1088 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1090 const u64 page_start = page_offset(locked_page);
1092 set_page_writeback(locked_page);
1093 end_page_writeback(locked_page);
1094 btrfs_mark_ordered_io_finished(inode, locked_page,
1095 page_start, PAGE_SIZE,
1097 btrfs_page_clear_uptodate(inode->root->fs_info,
1098 locked_page, page_start,
1100 mapping_set_error(locked_page->mapping, ret);
1101 unlock_page(locked_page);
1106 /* All pages will be unlocked, including @locked_page */
1107 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1108 extent_write_locked_range(&inode->vfs_inode, start, end, &wbc, false);
1109 wbc_detach_inode(&wbc);
1112 static void submit_one_async_extent(struct async_chunk *async_chunk,
1113 struct async_extent *async_extent,
1116 struct btrfs_inode *inode = async_chunk->inode;
1117 struct extent_io_tree *io_tree = &inode->io_tree;
1118 struct btrfs_root *root = inode->root;
1119 struct btrfs_fs_info *fs_info = root->fs_info;
1120 struct btrfs_ordered_extent *ordered;
1121 struct btrfs_key ins;
1122 struct page *locked_page = NULL;
1123 struct extent_map *em;
1125 u64 start = async_extent->start;
1126 u64 end = async_extent->start + async_extent->ram_size - 1;
1128 if (async_chunk->blkcg_css)
1129 kthread_associate_blkcg(async_chunk->blkcg_css);
1132 * If async_chunk->locked_page is in the async_extent range, we need to
1135 if (async_chunk->locked_page) {
1136 u64 locked_page_start = page_offset(async_chunk->locked_page);
1137 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1139 if (!(start >= locked_page_end || end <= locked_page_start))
1140 locked_page = async_chunk->locked_page;
1142 lock_extent(io_tree, start, end, NULL);
1144 if (async_extent->compress_type == BTRFS_COMPRESS_NONE) {
1145 submit_uncompressed_range(inode, async_extent, locked_page);
1149 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1150 async_extent->compressed_size,
1151 async_extent->compressed_size,
1152 0, *alloc_hint, &ins, 1, 1);
1155 * Here we used to try again by going back to non-compressed
1156 * path for ENOSPC. But we can't reserve space even for
1157 * compressed size, how could it work for uncompressed size
1158 * which requires larger size? So here we directly go error
1164 /* Here we're doing allocation and writeback of the compressed pages */
1165 em = create_io_em(inode, start,
1166 async_extent->ram_size, /* len */
1167 start, /* orig_start */
1168 ins.objectid, /* block_start */
1169 ins.offset, /* block_len */
1170 ins.offset, /* orig_block_len */
1171 async_extent->ram_size, /* ram_bytes */
1172 async_extent->compress_type,
1173 BTRFS_ORDERED_COMPRESSED);
1176 goto out_free_reserve;
1178 free_extent_map(em);
1180 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1181 async_extent->ram_size, /* num_bytes */
1182 async_extent->ram_size, /* ram_bytes */
1183 ins.objectid, /* disk_bytenr */
1184 ins.offset, /* disk_num_bytes */
1186 1 << BTRFS_ORDERED_COMPRESSED,
1187 async_extent->compress_type);
1188 if (IS_ERR(ordered)) {
1189 btrfs_drop_extent_map_range(inode, start, end, false);
1190 ret = PTR_ERR(ordered);
1191 goto out_free_reserve;
1193 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1195 /* Clear dirty, set writeback and unlock the pages. */
1196 extent_clear_unlock_delalloc(inode, start, end,
1197 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1198 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1199 btrfs_submit_compressed_write(ordered,
1200 async_extent->pages, /* compressed_pages */
1201 async_extent->nr_pages,
1202 async_chunk->write_flags, true);
1203 *alloc_hint = ins.objectid + ins.offset;
1205 if (async_chunk->blkcg_css)
1206 kthread_associate_blkcg(NULL);
1207 kfree(async_extent);
1211 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1212 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1214 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1215 extent_clear_unlock_delalloc(inode, start, end,
1216 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1217 EXTENT_DELALLOC_NEW |
1218 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1219 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1220 PAGE_END_WRITEBACK);
1221 free_async_extent_pages(async_extent);
1222 if (async_chunk->blkcg_css)
1223 kthread_associate_blkcg(NULL);
1224 btrfs_debug(fs_info,
1225 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1226 root->root_key.objectid, btrfs_ino(inode), start,
1227 async_extent->ram_size, ret);
1228 kfree(async_extent);
1231 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1234 struct extent_map_tree *em_tree = &inode->extent_tree;
1235 struct extent_map *em;
1238 read_lock(&em_tree->lock);
1239 em = search_extent_mapping(em_tree, start, num_bytes);
1242 * if block start isn't an actual block number then find the
1243 * first block in this inode and use that as a hint. If that
1244 * block is also bogus then just don't worry about it.
1246 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1247 free_extent_map(em);
1248 em = search_extent_mapping(em_tree, 0, 0);
1249 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1250 alloc_hint = em->block_start;
1252 free_extent_map(em);
1254 alloc_hint = em->block_start;
1255 free_extent_map(em);
1258 read_unlock(&em_tree->lock);
1264 * when extent_io.c finds a delayed allocation range in the file,
1265 * the call backs end up in this code. The basic idea is to
1266 * allocate extents on disk for the range, and create ordered data structs
1267 * in ram to track those extents.
1269 * locked_page is the page that writepage had locked already. We use
1270 * it to make sure we don't do extra locks or unlocks.
1272 * When this function fails, it unlocks all pages except @locked_page.
1274 * When this function successfully creates an inline extent, it returns 1 and
1275 * unlocks all pages including locked_page and starts I/O on them.
1276 * (In reality inline extents are limited to a single page, so locked_page is
1277 * the only page handled anyway).
1279 * When this function succeed and creates a normal extent, the page locking
1280 * status depends on the passed in flags:
1282 * - If @keep_locked is set, all pages are kept locked.
1283 * - Else all pages except for @locked_page are unlocked.
1285 * When a failure happens in the second or later iteration of the
1286 * while-loop, the ordered extents created in previous iterations are kept
1287 * intact. So, the caller must clean them up by calling
1288 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1291 static noinline int cow_file_range(struct btrfs_inode *inode,
1292 struct page *locked_page, u64 start, u64 end,
1294 bool keep_locked, bool no_inline)
1296 struct btrfs_root *root = inode->root;
1297 struct btrfs_fs_info *fs_info = root->fs_info;
1299 u64 orig_start = start;
1301 unsigned long ram_size;
1302 u64 cur_alloc_size = 0;
1304 u64 blocksize = fs_info->sectorsize;
1305 struct btrfs_key ins;
1306 struct extent_map *em;
1307 unsigned clear_bits;
1308 unsigned long page_ops;
1309 bool extent_reserved = false;
1312 if (btrfs_is_free_space_inode(inode)) {
1317 num_bytes = ALIGN(end - start + 1, blocksize);
1318 num_bytes = max(blocksize, num_bytes);
1319 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1321 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1324 * Due to the page size limit, for subpage we can only trigger the
1325 * writeback for the dirty sectors of page, that means data writeback
1326 * is doing more writeback than what we want.
1328 * This is especially unexpected for some call sites like fallocate,
1329 * where we only increase i_size after everything is done.
1330 * This means we can trigger inline extent even if we didn't want to.
1331 * So here we skip inline extent creation completely.
1333 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1334 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1337 /* lets try to make an inline extent */
1338 ret = cow_file_range_inline(inode, actual_end, 0,
1339 BTRFS_COMPRESS_NONE, NULL, false);
1342 * We use DO_ACCOUNTING here because we need the
1343 * delalloc_release_metadata to be run _after_ we drop
1344 * our outstanding extent for clearing delalloc for this
1347 extent_clear_unlock_delalloc(inode, start, end,
1349 EXTENT_LOCKED | EXTENT_DELALLOC |
1350 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1351 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1352 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1354 * locked_page is locked by the caller of
1355 * writepage_delalloc(), not locked by
1356 * __process_pages_contig().
1358 * We can't let __process_pages_contig() to unlock it,
1359 * as it doesn't have any subpage::writers recorded.
1361 * Here we manually unlock the page, since the caller
1362 * can't determine if it's an inline extent or a
1363 * compressed extent.
1365 unlock_page(locked_page);
1367 } else if (ret < 0) {
1372 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1375 * Relocation relies on the relocated extents to have exactly the same
1376 * size as the original extents. Normally writeback for relocation data
1377 * extents follows a NOCOW path because relocation preallocates the
1378 * extents. However, due to an operation such as scrub turning a block
1379 * group to RO mode, it may fallback to COW mode, so we must make sure
1380 * an extent allocated during COW has exactly the requested size and can
1381 * not be split into smaller extents, otherwise relocation breaks and
1382 * fails during the stage where it updates the bytenr of file extent
1385 if (btrfs_is_data_reloc_root(root))
1386 min_alloc_size = num_bytes;
1388 min_alloc_size = fs_info->sectorsize;
1390 while (num_bytes > 0) {
1391 struct btrfs_ordered_extent *ordered;
1393 cur_alloc_size = num_bytes;
1394 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1395 min_alloc_size, 0, alloc_hint,
1399 cur_alloc_size = ins.offset;
1400 extent_reserved = true;
1402 ram_size = ins.offset;
1403 em = create_io_em(inode, start, ins.offset, /* len */
1404 start, /* orig_start */
1405 ins.objectid, /* block_start */
1406 ins.offset, /* block_len */
1407 ins.offset, /* orig_block_len */
1408 ram_size, /* ram_bytes */
1409 BTRFS_COMPRESS_NONE, /* compress_type */
1410 BTRFS_ORDERED_REGULAR /* type */);
1415 free_extent_map(em);
1417 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1418 ram_size, ins.objectid, cur_alloc_size,
1419 0, 1 << BTRFS_ORDERED_REGULAR,
1420 BTRFS_COMPRESS_NONE);
1421 if (IS_ERR(ordered)) {
1422 ret = PTR_ERR(ordered);
1423 goto out_drop_extent_cache;
1426 if (btrfs_is_data_reloc_root(root)) {
1427 ret = btrfs_reloc_clone_csums(ordered);
1430 * Only drop cache here, and process as normal.
1432 * We must not allow extent_clear_unlock_delalloc()
1433 * at out_unlock label to free meta of this ordered
1434 * extent, as its meta should be freed by
1435 * btrfs_finish_ordered_io().
1437 * So we must continue until @start is increased to
1438 * skip current ordered extent.
1441 btrfs_drop_extent_map_range(inode, start,
1442 start + ram_size - 1,
1445 btrfs_put_ordered_extent(ordered);
1447 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1450 * We're not doing compressed IO, don't unlock the first page
1451 * (which the caller expects to stay locked), don't clear any
1452 * dirty bits and don't set any writeback bits
1454 * Do set the Ordered (Private2) bit so we know this page was
1455 * properly setup for writepage.
1457 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1458 page_ops |= PAGE_SET_ORDERED;
1460 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1462 EXTENT_LOCKED | EXTENT_DELALLOC,
1464 if (num_bytes < cur_alloc_size)
1467 num_bytes -= cur_alloc_size;
1468 alloc_hint = ins.objectid + ins.offset;
1469 start += cur_alloc_size;
1470 extent_reserved = false;
1473 * btrfs_reloc_clone_csums() error, since start is increased
1474 * extent_clear_unlock_delalloc() at out_unlock label won't
1475 * free metadata of current ordered extent, we're OK to exit.
1482 out_drop_extent_cache:
1483 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1485 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1486 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1489 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1490 * caller to write out the successfully allocated region and retry.
1492 if (done_offset && ret == -EAGAIN) {
1493 if (orig_start < start)
1494 *done_offset = start - 1;
1496 *done_offset = start;
1498 } else if (ret == -EAGAIN) {
1499 /* Convert to -ENOSPC since the caller cannot retry. */
1504 * Now, we have three regions to clean up:
1506 * |-------(1)----|---(2)---|-------------(3)----------|
1507 * `- orig_start `- start `- start + cur_alloc_size `- end
1509 * We process each region below.
1512 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1513 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1514 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1517 * For the range (1). We have already instantiated the ordered extents
1518 * for this region. They are cleaned up by
1519 * btrfs_cleanup_ordered_extents() in e.g,
1520 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1521 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1522 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1525 * However, in case of @keep_locked, we still need to unlock the pages
1526 * (except @locked_page) to ensure all the pages are unlocked.
1528 if (keep_locked && orig_start < start) {
1530 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1531 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1532 locked_page, 0, page_ops);
1536 * For the range (2). If we reserved an extent for our delalloc range
1537 * (or a subrange) and failed to create the respective ordered extent,
1538 * then it means that when we reserved the extent we decremented the
1539 * extent's size from the data space_info's bytes_may_use counter and
1540 * incremented the space_info's bytes_reserved counter by the same
1541 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1542 * to decrement again the data space_info's bytes_may_use counter,
1543 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1545 if (extent_reserved) {
1546 extent_clear_unlock_delalloc(inode, start,
1547 start + cur_alloc_size - 1,
1551 start += cur_alloc_size;
1555 * For the range (3). We never touched the region. In addition to the
1556 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1557 * space_info's bytes_may_use counter, reserved in
1558 * btrfs_check_data_free_space().
1561 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1562 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1563 clear_bits, page_ops);
1569 * Phase two of compressed writeback. This is the ordered portion of the code,
1570 * which only gets called in the order the work was queued. We walk all the
1571 * async extents created by compress_file_range and send them down to the disk.
1573 static noinline void submit_compressed_extents(struct btrfs_work *work)
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;
1582 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1585 while (!list_empty(&async_chunk->extents)) {
1586 async_extent = list_entry(async_chunk->extents.next,
1587 struct async_extent, list);
1588 list_del(&async_extent->list);
1589 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1592 /* atomic_sub_return implies a barrier */
1593 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1595 cond_wake_up_nomb(&fs_info->async_submit_wait);
1598 static noinline void async_cow_free(struct btrfs_work *work)
1600 struct async_chunk *async_chunk;
1601 struct async_cow *async_cow;
1603 async_chunk = container_of(work, struct async_chunk, work);
1604 btrfs_add_delayed_iput(async_chunk->inode);
1605 if (async_chunk->blkcg_css)
1606 css_put(async_chunk->blkcg_css);
1608 async_cow = async_chunk->async_cow;
1609 if (atomic_dec_and_test(&async_cow->num_chunks))
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, async_cow_free);
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;
1702 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1703 struct page *locked_page, u64 start,
1704 u64 end, struct writeback_control *wbc)
1706 u64 done_offset = end;
1708 bool locked_page_done = false;
1710 while (start <= end) {
1711 ret = cow_file_range(inode, locked_page, start, end, &done_offset,
1713 if (ret && ret != -EAGAIN)
1719 if (done_offset == start) {
1720 wait_on_bit_io(&inode->root->fs_info->flags,
1721 BTRFS_FS_NEED_ZONE_FINISH,
1722 TASK_UNINTERRUPTIBLE);
1726 if (!locked_page_done) {
1727 __set_page_dirty_nobuffers(locked_page);
1728 account_page_redirty(locked_page);
1730 locked_page_done = true;
1731 extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1733 start = done_offset + 1;
1739 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1740 u64 bytenr, u64 num_bytes, bool nowait)
1742 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1743 struct btrfs_ordered_sum *sums;
1747 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1749 if (ret == 0 && list_empty(&list))
1752 while (!list_empty(&list)) {
1753 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1754 list_del(&sums->list);
1762 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1763 const u64 start, const u64 end)
1765 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1766 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1767 const u64 range_bytes = end + 1 - start;
1768 struct extent_io_tree *io_tree = &inode->io_tree;
1769 u64 range_start = start;
1774 * If EXTENT_NORESERVE is set it means that when the buffered write was
1775 * made we had not enough available data space and therefore we did not
1776 * reserve data space for it, since we though we could do NOCOW for the
1777 * respective file range (either there is prealloc extent or the inode
1778 * has the NOCOW bit set).
1780 * However when we need to fallback to COW mode (because for example the
1781 * block group for the corresponding extent was turned to RO mode by a
1782 * scrub or relocation) we need to do the following:
1784 * 1) We increment the bytes_may_use counter of the data space info.
1785 * If COW succeeds, it allocates a new data extent and after doing
1786 * that it decrements the space info's bytes_may_use counter and
1787 * increments its bytes_reserved counter by the same amount (we do
1788 * this at btrfs_add_reserved_bytes()). So we need to increment the
1789 * bytes_may_use counter to compensate (when space is reserved at
1790 * buffered write time, the bytes_may_use counter is incremented);
1792 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1793 * that if the COW path fails for any reason, it decrements (through
1794 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1795 * data space info, which we incremented in the step above.
1797 * If we need to fallback to cow and the inode corresponds to a free
1798 * space cache inode or an inode of the data relocation tree, we must
1799 * also increment bytes_may_use of the data space_info for the same
1800 * reason. Space caches and relocated data extents always get a prealloc
1801 * extent for them, however scrub or balance may have set the block
1802 * group that contains that extent to RO mode and therefore force COW
1803 * when starting writeback.
1805 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1806 EXTENT_NORESERVE, 0, NULL);
1807 if (count > 0 || is_space_ino || is_reloc_ino) {
1809 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1810 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1812 if (is_space_ino || is_reloc_ino)
1813 bytes = range_bytes;
1815 spin_lock(&sinfo->lock);
1816 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1817 spin_unlock(&sinfo->lock);
1820 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1825 * Don't try to create inline extents, as a mix of inline extent that
1826 * is written out and unlocked directly and a normal NOCOW extent
1829 ret = cow_file_range(inode, locked_page, start, end, NULL, false, true);
1834 struct can_nocow_file_extent_args {
1837 /* Start file offset of the range we want to NOCOW. */
1839 /* End file offset (inclusive) of the range we want to NOCOW. */
1841 bool writeback_path;
1844 * Free the path passed to can_nocow_file_extent() once it's not needed
1849 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1854 /* Number of bytes that can be written to in NOCOW mode. */
1859 * Check if we can NOCOW the file extent that the path points to.
1860 * This function may return with the path released, so the caller should check
1861 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1863 * Returns: < 0 on error
1864 * 0 if we can not NOCOW
1867 static int can_nocow_file_extent(struct btrfs_path *path,
1868 struct btrfs_key *key,
1869 struct btrfs_inode *inode,
1870 struct can_nocow_file_extent_args *args)
1872 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
1873 struct extent_buffer *leaf = path->nodes[0];
1874 struct btrfs_root *root = inode->root;
1875 struct btrfs_file_extent_item *fi;
1880 bool nowait = path->nowait;
1882 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
1883 extent_type = btrfs_file_extent_type(leaf, fi);
1885 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1888 /* Can't access these fields unless we know it's not an inline extent. */
1889 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1890 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1891 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
1893 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
1894 extent_type == BTRFS_FILE_EXTENT_REG)
1898 * If the extent was created before the generation where the last snapshot
1899 * for its subvolume was created, then this implies the extent is shared,
1900 * hence we must COW.
1902 if (!args->strict &&
1903 btrfs_file_extent_generation(leaf, fi) <=
1904 btrfs_root_last_snapshot(&root->root_item))
1907 /* An explicit hole, must COW. */
1908 if (args->disk_bytenr == 0)
1911 /* Compressed/encrypted/encoded extents must be COWed. */
1912 if (btrfs_file_extent_compression(leaf, fi) ||
1913 btrfs_file_extent_encryption(leaf, fi) ||
1914 btrfs_file_extent_other_encoding(leaf, fi))
1917 extent_end = btrfs_file_extent_end(path);
1920 * The following checks can be expensive, as they need to take other
1921 * locks and do btree or rbtree searches, so release the path to avoid
1922 * blocking other tasks for too long.
1924 btrfs_release_path(path);
1926 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
1927 key->offset - args->extent_offset,
1928 args->disk_bytenr, args->strict, path);
1929 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1933 if (args->free_path) {
1935 * We don't need the path anymore, plus through the
1936 * csum_exist_in_range() call below we will end up allocating
1937 * another path. So free the path to avoid unnecessary extra
1940 btrfs_free_path(path);
1944 /* If there are pending snapshots for this root, we must COW. */
1945 if (args->writeback_path && !is_freespace_inode &&
1946 atomic_read(&root->snapshot_force_cow))
1949 args->disk_bytenr += args->extent_offset;
1950 args->disk_bytenr += args->start - key->offset;
1951 args->num_bytes = min(args->end + 1, extent_end) - args->start;
1954 * Force COW if csums exist in the range. This ensures that csums for a
1955 * given extent are either valid or do not exist.
1957 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
1959 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
1965 if (args->free_path && path)
1966 btrfs_free_path(path);
1968 return ret < 0 ? ret : can_nocow;
1972 * when nowcow writeback call back. This checks for snapshots or COW copies
1973 * of the extents that exist in the file, and COWs the file as required.
1975 * If no cow copies or snapshots exist, we write directly to the existing
1978 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
1979 struct page *locked_page,
1980 const u64 start, const u64 end)
1982 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1983 struct btrfs_root *root = inode->root;
1984 struct btrfs_path *path;
1985 u64 cow_start = (u64)-1;
1986 u64 cur_offset = start;
1988 bool check_prev = true;
1989 u64 ino = btrfs_ino(inode);
1990 struct btrfs_block_group *bg;
1992 struct can_nocow_file_extent_args nocow_args = { 0 };
1994 path = btrfs_alloc_path();
1996 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1997 EXTENT_LOCKED | EXTENT_DELALLOC |
1998 EXTENT_DO_ACCOUNTING |
1999 EXTENT_DEFRAG, PAGE_UNLOCK |
2000 PAGE_START_WRITEBACK |
2001 PAGE_END_WRITEBACK);
2005 nocow_args.end = end;
2006 nocow_args.writeback_path = true;
2009 struct btrfs_ordered_extent *ordered;
2010 struct btrfs_key found_key;
2011 struct btrfs_file_extent_item *fi;
2012 struct extent_buffer *leaf;
2021 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2027 * If there is no extent for our range when doing the initial
2028 * search, then go back to the previous slot as it will be the
2029 * one containing the search offset
2031 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2032 leaf = path->nodes[0];
2033 btrfs_item_key_to_cpu(leaf, &found_key,
2034 path->slots[0] - 1);
2035 if (found_key.objectid == ino &&
2036 found_key.type == BTRFS_EXTENT_DATA_KEY)
2041 /* Go to next leaf if we have exhausted the current one */
2042 leaf = path->nodes[0];
2043 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2044 ret = btrfs_next_leaf(root, path);
2046 if (cow_start != (u64)-1)
2047 cur_offset = cow_start;
2052 leaf = path->nodes[0];
2055 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2057 /* Didn't find anything for our INO */
2058 if (found_key.objectid > ino)
2061 * Keep searching until we find an EXTENT_ITEM or there are no
2062 * more extents for this inode
2064 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2065 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2070 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2071 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2072 found_key.offset > end)
2076 * If the found extent starts after requested offset, then
2077 * adjust extent_end to be right before this extent begins
2079 if (found_key.offset > cur_offset) {
2080 extent_end = found_key.offset;
2086 * Found extent which begins before our range and potentially
2089 fi = btrfs_item_ptr(leaf, path->slots[0],
2090 struct btrfs_file_extent_item);
2091 extent_type = btrfs_file_extent_type(leaf, fi);
2092 /* If this is triggered then we have a memory corruption. */
2093 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2094 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2098 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2099 extent_end = btrfs_file_extent_end(path);
2102 * If the extent we got ends before our current offset, skip to
2105 if (extent_end <= cur_offset) {
2110 nocow_args.start = cur_offset;
2111 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2113 if (cow_start != (u64)-1)
2114 cur_offset = cow_start;
2116 } else if (ret == 0) {
2121 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2126 * If nocow is false then record the beginning of the range
2127 * that needs to be COWed
2130 if (cow_start == (u64)-1)
2131 cow_start = cur_offset;
2132 cur_offset = extent_end;
2133 if (cur_offset > end)
2135 if (!path->nodes[0])
2142 * COW range from cow_start to found_key.offset - 1. As the key
2143 * will contain the beginning of the first extent that can be
2144 * NOCOW, following one which needs to be COW'ed
2146 if (cow_start != (u64)-1) {
2147 ret = fallback_to_cow(inode, locked_page,
2148 cow_start, found_key.offset - 1);
2151 cow_start = (u64)-1;
2154 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2155 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2157 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2158 struct extent_map *em;
2160 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2162 nocow_args.disk_bytenr, /* block_start */
2163 nocow_args.num_bytes, /* block_len */
2164 nocow_args.disk_num_bytes, /* orig_block_len */
2165 ram_bytes, BTRFS_COMPRESS_NONE,
2166 BTRFS_ORDERED_PREALLOC);
2171 free_extent_map(em);
2174 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2175 nocow_args.num_bytes, nocow_args.num_bytes,
2176 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2178 ? (1 << BTRFS_ORDERED_PREALLOC)
2179 : (1 << BTRFS_ORDERED_NOCOW),
2180 BTRFS_COMPRESS_NONE);
2181 if (IS_ERR(ordered)) {
2183 btrfs_drop_extent_map_range(inode, cur_offset,
2186 ret = PTR_ERR(ordered);
2191 btrfs_dec_nocow_writers(bg);
2195 if (btrfs_is_data_reloc_root(root))
2197 * Error handled later, as we must prevent
2198 * extent_clear_unlock_delalloc() in error handler
2199 * from freeing metadata of created ordered extent.
2201 ret = btrfs_reloc_clone_csums(ordered);
2202 btrfs_put_ordered_extent(ordered);
2204 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2205 locked_page, EXTENT_LOCKED |
2207 EXTENT_CLEAR_DATA_RESV,
2208 PAGE_UNLOCK | PAGE_SET_ORDERED);
2210 cur_offset = extent_end;
2213 * btrfs_reloc_clone_csums() error, now we're OK to call error
2214 * handler, as metadata for created ordered extent will only
2215 * be freed by btrfs_finish_ordered_io().
2219 if (cur_offset > end)
2222 btrfs_release_path(path);
2224 if (cur_offset <= end && cow_start == (u64)-1)
2225 cow_start = cur_offset;
2227 if (cow_start != (u64)-1) {
2229 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2236 btrfs_dec_nocow_writers(bg);
2238 if (ret && cur_offset < end)
2239 extent_clear_unlock_delalloc(inode, cur_offset, end,
2240 locked_page, EXTENT_LOCKED |
2241 EXTENT_DELALLOC | EXTENT_DEFRAG |
2242 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2243 PAGE_START_WRITEBACK |
2244 PAGE_END_WRITEBACK);
2245 btrfs_free_path(path);
2249 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2251 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2252 if (inode->defrag_bytes &&
2253 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2262 * Function to process delayed allocation (create CoW) for ranges which are
2263 * being touched for the first time.
2265 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2266 u64 start, u64 end, struct writeback_control *wbc)
2268 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2272 * The range must cover part of the @locked_page, or a return of 1
2273 * can confuse the caller.
2275 ASSERT(!(end <= page_offset(locked_page) ||
2276 start >= page_offset(locked_page) + PAGE_SIZE));
2278 if (should_nocow(inode, start, end)) {
2280 * Normally on a zoned device we're only doing COW writes, but
2281 * in case of relocation on a zoned filesystem we have taken
2282 * precaution, that we're only writing sequentially. It's safe
2283 * to use run_delalloc_nocow() here, like for regular
2284 * preallocated inodes.
2286 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2287 ret = run_delalloc_nocow(inode, locked_page, start, end);
2291 if (btrfs_inode_can_compress(inode) &&
2292 inode_need_compress(inode, start, end) &&
2293 run_delalloc_compressed(inode, locked_page, start, end, wbc))
2297 ret = run_delalloc_zoned(inode, locked_page, start, end, wbc);
2299 ret = cow_file_range(inode, locked_page, start, end, NULL,
2304 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2309 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2310 struct extent_state *orig, u64 split)
2312 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2315 /* not delalloc, ignore it */
2316 if (!(orig->state & EXTENT_DELALLOC))
2319 size = orig->end - orig->start + 1;
2320 if (size > fs_info->max_extent_size) {
2325 * See the explanation in btrfs_merge_delalloc_extent, the same
2326 * applies here, just in reverse.
2328 new_size = orig->end - split + 1;
2329 num_extents = count_max_extents(fs_info, new_size);
2330 new_size = split - orig->start;
2331 num_extents += count_max_extents(fs_info, new_size);
2332 if (count_max_extents(fs_info, size) >= num_extents)
2336 spin_lock(&inode->lock);
2337 btrfs_mod_outstanding_extents(inode, 1);
2338 spin_unlock(&inode->lock);
2342 * Handle merged delayed allocation extents so we can keep track of new extents
2343 * that are just merged onto old extents, such as when we are doing sequential
2344 * writes, so we can properly account for the metadata space we'll need.
2346 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2347 struct extent_state *other)
2349 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2350 u64 new_size, old_size;
2353 /* not delalloc, ignore it */
2354 if (!(other->state & EXTENT_DELALLOC))
2357 if (new->start > other->start)
2358 new_size = new->end - other->start + 1;
2360 new_size = other->end - new->start + 1;
2362 /* we're not bigger than the max, unreserve the space and go */
2363 if (new_size <= fs_info->max_extent_size) {
2364 spin_lock(&inode->lock);
2365 btrfs_mod_outstanding_extents(inode, -1);
2366 spin_unlock(&inode->lock);
2371 * We have to add up either side to figure out how many extents were
2372 * accounted for before we merged into one big extent. If the number of
2373 * extents we accounted for is <= the amount we need for the new range
2374 * then we can return, otherwise drop. Think of it like this
2378 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2379 * need 2 outstanding extents, on one side we have 1 and the other side
2380 * we have 1 so they are == and we can return. But in this case
2382 * [MAX_SIZE+4k][MAX_SIZE+4k]
2384 * Each range on their own accounts for 2 extents, but merged together
2385 * they are only 3 extents worth of accounting, so we need to drop in
2388 old_size = other->end - other->start + 1;
2389 num_extents = count_max_extents(fs_info, old_size);
2390 old_size = new->end - new->start + 1;
2391 num_extents += count_max_extents(fs_info, old_size);
2392 if (count_max_extents(fs_info, new_size) >= num_extents)
2395 spin_lock(&inode->lock);
2396 btrfs_mod_outstanding_extents(inode, -1);
2397 spin_unlock(&inode->lock);
2400 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2401 struct btrfs_inode *inode)
2403 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2405 spin_lock(&root->delalloc_lock);
2406 if (list_empty(&inode->delalloc_inodes)) {
2407 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2408 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2409 root->nr_delalloc_inodes++;
2410 if (root->nr_delalloc_inodes == 1) {
2411 spin_lock(&fs_info->delalloc_root_lock);
2412 BUG_ON(!list_empty(&root->delalloc_root));
2413 list_add_tail(&root->delalloc_root,
2414 &fs_info->delalloc_roots);
2415 spin_unlock(&fs_info->delalloc_root_lock);
2418 spin_unlock(&root->delalloc_lock);
2421 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2422 struct btrfs_inode *inode)
2424 struct btrfs_fs_info *fs_info = root->fs_info;
2426 if (!list_empty(&inode->delalloc_inodes)) {
2427 list_del_init(&inode->delalloc_inodes);
2428 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2429 &inode->runtime_flags);
2430 root->nr_delalloc_inodes--;
2431 if (!root->nr_delalloc_inodes) {
2432 ASSERT(list_empty(&root->delalloc_inodes));
2433 spin_lock(&fs_info->delalloc_root_lock);
2434 BUG_ON(list_empty(&root->delalloc_root));
2435 list_del_init(&root->delalloc_root);
2436 spin_unlock(&fs_info->delalloc_root_lock);
2441 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2442 struct btrfs_inode *inode)
2444 spin_lock(&root->delalloc_lock);
2445 __btrfs_del_delalloc_inode(root, inode);
2446 spin_unlock(&root->delalloc_lock);
2450 * Properly track delayed allocation bytes in the inode and to maintain the
2451 * list of inodes that have pending delalloc work to be done.
2453 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2456 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2458 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2461 * set_bit and clear bit hooks normally require _irqsave/restore
2462 * but in this case, we are only testing for the DELALLOC
2463 * bit, which is only set or cleared with irqs on
2465 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2466 struct btrfs_root *root = inode->root;
2467 u64 len = state->end + 1 - state->start;
2468 u32 num_extents = count_max_extents(fs_info, len);
2469 bool do_list = !btrfs_is_free_space_inode(inode);
2471 spin_lock(&inode->lock);
2472 btrfs_mod_outstanding_extents(inode, num_extents);
2473 spin_unlock(&inode->lock);
2475 /* For sanity tests */
2476 if (btrfs_is_testing(fs_info))
2479 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2480 fs_info->delalloc_batch);
2481 spin_lock(&inode->lock);
2482 inode->delalloc_bytes += len;
2483 if (bits & EXTENT_DEFRAG)
2484 inode->defrag_bytes += len;
2485 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2486 &inode->runtime_flags))
2487 btrfs_add_delalloc_inodes(root, inode);
2488 spin_unlock(&inode->lock);
2491 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2492 (bits & EXTENT_DELALLOC_NEW)) {
2493 spin_lock(&inode->lock);
2494 inode->new_delalloc_bytes += state->end + 1 - state->start;
2495 spin_unlock(&inode->lock);
2500 * Once a range is no longer delalloc this function ensures that proper
2501 * accounting happens.
2503 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2504 struct extent_state *state, u32 bits)
2506 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2507 u64 len = state->end + 1 - state->start;
2508 u32 num_extents = count_max_extents(fs_info, len);
2510 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2511 spin_lock(&inode->lock);
2512 inode->defrag_bytes -= len;
2513 spin_unlock(&inode->lock);
2517 * set_bit and clear bit hooks normally require _irqsave/restore
2518 * but in this case, we are only testing for the DELALLOC
2519 * bit, which is only set or cleared with irqs on
2521 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2522 struct btrfs_root *root = inode->root;
2523 bool do_list = !btrfs_is_free_space_inode(inode);
2525 spin_lock(&inode->lock);
2526 btrfs_mod_outstanding_extents(inode, -num_extents);
2527 spin_unlock(&inode->lock);
2530 * We don't reserve metadata space for space cache inodes so we
2531 * don't need to call delalloc_release_metadata if there is an
2534 if (bits & EXTENT_CLEAR_META_RESV &&
2535 root != fs_info->tree_root)
2536 btrfs_delalloc_release_metadata(inode, len, false);
2538 /* For sanity tests. */
2539 if (btrfs_is_testing(fs_info))
2542 if (!btrfs_is_data_reloc_root(root) &&
2543 do_list && !(state->state & EXTENT_NORESERVE) &&
2544 (bits & EXTENT_CLEAR_DATA_RESV))
2545 btrfs_free_reserved_data_space_noquota(fs_info, len);
2547 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2548 fs_info->delalloc_batch);
2549 spin_lock(&inode->lock);
2550 inode->delalloc_bytes -= len;
2551 if (do_list && inode->delalloc_bytes == 0 &&
2552 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2553 &inode->runtime_flags))
2554 btrfs_del_delalloc_inode(root, inode);
2555 spin_unlock(&inode->lock);
2558 if ((state->state & EXTENT_DELALLOC_NEW) &&
2559 (bits & EXTENT_DELALLOC_NEW)) {
2560 spin_lock(&inode->lock);
2561 ASSERT(inode->new_delalloc_bytes >= len);
2562 inode->new_delalloc_bytes -= len;
2563 if (bits & EXTENT_ADD_INODE_BYTES)
2564 inode_add_bytes(&inode->vfs_inode, len);
2565 spin_unlock(&inode->lock);
2569 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2570 struct btrfs_ordered_extent *ordered)
2572 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2573 u64 len = bbio->bio.bi_iter.bi_size;
2574 struct btrfs_ordered_extent *new;
2577 /* Must always be called for the beginning of an ordered extent. */
2578 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2581 /* No need to split if the ordered extent covers the entire bio. */
2582 if (ordered->disk_num_bytes == len) {
2583 refcount_inc(&ordered->refs);
2584 bbio->ordered = ordered;
2589 * Don't split the extent_map for NOCOW extents, as we're writing into
2590 * a pre-existing one.
2592 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2593 ret = split_extent_map(bbio->inode, bbio->file_offset,
2594 ordered->num_bytes, len,
2595 ordered->disk_bytenr);
2600 new = btrfs_split_ordered_extent(ordered, len);
2602 return PTR_ERR(new);
2603 bbio->ordered = new;
2608 * given a list of ordered sums record them in the inode. This happens
2609 * at IO completion time based on sums calculated at bio submission time.
2611 static int add_pending_csums(struct btrfs_trans_handle *trans,
2612 struct list_head *list)
2614 struct btrfs_ordered_sum *sum;
2615 struct btrfs_root *csum_root = NULL;
2618 list_for_each_entry(sum, list, list) {
2619 trans->adding_csums = true;
2621 csum_root = btrfs_csum_root(trans->fs_info,
2623 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2624 trans->adding_csums = false;
2631 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2634 struct extent_state **cached_state)
2636 u64 search_start = start;
2637 const u64 end = start + len - 1;
2639 while (search_start < end) {
2640 const u64 search_len = end - search_start + 1;
2641 struct extent_map *em;
2645 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2649 if (em->block_start != EXTENT_MAP_HOLE)
2653 if (em->start < search_start)
2654 em_len -= search_start - em->start;
2655 if (em_len > search_len)
2656 em_len = search_len;
2658 ret = set_extent_bit(&inode->io_tree, search_start,
2659 search_start + em_len - 1,
2660 EXTENT_DELALLOC_NEW, cached_state);
2662 search_start = extent_map_end(em);
2663 free_extent_map(em);
2670 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2671 unsigned int extra_bits,
2672 struct extent_state **cached_state)
2674 WARN_ON(PAGE_ALIGNED(end));
2676 if (start >= i_size_read(&inode->vfs_inode) &&
2677 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2679 * There can't be any extents following eof in this case so just
2680 * set the delalloc new bit for the range directly.
2682 extra_bits |= EXTENT_DELALLOC_NEW;
2686 ret = btrfs_find_new_delalloc_bytes(inode, start,
2693 return set_extent_bit(&inode->io_tree, start, end,
2694 EXTENT_DELALLOC | extra_bits, cached_state);
2697 /* see btrfs_writepage_start_hook for details on why this is required */
2698 struct btrfs_writepage_fixup {
2700 struct btrfs_inode *inode;
2701 struct btrfs_work work;
2704 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2706 struct btrfs_writepage_fixup *fixup =
2707 container_of(work, struct btrfs_writepage_fixup, work);
2708 struct btrfs_ordered_extent *ordered;
2709 struct extent_state *cached_state = NULL;
2710 struct extent_changeset *data_reserved = NULL;
2711 struct page *page = fixup->page;
2712 struct btrfs_inode *inode = fixup->inode;
2713 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2714 u64 page_start = page_offset(page);
2715 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2717 bool free_delalloc_space = true;
2720 * This is similar to page_mkwrite, we need to reserve the space before
2721 * we take the page lock.
2723 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2729 * Before we queued this fixup, we took a reference on the page.
2730 * page->mapping may go NULL, but it shouldn't be moved to a different
2733 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2735 * Unfortunately this is a little tricky, either
2737 * 1) We got here and our page had already been dealt with and
2738 * we reserved our space, thus ret == 0, so we need to just
2739 * drop our space reservation and bail. This can happen the
2740 * first time we come into the fixup worker, or could happen
2741 * while waiting for the ordered extent.
2742 * 2) Our page was already dealt with, but we happened to get an
2743 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2744 * this case we obviously don't have anything to release, but
2745 * because the page was already dealt with we don't want to
2746 * mark the page with an error, so make sure we're resetting
2747 * ret to 0. This is why we have this check _before_ the ret
2748 * check, because we do not want to have a surprise ENOSPC
2749 * when the page was already properly dealt with.
2752 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2753 btrfs_delalloc_release_space(inode, data_reserved,
2754 page_start, PAGE_SIZE,
2762 * We can't mess with the page state unless it is locked, so now that
2763 * it is locked bail if we failed to make our space reservation.
2768 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2770 /* already ordered? We're done */
2771 if (PageOrdered(page))
2774 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2776 unlock_extent(&inode->io_tree, page_start, page_end,
2779 btrfs_start_ordered_extent(ordered);
2780 btrfs_put_ordered_extent(ordered);
2784 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2790 * Everything went as planned, we're now the owner of a dirty page with
2791 * delayed allocation bits set and space reserved for our COW
2794 * The page was dirty when we started, nothing should have cleaned it.
2796 BUG_ON(!PageDirty(page));
2797 free_delalloc_space = false;
2799 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2800 if (free_delalloc_space)
2801 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2803 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2807 * We hit ENOSPC or other errors. Update the mapping and page
2808 * to reflect the errors and clean the page.
2810 mapping_set_error(page->mapping, ret);
2811 btrfs_mark_ordered_io_finished(inode, page, page_start,
2813 btrfs_page_clear_uptodate(fs_info, page, page_start, PAGE_SIZE);
2814 clear_page_dirty_for_io(page);
2816 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2820 extent_changeset_free(data_reserved);
2822 * As a precaution, do a delayed iput in case it would be the last iput
2823 * that could need flushing space. Recursing back to fixup worker would
2826 btrfs_add_delayed_iput(inode);
2830 * There are a few paths in the higher layers of the kernel that directly
2831 * set the page dirty bit without asking the filesystem if it is a
2832 * good idea. This causes problems because we want to make sure COW
2833 * properly happens and the data=ordered rules are followed.
2835 * In our case any range that doesn't have the ORDERED bit set
2836 * hasn't been properly setup for IO. We kick off an async process
2837 * to fix it up. The async helper will wait for ordered extents, set
2838 * the delalloc bit and make it safe to write the page.
2840 int btrfs_writepage_cow_fixup(struct page *page)
2842 struct inode *inode = page->mapping->host;
2843 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2844 struct btrfs_writepage_fixup *fixup;
2846 /* This page has ordered extent covering it already */
2847 if (PageOrdered(page))
2851 * PageChecked is set below when we create a fixup worker for this page,
2852 * don't try to create another one if we're already PageChecked()
2854 * The extent_io writepage code will redirty the page if we send back
2857 if (PageChecked(page))
2860 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2865 * We are already holding a reference to this inode from
2866 * write_cache_pages. We need to hold it because the space reservation
2867 * takes place outside of the page lock, and we can't trust
2868 * page->mapping outside of the page lock.
2871 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
2873 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
2875 fixup->inode = BTRFS_I(inode);
2876 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2881 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2882 struct btrfs_inode *inode, u64 file_pos,
2883 struct btrfs_file_extent_item *stack_fi,
2884 const bool update_inode_bytes,
2885 u64 qgroup_reserved)
2887 struct btrfs_root *root = inode->root;
2888 const u64 sectorsize = root->fs_info->sectorsize;
2889 struct btrfs_path *path;
2890 struct extent_buffer *leaf;
2891 struct btrfs_key ins;
2892 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
2893 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
2894 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
2895 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
2896 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
2897 struct btrfs_drop_extents_args drop_args = { 0 };
2900 path = btrfs_alloc_path();
2905 * we may be replacing one extent in the tree with another.
2906 * The new extent is pinned in the extent map, and we don't want
2907 * to drop it from the cache until it is completely in the btree.
2909 * So, tell btrfs_drop_extents to leave this extent in the cache.
2910 * the caller is expected to unpin it and allow it to be merged
2913 drop_args.path = path;
2914 drop_args.start = file_pos;
2915 drop_args.end = file_pos + num_bytes;
2916 drop_args.replace_extent = true;
2917 drop_args.extent_item_size = sizeof(*stack_fi);
2918 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
2922 if (!drop_args.extent_inserted) {
2923 ins.objectid = btrfs_ino(inode);
2924 ins.offset = file_pos;
2925 ins.type = BTRFS_EXTENT_DATA_KEY;
2927 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2932 leaf = path->nodes[0];
2933 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
2934 write_extent_buffer(leaf, stack_fi,
2935 btrfs_item_ptr_offset(leaf, path->slots[0]),
2936 sizeof(struct btrfs_file_extent_item));
2938 btrfs_mark_buffer_dirty(leaf);
2939 btrfs_release_path(path);
2942 * If we dropped an inline extent here, we know the range where it is
2943 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
2944 * number of bytes only for that range containing the inline extent.
2945 * The remaining of the range will be processed when clearning the
2946 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
2948 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
2949 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
2951 inline_size = drop_args.bytes_found - inline_size;
2952 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
2953 drop_args.bytes_found -= inline_size;
2954 num_bytes -= sectorsize;
2957 if (update_inode_bytes)
2958 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
2960 ins.objectid = disk_bytenr;
2961 ins.offset = disk_num_bytes;
2962 ins.type = BTRFS_EXTENT_ITEM_KEY;
2964 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
2968 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
2970 qgroup_reserved, &ins);
2972 btrfs_free_path(path);
2977 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2980 struct btrfs_block_group *cache;
2982 cache = btrfs_lookup_block_group(fs_info, start);
2985 spin_lock(&cache->lock);
2986 cache->delalloc_bytes -= len;
2987 spin_unlock(&cache->lock);
2989 btrfs_put_block_group(cache);
2992 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
2993 struct btrfs_ordered_extent *oe)
2995 struct btrfs_file_extent_item stack_fi;
2996 bool update_inode_bytes;
2997 u64 num_bytes = oe->num_bytes;
2998 u64 ram_bytes = oe->ram_bytes;
3000 memset(&stack_fi, 0, sizeof(stack_fi));
3001 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3002 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3003 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3004 oe->disk_num_bytes);
3005 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3006 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3007 num_bytes = oe->truncated_len;
3008 ram_bytes = num_bytes;
3010 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3011 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3012 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3013 /* Encryption and other encoding is reserved and all 0 */
3016 * For delalloc, when completing an ordered extent we update the inode's
3017 * bytes when clearing the range in the inode's io tree, so pass false
3018 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3019 * except if the ordered extent was truncated.
3021 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3022 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3023 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3025 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3026 oe->file_offset, &stack_fi,
3027 update_inode_bytes, oe->qgroup_rsv);
3031 * As ordered data IO finishes, this gets called so we can finish
3032 * an ordered extent if the range of bytes in the file it covers are
3035 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3037 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3038 struct btrfs_root *root = inode->root;
3039 struct btrfs_fs_info *fs_info = root->fs_info;
3040 struct btrfs_trans_handle *trans = NULL;
3041 struct extent_io_tree *io_tree = &inode->io_tree;
3042 struct extent_state *cached_state = NULL;
3044 int compress_type = 0;
3046 u64 logical_len = ordered_extent->num_bytes;
3047 bool freespace_inode;
3048 bool truncated = false;
3049 bool clear_reserved_extent = true;
3050 unsigned int clear_bits = EXTENT_DEFRAG;
3052 start = ordered_extent->file_offset;
3053 end = start + ordered_extent->num_bytes - 1;
3055 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3056 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3057 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3058 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3059 clear_bits |= EXTENT_DELALLOC_NEW;
3061 freespace_inode = btrfs_is_free_space_inode(inode);
3062 if (!freespace_inode)
3063 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3065 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3070 if (btrfs_is_zoned(fs_info))
3071 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3072 ordered_extent->disk_num_bytes);
3074 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3076 logical_len = ordered_extent->truncated_len;
3077 /* Truncated the entire extent, don't bother adding */
3082 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3083 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3085 btrfs_inode_safe_disk_i_size_write(inode, 0);
3086 if (freespace_inode)
3087 trans = btrfs_join_transaction_spacecache(root);
3089 trans = btrfs_join_transaction(root);
3090 if (IS_ERR(trans)) {
3091 ret = PTR_ERR(trans);
3095 trans->block_rsv = &inode->block_rsv;
3096 ret = btrfs_update_inode_fallback(trans, root, inode);
3097 if (ret) /* -ENOMEM or corruption */
3098 btrfs_abort_transaction(trans, ret);
3102 clear_bits |= EXTENT_LOCKED;
3103 lock_extent(io_tree, start, end, &cached_state);
3105 if (freespace_inode)
3106 trans = btrfs_join_transaction_spacecache(root);
3108 trans = btrfs_join_transaction(root);
3109 if (IS_ERR(trans)) {
3110 ret = PTR_ERR(trans);
3115 trans->block_rsv = &inode->block_rsv;
3117 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3118 compress_type = ordered_extent->compress_type;
3119 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3120 BUG_ON(compress_type);
3121 ret = btrfs_mark_extent_written(trans, inode,
3122 ordered_extent->file_offset,
3123 ordered_extent->file_offset +
3125 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3126 ordered_extent->disk_num_bytes);
3128 BUG_ON(root == fs_info->tree_root);
3129 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3131 clear_reserved_extent = false;
3132 btrfs_release_delalloc_bytes(fs_info,
3133 ordered_extent->disk_bytenr,
3134 ordered_extent->disk_num_bytes);
3137 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3138 ordered_extent->num_bytes, trans->transid);
3140 btrfs_abort_transaction(trans, ret);
3144 ret = add_pending_csums(trans, &ordered_extent->list);
3146 btrfs_abort_transaction(trans, ret);
3151 * If this is a new delalloc range, clear its new delalloc flag to
3152 * update the inode's number of bytes. This needs to be done first
3153 * before updating the inode item.
3155 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3156 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3157 clear_extent_bit(&inode->io_tree, start, end,
3158 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3161 btrfs_inode_safe_disk_i_size_write(inode, 0);
3162 ret = btrfs_update_inode_fallback(trans, root, inode);
3163 if (ret) { /* -ENOMEM or corruption */
3164 btrfs_abort_transaction(trans, ret);
3169 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3173 btrfs_end_transaction(trans);
3175 if (ret || truncated) {
3176 u64 unwritten_start = start;
3179 * If we failed to finish this ordered extent for any reason we
3180 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3181 * extent, and mark the inode with the error if it wasn't
3182 * already set. Any error during writeback would have already
3183 * set the mapping error, so we need to set it if we're the ones
3184 * marking this ordered extent as failed.
3186 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3187 &ordered_extent->flags))
3188 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3191 unwritten_start += logical_len;
3192 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3194 /* Drop extent maps for the part of the extent we didn't write. */
3195 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3198 * If the ordered extent had an IOERR or something else went
3199 * wrong we need to return the space for this ordered extent
3200 * back to the allocator. We only free the extent in the
3201 * truncated case if we didn't write out the extent at all.
3203 * If we made it past insert_reserved_file_extent before we
3204 * errored out then we don't need to do this as the accounting
3205 * has already been done.
3207 if ((ret || !logical_len) &&
3208 clear_reserved_extent &&
3209 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3210 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3212 * Discard the range before returning it back to the
3215 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3216 btrfs_discard_extent(fs_info,
3217 ordered_extent->disk_bytenr,
3218 ordered_extent->disk_num_bytes,
3220 btrfs_free_reserved_extent(fs_info,
3221 ordered_extent->disk_bytenr,
3222 ordered_extent->disk_num_bytes, 1);
3224 * Actually free the qgroup rsv which was released when
3225 * the ordered extent was created.
3227 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3228 ordered_extent->qgroup_rsv,
3229 BTRFS_QGROUP_RSV_DATA);
3234 * This needs to be done to make sure anybody waiting knows we are done
3235 * updating everything for this ordered extent.
3237 btrfs_remove_ordered_extent(inode, ordered_extent);
3240 btrfs_put_ordered_extent(ordered_extent);
3241 /* once for the tree */
3242 btrfs_put_ordered_extent(ordered_extent);
3247 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3249 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3250 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3251 btrfs_finish_ordered_zoned(ordered);
3252 return btrfs_finish_one_ordered(ordered);
3256 * Verify the checksum for a single sector without any extra action that depend
3257 * on the type of I/O.
3259 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3260 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3262 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3265 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3267 shash->tfm = fs_info->csum_shash;
3269 kaddr = kmap_local_page(page) + pgoff;
3270 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3271 kunmap_local(kaddr);
3273 if (memcmp(csum, csum_expected, fs_info->csum_size))
3279 * Verify the checksum of a single data sector.
3281 * @bbio: btrfs_io_bio which contains the csum
3282 * @dev: device the sector is on
3283 * @bio_offset: offset to the beginning of the bio (in bytes)
3284 * @bv: bio_vec to check
3286 * Check if the checksum on a data block is valid. When a checksum mismatch is
3287 * detected, report the error and fill the corrupted range with zero.
3289 * Return %true if the sector is ok or had no checksum to start with, else %false.
3291 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3292 u32 bio_offset, struct bio_vec *bv)
3294 struct btrfs_inode *inode = bbio->inode;
3295 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3296 u64 file_offset = bbio->file_offset + bio_offset;
3297 u64 end = file_offset + bv->bv_len - 1;
3299 u8 csum[BTRFS_CSUM_SIZE];
3301 ASSERT(bv->bv_len == fs_info->sectorsize);
3306 if (btrfs_is_data_reloc_root(inode->root) &&
3307 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3309 /* Skip the range without csum for data reloc inode */
3310 clear_extent_bits(&inode->io_tree, file_offset, end,
3315 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3317 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3323 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3326 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3332 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3334 * @inode: The inode we want to perform iput on
3336 * This function uses the generic vfs_inode::i_count to track whether we should
3337 * just decrement it (in case it's > 1) or if this is the last iput then link
3338 * the inode to the delayed iput machinery. Delayed iputs are processed at
3339 * transaction commit time/superblock commit/cleaner kthread.
3341 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3343 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3344 unsigned long flags;
3346 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3349 atomic_inc(&fs_info->nr_delayed_iputs);
3351 * Need to be irq safe here because we can be called from either an irq
3352 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3355 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3356 ASSERT(list_empty(&inode->delayed_iput));
3357 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3358 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3359 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3360 wake_up_process(fs_info->cleaner_kthread);
3363 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3364 struct btrfs_inode *inode)
3366 list_del_init(&inode->delayed_iput);
3367 spin_unlock_irq(&fs_info->delayed_iput_lock);
3368 iput(&inode->vfs_inode);
3369 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3370 wake_up(&fs_info->delayed_iputs_wait);
3371 spin_lock_irq(&fs_info->delayed_iput_lock);
3374 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3375 struct btrfs_inode *inode)
3377 if (!list_empty(&inode->delayed_iput)) {
3378 spin_lock_irq(&fs_info->delayed_iput_lock);
3379 if (!list_empty(&inode->delayed_iput))
3380 run_delayed_iput_locked(fs_info, inode);
3381 spin_unlock_irq(&fs_info->delayed_iput_lock);
3385 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3388 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3389 * calls btrfs_add_delayed_iput() and that needs to lock
3390 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3391 * prevent a deadlock.
3393 spin_lock_irq(&fs_info->delayed_iput_lock);
3394 while (!list_empty(&fs_info->delayed_iputs)) {
3395 struct btrfs_inode *inode;
3397 inode = list_first_entry(&fs_info->delayed_iputs,
3398 struct btrfs_inode, delayed_iput);
3399 run_delayed_iput_locked(fs_info, inode);
3400 if (need_resched()) {
3401 spin_unlock_irq(&fs_info->delayed_iput_lock);
3403 spin_lock_irq(&fs_info->delayed_iput_lock);
3406 spin_unlock_irq(&fs_info->delayed_iput_lock);
3410 * Wait for flushing all delayed iputs
3412 * @fs_info: the filesystem
3414 * This will wait on any delayed iputs that are currently running with KILLABLE
3415 * set. Once they are all done running we will return, unless we are killed in
3416 * which case we return EINTR. This helps in user operations like fallocate etc
3417 * that might get blocked on the iputs.
3419 * Return EINTR if we were killed, 0 if nothing's pending
3421 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3423 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3424 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3431 * This creates an orphan entry for the given inode in case something goes wrong
3432 * in the middle of an unlink.
3434 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3435 struct btrfs_inode *inode)
3439 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3440 if (ret && ret != -EEXIST) {
3441 btrfs_abort_transaction(trans, ret);
3449 * We have done the delete so we can go ahead and remove the orphan item for
3450 * this particular inode.
3452 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3453 struct btrfs_inode *inode)
3455 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3459 * this cleans up any orphans that may be left on the list from the last use
3462 int btrfs_orphan_cleanup(struct btrfs_root *root)
3464 struct btrfs_fs_info *fs_info = root->fs_info;
3465 struct btrfs_path *path;
3466 struct extent_buffer *leaf;
3467 struct btrfs_key key, found_key;
3468 struct btrfs_trans_handle *trans;
3469 struct inode *inode;
3470 u64 last_objectid = 0;
3471 int ret = 0, nr_unlink = 0;
3473 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3476 path = btrfs_alloc_path();
3481 path->reada = READA_BACK;
3483 key.objectid = BTRFS_ORPHAN_OBJECTID;
3484 key.type = BTRFS_ORPHAN_ITEM_KEY;
3485 key.offset = (u64)-1;
3488 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3493 * if ret == 0 means we found what we were searching for, which
3494 * is weird, but possible, so only screw with path if we didn't
3495 * find the key and see if we have stuff that matches
3499 if (path->slots[0] == 0)
3504 /* pull out the item */
3505 leaf = path->nodes[0];
3506 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3508 /* make sure the item matches what we want */
3509 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3511 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3514 /* release the path since we're done with it */
3515 btrfs_release_path(path);
3518 * this is where we are basically btrfs_lookup, without the
3519 * crossing root thing. we store the inode number in the
3520 * offset of the orphan item.
3523 if (found_key.offset == last_objectid) {
3525 "Error removing orphan entry, stopping orphan cleanup");
3530 last_objectid = found_key.offset;
3532 found_key.objectid = found_key.offset;
3533 found_key.type = BTRFS_INODE_ITEM_KEY;
3534 found_key.offset = 0;
3535 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3536 if (IS_ERR(inode)) {
3537 ret = PTR_ERR(inode);
3543 if (!inode && root == fs_info->tree_root) {
3544 struct btrfs_root *dead_root;
3545 int is_dead_root = 0;
3548 * This is an orphan in the tree root. Currently these
3549 * could come from 2 sources:
3550 * a) a root (snapshot/subvolume) deletion in progress
3551 * b) a free space cache inode
3552 * We need to distinguish those two, as the orphan item
3553 * for a root must not get deleted before the deletion
3554 * of the snapshot/subvolume's tree completes.
3556 * btrfs_find_orphan_roots() ran before us, which has
3557 * found all deleted roots and loaded them into
3558 * fs_info->fs_roots_radix. So here we can find if an
3559 * orphan item corresponds to a deleted root by looking
3560 * up the root from that radix tree.
3563 spin_lock(&fs_info->fs_roots_radix_lock);
3564 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3565 (unsigned long)found_key.objectid);
3566 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3568 spin_unlock(&fs_info->fs_roots_radix_lock);
3571 /* prevent this orphan from being found again */
3572 key.offset = found_key.objectid - 1;
3579 * If we have an inode with links, there are a couple of
3582 * 1. We were halfway through creating fsverity metadata for the
3583 * file. In that case, the orphan item represents incomplete
3584 * fsverity metadata which must be cleaned up with
3585 * btrfs_drop_verity_items and deleting the orphan item.
3587 * 2. Old kernels (before v3.12) used to create an
3588 * orphan item for truncate indicating that there were possibly
3589 * extent items past i_size that needed to be deleted. In v3.12,
3590 * truncate was changed to update i_size in sync with the extent
3591 * items, but the (useless) orphan item was still created. Since
3592 * v4.18, we don't create the orphan item for truncate at all.
3594 * So, this item could mean that we need to do a truncate, but
3595 * only if this filesystem was last used on a pre-v3.12 kernel
3596 * and was not cleanly unmounted. The odds of that are quite
3597 * slim, and it's a pain to do the truncate now, so just delete
3600 * It's also possible that this orphan item was supposed to be
3601 * deleted but wasn't. The inode number may have been reused,
3602 * but either way, we can delete the orphan item.
3604 if (!inode || inode->i_nlink) {
3606 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3612 trans = btrfs_start_transaction(root, 1);
3613 if (IS_ERR(trans)) {
3614 ret = PTR_ERR(trans);
3617 btrfs_debug(fs_info, "auto deleting %Lu",
3618 found_key.objectid);
3619 ret = btrfs_del_orphan_item(trans, root,
3620 found_key.objectid);
3621 btrfs_end_transaction(trans);
3629 /* this will do delete_inode and everything for us */
3632 /* release the path since we're done with it */
3633 btrfs_release_path(path);
3635 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3636 trans = btrfs_join_transaction(root);
3638 btrfs_end_transaction(trans);
3642 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3646 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3647 btrfs_free_path(path);
3652 * very simple check to peek ahead in the leaf looking for xattrs. If we
3653 * don't find any xattrs, we know there can't be any acls.
3655 * slot is the slot the inode is in, objectid is the objectid of the inode
3657 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3658 int slot, u64 objectid,
3659 int *first_xattr_slot)
3661 u32 nritems = btrfs_header_nritems(leaf);
3662 struct btrfs_key found_key;
3663 static u64 xattr_access = 0;
3664 static u64 xattr_default = 0;
3667 if (!xattr_access) {
3668 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3669 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3670 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3671 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3675 *first_xattr_slot = -1;
3676 while (slot < nritems) {
3677 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3679 /* we found a different objectid, there must not be acls */
3680 if (found_key.objectid != objectid)
3683 /* we found an xattr, assume we've got an acl */
3684 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3685 if (*first_xattr_slot == -1)
3686 *first_xattr_slot = slot;
3687 if (found_key.offset == xattr_access ||
3688 found_key.offset == xattr_default)
3693 * we found a key greater than an xattr key, there can't
3694 * be any acls later on
3696 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3703 * it goes inode, inode backrefs, xattrs, extents,
3704 * so if there are a ton of hard links to an inode there can
3705 * be a lot of backrefs. Don't waste time searching too hard,
3706 * this is just an optimization
3711 /* we hit the end of the leaf before we found an xattr or
3712 * something larger than an xattr. We have to assume the inode
3715 if (*first_xattr_slot == -1)
3716 *first_xattr_slot = slot;
3721 * read an inode from the btree into the in-memory inode
3723 static int btrfs_read_locked_inode(struct inode *inode,
3724 struct btrfs_path *in_path)
3726 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3727 struct btrfs_path *path = in_path;
3728 struct extent_buffer *leaf;
3729 struct btrfs_inode_item *inode_item;
3730 struct btrfs_root *root = BTRFS_I(inode)->root;
3731 struct btrfs_key location;
3736 bool filled = false;
3737 int first_xattr_slot;
3739 ret = btrfs_fill_inode(inode, &rdev);
3744 path = btrfs_alloc_path();
3749 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3751 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3753 if (path != in_path)
3754 btrfs_free_path(path);
3758 leaf = path->nodes[0];
3763 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3764 struct btrfs_inode_item);
3765 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3766 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3767 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3768 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3769 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3770 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3771 round_up(i_size_read(inode), fs_info->sectorsize));
3773 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3774 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3776 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3777 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3779 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3780 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3782 BTRFS_I(inode)->i_otime.tv_sec =
3783 btrfs_timespec_sec(leaf, &inode_item->otime);
3784 BTRFS_I(inode)->i_otime.tv_nsec =
3785 btrfs_timespec_nsec(leaf, &inode_item->otime);
3787 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3788 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3789 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3791 inode_set_iversion_queried(inode,
3792 btrfs_inode_sequence(leaf, inode_item));
3793 inode->i_generation = BTRFS_I(inode)->generation;
3795 rdev = btrfs_inode_rdev(leaf, inode_item);
3797 BTRFS_I(inode)->index_cnt = (u64)-1;
3798 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3799 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3803 * If we were modified in the current generation and evicted from memory
3804 * and then re-read we need to do a full sync since we don't have any
3805 * idea about which extents were modified before we were evicted from
3808 * This is required for both inode re-read from disk and delayed inode
3809 * in delayed_nodes_tree.
3811 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3812 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3813 &BTRFS_I(inode)->runtime_flags);
3816 * We don't persist the id of the transaction where an unlink operation
3817 * against the inode was last made. So here we assume the inode might
3818 * have been evicted, and therefore the exact value of last_unlink_trans
3819 * lost, and set it to last_trans to avoid metadata inconsistencies
3820 * between the inode and its parent if the inode is fsync'ed and the log
3821 * replayed. For example, in the scenario:
3824 * ln mydir/foo mydir/bar
3827 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3828 * xfs_io -c fsync mydir/foo
3830 * mount fs, triggers fsync log replay
3832 * We must make sure that when we fsync our inode foo we also log its
3833 * parent inode, otherwise after log replay the parent still has the
3834 * dentry with the "bar" name but our inode foo has a link count of 1
3835 * and doesn't have an inode ref with the name "bar" anymore.
3837 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3838 * but it guarantees correctness at the expense of occasional full
3839 * transaction commits on fsync if our inode is a directory, or if our
3840 * inode is not a directory, logging its parent unnecessarily.
3842 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3845 * Same logic as for last_unlink_trans. We don't persist the generation
3846 * of the last transaction where this inode was used for a reflink
3847 * operation, so after eviction and reloading the inode we must be
3848 * pessimistic and assume the last transaction that modified the inode.
3850 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3853 if (inode->i_nlink != 1 ||
3854 path->slots[0] >= btrfs_header_nritems(leaf))
3857 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3858 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3861 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3862 if (location.type == BTRFS_INODE_REF_KEY) {
3863 struct btrfs_inode_ref *ref;
3865 ref = (struct btrfs_inode_ref *)ptr;
3866 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3867 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3868 struct btrfs_inode_extref *extref;
3870 extref = (struct btrfs_inode_extref *)ptr;
3871 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3876 * try to precache a NULL acl entry for files that don't have
3877 * any xattrs or acls
3879 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3880 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3881 if (first_xattr_slot != -1) {
3882 path->slots[0] = first_xattr_slot;
3883 ret = btrfs_load_inode_props(inode, path);
3886 "error loading props for ino %llu (root %llu): %d",
3887 btrfs_ino(BTRFS_I(inode)),
3888 root->root_key.objectid, ret);
3890 if (path != in_path)
3891 btrfs_free_path(path);
3894 cache_no_acl(inode);
3896 switch (inode->i_mode & S_IFMT) {
3898 inode->i_mapping->a_ops = &btrfs_aops;
3899 inode->i_fop = &btrfs_file_operations;
3900 inode->i_op = &btrfs_file_inode_operations;
3903 inode->i_fop = &btrfs_dir_file_operations;
3904 inode->i_op = &btrfs_dir_inode_operations;
3907 inode->i_op = &btrfs_symlink_inode_operations;
3908 inode_nohighmem(inode);
3909 inode->i_mapping->a_ops = &btrfs_aops;
3912 inode->i_op = &btrfs_special_inode_operations;
3913 init_special_inode(inode, inode->i_mode, rdev);
3917 btrfs_sync_inode_flags_to_i_flags(inode);
3922 * given a leaf and an inode, copy the inode fields into the leaf
3924 static void fill_inode_item(struct btrfs_trans_handle *trans,
3925 struct extent_buffer *leaf,
3926 struct btrfs_inode_item *item,
3927 struct inode *inode)
3929 struct btrfs_map_token token;
3932 btrfs_init_map_token(&token, leaf);
3934 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
3935 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
3936 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
3937 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
3938 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
3940 btrfs_set_token_timespec_sec(&token, &item->atime,
3941 inode->i_atime.tv_sec);
3942 btrfs_set_token_timespec_nsec(&token, &item->atime,
3943 inode->i_atime.tv_nsec);
3945 btrfs_set_token_timespec_sec(&token, &item->mtime,
3946 inode->i_mtime.tv_sec);
3947 btrfs_set_token_timespec_nsec(&token, &item->mtime,
3948 inode->i_mtime.tv_nsec);
3950 btrfs_set_token_timespec_sec(&token, &item->ctime,
3951 inode->i_ctime.tv_sec);
3952 btrfs_set_token_timespec_nsec(&token, &item->ctime,
3953 inode->i_ctime.tv_nsec);
3955 btrfs_set_token_timespec_sec(&token, &item->otime,
3956 BTRFS_I(inode)->i_otime.tv_sec);
3957 btrfs_set_token_timespec_nsec(&token, &item->otime,
3958 BTRFS_I(inode)->i_otime.tv_nsec);
3960 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
3961 btrfs_set_token_inode_generation(&token, item,
3962 BTRFS_I(inode)->generation);
3963 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
3964 btrfs_set_token_inode_transid(&token, item, trans->transid);
3965 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
3966 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
3967 BTRFS_I(inode)->ro_flags);
3968 btrfs_set_token_inode_flags(&token, item, flags);
3969 btrfs_set_token_inode_block_group(&token, item, 0);
3973 * copy everything in the in-memory inode into the btree.
3975 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3976 struct btrfs_root *root,
3977 struct btrfs_inode *inode)
3979 struct btrfs_inode_item *inode_item;
3980 struct btrfs_path *path;
3981 struct extent_buffer *leaf;
3984 path = btrfs_alloc_path();
3988 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
3995 leaf = path->nodes[0];
3996 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3997 struct btrfs_inode_item);
3999 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4000 btrfs_mark_buffer_dirty(leaf);
4001 btrfs_set_inode_last_trans(trans, inode);
4004 btrfs_free_path(path);
4009 * copy everything in the in-memory inode into the btree.
4011 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4012 struct btrfs_root *root,
4013 struct btrfs_inode *inode)
4015 struct btrfs_fs_info *fs_info = root->fs_info;
4019 * If the inode is a free space inode, we can deadlock during commit
4020 * if we put it into the delayed code.
4022 * The data relocation inode should also be directly updated
4025 if (!btrfs_is_free_space_inode(inode)
4026 && !btrfs_is_data_reloc_root(root)
4027 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4028 btrfs_update_root_times(trans, root);
4030 ret = btrfs_delayed_update_inode(trans, root, inode);
4032 btrfs_set_inode_last_trans(trans, inode);
4036 return btrfs_update_inode_item(trans, root, inode);
4039 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4040 struct btrfs_root *root, struct btrfs_inode *inode)
4044 ret = btrfs_update_inode(trans, root, inode);
4046 return btrfs_update_inode_item(trans, root, inode);
4051 * unlink helper that gets used here in inode.c and in the tree logging
4052 * recovery code. It remove a link in a directory with a given name, and
4053 * also drops the back refs in the inode to the directory
4055 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4056 struct btrfs_inode *dir,
4057 struct btrfs_inode *inode,
4058 const struct fscrypt_str *name,
4059 struct btrfs_rename_ctx *rename_ctx)
4061 struct btrfs_root *root = dir->root;
4062 struct btrfs_fs_info *fs_info = root->fs_info;
4063 struct btrfs_path *path;
4065 struct btrfs_dir_item *di;
4067 u64 ino = btrfs_ino(inode);
4068 u64 dir_ino = btrfs_ino(dir);
4070 path = btrfs_alloc_path();
4076 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4077 if (IS_ERR_OR_NULL(di)) {
4078 ret = di ? PTR_ERR(di) : -ENOENT;
4081 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4084 btrfs_release_path(path);
4087 * If we don't have dir index, we have to get it by looking up
4088 * the inode ref, since we get the inode ref, remove it directly,
4089 * it is unnecessary to do delayed deletion.
4091 * But if we have dir index, needn't search inode ref to get it.
4092 * Since the inode ref is close to the inode item, it is better
4093 * that we delay to delete it, and just do this deletion when
4094 * we update the inode item.
4096 if (inode->dir_index) {
4097 ret = btrfs_delayed_delete_inode_ref(inode);
4099 index = inode->dir_index;
4104 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4107 "failed to delete reference to %.*s, inode %llu parent %llu",
4108 name->len, name->name, ino, dir_ino);
4109 btrfs_abort_transaction(trans, ret);
4114 rename_ctx->index = index;
4116 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4118 btrfs_abort_transaction(trans, ret);
4123 * If we are in a rename context, we don't need to update anything in the
4124 * log. That will be done later during the rename by btrfs_log_new_name().
4125 * Besides that, doing it here would only cause extra unnecessary btree
4126 * operations on the log tree, increasing latency for applications.
4129 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4130 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4134 * If we have a pending delayed iput we could end up with the final iput
4135 * being run in btrfs-cleaner context. If we have enough of these built
4136 * up we can end up burning a lot of time in btrfs-cleaner without any
4137 * way to throttle the unlinks. Since we're currently holding a ref on
4138 * the inode we can run the delayed iput here without any issues as the
4139 * final iput won't be done until after we drop the ref we're currently
4142 btrfs_run_delayed_iput(fs_info, inode);
4144 btrfs_free_path(path);
4148 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4149 inode_inc_iversion(&inode->vfs_inode);
4150 inode_inc_iversion(&dir->vfs_inode);
4151 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4152 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4153 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4154 ret = btrfs_update_inode(trans, root, dir);
4159 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4160 struct btrfs_inode *dir, struct btrfs_inode *inode,
4161 const struct fscrypt_str *name)
4165 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4167 drop_nlink(&inode->vfs_inode);
4168 ret = btrfs_update_inode(trans, inode->root, inode);
4174 * helper to start transaction for unlink and rmdir.
4176 * unlink and rmdir are special in btrfs, they do not always free space, so
4177 * if we cannot make our reservations the normal way try and see if there is
4178 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4179 * allow the unlink to occur.
4181 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4183 struct btrfs_root *root = dir->root;
4185 return btrfs_start_transaction_fallback_global_rsv(root,
4186 BTRFS_UNLINK_METADATA_UNITS);
4189 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4191 struct btrfs_trans_handle *trans;
4192 struct inode *inode = d_inode(dentry);
4194 struct fscrypt_name fname;
4196 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4200 /* This needs to handle no-key deletions later on */
4202 trans = __unlink_start_trans(BTRFS_I(dir));
4203 if (IS_ERR(trans)) {
4204 ret = PTR_ERR(trans);
4208 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4211 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4216 if (inode->i_nlink == 0) {
4217 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4223 btrfs_end_transaction(trans);
4224 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4226 fscrypt_free_filename(&fname);
4230 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4231 struct btrfs_inode *dir, struct dentry *dentry)
4233 struct btrfs_root *root = dir->root;
4234 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4235 struct btrfs_path *path;
4236 struct extent_buffer *leaf;
4237 struct btrfs_dir_item *di;
4238 struct btrfs_key key;
4242 u64 dir_ino = btrfs_ino(dir);
4243 struct fscrypt_name fname;
4245 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4249 /* This needs to handle no-key deletions later on */
4251 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4252 objectid = inode->root->root_key.objectid;
4253 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4254 objectid = inode->location.objectid;
4257 fscrypt_free_filename(&fname);
4261 path = btrfs_alloc_path();
4267 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4268 &fname.disk_name, -1);
4269 if (IS_ERR_OR_NULL(di)) {
4270 ret = di ? PTR_ERR(di) : -ENOENT;
4274 leaf = path->nodes[0];
4275 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4276 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4277 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4279 btrfs_abort_transaction(trans, ret);
4282 btrfs_release_path(path);
4285 * This is a placeholder inode for a subvolume we didn't have a
4286 * reference to at the time of the snapshot creation. In the meantime
4287 * we could have renamed the real subvol link into our snapshot, so
4288 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4289 * Instead simply lookup the dir_index_item for this entry so we can
4290 * remove it. Otherwise we know we have a ref to the root and we can
4291 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4293 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4294 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4295 if (IS_ERR_OR_NULL(di)) {
4300 btrfs_abort_transaction(trans, ret);
4304 leaf = path->nodes[0];
4305 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4307 btrfs_release_path(path);
4309 ret = btrfs_del_root_ref(trans, objectid,
4310 root->root_key.objectid, dir_ino,
4311 &index, &fname.disk_name);
4313 btrfs_abort_transaction(trans, ret);
4318 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4320 btrfs_abort_transaction(trans, ret);
4324 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4325 inode_inc_iversion(&dir->vfs_inode);
4326 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4327 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4328 ret = btrfs_update_inode_fallback(trans, root, dir);
4330 btrfs_abort_transaction(trans, ret);
4332 btrfs_free_path(path);
4333 fscrypt_free_filename(&fname);
4338 * Helper to check if the subvolume references other subvolumes or if it's
4341 static noinline int may_destroy_subvol(struct btrfs_root *root)
4343 struct btrfs_fs_info *fs_info = root->fs_info;
4344 struct btrfs_path *path;
4345 struct btrfs_dir_item *di;
4346 struct btrfs_key key;
4347 struct fscrypt_str name = FSTR_INIT("default", 7);
4351 path = btrfs_alloc_path();
4355 /* Make sure this root isn't set as the default subvol */
4356 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4357 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4359 if (di && !IS_ERR(di)) {
4360 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4361 if (key.objectid == root->root_key.objectid) {
4364 "deleting default subvolume %llu is not allowed",
4368 btrfs_release_path(path);
4371 key.objectid = root->root_key.objectid;
4372 key.type = BTRFS_ROOT_REF_KEY;
4373 key.offset = (u64)-1;
4375 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4381 if (path->slots[0] > 0) {
4383 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4384 if (key.objectid == root->root_key.objectid &&
4385 key.type == BTRFS_ROOT_REF_KEY)
4389 btrfs_free_path(path);
4393 /* Delete all dentries for inodes belonging to the root */
4394 static void btrfs_prune_dentries(struct btrfs_root *root)
4396 struct btrfs_fs_info *fs_info = root->fs_info;
4397 struct rb_node *node;
4398 struct rb_node *prev;
4399 struct btrfs_inode *entry;
4400 struct inode *inode;
4403 if (!BTRFS_FS_ERROR(fs_info))
4404 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4406 spin_lock(&root->inode_lock);
4408 node = root->inode_tree.rb_node;
4412 entry = rb_entry(node, struct btrfs_inode, rb_node);
4414 if (objectid < btrfs_ino(entry))
4415 node = node->rb_left;
4416 else if (objectid > btrfs_ino(entry))
4417 node = node->rb_right;
4423 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4424 if (objectid <= btrfs_ino(entry)) {
4428 prev = rb_next(prev);
4432 entry = rb_entry(node, struct btrfs_inode, rb_node);
4433 objectid = btrfs_ino(entry) + 1;
4434 inode = igrab(&entry->vfs_inode);
4436 spin_unlock(&root->inode_lock);
4437 if (atomic_read(&inode->i_count) > 1)
4438 d_prune_aliases(inode);
4440 * btrfs_drop_inode will have it removed from the inode
4441 * cache when its usage count hits zero.
4445 spin_lock(&root->inode_lock);
4449 if (cond_resched_lock(&root->inode_lock))
4452 node = rb_next(node);
4454 spin_unlock(&root->inode_lock);
4457 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4459 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4460 struct btrfs_root *root = dir->root;
4461 struct inode *inode = d_inode(dentry);
4462 struct btrfs_root *dest = BTRFS_I(inode)->root;
4463 struct btrfs_trans_handle *trans;
4464 struct btrfs_block_rsv block_rsv;
4469 * Don't allow to delete a subvolume with send in progress. This is
4470 * inside the inode lock so the error handling that has to drop the bit
4471 * again is not run concurrently.
4473 spin_lock(&dest->root_item_lock);
4474 if (dest->send_in_progress) {
4475 spin_unlock(&dest->root_item_lock);
4477 "attempt to delete subvolume %llu during send",
4478 dest->root_key.objectid);
4481 if (atomic_read(&dest->nr_swapfiles)) {
4482 spin_unlock(&dest->root_item_lock);
4484 "attempt to delete subvolume %llu with active swapfile",
4485 root->root_key.objectid);
4488 root_flags = btrfs_root_flags(&dest->root_item);
4489 btrfs_set_root_flags(&dest->root_item,
4490 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4491 spin_unlock(&dest->root_item_lock);
4493 down_write(&fs_info->subvol_sem);
4495 ret = may_destroy_subvol(dest);
4499 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4501 * One for dir inode,
4502 * two for dir entries,
4503 * two for root ref/backref.
4505 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4509 trans = btrfs_start_transaction(root, 0);
4510 if (IS_ERR(trans)) {
4511 ret = PTR_ERR(trans);
4514 trans->block_rsv = &block_rsv;
4515 trans->bytes_reserved = block_rsv.size;
4517 btrfs_record_snapshot_destroy(trans, dir);
4519 ret = btrfs_unlink_subvol(trans, dir, dentry);
4521 btrfs_abort_transaction(trans, ret);
4525 ret = btrfs_record_root_in_trans(trans, dest);
4527 btrfs_abort_transaction(trans, ret);
4531 memset(&dest->root_item.drop_progress, 0,
4532 sizeof(dest->root_item.drop_progress));
4533 btrfs_set_root_drop_level(&dest->root_item, 0);
4534 btrfs_set_root_refs(&dest->root_item, 0);
4536 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4537 ret = btrfs_insert_orphan_item(trans,
4539 dest->root_key.objectid);
4541 btrfs_abort_transaction(trans, ret);
4546 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4547 BTRFS_UUID_KEY_SUBVOL,
4548 dest->root_key.objectid);
4549 if (ret && ret != -ENOENT) {
4550 btrfs_abort_transaction(trans, ret);
4553 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4554 ret = btrfs_uuid_tree_remove(trans,
4555 dest->root_item.received_uuid,
4556 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4557 dest->root_key.objectid);
4558 if (ret && ret != -ENOENT) {
4559 btrfs_abort_transaction(trans, ret);
4564 free_anon_bdev(dest->anon_dev);
4567 trans->block_rsv = NULL;
4568 trans->bytes_reserved = 0;
4569 ret = btrfs_end_transaction(trans);
4570 inode->i_flags |= S_DEAD;
4572 btrfs_subvolume_release_metadata(root, &block_rsv);
4574 up_write(&fs_info->subvol_sem);
4576 spin_lock(&dest->root_item_lock);
4577 root_flags = btrfs_root_flags(&dest->root_item);
4578 btrfs_set_root_flags(&dest->root_item,
4579 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4580 spin_unlock(&dest->root_item_lock);
4582 d_invalidate(dentry);
4583 btrfs_prune_dentries(dest);
4584 ASSERT(dest->send_in_progress == 0);
4590 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4592 struct inode *inode = d_inode(dentry);
4593 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4595 struct btrfs_trans_handle *trans;
4596 u64 last_unlink_trans;
4597 struct fscrypt_name fname;
4599 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4601 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4602 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4604 "extent tree v2 doesn't support snapshot deletion yet");
4607 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4610 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4614 /* This needs to handle no-key deletions later on */
4616 trans = __unlink_start_trans(BTRFS_I(dir));
4617 if (IS_ERR(trans)) {
4618 err = PTR_ERR(trans);
4622 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4623 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4627 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4631 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4633 /* now the directory is empty */
4634 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4637 btrfs_i_size_write(BTRFS_I(inode), 0);
4639 * Propagate the last_unlink_trans value of the deleted dir to
4640 * its parent directory. This is to prevent an unrecoverable
4641 * log tree in the case we do something like this:
4643 * 2) create snapshot under dir foo
4644 * 3) delete the snapshot
4647 * 6) fsync foo or some file inside foo
4649 if (last_unlink_trans >= trans->transid)
4650 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4653 btrfs_end_transaction(trans);
4655 btrfs_btree_balance_dirty(fs_info);
4656 fscrypt_free_filename(&fname);
4662 * btrfs_truncate_block - read, zero a chunk and write a block
4663 * @inode - inode that we're zeroing
4664 * @from - the offset to start zeroing
4665 * @len - the length to zero, 0 to zero the entire range respective to the
4667 * @front - zero up to the offset instead of from the offset on
4669 * This will find the block for the "from" offset and cow the block and zero the
4670 * part we want to zero. This is used with truncate and hole punching.
4672 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4675 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4676 struct address_space *mapping = inode->vfs_inode.i_mapping;
4677 struct extent_io_tree *io_tree = &inode->io_tree;
4678 struct btrfs_ordered_extent *ordered;
4679 struct extent_state *cached_state = NULL;
4680 struct extent_changeset *data_reserved = NULL;
4681 bool only_release_metadata = false;
4682 u32 blocksize = fs_info->sectorsize;
4683 pgoff_t index = from >> PAGE_SHIFT;
4684 unsigned offset = from & (blocksize - 1);
4686 gfp_t mask = btrfs_alloc_write_mask(mapping);
4687 size_t write_bytes = blocksize;
4692 if (IS_ALIGNED(offset, blocksize) &&
4693 (!len || IS_ALIGNED(len, blocksize)))
4696 block_start = round_down(from, blocksize);
4697 block_end = block_start + blocksize - 1;
4699 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4702 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4703 /* For nocow case, no need to reserve data space */
4704 only_release_metadata = true;
4709 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4711 if (!only_release_metadata)
4712 btrfs_free_reserved_data_space(inode, data_reserved,
4713 block_start, blocksize);
4717 page = find_or_create_page(mapping, index, mask);
4719 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4721 btrfs_delalloc_release_extents(inode, blocksize);
4726 if (!PageUptodate(page)) {
4727 ret = btrfs_read_folio(NULL, page_folio(page));
4729 if (page->mapping != mapping) {
4734 if (!PageUptodate(page)) {
4741 * We unlock the page after the io is completed and then re-lock it
4742 * above. release_folio() could have come in between that and cleared
4743 * PagePrivate(), but left the page in the mapping. Set the page mapped
4744 * here to make sure it's properly set for the subpage stuff.
4746 ret = set_page_extent_mapped(page);
4750 wait_on_page_writeback(page);
4752 lock_extent(io_tree, block_start, block_end, &cached_state);
4754 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4756 unlock_extent(io_tree, block_start, block_end, &cached_state);
4759 btrfs_start_ordered_extent(ordered);
4760 btrfs_put_ordered_extent(ordered);
4764 clear_extent_bit(&inode->io_tree, block_start, block_end,
4765 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4768 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4771 unlock_extent(io_tree, block_start, block_end, &cached_state);
4775 if (offset != blocksize) {
4777 len = blocksize - offset;
4779 memzero_page(page, (block_start - page_offset(page)),
4782 memzero_page(page, (block_start - page_offset(page)) + offset,
4785 btrfs_page_clear_checked(fs_info, page, block_start,
4786 block_end + 1 - block_start);
4787 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4788 unlock_extent(io_tree, block_start, block_end, &cached_state);
4790 if (only_release_metadata)
4791 set_extent_bit(&inode->io_tree, block_start, block_end,
4792 EXTENT_NORESERVE, NULL);
4796 if (only_release_metadata)
4797 btrfs_delalloc_release_metadata(inode, blocksize, true);
4799 btrfs_delalloc_release_space(inode, data_reserved,
4800 block_start, blocksize, true);
4802 btrfs_delalloc_release_extents(inode, blocksize);
4806 if (only_release_metadata)
4807 btrfs_check_nocow_unlock(inode);
4808 extent_changeset_free(data_reserved);
4812 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4813 u64 offset, u64 len)
4815 struct btrfs_fs_info *fs_info = root->fs_info;
4816 struct btrfs_trans_handle *trans;
4817 struct btrfs_drop_extents_args drop_args = { 0 };
4821 * If NO_HOLES is enabled, we don't need to do anything.
4822 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4823 * or btrfs_update_inode() will be called, which guarantee that the next
4824 * fsync will know this inode was changed and needs to be logged.
4826 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4830 * 1 - for the one we're dropping
4831 * 1 - for the one we're adding
4832 * 1 - for updating the inode.
4834 trans = btrfs_start_transaction(root, 3);
4836 return PTR_ERR(trans);
4838 drop_args.start = offset;
4839 drop_args.end = offset + len;
4840 drop_args.drop_cache = true;
4842 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4844 btrfs_abort_transaction(trans, ret);
4845 btrfs_end_transaction(trans);
4849 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4851 btrfs_abort_transaction(trans, ret);
4853 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4854 btrfs_update_inode(trans, root, inode);
4856 btrfs_end_transaction(trans);
4861 * This function puts in dummy file extents for the area we're creating a hole
4862 * for. So if we are truncating this file to a larger size we need to insert
4863 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4864 * the range between oldsize and size
4866 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4868 struct btrfs_root *root = inode->root;
4869 struct btrfs_fs_info *fs_info = root->fs_info;
4870 struct extent_io_tree *io_tree = &inode->io_tree;
4871 struct extent_map *em = NULL;
4872 struct extent_state *cached_state = NULL;
4873 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4874 u64 block_end = ALIGN(size, fs_info->sectorsize);
4881 * If our size started in the middle of a block we need to zero out the
4882 * rest of the block before we expand the i_size, otherwise we could
4883 * expose stale data.
4885 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4889 if (size <= hole_start)
4892 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
4894 cur_offset = hole_start;
4896 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4897 block_end - cur_offset);
4903 last_byte = min(extent_map_end(em), block_end);
4904 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4905 hole_size = last_byte - cur_offset;
4907 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4908 struct extent_map *hole_em;
4910 err = maybe_insert_hole(root, inode, cur_offset,
4915 err = btrfs_inode_set_file_extent_range(inode,
4916 cur_offset, hole_size);
4920 hole_em = alloc_extent_map();
4922 btrfs_drop_extent_map_range(inode, cur_offset,
4923 cur_offset + hole_size - 1,
4925 btrfs_set_inode_full_sync(inode);
4928 hole_em->start = cur_offset;
4929 hole_em->len = hole_size;
4930 hole_em->orig_start = cur_offset;
4932 hole_em->block_start = EXTENT_MAP_HOLE;
4933 hole_em->block_len = 0;
4934 hole_em->orig_block_len = 0;
4935 hole_em->ram_bytes = hole_size;
4936 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4937 hole_em->generation = fs_info->generation;
4939 err = btrfs_replace_extent_map_range(inode, hole_em, true);
4940 free_extent_map(hole_em);
4942 err = btrfs_inode_set_file_extent_range(inode,
4943 cur_offset, hole_size);
4948 free_extent_map(em);
4950 cur_offset = last_byte;
4951 if (cur_offset >= block_end)
4954 free_extent_map(em);
4955 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
4959 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4961 struct btrfs_root *root = BTRFS_I(inode)->root;
4962 struct btrfs_trans_handle *trans;
4963 loff_t oldsize = i_size_read(inode);
4964 loff_t newsize = attr->ia_size;
4965 int mask = attr->ia_valid;
4969 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4970 * special case where we need to update the times despite not having
4971 * these flags set. For all other operations the VFS set these flags
4972 * explicitly if it wants a timestamp update.
4974 if (newsize != oldsize) {
4975 inode_inc_iversion(inode);
4976 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
4977 inode->i_mtime = current_time(inode);
4978 inode->i_ctime = inode->i_mtime;
4982 if (newsize > oldsize) {
4984 * Don't do an expanding truncate while snapshotting is ongoing.
4985 * This is to ensure the snapshot captures a fully consistent
4986 * state of this file - if the snapshot captures this expanding
4987 * truncation, it must capture all writes that happened before
4990 btrfs_drew_write_lock(&root->snapshot_lock);
4991 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
4993 btrfs_drew_write_unlock(&root->snapshot_lock);
4997 trans = btrfs_start_transaction(root, 1);
4998 if (IS_ERR(trans)) {
4999 btrfs_drew_write_unlock(&root->snapshot_lock);
5000 return PTR_ERR(trans);
5003 i_size_write(inode, newsize);
5004 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5005 pagecache_isize_extended(inode, oldsize, newsize);
5006 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5007 btrfs_drew_write_unlock(&root->snapshot_lock);
5008 btrfs_end_transaction(trans);
5010 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5012 if (btrfs_is_zoned(fs_info)) {
5013 ret = btrfs_wait_ordered_range(inode,
5014 ALIGN(newsize, fs_info->sectorsize),
5021 * We're truncating a file that used to have good data down to
5022 * zero. Make sure any new writes to the file get on disk
5026 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5027 &BTRFS_I(inode)->runtime_flags);
5029 truncate_setsize(inode, newsize);
5031 inode_dio_wait(inode);
5033 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5034 if (ret && inode->i_nlink) {
5038 * Truncate failed, so fix up the in-memory size. We
5039 * adjusted disk_i_size down as we removed extents, so
5040 * wait for disk_i_size to be stable and then update the
5041 * in-memory size to match.
5043 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5046 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5053 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5056 struct inode *inode = d_inode(dentry);
5057 struct btrfs_root *root = BTRFS_I(inode)->root;
5060 if (btrfs_root_readonly(root))
5063 err = setattr_prepare(idmap, dentry, attr);
5067 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5068 err = btrfs_setsize(inode, attr);
5073 if (attr->ia_valid) {
5074 setattr_copy(idmap, inode, attr);
5075 inode_inc_iversion(inode);
5076 err = btrfs_dirty_inode(BTRFS_I(inode));
5078 if (!err && attr->ia_valid & ATTR_MODE)
5079 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5086 * While truncating the inode pages during eviction, we get the VFS
5087 * calling btrfs_invalidate_folio() against each folio of the inode. This
5088 * is slow because the calls to btrfs_invalidate_folio() result in a
5089 * huge amount of calls to lock_extent() and clear_extent_bit(),
5090 * which keep merging and splitting extent_state structures over and over,
5091 * wasting lots of time.
5093 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5094 * skip all those expensive operations on a per folio basis and do only
5095 * the ordered io finishing, while we release here the extent_map and
5096 * extent_state structures, without the excessive merging and splitting.
5098 static void evict_inode_truncate_pages(struct inode *inode)
5100 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5101 struct rb_node *node;
5103 ASSERT(inode->i_state & I_FREEING);
5104 truncate_inode_pages_final(&inode->i_data);
5106 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5109 * Keep looping until we have no more ranges in the io tree.
5110 * We can have ongoing bios started by readahead that have
5111 * their endio callback (extent_io.c:end_bio_extent_readpage)
5112 * still in progress (unlocked the pages in the bio but did not yet
5113 * unlocked the ranges in the io tree). Therefore this means some
5114 * ranges can still be locked and eviction started because before
5115 * submitting those bios, which are executed by a separate task (work
5116 * queue kthread), inode references (inode->i_count) were not taken
5117 * (which would be dropped in the end io callback of each bio).
5118 * Therefore here we effectively end up waiting for those bios and
5119 * anyone else holding locked ranges without having bumped the inode's
5120 * reference count - if we don't do it, when they access the inode's
5121 * io_tree to unlock a range it may be too late, leading to an
5122 * use-after-free issue.
5124 spin_lock(&io_tree->lock);
5125 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5126 struct extent_state *state;
5127 struct extent_state *cached_state = NULL;
5130 unsigned state_flags;
5132 node = rb_first(&io_tree->state);
5133 state = rb_entry(node, struct extent_state, rb_node);
5134 start = state->start;
5136 state_flags = state->state;
5137 spin_unlock(&io_tree->lock);
5139 lock_extent(io_tree, start, end, &cached_state);
5142 * If still has DELALLOC flag, the extent didn't reach disk,
5143 * and its reserved space won't be freed by delayed_ref.
5144 * So we need to free its reserved space here.
5145 * (Refer to comment in btrfs_invalidate_folio, case 2)
5147 * Note, end is the bytenr of last byte, so we need + 1 here.
5149 if (state_flags & EXTENT_DELALLOC)
5150 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5153 clear_extent_bit(io_tree, start, end,
5154 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5158 spin_lock(&io_tree->lock);
5160 spin_unlock(&io_tree->lock);
5163 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5164 struct btrfs_block_rsv *rsv)
5166 struct btrfs_fs_info *fs_info = root->fs_info;
5167 struct btrfs_trans_handle *trans;
5168 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5172 * Eviction should be taking place at some place safe because of our
5173 * delayed iputs. However the normal flushing code will run delayed
5174 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5176 * We reserve the delayed_refs_extra here again because we can't use
5177 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5178 * above. We reserve our extra bit here because we generate a ton of
5179 * delayed refs activity by truncating.
5181 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5182 * if we fail to make this reservation we can re-try without the
5183 * delayed_refs_extra so we can make some forward progress.
5185 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5186 BTRFS_RESERVE_FLUSH_EVICT);
5188 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5189 BTRFS_RESERVE_FLUSH_EVICT);
5192 "could not allocate space for delete; will truncate on mount");
5193 return ERR_PTR(-ENOSPC);
5195 delayed_refs_extra = 0;
5198 trans = btrfs_join_transaction(root);
5202 if (delayed_refs_extra) {
5203 trans->block_rsv = &fs_info->trans_block_rsv;
5204 trans->bytes_reserved = delayed_refs_extra;
5205 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5206 delayed_refs_extra, true);
5211 void btrfs_evict_inode(struct inode *inode)
5213 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5214 struct btrfs_trans_handle *trans;
5215 struct btrfs_root *root = BTRFS_I(inode)->root;
5216 struct btrfs_block_rsv *rsv = NULL;
5219 trace_btrfs_inode_evict(inode);
5222 fsverity_cleanup_inode(inode);
5227 evict_inode_truncate_pages(inode);
5229 if (inode->i_nlink &&
5230 ((btrfs_root_refs(&root->root_item) != 0 &&
5231 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5232 btrfs_is_free_space_inode(BTRFS_I(inode))))
5235 if (is_bad_inode(inode))
5238 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5241 if (inode->i_nlink > 0) {
5242 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5243 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5248 * This makes sure the inode item in tree is uptodate and the space for
5249 * the inode update is released.
5251 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5256 * This drops any pending insert or delete operations we have for this
5257 * inode. We could have a delayed dir index deletion queued up, but
5258 * we're removing the inode completely so that'll be taken care of in
5261 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5263 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5266 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5267 rsv->failfast = true;
5269 btrfs_i_size_write(BTRFS_I(inode), 0);
5272 struct btrfs_truncate_control control = {
5273 .inode = BTRFS_I(inode),
5274 .ino = btrfs_ino(BTRFS_I(inode)),
5279 trans = evict_refill_and_join(root, rsv);
5283 trans->block_rsv = rsv;
5285 ret = btrfs_truncate_inode_items(trans, root, &control);
5286 trans->block_rsv = &fs_info->trans_block_rsv;
5287 btrfs_end_transaction(trans);
5289 * We have not added new delayed items for our inode after we
5290 * have flushed its delayed items, so no need to throttle on
5291 * delayed items. However we have modified extent buffers.
5293 btrfs_btree_balance_dirty_nodelay(fs_info);
5294 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5301 * Errors here aren't a big deal, it just means we leave orphan items in
5302 * the tree. They will be cleaned up on the next mount. If the inode
5303 * number gets reused, cleanup deletes the orphan item without doing
5304 * anything, and unlink reuses the existing orphan item.
5306 * If it turns out that we are dropping too many of these, we might want
5307 * to add a mechanism for retrying these after a commit.
5309 trans = evict_refill_and_join(root, rsv);
5310 if (!IS_ERR(trans)) {
5311 trans->block_rsv = rsv;
5312 btrfs_orphan_del(trans, BTRFS_I(inode));
5313 trans->block_rsv = &fs_info->trans_block_rsv;
5314 btrfs_end_transaction(trans);
5318 btrfs_free_block_rsv(fs_info, rsv);
5320 * If we didn't successfully delete, the orphan item will still be in
5321 * the tree and we'll retry on the next mount. Again, we might also want
5322 * to retry these periodically in the future.
5324 btrfs_remove_delayed_node(BTRFS_I(inode));
5325 fsverity_cleanup_inode(inode);
5330 * Return the key found in the dir entry in the location pointer, fill @type
5331 * with BTRFS_FT_*, and return 0.
5333 * If no dir entries were found, returns -ENOENT.
5334 * If found a corrupted location in dir entry, returns -EUCLEAN.
5336 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5337 struct btrfs_key *location, u8 *type)
5339 struct btrfs_dir_item *di;
5340 struct btrfs_path *path;
5341 struct btrfs_root *root = dir->root;
5343 struct fscrypt_name fname;
5345 path = btrfs_alloc_path();
5349 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5353 * fscrypt_setup_filename() should never return a positive value, but
5354 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5358 /* This needs to handle no-key deletions later on */
5360 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5361 &fname.disk_name, 0);
5362 if (IS_ERR_OR_NULL(di)) {
5363 ret = di ? PTR_ERR(di) : -ENOENT;
5367 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5368 if (location->type != BTRFS_INODE_ITEM_KEY &&
5369 location->type != BTRFS_ROOT_ITEM_KEY) {
5371 btrfs_warn(root->fs_info,
5372 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5373 __func__, fname.disk_name.name, btrfs_ino(dir),
5374 location->objectid, location->type, location->offset);
5377 *type = btrfs_dir_ftype(path->nodes[0], di);
5379 fscrypt_free_filename(&fname);
5380 btrfs_free_path(path);
5385 * when we hit a tree root in a directory, the btrfs part of the inode
5386 * needs to be changed to reflect the root directory of the tree root. This
5387 * is kind of like crossing a mount point.
5389 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5390 struct btrfs_inode *dir,
5391 struct dentry *dentry,
5392 struct btrfs_key *location,
5393 struct btrfs_root **sub_root)
5395 struct btrfs_path *path;
5396 struct btrfs_root *new_root;
5397 struct btrfs_root_ref *ref;
5398 struct extent_buffer *leaf;
5399 struct btrfs_key key;
5402 struct fscrypt_name fname;
5404 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5408 path = btrfs_alloc_path();
5415 key.objectid = dir->root->root_key.objectid;
5416 key.type = BTRFS_ROOT_REF_KEY;
5417 key.offset = location->objectid;
5419 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5426 leaf = path->nodes[0];
5427 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5428 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5429 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5432 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5433 (unsigned long)(ref + 1), fname.disk_name.len);
5437 btrfs_release_path(path);
5439 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5440 if (IS_ERR(new_root)) {
5441 err = PTR_ERR(new_root);
5445 *sub_root = new_root;
5446 location->objectid = btrfs_root_dirid(&new_root->root_item);
5447 location->type = BTRFS_INODE_ITEM_KEY;
5448 location->offset = 0;
5451 btrfs_free_path(path);
5452 fscrypt_free_filename(&fname);
5456 static void inode_tree_add(struct btrfs_inode *inode)
5458 struct btrfs_root *root = inode->root;
5459 struct btrfs_inode *entry;
5461 struct rb_node *parent;
5462 struct rb_node *new = &inode->rb_node;
5463 u64 ino = btrfs_ino(inode);
5465 if (inode_unhashed(&inode->vfs_inode))
5468 spin_lock(&root->inode_lock);
5469 p = &root->inode_tree.rb_node;
5472 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5474 if (ino < btrfs_ino(entry))
5475 p = &parent->rb_left;
5476 else if (ino > btrfs_ino(entry))
5477 p = &parent->rb_right;
5479 WARN_ON(!(entry->vfs_inode.i_state &
5480 (I_WILL_FREE | I_FREEING)));
5481 rb_replace_node(parent, new, &root->inode_tree);
5482 RB_CLEAR_NODE(parent);
5483 spin_unlock(&root->inode_lock);
5487 rb_link_node(new, parent, p);
5488 rb_insert_color(new, &root->inode_tree);
5489 spin_unlock(&root->inode_lock);
5492 static void inode_tree_del(struct btrfs_inode *inode)
5494 struct btrfs_root *root = inode->root;
5497 spin_lock(&root->inode_lock);
5498 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5499 rb_erase(&inode->rb_node, &root->inode_tree);
5500 RB_CLEAR_NODE(&inode->rb_node);
5501 empty = RB_EMPTY_ROOT(&root->inode_tree);
5503 spin_unlock(&root->inode_lock);
5505 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5506 spin_lock(&root->inode_lock);
5507 empty = RB_EMPTY_ROOT(&root->inode_tree);
5508 spin_unlock(&root->inode_lock);
5510 btrfs_add_dead_root(root);
5515 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5517 struct btrfs_iget_args *args = p;
5519 inode->i_ino = args->ino;
5520 BTRFS_I(inode)->location.objectid = args->ino;
5521 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5522 BTRFS_I(inode)->location.offset = 0;
5523 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5524 BUG_ON(args->root && !BTRFS_I(inode)->root);
5526 if (args->root && args->root == args->root->fs_info->tree_root &&
5527 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5528 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5529 &BTRFS_I(inode)->runtime_flags);
5533 static int btrfs_find_actor(struct inode *inode, void *opaque)
5535 struct btrfs_iget_args *args = opaque;
5537 return args->ino == BTRFS_I(inode)->location.objectid &&
5538 args->root == BTRFS_I(inode)->root;
5541 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5542 struct btrfs_root *root)
5544 struct inode *inode;
5545 struct btrfs_iget_args args;
5546 unsigned long hashval = btrfs_inode_hash(ino, root);
5551 inode = iget5_locked(s, hashval, btrfs_find_actor,
5552 btrfs_init_locked_inode,
5558 * Get an inode object given its inode number and corresponding root.
5559 * Path can be preallocated to prevent recursing back to iget through
5560 * allocator. NULL is also valid but may require an additional allocation
5563 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5564 struct btrfs_root *root, struct btrfs_path *path)
5566 struct inode *inode;
5568 inode = btrfs_iget_locked(s, ino, root);
5570 return ERR_PTR(-ENOMEM);
5572 if (inode->i_state & I_NEW) {
5575 ret = btrfs_read_locked_inode(inode, path);
5577 inode_tree_add(BTRFS_I(inode));
5578 unlock_new_inode(inode);
5582 * ret > 0 can come from btrfs_search_slot called by
5583 * btrfs_read_locked_inode, this means the inode item
5588 inode = ERR_PTR(ret);
5595 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5597 return btrfs_iget_path(s, ino, root, NULL);
5600 static struct inode *new_simple_dir(struct super_block *s,
5601 struct btrfs_key *key,
5602 struct btrfs_root *root)
5604 struct inode *inode = new_inode(s);
5607 return ERR_PTR(-ENOMEM);
5609 BTRFS_I(inode)->root = btrfs_grab_root(root);
5610 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5611 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5613 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5615 * We only need lookup, the rest is read-only and there's no inode
5616 * associated with the dentry
5618 inode->i_op = &simple_dir_inode_operations;
5619 inode->i_opflags &= ~IOP_XATTR;
5620 inode->i_fop = &simple_dir_operations;
5621 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5622 inode->i_mtime = current_time(inode);
5623 inode->i_atime = inode->i_mtime;
5624 inode->i_ctime = inode->i_mtime;
5625 BTRFS_I(inode)->i_otime = inode->i_mtime;
5630 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5631 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5632 static_assert(BTRFS_FT_DIR == FT_DIR);
5633 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5634 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5635 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5636 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5637 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5639 static inline u8 btrfs_inode_type(struct inode *inode)
5641 return fs_umode_to_ftype(inode->i_mode);
5644 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5646 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5647 struct inode *inode;
5648 struct btrfs_root *root = BTRFS_I(dir)->root;
5649 struct btrfs_root *sub_root = root;
5650 struct btrfs_key location;
5654 if (dentry->d_name.len > BTRFS_NAME_LEN)
5655 return ERR_PTR(-ENAMETOOLONG);
5657 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5659 return ERR_PTR(ret);
5661 if (location.type == BTRFS_INODE_ITEM_KEY) {
5662 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5666 /* Do extra check against inode mode with di_type */
5667 if (btrfs_inode_type(inode) != di_type) {
5669 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5670 inode->i_mode, btrfs_inode_type(inode),
5673 return ERR_PTR(-EUCLEAN);
5678 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5679 &location, &sub_root);
5682 inode = ERR_PTR(ret);
5684 inode = new_simple_dir(dir->i_sb, &location, root);
5686 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5687 btrfs_put_root(sub_root);
5692 down_read(&fs_info->cleanup_work_sem);
5693 if (!sb_rdonly(inode->i_sb))
5694 ret = btrfs_orphan_cleanup(sub_root);
5695 up_read(&fs_info->cleanup_work_sem);
5698 inode = ERR_PTR(ret);
5705 static int btrfs_dentry_delete(const struct dentry *dentry)
5707 struct btrfs_root *root;
5708 struct inode *inode = d_inode(dentry);
5710 if (!inode && !IS_ROOT(dentry))
5711 inode = d_inode(dentry->d_parent);
5714 root = BTRFS_I(inode)->root;
5715 if (btrfs_root_refs(&root->root_item) == 0)
5718 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5724 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5727 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5729 if (inode == ERR_PTR(-ENOENT))
5731 return d_splice_alias(inode, dentry);
5735 * Find the highest existing sequence number in a directory and then set the
5736 * in-memory index_cnt variable to the first free sequence number.
5738 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5740 struct btrfs_root *root = inode->root;
5741 struct btrfs_key key, found_key;
5742 struct btrfs_path *path;
5743 struct extent_buffer *leaf;
5746 key.objectid = btrfs_ino(inode);
5747 key.type = BTRFS_DIR_INDEX_KEY;
5748 key.offset = (u64)-1;
5750 path = btrfs_alloc_path();
5754 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5757 /* FIXME: we should be able to handle this */
5762 if (path->slots[0] == 0) {
5763 inode->index_cnt = BTRFS_DIR_START_INDEX;
5769 leaf = path->nodes[0];
5770 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5772 if (found_key.objectid != btrfs_ino(inode) ||
5773 found_key.type != BTRFS_DIR_INDEX_KEY) {
5774 inode->index_cnt = BTRFS_DIR_START_INDEX;
5778 inode->index_cnt = found_key.offset + 1;
5780 btrfs_free_path(path);
5784 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5786 if (dir->index_cnt == (u64)-1) {
5789 ret = btrfs_inode_delayed_dir_index_count(dir);
5791 ret = btrfs_set_inode_index_count(dir);
5797 *index = dir->index_cnt;
5803 * All this infrastructure exists because dir_emit can fault, and we are holding
5804 * the tree lock when doing readdir. For now just allocate a buffer and copy
5805 * our information into that, and then dir_emit from the buffer. This is
5806 * similar to what NFS does, only we don't keep the buffer around in pagecache
5807 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5808 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5811 static int btrfs_opendir(struct inode *inode, struct file *file)
5813 struct btrfs_file_private *private;
5817 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5821 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5824 private->last_index = last_index;
5825 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5826 if (!private->filldir_buf) {
5830 file->private_data = private;
5841 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5844 struct dir_entry *entry = addr;
5845 char *name = (char *)(entry + 1);
5847 ctx->pos = get_unaligned(&entry->offset);
5848 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5849 get_unaligned(&entry->ino),
5850 get_unaligned(&entry->type)))
5852 addr += sizeof(struct dir_entry) +
5853 get_unaligned(&entry->name_len);
5859 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5861 struct inode *inode = file_inode(file);
5862 struct btrfs_root *root = BTRFS_I(inode)->root;
5863 struct btrfs_file_private *private = file->private_data;
5864 struct btrfs_dir_item *di;
5865 struct btrfs_key key;
5866 struct btrfs_key found_key;
5867 struct btrfs_path *path;
5869 struct list_head ins_list;
5870 struct list_head del_list;
5877 struct btrfs_key location;
5879 if (!dir_emit_dots(file, ctx))
5882 path = btrfs_alloc_path();
5886 addr = private->filldir_buf;
5887 path->reada = READA_FORWARD;
5889 INIT_LIST_HEAD(&ins_list);
5890 INIT_LIST_HEAD(&del_list);
5891 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
5892 &ins_list, &del_list);
5895 key.type = BTRFS_DIR_INDEX_KEY;
5896 key.offset = ctx->pos;
5897 key.objectid = btrfs_ino(BTRFS_I(inode));
5899 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
5900 struct dir_entry *entry;
5901 struct extent_buffer *leaf = path->nodes[0];
5904 if (found_key.objectid != key.objectid)
5906 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5908 if (found_key.offset < ctx->pos)
5910 if (found_key.offset > private->last_index)
5912 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5914 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
5915 name_len = btrfs_dir_name_len(leaf, di);
5916 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5918 btrfs_release_path(path);
5919 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5922 addr = private->filldir_buf;
5928 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
5930 name_ptr = (char *)(entry + 1);
5931 read_extent_buffer(leaf, name_ptr,
5932 (unsigned long)(di + 1), name_len);
5933 put_unaligned(name_len, &entry->name_len);
5934 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
5935 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5936 put_unaligned(location.objectid, &entry->ino);
5937 put_unaligned(found_key.offset, &entry->offset);
5939 addr += sizeof(struct dir_entry) + name_len;
5940 total_len += sizeof(struct dir_entry) + name_len;
5942 /* Catch error encountered during iteration */
5946 btrfs_release_path(path);
5948 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5952 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5957 * Stop new entries from being returned after we return the last
5960 * New directory entries are assigned a strictly increasing
5961 * offset. This means that new entries created during readdir
5962 * are *guaranteed* to be seen in the future by that readdir.
5963 * This has broken buggy programs which operate on names as
5964 * they're returned by readdir. Until we re-use freed offsets
5965 * we have this hack to stop new entries from being returned
5966 * under the assumption that they'll never reach this huge
5969 * This is being careful not to overflow 32bit loff_t unless the
5970 * last entry requires it because doing so has broken 32bit apps
5973 if (ctx->pos >= INT_MAX)
5974 ctx->pos = LLONG_MAX;
5981 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5982 btrfs_free_path(path);
5987 * This is somewhat expensive, updating the tree every time the
5988 * inode changes. But, it is most likely to find the inode in cache.
5989 * FIXME, needs more benchmarking...there are no reasons other than performance
5990 * to keep or drop this code.
5992 static int btrfs_dirty_inode(struct btrfs_inode *inode)
5994 struct btrfs_root *root = inode->root;
5995 struct btrfs_fs_info *fs_info = root->fs_info;
5996 struct btrfs_trans_handle *trans;
5999 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6002 trans = btrfs_join_transaction(root);
6004 return PTR_ERR(trans);
6006 ret = btrfs_update_inode(trans, root, inode);
6007 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6008 /* whoops, lets try again with the full transaction */
6009 btrfs_end_transaction(trans);
6010 trans = btrfs_start_transaction(root, 1);
6012 return PTR_ERR(trans);
6014 ret = btrfs_update_inode(trans, root, inode);
6016 btrfs_end_transaction(trans);
6017 if (inode->delayed_node)
6018 btrfs_balance_delayed_items(fs_info);
6024 * This is a copy of file_update_time. We need this so we can return error on
6025 * ENOSPC for updating the inode in the case of file write and mmap writes.
6027 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6030 struct btrfs_root *root = BTRFS_I(inode)->root;
6031 bool dirty = flags & ~S_VERSION;
6033 if (btrfs_root_readonly(root))
6036 if (flags & S_VERSION)
6037 dirty |= inode_maybe_inc_iversion(inode, dirty);
6038 if (flags & S_CTIME)
6039 inode->i_ctime = *now;
6040 if (flags & S_MTIME)
6041 inode->i_mtime = *now;
6042 if (flags & S_ATIME)
6043 inode->i_atime = *now;
6044 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6048 * helper to find a free sequence number in a given directory. This current
6049 * code is very simple, later versions will do smarter things in the btree
6051 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6055 if (dir->index_cnt == (u64)-1) {
6056 ret = btrfs_inode_delayed_dir_index_count(dir);
6058 ret = btrfs_set_inode_index_count(dir);
6064 *index = dir->index_cnt;
6070 static int btrfs_insert_inode_locked(struct inode *inode)
6072 struct btrfs_iget_args args;
6074 args.ino = BTRFS_I(inode)->location.objectid;
6075 args.root = BTRFS_I(inode)->root;
6077 return insert_inode_locked4(inode,
6078 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6079 btrfs_find_actor, &args);
6082 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6083 unsigned int *trans_num_items)
6085 struct inode *dir = args->dir;
6086 struct inode *inode = args->inode;
6089 if (!args->orphan) {
6090 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6096 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6098 fscrypt_free_filename(&args->fname);
6102 /* 1 to add inode item */
6103 *trans_num_items = 1;
6104 /* 1 to add compression property */
6105 if (BTRFS_I(dir)->prop_compress)
6106 (*trans_num_items)++;
6107 /* 1 to add default ACL xattr */
6108 if (args->default_acl)
6109 (*trans_num_items)++;
6110 /* 1 to add access ACL xattr */
6112 (*trans_num_items)++;
6113 #ifdef CONFIG_SECURITY
6114 /* 1 to add LSM xattr */
6115 if (dir->i_security)
6116 (*trans_num_items)++;
6119 /* 1 to add orphan item */
6120 (*trans_num_items)++;
6124 * 1 to add dir index
6125 * 1 to update parent inode item
6127 * No need for 1 unit for the inode ref item because it is
6128 * inserted in a batch together with the inode item at
6129 * btrfs_create_new_inode().
6131 *trans_num_items += 3;
6136 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6138 posix_acl_release(args->acl);
6139 posix_acl_release(args->default_acl);
6140 fscrypt_free_filename(&args->fname);
6144 * Inherit flags from the parent inode.
6146 * Currently only the compression flags and the cow flags are inherited.
6148 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6154 if (flags & BTRFS_INODE_NOCOMPRESS) {
6155 inode->flags &= ~BTRFS_INODE_COMPRESS;
6156 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6157 } else if (flags & BTRFS_INODE_COMPRESS) {
6158 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6159 inode->flags |= BTRFS_INODE_COMPRESS;
6162 if (flags & BTRFS_INODE_NODATACOW) {
6163 inode->flags |= BTRFS_INODE_NODATACOW;
6164 if (S_ISREG(inode->vfs_inode.i_mode))
6165 inode->flags |= BTRFS_INODE_NODATASUM;
6168 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6171 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6172 struct btrfs_new_inode_args *args)
6174 struct inode *dir = args->dir;
6175 struct inode *inode = args->inode;
6176 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6177 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6178 struct btrfs_root *root;
6179 struct btrfs_inode_item *inode_item;
6180 struct btrfs_key *location;
6181 struct btrfs_path *path;
6183 struct btrfs_inode_ref *ref;
6184 struct btrfs_key key[2];
6186 struct btrfs_item_batch batch;
6190 path = btrfs_alloc_path();
6195 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6196 root = BTRFS_I(inode)->root;
6198 ret = btrfs_get_free_objectid(root, &objectid);
6201 inode->i_ino = objectid;
6205 * O_TMPFILE, set link count to 0, so that after this point, we
6206 * fill in an inode item with the correct link count.
6208 set_nlink(inode, 0);
6210 trace_btrfs_inode_request(dir);
6212 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6216 /* index_cnt is ignored for everything but a dir. */
6217 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6218 BTRFS_I(inode)->generation = trans->transid;
6219 inode->i_generation = BTRFS_I(inode)->generation;
6222 * Subvolumes don't inherit flags from their parent directory.
6223 * Originally this was probably by accident, but we probably can't
6224 * change it now without compatibility issues.
6227 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6229 if (S_ISREG(inode->i_mode)) {
6230 if (btrfs_test_opt(fs_info, NODATASUM))
6231 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6232 if (btrfs_test_opt(fs_info, NODATACOW))
6233 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6234 BTRFS_INODE_NODATASUM;
6237 location = &BTRFS_I(inode)->location;
6238 location->objectid = objectid;
6239 location->offset = 0;
6240 location->type = BTRFS_INODE_ITEM_KEY;
6242 ret = btrfs_insert_inode_locked(inode);
6245 BTRFS_I(dir)->index_cnt--;
6250 * We could have gotten an inode number from somebody who was fsynced
6251 * and then removed in this same transaction, so let's just set full
6252 * sync since it will be a full sync anyway and this will blow away the
6253 * old info in the log.
6255 btrfs_set_inode_full_sync(BTRFS_I(inode));
6257 key[0].objectid = objectid;
6258 key[0].type = BTRFS_INODE_ITEM_KEY;
6261 sizes[0] = sizeof(struct btrfs_inode_item);
6263 if (!args->orphan) {
6265 * Start new inodes with an inode_ref. This is slightly more
6266 * efficient for small numbers of hard links since they will
6267 * be packed into one item. Extended refs will kick in if we
6268 * add more hard links than can fit in the ref item.
6270 key[1].objectid = objectid;
6271 key[1].type = BTRFS_INODE_REF_KEY;
6273 key[1].offset = objectid;
6274 sizes[1] = 2 + sizeof(*ref);
6276 key[1].offset = btrfs_ino(BTRFS_I(dir));
6277 sizes[1] = name->len + sizeof(*ref);
6281 batch.keys = &key[0];
6282 batch.data_sizes = &sizes[0];
6283 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6284 batch.nr = args->orphan ? 1 : 2;
6285 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6287 btrfs_abort_transaction(trans, ret);
6291 inode->i_mtime = current_time(inode);
6292 inode->i_atime = inode->i_mtime;
6293 inode->i_ctime = inode->i_mtime;
6294 BTRFS_I(inode)->i_otime = inode->i_mtime;
6297 * We're going to fill the inode item now, so at this point the inode
6298 * must be fully initialized.
6301 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6302 struct btrfs_inode_item);
6303 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6304 sizeof(*inode_item));
6305 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6307 if (!args->orphan) {
6308 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6309 struct btrfs_inode_ref);
6310 ptr = (unsigned long)(ref + 1);
6312 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6313 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6314 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6316 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6318 btrfs_set_inode_ref_index(path->nodes[0], ref,
6319 BTRFS_I(inode)->dir_index);
6320 write_extent_buffer(path->nodes[0], name->name, ptr,
6325 btrfs_mark_buffer_dirty(path->nodes[0]);
6327 * We don't need the path anymore, plus inheriting properties, adding
6328 * ACLs, security xattrs, orphan item or adding the link, will result in
6329 * allocating yet another path. So just free our path.
6331 btrfs_free_path(path);
6335 struct inode *parent;
6338 * Subvolumes inherit properties from their parent subvolume,
6339 * not the directory they were created in.
6341 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6342 BTRFS_I(dir)->root);
6343 if (IS_ERR(parent)) {
6344 ret = PTR_ERR(parent);
6346 ret = btrfs_inode_inherit_props(trans, inode, parent);
6350 ret = btrfs_inode_inherit_props(trans, inode, dir);
6354 "error inheriting props for ino %llu (root %llu): %d",
6355 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6360 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6363 if (!args->subvol) {
6364 ret = btrfs_init_inode_security(trans, args);
6366 btrfs_abort_transaction(trans, ret);
6371 inode_tree_add(BTRFS_I(inode));
6373 trace_btrfs_inode_new(inode);
6374 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6376 btrfs_update_root_times(trans, root);
6379 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6381 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6382 0, BTRFS_I(inode)->dir_index);
6385 btrfs_abort_transaction(trans, ret);
6393 * discard_new_inode() calls iput(), but the caller owns the reference
6397 discard_new_inode(inode);
6399 btrfs_free_path(path);
6404 * utility function to add 'inode' into 'parent_inode' with
6405 * a give name and a given sequence number.
6406 * if 'add_backref' is true, also insert a backref from the
6407 * inode to the parent directory.
6409 int btrfs_add_link(struct btrfs_trans_handle *trans,
6410 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6411 const struct fscrypt_str *name, int add_backref, u64 index)
6414 struct btrfs_key key;
6415 struct btrfs_root *root = parent_inode->root;
6416 u64 ino = btrfs_ino(inode);
6417 u64 parent_ino = btrfs_ino(parent_inode);
6419 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6420 memcpy(&key, &inode->root->root_key, sizeof(key));
6423 key.type = BTRFS_INODE_ITEM_KEY;
6427 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6428 ret = btrfs_add_root_ref(trans, key.objectid,
6429 root->root_key.objectid, parent_ino,
6431 } else if (add_backref) {
6432 ret = btrfs_insert_inode_ref(trans, root, name,
6433 ino, parent_ino, index);
6436 /* Nothing to clean up yet */
6440 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6441 btrfs_inode_type(&inode->vfs_inode), index);
6442 if (ret == -EEXIST || ret == -EOVERFLOW)
6445 btrfs_abort_transaction(trans, ret);
6449 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6451 inode_inc_iversion(&parent_inode->vfs_inode);
6453 * If we are replaying a log tree, we do not want to update the mtime
6454 * and ctime of the parent directory with the current time, since the
6455 * log replay procedure is responsible for setting them to their correct
6456 * values (the ones it had when the fsync was done).
6458 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6459 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6461 parent_inode->vfs_inode.i_mtime = now;
6462 parent_inode->vfs_inode.i_ctime = now;
6464 ret = btrfs_update_inode(trans, root, parent_inode);
6466 btrfs_abort_transaction(trans, ret);
6470 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6473 err = btrfs_del_root_ref(trans, key.objectid,
6474 root->root_key.objectid, parent_ino,
6475 &local_index, name);
6477 btrfs_abort_transaction(trans, err);
6478 } else if (add_backref) {
6482 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6485 btrfs_abort_transaction(trans, err);
6488 /* Return the original error code */
6492 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6493 struct inode *inode)
6495 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6496 struct btrfs_root *root = BTRFS_I(dir)->root;
6497 struct btrfs_new_inode_args new_inode_args = {
6502 unsigned int trans_num_items;
6503 struct btrfs_trans_handle *trans;
6506 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6510 trans = btrfs_start_transaction(root, trans_num_items);
6511 if (IS_ERR(trans)) {
6512 err = PTR_ERR(trans);
6513 goto out_new_inode_args;
6516 err = btrfs_create_new_inode(trans, &new_inode_args);
6518 d_instantiate_new(dentry, inode);
6520 btrfs_end_transaction(trans);
6521 btrfs_btree_balance_dirty(fs_info);
6523 btrfs_new_inode_args_destroy(&new_inode_args);
6530 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6531 struct dentry *dentry, umode_t mode, dev_t rdev)
6533 struct inode *inode;
6535 inode = new_inode(dir->i_sb);
6538 inode_init_owner(idmap, inode, dir, mode);
6539 inode->i_op = &btrfs_special_inode_operations;
6540 init_special_inode(inode, inode->i_mode, rdev);
6541 return btrfs_create_common(dir, dentry, inode);
6544 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6545 struct dentry *dentry, umode_t mode, bool excl)
6547 struct inode *inode;
6549 inode = new_inode(dir->i_sb);
6552 inode_init_owner(idmap, inode, dir, mode);
6553 inode->i_fop = &btrfs_file_operations;
6554 inode->i_op = &btrfs_file_inode_operations;
6555 inode->i_mapping->a_ops = &btrfs_aops;
6556 return btrfs_create_common(dir, dentry, inode);
6559 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6560 struct dentry *dentry)
6562 struct btrfs_trans_handle *trans = NULL;
6563 struct btrfs_root *root = BTRFS_I(dir)->root;
6564 struct inode *inode = d_inode(old_dentry);
6565 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6566 struct fscrypt_name fname;
6571 /* do not allow sys_link's with other subvols of the same device */
6572 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6575 if (inode->i_nlink >= BTRFS_LINK_MAX)
6578 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6582 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6587 * 2 items for inode and inode ref
6588 * 2 items for dir items
6589 * 1 item for parent inode
6590 * 1 item for orphan item deletion if O_TMPFILE
6592 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6593 if (IS_ERR(trans)) {
6594 err = PTR_ERR(trans);
6599 /* There are several dir indexes for this inode, clear the cache. */
6600 BTRFS_I(inode)->dir_index = 0ULL;
6602 inode_inc_iversion(inode);
6603 inode->i_ctime = current_time(inode);
6605 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6607 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6608 &fname.disk_name, 1, index);
6613 struct dentry *parent = dentry->d_parent;
6615 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6618 if (inode->i_nlink == 1) {
6620 * If new hard link count is 1, it's a file created
6621 * with open(2) O_TMPFILE flag.
6623 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6627 d_instantiate(dentry, inode);
6628 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6632 fscrypt_free_filename(&fname);
6634 btrfs_end_transaction(trans);
6636 inode_dec_link_count(inode);
6639 btrfs_btree_balance_dirty(fs_info);
6643 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6644 struct dentry *dentry, umode_t mode)
6646 struct inode *inode;
6648 inode = new_inode(dir->i_sb);
6651 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6652 inode->i_op = &btrfs_dir_inode_operations;
6653 inode->i_fop = &btrfs_dir_file_operations;
6654 return btrfs_create_common(dir, dentry, inode);
6657 static noinline int uncompress_inline(struct btrfs_path *path,
6659 struct btrfs_file_extent_item *item)
6662 struct extent_buffer *leaf = path->nodes[0];
6665 unsigned long inline_size;
6669 compress_type = btrfs_file_extent_compression(leaf, item);
6670 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6671 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6672 tmp = kmalloc(inline_size, GFP_NOFS);
6675 ptr = btrfs_file_extent_inline_start(item);
6677 read_extent_buffer(leaf, tmp, ptr, inline_size);
6679 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6680 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6683 * decompression code contains a memset to fill in any space between the end
6684 * of the uncompressed data and the end of max_size in case the decompressed
6685 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6686 * the end of an inline extent and the beginning of the next block, so we
6687 * cover that region here.
6690 if (max_size < PAGE_SIZE)
6691 memzero_page(page, max_size, PAGE_SIZE - max_size);
6696 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6699 struct btrfs_file_extent_item *fi;
6703 if (!page || PageUptodate(page))
6706 ASSERT(page_offset(page) == 0);
6708 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6709 struct btrfs_file_extent_item);
6710 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6711 return uncompress_inline(path, page, fi);
6713 copy_size = min_t(u64, PAGE_SIZE,
6714 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6715 kaddr = kmap_local_page(page);
6716 read_extent_buffer(path->nodes[0], kaddr,
6717 btrfs_file_extent_inline_start(fi), copy_size);
6718 kunmap_local(kaddr);
6719 if (copy_size < PAGE_SIZE)
6720 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6725 * Lookup the first extent overlapping a range in a file.
6727 * @inode: file to search in
6728 * @page: page to read extent data into if the extent is inline
6729 * @pg_offset: offset into @page to copy to
6730 * @start: file offset
6731 * @len: length of range starting at @start
6733 * Return the first &struct extent_map which overlaps the given range, reading
6734 * it from the B-tree and caching it if necessary. Note that there may be more
6735 * extents which overlap the given range after the returned extent_map.
6737 * If @page is not NULL and the extent is inline, this also reads the extent
6738 * data directly into the page and marks the extent up to date in the io_tree.
6740 * Return: ERR_PTR on error, non-NULL extent_map on success.
6742 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6743 struct page *page, size_t pg_offset,
6746 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6748 u64 extent_start = 0;
6750 u64 objectid = btrfs_ino(inode);
6751 int extent_type = -1;
6752 struct btrfs_path *path = NULL;
6753 struct btrfs_root *root = inode->root;
6754 struct btrfs_file_extent_item *item;
6755 struct extent_buffer *leaf;
6756 struct btrfs_key found_key;
6757 struct extent_map *em = NULL;
6758 struct extent_map_tree *em_tree = &inode->extent_tree;
6760 read_lock(&em_tree->lock);
6761 em = lookup_extent_mapping(em_tree, start, len);
6762 read_unlock(&em_tree->lock);
6765 if (em->start > start || em->start + em->len <= start)
6766 free_extent_map(em);
6767 else if (em->block_start == EXTENT_MAP_INLINE && page)
6768 free_extent_map(em);
6772 em = alloc_extent_map();
6777 em->start = EXTENT_MAP_HOLE;
6778 em->orig_start = EXTENT_MAP_HOLE;
6780 em->block_len = (u64)-1;
6782 path = btrfs_alloc_path();
6788 /* Chances are we'll be called again, so go ahead and do readahead */
6789 path->reada = READA_FORWARD;
6792 * The same explanation in load_free_space_cache applies here as well,
6793 * we only read when we're loading the free space cache, and at that
6794 * point the commit_root has everything we need.
6796 if (btrfs_is_free_space_inode(inode)) {
6797 path->search_commit_root = 1;
6798 path->skip_locking = 1;
6801 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6804 } else if (ret > 0) {
6805 if (path->slots[0] == 0)
6811 leaf = path->nodes[0];
6812 item = btrfs_item_ptr(leaf, path->slots[0],
6813 struct btrfs_file_extent_item);
6814 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6815 if (found_key.objectid != objectid ||
6816 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6818 * If we backup past the first extent we want to move forward
6819 * and see if there is an extent in front of us, otherwise we'll
6820 * say there is a hole for our whole search range which can
6827 extent_type = btrfs_file_extent_type(leaf, item);
6828 extent_start = found_key.offset;
6829 extent_end = btrfs_file_extent_end(path);
6830 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6831 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6832 /* Only regular file could have regular/prealloc extent */
6833 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6836 "regular/prealloc extent found for non-regular inode %llu",
6840 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6842 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6843 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6848 if (start >= extent_end) {
6850 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6851 ret = btrfs_next_leaf(root, path);
6857 leaf = path->nodes[0];
6859 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6860 if (found_key.objectid != objectid ||
6861 found_key.type != BTRFS_EXTENT_DATA_KEY)
6863 if (start + len <= found_key.offset)
6865 if (start > found_key.offset)
6868 /* New extent overlaps with existing one */
6870 em->orig_start = start;
6871 em->len = found_key.offset - start;
6872 em->block_start = EXTENT_MAP_HOLE;
6876 btrfs_extent_item_to_extent_map(inode, path, item, em);
6878 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6879 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6881 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6883 * Inline extent can only exist at file offset 0. This is
6884 * ensured by tree-checker and inline extent creation path.
6885 * Thus all members representing file offsets should be zero.
6887 ASSERT(pg_offset == 0);
6888 ASSERT(extent_start == 0);
6889 ASSERT(em->start == 0);
6892 * btrfs_extent_item_to_extent_map() should have properly
6893 * initialized em members already.
6895 * Other members are not utilized for inline extents.
6897 ASSERT(em->block_start == EXTENT_MAP_INLINE);
6898 ASSERT(em->len == fs_info->sectorsize);
6900 ret = read_inline_extent(inode, path, page);
6907 em->orig_start = start;
6909 em->block_start = EXTENT_MAP_HOLE;
6912 btrfs_release_path(path);
6913 if (em->start > start || extent_map_end(em) <= start) {
6915 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6916 em->start, em->len, start, len);
6921 write_lock(&em_tree->lock);
6922 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6923 write_unlock(&em_tree->lock);
6925 btrfs_free_path(path);
6927 trace_btrfs_get_extent(root, inode, em);
6930 free_extent_map(em);
6931 return ERR_PTR(ret);
6936 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
6937 struct btrfs_dio_data *dio_data,
6940 const u64 orig_start,
6941 const u64 block_start,
6942 const u64 block_len,
6943 const u64 orig_block_len,
6944 const u64 ram_bytes,
6947 struct extent_map *em = NULL;
6948 struct btrfs_ordered_extent *ordered;
6950 if (type != BTRFS_ORDERED_NOCOW) {
6951 em = create_io_em(inode, start, len, orig_start, block_start,
6952 block_len, orig_block_len, ram_bytes,
6953 BTRFS_COMPRESS_NONE, /* compress_type */
6958 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
6959 block_start, block_len, 0,
6961 (1 << BTRFS_ORDERED_DIRECT),
6962 BTRFS_COMPRESS_NONE);
6963 if (IS_ERR(ordered)) {
6965 free_extent_map(em);
6966 btrfs_drop_extent_map_range(inode, start,
6967 start + len - 1, false);
6969 em = ERR_CAST(ordered);
6971 ASSERT(!dio_data->ordered);
6972 dio_data->ordered = ordered;
6979 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
6980 struct btrfs_dio_data *dio_data,
6983 struct btrfs_root *root = inode->root;
6984 struct btrfs_fs_info *fs_info = root->fs_info;
6985 struct extent_map *em;
6986 struct btrfs_key ins;
6990 alloc_hint = get_extent_allocation_hint(inode, start, len);
6991 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
6992 0, alloc_hint, &ins, 1, 1);
6994 return ERR_PTR(ret);
6996 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
6997 ins.objectid, ins.offset, ins.offset,
6998 ins.offset, BTRFS_ORDERED_REGULAR);
6999 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7001 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7007 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7009 struct btrfs_block_group *block_group;
7010 bool readonly = false;
7012 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7013 if (!block_group || block_group->ro)
7016 btrfs_put_block_group(block_group);
7021 * Check if we can do nocow write into the range [@offset, @offset + @len)
7023 * @offset: File offset
7024 * @len: The length to write, will be updated to the nocow writeable
7026 * @orig_start: (optional) Return the original file offset of the file extent
7027 * @orig_len: (optional) Return the original on-disk length of the file extent
7028 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7029 * @strict: if true, omit optimizations that might force us into unnecessary
7030 * cow. e.g., don't trust generation number.
7033 * >0 and update @len if we can do nocow write
7034 * 0 if we can't do nocow write
7035 * <0 if error happened
7037 * NOTE: This only checks the file extents, caller is responsible to wait for
7038 * any ordered extents.
7040 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7041 u64 *orig_start, u64 *orig_block_len,
7042 u64 *ram_bytes, bool nowait, bool strict)
7044 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7045 struct can_nocow_file_extent_args nocow_args = { 0 };
7046 struct btrfs_path *path;
7048 struct extent_buffer *leaf;
7049 struct btrfs_root *root = BTRFS_I(inode)->root;
7050 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7051 struct btrfs_file_extent_item *fi;
7052 struct btrfs_key key;
7055 path = btrfs_alloc_path();
7058 path->nowait = nowait;
7060 ret = btrfs_lookup_file_extent(NULL, root, path,
7061 btrfs_ino(BTRFS_I(inode)), offset, 0);
7066 if (path->slots[0] == 0) {
7067 /* can't find the item, must cow */
7074 leaf = path->nodes[0];
7075 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7076 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7077 key.type != BTRFS_EXTENT_DATA_KEY) {
7078 /* not our file or wrong item type, must cow */
7082 if (key.offset > offset) {
7083 /* Wrong offset, must cow */
7087 if (btrfs_file_extent_end(path) <= offset)
7090 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7091 found_type = btrfs_file_extent_type(leaf, fi);
7093 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7095 nocow_args.start = offset;
7096 nocow_args.end = offset + *len - 1;
7097 nocow_args.strict = strict;
7098 nocow_args.free_path = true;
7100 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7101 /* can_nocow_file_extent() has freed the path. */
7105 /* Treat errors as not being able to NOCOW. */
7111 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7114 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7115 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7118 range_end = round_up(offset + nocow_args.num_bytes,
7119 root->fs_info->sectorsize) - 1;
7120 ret = test_range_bit(io_tree, offset, range_end,
7121 EXTENT_DELALLOC, 0, NULL);
7129 *orig_start = key.offset - nocow_args.extent_offset;
7131 *orig_block_len = nocow_args.disk_num_bytes;
7133 *len = nocow_args.num_bytes;
7136 btrfs_free_path(path);
7140 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7141 struct extent_state **cached_state,
7142 unsigned int iomap_flags)
7144 const bool writing = (iomap_flags & IOMAP_WRITE);
7145 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7146 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7147 struct btrfs_ordered_extent *ordered;
7152 if (!try_lock_extent(io_tree, lockstart, lockend,
7156 lock_extent(io_tree, lockstart, lockend, cached_state);
7159 * We're concerned with the entire range that we're going to be
7160 * doing DIO to, so we need to make sure there's no ordered
7161 * extents in this range.
7163 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7164 lockend - lockstart + 1);
7167 * We need to make sure there are no buffered pages in this
7168 * range either, we could have raced between the invalidate in
7169 * generic_file_direct_write and locking the extent. The
7170 * invalidate needs to happen so that reads after a write do not
7174 (!writing || !filemap_range_has_page(inode->i_mapping,
7175 lockstart, lockend)))
7178 unlock_extent(io_tree, lockstart, lockend, cached_state);
7182 btrfs_put_ordered_extent(ordered);
7187 * If we are doing a DIO read and the ordered extent we
7188 * found is for a buffered write, we can not wait for it
7189 * to complete and retry, because if we do so we can
7190 * deadlock with concurrent buffered writes on page
7191 * locks. This happens only if our DIO read covers more
7192 * than one extent map, if at this point has already
7193 * created an ordered extent for a previous extent map
7194 * and locked its range in the inode's io tree, and a
7195 * concurrent write against that previous extent map's
7196 * range and this range started (we unlock the ranges
7197 * in the io tree only when the bios complete and
7198 * buffered writes always lock pages before attempting
7199 * to lock range in the io tree).
7202 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7203 btrfs_start_ordered_extent(ordered);
7205 ret = nowait ? -EAGAIN : -ENOTBLK;
7206 btrfs_put_ordered_extent(ordered);
7209 * We could trigger writeback for this range (and wait
7210 * for it to complete) and then invalidate the pages for
7211 * this range (through invalidate_inode_pages2_range()),
7212 * but that can lead us to a deadlock with a concurrent
7213 * call to readahead (a buffered read or a defrag call
7214 * triggered a readahead) on a page lock due to an
7215 * ordered dio extent we created before but did not have
7216 * yet a corresponding bio submitted (whence it can not
7217 * complete), which makes readahead wait for that
7218 * ordered extent to complete while holding a lock on
7221 ret = nowait ? -EAGAIN : -ENOTBLK;
7233 /* The callers of this must take lock_extent() */
7234 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7235 u64 len, u64 orig_start, u64 block_start,
7236 u64 block_len, u64 orig_block_len,
7237 u64 ram_bytes, int compress_type,
7240 struct extent_map *em;
7243 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7244 type == BTRFS_ORDERED_COMPRESSED ||
7245 type == BTRFS_ORDERED_NOCOW ||
7246 type == BTRFS_ORDERED_REGULAR);
7248 em = alloc_extent_map();
7250 return ERR_PTR(-ENOMEM);
7253 em->orig_start = orig_start;
7255 em->block_len = block_len;
7256 em->block_start = block_start;
7257 em->orig_block_len = orig_block_len;
7258 em->ram_bytes = ram_bytes;
7259 em->generation = -1;
7260 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7261 if (type == BTRFS_ORDERED_PREALLOC) {
7262 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7263 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7264 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7265 em->compress_type = compress_type;
7268 ret = btrfs_replace_extent_map_range(inode, em, true);
7270 free_extent_map(em);
7271 return ERR_PTR(ret);
7274 /* em got 2 refs now, callers needs to do free_extent_map once. */
7279 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7280 struct inode *inode,
7281 struct btrfs_dio_data *dio_data,
7282 u64 start, u64 *lenp,
7283 unsigned int iomap_flags)
7285 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7286 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7287 struct extent_map *em = *map;
7289 u64 block_start, orig_start, orig_block_len, ram_bytes;
7290 struct btrfs_block_group *bg;
7291 bool can_nocow = false;
7292 bool space_reserved = false;
7298 * We don't allocate a new extent in the following cases
7300 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7302 * 2) The extent is marked as PREALLOC. We're good to go here and can
7303 * just use the extent.
7306 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7307 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7308 em->block_start != EXTENT_MAP_HOLE)) {
7309 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7310 type = BTRFS_ORDERED_PREALLOC;
7312 type = BTRFS_ORDERED_NOCOW;
7313 len = min(len, em->len - (start - em->start));
7314 block_start = em->block_start + (start - em->start);
7316 if (can_nocow_extent(inode, start, &len, &orig_start,
7317 &orig_block_len, &ram_bytes, false, false) == 1) {
7318 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7326 struct extent_map *em2;
7328 /* We can NOCOW, so only need to reserve metadata space. */
7329 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7332 /* Our caller expects us to free the input extent map. */
7333 free_extent_map(em);
7335 btrfs_dec_nocow_writers(bg);
7336 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7340 space_reserved = true;
7342 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7343 orig_start, block_start,
7344 len, orig_block_len,
7346 btrfs_dec_nocow_writers(bg);
7347 if (type == BTRFS_ORDERED_PREALLOC) {
7348 free_extent_map(em);
7358 dio_data->nocow_done = true;
7360 /* Our caller expects us to free the input extent map. */
7361 free_extent_map(em);
7370 * If we could not allocate data space before locking the file
7371 * range and we can't do a NOCOW write, then we have to fail.
7373 if (!dio_data->data_space_reserved) {
7379 * We have to COW and we have already reserved data space before,
7380 * so now we reserve only metadata.
7382 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7386 space_reserved = true;
7388 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7394 len = min(len, em->len - (start - em->start));
7396 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7397 prev_len - len, true);
7401 * We have created our ordered extent, so we can now release our reservation
7402 * for an outstanding extent.
7404 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7407 * Need to update the i_size under the extent lock so buffered
7408 * readers will get the updated i_size when we unlock.
7410 if (start + len > i_size_read(inode))
7411 i_size_write(inode, start + len);
7413 if (ret && space_reserved) {
7414 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7415 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7421 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7422 loff_t length, unsigned int flags, struct iomap *iomap,
7423 struct iomap *srcmap)
7425 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7426 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7427 struct extent_map *em;
7428 struct extent_state *cached_state = NULL;
7429 struct btrfs_dio_data *dio_data = iter->private;
7430 u64 lockstart, lockend;
7431 const bool write = !!(flags & IOMAP_WRITE);
7434 const u64 data_alloc_len = length;
7435 bool unlock_extents = false;
7438 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7439 * we're NOWAIT we may submit a bio for a partial range and return
7440 * EIOCBQUEUED, which would result in an errant short read.
7442 * The best way to handle this would be to allow for partial completions
7443 * of iocb's, so we could submit the partial bio, return and fault in
7444 * the rest of the pages, and then submit the io for the rest of the
7445 * range. However we don't have that currently, so simply return
7446 * -EAGAIN at this point so that the normal path is used.
7448 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7452 * Cap the size of reads to that usually seen in buffered I/O as we need
7453 * to allocate a contiguous array for the checksums.
7456 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7459 lockend = start + len - 1;
7462 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7463 * enough if we've written compressed pages to this area, so we need to
7464 * flush the dirty pages again to make absolutely sure that any
7465 * outstanding dirty pages are on disk - the first flush only starts
7466 * compression on the data, while keeping the pages locked, so by the
7467 * time the second flush returns we know bios for the compressed pages
7468 * were submitted and finished, and the pages no longer under writeback.
7470 * If we have a NOWAIT request and we have any pages in the range that
7471 * are locked, likely due to compression still in progress, we don't want
7472 * to block on page locks. We also don't want to block on pages marked as
7473 * dirty or under writeback (same as for the non-compression case).
7474 * iomap_dio_rw() did the same check, but after that and before we got
7475 * here, mmap'ed writes may have happened or buffered reads started
7476 * (readpage() and readahead(), which lock pages), as we haven't locked
7477 * the file range yet.
7479 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7480 &BTRFS_I(inode)->runtime_flags)) {
7481 if (flags & IOMAP_NOWAIT) {
7482 if (filemap_range_needs_writeback(inode->i_mapping,
7483 lockstart, lockend))
7486 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7487 start + length - 1);
7493 memset(dio_data, 0, sizeof(*dio_data));
7496 * We always try to allocate data space and must do it before locking
7497 * the file range, to avoid deadlocks with concurrent writes to the same
7498 * range if the range has several extents and the writes don't expand the
7499 * current i_size (the inode lock is taken in shared mode). If we fail to
7500 * allocate data space here we continue and later, after locking the
7501 * file range, we fail with ENOSPC only if we figure out we can not do a
7504 if (write && !(flags & IOMAP_NOWAIT)) {
7505 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7506 &dio_data->data_reserved,
7507 start, data_alloc_len, false);
7509 dio_data->data_space_reserved = true;
7510 else if (ret && !(BTRFS_I(inode)->flags &
7511 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7516 * If this errors out it's because we couldn't invalidate pagecache for
7517 * this range and we need to fallback to buffered IO, or we are doing a
7518 * NOWAIT read/write and we need to block.
7520 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7524 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7531 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7532 * io. INLINE is special, and we could probably kludge it in here, but
7533 * it's still buffered so for safety lets just fall back to the generic
7536 * For COMPRESSED we _have_ to read the entire extent in so we can
7537 * decompress it, so there will be buffering required no matter what we
7538 * do, so go ahead and fallback to buffered.
7540 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7541 * to buffered IO. Don't blame me, this is the price we pay for using
7544 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7545 em->block_start == EXTENT_MAP_INLINE) {
7546 free_extent_map(em);
7548 * If we are in a NOWAIT context, return -EAGAIN in order to
7549 * fallback to buffered IO. This is not only because we can
7550 * block with buffered IO (no support for NOWAIT semantics at
7551 * the moment) but also to avoid returning short reads to user
7552 * space - this happens if we were able to read some data from
7553 * previous non-compressed extents and then when we fallback to
7554 * buffered IO, at btrfs_file_read_iter() by calling
7555 * filemap_read(), we fail to fault in pages for the read buffer,
7556 * in which case filemap_read() returns a short read (the number
7557 * of bytes previously read is > 0, so it does not return -EFAULT).
7559 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7563 len = min(len, em->len - (start - em->start));
7566 * If we have a NOWAIT request and the range contains multiple extents
7567 * (or a mix of extents and holes), then we return -EAGAIN to make the
7568 * caller fallback to a context where it can do a blocking (without
7569 * NOWAIT) request. This way we avoid doing partial IO and returning
7570 * success to the caller, which is not optimal for writes and for reads
7571 * it can result in unexpected behaviour for an application.
7573 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7574 * iomap_dio_rw(), we can end up returning less data then what the caller
7575 * asked for, resulting in an unexpected, and incorrect, short read.
7576 * That is, the caller asked to read N bytes and we return less than that,
7577 * which is wrong unless we are crossing EOF. This happens if we get a
7578 * page fault error when trying to fault in pages for the buffer that is
7579 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7580 * have previously submitted bios for other extents in the range, in
7581 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7582 * those bios have completed by the time we get the page fault error,
7583 * which we return back to our caller - we should only return EIOCBQUEUED
7584 * after we have submitted bios for all the extents in the range.
7586 if ((flags & IOMAP_NOWAIT) && len < length) {
7587 free_extent_map(em);
7593 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7594 start, &len, flags);
7597 unlock_extents = true;
7598 /* Recalc len in case the new em is smaller than requested */
7599 len = min(len, em->len - (start - em->start));
7600 if (dio_data->data_space_reserved) {
7602 u64 release_len = 0;
7604 if (dio_data->nocow_done) {
7605 release_offset = start;
7606 release_len = data_alloc_len;
7607 } else if (len < data_alloc_len) {
7608 release_offset = start + len;
7609 release_len = data_alloc_len - len;
7612 if (release_len > 0)
7613 btrfs_free_reserved_data_space(BTRFS_I(inode),
7614 dio_data->data_reserved,
7620 * We need to unlock only the end area that we aren't using.
7621 * The rest is going to be unlocked by the endio routine.
7623 lockstart = start + len;
7624 if (lockstart < lockend)
7625 unlock_extents = true;
7629 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7632 free_extent_state(cached_state);
7635 * Translate extent map information to iomap.
7636 * We trim the extents (and move the addr) even though iomap code does
7637 * that, since we have locked only the parts we are performing I/O in.
7639 if ((em->block_start == EXTENT_MAP_HOLE) ||
7640 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7641 iomap->addr = IOMAP_NULL_ADDR;
7642 iomap->type = IOMAP_HOLE;
7644 iomap->addr = em->block_start + (start - em->start);
7645 iomap->type = IOMAP_MAPPED;
7647 iomap->offset = start;
7648 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7649 iomap->length = len;
7650 free_extent_map(em);
7655 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7658 if (dio_data->data_space_reserved) {
7659 btrfs_free_reserved_data_space(BTRFS_I(inode),
7660 dio_data->data_reserved,
7661 start, data_alloc_len);
7662 extent_changeset_free(dio_data->data_reserved);
7668 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7669 ssize_t written, unsigned int flags, struct iomap *iomap)
7671 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7672 struct btrfs_dio_data *dio_data = iter->private;
7673 size_t submitted = dio_data->submitted;
7674 const bool write = !!(flags & IOMAP_WRITE);
7677 if (!write && (iomap->type == IOMAP_HOLE)) {
7678 /* If reading from a hole, unlock and return */
7679 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7684 if (submitted < length) {
7686 length -= submitted;
7688 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7689 pos, length, false);
7691 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7692 pos + length - 1, NULL);
7696 btrfs_put_ordered_extent(dio_data->ordered);
7697 dio_data->ordered = NULL;
7701 extent_changeset_free(dio_data->data_reserved);
7705 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7707 struct btrfs_dio_private *dip =
7708 container_of(bbio, struct btrfs_dio_private, bbio);
7709 struct btrfs_inode *inode = bbio->inode;
7710 struct bio *bio = &bbio->bio;
7712 if (bio->bi_status) {
7713 btrfs_warn(inode->root->fs_info,
7714 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7715 btrfs_ino(inode), bio->bi_opf,
7716 dip->file_offset, dip->bytes, bio->bi_status);
7719 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7720 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7721 dip->file_offset, dip->bytes,
7724 unlock_extent(&inode->io_tree, dip->file_offset,
7725 dip->file_offset + dip->bytes - 1, NULL);
7728 bbio->bio.bi_private = bbio->private;
7729 iomap_dio_bio_end_io(bio);
7732 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7735 struct btrfs_bio *bbio = btrfs_bio(bio);
7736 struct btrfs_dio_private *dip =
7737 container_of(bbio, struct btrfs_dio_private, bbio);
7738 struct btrfs_dio_data *dio_data = iter->private;
7740 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7741 btrfs_dio_end_io, bio->bi_private);
7742 bbio->inode = BTRFS_I(iter->inode);
7743 bbio->file_offset = file_offset;
7745 dip->file_offset = file_offset;
7746 dip->bytes = bio->bi_iter.bi_size;
7748 dio_data->submitted += bio->bi_iter.bi_size;
7751 * Check if we are doing a partial write. If we are, we need to split
7752 * the ordered extent to match the submitted bio. Hang on to the
7753 * remaining unfinishable ordered_extent in dio_data so that it can be
7754 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7755 * remaining pages is blocked on the outstanding ordered extent.
7757 if (iter->flags & IOMAP_WRITE) {
7760 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7762 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7763 file_offset, dip->bytes,
7765 bio->bi_status = errno_to_blk_status(ret);
7766 iomap_dio_bio_end_io(bio);
7771 btrfs_submit_bio(bbio, 0);
7774 static const struct iomap_ops btrfs_dio_iomap_ops = {
7775 .iomap_begin = btrfs_dio_iomap_begin,
7776 .iomap_end = btrfs_dio_iomap_end,
7779 static const struct iomap_dio_ops btrfs_dio_ops = {
7780 .submit_io = btrfs_dio_submit_io,
7781 .bio_set = &btrfs_dio_bioset,
7784 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7786 struct btrfs_dio_data data = { 0 };
7788 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7789 IOMAP_DIO_PARTIAL, &data, done_before);
7792 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7795 struct btrfs_dio_data data = { 0 };
7797 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7798 IOMAP_DIO_PARTIAL, &data, done_before);
7801 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7806 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7811 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7812 * file range (0 to LLONG_MAX), but that is not enough if we have
7813 * compression enabled. The first filemap_fdatawrite_range() only kicks
7814 * in the compression of data (in an async thread) and will return
7815 * before the compression is done and writeback is started. A second
7816 * filemap_fdatawrite_range() is needed to wait for the compression to
7817 * complete and writeback to start. We also need to wait for ordered
7818 * extents to complete, because our fiemap implementation uses mainly
7819 * file extent items to list the extents, searching for extent maps
7820 * only for file ranges with holes or prealloc extents to figure out
7821 * if we have delalloc in those ranges.
7823 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7824 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7829 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7832 static int btrfs_writepages(struct address_space *mapping,
7833 struct writeback_control *wbc)
7835 return extent_writepages(mapping, wbc);
7838 static void btrfs_readahead(struct readahead_control *rac)
7840 extent_readahead(rac);
7844 * For release_folio() and invalidate_folio() we have a race window where
7845 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7846 * If we continue to release/invalidate the page, we could cause use-after-free
7847 * for subpage spinlock. So this function is to spin and wait for subpage
7850 static void wait_subpage_spinlock(struct page *page)
7852 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7853 struct btrfs_subpage *subpage;
7855 if (!btrfs_is_subpage(fs_info, page))
7858 ASSERT(PagePrivate(page) && page->private);
7859 subpage = (struct btrfs_subpage *)page->private;
7862 * This may look insane as we just acquire the spinlock and release it,
7863 * without doing anything. But we just want to make sure no one is
7864 * still holding the subpage spinlock.
7865 * And since the page is not dirty nor writeback, and we have page
7866 * locked, the only possible way to hold a spinlock is from the endio
7867 * function to clear page writeback.
7869 * Here we just acquire the spinlock so that all existing callers
7870 * should exit and we're safe to release/invalidate the page.
7872 spin_lock_irq(&subpage->lock);
7873 spin_unlock_irq(&subpage->lock);
7876 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7878 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
7881 wait_subpage_spinlock(&folio->page);
7882 clear_page_extent_mapped(&folio->page);
7887 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
7889 if (folio_test_writeback(folio) || folio_test_dirty(folio))
7891 return __btrfs_release_folio(folio, gfp_flags);
7894 #ifdef CONFIG_MIGRATION
7895 static int btrfs_migrate_folio(struct address_space *mapping,
7896 struct folio *dst, struct folio *src,
7897 enum migrate_mode mode)
7899 int ret = filemap_migrate_folio(mapping, dst, src, mode);
7901 if (ret != MIGRATEPAGE_SUCCESS)
7904 if (folio_test_ordered(src)) {
7905 folio_clear_ordered(src);
7906 folio_set_ordered(dst);
7909 return MIGRATEPAGE_SUCCESS;
7912 #define btrfs_migrate_folio NULL
7915 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
7918 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
7919 struct btrfs_fs_info *fs_info = inode->root->fs_info;
7920 struct extent_io_tree *tree = &inode->io_tree;
7921 struct extent_state *cached_state = NULL;
7922 u64 page_start = folio_pos(folio);
7923 u64 page_end = page_start + folio_size(folio) - 1;
7925 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
7928 * We have folio locked so no new ordered extent can be created on this
7929 * page, nor bio can be submitted for this folio.
7931 * But already submitted bio can still be finished on this folio.
7932 * Furthermore, endio function won't skip folio which has Ordered
7933 * (Private2) already cleared, so it's possible for endio and
7934 * invalidate_folio to do the same ordered extent accounting twice
7937 * So here we wait for any submitted bios to finish, so that we won't
7938 * do double ordered extent accounting on the same folio.
7940 folio_wait_writeback(folio);
7941 wait_subpage_spinlock(&folio->page);
7944 * For subpage case, we have call sites like
7945 * btrfs_punch_hole_lock_range() which passes range not aligned to
7947 * If the range doesn't cover the full folio, we don't need to and
7948 * shouldn't clear page extent mapped, as folio->private can still
7949 * record subpage dirty bits for other part of the range.
7951 * For cases that invalidate the full folio even the range doesn't
7952 * cover the full folio, like invalidating the last folio, we're
7953 * still safe to wait for ordered extent to finish.
7955 if (!(offset == 0 && length == folio_size(folio))) {
7956 btrfs_release_folio(folio, GFP_NOFS);
7960 if (!inode_evicting)
7961 lock_extent(tree, page_start, page_end, &cached_state);
7964 while (cur < page_end) {
7965 struct btrfs_ordered_extent *ordered;
7968 u32 extra_flags = 0;
7970 ordered = btrfs_lookup_first_ordered_range(inode, cur,
7971 page_end + 1 - cur);
7973 range_end = page_end;
7975 * No ordered extent covering this range, we are safe
7976 * to delete all extent states in the range.
7978 extra_flags = EXTENT_CLEAR_ALL_BITS;
7981 if (ordered->file_offset > cur) {
7983 * There is a range between [cur, oe->file_offset) not
7984 * covered by any ordered extent.
7985 * We are safe to delete all extent states, and handle
7986 * the ordered extent in the next iteration.
7988 range_end = ordered->file_offset - 1;
7989 extra_flags = EXTENT_CLEAR_ALL_BITS;
7993 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
7995 ASSERT(range_end + 1 - cur < U32_MAX);
7996 range_len = range_end + 1 - cur;
7997 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
7999 * If Ordered (Private2) is cleared, it means endio has
8000 * already been executed for the range.
8001 * We can't delete the extent states as
8002 * btrfs_finish_ordered_io() may still use some of them.
8006 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8009 * IO on this page will never be started, so we need to account
8010 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8011 * here, must leave that up for the ordered extent completion.
8013 * This will also unlock the range for incoming
8014 * btrfs_finish_ordered_io().
8016 if (!inode_evicting)
8017 clear_extent_bit(tree, cur, range_end,
8019 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8020 EXTENT_DEFRAG, &cached_state);
8022 spin_lock_irq(&inode->ordered_tree.lock);
8023 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8024 ordered->truncated_len = min(ordered->truncated_len,
8025 cur - ordered->file_offset);
8026 spin_unlock_irq(&inode->ordered_tree.lock);
8029 * If the ordered extent has finished, we're safe to delete all
8030 * the extent states of the range, otherwise
8031 * btrfs_finish_ordered_io() will get executed by endio for
8032 * other pages, so we can't delete extent states.
8034 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8035 cur, range_end + 1 - cur)) {
8036 btrfs_finish_ordered_io(ordered);
8038 * The ordered extent has finished, now we're again
8039 * safe to delete all extent states of the range.
8041 extra_flags = EXTENT_CLEAR_ALL_BITS;
8045 btrfs_put_ordered_extent(ordered);
8047 * Qgroup reserved space handler
8048 * Sector(s) here will be either:
8050 * 1) Already written to disk or bio already finished
8051 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8052 * Qgroup will be handled by its qgroup_record then.
8053 * btrfs_qgroup_free_data() call will do nothing here.
8055 * 2) Not written to disk yet
8056 * Then btrfs_qgroup_free_data() call will clear the
8057 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8058 * reserved data space.
8059 * Since the IO will never happen for this page.
8061 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8062 if (!inode_evicting) {
8063 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8064 EXTENT_DELALLOC | EXTENT_UPTODATE |
8065 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8066 extra_flags, &cached_state);
8068 cur = range_end + 1;
8071 * We have iterated through all ordered extents of the page, the page
8072 * should not have Ordered (Private2) anymore, or the above iteration
8073 * did something wrong.
8075 ASSERT(!folio_test_ordered(folio));
8076 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8077 if (!inode_evicting)
8078 __btrfs_release_folio(folio, GFP_NOFS);
8079 clear_page_extent_mapped(&folio->page);
8083 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8084 * called from a page fault handler when a page is first dirtied. Hence we must
8085 * be careful to check for EOF conditions here. We set the page up correctly
8086 * for a written page which means we get ENOSPC checking when writing into
8087 * holes and correct delalloc and unwritten extent mapping on filesystems that
8088 * support these features.
8090 * We are not allowed to take the i_mutex here so we have to play games to
8091 * protect against truncate races as the page could now be beyond EOF. Because
8092 * truncate_setsize() writes the inode size before removing pages, once we have
8093 * the page lock we can determine safely if the page is beyond EOF. If it is not
8094 * beyond EOF, then the page is guaranteed safe against truncation until we
8097 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8099 struct page *page = vmf->page;
8100 struct inode *inode = file_inode(vmf->vma->vm_file);
8101 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8102 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8103 struct btrfs_ordered_extent *ordered;
8104 struct extent_state *cached_state = NULL;
8105 struct extent_changeset *data_reserved = NULL;
8106 unsigned long zero_start;
8116 reserved_space = PAGE_SIZE;
8118 sb_start_pagefault(inode->i_sb);
8119 page_start = page_offset(page);
8120 page_end = page_start + PAGE_SIZE - 1;
8124 * Reserving delalloc space after obtaining the page lock can lead to
8125 * deadlock. For example, if a dirty page is locked by this function
8126 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8127 * dirty page write out, then the btrfs_writepages() function could
8128 * end up waiting indefinitely to get a lock on the page currently
8129 * being processed by btrfs_page_mkwrite() function.
8131 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8132 page_start, reserved_space);
8134 ret2 = file_update_time(vmf->vma->vm_file);
8138 ret = vmf_error(ret2);
8144 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8146 down_read(&BTRFS_I(inode)->i_mmap_lock);
8148 size = i_size_read(inode);
8150 if ((page->mapping != inode->i_mapping) ||
8151 (page_start >= size)) {
8152 /* page got truncated out from underneath us */
8155 wait_on_page_writeback(page);
8157 lock_extent(io_tree, page_start, page_end, &cached_state);
8158 ret2 = set_page_extent_mapped(page);
8160 ret = vmf_error(ret2);
8161 unlock_extent(io_tree, page_start, page_end, &cached_state);
8166 * we can't set the delalloc bits if there are pending ordered
8167 * extents. Drop our locks and wait for them to finish
8169 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8172 unlock_extent(io_tree, page_start, page_end, &cached_state);
8174 up_read(&BTRFS_I(inode)->i_mmap_lock);
8175 btrfs_start_ordered_extent(ordered);
8176 btrfs_put_ordered_extent(ordered);
8180 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8181 reserved_space = round_up(size - page_start,
8182 fs_info->sectorsize);
8183 if (reserved_space < PAGE_SIZE) {
8184 end = page_start + reserved_space - 1;
8185 btrfs_delalloc_release_space(BTRFS_I(inode),
8186 data_reserved, page_start,
8187 PAGE_SIZE - reserved_space, true);
8192 * page_mkwrite gets called when the page is firstly dirtied after it's
8193 * faulted in, but write(2) could also dirty a page and set delalloc
8194 * bits, thus in this case for space account reason, we still need to
8195 * clear any delalloc bits within this page range since we have to
8196 * reserve data&meta space before lock_page() (see above comments).
8198 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8199 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8200 EXTENT_DEFRAG, &cached_state);
8202 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8205 unlock_extent(io_tree, page_start, page_end, &cached_state);
8206 ret = VM_FAULT_SIGBUS;
8210 /* page is wholly or partially inside EOF */
8211 if (page_start + PAGE_SIZE > size)
8212 zero_start = offset_in_page(size);
8214 zero_start = PAGE_SIZE;
8216 if (zero_start != PAGE_SIZE)
8217 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8219 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8220 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8221 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8223 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8225 unlock_extent(io_tree, page_start, page_end, &cached_state);
8226 up_read(&BTRFS_I(inode)->i_mmap_lock);
8228 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8229 sb_end_pagefault(inode->i_sb);
8230 extent_changeset_free(data_reserved);
8231 return VM_FAULT_LOCKED;
8235 up_read(&BTRFS_I(inode)->i_mmap_lock);
8237 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8238 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8239 reserved_space, (ret != 0));
8241 sb_end_pagefault(inode->i_sb);
8242 extent_changeset_free(data_reserved);
8246 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8248 struct btrfs_truncate_control control = {
8250 .ino = btrfs_ino(inode),
8251 .min_type = BTRFS_EXTENT_DATA_KEY,
8252 .clear_extent_range = true,
8254 struct btrfs_root *root = inode->root;
8255 struct btrfs_fs_info *fs_info = root->fs_info;
8256 struct btrfs_block_rsv *rsv;
8258 struct btrfs_trans_handle *trans;
8259 u64 mask = fs_info->sectorsize - 1;
8260 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8262 if (!skip_writeback) {
8263 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8264 inode->vfs_inode.i_size & (~mask),
8271 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8272 * things going on here:
8274 * 1) We need to reserve space to update our inode.
8276 * 2) We need to have something to cache all the space that is going to
8277 * be free'd up by the truncate operation, but also have some slack
8278 * space reserved in case it uses space during the truncate (thank you
8279 * very much snapshotting).
8281 * And we need these to be separate. The fact is we can use a lot of
8282 * space doing the truncate, and we have no earthly idea how much space
8283 * we will use, so we need the truncate reservation to be separate so it
8284 * doesn't end up using space reserved for updating the inode. We also
8285 * need to be able to stop the transaction and start a new one, which
8286 * means we need to be able to update the inode several times, and we
8287 * have no idea of knowing how many times that will be, so we can't just
8288 * reserve 1 item for the entirety of the operation, so that has to be
8289 * done separately as well.
8291 * So that leaves us with
8293 * 1) rsv - for the truncate reservation, which we will steal from the
8294 * transaction reservation.
8295 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8296 * updating the inode.
8298 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8301 rsv->size = min_size;
8302 rsv->failfast = true;
8305 * 1 for the truncate slack space
8306 * 1 for updating the inode.
8308 trans = btrfs_start_transaction(root, 2);
8309 if (IS_ERR(trans)) {
8310 ret = PTR_ERR(trans);
8314 /* Migrate the slack space for the truncate to our reserve */
8315 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8318 * We have reserved 2 metadata units when we started the transaction and
8319 * min_size matches 1 unit, so this should never fail, but if it does,
8320 * it's not critical we just fail truncation.
8323 btrfs_end_transaction(trans);
8327 trans->block_rsv = rsv;
8330 struct extent_state *cached_state = NULL;
8331 const u64 new_size = inode->vfs_inode.i_size;
8332 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8334 control.new_size = new_size;
8335 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8337 * We want to drop from the next block forward in case this new
8338 * size is not block aligned since we will be keeping the last
8339 * block of the extent just the way it is.
8341 btrfs_drop_extent_map_range(inode,
8342 ALIGN(new_size, fs_info->sectorsize),
8345 ret = btrfs_truncate_inode_items(trans, root, &control);
8347 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8348 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8350 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8352 trans->block_rsv = &fs_info->trans_block_rsv;
8353 if (ret != -ENOSPC && ret != -EAGAIN)
8356 ret = btrfs_update_inode(trans, root, inode);
8360 btrfs_end_transaction(trans);
8361 btrfs_btree_balance_dirty(fs_info);
8363 trans = btrfs_start_transaction(root, 2);
8364 if (IS_ERR(trans)) {
8365 ret = PTR_ERR(trans);
8370 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8371 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8372 rsv, min_size, false);
8374 * We have reserved 2 metadata units when we started the
8375 * transaction and min_size matches 1 unit, so this should never
8376 * fail, but if it does, it's not critical we just fail truncation.
8381 trans->block_rsv = rsv;
8385 * We can't call btrfs_truncate_block inside a trans handle as we could
8386 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8387 * know we've truncated everything except the last little bit, and can
8388 * do btrfs_truncate_block and then update the disk_i_size.
8390 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8391 btrfs_end_transaction(trans);
8392 btrfs_btree_balance_dirty(fs_info);
8394 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8397 trans = btrfs_start_transaction(root, 1);
8398 if (IS_ERR(trans)) {
8399 ret = PTR_ERR(trans);
8402 btrfs_inode_safe_disk_i_size_write(inode, 0);
8408 trans->block_rsv = &fs_info->trans_block_rsv;
8409 ret2 = btrfs_update_inode(trans, root, inode);
8413 ret2 = btrfs_end_transaction(trans);
8416 btrfs_btree_balance_dirty(fs_info);
8419 btrfs_free_block_rsv(fs_info, rsv);
8421 * So if we truncate and then write and fsync we normally would just
8422 * write the extents that changed, which is a problem if we need to
8423 * first truncate that entire inode. So set this flag so we write out
8424 * all of the extents in the inode to the sync log so we're completely
8427 * If no extents were dropped or trimmed we don't need to force the next
8428 * fsync to truncate all the inode's items from the log and re-log them
8429 * all. This means the truncate operation did not change the file size,
8430 * or changed it to a smaller size but there was only an implicit hole
8431 * between the old i_size and the new i_size, and there were no prealloc
8432 * extents beyond i_size to drop.
8434 if (control.extents_found > 0)
8435 btrfs_set_inode_full_sync(inode);
8440 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8443 struct inode *inode;
8445 inode = new_inode(dir->i_sb);
8448 * Subvolumes don't inherit the sgid bit or the parent's gid if
8449 * the parent's sgid bit is set. This is probably a bug.
8451 inode_init_owner(idmap, inode, NULL,
8452 S_IFDIR | (~current_umask() & S_IRWXUGO));
8453 inode->i_op = &btrfs_dir_inode_operations;
8454 inode->i_fop = &btrfs_dir_file_operations;
8459 struct inode *btrfs_alloc_inode(struct super_block *sb)
8461 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8462 struct btrfs_inode *ei;
8463 struct inode *inode;
8465 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8472 ei->last_sub_trans = 0;
8473 ei->logged_trans = 0;
8474 ei->delalloc_bytes = 0;
8475 ei->new_delalloc_bytes = 0;
8476 ei->defrag_bytes = 0;
8477 ei->disk_i_size = 0;
8481 ei->index_cnt = (u64)-1;
8483 ei->last_unlink_trans = 0;
8484 ei->last_reflink_trans = 0;
8485 ei->last_log_commit = 0;
8487 spin_lock_init(&ei->lock);
8488 ei->outstanding_extents = 0;
8489 if (sb->s_magic != BTRFS_TEST_MAGIC)
8490 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8491 BTRFS_BLOCK_RSV_DELALLOC);
8492 ei->runtime_flags = 0;
8493 ei->prop_compress = BTRFS_COMPRESS_NONE;
8494 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8496 ei->delayed_node = NULL;
8498 ei->i_otime.tv_sec = 0;
8499 ei->i_otime.tv_nsec = 0;
8501 inode = &ei->vfs_inode;
8502 extent_map_tree_init(&ei->extent_tree);
8503 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8504 ei->io_tree.inode = ei;
8505 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8506 IO_TREE_INODE_FILE_EXTENT);
8507 mutex_init(&ei->log_mutex);
8508 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8509 INIT_LIST_HEAD(&ei->delalloc_inodes);
8510 INIT_LIST_HEAD(&ei->delayed_iput);
8511 RB_CLEAR_NODE(&ei->rb_node);
8512 init_rwsem(&ei->i_mmap_lock);
8517 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8518 void btrfs_test_destroy_inode(struct inode *inode)
8520 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8521 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8525 void btrfs_free_inode(struct inode *inode)
8527 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8530 void btrfs_destroy_inode(struct inode *vfs_inode)
8532 struct btrfs_ordered_extent *ordered;
8533 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8534 struct btrfs_root *root = inode->root;
8535 bool freespace_inode;
8537 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8538 WARN_ON(vfs_inode->i_data.nrpages);
8539 WARN_ON(inode->block_rsv.reserved);
8540 WARN_ON(inode->block_rsv.size);
8541 WARN_ON(inode->outstanding_extents);
8542 if (!S_ISDIR(vfs_inode->i_mode)) {
8543 WARN_ON(inode->delalloc_bytes);
8544 WARN_ON(inode->new_delalloc_bytes);
8546 WARN_ON(inode->csum_bytes);
8547 WARN_ON(inode->defrag_bytes);
8550 * This can happen where we create an inode, but somebody else also
8551 * created the same inode and we need to destroy the one we already
8558 * If this is a free space inode do not take the ordered extents lockdep
8561 freespace_inode = btrfs_is_free_space_inode(inode);
8564 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8568 btrfs_err(root->fs_info,
8569 "found ordered extent %llu %llu on inode cleanup",
8570 ordered->file_offset, ordered->num_bytes);
8572 if (!freespace_inode)
8573 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8575 btrfs_remove_ordered_extent(inode, ordered);
8576 btrfs_put_ordered_extent(ordered);
8577 btrfs_put_ordered_extent(ordered);
8580 btrfs_qgroup_check_reserved_leak(inode);
8581 inode_tree_del(inode);
8582 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8583 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8584 btrfs_put_root(inode->root);
8587 int btrfs_drop_inode(struct inode *inode)
8589 struct btrfs_root *root = BTRFS_I(inode)->root;
8594 /* the snap/subvol tree is on deleting */
8595 if (btrfs_root_refs(&root->root_item) == 0)
8598 return generic_drop_inode(inode);
8601 static void init_once(void *foo)
8603 struct btrfs_inode *ei = foo;
8605 inode_init_once(&ei->vfs_inode);
8608 void __cold btrfs_destroy_cachep(void)
8611 * Make sure all delayed rcu free inodes are flushed before we
8615 bioset_exit(&btrfs_dio_bioset);
8616 kmem_cache_destroy(btrfs_inode_cachep);
8619 int __init btrfs_init_cachep(void)
8621 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8622 sizeof(struct btrfs_inode), 0,
8623 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8625 if (!btrfs_inode_cachep)
8628 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8629 offsetof(struct btrfs_dio_private, bbio.bio),
8635 btrfs_destroy_cachep();
8639 static int btrfs_getattr(struct mnt_idmap *idmap,
8640 const struct path *path, struct kstat *stat,
8641 u32 request_mask, unsigned int flags)
8645 struct inode *inode = d_inode(path->dentry);
8646 u32 blocksize = inode->i_sb->s_blocksize;
8647 u32 bi_flags = BTRFS_I(inode)->flags;
8648 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8650 stat->result_mask |= STATX_BTIME;
8651 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8652 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8653 if (bi_flags & BTRFS_INODE_APPEND)
8654 stat->attributes |= STATX_ATTR_APPEND;
8655 if (bi_flags & BTRFS_INODE_COMPRESS)
8656 stat->attributes |= STATX_ATTR_COMPRESSED;
8657 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8658 stat->attributes |= STATX_ATTR_IMMUTABLE;
8659 if (bi_flags & BTRFS_INODE_NODUMP)
8660 stat->attributes |= STATX_ATTR_NODUMP;
8661 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8662 stat->attributes |= STATX_ATTR_VERITY;
8664 stat->attributes_mask |= (STATX_ATTR_APPEND |
8665 STATX_ATTR_COMPRESSED |
8666 STATX_ATTR_IMMUTABLE |
8669 generic_fillattr(idmap, inode, stat);
8670 stat->dev = BTRFS_I(inode)->root->anon_dev;
8672 spin_lock(&BTRFS_I(inode)->lock);
8673 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8674 inode_bytes = inode_get_bytes(inode);
8675 spin_unlock(&BTRFS_I(inode)->lock);
8676 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8677 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8681 static int btrfs_rename_exchange(struct inode *old_dir,
8682 struct dentry *old_dentry,
8683 struct inode *new_dir,
8684 struct dentry *new_dentry)
8686 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8687 struct btrfs_trans_handle *trans;
8688 unsigned int trans_num_items;
8689 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8690 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8691 struct inode *new_inode = new_dentry->d_inode;
8692 struct inode *old_inode = old_dentry->d_inode;
8693 struct timespec64 ctime = current_time(old_inode);
8694 struct btrfs_rename_ctx old_rename_ctx;
8695 struct btrfs_rename_ctx new_rename_ctx;
8696 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8697 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8702 bool need_abort = false;
8703 struct fscrypt_name old_fname, new_fname;
8704 struct fscrypt_str *old_name, *new_name;
8707 * For non-subvolumes allow exchange only within one subvolume, in the
8708 * same inode namespace. Two subvolumes (represented as directory) can
8709 * be exchanged as they're a logical link and have a fixed inode number.
8712 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8713 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8716 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8720 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8722 fscrypt_free_filename(&old_fname);
8726 old_name = &old_fname.disk_name;
8727 new_name = &new_fname.disk_name;
8729 /* close the race window with snapshot create/destroy ioctl */
8730 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8731 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8732 down_read(&fs_info->subvol_sem);
8736 * 1 to remove old dir item
8737 * 1 to remove old dir index
8738 * 1 to add new dir item
8739 * 1 to add new dir index
8740 * 1 to update parent inode
8742 * If the parents are the same, we only need to account for one
8744 trans_num_items = (old_dir == new_dir ? 9 : 10);
8745 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8747 * 1 to remove old root ref
8748 * 1 to remove old root backref
8749 * 1 to add new root ref
8750 * 1 to add new root backref
8752 trans_num_items += 4;
8755 * 1 to update inode item
8756 * 1 to remove old inode ref
8757 * 1 to add new inode ref
8759 trans_num_items += 3;
8761 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8762 trans_num_items += 4;
8764 trans_num_items += 3;
8765 trans = btrfs_start_transaction(root, trans_num_items);
8766 if (IS_ERR(trans)) {
8767 ret = PTR_ERR(trans);
8772 ret = btrfs_record_root_in_trans(trans, dest);
8778 * We need to find a free sequence number both in the source and
8779 * in the destination directory for the exchange.
8781 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8784 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8788 BTRFS_I(old_inode)->dir_index = 0ULL;
8789 BTRFS_I(new_inode)->dir_index = 0ULL;
8791 /* Reference for the source. */
8792 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8793 /* force full log commit if subvolume involved. */
8794 btrfs_set_log_full_commit(trans);
8796 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8797 btrfs_ino(BTRFS_I(new_dir)),
8804 /* And now for the dest. */
8805 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8806 /* force full log commit if subvolume involved. */
8807 btrfs_set_log_full_commit(trans);
8809 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8810 btrfs_ino(BTRFS_I(old_dir)),
8814 btrfs_abort_transaction(trans, ret);
8819 /* Update inode version and ctime/mtime. */
8820 inode_inc_iversion(old_dir);
8821 inode_inc_iversion(new_dir);
8822 inode_inc_iversion(old_inode);
8823 inode_inc_iversion(new_inode);
8824 old_dir->i_mtime = ctime;
8825 old_dir->i_ctime = ctime;
8826 new_dir->i_mtime = ctime;
8827 new_dir->i_ctime = ctime;
8828 old_inode->i_ctime = ctime;
8829 new_inode->i_ctime = ctime;
8831 if (old_dentry->d_parent != new_dentry->d_parent) {
8832 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8833 BTRFS_I(old_inode), true);
8834 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8835 BTRFS_I(new_inode), true);
8838 /* src is a subvolume */
8839 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8840 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8841 } else { /* src is an inode */
8842 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8843 BTRFS_I(old_dentry->d_inode),
8844 old_name, &old_rename_ctx);
8846 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8849 btrfs_abort_transaction(trans, ret);
8853 /* dest is a subvolume */
8854 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8855 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8856 } else { /* dest is an inode */
8857 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8858 BTRFS_I(new_dentry->d_inode),
8859 new_name, &new_rename_ctx);
8861 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8864 btrfs_abort_transaction(trans, ret);
8868 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
8869 new_name, 0, old_idx);
8871 btrfs_abort_transaction(trans, ret);
8875 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
8876 old_name, 0, new_idx);
8878 btrfs_abort_transaction(trans, ret);
8882 if (old_inode->i_nlink == 1)
8883 BTRFS_I(old_inode)->dir_index = old_idx;
8884 if (new_inode->i_nlink == 1)
8885 BTRFS_I(new_inode)->dir_index = new_idx;
8888 * Now pin the logs of the roots. We do it to ensure that no other task
8889 * can sync the logs while we are in progress with the rename, because
8890 * that could result in an inconsistency in case any of the inodes that
8891 * are part of this rename operation were logged before.
8893 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8894 btrfs_pin_log_trans(root);
8895 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8896 btrfs_pin_log_trans(dest);
8898 /* Do the log updates for all inodes. */
8899 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8900 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
8901 old_rename_ctx.index, new_dentry->d_parent);
8902 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8903 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
8904 new_rename_ctx.index, old_dentry->d_parent);
8906 /* Now unpin the logs. */
8907 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
8908 btrfs_end_log_trans(root);
8909 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
8910 btrfs_end_log_trans(dest);
8912 ret2 = btrfs_end_transaction(trans);
8913 ret = ret ? ret : ret2;
8915 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
8916 old_ino == BTRFS_FIRST_FREE_OBJECTID)
8917 up_read(&fs_info->subvol_sem);
8919 fscrypt_free_filename(&new_fname);
8920 fscrypt_free_filename(&old_fname);
8924 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
8927 struct inode *inode;
8929 inode = new_inode(dir->i_sb);
8931 inode_init_owner(idmap, inode, dir,
8932 S_IFCHR | WHITEOUT_MODE);
8933 inode->i_op = &btrfs_special_inode_operations;
8934 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
8939 static int btrfs_rename(struct mnt_idmap *idmap,
8940 struct inode *old_dir, struct dentry *old_dentry,
8941 struct inode *new_dir, struct dentry *new_dentry,
8944 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8945 struct btrfs_new_inode_args whiteout_args = {
8947 .dentry = old_dentry,
8949 struct btrfs_trans_handle *trans;
8950 unsigned int trans_num_items;
8951 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8952 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8953 struct inode *new_inode = d_inode(new_dentry);
8954 struct inode *old_inode = d_inode(old_dentry);
8955 struct btrfs_rename_ctx rename_ctx;
8959 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8960 struct fscrypt_name old_fname, new_fname;
8962 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8965 /* we only allow rename subvolume link between subvolumes */
8966 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8969 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8970 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
8973 if (S_ISDIR(old_inode->i_mode) && new_inode &&
8974 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8977 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8981 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8983 fscrypt_free_filename(&old_fname);
8987 /* check for collisions, even if the name isn't there */
8988 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
8990 if (ret == -EEXIST) {
8992 * eexist without a new_inode */
8993 if (WARN_ON(!new_inode)) {
8994 goto out_fscrypt_names;
8997 /* maybe -EOVERFLOW */
8998 goto out_fscrypt_names;
9004 * we're using rename to replace one file with another. Start IO on it
9005 * now so we don't add too much work to the end of the transaction
9007 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9008 filemap_flush(old_inode->i_mapping);
9010 if (flags & RENAME_WHITEOUT) {
9011 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9012 if (!whiteout_args.inode) {
9014 goto out_fscrypt_names;
9016 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9018 goto out_whiteout_inode;
9020 /* 1 to update the old parent inode. */
9021 trans_num_items = 1;
9024 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9025 /* Close the race window with snapshot create/destroy ioctl */
9026 down_read(&fs_info->subvol_sem);
9028 * 1 to remove old root ref
9029 * 1 to remove old root backref
9030 * 1 to add new root ref
9031 * 1 to add new root backref
9033 trans_num_items += 4;
9037 * 1 to remove old inode ref
9038 * 1 to add new inode ref
9040 trans_num_items += 3;
9043 * 1 to remove old dir item
9044 * 1 to remove old dir index
9045 * 1 to add new dir item
9046 * 1 to add new dir index
9048 trans_num_items += 4;
9049 /* 1 to update new parent inode if it's not the same as the old parent */
9050 if (new_dir != old_dir)
9055 * 1 to remove inode ref
9056 * 1 to remove dir item
9057 * 1 to remove dir index
9058 * 1 to possibly add orphan item
9060 trans_num_items += 5;
9062 trans = btrfs_start_transaction(root, trans_num_items);
9063 if (IS_ERR(trans)) {
9064 ret = PTR_ERR(trans);
9069 ret = btrfs_record_root_in_trans(trans, dest);
9074 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9078 BTRFS_I(old_inode)->dir_index = 0ULL;
9079 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9080 /* force full log commit if subvolume involved. */
9081 btrfs_set_log_full_commit(trans);
9083 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9084 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9090 inode_inc_iversion(old_dir);
9091 inode_inc_iversion(new_dir);
9092 inode_inc_iversion(old_inode);
9093 old_dir->i_mtime = current_time(old_dir);
9094 old_dir->i_ctime = old_dir->i_mtime;
9095 new_dir->i_mtime = old_dir->i_mtime;
9096 new_dir->i_ctime = old_dir->i_mtime;
9097 old_inode->i_ctime = old_dir->i_mtime;
9099 if (old_dentry->d_parent != new_dentry->d_parent)
9100 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9101 BTRFS_I(old_inode), true);
9103 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9104 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9106 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9107 BTRFS_I(d_inode(old_dentry)),
9108 &old_fname.disk_name, &rename_ctx);
9110 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9113 btrfs_abort_transaction(trans, ret);
9118 inode_inc_iversion(new_inode);
9119 new_inode->i_ctime = current_time(new_inode);
9120 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9121 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9122 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9123 BUG_ON(new_inode->i_nlink == 0);
9125 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9126 BTRFS_I(d_inode(new_dentry)),
9127 &new_fname.disk_name);
9129 if (!ret && new_inode->i_nlink == 0)
9130 ret = btrfs_orphan_add(trans,
9131 BTRFS_I(d_inode(new_dentry)));
9133 btrfs_abort_transaction(trans, ret);
9138 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9139 &new_fname.disk_name, 0, index);
9141 btrfs_abort_transaction(trans, ret);
9145 if (old_inode->i_nlink == 1)
9146 BTRFS_I(old_inode)->dir_index = index;
9148 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9149 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9150 rename_ctx.index, new_dentry->d_parent);
9152 if (flags & RENAME_WHITEOUT) {
9153 ret = btrfs_create_new_inode(trans, &whiteout_args);
9155 btrfs_abort_transaction(trans, ret);
9158 unlock_new_inode(whiteout_args.inode);
9159 iput(whiteout_args.inode);
9160 whiteout_args.inode = NULL;
9164 ret2 = btrfs_end_transaction(trans);
9165 ret = ret ? ret : ret2;
9167 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9168 up_read(&fs_info->subvol_sem);
9169 if (flags & RENAME_WHITEOUT)
9170 btrfs_new_inode_args_destroy(&whiteout_args);
9172 if (flags & RENAME_WHITEOUT)
9173 iput(whiteout_args.inode);
9175 fscrypt_free_filename(&old_fname);
9176 fscrypt_free_filename(&new_fname);
9180 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9181 struct dentry *old_dentry, struct inode *new_dir,
9182 struct dentry *new_dentry, unsigned int flags)
9186 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9189 if (flags & RENAME_EXCHANGE)
9190 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9193 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9196 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9201 struct btrfs_delalloc_work {
9202 struct inode *inode;
9203 struct completion completion;
9204 struct list_head list;
9205 struct btrfs_work work;
9208 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9210 struct btrfs_delalloc_work *delalloc_work;
9211 struct inode *inode;
9213 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9215 inode = delalloc_work->inode;
9216 filemap_flush(inode->i_mapping);
9217 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9218 &BTRFS_I(inode)->runtime_flags))
9219 filemap_flush(inode->i_mapping);
9222 complete(&delalloc_work->completion);
9225 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9227 struct btrfs_delalloc_work *work;
9229 work = kmalloc(sizeof(*work), GFP_NOFS);
9233 init_completion(&work->completion);
9234 INIT_LIST_HEAD(&work->list);
9235 work->inode = inode;
9236 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9242 * some fairly slow code that needs optimization. This walks the list
9243 * of all the inodes with pending delalloc and forces them to disk.
9245 static int start_delalloc_inodes(struct btrfs_root *root,
9246 struct writeback_control *wbc, bool snapshot,
9247 bool in_reclaim_context)
9249 struct btrfs_inode *binode;
9250 struct inode *inode;
9251 struct btrfs_delalloc_work *work, *next;
9252 struct list_head works;
9253 struct list_head splice;
9255 bool full_flush = wbc->nr_to_write == LONG_MAX;
9257 INIT_LIST_HEAD(&works);
9258 INIT_LIST_HEAD(&splice);
9260 mutex_lock(&root->delalloc_mutex);
9261 spin_lock(&root->delalloc_lock);
9262 list_splice_init(&root->delalloc_inodes, &splice);
9263 while (!list_empty(&splice)) {
9264 binode = list_entry(splice.next, struct btrfs_inode,
9267 list_move_tail(&binode->delalloc_inodes,
9268 &root->delalloc_inodes);
9270 if (in_reclaim_context &&
9271 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9274 inode = igrab(&binode->vfs_inode);
9276 cond_resched_lock(&root->delalloc_lock);
9279 spin_unlock(&root->delalloc_lock);
9282 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9283 &binode->runtime_flags);
9285 work = btrfs_alloc_delalloc_work(inode);
9291 list_add_tail(&work->list, &works);
9292 btrfs_queue_work(root->fs_info->flush_workers,
9295 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9296 btrfs_add_delayed_iput(BTRFS_I(inode));
9297 if (ret || wbc->nr_to_write <= 0)
9301 spin_lock(&root->delalloc_lock);
9303 spin_unlock(&root->delalloc_lock);
9306 list_for_each_entry_safe(work, next, &works, list) {
9307 list_del_init(&work->list);
9308 wait_for_completion(&work->completion);
9312 if (!list_empty(&splice)) {
9313 spin_lock(&root->delalloc_lock);
9314 list_splice_tail(&splice, &root->delalloc_inodes);
9315 spin_unlock(&root->delalloc_lock);
9317 mutex_unlock(&root->delalloc_mutex);
9321 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9323 struct writeback_control wbc = {
9324 .nr_to_write = LONG_MAX,
9325 .sync_mode = WB_SYNC_NONE,
9327 .range_end = LLONG_MAX,
9329 struct btrfs_fs_info *fs_info = root->fs_info;
9331 if (BTRFS_FS_ERROR(fs_info))
9334 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9337 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9338 bool in_reclaim_context)
9340 struct writeback_control wbc = {
9342 .sync_mode = WB_SYNC_NONE,
9344 .range_end = LLONG_MAX,
9346 struct btrfs_root *root;
9347 struct list_head splice;
9350 if (BTRFS_FS_ERROR(fs_info))
9353 INIT_LIST_HEAD(&splice);
9355 mutex_lock(&fs_info->delalloc_root_mutex);
9356 spin_lock(&fs_info->delalloc_root_lock);
9357 list_splice_init(&fs_info->delalloc_roots, &splice);
9358 while (!list_empty(&splice)) {
9360 * Reset nr_to_write here so we know that we're doing a full
9364 wbc.nr_to_write = LONG_MAX;
9366 root = list_first_entry(&splice, struct btrfs_root,
9368 root = btrfs_grab_root(root);
9370 list_move_tail(&root->delalloc_root,
9371 &fs_info->delalloc_roots);
9372 spin_unlock(&fs_info->delalloc_root_lock);
9374 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9375 btrfs_put_root(root);
9376 if (ret < 0 || wbc.nr_to_write <= 0)
9378 spin_lock(&fs_info->delalloc_root_lock);
9380 spin_unlock(&fs_info->delalloc_root_lock);
9384 if (!list_empty(&splice)) {
9385 spin_lock(&fs_info->delalloc_root_lock);
9386 list_splice_tail(&splice, &fs_info->delalloc_roots);
9387 spin_unlock(&fs_info->delalloc_root_lock);
9389 mutex_unlock(&fs_info->delalloc_root_mutex);
9393 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9394 struct dentry *dentry, const char *symname)
9396 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9397 struct btrfs_trans_handle *trans;
9398 struct btrfs_root *root = BTRFS_I(dir)->root;
9399 struct btrfs_path *path;
9400 struct btrfs_key key;
9401 struct inode *inode;
9402 struct btrfs_new_inode_args new_inode_args = {
9406 unsigned int trans_num_items;
9411 struct btrfs_file_extent_item *ei;
9412 struct extent_buffer *leaf;
9414 name_len = strlen(symname);
9415 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9416 return -ENAMETOOLONG;
9418 inode = new_inode(dir->i_sb);
9421 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9422 inode->i_op = &btrfs_symlink_inode_operations;
9423 inode_nohighmem(inode);
9424 inode->i_mapping->a_ops = &btrfs_aops;
9425 btrfs_i_size_write(BTRFS_I(inode), name_len);
9426 inode_set_bytes(inode, name_len);
9428 new_inode_args.inode = inode;
9429 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9432 /* 1 additional item for the inline extent */
9435 trans = btrfs_start_transaction(root, trans_num_items);
9436 if (IS_ERR(trans)) {
9437 err = PTR_ERR(trans);
9438 goto out_new_inode_args;
9441 err = btrfs_create_new_inode(trans, &new_inode_args);
9445 path = btrfs_alloc_path();
9448 btrfs_abort_transaction(trans, err);
9449 discard_new_inode(inode);
9453 key.objectid = btrfs_ino(BTRFS_I(inode));
9455 key.type = BTRFS_EXTENT_DATA_KEY;
9456 datasize = btrfs_file_extent_calc_inline_size(name_len);
9457 err = btrfs_insert_empty_item(trans, root, path, &key,
9460 btrfs_abort_transaction(trans, err);
9461 btrfs_free_path(path);
9462 discard_new_inode(inode);
9466 leaf = path->nodes[0];
9467 ei = btrfs_item_ptr(leaf, path->slots[0],
9468 struct btrfs_file_extent_item);
9469 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9470 btrfs_set_file_extent_type(leaf, ei,
9471 BTRFS_FILE_EXTENT_INLINE);
9472 btrfs_set_file_extent_encryption(leaf, ei, 0);
9473 btrfs_set_file_extent_compression(leaf, ei, 0);
9474 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9475 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9477 ptr = btrfs_file_extent_inline_start(ei);
9478 write_extent_buffer(leaf, symname, ptr, name_len);
9479 btrfs_mark_buffer_dirty(leaf);
9480 btrfs_free_path(path);
9482 d_instantiate_new(dentry, inode);
9485 btrfs_end_transaction(trans);
9486 btrfs_btree_balance_dirty(fs_info);
9488 btrfs_new_inode_args_destroy(&new_inode_args);
9495 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9496 struct btrfs_trans_handle *trans_in,
9497 struct btrfs_inode *inode,
9498 struct btrfs_key *ins,
9501 struct btrfs_file_extent_item stack_fi;
9502 struct btrfs_replace_extent_info extent_info;
9503 struct btrfs_trans_handle *trans = trans_in;
9504 struct btrfs_path *path;
9505 u64 start = ins->objectid;
9506 u64 len = ins->offset;
9507 int qgroup_released;
9510 memset(&stack_fi, 0, sizeof(stack_fi));
9512 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9513 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9514 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9515 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9516 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9517 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9518 /* Encryption and other encoding is reserved and all 0 */
9520 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9521 if (qgroup_released < 0)
9522 return ERR_PTR(qgroup_released);
9525 ret = insert_reserved_file_extent(trans, inode,
9526 file_offset, &stack_fi,
9527 true, qgroup_released);
9533 extent_info.disk_offset = start;
9534 extent_info.disk_len = len;
9535 extent_info.data_offset = 0;
9536 extent_info.data_len = len;
9537 extent_info.file_offset = file_offset;
9538 extent_info.extent_buf = (char *)&stack_fi;
9539 extent_info.is_new_extent = true;
9540 extent_info.update_times = true;
9541 extent_info.qgroup_reserved = qgroup_released;
9542 extent_info.insertions = 0;
9544 path = btrfs_alloc_path();
9550 ret = btrfs_replace_file_extents(inode, path, file_offset,
9551 file_offset + len - 1, &extent_info,
9553 btrfs_free_path(path);
9560 * We have released qgroup data range at the beginning of the function,
9561 * and normally qgroup_released bytes will be freed when committing
9563 * But if we error out early, we have to free what we have released
9564 * or we leak qgroup data reservation.
9566 btrfs_qgroup_free_refroot(inode->root->fs_info,
9567 inode->root->root_key.objectid, qgroup_released,
9568 BTRFS_QGROUP_RSV_DATA);
9569 return ERR_PTR(ret);
9572 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9573 u64 start, u64 num_bytes, u64 min_size,
9574 loff_t actual_len, u64 *alloc_hint,
9575 struct btrfs_trans_handle *trans)
9577 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9578 struct extent_map *em;
9579 struct btrfs_root *root = BTRFS_I(inode)->root;
9580 struct btrfs_key ins;
9581 u64 cur_offset = start;
9582 u64 clear_offset = start;
9585 u64 last_alloc = (u64)-1;
9587 bool own_trans = true;
9588 u64 end = start + num_bytes - 1;
9592 while (num_bytes > 0) {
9593 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9594 cur_bytes = max(cur_bytes, min_size);
9596 * If we are severely fragmented we could end up with really
9597 * small allocations, so if the allocator is returning small
9598 * chunks lets make its job easier by only searching for those
9601 cur_bytes = min(cur_bytes, last_alloc);
9602 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9603 min_size, 0, *alloc_hint, &ins, 1, 0);
9608 * We've reserved this space, and thus converted it from
9609 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9610 * from here on out we will only need to clear our reservation
9611 * for the remaining unreserved area, so advance our
9612 * clear_offset by our extent size.
9614 clear_offset += ins.offset;
9616 last_alloc = ins.offset;
9617 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9620 * Now that we inserted the prealloc extent we can finally
9621 * decrement the number of reservations in the block group.
9622 * If we did it before, we could race with relocation and have
9623 * relocation miss the reserved extent, making it fail later.
9625 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9626 if (IS_ERR(trans)) {
9627 ret = PTR_ERR(trans);
9628 btrfs_free_reserved_extent(fs_info, ins.objectid,
9633 em = alloc_extent_map();
9635 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9636 cur_offset + ins.offset - 1, false);
9637 btrfs_set_inode_full_sync(BTRFS_I(inode));
9641 em->start = cur_offset;
9642 em->orig_start = cur_offset;
9643 em->len = ins.offset;
9644 em->block_start = ins.objectid;
9645 em->block_len = ins.offset;
9646 em->orig_block_len = ins.offset;
9647 em->ram_bytes = ins.offset;
9648 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9649 em->generation = trans->transid;
9651 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9652 free_extent_map(em);
9654 num_bytes -= ins.offset;
9655 cur_offset += ins.offset;
9656 *alloc_hint = ins.objectid + ins.offset;
9658 inode_inc_iversion(inode);
9659 inode->i_ctime = current_time(inode);
9660 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9661 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9662 (actual_len > inode->i_size) &&
9663 (cur_offset > inode->i_size)) {
9664 if (cur_offset > actual_len)
9665 i_size = actual_len;
9667 i_size = cur_offset;
9668 i_size_write(inode, i_size);
9669 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9672 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9675 btrfs_abort_transaction(trans, ret);
9677 btrfs_end_transaction(trans);
9682 btrfs_end_transaction(trans);
9686 if (clear_offset < end)
9687 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9688 end - clear_offset + 1);
9692 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9693 u64 start, u64 num_bytes, u64 min_size,
9694 loff_t actual_len, u64 *alloc_hint)
9696 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9697 min_size, actual_len, alloc_hint,
9701 int btrfs_prealloc_file_range_trans(struct inode *inode,
9702 struct btrfs_trans_handle *trans, int mode,
9703 u64 start, u64 num_bytes, u64 min_size,
9704 loff_t actual_len, u64 *alloc_hint)
9706 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9707 min_size, actual_len, alloc_hint, trans);
9710 static int btrfs_permission(struct mnt_idmap *idmap,
9711 struct inode *inode, int mask)
9713 struct btrfs_root *root = BTRFS_I(inode)->root;
9714 umode_t mode = inode->i_mode;
9716 if (mask & MAY_WRITE &&
9717 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9718 if (btrfs_root_readonly(root))
9720 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9723 return generic_permission(idmap, inode, mask);
9726 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9727 struct file *file, umode_t mode)
9729 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9730 struct btrfs_trans_handle *trans;
9731 struct btrfs_root *root = BTRFS_I(dir)->root;
9732 struct inode *inode;
9733 struct btrfs_new_inode_args new_inode_args = {
9735 .dentry = file->f_path.dentry,
9738 unsigned int trans_num_items;
9741 inode = new_inode(dir->i_sb);
9744 inode_init_owner(idmap, inode, dir, mode);
9745 inode->i_fop = &btrfs_file_operations;
9746 inode->i_op = &btrfs_file_inode_operations;
9747 inode->i_mapping->a_ops = &btrfs_aops;
9749 new_inode_args.inode = inode;
9750 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9754 trans = btrfs_start_transaction(root, trans_num_items);
9755 if (IS_ERR(trans)) {
9756 ret = PTR_ERR(trans);
9757 goto out_new_inode_args;
9760 ret = btrfs_create_new_inode(trans, &new_inode_args);
9763 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9764 * set it to 1 because d_tmpfile() will issue a warning if the count is
9767 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9769 set_nlink(inode, 1);
9772 d_tmpfile(file, inode);
9773 unlock_new_inode(inode);
9774 mark_inode_dirty(inode);
9777 btrfs_end_transaction(trans);
9778 btrfs_btree_balance_dirty(fs_info);
9780 btrfs_new_inode_args_destroy(&new_inode_args);
9784 return finish_open_simple(file, ret);
9787 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9789 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9790 unsigned long index = start >> PAGE_SHIFT;
9791 unsigned long end_index = end >> PAGE_SHIFT;
9795 ASSERT(end + 1 - start <= U32_MAX);
9796 len = end + 1 - start;
9797 while (index <= end_index) {
9798 page = find_get_page(inode->vfs_inode.i_mapping, index);
9799 ASSERT(page); /* Pages should be in the extent_io_tree */
9801 btrfs_page_set_writeback(fs_info, page, start, len);
9807 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9810 switch (compress_type) {
9811 case BTRFS_COMPRESS_NONE:
9812 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9813 case BTRFS_COMPRESS_ZLIB:
9814 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9815 case BTRFS_COMPRESS_LZO:
9817 * The LZO format depends on the sector size. 64K is the maximum
9818 * sector size that we support.
9820 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9822 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9823 (fs_info->sectorsize_bits - 12);
9824 case BTRFS_COMPRESS_ZSTD:
9825 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9831 static ssize_t btrfs_encoded_read_inline(
9833 struct iov_iter *iter, u64 start,
9835 struct extent_state **cached_state,
9836 u64 extent_start, size_t count,
9837 struct btrfs_ioctl_encoded_io_args *encoded,
9840 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9841 struct btrfs_root *root = inode->root;
9842 struct btrfs_fs_info *fs_info = root->fs_info;
9843 struct extent_io_tree *io_tree = &inode->io_tree;
9844 struct btrfs_path *path;
9845 struct extent_buffer *leaf;
9846 struct btrfs_file_extent_item *item;
9852 path = btrfs_alloc_path();
9857 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9861 /* The extent item disappeared? */
9866 leaf = path->nodes[0];
9867 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
9869 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
9870 ptr = btrfs_file_extent_inline_start(item);
9872 encoded->len = min_t(u64, extent_start + ram_bytes,
9873 inode->vfs_inode.i_size) - iocb->ki_pos;
9874 ret = btrfs_encoded_io_compression_from_extent(fs_info,
9875 btrfs_file_extent_compression(leaf, item));
9878 encoded->compression = ret;
9879 if (encoded->compression) {
9882 inline_size = btrfs_file_extent_inline_item_len(leaf,
9884 if (inline_size > count) {
9888 count = inline_size;
9889 encoded->unencoded_len = ram_bytes;
9890 encoded->unencoded_offset = iocb->ki_pos - extent_start;
9892 count = min_t(u64, count, encoded->len);
9893 encoded->len = count;
9894 encoded->unencoded_len = count;
9895 ptr += iocb->ki_pos - extent_start;
9898 tmp = kmalloc(count, GFP_NOFS);
9903 read_extent_buffer(leaf, tmp, ptr, count);
9904 btrfs_release_path(path);
9905 unlock_extent(io_tree, start, lockend, cached_state);
9906 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
9909 ret = copy_to_iter(tmp, count, iter);
9914 btrfs_free_path(path);
9918 struct btrfs_encoded_read_private {
9919 wait_queue_head_t wait;
9921 blk_status_t status;
9924 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
9926 struct btrfs_encoded_read_private *priv = bbio->private;
9928 if (bbio->bio.bi_status) {
9930 * The memory barrier implied by the atomic_dec_return() here
9931 * pairs with the memory barrier implied by the
9932 * atomic_dec_return() or io_wait_event() in
9933 * btrfs_encoded_read_regular_fill_pages() to ensure that this
9934 * write is observed before the load of status in
9935 * btrfs_encoded_read_regular_fill_pages().
9937 WRITE_ONCE(priv->status, bbio->bio.bi_status);
9939 if (!atomic_dec_return(&priv->pending))
9940 wake_up(&priv->wait);
9941 bio_put(&bbio->bio);
9944 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
9945 u64 file_offset, u64 disk_bytenr,
9946 u64 disk_io_size, struct page **pages)
9948 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9949 struct btrfs_encoded_read_private priv = {
9950 .pending = ATOMIC_INIT(1),
9952 unsigned long i = 0;
9953 struct btrfs_bio *bbio;
9955 init_waitqueue_head(&priv.wait);
9957 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9958 btrfs_encoded_read_endio, &priv);
9959 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9960 bbio->inode = inode;
9963 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
9965 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
9966 atomic_inc(&priv.pending);
9967 btrfs_submit_bio(bbio, 0);
9969 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
9970 btrfs_encoded_read_endio, &priv);
9971 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
9972 bbio->inode = inode;
9977 disk_bytenr += bytes;
9978 disk_io_size -= bytes;
9979 } while (disk_io_size);
9981 atomic_inc(&priv.pending);
9982 btrfs_submit_bio(bbio, 0);
9984 if (atomic_dec_return(&priv.pending))
9985 io_wait_event(priv.wait, !atomic_read(&priv.pending));
9986 /* See btrfs_encoded_read_endio() for ordering. */
9987 return blk_status_to_errno(READ_ONCE(priv.status));
9990 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
9991 struct iov_iter *iter,
9992 u64 start, u64 lockend,
9993 struct extent_state **cached_state,
9994 u64 disk_bytenr, u64 disk_io_size,
9995 size_t count, bool compressed,
9998 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9999 struct extent_io_tree *io_tree = &inode->io_tree;
10000 struct page **pages;
10001 unsigned long nr_pages, i;
10003 size_t page_offset;
10006 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10007 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10010 ret = btrfs_alloc_page_array(nr_pages, pages);
10016 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10017 disk_io_size, pages);
10021 unlock_extent(io_tree, start, lockend, cached_state);
10022 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10029 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10030 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10033 while (cur < count) {
10034 size_t bytes = min_t(size_t, count - cur,
10035 PAGE_SIZE - page_offset);
10037 if (copy_page_to_iter(pages[i], page_offset, bytes,
10048 for (i = 0; i < nr_pages; i++) {
10050 __free_page(pages[i]);
10056 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10057 struct btrfs_ioctl_encoded_io_args *encoded)
10059 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10060 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10061 struct extent_io_tree *io_tree = &inode->io_tree;
10063 size_t count = iov_iter_count(iter);
10064 u64 start, lockend, disk_bytenr, disk_io_size;
10065 struct extent_state *cached_state = NULL;
10066 struct extent_map *em;
10067 bool unlocked = false;
10069 file_accessed(iocb->ki_filp);
10071 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10073 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10074 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10077 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10079 * We don't know how long the extent containing iocb->ki_pos is, but if
10080 * it's compressed we know that it won't be longer than this.
10082 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10085 struct btrfs_ordered_extent *ordered;
10087 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10088 lockend - start + 1);
10090 goto out_unlock_inode;
10091 lock_extent(io_tree, start, lockend, &cached_state);
10092 ordered = btrfs_lookup_ordered_range(inode, start,
10093 lockend - start + 1);
10096 btrfs_put_ordered_extent(ordered);
10097 unlock_extent(io_tree, start, lockend, &cached_state);
10101 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10104 goto out_unlock_extent;
10107 if (em->block_start == EXTENT_MAP_INLINE) {
10108 u64 extent_start = em->start;
10111 * For inline extents we get everything we need out of the
10114 free_extent_map(em);
10116 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10117 &cached_state, extent_start,
10118 count, encoded, &unlocked);
10123 * We only want to return up to EOF even if the extent extends beyond
10126 encoded->len = min_t(u64, extent_map_end(em),
10127 inode->vfs_inode.i_size) - iocb->ki_pos;
10128 if (em->block_start == EXTENT_MAP_HOLE ||
10129 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10130 disk_bytenr = EXTENT_MAP_HOLE;
10131 count = min_t(u64, count, encoded->len);
10132 encoded->len = count;
10133 encoded->unencoded_len = count;
10134 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10135 disk_bytenr = em->block_start;
10137 * Bail if the buffer isn't large enough to return the whole
10138 * compressed extent.
10140 if (em->block_len > count) {
10144 disk_io_size = em->block_len;
10145 count = em->block_len;
10146 encoded->unencoded_len = em->ram_bytes;
10147 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10148 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10149 em->compress_type);
10152 encoded->compression = ret;
10154 disk_bytenr = em->block_start + (start - em->start);
10155 if (encoded->len > count)
10156 encoded->len = count;
10158 * Don't read beyond what we locked. This also limits the page
10159 * allocations that we'll do.
10161 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10162 count = start + disk_io_size - iocb->ki_pos;
10163 encoded->len = count;
10164 encoded->unencoded_len = count;
10165 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10167 free_extent_map(em);
10170 if (disk_bytenr == EXTENT_MAP_HOLE) {
10171 unlock_extent(io_tree, start, lockend, &cached_state);
10172 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10174 ret = iov_iter_zero(count, iter);
10178 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10179 &cached_state, disk_bytenr,
10180 disk_io_size, count,
10181 encoded->compression,
10187 iocb->ki_pos += encoded->len;
10189 free_extent_map(em);
10192 unlock_extent(io_tree, start, lockend, &cached_state);
10195 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10199 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10200 const struct btrfs_ioctl_encoded_io_args *encoded)
10202 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10203 struct btrfs_root *root = inode->root;
10204 struct btrfs_fs_info *fs_info = root->fs_info;
10205 struct extent_io_tree *io_tree = &inode->io_tree;
10206 struct extent_changeset *data_reserved = NULL;
10207 struct extent_state *cached_state = NULL;
10208 struct btrfs_ordered_extent *ordered;
10212 u64 num_bytes, ram_bytes, disk_num_bytes;
10213 unsigned long nr_pages, i;
10214 struct page **pages;
10215 struct btrfs_key ins;
10216 bool extent_reserved = false;
10217 struct extent_map *em;
10220 switch (encoded->compression) {
10221 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10222 compression = BTRFS_COMPRESS_ZLIB;
10224 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10225 compression = BTRFS_COMPRESS_ZSTD;
10227 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10228 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10229 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10230 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10231 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10232 /* The sector size must match for LZO. */
10233 if (encoded->compression -
10234 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10235 fs_info->sectorsize_bits)
10237 compression = BTRFS_COMPRESS_LZO;
10242 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10245 orig_count = iov_iter_count(from);
10247 /* The extent size must be sane. */
10248 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10249 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10253 * The compressed data must be smaller than the decompressed data.
10255 * It's of course possible for data to compress to larger or the same
10256 * size, but the buffered I/O path falls back to no compression for such
10257 * data, and we don't want to break any assumptions by creating these
10260 * Note that this is less strict than the current check we have that the
10261 * compressed data must be at least one sector smaller than the
10262 * decompressed data. We only want to enforce the weaker requirement
10263 * from old kernels that it is at least one byte smaller.
10265 if (orig_count >= encoded->unencoded_len)
10268 /* The extent must start on a sector boundary. */
10269 start = iocb->ki_pos;
10270 if (!IS_ALIGNED(start, fs_info->sectorsize))
10274 * The extent must end on a sector boundary. However, we allow a write
10275 * which ends at or extends i_size to have an unaligned length; we round
10276 * up the extent size and set i_size to the unaligned end.
10278 if (start + encoded->len < inode->vfs_inode.i_size &&
10279 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10282 /* Finally, the offset in the unencoded data must be sector-aligned. */
10283 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10286 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10287 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10288 end = start + num_bytes - 1;
10291 * If the extent cannot be inline, the compressed data on disk must be
10292 * sector-aligned. For convenience, we extend it with zeroes if it
10295 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10296 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10297 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10300 for (i = 0; i < nr_pages; i++) {
10301 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10304 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10309 kaddr = kmap_local_page(pages[i]);
10310 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10311 kunmap_local(kaddr);
10315 if (bytes < PAGE_SIZE)
10316 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10317 kunmap_local(kaddr);
10321 struct btrfs_ordered_extent *ordered;
10323 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10326 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10327 start >> PAGE_SHIFT,
10328 end >> PAGE_SHIFT);
10331 lock_extent(io_tree, start, end, &cached_state);
10332 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10334 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10337 btrfs_put_ordered_extent(ordered);
10338 unlock_extent(io_tree, start, end, &cached_state);
10343 * We don't use the higher-level delalloc space functions because our
10344 * num_bytes and disk_num_bytes are different.
10346 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10349 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10351 goto out_free_data_space;
10352 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10355 goto out_qgroup_free_data;
10357 /* Try an inline extent first. */
10358 if (start == 0 && encoded->unencoded_len == encoded->len &&
10359 encoded->unencoded_offset == 0) {
10360 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10361 compression, pages, true);
10365 goto out_delalloc_release;
10369 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10370 disk_num_bytes, 0, 0, &ins, 1, 1);
10372 goto out_delalloc_release;
10373 extent_reserved = true;
10375 em = create_io_em(inode, start, num_bytes,
10376 start - encoded->unencoded_offset, ins.objectid,
10377 ins.offset, ins.offset, ram_bytes, compression,
10378 BTRFS_ORDERED_COMPRESSED);
10381 goto out_free_reserved;
10383 free_extent_map(em);
10385 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10386 ins.objectid, ins.offset,
10387 encoded->unencoded_offset,
10388 (1 << BTRFS_ORDERED_ENCODED) |
10389 (1 << BTRFS_ORDERED_COMPRESSED),
10391 if (IS_ERR(ordered)) {
10392 btrfs_drop_extent_map_range(inode, start, end, false);
10393 ret = PTR_ERR(ordered);
10394 goto out_free_reserved;
10396 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10398 if (start + encoded->len > inode->vfs_inode.i_size)
10399 i_size_write(&inode->vfs_inode, start + encoded->len);
10401 unlock_extent(io_tree, start, end, &cached_state);
10403 btrfs_delalloc_release_extents(inode, num_bytes);
10405 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10410 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10411 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10412 out_delalloc_release:
10413 btrfs_delalloc_release_extents(inode, num_bytes);
10414 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10415 out_qgroup_free_data:
10417 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10418 out_free_data_space:
10420 * If btrfs_reserve_extent() succeeded, then we already decremented
10423 if (!extent_reserved)
10424 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10426 unlock_extent(io_tree, start, end, &cached_state);
10428 for (i = 0; i < nr_pages; i++) {
10430 __free_page(pages[i]);
10435 iocb->ki_pos += encoded->len;
10441 * Add an entry indicating a block group or device which is pinned by a
10442 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10443 * negative errno on failure.
10445 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10446 bool is_block_group)
10448 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10449 struct btrfs_swapfile_pin *sp, *entry;
10450 struct rb_node **p;
10451 struct rb_node *parent = NULL;
10453 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10458 sp->is_block_group = is_block_group;
10459 sp->bg_extent_count = 1;
10461 spin_lock(&fs_info->swapfile_pins_lock);
10462 p = &fs_info->swapfile_pins.rb_node;
10465 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10466 if (sp->ptr < entry->ptr ||
10467 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10468 p = &(*p)->rb_left;
10469 } else if (sp->ptr > entry->ptr ||
10470 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10471 p = &(*p)->rb_right;
10473 if (is_block_group)
10474 entry->bg_extent_count++;
10475 spin_unlock(&fs_info->swapfile_pins_lock);
10480 rb_link_node(&sp->node, parent, p);
10481 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10482 spin_unlock(&fs_info->swapfile_pins_lock);
10486 /* Free all of the entries pinned by this swapfile. */
10487 static void btrfs_free_swapfile_pins(struct inode *inode)
10489 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10490 struct btrfs_swapfile_pin *sp;
10491 struct rb_node *node, *next;
10493 spin_lock(&fs_info->swapfile_pins_lock);
10494 node = rb_first(&fs_info->swapfile_pins);
10496 next = rb_next(node);
10497 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10498 if (sp->inode == inode) {
10499 rb_erase(&sp->node, &fs_info->swapfile_pins);
10500 if (sp->is_block_group) {
10501 btrfs_dec_block_group_swap_extents(sp->ptr,
10502 sp->bg_extent_count);
10503 btrfs_put_block_group(sp->ptr);
10509 spin_unlock(&fs_info->swapfile_pins_lock);
10512 struct btrfs_swap_info {
10518 unsigned long nr_pages;
10522 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10523 struct btrfs_swap_info *bsi)
10525 unsigned long nr_pages;
10526 unsigned long max_pages;
10527 u64 first_ppage, first_ppage_reported, next_ppage;
10531 * Our swapfile may have had its size extended after the swap header was
10532 * written. In that case activating the swapfile should not go beyond
10533 * the max size set in the swap header.
10535 if (bsi->nr_pages >= sis->max)
10538 max_pages = sis->max - bsi->nr_pages;
10539 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10540 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10542 if (first_ppage >= next_ppage)
10544 nr_pages = next_ppage - first_ppage;
10545 nr_pages = min(nr_pages, max_pages);
10547 first_ppage_reported = first_ppage;
10548 if (bsi->start == 0)
10549 first_ppage_reported++;
10550 if (bsi->lowest_ppage > first_ppage_reported)
10551 bsi->lowest_ppage = first_ppage_reported;
10552 if (bsi->highest_ppage < (next_ppage - 1))
10553 bsi->highest_ppage = next_ppage - 1;
10555 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10558 bsi->nr_extents += ret;
10559 bsi->nr_pages += nr_pages;
10563 static void btrfs_swap_deactivate(struct file *file)
10565 struct inode *inode = file_inode(file);
10567 btrfs_free_swapfile_pins(inode);
10568 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10571 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10574 struct inode *inode = file_inode(file);
10575 struct btrfs_root *root = BTRFS_I(inode)->root;
10576 struct btrfs_fs_info *fs_info = root->fs_info;
10577 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10578 struct extent_state *cached_state = NULL;
10579 struct extent_map *em = NULL;
10580 struct btrfs_device *device = NULL;
10581 struct btrfs_swap_info bsi = {
10582 .lowest_ppage = (sector_t)-1ULL,
10589 * If the swap file was just created, make sure delalloc is done. If the
10590 * file changes again after this, the user is doing something stupid and
10591 * we don't really care.
10593 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10598 * The inode is locked, so these flags won't change after we check them.
10600 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10601 btrfs_warn(fs_info, "swapfile must not be compressed");
10604 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10605 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10608 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10609 btrfs_warn(fs_info, "swapfile must not be checksummed");
10614 * Balance or device remove/replace/resize can move stuff around from
10615 * under us. The exclop protection makes sure they aren't running/won't
10616 * run concurrently while we are mapping the swap extents, and
10617 * fs_info->swapfile_pins prevents them from running while the swap
10618 * file is active and moving the extents. Note that this also prevents
10619 * a concurrent device add which isn't actually necessary, but it's not
10620 * really worth the trouble to allow it.
10622 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10623 btrfs_warn(fs_info,
10624 "cannot activate swapfile while exclusive operation is running");
10629 * Prevent snapshot creation while we are activating the swap file.
10630 * We do not want to race with snapshot creation. If snapshot creation
10631 * already started before we bumped nr_swapfiles from 0 to 1 and
10632 * completes before the first write into the swap file after it is
10633 * activated, than that write would fallback to COW.
10635 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10636 btrfs_exclop_finish(fs_info);
10637 btrfs_warn(fs_info,
10638 "cannot activate swapfile because snapshot creation is in progress");
10642 * Snapshots can create extents which require COW even if NODATACOW is
10643 * set. We use this counter to prevent snapshots. We must increment it
10644 * before walking the extents because we don't want a concurrent
10645 * snapshot to run after we've already checked the extents.
10647 * It is possible that subvolume is marked for deletion but still not
10648 * removed yet. To prevent this race, we check the root status before
10649 * activating the swapfile.
10651 spin_lock(&root->root_item_lock);
10652 if (btrfs_root_dead(root)) {
10653 spin_unlock(&root->root_item_lock);
10655 btrfs_exclop_finish(fs_info);
10656 btrfs_warn(fs_info,
10657 "cannot activate swapfile because subvolume %llu is being deleted",
10658 root->root_key.objectid);
10661 atomic_inc(&root->nr_swapfiles);
10662 spin_unlock(&root->root_item_lock);
10664 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10666 lock_extent(io_tree, 0, isize - 1, &cached_state);
10668 while (start < isize) {
10669 u64 logical_block_start, physical_block_start;
10670 struct btrfs_block_group *bg;
10671 u64 len = isize - start;
10673 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10679 if (em->block_start == EXTENT_MAP_HOLE) {
10680 btrfs_warn(fs_info, "swapfile must not have holes");
10684 if (em->block_start == EXTENT_MAP_INLINE) {
10686 * It's unlikely we'll ever actually find ourselves
10687 * here, as a file small enough to fit inline won't be
10688 * big enough to store more than the swap header, but in
10689 * case something changes in the future, let's catch it
10690 * here rather than later.
10692 btrfs_warn(fs_info, "swapfile must not be inline");
10696 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10697 btrfs_warn(fs_info, "swapfile must not be compressed");
10702 logical_block_start = em->block_start + (start - em->start);
10703 len = min(len, em->len - (start - em->start));
10704 free_extent_map(em);
10707 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10713 btrfs_warn(fs_info,
10714 "swapfile must not be copy-on-write");
10719 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10725 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10726 btrfs_warn(fs_info,
10727 "swapfile must have single data profile");
10732 if (device == NULL) {
10733 device = em->map_lookup->stripes[0].dev;
10734 ret = btrfs_add_swapfile_pin(inode, device, false);
10739 } else if (device != em->map_lookup->stripes[0].dev) {
10740 btrfs_warn(fs_info, "swapfile must be on one device");
10745 physical_block_start = (em->map_lookup->stripes[0].physical +
10746 (logical_block_start - em->start));
10747 len = min(len, em->len - (logical_block_start - em->start));
10748 free_extent_map(em);
10751 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10753 btrfs_warn(fs_info,
10754 "could not find block group containing swapfile");
10759 if (!btrfs_inc_block_group_swap_extents(bg)) {
10760 btrfs_warn(fs_info,
10761 "block group for swapfile at %llu is read-only%s",
10763 atomic_read(&fs_info->scrubs_running) ?
10764 " (scrub running)" : "");
10765 btrfs_put_block_group(bg);
10770 ret = btrfs_add_swapfile_pin(inode, bg, true);
10772 btrfs_put_block_group(bg);
10779 if (bsi.block_len &&
10780 bsi.block_start + bsi.block_len == physical_block_start) {
10781 bsi.block_len += len;
10783 if (bsi.block_len) {
10784 ret = btrfs_add_swap_extent(sis, &bsi);
10789 bsi.block_start = physical_block_start;
10790 bsi.block_len = len;
10797 ret = btrfs_add_swap_extent(sis, &bsi);
10800 if (!IS_ERR_OR_NULL(em))
10801 free_extent_map(em);
10803 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10806 btrfs_swap_deactivate(file);
10808 btrfs_drew_write_unlock(&root->snapshot_lock);
10810 btrfs_exclop_finish(fs_info);
10816 sis->bdev = device->bdev;
10817 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10818 sis->max = bsi.nr_pages;
10819 sis->pages = bsi.nr_pages - 1;
10820 sis->highest_bit = bsi.nr_pages - 1;
10821 return bsi.nr_extents;
10824 static void btrfs_swap_deactivate(struct file *file)
10828 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10831 return -EOPNOTSUPP;
10836 * Update the number of bytes used in the VFS' inode. When we replace extents in
10837 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10838 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10839 * always get a correct value.
10841 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10842 const u64 add_bytes,
10843 const u64 del_bytes)
10845 if (add_bytes == del_bytes)
10848 spin_lock(&inode->lock);
10850 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10852 inode_add_bytes(&inode->vfs_inode, add_bytes);
10853 spin_unlock(&inode->lock);
10857 * Verify that there are no ordered extents for a given file range.
10859 * @inode: The target inode.
10860 * @start: Start offset of the file range, should be sector size aligned.
10861 * @end: End offset (inclusive) of the file range, its value +1 should be
10862 * sector size aligned.
10864 * This should typically be used for cases where we locked an inode's VFS lock in
10865 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10866 * we have flushed all delalloc in the range, we have waited for all ordered
10867 * extents in the range to complete and finally we have locked the file range in
10868 * the inode's io_tree.
10870 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
10872 struct btrfs_root *root = inode->root;
10873 struct btrfs_ordered_extent *ordered;
10875 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
10878 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
10880 btrfs_err(root->fs_info,
10881 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10882 start, end, btrfs_ino(inode), root->root_key.objectid,
10883 ordered->file_offset,
10884 ordered->file_offset + ordered->num_bytes - 1);
10885 btrfs_put_ordered_extent(ordered);
10888 ASSERT(ordered == NULL);
10891 static const struct inode_operations btrfs_dir_inode_operations = {
10892 .getattr = btrfs_getattr,
10893 .lookup = btrfs_lookup,
10894 .create = btrfs_create,
10895 .unlink = btrfs_unlink,
10896 .link = btrfs_link,
10897 .mkdir = btrfs_mkdir,
10898 .rmdir = btrfs_rmdir,
10899 .rename = btrfs_rename2,
10900 .symlink = btrfs_symlink,
10901 .setattr = btrfs_setattr,
10902 .mknod = btrfs_mknod,
10903 .listxattr = btrfs_listxattr,
10904 .permission = btrfs_permission,
10905 .get_inode_acl = btrfs_get_acl,
10906 .set_acl = btrfs_set_acl,
10907 .update_time = btrfs_update_time,
10908 .tmpfile = btrfs_tmpfile,
10909 .fileattr_get = btrfs_fileattr_get,
10910 .fileattr_set = btrfs_fileattr_set,
10913 static const struct file_operations btrfs_dir_file_operations = {
10914 .llseek = generic_file_llseek,
10915 .read = generic_read_dir,
10916 .iterate_shared = btrfs_real_readdir,
10917 .open = btrfs_opendir,
10918 .unlocked_ioctl = btrfs_ioctl,
10919 #ifdef CONFIG_COMPAT
10920 .compat_ioctl = btrfs_compat_ioctl,
10922 .release = btrfs_release_file,
10923 .fsync = btrfs_sync_file,
10927 * btrfs doesn't support the bmap operation because swapfiles
10928 * use bmap to make a mapping of extents in the file. They assume
10929 * these extents won't change over the life of the file and they
10930 * use the bmap result to do IO directly to the drive.
10932 * the btrfs bmap call would return logical addresses that aren't
10933 * suitable for IO and they also will change frequently as COW
10934 * operations happen. So, swapfile + btrfs == corruption.
10936 * For now we're avoiding this by dropping bmap.
10938 static const struct address_space_operations btrfs_aops = {
10939 .read_folio = btrfs_read_folio,
10940 .writepages = btrfs_writepages,
10941 .readahead = btrfs_readahead,
10942 .invalidate_folio = btrfs_invalidate_folio,
10943 .release_folio = btrfs_release_folio,
10944 .migrate_folio = btrfs_migrate_folio,
10945 .dirty_folio = filemap_dirty_folio,
10946 .error_remove_page = generic_error_remove_page,
10947 .swap_activate = btrfs_swap_activate,
10948 .swap_deactivate = btrfs_swap_deactivate,
10951 static const struct inode_operations btrfs_file_inode_operations = {
10952 .getattr = btrfs_getattr,
10953 .setattr = btrfs_setattr,
10954 .listxattr = btrfs_listxattr,
10955 .permission = btrfs_permission,
10956 .fiemap = btrfs_fiemap,
10957 .get_inode_acl = btrfs_get_acl,
10958 .set_acl = btrfs_set_acl,
10959 .update_time = btrfs_update_time,
10960 .fileattr_get = btrfs_fileattr_get,
10961 .fileattr_set = btrfs_fileattr_set,
10963 static const struct inode_operations btrfs_special_inode_operations = {
10964 .getattr = btrfs_getattr,
10965 .setattr = btrfs_setattr,
10966 .permission = btrfs_permission,
10967 .listxattr = btrfs_listxattr,
10968 .get_inode_acl = btrfs_get_acl,
10969 .set_acl = btrfs_set_acl,
10970 .update_time = btrfs_update_time,
10972 static const struct inode_operations btrfs_symlink_inode_operations = {
10973 .get_link = page_get_link,
10974 .getattr = btrfs_getattr,
10975 .setattr = btrfs_setattr,
10976 .permission = btrfs_permission,
10977 .listxattr = btrfs_listxattr,
10978 .update_time = btrfs_update_time,
10981 const struct dentry_operations btrfs_dentry_operations = {
10982 .d_delete = btrfs_dentry_delete,