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, int *page_started,
131 unsigned long *nr_written, u64 *done_offset,
132 bool keep_locked, bool no_inline);
133 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
134 u64 len, u64 orig_start, u64 block_start,
135 u64 block_len, u64 orig_block_len,
136 u64 ram_bytes, int compress_type,
139 static int data_reloc_print_warning_inode(u64 inum, u64 offset, u64 num_bytes,
140 u64 root, void *warn_ctx)
142 struct data_reloc_warn *warn = warn_ctx;
143 struct btrfs_fs_info *fs_info = warn->fs_info;
144 struct extent_buffer *eb;
145 struct btrfs_inode_item *inode_item;
146 struct inode_fs_paths *ipath = NULL;
147 struct btrfs_root *local_root;
148 struct btrfs_key key;
149 unsigned int nofs_flag;
153 local_root = btrfs_get_fs_root(fs_info, root, true);
154 if (IS_ERR(local_root)) {
155 ret = PTR_ERR(local_root);
159 /* This makes the path point to (inum INODE_ITEM ioff). */
161 key.type = BTRFS_INODE_ITEM_KEY;
164 ret = btrfs_search_slot(NULL, local_root, &key, &warn->path, 0, 0);
166 btrfs_put_root(local_root);
167 btrfs_release_path(&warn->path);
171 eb = warn->path.nodes[0];
172 inode_item = btrfs_item_ptr(eb, warn->path.slots[0], struct btrfs_inode_item);
173 nlink = btrfs_inode_nlink(eb, inode_item);
174 btrfs_release_path(&warn->path);
176 nofs_flag = memalloc_nofs_save();
177 ipath = init_ipath(4096, local_root, &warn->path);
178 memalloc_nofs_restore(nofs_flag);
180 btrfs_put_root(local_root);
181 ret = PTR_ERR(ipath);
184 * -ENOMEM, not a critical error, just output an generic error
188 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
189 warn->logical, warn->mirror_num, root, inum, offset);
192 ret = paths_from_inode(inum, ipath);
197 * We deliberately ignore the bit ipath might have been too small to
198 * hold all of the paths here
200 for (int i = 0; i < ipath->fspath->elem_cnt; i++) {
202 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
203 warn->logical, warn->mirror_num, root, inum, offset,
204 fs_info->sectorsize, nlink,
205 (char *)(unsigned long)ipath->fspath->val[i]);
208 btrfs_put_root(local_root);
214 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
215 warn->logical, warn->mirror_num, root, inum, offset, ret);
222 * Do extra user-friendly error output (e.g. lookup all the affected files).
224 * Return true if we succeeded doing the backref lookup.
225 * Return false if such lookup failed, and has to fallback to the old error message.
227 static void print_data_reloc_error(const struct btrfs_inode *inode, u64 file_off,
228 const u8 *csum, const u8 *csum_expected,
231 struct btrfs_fs_info *fs_info = inode->root->fs_info;
232 struct btrfs_path path = { 0 };
233 struct btrfs_key found_key = { 0 };
234 struct extent_buffer *eb;
235 struct btrfs_extent_item *ei;
236 const u32 csum_size = fs_info->csum_size;
242 mutex_lock(&fs_info->reloc_mutex);
243 logical = btrfs_get_reloc_bg_bytenr(fs_info);
244 mutex_unlock(&fs_info->reloc_mutex);
246 if (logical == U64_MAX) {
247 btrfs_warn_rl(fs_info, "has data reloc tree but no running relocation");
248 btrfs_warn_rl(fs_info,
249 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
250 inode->root->root_key.objectid, btrfs_ino(inode), file_off,
251 CSUM_FMT_VALUE(csum_size, csum),
252 CSUM_FMT_VALUE(csum_size, csum_expected),
258 btrfs_warn_rl(fs_info,
259 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
260 inode->root->root_key.objectid,
261 btrfs_ino(inode), file_off, logical,
262 CSUM_FMT_VALUE(csum_size, csum),
263 CSUM_FMT_VALUE(csum_size, csum_expected),
266 ret = extent_from_logical(fs_info, logical, &path, &found_key, &flags);
268 btrfs_err_rl(fs_info, "failed to lookup extent item for logical %llu: %d",
273 ei = btrfs_item_ptr(eb, path.slots[0], struct btrfs_extent_item);
274 item_size = btrfs_item_size(eb, path.slots[0]);
275 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
276 unsigned long ptr = 0;
281 ret = tree_backref_for_extent(&ptr, eb, &found_key, ei,
282 item_size, &ref_root,
285 btrfs_warn_rl(fs_info,
286 "failed to resolve tree backref for logical %llu: %d",
293 btrfs_warn_rl(fs_info,
294 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
296 (ref_level ? "node" : "leaf"),
297 ref_level, ref_root);
299 btrfs_release_path(&path);
301 struct btrfs_backref_walk_ctx ctx = { 0 };
302 struct data_reloc_warn reloc_warn = { 0 };
304 btrfs_release_path(&path);
306 ctx.bytenr = found_key.objectid;
307 ctx.extent_item_pos = logical - found_key.objectid;
308 ctx.fs_info = fs_info;
310 reloc_warn.logical = logical;
311 reloc_warn.extent_item_size = found_key.offset;
312 reloc_warn.mirror_num = mirror_num;
313 reloc_warn.fs_info = fs_info;
315 iterate_extent_inodes(&ctx, true,
316 data_reloc_print_warning_inode, &reloc_warn);
320 static void __cold btrfs_print_data_csum_error(struct btrfs_inode *inode,
321 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
323 struct btrfs_root *root = inode->root;
324 const u32 csum_size = root->fs_info->csum_size;
326 /* For data reloc tree, it's better to do a backref lookup instead. */
327 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
328 return print_data_reloc_error(inode, logical_start, csum,
329 csum_expected, mirror_num);
331 /* Output without objectid, which is more meaningful */
332 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID) {
333 btrfs_warn_rl(root->fs_info,
334 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
335 root->root_key.objectid, btrfs_ino(inode),
337 CSUM_FMT_VALUE(csum_size, csum),
338 CSUM_FMT_VALUE(csum_size, csum_expected),
341 btrfs_warn_rl(root->fs_info,
342 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
343 root->root_key.objectid, btrfs_ino(inode),
345 CSUM_FMT_VALUE(csum_size, csum),
346 CSUM_FMT_VALUE(csum_size, csum_expected),
352 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
354 * ilock_flags can have the following bit set:
356 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
357 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
359 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
361 int btrfs_inode_lock(struct btrfs_inode *inode, unsigned int ilock_flags)
363 if (ilock_flags & BTRFS_ILOCK_SHARED) {
364 if (ilock_flags & BTRFS_ILOCK_TRY) {
365 if (!inode_trylock_shared(&inode->vfs_inode))
370 inode_lock_shared(&inode->vfs_inode);
372 if (ilock_flags & BTRFS_ILOCK_TRY) {
373 if (!inode_trylock(&inode->vfs_inode))
378 inode_lock(&inode->vfs_inode);
380 if (ilock_flags & BTRFS_ILOCK_MMAP)
381 down_write(&inode->i_mmap_lock);
386 * btrfs_inode_unlock - unock inode i_rwsem
388 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
389 * to decide whether the lock acquired is shared or exclusive.
391 void btrfs_inode_unlock(struct btrfs_inode *inode, unsigned int ilock_flags)
393 if (ilock_flags & BTRFS_ILOCK_MMAP)
394 up_write(&inode->i_mmap_lock);
395 if (ilock_flags & BTRFS_ILOCK_SHARED)
396 inode_unlock_shared(&inode->vfs_inode);
398 inode_unlock(&inode->vfs_inode);
402 * Cleanup all submitted ordered extents in specified range to handle errors
403 * from the btrfs_run_delalloc_range() callback.
405 * NOTE: caller must ensure that when an error happens, it can not call
406 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
407 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
408 * to be released, which we want to happen only when finishing the ordered
409 * extent (btrfs_finish_ordered_io()).
411 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode *inode,
412 struct page *locked_page,
413 u64 offset, u64 bytes)
415 unsigned long index = offset >> PAGE_SHIFT;
416 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
417 u64 page_start = 0, page_end = 0;
421 page_start = page_offset(locked_page);
422 page_end = page_start + PAGE_SIZE - 1;
425 while (index <= end_index) {
427 * For locked page, we will call btrfs_mark_ordered_io_finished
428 * through btrfs_mark_ordered_io_finished() on it
429 * in run_delalloc_range() for the error handling, which will
430 * clear page Ordered and run the ordered extent accounting.
432 * Here we can't just clear the Ordered bit, or
433 * btrfs_mark_ordered_io_finished() would skip the accounting
434 * for the page range, and the ordered extent will never finish.
436 if (locked_page && index == (page_start >> PAGE_SHIFT)) {
440 page = find_get_page(inode->vfs_inode.i_mapping, index);
446 * Here we just clear all Ordered bits for every page in the
447 * range, then btrfs_mark_ordered_io_finished() will handle
448 * the ordered extent accounting for the range.
450 btrfs_page_clamp_clear_ordered(inode->root->fs_info, page,
456 /* The locked page covers the full range, nothing needs to be done */
457 if (bytes + offset <= page_start + PAGE_SIZE)
460 * In case this page belongs to the delalloc range being
461 * instantiated then skip it, since the first page of a range is
462 * going to be properly cleaned up by the caller of
465 if (page_start >= offset && page_end <= (offset + bytes - 1)) {
466 bytes = offset + bytes - page_offset(locked_page) - PAGE_SIZE;
467 offset = page_offset(locked_page) + PAGE_SIZE;
471 return btrfs_mark_ordered_io_finished(inode, NULL, offset, bytes, false);
474 static int btrfs_dirty_inode(struct btrfs_inode *inode);
476 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
477 struct btrfs_new_inode_args *args)
481 if (args->default_acl) {
482 err = __btrfs_set_acl(trans, args->inode, args->default_acl,
488 err = __btrfs_set_acl(trans, args->inode, args->acl, ACL_TYPE_ACCESS);
492 if (!args->default_acl && !args->acl)
493 cache_no_acl(args->inode);
494 return btrfs_xattr_security_init(trans, args->inode, args->dir,
495 &args->dentry->d_name);
499 * this does all the hard work for inserting an inline extent into
500 * the btree. The caller should have done a btrfs_drop_extents so that
501 * no overlapping inline items exist in the btree
503 static int insert_inline_extent(struct btrfs_trans_handle *trans,
504 struct btrfs_path *path,
505 struct btrfs_inode *inode, bool extent_inserted,
506 size_t size, size_t compressed_size,
508 struct page **compressed_pages,
511 struct btrfs_root *root = inode->root;
512 struct extent_buffer *leaf;
513 struct page *page = NULL;
516 struct btrfs_file_extent_item *ei;
518 size_t cur_size = size;
521 ASSERT((compressed_size > 0 && compressed_pages) ||
522 (compressed_size == 0 && !compressed_pages));
524 if (compressed_size && compressed_pages)
525 cur_size = compressed_size;
527 if (!extent_inserted) {
528 struct btrfs_key key;
531 key.objectid = btrfs_ino(inode);
533 key.type = BTRFS_EXTENT_DATA_KEY;
535 datasize = btrfs_file_extent_calc_inline_size(cur_size);
536 ret = btrfs_insert_empty_item(trans, root, path, &key,
541 leaf = path->nodes[0];
542 ei = btrfs_item_ptr(leaf, path->slots[0],
543 struct btrfs_file_extent_item);
544 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
545 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
546 btrfs_set_file_extent_encryption(leaf, ei, 0);
547 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
548 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
549 ptr = btrfs_file_extent_inline_start(ei);
551 if (compress_type != BTRFS_COMPRESS_NONE) {
554 while (compressed_size > 0) {
555 cpage = compressed_pages[i];
556 cur_size = min_t(unsigned long, compressed_size,
559 kaddr = kmap_local_page(cpage);
560 write_extent_buffer(leaf, kaddr, ptr, cur_size);
565 compressed_size -= cur_size;
567 btrfs_set_file_extent_compression(leaf, ei,
570 page = find_get_page(inode->vfs_inode.i_mapping, 0);
571 btrfs_set_file_extent_compression(leaf, ei, 0);
572 kaddr = kmap_local_page(page);
573 write_extent_buffer(leaf, kaddr, ptr, size);
577 btrfs_mark_buffer_dirty(leaf);
578 btrfs_release_path(path);
581 * We align size to sectorsize for inline extents just for simplicity
584 ret = btrfs_inode_set_file_extent_range(inode, 0,
585 ALIGN(size, root->fs_info->sectorsize));
590 * We're an inline extent, so nobody can extend the file past i_size
591 * without locking a page we already have locked.
593 * We must do any i_size and inode updates before we unlock the pages.
594 * Otherwise we could end up racing with unlink.
596 i_size = i_size_read(&inode->vfs_inode);
597 if (update_i_size && size > i_size) {
598 i_size_write(&inode->vfs_inode, size);
601 inode->disk_i_size = i_size;
609 * conditionally insert an inline extent into the file. This
610 * does the checks required to make sure the data is small enough
611 * to fit as an inline extent.
613 static noinline int cow_file_range_inline(struct btrfs_inode *inode, u64 size,
614 size_t compressed_size,
616 struct page **compressed_pages,
619 struct btrfs_drop_extents_args drop_args = { 0 };
620 struct btrfs_root *root = inode->root;
621 struct btrfs_fs_info *fs_info = root->fs_info;
622 struct btrfs_trans_handle *trans;
623 u64 data_len = (compressed_size ?: size);
625 struct btrfs_path *path;
628 * We can create an inline extent if it ends at or beyond the current
629 * i_size, is no larger than a sector (decompressed), and the (possibly
630 * compressed) data fits in a leaf and the configured maximum inline
633 if (size < i_size_read(&inode->vfs_inode) ||
634 size > fs_info->sectorsize ||
635 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
636 data_len > fs_info->max_inline)
639 path = btrfs_alloc_path();
643 trans = btrfs_join_transaction(root);
645 btrfs_free_path(path);
646 return PTR_ERR(trans);
648 trans->block_rsv = &inode->block_rsv;
650 drop_args.path = path;
652 drop_args.end = fs_info->sectorsize;
653 drop_args.drop_cache = true;
654 drop_args.replace_extent = true;
655 drop_args.extent_item_size = btrfs_file_extent_calc_inline_size(data_len);
656 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
658 btrfs_abort_transaction(trans, ret);
662 ret = insert_inline_extent(trans, path, inode, drop_args.extent_inserted,
663 size, compressed_size, compress_type,
664 compressed_pages, update_i_size);
665 if (ret && ret != -ENOSPC) {
666 btrfs_abort_transaction(trans, ret);
668 } else if (ret == -ENOSPC) {
673 btrfs_update_inode_bytes(inode, size, drop_args.bytes_found);
674 ret = btrfs_update_inode(trans, root, inode);
675 if (ret && ret != -ENOSPC) {
676 btrfs_abort_transaction(trans, ret);
678 } else if (ret == -ENOSPC) {
683 btrfs_set_inode_full_sync(inode);
686 * Don't forget to free the reserved space, as for inlined extent
687 * it won't count as data extent, free them directly here.
688 * And at reserve time, it's always aligned to page size, so
689 * just free one page here.
691 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
692 btrfs_free_path(path);
693 btrfs_end_transaction(trans);
697 struct async_extent {
702 unsigned long nr_pages;
704 struct list_head list;
708 struct btrfs_inode *inode;
709 struct page *locked_page;
712 blk_opf_t write_flags;
713 struct list_head extents;
714 struct cgroup_subsys_state *blkcg_css;
715 struct btrfs_work work;
716 struct async_cow *async_cow;
721 struct async_chunk chunks[];
724 static noinline int add_async_extent(struct async_chunk *cow,
725 u64 start, u64 ram_size,
728 unsigned long nr_pages,
731 struct async_extent *async_extent;
733 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
734 BUG_ON(!async_extent); /* -ENOMEM */
735 async_extent->start = start;
736 async_extent->ram_size = ram_size;
737 async_extent->compressed_size = compressed_size;
738 async_extent->pages = pages;
739 async_extent->nr_pages = nr_pages;
740 async_extent->compress_type = compress_type;
741 list_add_tail(&async_extent->list, &cow->extents);
746 * Check if the inode needs to be submitted to compression, based on mount
747 * options, defragmentation, properties or heuristics.
749 static inline int inode_need_compress(struct btrfs_inode *inode, u64 start,
752 struct btrfs_fs_info *fs_info = inode->root->fs_info;
754 if (!btrfs_inode_can_compress(inode)) {
755 WARN(IS_ENABLED(CONFIG_BTRFS_DEBUG),
756 KERN_ERR "BTRFS: unexpected compression for ino %llu\n",
761 * Special check for subpage.
763 * We lock the full page then run each delalloc range in the page, thus
764 * for the following case, we will hit some subpage specific corner case:
767 * | |///////| |///////|
770 * In above case, both range A and range B will try to unlock the full
771 * page [0, 64K), causing the one finished later will have page
772 * unlocked already, triggering various page lock requirement BUG_ON()s.
774 * So here we add an artificial limit that subpage compression can only
775 * if the range is fully page aligned.
777 * In theory we only need to ensure the first page is fully covered, but
778 * the tailing partial page will be locked until the full compression
779 * finishes, delaying the write of other range.
781 * TODO: Make btrfs_run_delalloc_range() to lock all delalloc range
782 * first to prevent any submitted async extent to unlock the full page.
783 * By this, we can ensure for subpage case that only the last async_cow
784 * will unlock the full page.
786 if (fs_info->sectorsize < PAGE_SIZE) {
787 if (!PAGE_ALIGNED(start) ||
788 !PAGE_ALIGNED(end + 1))
793 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
796 if (inode->defrag_compress)
798 /* bad compression ratios */
799 if (inode->flags & BTRFS_INODE_NOCOMPRESS)
801 if (btrfs_test_opt(fs_info, COMPRESS) ||
802 inode->flags & BTRFS_INODE_COMPRESS ||
803 inode->prop_compress)
804 return btrfs_compress_heuristic(&inode->vfs_inode, start, end);
808 static inline void inode_should_defrag(struct btrfs_inode *inode,
809 u64 start, u64 end, u64 num_bytes, u32 small_write)
811 /* If this is a small write inside eof, kick off a defrag */
812 if (num_bytes < small_write &&
813 (start > 0 || end + 1 < inode->disk_i_size))
814 btrfs_add_inode_defrag(NULL, inode, small_write);
818 * we create compressed extents in two phases. The first
819 * phase compresses a range of pages that have already been
820 * locked (both pages and state bits are locked).
822 * This is done inside an ordered work queue, and the compression
823 * is spread across many cpus. The actual IO submission is step
824 * two, and the ordered work queue takes care of making sure that
825 * happens in the same order things were put onto the queue by
826 * writepages and friends.
828 * If this code finds it can't get good compression, it puts an
829 * entry onto the work queue to write the uncompressed bytes. This
830 * makes sure that both compressed inodes and uncompressed inodes
831 * are written in the same order that the flusher thread sent them
834 static noinline int compress_file_range(struct async_chunk *async_chunk)
836 struct btrfs_inode *inode = async_chunk->inode;
837 struct btrfs_fs_info *fs_info = inode->root->fs_info;
838 struct address_space *mapping = inode->vfs_inode.i_mapping;
839 u64 blocksize = fs_info->sectorsize;
840 u64 start = async_chunk->start;
841 u64 end = async_chunk->end;
845 struct page **pages = NULL;
846 unsigned long nr_pages;
847 unsigned long total_compressed = 0;
848 unsigned long total_in = 0;
851 int compress_type = fs_info->compress_type;
852 int compressed_extents = 0;
855 inode_should_defrag(inode, start, end, end - start + 1, SZ_16K);
858 * We need to save i_size before now because it could change in between
859 * us evaluating the size and assigning it. This is because we lock and
860 * unlock the page in truncate and fallocate, and then modify the i_size
863 * The barriers are to emulate READ_ONCE, remove that once i_size_read
867 i_size = i_size_read(&inode->vfs_inode);
869 actual_end = min_t(u64, i_size, end + 1);
872 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
873 nr_pages = min_t(unsigned long, nr_pages, BTRFS_MAX_COMPRESSED_PAGES);
876 * we don't want to send crud past the end of i_size through
877 * compression, that's just a waste of CPU time. So, if the
878 * end of the file is before the start of our current
879 * requested range of bytes, we bail out to the uncompressed
880 * cleanup code that can deal with all of this.
882 * It isn't really the fastest way to fix things, but this is a
883 * very uncommon corner.
885 if (actual_end <= start)
886 goto cleanup_and_bail_uncompressed;
888 total_compressed = actual_end - start;
891 * Skip compression for a small file range(<=blocksize) that
892 * isn't an inline extent, since it doesn't save disk space at all.
894 if (total_compressed <= blocksize &&
895 (start > 0 || end + 1 < inode->disk_i_size))
896 goto cleanup_and_bail_uncompressed;
899 * For subpage case, we require full page alignment for the sector
901 * Thus we must also check against @actual_end, not just @end.
903 if (blocksize < PAGE_SIZE) {
904 if (!PAGE_ALIGNED(start) ||
905 !PAGE_ALIGNED(round_up(actual_end, blocksize)))
906 goto cleanup_and_bail_uncompressed;
909 total_compressed = min_t(unsigned long, total_compressed,
910 BTRFS_MAX_UNCOMPRESSED);
915 * we do compression for mount -o compress and when the
916 * inode has not been flagged as nocompress. This flag can
917 * change at any time if we discover bad compression ratios.
919 if (inode_need_compress(inode, start, end)) {
921 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
923 /* just bail out to the uncompressed code */
928 if (inode->defrag_compress)
929 compress_type = inode->defrag_compress;
930 else if (inode->prop_compress)
931 compress_type = inode->prop_compress;
934 * we need to call clear_page_dirty_for_io on each
935 * page in the range. Otherwise applications with the file
936 * mmap'd can wander in and change the page contents while
937 * we are compressing them.
939 * If the compression fails for any reason, we set the pages
940 * dirty again later on.
942 * Note that the remaining part is redirtied, the start pointer
943 * has moved, the end is the original one.
946 extent_range_clear_dirty_for_io(&inode->vfs_inode, start, end);
950 /* Compression level is applied here and only here */
951 ret = btrfs_compress_pages(
952 compress_type | (fs_info->compress_level << 4),
960 unsigned long offset = offset_in_page(total_compressed);
961 struct page *page = pages[nr_pages - 1];
963 /* zero the tail end of the last page, we might be
964 * sending it down to disk
967 memzero_page(page, offset, PAGE_SIZE - offset);
973 * Check cow_file_range() for why we don't even try to create inline
974 * extent for subpage case.
976 if (start == 0 && fs_info->sectorsize == PAGE_SIZE) {
977 /* lets try to make an inline extent */
978 if (ret || total_in < actual_end) {
979 /* we didn't compress the entire range, try
980 * to make an uncompressed inline extent.
982 ret = cow_file_range_inline(inode, actual_end,
983 0, BTRFS_COMPRESS_NONE,
986 /* try making a compressed inline extent */
987 ret = cow_file_range_inline(inode, actual_end,
989 compress_type, pages,
993 unsigned long clear_flags = EXTENT_DELALLOC |
994 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
995 EXTENT_DO_ACCOUNTING;
998 mapping_set_error(mapping, -EIO);
1001 * inline extent creation worked or returned error,
1002 * we don't need to create any more async work items.
1003 * Unlock and free up our temp pages.
1005 * We use DO_ACCOUNTING here because we need the
1006 * delalloc_release_metadata to be done _after_ we drop
1007 * our outstanding extent for clearing delalloc for this
1010 extent_clear_unlock_delalloc(inode, start, end,
1014 PAGE_START_WRITEBACK |
1015 PAGE_END_WRITEBACK);
1018 * Ensure we only free the compressed pages if we have
1019 * them allocated, as we can still reach here with
1020 * inode_need_compress() == false.
1023 for (i = 0; i < nr_pages; i++) {
1024 WARN_ON(pages[i]->mapping);
1033 if (will_compress) {
1035 * we aren't doing an inline extent round the compressed size
1036 * up to a block size boundary so the allocator does sane
1039 total_compressed = ALIGN(total_compressed, blocksize);
1042 * one last check to make sure the compression is really a
1043 * win, compare the page count read with the blocks on disk,
1044 * compression must free at least one sector size
1046 total_in = round_up(total_in, fs_info->sectorsize);
1047 if (total_compressed + blocksize <= total_in) {
1048 compressed_extents++;
1051 * The async work queues will take care of doing actual
1052 * allocation on disk for these compressed pages, and
1053 * will submit them to the elevator.
1055 add_async_extent(async_chunk, start, total_in,
1056 total_compressed, pages, nr_pages,
1059 if (start + total_in < end) {
1065 return compressed_extents;
1070 * the compression code ran but failed to make things smaller,
1071 * free any pages it allocated and our page pointer array
1073 for (i = 0; i < nr_pages; i++) {
1074 WARN_ON(pages[i]->mapping);
1079 total_compressed = 0;
1082 /* flag the file so we don't compress in the future */
1083 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
1084 !(inode->prop_compress)) {
1085 inode->flags |= BTRFS_INODE_NOCOMPRESS;
1088 cleanup_and_bail_uncompressed:
1090 * No compression, but we still need to write the pages in the file
1091 * we've been given so far. redirty the locked page if it corresponds
1092 * to our extent and set things up for the async work queue to run
1093 * cow_file_range to do the normal delalloc dance.
1095 if (async_chunk->locked_page &&
1096 (page_offset(async_chunk->locked_page) >= start &&
1097 page_offset(async_chunk->locked_page)) <= end) {
1098 __set_page_dirty_nobuffers(async_chunk->locked_page);
1099 /* unlocked later on in the async handlers */
1103 extent_range_redirty_for_io(&inode->vfs_inode, start, end);
1104 add_async_extent(async_chunk, start, end - start + 1, 0, NULL, 0,
1105 BTRFS_COMPRESS_NONE);
1106 compressed_extents++;
1108 return compressed_extents;
1111 static void free_async_extent_pages(struct async_extent *async_extent)
1115 if (!async_extent->pages)
1118 for (i = 0; i < async_extent->nr_pages; i++) {
1119 WARN_ON(async_extent->pages[i]->mapping);
1120 put_page(async_extent->pages[i]);
1122 kfree(async_extent->pages);
1123 async_extent->nr_pages = 0;
1124 async_extent->pages = NULL;
1127 static int submit_uncompressed_range(struct btrfs_inode *inode,
1128 struct async_extent *async_extent,
1129 struct page *locked_page)
1131 u64 start = async_extent->start;
1132 u64 end = async_extent->start + async_extent->ram_size - 1;
1133 unsigned long nr_written = 0;
1134 int page_started = 0;
1136 struct writeback_control wbc = {
1137 .sync_mode = WB_SYNC_ALL,
1138 .range_start = start,
1140 .no_cgroup_owner = 1,
1144 * Call cow_file_range() to run the delalloc range directly, since we
1145 * won't go to NOCOW or async path again.
1147 * Also we call cow_file_range() with @unlock_page == 0, so that we
1148 * can directly submit them without interruption.
1150 ret = cow_file_range(inode, locked_page, start, end, &page_started,
1151 &nr_written, NULL, true, false);
1152 /* Inline extent inserted, page gets unlocked and everything is done */
1157 btrfs_cleanup_ordered_extents(inode, locked_page, start, end - start + 1);
1159 const u64 page_start = page_offset(locked_page);
1161 set_page_writeback(locked_page);
1162 end_page_writeback(locked_page);
1163 btrfs_mark_ordered_io_finished(inode, locked_page,
1164 page_start, PAGE_SIZE,
1166 btrfs_page_clear_uptodate(inode->root->fs_info,
1167 locked_page, page_start,
1169 mapping_set_error(locked_page->mapping, ret);
1170 unlock_page(locked_page);
1175 /* All pages will be unlocked, including @locked_page */
1176 wbc_attach_fdatawrite_inode(&wbc, &inode->vfs_inode);
1177 ret = extent_write_locked_range(&inode->vfs_inode, start, end, &wbc);
1178 wbc_detach_inode(&wbc);
1182 static int submit_one_async_extent(struct btrfs_inode *inode,
1183 struct async_chunk *async_chunk,
1184 struct async_extent *async_extent,
1187 struct extent_io_tree *io_tree = &inode->io_tree;
1188 struct btrfs_root *root = inode->root;
1189 struct btrfs_fs_info *fs_info = root->fs_info;
1190 struct btrfs_ordered_extent *ordered;
1191 struct btrfs_key ins;
1192 struct page *locked_page = NULL;
1193 struct extent_map *em;
1195 u64 start = async_extent->start;
1196 u64 end = async_extent->start + async_extent->ram_size - 1;
1198 if (async_chunk->blkcg_css)
1199 kthread_associate_blkcg(async_chunk->blkcg_css);
1202 * If async_chunk->locked_page is in the async_extent range, we need to
1205 if (async_chunk->locked_page) {
1206 u64 locked_page_start = page_offset(async_chunk->locked_page);
1207 u64 locked_page_end = locked_page_start + PAGE_SIZE - 1;
1209 if (!(start >= locked_page_end || end <= locked_page_start))
1210 locked_page = async_chunk->locked_page;
1212 lock_extent(io_tree, start, end, NULL);
1214 /* We have fall back to uncompressed write */
1215 if (!async_extent->pages) {
1216 ret = submit_uncompressed_range(inode, async_extent, locked_page);
1220 ret = btrfs_reserve_extent(root, async_extent->ram_size,
1221 async_extent->compressed_size,
1222 async_extent->compressed_size,
1223 0, *alloc_hint, &ins, 1, 1);
1226 * Here we used to try again by going back to non-compressed
1227 * path for ENOSPC. But we can't reserve space even for
1228 * compressed size, how could it work for uncompressed size
1229 * which requires larger size? So here we directly go error
1235 /* Here we're doing allocation and writeback of the compressed pages */
1236 em = create_io_em(inode, start,
1237 async_extent->ram_size, /* len */
1238 start, /* orig_start */
1239 ins.objectid, /* block_start */
1240 ins.offset, /* block_len */
1241 ins.offset, /* orig_block_len */
1242 async_extent->ram_size, /* ram_bytes */
1243 async_extent->compress_type,
1244 BTRFS_ORDERED_COMPRESSED);
1247 goto out_free_reserve;
1249 free_extent_map(em);
1251 ordered = btrfs_alloc_ordered_extent(inode, start, /* file_offset */
1252 async_extent->ram_size, /* num_bytes */
1253 async_extent->ram_size, /* ram_bytes */
1254 ins.objectid, /* disk_bytenr */
1255 ins.offset, /* disk_num_bytes */
1257 1 << BTRFS_ORDERED_COMPRESSED,
1258 async_extent->compress_type);
1259 if (IS_ERR(ordered)) {
1260 btrfs_drop_extent_map_range(inode, start, end, false);
1261 ret = PTR_ERR(ordered);
1262 goto out_free_reserve;
1264 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1266 /* Clear dirty, set writeback and unlock the pages. */
1267 extent_clear_unlock_delalloc(inode, start, end,
1268 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
1269 PAGE_UNLOCK | PAGE_START_WRITEBACK);
1270 btrfs_submit_compressed_write(ordered,
1271 async_extent->pages, /* compressed_pages */
1272 async_extent->nr_pages,
1273 async_chunk->write_flags, true);
1274 *alloc_hint = ins.objectid + ins.offset;
1276 if (async_chunk->blkcg_css)
1277 kthread_associate_blkcg(NULL);
1278 kfree(async_extent);
1282 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1283 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1285 mapping_set_error(inode->vfs_inode.i_mapping, -EIO);
1286 extent_clear_unlock_delalloc(inode, start, end,
1287 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
1288 EXTENT_DELALLOC_NEW |
1289 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
1290 PAGE_UNLOCK | PAGE_START_WRITEBACK |
1291 PAGE_END_WRITEBACK);
1292 free_async_extent_pages(async_extent);
1297 * Phase two of compressed writeback. This is the ordered portion of the code,
1298 * which only gets called in the order the work was queued. We walk all the
1299 * async extents created by compress_file_range and send them down to the disk.
1301 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1303 struct btrfs_inode *inode = async_chunk->inode;
1304 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1305 struct async_extent *async_extent;
1309 while (!list_empty(&async_chunk->extents)) {
1313 async_extent = list_entry(async_chunk->extents.next,
1314 struct async_extent, list);
1315 list_del(&async_extent->list);
1316 extent_start = async_extent->start;
1317 ram_size = async_extent->ram_size;
1319 ret = submit_one_async_extent(inode, async_chunk, async_extent,
1321 btrfs_debug(fs_info,
1322 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1323 inode->root->root_key.objectid,
1324 btrfs_ino(inode), extent_start, ram_size, ret);
1328 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1331 struct extent_map_tree *em_tree = &inode->extent_tree;
1332 struct extent_map *em;
1335 read_lock(&em_tree->lock);
1336 em = search_extent_mapping(em_tree, start, num_bytes);
1339 * if block start isn't an actual block number then find the
1340 * first block in this inode and use that as a hint. If that
1341 * block is also bogus then just don't worry about it.
1343 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1344 free_extent_map(em);
1345 em = search_extent_mapping(em_tree, 0, 0);
1346 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1347 alloc_hint = em->block_start;
1349 free_extent_map(em);
1351 alloc_hint = em->block_start;
1352 free_extent_map(em);
1355 read_unlock(&em_tree->lock);
1361 * when extent_io.c finds a delayed allocation range in the file,
1362 * the call backs end up in this code. The basic idea is to
1363 * allocate extents on disk for the range, and create ordered data structs
1364 * in ram to track those extents.
1366 * locked_page is the page that writepage had locked already. We use
1367 * it to make sure we don't do extra locks or unlocks.
1369 * When this function fails, it unlocks all pages except @locked_page.
1371 * When this function successfully creates an inline extent, it sets page_started
1372 * to 1 and unlocks all pages including locked_page and starts I/O on them.
1373 * (In reality inline extents are limited to a single page, so locked_page is
1374 * the only page handled anyway).
1376 * When this function succeed and creates a normal extent, the page locking
1377 * status depends on the passed in flags:
1379 * - If @keep_locked is set, all pages are kept locked.
1380 * - Else all pages except for @locked_page are unlocked.
1382 * When a failure happens in the second or later iteration of the
1383 * while-loop, the ordered extents created in previous iterations are kept
1384 * intact. So, the caller must clean them up by calling
1385 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1388 static noinline int cow_file_range(struct btrfs_inode *inode,
1389 struct page *locked_page,
1390 u64 start, u64 end, int *page_started,
1391 unsigned long *nr_written, u64 *done_offset,
1392 bool keep_locked, bool no_inline)
1394 struct btrfs_root *root = inode->root;
1395 struct btrfs_fs_info *fs_info = root->fs_info;
1397 u64 orig_start = start;
1399 unsigned long ram_size;
1400 u64 cur_alloc_size = 0;
1402 u64 blocksize = fs_info->sectorsize;
1403 struct btrfs_key ins;
1404 struct extent_map *em;
1405 unsigned clear_bits;
1406 unsigned long page_ops;
1407 bool extent_reserved = false;
1410 if (btrfs_is_free_space_inode(inode)) {
1415 num_bytes = ALIGN(end - start + 1, blocksize);
1416 num_bytes = max(blocksize, num_bytes);
1417 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1419 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1422 * Due to the page size limit, for subpage we can only trigger the
1423 * writeback for the dirty sectors of page, that means data writeback
1424 * is doing more writeback than what we want.
1426 * This is especially unexpected for some call sites like fallocate,
1427 * where we only increase i_size after everything is done.
1428 * This means we can trigger inline extent even if we didn't want to.
1429 * So here we skip inline extent creation completely.
1431 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1432 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1435 /* lets try to make an inline extent */
1436 ret = cow_file_range_inline(inode, actual_end, 0,
1437 BTRFS_COMPRESS_NONE, NULL, false);
1440 * We use DO_ACCOUNTING here because we need the
1441 * delalloc_release_metadata to be run _after_ we drop
1442 * our outstanding extent for clearing delalloc for this
1445 extent_clear_unlock_delalloc(inode, start, end,
1447 EXTENT_LOCKED | EXTENT_DELALLOC |
1448 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1449 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1450 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1451 *nr_written = *nr_written +
1452 (end - start + PAGE_SIZE) / PAGE_SIZE;
1455 * locked_page is locked by the caller of
1456 * writepage_delalloc(), not locked by
1457 * __process_pages_contig().
1459 * We can't let __process_pages_contig() to unlock it,
1460 * as it doesn't have any subpage::writers recorded.
1462 * Here we manually unlock the page, since the caller
1463 * can't use page_started to determine if it's an
1464 * inline extent or a compressed extent.
1466 unlock_page(locked_page);
1468 } else if (ret < 0) {
1473 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1476 * Relocation relies on the relocated extents to have exactly the same
1477 * size as the original extents. Normally writeback for relocation data
1478 * extents follows a NOCOW path because relocation preallocates the
1479 * extents. However, due to an operation such as scrub turning a block
1480 * group to RO mode, it may fallback to COW mode, so we must make sure
1481 * an extent allocated during COW has exactly the requested size and can
1482 * not be split into smaller extents, otherwise relocation breaks and
1483 * fails during the stage where it updates the bytenr of file extent
1486 if (btrfs_is_data_reloc_root(root))
1487 min_alloc_size = num_bytes;
1489 min_alloc_size = fs_info->sectorsize;
1491 while (num_bytes > 0) {
1492 struct btrfs_ordered_extent *ordered;
1494 cur_alloc_size = num_bytes;
1495 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1496 min_alloc_size, 0, alloc_hint,
1500 cur_alloc_size = ins.offset;
1501 extent_reserved = true;
1503 ram_size = ins.offset;
1504 em = create_io_em(inode, start, ins.offset, /* len */
1505 start, /* orig_start */
1506 ins.objectid, /* block_start */
1507 ins.offset, /* block_len */
1508 ins.offset, /* orig_block_len */
1509 ram_size, /* ram_bytes */
1510 BTRFS_COMPRESS_NONE, /* compress_type */
1511 BTRFS_ORDERED_REGULAR /* type */);
1516 free_extent_map(em);
1518 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1519 ram_size, ins.objectid, cur_alloc_size,
1520 0, 1 << BTRFS_ORDERED_REGULAR,
1521 BTRFS_COMPRESS_NONE);
1522 if (IS_ERR(ordered)) {
1523 ret = PTR_ERR(ordered);
1524 goto out_drop_extent_cache;
1527 if (btrfs_is_data_reloc_root(root)) {
1528 ret = btrfs_reloc_clone_csums(ordered);
1531 * Only drop cache here, and process as normal.
1533 * We must not allow extent_clear_unlock_delalloc()
1534 * at out_unlock label to free meta of this ordered
1535 * extent, as its meta should be freed by
1536 * btrfs_finish_ordered_io().
1538 * So we must continue until @start is increased to
1539 * skip current ordered extent.
1542 btrfs_drop_extent_map_range(inode, start,
1543 start + ram_size - 1,
1546 btrfs_put_ordered_extent(ordered);
1548 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1551 * We're not doing compressed IO, don't unlock the first page
1552 * (which the caller expects to stay locked), don't clear any
1553 * dirty bits and don't set any writeback bits
1555 * Do set the Ordered (Private2) bit so we know this page was
1556 * properly setup for writepage.
1558 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1559 page_ops |= PAGE_SET_ORDERED;
1561 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1563 EXTENT_LOCKED | EXTENT_DELALLOC,
1565 if (num_bytes < cur_alloc_size)
1568 num_bytes -= cur_alloc_size;
1569 alloc_hint = ins.objectid + ins.offset;
1570 start += cur_alloc_size;
1571 extent_reserved = false;
1574 * btrfs_reloc_clone_csums() error, since start is increased
1575 * extent_clear_unlock_delalloc() at out_unlock label won't
1576 * free metadata of current ordered extent, we're OK to exit.
1584 out_drop_extent_cache:
1585 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1587 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1588 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1591 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1592 * caller to write out the successfully allocated region and retry.
1594 if (done_offset && ret == -EAGAIN) {
1595 if (orig_start < start)
1596 *done_offset = start - 1;
1598 *done_offset = start;
1600 } else if (ret == -EAGAIN) {
1601 /* Convert to -ENOSPC since the caller cannot retry. */
1606 * Now, we have three regions to clean up:
1608 * |-------(1)----|---(2)---|-------------(3)----------|
1609 * `- orig_start `- start `- start + cur_alloc_size `- end
1611 * We process each region below.
1614 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1615 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1616 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1619 * For the range (1). We have already instantiated the ordered extents
1620 * for this region. They are cleaned up by
1621 * btrfs_cleanup_ordered_extents() in e.g,
1622 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1623 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1624 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1627 * However, in case of @keep_locked, we still need to unlock the pages
1628 * (except @locked_page) to ensure all the pages are unlocked.
1630 if (keep_locked && orig_start < start) {
1632 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1633 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1634 locked_page, 0, page_ops);
1638 * For the range (2). If we reserved an extent for our delalloc range
1639 * (or a subrange) and failed to create the respective ordered extent,
1640 * then it means that when we reserved the extent we decremented the
1641 * extent's size from the data space_info's bytes_may_use counter and
1642 * incremented the space_info's bytes_reserved counter by the same
1643 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1644 * to decrement again the data space_info's bytes_may_use counter,
1645 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1647 if (extent_reserved) {
1648 extent_clear_unlock_delalloc(inode, start,
1649 start + cur_alloc_size - 1,
1653 start += cur_alloc_size;
1657 * For the range (3). We never touched the region. In addition to the
1658 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1659 * space_info's bytes_may_use counter, reserved in
1660 * btrfs_check_data_free_space().
1663 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1664 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1665 clear_bits, page_ops);
1671 * work queue call back to started compression on a file and pages
1673 static noinline void async_cow_start(struct btrfs_work *work)
1675 struct async_chunk *async_chunk;
1676 int compressed_extents;
1678 async_chunk = container_of(work, struct async_chunk, work);
1680 compressed_extents = compress_file_range(async_chunk);
1681 if (compressed_extents == 0) {
1682 btrfs_add_delayed_iput(async_chunk->inode);
1683 async_chunk->inode = NULL;
1688 * work queue call back to submit previously compressed pages
1690 static noinline void async_cow_submit(struct btrfs_work *work)
1692 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1694 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1695 unsigned long nr_pages;
1697 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1701 * ->inode could be NULL if async_chunk_start has failed to compress,
1702 * in which case we don't have anything to submit, yet we need to
1703 * always adjust ->async_delalloc_pages as its paired with the init
1704 * happening in run_delalloc_compressed
1706 if (async_chunk->inode)
1707 submit_compressed_extents(async_chunk);
1709 /* atomic_sub_return implies a barrier */
1710 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1712 cond_wake_up_nomb(&fs_info->async_submit_wait);
1715 static noinline void async_cow_free(struct btrfs_work *work)
1717 struct async_chunk *async_chunk;
1718 struct async_cow *async_cow;
1720 async_chunk = container_of(work, struct async_chunk, work);
1721 if (async_chunk->inode)
1722 btrfs_add_delayed_iput(async_chunk->inode);
1723 if (async_chunk->blkcg_css)
1724 css_put(async_chunk->blkcg_css);
1726 async_cow = async_chunk->async_cow;
1727 if (atomic_dec_and_test(&async_cow->num_chunks))
1731 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1732 struct writeback_control *wbc,
1733 struct page *locked_page,
1734 u64 start, u64 end, int *page_started,
1735 unsigned long *nr_written)
1737 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1738 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1739 struct async_cow *ctx;
1740 struct async_chunk *async_chunk;
1741 unsigned long nr_pages;
1742 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1745 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1747 nofs_flag = memalloc_nofs_save();
1748 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1749 memalloc_nofs_restore(nofs_flag);
1753 unlock_extent(&inode->io_tree, start, end, NULL);
1754 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1756 async_chunk = ctx->chunks;
1757 atomic_set(&ctx->num_chunks, num_chunks);
1759 for (i = 0; i < num_chunks; i++) {
1760 u64 cur_end = min(end, start + SZ_512K - 1);
1763 * igrab is called higher up in the call chain, take only the
1764 * lightweight reference for the callback lifetime
1766 ihold(&inode->vfs_inode);
1767 async_chunk[i].async_cow = ctx;
1768 async_chunk[i].inode = inode;
1769 async_chunk[i].start = start;
1770 async_chunk[i].end = cur_end;
1771 async_chunk[i].write_flags = write_flags;
1772 INIT_LIST_HEAD(&async_chunk[i].extents);
1775 * The locked_page comes all the way from writepage and its
1776 * the original page we were actually given. As we spread
1777 * this large delalloc region across multiple async_chunk
1778 * structs, only the first struct needs a pointer to locked_page
1780 * This way we don't need racey decisions about who is supposed
1785 * Depending on the compressibility, the pages might or
1786 * might not go through async. We want all of them to
1787 * be accounted against wbc once. Let's do it here
1788 * before the paths diverge. wbc accounting is used
1789 * only for foreign writeback detection and doesn't
1790 * need full accuracy. Just account the whole thing
1791 * against the first page.
1793 wbc_account_cgroup_owner(wbc, locked_page,
1795 async_chunk[i].locked_page = locked_page;
1798 async_chunk[i].locked_page = NULL;
1801 if (blkcg_css != blkcg_root_css) {
1803 async_chunk[i].blkcg_css = blkcg_css;
1804 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1806 async_chunk[i].blkcg_css = NULL;
1809 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1810 async_cow_submit, async_cow_free);
1812 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1813 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1815 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1817 *nr_written += nr_pages;
1818 start = cur_end + 1;
1824 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1825 struct page *locked_page, u64 start,
1826 u64 end, int *page_started,
1827 unsigned long *nr_written,
1828 struct writeback_control *wbc)
1830 u64 done_offset = end;
1832 bool locked_page_done = false;
1834 while (start <= end) {
1835 ret = cow_file_range(inode, locked_page, start, end, page_started,
1836 nr_written, &done_offset, true, false);
1837 if (ret && ret != -EAGAIN)
1840 if (*page_started) {
1848 if (done_offset == start) {
1849 wait_on_bit_io(&inode->root->fs_info->flags,
1850 BTRFS_FS_NEED_ZONE_FINISH,
1851 TASK_UNINTERRUPTIBLE);
1855 if (!locked_page_done) {
1856 __set_page_dirty_nobuffers(locked_page);
1857 account_page_redirty(locked_page);
1859 locked_page_done = true;
1860 extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1862 start = done_offset + 1;
1870 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1871 u64 bytenr, u64 num_bytes, bool nowait)
1873 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1874 struct btrfs_ordered_sum *sums;
1878 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1880 if (ret == 0 && list_empty(&list))
1883 while (!list_empty(&list)) {
1884 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1885 list_del(&sums->list);
1893 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1894 const u64 start, const u64 end)
1896 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1897 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1898 const u64 range_bytes = end + 1 - start;
1899 struct extent_io_tree *io_tree = &inode->io_tree;
1900 int page_started = 0;
1901 unsigned long nr_written;
1902 u64 range_start = start;
1907 * If EXTENT_NORESERVE is set it means that when the buffered write was
1908 * made we had not enough available data space and therefore we did not
1909 * reserve data space for it, since we though we could do NOCOW for the
1910 * respective file range (either there is prealloc extent or the inode
1911 * has the NOCOW bit set).
1913 * However when we need to fallback to COW mode (because for example the
1914 * block group for the corresponding extent was turned to RO mode by a
1915 * scrub or relocation) we need to do the following:
1917 * 1) We increment the bytes_may_use counter of the data space info.
1918 * If COW succeeds, it allocates a new data extent and after doing
1919 * that it decrements the space info's bytes_may_use counter and
1920 * increments its bytes_reserved counter by the same amount (we do
1921 * this at btrfs_add_reserved_bytes()). So we need to increment the
1922 * bytes_may_use counter to compensate (when space is reserved at
1923 * buffered write time, the bytes_may_use counter is incremented);
1925 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1926 * that if the COW path fails for any reason, it decrements (through
1927 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1928 * data space info, which we incremented in the step above.
1930 * If we need to fallback to cow and the inode corresponds to a free
1931 * space cache inode or an inode of the data relocation tree, we must
1932 * also increment bytes_may_use of the data space_info for the same
1933 * reason. Space caches and relocated data extents always get a prealloc
1934 * extent for them, however scrub or balance may have set the block
1935 * group that contains that extent to RO mode and therefore force COW
1936 * when starting writeback.
1938 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1939 EXTENT_NORESERVE, 0, NULL);
1940 if (count > 0 || is_space_ino || is_reloc_ino) {
1942 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1943 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1945 if (is_space_ino || is_reloc_ino)
1946 bytes = range_bytes;
1948 spin_lock(&sinfo->lock);
1949 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1950 spin_unlock(&sinfo->lock);
1953 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1958 * Don't try to create inline extents, as a mix of inline extent that
1959 * is written out and unlocked directly and a normal NOCOW extent
1962 ret = cow_file_range(inode, locked_page, start, end, &page_started,
1963 &nr_written, NULL, false, true);
1964 ASSERT(!page_started);
1968 struct can_nocow_file_extent_args {
1971 /* Start file offset of the range we want to NOCOW. */
1973 /* End file offset (inclusive) of the range we want to NOCOW. */
1975 bool writeback_path;
1978 * Free the path passed to can_nocow_file_extent() once it's not needed
1983 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1988 /* Number of bytes that can be written to in NOCOW mode. */
1993 * Check if we can NOCOW the file extent that the path points to.
1994 * This function may return with the path released, so the caller should check
1995 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1997 * Returns: < 0 on error
1998 * 0 if we can not NOCOW
2001 static int can_nocow_file_extent(struct btrfs_path *path,
2002 struct btrfs_key *key,
2003 struct btrfs_inode *inode,
2004 struct can_nocow_file_extent_args *args)
2006 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
2007 struct extent_buffer *leaf = path->nodes[0];
2008 struct btrfs_root *root = inode->root;
2009 struct btrfs_file_extent_item *fi;
2014 bool nowait = path->nowait;
2016 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2017 extent_type = btrfs_file_extent_type(leaf, fi);
2019 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
2022 /* Can't access these fields unless we know it's not an inline extent. */
2023 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2024 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
2025 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
2027 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
2028 extent_type == BTRFS_FILE_EXTENT_REG)
2032 * If the extent was created before the generation where the last snapshot
2033 * for its subvolume was created, then this implies the extent is shared,
2034 * hence we must COW.
2036 if (!args->strict &&
2037 btrfs_file_extent_generation(leaf, fi) <=
2038 btrfs_root_last_snapshot(&root->root_item))
2041 /* An explicit hole, must COW. */
2042 if (args->disk_bytenr == 0)
2045 /* Compressed/encrypted/encoded extents must be COWed. */
2046 if (btrfs_file_extent_compression(leaf, fi) ||
2047 btrfs_file_extent_encryption(leaf, fi) ||
2048 btrfs_file_extent_other_encoding(leaf, fi))
2051 extent_end = btrfs_file_extent_end(path);
2054 * The following checks can be expensive, as they need to take other
2055 * locks and do btree or rbtree searches, so release the path to avoid
2056 * blocking other tasks for too long.
2058 btrfs_release_path(path);
2060 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
2061 key->offset - args->extent_offset,
2062 args->disk_bytenr, args->strict, path);
2063 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2067 if (args->free_path) {
2069 * We don't need the path anymore, plus through the
2070 * csum_exist_in_range() call below we will end up allocating
2071 * another path. So free the path to avoid unnecessary extra
2074 btrfs_free_path(path);
2078 /* If there are pending snapshots for this root, we must COW. */
2079 if (args->writeback_path && !is_freespace_inode &&
2080 atomic_read(&root->snapshot_force_cow))
2083 args->disk_bytenr += args->extent_offset;
2084 args->disk_bytenr += args->start - key->offset;
2085 args->num_bytes = min(args->end + 1, extent_end) - args->start;
2088 * Force COW if csums exist in the range. This ensures that csums for a
2089 * given extent are either valid or do not exist.
2091 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
2093 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2099 if (args->free_path && path)
2100 btrfs_free_path(path);
2102 return ret < 0 ? ret : can_nocow;
2106 * when nowcow writeback call back. This checks for snapshots or COW copies
2107 * of the extents that exist in the file, and COWs the file as required.
2109 * If no cow copies or snapshots exist, we write directly to the existing
2112 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
2113 struct page *locked_page,
2114 const u64 start, const u64 end)
2116 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2117 struct btrfs_root *root = inode->root;
2118 struct btrfs_path *path;
2119 u64 cow_start = (u64)-1;
2120 u64 cur_offset = start;
2122 bool check_prev = true;
2123 u64 ino = btrfs_ino(inode);
2124 struct btrfs_block_group *bg;
2126 struct can_nocow_file_extent_args nocow_args = { 0 };
2128 path = btrfs_alloc_path();
2130 extent_clear_unlock_delalloc(inode, start, end, locked_page,
2131 EXTENT_LOCKED | EXTENT_DELALLOC |
2132 EXTENT_DO_ACCOUNTING |
2133 EXTENT_DEFRAG, PAGE_UNLOCK |
2134 PAGE_START_WRITEBACK |
2135 PAGE_END_WRITEBACK);
2139 nocow_args.end = end;
2140 nocow_args.writeback_path = true;
2143 struct btrfs_ordered_extent *ordered;
2144 struct btrfs_key found_key;
2145 struct btrfs_file_extent_item *fi;
2146 struct extent_buffer *leaf;
2155 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2161 * If there is no extent for our range when doing the initial
2162 * search, then go back to the previous slot as it will be the
2163 * one containing the search offset
2165 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2166 leaf = path->nodes[0];
2167 btrfs_item_key_to_cpu(leaf, &found_key,
2168 path->slots[0] - 1);
2169 if (found_key.objectid == ino &&
2170 found_key.type == BTRFS_EXTENT_DATA_KEY)
2175 /* Go to next leaf if we have exhausted the current one */
2176 leaf = path->nodes[0];
2177 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2178 ret = btrfs_next_leaf(root, path);
2180 if (cow_start != (u64)-1)
2181 cur_offset = cow_start;
2186 leaf = path->nodes[0];
2189 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2191 /* Didn't find anything for our INO */
2192 if (found_key.objectid > ino)
2195 * Keep searching until we find an EXTENT_ITEM or there are no
2196 * more extents for this inode
2198 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2199 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2204 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2205 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2206 found_key.offset > end)
2210 * If the found extent starts after requested offset, then
2211 * adjust extent_end to be right before this extent begins
2213 if (found_key.offset > cur_offset) {
2214 extent_end = found_key.offset;
2220 * Found extent which begins before our range and potentially
2223 fi = btrfs_item_ptr(leaf, path->slots[0],
2224 struct btrfs_file_extent_item);
2225 extent_type = btrfs_file_extent_type(leaf, fi);
2226 /* If this is triggered then we have a memory corruption. */
2227 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2228 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2232 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2233 extent_end = btrfs_file_extent_end(path);
2236 * If the extent we got ends before our current offset, skip to
2239 if (extent_end <= cur_offset) {
2244 nocow_args.start = cur_offset;
2245 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2247 if (cow_start != (u64)-1)
2248 cur_offset = cow_start;
2250 } else if (ret == 0) {
2255 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2260 * If nocow is false then record the beginning of the range
2261 * that needs to be COWed
2264 if (cow_start == (u64)-1)
2265 cow_start = cur_offset;
2266 cur_offset = extent_end;
2267 if (cur_offset > end)
2269 if (!path->nodes[0])
2276 * COW range from cow_start to found_key.offset - 1. As the key
2277 * will contain the beginning of the first extent that can be
2278 * NOCOW, following one which needs to be COW'ed
2280 if (cow_start != (u64)-1) {
2281 ret = fallback_to_cow(inode, locked_page,
2282 cow_start, found_key.offset - 1);
2285 cow_start = (u64)-1;
2288 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2289 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2291 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2292 struct extent_map *em;
2294 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2296 nocow_args.disk_bytenr, /* block_start */
2297 nocow_args.num_bytes, /* block_len */
2298 nocow_args.disk_num_bytes, /* orig_block_len */
2299 ram_bytes, BTRFS_COMPRESS_NONE,
2300 BTRFS_ORDERED_PREALLOC);
2305 free_extent_map(em);
2308 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2309 nocow_args.num_bytes, nocow_args.num_bytes,
2310 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2312 ? (1 << BTRFS_ORDERED_PREALLOC)
2313 : (1 << BTRFS_ORDERED_NOCOW),
2314 BTRFS_COMPRESS_NONE);
2315 if (IS_ERR(ordered)) {
2317 btrfs_drop_extent_map_range(inode, cur_offset,
2320 ret = PTR_ERR(ordered);
2325 btrfs_dec_nocow_writers(bg);
2329 if (btrfs_is_data_reloc_root(root))
2331 * Error handled later, as we must prevent
2332 * extent_clear_unlock_delalloc() in error handler
2333 * from freeing metadata of created ordered extent.
2335 ret = btrfs_reloc_clone_csums(ordered);
2336 btrfs_put_ordered_extent(ordered);
2338 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2339 locked_page, EXTENT_LOCKED |
2341 EXTENT_CLEAR_DATA_RESV,
2342 PAGE_UNLOCK | PAGE_SET_ORDERED);
2344 cur_offset = extent_end;
2347 * btrfs_reloc_clone_csums() error, now we're OK to call error
2348 * handler, as metadata for created ordered extent will only
2349 * be freed by btrfs_finish_ordered_io().
2353 if (cur_offset > end)
2356 btrfs_release_path(path);
2358 if (cur_offset <= end && cow_start == (u64)-1)
2359 cow_start = cur_offset;
2361 if (cow_start != (u64)-1) {
2363 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2370 btrfs_dec_nocow_writers(bg);
2372 if (ret && cur_offset < end)
2373 extent_clear_unlock_delalloc(inode, cur_offset, end,
2374 locked_page, EXTENT_LOCKED |
2375 EXTENT_DELALLOC | EXTENT_DEFRAG |
2376 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2377 PAGE_START_WRITEBACK |
2378 PAGE_END_WRITEBACK);
2379 btrfs_free_path(path);
2383 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2385 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2386 if (inode->defrag_bytes &&
2387 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2396 * Function to process delayed allocation (create CoW) for ranges which are
2397 * being touched for the first time.
2399 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2400 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2401 struct writeback_control *wbc)
2404 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2407 * The range must cover part of the @locked_page, or the returned
2408 * @page_started can confuse the caller.
2410 ASSERT(!(end <= page_offset(locked_page) ||
2411 start >= page_offset(locked_page) + PAGE_SIZE));
2413 if (should_nocow(inode, start, end)) {
2415 * Normally on a zoned device we're only doing COW writes, but
2416 * in case of relocation on a zoned filesystem we have taken
2417 * precaution, that we're only writing sequentially. It's safe
2418 * to use run_delalloc_nocow() here, like for regular
2419 * preallocated inodes.
2421 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2422 ret = run_delalloc_nocow(inode, locked_page, start, end);
2426 if (btrfs_inode_can_compress(inode) &&
2427 inode_need_compress(inode, start, end) &&
2428 run_delalloc_compressed(inode, wbc, locked_page, start,
2429 end, page_started, nr_written))
2433 ret = run_delalloc_zoned(inode, locked_page, start, end,
2434 page_started, nr_written, wbc);
2436 ret = cow_file_range(inode, locked_page, start, end,
2437 page_started, nr_written, NULL, false, false);
2442 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2447 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2448 struct extent_state *orig, u64 split)
2450 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2453 /* not delalloc, ignore it */
2454 if (!(orig->state & EXTENT_DELALLOC))
2457 size = orig->end - orig->start + 1;
2458 if (size > fs_info->max_extent_size) {
2463 * See the explanation in btrfs_merge_delalloc_extent, the same
2464 * applies here, just in reverse.
2466 new_size = orig->end - split + 1;
2467 num_extents = count_max_extents(fs_info, new_size);
2468 new_size = split - orig->start;
2469 num_extents += count_max_extents(fs_info, new_size);
2470 if (count_max_extents(fs_info, size) >= num_extents)
2474 spin_lock(&inode->lock);
2475 btrfs_mod_outstanding_extents(inode, 1);
2476 spin_unlock(&inode->lock);
2480 * Handle merged delayed allocation extents so we can keep track of new extents
2481 * that are just merged onto old extents, such as when we are doing sequential
2482 * writes, so we can properly account for the metadata space we'll need.
2484 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2485 struct extent_state *other)
2487 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2488 u64 new_size, old_size;
2491 /* not delalloc, ignore it */
2492 if (!(other->state & EXTENT_DELALLOC))
2495 if (new->start > other->start)
2496 new_size = new->end - other->start + 1;
2498 new_size = other->end - new->start + 1;
2500 /* we're not bigger than the max, unreserve the space and go */
2501 if (new_size <= fs_info->max_extent_size) {
2502 spin_lock(&inode->lock);
2503 btrfs_mod_outstanding_extents(inode, -1);
2504 spin_unlock(&inode->lock);
2509 * We have to add up either side to figure out how many extents were
2510 * accounted for before we merged into one big extent. If the number of
2511 * extents we accounted for is <= the amount we need for the new range
2512 * then we can return, otherwise drop. Think of it like this
2516 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2517 * need 2 outstanding extents, on one side we have 1 and the other side
2518 * we have 1 so they are == and we can return. But in this case
2520 * [MAX_SIZE+4k][MAX_SIZE+4k]
2522 * Each range on their own accounts for 2 extents, but merged together
2523 * they are only 3 extents worth of accounting, so we need to drop in
2526 old_size = other->end - other->start + 1;
2527 num_extents = count_max_extents(fs_info, old_size);
2528 old_size = new->end - new->start + 1;
2529 num_extents += count_max_extents(fs_info, old_size);
2530 if (count_max_extents(fs_info, new_size) >= num_extents)
2533 spin_lock(&inode->lock);
2534 btrfs_mod_outstanding_extents(inode, -1);
2535 spin_unlock(&inode->lock);
2538 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2539 struct btrfs_inode *inode)
2541 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2543 spin_lock(&root->delalloc_lock);
2544 if (list_empty(&inode->delalloc_inodes)) {
2545 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2546 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2547 root->nr_delalloc_inodes++;
2548 if (root->nr_delalloc_inodes == 1) {
2549 spin_lock(&fs_info->delalloc_root_lock);
2550 BUG_ON(!list_empty(&root->delalloc_root));
2551 list_add_tail(&root->delalloc_root,
2552 &fs_info->delalloc_roots);
2553 spin_unlock(&fs_info->delalloc_root_lock);
2556 spin_unlock(&root->delalloc_lock);
2559 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2560 struct btrfs_inode *inode)
2562 struct btrfs_fs_info *fs_info = root->fs_info;
2564 if (!list_empty(&inode->delalloc_inodes)) {
2565 list_del_init(&inode->delalloc_inodes);
2566 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2567 &inode->runtime_flags);
2568 root->nr_delalloc_inodes--;
2569 if (!root->nr_delalloc_inodes) {
2570 ASSERT(list_empty(&root->delalloc_inodes));
2571 spin_lock(&fs_info->delalloc_root_lock);
2572 BUG_ON(list_empty(&root->delalloc_root));
2573 list_del_init(&root->delalloc_root);
2574 spin_unlock(&fs_info->delalloc_root_lock);
2579 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2580 struct btrfs_inode *inode)
2582 spin_lock(&root->delalloc_lock);
2583 __btrfs_del_delalloc_inode(root, inode);
2584 spin_unlock(&root->delalloc_lock);
2588 * Properly track delayed allocation bytes in the inode and to maintain the
2589 * list of inodes that have pending delalloc work to be done.
2591 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2594 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2596 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2599 * set_bit and clear bit hooks normally require _irqsave/restore
2600 * but in this case, we are only testing for the DELALLOC
2601 * bit, which is only set or cleared with irqs on
2603 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2604 struct btrfs_root *root = inode->root;
2605 u64 len = state->end + 1 - state->start;
2606 u32 num_extents = count_max_extents(fs_info, len);
2607 bool do_list = !btrfs_is_free_space_inode(inode);
2609 spin_lock(&inode->lock);
2610 btrfs_mod_outstanding_extents(inode, num_extents);
2611 spin_unlock(&inode->lock);
2613 /* For sanity tests */
2614 if (btrfs_is_testing(fs_info))
2617 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2618 fs_info->delalloc_batch);
2619 spin_lock(&inode->lock);
2620 inode->delalloc_bytes += len;
2621 if (bits & EXTENT_DEFRAG)
2622 inode->defrag_bytes += len;
2623 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2624 &inode->runtime_flags))
2625 btrfs_add_delalloc_inodes(root, inode);
2626 spin_unlock(&inode->lock);
2629 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2630 (bits & EXTENT_DELALLOC_NEW)) {
2631 spin_lock(&inode->lock);
2632 inode->new_delalloc_bytes += state->end + 1 - state->start;
2633 spin_unlock(&inode->lock);
2638 * Once a range is no longer delalloc this function ensures that proper
2639 * accounting happens.
2641 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2642 struct extent_state *state, u32 bits)
2644 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2645 u64 len = state->end + 1 - state->start;
2646 u32 num_extents = count_max_extents(fs_info, len);
2648 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2649 spin_lock(&inode->lock);
2650 inode->defrag_bytes -= len;
2651 spin_unlock(&inode->lock);
2655 * set_bit and clear bit hooks normally require _irqsave/restore
2656 * but in this case, we are only testing for the DELALLOC
2657 * bit, which is only set or cleared with irqs on
2659 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2660 struct btrfs_root *root = inode->root;
2661 bool do_list = !btrfs_is_free_space_inode(inode);
2663 spin_lock(&inode->lock);
2664 btrfs_mod_outstanding_extents(inode, -num_extents);
2665 spin_unlock(&inode->lock);
2668 * We don't reserve metadata space for space cache inodes so we
2669 * don't need to call delalloc_release_metadata if there is an
2672 if (bits & EXTENT_CLEAR_META_RESV &&
2673 root != fs_info->tree_root)
2674 btrfs_delalloc_release_metadata(inode, len, false);
2676 /* For sanity tests. */
2677 if (btrfs_is_testing(fs_info))
2680 if (!btrfs_is_data_reloc_root(root) &&
2681 do_list && !(state->state & EXTENT_NORESERVE) &&
2682 (bits & EXTENT_CLEAR_DATA_RESV))
2683 btrfs_free_reserved_data_space_noquota(fs_info, len);
2685 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2686 fs_info->delalloc_batch);
2687 spin_lock(&inode->lock);
2688 inode->delalloc_bytes -= len;
2689 if (do_list && inode->delalloc_bytes == 0 &&
2690 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2691 &inode->runtime_flags))
2692 btrfs_del_delalloc_inode(root, inode);
2693 spin_unlock(&inode->lock);
2696 if ((state->state & EXTENT_DELALLOC_NEW) &&
2697 (bits & EXTENT_DELALLOC_NEW)) {
2698 spin_lock(&inode->lock);
2699 ASSERT(inode->new_delalloc_bytes >= len);
2700 inode->new_delalloc_bytes -= len;
2701 if (bits & EXTENT_ADD_INODE_BYTES)
2702 inode_add_bytes(&inode->vfs_inode, len);
2703 spin_unlock(&inode->lock);
2707 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2708 struct btrfs_ordered_extent *ordered)
2710 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2711 u64 len = bbio->bio.bi_iter.bi_size;
2712 struct btrfs_ordered_extent *new;
2715 /* Must always be called for the beginning of an ordered extent. */
2716 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2719 /* No need to split if the ordered extent covers the entire bio. */
2720 if (ordered->disk_num_bytes == len) {
2721 refcount_inc(&ordered->refs);
2722 bbio->ordered = ordered;
2727 * Don't split the extent_map for NOCOW extents, as we're writing into
2728 * a pre-existing one.
2730 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2731 ret = split_extent_map(bbio->inode, bbio->file_offset,
2732 ordered->num_bytes, len,
2733 ordered->disk_bytenr);
2738 new = btrfs_split_ordered_extent(ordered, len);
2740 return PTR_ERR(new);
2741 bbio->ordered = new;
2746 * given a list of ordered sums record them in the inode. This happens
2747 * at IO completion time based on sums calculated at bio submission time.
2749 static int add_pending_csums(struct btrfs_trans_handle *trans,
2750 struct list_head *list)
2752 struct btrfs_ordered_sum *sum;
2753 struct btrfs_root *csum_root = NULL;
2756 list_for_each_entry(sum, list, list) {
2757 trans->adding_csums = true;
2759 csum_root = btrfs_csum_root(trans->fs_info,
2761 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2762 trans->adding_csums = false;
2769 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2772 struct extent_state **cached_state)
2774 u64 search_start = start;
2775 const u64 end = start + len - 1;
2777 while (search_start < end) {
2778 const u64 search_len = end - search_start + 1;
2779 struct extent_map *em;
2783 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2787 if (em->block_start != EXTENT_MAP_HOLE)
2791 if (em->start < search_start)
2792 em_len -= search_start - em->start;
2793 if (em_len > search_len)
2794 em_len = search_len;
2796 ret = set_extent_bit(&inode->io_tree, search_start,
2797 search_start + em_len - 1,
2798 EXTENT_DELALLOC_NEW, cached_state);
2800 search_start = extent_map_end(em);
2801 free_extent_map(em);
2808 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2809 unsigned int extra_bits,
2810 struct extent_state **cached_state)
2812 WARN_ON(PAGE_ALIGNED(end));
2814 if (start >= i_size_read(&inode->vfs_inode) &&
2815 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2817 * There can't be any extents following eof in this case so just
2818 * set the delalloc new bit for the range directly.
2820 extra_bits |= EXTENT_DELALLOC_NEW;
2824 ret = btrfs_find_new_delalloc_bytes(inode, start,
2831 return set_extent_bit(&inode->io_tree, start, end,
2832 EXTENT_DELALLOC | extra_bits, cached_state);
2835 /* see btrfs_writepage_start_hook for details on why this is required */
2836 struct btrfs_writepage_fixup {
2838 struct btrfs_inode *inode;
2839 struct btrfs_work work;
2842 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2844 struct btrfs_writepage_fixup *fixup =
2845 container_of(work, struct btrfs_writepage_fixup, work);
2846 struct btrfs_ordered_extent *ordered;
2847 struct extent_state *cached_state = NULL;
2848 struct extent_changeset *data_reserved = NULL;
2849 struct page *page = fixup->page;
2850 struct btrfs_inode *inode = fixup->inode;
2851 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2852 u64 page_start = page_offset(page);
2853 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2855 bool free_delalloc_space = true;
2858 * This is similar to page_mkwrite, we need to reserve the space before
2859 * we take the page lock.
2861 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2867 * Before we queued this fixup, we took a reference on the page.
2868 * page->mapping may go NULL, but it shouldn't be moved to a different
2871 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2873 * Unfortunately this is a little tricky, either
2875 * 1) We got here and our page had already been dealt with and
2876 * we reserved our space, thus ret == 0, so we need to just
2877 * drop our space reservation and bail. This can happen the
2878 * first time we come into the fixup worker, or could happen
2879 * while waiting for the ordered extent.
2880 * 2) Our page was already dealt with, but we happened to get an
2881 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2882 * this case we obviously don't have anything to release, but
2883 * because the page was already dealt with we don't want to
2884 * mark the page with an error, so make sure we're resetting
2885 * ret to 0. This is why we have this check _before_ the ret
2886 * check, because we do not want to have a surprise ENOSPC
2887 * when the page was already properly dealt with.
2890 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2891 btrfs_delalloc_release_space(inode, data_reserved,
2892 page_start, PAGE_SIZE,
2900 * We can't mess with the page state unless it is locked, so now that
2901 * it is locked bail if we failed to make our space reservation.
2906 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2908 /* already ordered? We're done */
2909 if (PageOrdered(page))
2912 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2914 unlock_extent(&inode->io_tree, page_start, page_end,
2917 btrfs_start_ordered_extent(ordered);
2918 btrfs_put_ordered_extent(ordered);
2922 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2928 * Everything went as planned, we're now the owner of a dirty page with
2929 * delayed allocation bits set and space reserved for our COW
2932 * The page was dirty when we started, nothing should have cleaned it.
2934 BUG_ON(!PageDirty(page));
2935 free_delalloc_space = false;
2937 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2938 if (free_delalloc_space)
2939 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2941 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2945 * We hit ENOSPC or other errors. Update the mapping and page
2946 * to reflect the errors and clean the page.
2948 mapping_set_error(page->mapping, ret);
2949 btrfs_mark_ordered_io_finished(inode, page, page_start,
2951 btrfs_page_clear_uptodate(fs_info, page, page_start, PAGE_SIZE);
2952 clear_page_dirty_for_io(page);
2954 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2958 extent_changeset_free(data_reserved);
2960 * As a precaution, do a delayed iput in case it would be the last iput
2961 * that could need flushing space. Recursing back to fixup worker would
2964 btrfs_add_delayed_iput(inode);
2968 * There are a few paths in the higher layers of the kernel that directly
2969 * set the page dirty bit without asking the filesystem if it is a
2970 * good idea. This causes problems because we want to make sure COW
2971 * properly happens and the data=ordered rules are followed.
2973 * In our case any range that doesn't have the ORDERED bit set
2974 * hasn't been properly setup for IO. We kick off an async process
2975 * to fix it up. The async helper will wait for ordered extents, set
2976 * the delalloc bit and make it safe to write the page.
2978 int btrfs_writepage_cow_fixup(struct page *page)
2980 struct inode *inode = page->mapping->host;
2981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2982 struct btrfs_writepage_fixup *fixup;
2984 /* This page has ordered extent covering it already */
2985 if (PageOrdered(page))
2989 * PageChecked is set below when we create a fixup worker for this page,
2990 * don't try to create another one if we're already PageChecked()
2992 * The extent_io writepage code will redirty the page if we send back
2995 if (PageChecked(page))
2998 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
3003 * We are already holding a reference to this inode from
3004 * write_cache_pages. We need to hold it because the space reservation
3005 * takes place outside of the page lock, and we can't trust
3006 * page->mapping outside of the page lock.
3009 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
3011 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
3013 fixup->inode = BTRFS_I(inode);
3014 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
3019 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
3020 struct btrfs_inode *inode, u64 file_pos,
3021 struct btrfs_file_extent_item *stack_fi,
3022 const bool update_inode_bytes,
3023 u64 qgroup_reserved)
3025 struct btrfs_root *root = inode->root;
3026 const u64 sectorsize = root->fs_info->sectorsize;
3027 struct btrfs_path *path;
3028 struct extent_buffer *leaf;
3029 struct btrfs_key ins;
3030 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
3031 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
3032 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
3033 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
3034 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
3035 struct btrfs_drop_extents_args drop_args = { 0 };
3038 path = btrfs_alloc_path();
3043 * we may be replacing one extent in the tree with another.
3044 * The new extent is pinned in the extent map, and we don't want
3045 * to drop it from the cache until it is completely in the btree.
3047 * So, tell btrfs_drop_extents to leave this extent in the cache.
3048 * the caller is expected to unpin it and allow it to be merged
3051 drop_args.path = path;
3052 drop_args.start = file_pos;
3053 drop_args.end = file_pos + num_bytes;
3054 drop_args.replace_extent = true;
3055 drop_args.extent_item_size = sizeof(*stack_fi);
3056 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3060 if (!drop_args.extent_inserted) {
3061 ins.objectid = btrfs_ino(inode);
3062 ins.offset = file_pos;
3063 ins.type = BTRFS_EXTENT_DATA_KEY;
3065 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3070 leaf = path->nodes[0];
3071 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3072 write_extent_buffer(leaf, stack_fi,
3073 btrfs_item_ptr_offset(leaf, path->slots[0]),
3074 sizeof(struct btrfs_file_extent_item));
3076 btrfs_mark_buffer_dirty(leaf);
3077 btrfs_release_path(path);
3080 * If we dropped an inline extent here, we know the range where it is
3081 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3082 * number of bytes only for that range containing the inline extent.
3083 * The remaining of the range will be processed when clearning the
3084 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3086 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3087 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3089 inline_size = drop_args.bytes_found - inline_size;
3090 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3091 drop_args.bytes_found -= inline_size;
3092 num_bytes -= sectorsize;
3095 if (update_inode_bytes)
3096 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3098 ins.objectid = disk_bytenr;
3099 ins.offset = disk_num_bytes;
3100 ins.type = BTRFS_EXTENT_ITEM_KEY;
3102 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3106 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3108 qgroup_reserved, &ins);
3110 btrfs_free_path(path);
3115 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3118 struct btrfs_block_group *cache;
3120 cache = btrfs_lookup_block_group(fs_info, start);
3123 spin_lock(&cache->lock);
3124 cache->delalloc_bytes -= len;
3125 spin_unlock(&cache->lock);
3127 btrfs_put_block_group(cache);
3130 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3131 struct btrfs_ordered_extent *oe)
3133 struct btrfs_file_extent_item stack_fi;
3134 bool update_inode_bytes;
3135 u64 num_bytes = oe->num_bytes;
3136 u64 ram_bytes = oe->ram_bytes;
3138 memset(&stack_fi, 0, sizeof(stack_fi));
3139 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3140 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3141 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3142 oe->disk_num_bytes);
3143 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3144 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3145 num_bytes = oe->truncated_len;
3146 ram_bytes = num_bytes;
3148 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3149 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3150 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3151 /* Encryption and other encoding is reserved and all 0 */
3154 * For delalloc, when completing an ordered extent we update the inode's
3155 * bytes when clearing the range in the inode's io tree, so pass false
3156 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3157 * except if the ordered extent was truncated.
3159 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3160 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3161 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3163 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3164 oe->file_offset, &stack_fi,
3165 update_inode_bytes, oe->qgroup_rsv);
3169 * As ordered data IO finishes, this gets called so we can finish
3170 * an ordered extent if the range of bytes in the file it covers are
3173 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3175 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3176 struct btrfs_root *root = inode->root;
3177 struct btrfs_fs_info *fs_info = root->fs_info;
3178 struct btrfs_trans_handle *trans = NULL;
3179 struct extent_io_tree *io_tree = &inode->io_tree;
3180 struct extent_state *cached_state = NULL;
3182 int compress_type = 0;
3184 u64 logical_len = ordered_extent->num_bytes;
3185 bool freespace_inode;
3186 bool truncated = false;
3187 bool clear_reserved_extent = true;
3188 unsigned int clear_bits = EXTENT_DEFRAG;
3190 start = ordered_extent->file_offset;
3191 end = start + ordered_extent->num_bytes - 1;
3193 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3194 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3195 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3196 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3197 clear_bits |= EXTENT_DELALLOC_NEW;
3199 freespace_inode = btrfs_is_free_space_inode(inode);
3200 if (!freespace_inode)
3201 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3203 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3208 if (btrfs_is_zoned(fs_info))
3209 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3210 ordered_extent->disk_num_bytes);
3212 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3214 logical_len = ordered_extent->truncated_len;
3215 /* Truncated the entire extent, don't bother adding */
3220 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3221 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3223 btrfs_inode_safe_disk_i_size_write(inode, 0);
3224 if (freespace_inode)
3225 trans = btrfs_join_transaction_spacecache(root);
3227 trans = btrfs_join_transaction(root);
3228 if (IS_ERR(trans)) {
3229 ret = PTR_ERR(trans);
3233 trans->block_rsv = &inode->block_rsv;
3234 ret = btrfs_update_inode_fallback(trans, root, inode);
3235 if (ret) /* -ENOMEM or corruption */
3236 btrfs_abort_transaction(trans, ret);
3240 clear_bits |= EXTENT_LOCKED;
3241 lock_extent(io_tree, start, end, &cached_state);
3243 if (freespace_inode)
3244 trans = btrfs_join_transaction_spacecache(root);
3246 trans = btrfs_join_transaction(root);
3247 if (IS_ERR(trans)) {
3248 ret = PTR_ERR(trans);
3253 trans->block_rsv = &inode->block_rsv;
3255 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3256 compress_type = ordered_extent->compress_type;
3257 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3258 BUG_ON(compress_type);
3259 ret = btrfs_mark_extent_written(trans, inode,
3260 ordered_extent->file_offset,
3261 ordered_extent->file_offset +
3263 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3264 ordered_extent->disk_num_bytes);
3266 BUG_ON(root == fs_info->tree_root);
3267 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3269 clear_reserved_extent = false;
3270 btrfs_release_delalloc_bytes(fs_info,
3271 ordered_extent->disk_bytenr,
3272 ordered_extent->disk_num_bytes);
3275 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3276 ordered_extent->num_bytes, trans->transid);
3278 btrfs_abort_transaction(trans, ret);
3282 ret = add_pending_csums(trans, &ordered_extent->list);
3284 btrfs_abort_transaction(trans, ret);
3289 * If this is a new delalloc range, clear its new delalloc flag to
3290 * update the inode's number of bytes. This needs to be done first
3291 * before updating the inode item.
3293 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3294 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3295 clear_extent_bit(&inode->io_tree, start, end,
3296 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3299 btrfs_inode_safe_disk_i_size_write(inode, 0);
3300 ret = btrfs_update_inode_fallback(trans, root, inode);
3301 if (ret) { /* -ENOMEM or corruption */
3302 btrfs_abort_transaction(trans, ret);
3307 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3311 btrfs_end_transaction(trans);
3313 if (ret || truncated) {
3314 u64 unwritten_start = start;
3317 * If we failed to finish this ordered extent for any reason we
3318 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3319 * extent, and mark the inode with the error if it wasn't
3320 * already set. Any error during writeback would have already
3321 * set the mapping error, so we need to set it if we're the ones
3322 * marking this ordered extent as failed.
3324 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3325 &ordered_extent->flags))
3326 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3329 unwritten_start += logical_len;
3330 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3332 /* Drop extent maps for the part of the extent we didn't write. */
3333 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3336 * If the ordered extent had an IOERR or something else went
3337 * wrong we need to return the space for this ordered extent
3338 * back to the allocator. We only free the extent in the
3339 * truncated case if we didn't write out the extent at all.
3341 * If we made it past insert_reserved_file_extent before we
3342 * errored out then we don't need to do this as the accounting
3343 * has already been done.
3345 if ((ret || !logical_len) &&
3346 clear_reserved_extent &&
3347 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3348 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3350 * Discard the range before returning it back to the
3353 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3354 btrfs_discard_extent(fs_info,
3355 ordered_extent->disk_bytenr,
3356 ordered_extent->disk_num_bytes,
3358 btrfs_free_reserved_extent(fs_info,
3359 ordered_extent->disk_bytenr,
3360 ordered_extent->disk_num_bytes, 1);
3362 * Actually free the qgroup rsv which was released when
3363 * the ordered extent was created.
3365 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3366 ordered_extent->qgroup_rsv,
3367 BTRFS_QGROUP_RSV_DATA);
3372 * This needs to be done to make sure anybody waiting knows we are done
3373 * updating everything for this ordered extent.
3375 btrfs_remove_ordered_extent(inode, ordered_extent);
3378 btrfs_put_ordered_extent(ordered_extent);
3379 /* once for the tree */
3380 btrfs_put_ordered_extent(ordered_extent);
3385 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3387 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3388 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3389 btrfs_finish_ordered_zoned(ordered);
3390 return btrfs_finish_one_ordered(ordered);
3394 * Verify the checksum for a single sector without any extra action that depend
3395 * on the type of I/O.
3397 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3398 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3400 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3403 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3405 shash->tfm = fs_info->csum_shash;
3407 kaddr = kmap_local_page(page) + pgoff;
3408 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3409 kunmap_local(kaddr);
3411 if (memcmp(csum, csum_expected, fs_info->csum_size))
3417 * Verify the checksum of a single data sector.
3419 * @bbio: btrfs_io_bio which contains the csum
3420 * @dev: device the sector is on
3421 * @bio_offset: offset to the beginning of the bio (in bytes)
3422 * @bv: bio_vec to check
3424 * Check if the checksum on a data block is valid. When a checksum mismatch is
3425 * detected, report the error and fill the corrupted range with zero.
3427 * Return %true if the sector is ok or had no checksum to start with, else %false.
3429 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3430 u32 bio_offset, struct bio_vec *bv)
3432 struct btrfs_inode *inode = bbio->inode;
3433 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3434 u64 file_offset = bbio->file_offset + bio_offset;
3435 u64 end = file_offset + bv->bv_len - 1;
3437 u8 csum[BTRFS_CSUM_SIZE];
3439 ASSERT(bv->bv_len == fs_info->sectorsize);
3444 if (btrfs_is_data_reloc_root(inode->root) &&
3445 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3447 /* Skip the range without csum for data reloc inode */
3448 clear_extent_bits(&inode->io_tree, file_offset, end,
3453 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3455 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3461 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3464 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3470 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3472 * @inode: The inode we want to perform iput on
3474 * This function uses the generic vfs_inode::i_count to track whether we should
3475 * just decrement it (in case it's > 1) or if this is the last iput then link
3476 * the inode to the delayed iput machinery. Delayed iputs are processed at
3477 * transaction commit time/superblock commit/cleaner kthread.
3479 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3481 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3482 unsigned long flags;
3484 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3487 atomic_inc(&fs_info->nr_delayed_iputs);
3489 * Need to be irq safe here because we can be called from either an irq
3490 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3493 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3494 ASSERT(list_empty(&inode->delayed_iput));
3495 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3496 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3497 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3498 wake_up_process(fs_info->cleaner_kthread);
3501 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3502 struct btrfs_inode *inode)
3504 list_del_init(&inode->delayed_iput);
3505 spin_unlock_irq(&fs_info->delayed_iput_lock);
3506 iput(&inode->vfs_inode);
3507 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3508 wake_up(&fs_info->delayed_iputs_wait);
3509 spin_lock_irq(&fs_info->delayed_iput_lock);
3512 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3513 struct btrfs_inode *inode)
3515 if (!list_empty(&inode->delayed_iput)) {
3516 spin_lock_irq(&fs_info->delayed_iput_lock);
3517 if (!list_empty(&inode->delayed_iput))
3518 run_delayed_iput_locked(fs_info, inode);
3519 spin_unlock_irq(&fs_info->delayed_iput_lock);
3523 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3526 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3527 * calls btrfs_add_delayed_iput() and that needs to lock
3528 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3529 * prevent a deadlock.
3531 spin_lock_irq(&fs_info->delayed_iput_lock);
3532 while (!list_empty(&fs_info->delayed_iputs)) {
3533 struct btrfs_inode *inode;
3535 inode = list_first_entry(&fs_info->delayed_iputs,
3536 struct btrfs_inode, delayed_iput);
3537 run_delayed_iput_locked(fs_info, inode);
3538 if (need_resched()) {
3539 spin_unlock_irq(&fs_info->delayed_iput_lock);
3541 spin_lock_irq(&fs_info->delayed_iput_lock);
3544 spin_unlock_irq(&fs_info->delayed_iput_lock);
3548 * Wait for flushing all delayed iputs
3550 * @fs_info: the filesystem
3552 * This will wait on any delayed iputs that are currently running with KILLABLE
3553 * set. Once they are all done running we will return, unless we are killed in
3554 * which case we return EINTR. This helps in user operations like fallocate etc
3555 * that might get blocked on the iputs.
3557 * Return EINTR if we were killed, 0 if nothing's pending
3559 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3561 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3562 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3569 * This creates an orphan entry for the given inode in case something goes wrong
3570 * in the middle of an unlink.
3572 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3573 struct btrfs_inode *inode)
3577 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3578 if (ret && ret != -EEXIST) {
3579 btrfs_abort_transaction(trans, ret);
3587 * We have done the delete so we can go ahead and remove the orphan item for
3588 * this particular inode.
3590 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3591 struct btrfs_inode *inode)
3593 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3597 * this cleans up any orphans that may be left on the list from the last use
3600 int btrfs_orphan_cleanup(struct btrfs_root *root)
3602 struct btrfs_fs_info *fs_info = root->fs_info;
3603 struct btrfs_path *path;
3604 struct extent_buffer *leaf;
3605 struct btrfs_key key, found_key;
3606 struct btrfs_trans_handle *trans;
3607 struct inode *inode;
3608 u64 last_objectid = 0;
3609 int ret = 0, nr_unlink = 0;
3611 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3614 path = btrfs_alloc_path();
3619 path->reada = READA_BACK;
3621 key.objectid = BTRFS_ORPHAN_OBJECTID;
3622 key.type = BTRFS_ORPHAN_ITEM_KEY;
3623 key.offset = (u64)-1;
3626 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3631 * if ret == 0 means we found what we were searching for, which
3632 * is weird, but possible, so only screw with path if we didn't
3633 * find the key and see if we have stuff that matches
3637 if (path->slots[0] == 0)
3642 /* pull out the item */
3643 leaf = path->nodes[0];
3644 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3646 /* make sure the item matches what we want */
3647 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3649 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3652 /* release the path since we're done with it */
3653 btrfs_release_path(path);
3656 * this is where we are basically btrfs_lookup, without the
3657 * crossing root thing. we store the inode number in the
3658 * offset of the orphan item.
3661 if (found_key.offset == last_objectid) {
3663 "Error removing orphan entry, stopping orphan cleanup");
3668 last_objectid = found_key.offset;
3670 found_key.objectid = found_key.offset;
3671 found_key.type = BTRFS_INODE_ITEM_KEY;
3672 found_key.offset = 0;
3673 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3674 if (IS_ERR(inode)) {
3675 ret = PTR_ERR(inode);
3681 if (!inode && root == fs_info->tree_root) {
3682 struct btrfs_root *dead_root;
3683 int is_dead_root = 0;
3686 * This is an orphan in the tree root. Currently these
3687 * could come from 2 sources:
3688 * a) a root (snapshot/subvolume) deletion in progress
3689 * b) a free space cache inode
3690 * We need to distinguish those two, as the orphan item
3691 * for a root must not get deleted before the deletion
3692 * of the snapshot/subvolume's tree completes.
3694 * btrfs_find_orphan_roots() ran before us, which has
3695 * found all deleted roots and loaded them into
3696 * fs_info->fs_roots_radix. So here we can find if an
3697 * orphan item corresponds to a deleted root by looking
3698 * up the root from that radix tree.
3701 spin_lock(&fs_info->fs_roots_radix_lock);
3702 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3703 (unsigned long)found_key.objectid);
3704 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3706 spin_unlock(&fs_info->fs_roots_radix_lock);
3709 /* prevent this orphan from being found again */
3710 key.offset = found_key.objectid - 1;
3717 * If we have an inode with links, there are a couple of
3720 * 1. We were halfway through creating fsverity metadata for the
3721 * file. In that case, the orphan item represents incomplete
3722 * fsverity metadata which must be cleaned up with
3723 * btrfs_drop_verity_items and deleting the orphan item.
3725 * 2. Old kernels (before v3.12) used to create an
3726 * orphan item for truncate indicating that there were possibly
3727 * extent items past i_size that needed to be deleted. In v3.12,
3728 * truncate was changed to update i_size in sync with the extent
3729 * items, but the (useless) orphan item was still created. Since
3730 * v4.18, we don't create the orphan item for truncate at all.
3732 * So, this item could mean that we need to do a truncate, but
3733 * only if this filesystem was last used on a pre-v3.12 kernel
3734 * and was not cleanly unmounted. The odds of that are quite
3735 * slim, and it's a pain to do the truncate now, so just delete
3738 * It's also possible that this orphan item was supposed to be
3739 * deleted but wasn't. The inode number may have been reused,
3740 * but either way, we can delete the orphan item.
3742 if (!inode || inode->i_nlink) {
3744 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3750 trans = btrfs_start_transaction(root, 1);
3751 if (IS_ERR(trans)) {
3752 ret = PTR_ERR(trans);
3755 btrfs_debug(fs_info, "auto deleting %Lu",
3756 found_key.objectid);
3757 ret = btrfs_del_orphan_item(trans, root,
3758 found_key.objectid);
3759 btrfs_end_transaction(trans);
3767 /* this will do delete_inode and everything for us */
3770 /* release the path since we're done with it */
3771 btrfs_release_path(path);
3773 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3774 trans = btrfs_join_transaction(root);
3776 btrfs_end_transaction(trans);
3780 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3784 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3785 btrfs_free_path(path);
3790 * very simple check to peek ahead in the leaf looking for xattrs. If we
3791 * don't find any xattrs, we know there can't be any acls.
3793 * slot is the slot the inode is in, objectid is the objectid of the inode
3795 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3796 int slot, u64 objectid,
3797 int *first_xattr_slot)
3799 u32 nritems = btrfs_header_nritems(leaf);
3800 struct btrfs_key found_key;
3801 static u64 xattr_access = 0;
3802 static u64 xattr_default = 0;
3805 if (!xattr_access) {
3806 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3807 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3808 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3809 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3813 *first_xattr_slot = -1;
3814 while (slot < nritems) {
3815 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3817 /* we found a different objectid, there must not be acls */
3818 if (found_key.objectid != objectid)
3821 /* we found an xattr, assume we've got an acl */
3822 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3823 if (*first_xattr_slot == -1)
3824 *first_xattr_slot = slot;
3825 if (found_key.offset == xattr_access ||
3826 found_key.offset == xattr_default)
3831 * we found a key greater than an xattr key, there can't
3832 * be any acls later on
3834 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3841 * it goes inode, inode backrefs, xattrs, extents,
3842 * so if there are a ton of hard links to an inode there can
3843 * be a lot of backrefs. Don't waste time searching too hard,
3844 * this is just an optimization
3849 /* we hit the end of the leaf before we found an xattr or
3850 * something larger than an xattr. We have to assume the inode
3853 if (*first_xattr_slot == -1)
3854 *first_xattr_slot = slot;
3859 * read an inode from the btree into the in-memory inode
3861 static int btrfs_read_locked_inode(struct inode *inode,
3862 struct btrfs_path *in_path)
3864 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3865 struct btrfs_path *path = in_path;
3866 struct extent_buffer *leaf;
3867 struct btrfs_inode_item *inode_item;
3868 struct btrfs_root *root = BTRFS_I(inode)->root;
3869 struct btrfs_key location;
3874 bool filled = false;
3875 int first_xattr_slot;
3877 ret = btrfs_fill_inode(inode, &rdev);
3882 path = btrfs_alloc_path();
3887 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3889 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3891 if (path != in_path)
3892 btrfs_free_path(path);
3896 leaf = path->nodes[0];
3901 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3902 struct btrfs_inode_item);
3903 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3904 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3905 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3906 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3907 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3908 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3909 round_up(i_size_read(inode), fs_info->sectorsize));
3911 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3912 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3914 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3915 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3917 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3918 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3920 BTRFS_I(inode)->i_otime.tv_sec =
3921 btrfs_timespec_sec(leaf, &inode_item->otime);
3922 BTRFS_I(inode)->i_otime.tv_nsec =
3923 btrfs_timespec_nsec(leaf, &inode_item->otime);
3925 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3926 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3927 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3929 inode_set_iversion_queried(inode,
3930 btrfs_inode_sequence(leaf, inode_item));
3931 inode->i_generation = BTRFS_I(inode)->generation;
3933 rdev = btrfs_inode_rdev(leaf, inode_item);
3935 BTRFS_I(inode)->index_cnt = (u64)-1;
3936 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3937 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3941 * If we were modified in the current generation and evicted from memory
3942 * and then re-read we need to do a full sync since we don't have any
3943 * idea about which extents were modified before we were evicted from
3946 * This is required for both inode re-read from disk and delayed inode
3947 * in delayed_nodes_tree.
3949 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3950 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3951 &BTRFS_I(inode)->runtime_flags);
3954 * We don't persist the id of the transaction where an unlink operation
3955 * against the inode was last made. So here we assume the inode might
3956 * have been evicted, and therefore the exact value of last_unlink_trans
3957 * lost, and set it to last_trans to avoid metadata inconsistencies
3958 * between the inode and its parent if the inode is fsync'ed and the log
3959 * replayed. For example, in the scenario:
3962 * ln mydir/foo mydir/bar
3965 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3966 * xfs_io -c fsync mydir/foo
3968 * mount fs, triggers fsync log replay
3970 * We must make sure that when we fsync our inode foo we also log its
3971 * parent inode, otherwise after log replay the parent still has the
3972 * dentry with the "bar" name but our inode foo has a link count of 1
3973 * and doesn't have an inode ref with the name "bar" anymore.
3975 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3976 * but it guarantees correctness at the expense of occasional full
3977 * transaction commits on fsync if our inode is a directory, or if our
3978 * inode is not a directory, logging its parent unnecessarily.
3980 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3983 * Same logic as for last_unlink_trans. We don't persist the generation
3984 * of the last transaction where this inode was used for a reflink
3985 * operation, so after eviction and reloading the inode we must be
3986 * pessimistic and assume the last transaction that modified the inode.
3988 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3991 if (inode->i_nlink != 1 ||
3992 path->slots[0] >= btrfs_header_nritems(leaf))
3995 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3996 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3999 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4000 if (location.type == BTRFS_INODE_REF_KEY) {
4001 struct btrfs_inode_ref *ref;
4003 ref = (struct btrfs_inode_ref *)ptr;
4004 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
4005 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
4006 struct btrfs_inode_extref *extref;
4008 extref = (struct btrfs_inode_extref *)ptr;
4009 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
4014 * try to precache a NULL acl entry for files that don't have
4015 * any xattrs or acls
4017 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
4018 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
4019 if (first_xattr_slot != -1) {
4020 path->slots[0] = first_xattr_slot;
4021 ret = btrfs_load_inode_props(inode, path);
4024 "error loading props for ino %llu (root %llu): %d",
4025 btrfs_ino(BTRFS_I(inode)),
4026 root->root_key.objectid, ret);
4028 if (path != in_path)
4029 btrfs_free_path(path);
4032 cache_no_acl(inode);
4034 switch (inode->i_mode & S_IFMT) {
4036 inode->i_mapping->a_ops = &btrfs_aops;
4037 inode->i_fop = &btrfs_file_operations;
4038 inode->i_op = &btrfs_file_inode_operations;
4041 inode->i_fop = &btrfs_dir_file_operations;
4042 inode->i_op = &btrfs_dir_inode_operations;
4045 inode->i_op = &btrfs_symlink_inode_operations;
4046 inode_nohighmem(inode);
4047 inode->i_mapping->a_ops = &btrfs_aops;
4050 inode->i_op = &btrfs_special_inode_operations;
4051 init_special_inode(inode, inode->i_mode, rdev);
4055 btrfs_sync_inode_flags_to_i_flags(inode);
4060 * given a leaf and an inode, copy the inode fields into the leaf
4062 static void fill_inode_item(struct btrfs_trans_handle *trans,
4063 struct extent_buffer *leaf,
4064 struct btrfs_inode_item *item,
4065 struct inode *inode)
4067 struct btrfs_map_token token;
4070 btrfs_init_map_token(&token, leaf);
4072 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4073 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4074 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4075 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4076 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4078 btrfs_set_token_timespec_sec(&token, &item->atime,
4079 inode->i_atime.tv_sec);
4080 btrfs_set_token_timespec_nsec(&token, &item->atime,
4081 inode->i_atime.tv_nsec);
4083 btrfs_set_token_timespec_sec(&token, &item->mtime,
4084 inode->i_mtime.tv_sec);
4085 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4086 inode->i_mtime.tv_nsec);
4088 btrfs_set_token_timespec_sec(&token, &item->ctime,
4089 inode->i_ctime.tv_sec);
4090 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4091 inode->i_ctime.tv_nsec);
4093 btrfs_set_token_timespec_sec(&token, &item->otime,
4094 BTRFS_I(inode)->i_otime.tv_sec);
4095 btrfs_set_token_timespec_nsec(&token, &item->otime,
4096 BTRFS_I(inode)->i_otime.tv_nsec);
4098 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4099 btrfs_set_token_inode_generation(&token, item,
4100 BTRFS_I(inode)->generation);
4101 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4102 btrfs_set_token_inode_transid(&token, item, trans->transid);
4103 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4104 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4105 BTRFS_I(inode)->ro_flags);
4106 btrfs_set_token_inode_flags(&token, item, flags);
4107 btrfs_set_token_inode_block_group(&token, item, 0);
4111 * copy everything in the in-memory inode into the btree.
4113 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4114 struct btrfs_root *root,
4115 struct btrfs_inode *inode)
4117 struct btrfs_inode_item *inode_item;
4118 struct btrfs_path *path;
4119 struct extent_buffer *leaf;
4122 path = btrfs_alloc_path();
4126 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4133 leaf = path->nodes[0];
4134 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4135 struct btrfs_inode_item);
4137 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4138 btrfs_mark_buffer_dirty(leaf);
4139 btrfs_set_inode_last_trans(trans, inode);
4142 btrfs_free_path(path);
4147 * copy everything in the in-memory inode into the btree.
4149 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4150 struct btrfs_root *root,
4151 struct btrfs_inode *inode)
4153 struct btrfs_fs_info *fs_info = root->fs_info;
4157 * If the inode is a free space inode, we can deadlock during commit
4158 * if we put it into the delayed code.
4160 * The data relocation inode should also be directly updated
4163 if (!btrfs_is_free_space_inode(inode)
4164 && !btrfs_is_data_reloc_root(root)
4165 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4166 btrfs_update_root_times(trans, root);
4168 ret = btrfs_delayed_update_inode(trans, root, inode);
4170 btrfs_set_inode_last_trans(trans, inode);
4174 return btrfs_update_inode_item(trans, root, inode);
4177 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4178 struct btrfs_root *root, struct btrfs_inode *inode)
4182 ret = btrfs_update_inode(trans, root, inode);
4184 return btrfs_update_inode_item(trans, root, inode);
4189 * unlink helper that gets used here in inode.c and in the tree logging
4190 * recovery code. It remove a link in a directory with a given name, and
4191 * also drops the back refs in the inode to the directory
4193 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4194 struct btrfs_inode *dir,
4195 struct btrfs_inode *inode,
4196 const struct fscrypt_str *name,
4197 struct btrfs_rename_ctx *rename_ctx)
4199 struct btrfs_root *root = dir->root;
4200 struct btrfs_fs_info *fs_info = root->fs_info;
4201 struct btrfs_path *path;
4203 struct btrfs_dir_item *di;
4205 u64 ino = btrfs_ino(inode);
4206 u64 dir_ino = btrfs_ino(dir);
4208 path = btrfs_alloc_path();
4214 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4215 if (IS_ERR_OR_NULL(di)) {
4216 ret = di ? PTR_ERR(di) : -ENOENT;
4219 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4222 btrfs_release_path(path);
4225 * If we don't have dir index, we have to get it by looking up
4226 * the inode ref, since we get the inode ref, remove it directly,
4227 * it is unnecessary to do delayed deletion.
4229 * But if we have dir index, needn't search inode ref to get it.
4230 * Since the inode ref is close to the inode item, it is better
4231 * that we delay to delete it, and just do this deletion when
4232 * we update the inode item.
4234 if (inode->dir_index) {
4235 ret = btrfs_delayed_delete_inode_ref(inode);
4237 index = inode->dir_index;
4242 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4245 "failed to delete reference to %.*s, inode %llu parent %llu",
4246 name->len, name->name, ino, dir_ino);
4247 btrfs_abort_transaction(trans, ret);
4252 rename_ctx->index = index;
4254 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4256 btrfs_abort_transaction(trans, ret);
4261 * If we are in a rename context, we don't need to update anything in the
4262 * log. That will be done later during the rename by btrfs_log_new_name().
4263 * Besides that, doing it here would only cause extra unnecessary btree
4264 * operations on the log tree, increasing latency for applications.
4267 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4268 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4272 * If we have a pending delayed iput we could end up with the final iput
4273 * being run in btrfs-cleaner context. If we have enough of these built
4274 * up we can end up burning a lot of time in btrfs-cleaner without any
4275 * way to throttle the unlinks. Since we're currently holding a ref on
4276 * the inode we can run the delayed iput here without any issues as the
4277 * final iput won't be done until after we drop the ref we're currently
4280 btrfs_run_delayed_iput(fs_info, inode);
4282 btrfs_free_path(path);
4286 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4287 inode_inc_iversion(&inode->vfs_inode);
4288 inode_inc_iversion(&dir->vfs_inode);
4289 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4290 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4291 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4292 ret = btrfs_update_inode(trans, root, dir);
4297 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4298 struct btrfs_inode *dir, struct btrfs_inode *inode,
4299 const struct fscrypt_str *name)
4303 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4305 drop_nlink(&inode->vfs_inode);
4306 ret = btrfs_update_inode(trans, inode->root, inode);
4312 * helper to start transaction for unlink and rmdir.
4314 * unlink and rmdir are special in btrfs, they do not always free space, so
4315 * if we cannot make our reservations the normal way try and see if there is
4316 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4317 * allow the unlink to occur.
4319 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4321 struct btrfs_root *root = dir->root;
4323 return btrfs_start_transaction_fallback_global_rsv(root,
4324 BTRFS_UNLINK_METADATA_UNITS);
4327 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4329 struct btrfs_trans_handle *trans;
4330 struct inode *inode = d_inode(dentry);
4332 struct fscrypt_name fname;
4334 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4338 /* This needs to handle no-key deletions later on */
4340 trans = __unlink_start_trans(BTRFS_I(dir));
4341 if (IS_ERR(trans)) {
4342 ret = PTR_ERR(trans);
4346 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4349 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4354 if (inode->i_nlink == 0) {
4355 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4361 btrfs_end_transaction(trans);
4362 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4364 fscrypt_free_filename(&fname);
4368 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4369 struct btrfs_inode *dir, struct dentry *dentry)
4371 struct btrfs_root *root = dir->root;
4372 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4373 struct btrfs_path *path;
4374 struct extent_buffer *leaf;
4375 struct btrfs_dir_item *di;
4376 struct btrfs_key key;
4380 u64 dir_ino = btrfs_ino(dir);
4381 struct fscrypt_name fname;
4383 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4387 /* This needs to handle no-key deletions later on */
4389 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4390 objectid = inode->root->root_key.objectid;
4391 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4392 objectid = inode->location.objectid;
4395 fscrypt_free_filename(&fname);
4399 path = btrfs_alloc_path();
4405 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4406 &fname.disk_name, -1);
4407 if (IS_ERR_OR_NULL(di)) {
4408 ret = di ? PTR_ERR(di) : -ENOENT;
4412 leaf = path->nodes[0];
4413 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4414 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4415 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4417 btrfs_abort_transaction(trans, ret);
4420 btrfs_release_path(path);
4423 * This is a placeholder inode for a subvolume we didn't have a
4424 * reference to at the time of the snapshot creation. In the meantime
4425 * we could have renamed the real subvol link into our snapshot, so
4426 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4427 * Instead simply lookup the dir_index_item for this entry so we can
4428 * remove it. Otherwise we know we have a ref to the root and we can
4429 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4431 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4432 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4433 if (IS_ERR_OR_NULL(di)) {
4438 btrfs_abort_transaction(trans, ret);
4442 leaf = path->nodes[0];
4443 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4445 btrfs_release_path(path);
4447 ret = btrfs_del_root_ref(trans, objectid,
4448 root->root_key.objectid, dir_ino,
4449 &index, &fname.disk_name);
4451 btrfs_abort_transaction(trans, ret);
4456 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4458 btrfs_abort_transaction(trans, ret);
4462 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4463 inode_inc_iversion(&dir->vfs_inode);
4464 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4465 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4466 ret = btrfs_update_inode_fallback(trans, root, dir);
4468 btrfs_abort_transaction(trans, ret);
4470 btrfs_free_path(path);
4471 fscrypt_free_filename(&fname);
4476 * Helper to check if the subvolume references other subvolumes or if it's
4479 static noinline int may_destroy_subvol(struct btrfs_root *root)
4481 struct btrfs_fs_info *fs_info = root->fs_info;
4482 struct btrfs_path *path;
4483 struct btrfs_dir_item *di;
4484 struct btrfs_key key;
4485 struct fscrypt_str name = FSTR_INIT("default", 7);
4489 path = btrfs_alloc_path();
4493 /* Make sure this root isn't set as the default subvol */
4494 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4495 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4497 if (di && !IS_ERR(di)) {
4498 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4499 if (key.objectid == root->root_key.objectid) {
4502 "deleting default subvolume %llu is not allowed",
4506 btrfs_release_path(path);
4509 key.objectid = root->root_key.objectid;
4510 key.type = BTRFS_ROOT_REF_KEY;
4511 key.offset = (u64)-1;
4513 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4519 if (path->slots[0] > 0) {
4521 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4522 if (key.objectid == root->root_key.objectid &&
4523 key.type == BTRFS_ROOT_REF_KEY)
4527 btrfs_free_path(path);
4531 /* Delete all dentries for inodes belonging to the root */
4532 static void btrfs_prune_dentries(struct btrfs_root *root)
4534 struct btrfs_fs_info *fs_info = root->fs_info;
4535 struct rb_node *node;
4536 struct rb_node *prev;
4537 struct btrfs_inode *entry;
4538 struct inode *inode;
4541 if (!BTRFS_FS_ERROR(fs_info))
4542 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4544 spin_lock(&root->inode_lock);
4546 node = root->inode_tree.rb_node;
4550 entry = rb_entry(node, struct btrfs_inode, rb_node);
4552 if (objectid < btrfs_ino(entry))
4553 node = node->rb_left;
4554 else if (objectid > btrfs_ino(entry))
4555 node = node->rb_right;
4561 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4562 if (objectid <= btrfs_ino(entry)) {
4566 prev = rb_next(prev);
4570 entry = rb_entry(node, struct btrfs_inode, rb_node);
4571 objectid = btrfs_ino(entry) + 1;
4572 inode = igrab(&entry->vfs_inode);
4574 spin_unlock(&root->inode_lock);
4575 if (atomic_read(&inode->i_count) > 1)
4576 d_prune_aliases(inode);
4578 * btrfs_drop_inode will have it removed from the inode
4579 * cache when its usage count hits zero.
4583 spin_lock(&root->inode_lock);
4587 if (cond_resched_lock(&root->inode_lock))
4590 node = rb_next(node);
4592 spin_unlock(&root->inode_lock);
4595 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4597 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4598 struct btrfs_root *root = dir->root;
4599 struct inode *inode = d_inode(dentry);
4600 struct btrfs_root *dest = BTRFS_I(inode)->root;
4601 struct btrfs_trans_handle *trans;
4602 struct btrfs_block_rsv block_rsv;
4607 * Don't allow to delete a subvolume with send in progress. This is
4608 * inside the inode lock so the error handling that has to drop the bit
4609 * again is not run concurrently.
4611 spin_lock(&dest->root_item_lock);
4612 if (dest->send_in_progress) {
4613 spin_unlock(&dest->root_item_lock);
4615 "attempt to delete subvolume %llu during send",
4616 dest->root_key.objectid);
4619 if (atomic_read(&dest->nr_swapfiles)) {
4620 spin_unlock(&dest->root_item_lock);
4622 "attempt to delete subvolume %llu with active swapfile",
4623 root->root_key.objectid);
4626 root_flags = btrfs_root_flags(&dest->root_item);
4627 btrfs_set_root_flags(&dest->root_item,
4628 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4629 spin_unlock(&dest->root_item_lock);
4631 down_write(&fs_info->subvol_sem);
4633 ret = may_destroy_subvol(dest);
4637 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4639 * One for dir inode,
4640 * two for dir entries,
4641 * two for root ref/backref.
4643 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4647 trans = btrfs_start_transaction(root, 0);
4648 if (IS_ERR(trans)) {
4649 ret = PTR_ERR(trans);
4652 trans->block_rsv = &block_rsv;
4653 trans->bytes_reserved = block_rsv.size;
4655 btrfs_record_snapshot_destroy(trans, dir);
4657 ret = btrfs_unlink_subvol(trans, dir, dentry);
4659 btrfs_abort_transaction(trans, ret);
4663 ret = btrfs_record_root_in_trans(trans, dest);
4665 btrfs_abort_transaction(trans, ret);
4669 memset(&dest->root_item.drop_progress, 0,
4670 sizeof(dest->root_item.drop_progress));
4671 btrfs_set_root_drop_level(&dest->root_item, 0);
4672 btrfs_set_root_refs(&dest->root_item, 0);
4674 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4675 ret = btrfs_insert_orphan_item(trans,
4677 dest->root_key.objectid);
4679 btrfs_abort_transaction(trans, ret);
4684 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4685 BTRFS_UUID_KEY_SUBVOL,
4686 dest->root_key.objectid);
4687 if (ret && ret != -ENOENT) {
4688 btrfs_abort_transaction(trans, ret);
4691 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4692 ret = btrfs_uuid_tree_remove(trans,
4693 dest->root_item.received_uuid,
4694 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4695 dest->root_key.objectid);
4696 if (ret && ret != -ENOENT) {
4697 btrfs_abort_transaction(trans, ret);
4702 free_anon_bdev(dest->anon_dev);
4705 trans->block_rsv = NULL;
4706 trans->bytes_reserved = 0;
4707 ret = btrfs_end_transaction(trans);
4708 inode->i_flags |= S_DEAD;
4710 btrfs_subvolume_release_metadata(root, &block_rsv);
4712 up_write(&fs_info->subvol_sem);
4714 spin_lock(&dest->root_item_lock);
4715 root_flags = btrfs_root_flags(&dest->root_item);
4716 btrfs_set_root_flags(&dest->root_item,
4717 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4718 spin_unlock(&dest->root_item_lock);
4720 d_invalidate(dentry);
4721 btrfs_prune_dentries(dest);
4722 ASSERT(dest->send_in_progress == 0);
4728 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4730 struct inode *inode = d_inode(dentry);
4731 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4733 struct btrfs_trans_handle *trans;
4734 u64 last_unlink_trans;
4735 struct fscrypt_name fname;
4737 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4739 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4740 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4742 "extent tree v2 doesn't support snapshot deletion yet");
4745 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4748 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4752 /* This needs to handle no-key deletions later on */
4754 trans = __unlink_start_trans(BTRFS_I(dir));
4755 if (IS_ERR(trans)) {
4756 err = PTR_ERR(trans);
4760 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4761 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4765 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4769 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4771 /* now the directory is empty */
4772 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4775 btrfs_i_size_write(BTRFS_I(inode), 0);
4777 * Propagate the last_unlink_trans value of the deleted dir to
4778 * its parent directory. This is to prevent an unrecoverable
4779 * log tree in the case we do something like this:
4781 * 2) create snapshot under dir foo
4782 * 3) delete the snapshot
4785 * 6) fsync foo or some file inside foo
4787 if (last_unlink_trans >= trans->transid)
4788 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4791 btrfs_end_transaction(trans);
4793 btrfs_btree_balance_dirty(fs_info);
4794 fscrypt_free_filename(&fname);
4800 * btrfs_truncate_block - read, zero a chunk and write a block
4801 * @inode - inode that we're zeroing
4802 * @from - the offset to start zeroing
4803 * @len - the length to zero, 0 to zero the entire range respective to the
4805 * @front - zero up to the offset instead of from the offset on
4807 * This will find the block for the "from" offset and cow the block and zero the
4808 * part we want to zero. This is used with truncate and hole punching.
4810 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4813 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4814 struct address_space *mapping = inode->vfs_inode.i_mapping;
4815 struct extent_io_tree *io_tree = &inode->io_tree;
4816 struct btrfs_ordered_extent *ordered;
4817 struct extent_state *cached_state = NULL;
4818 struct extent_changeset *data_reserved = NULL;
4819 bool only_release_metadata = false;
4820 u32 blocksize = fs_info->sectorsize;
4821 pgoff_t index = from >> PAGE_SHIFT;
4822 unsigned offset = from & (blocksize - 1);
4824 gfp_t mask = btrfs_alloc_write_mask(mapping);
4825 size_t write_bytes = blocksize;
4830 if (IS_ALIGNED(offset, blocksize) &&
4831 (!len || IS_ALIGNED(len, blocksize)))
4834 block_start = round_down(from, blocksize);
4835 block_end = block_start + blocksize - 1;
4837 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4840 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4841 /* For nocow case, no need to reserve data space */
4842 only_release_metadata = true;
4847 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4849 if (!only_release_metadata)
4850 btrfs_free_reserved_data_space(inode, data_reserved,
4851 block_start, blocksize);
4855 page = find_or_create_page(mapping, index, mask);
4857 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4859 btrfs_delalloc_release_extents(inode, blocksize);
4864 if (!PageUptodate(page)) {
4865 ret = btrfs_read_folio(NULL, page_folio(page));
4867 if (page->mapping != mapping) {
4872 if (!PageUptodate(page)) {
4879 * We unlock the page after the io is completed and then re-lock it
4880 * above. release_folio() could have come in between that and cleared
4881 * PagePrivate(), but left the page in the mapping. Set the page mapped
4882 * here to make sure it's properly set for the subpage stuff.
4884 ret = set_page_extent_mapped(page);
4888 wait_on_page_writeback(page);
4890 lock_extent(io_tree, block_start, block_end, &cached_state);
4892 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4894 unlock_extent(io_tree, block_start, block_end, &cached_state);
4897 btrfs_start_ordered_extent(ordered);
4898 btrfs_put_ordered_extent(ordered);
4902 clear_extent_bit(&inode->io_tree, block_start, block_end,
4903 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4906 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4909 unlock_extent(io_tree, block_start, block_end, &cached_state);
4913 if (offset != blocksize) {
4915 len = blocksize - offset;
4917 memzero_page(page, (block_start - page_offset(page)),
4920 memzero_page(page, (block_start - page_offset(page)) + offset,
4923 btrfs_page_clear_checked(fs_info, page, block_start,
4924 block_end + 1 - block_start);
4925 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4926 unlock_extent(io_tree, block_start, block_end, &cached_state);
4928 if (only_release_metadata)
4929 set_extent_bit(&inode->io_tree, block_start, block_end,
4930 EXTENT_NORESERVE, NULL);
4934 if (only_release_metadata)
4935 btrfs_delalloc_release_metadata(inode, blocksize, true);
4937 btrfs_delalloc_release_space(inode, data_reserved,
4938 block_start, blocksize, true);
4940 btrfs_delalloc_release_extents(inode, blocksize);
4944 if (only_release_metadata)
4945 btrfs_check_nocow_unlock(inode);
4946 extent_changeset_free(data_reserved);
4950 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4951 u64 offset, u64 len)
4953 struct btrfs_fs_info *fs_info = root->fs_info;
4954 struct btrfs_trans_handle *trans;
4955 struct btrfs_drop_extents_args drop_args = { 0 };
4959 * If NO_HOLES is enabled, we don't need to do anything.
4960 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4961 * or btrfs_update_inode() will be called, which guarantee that the next
4962 * fsync will know this inode was changed and needs to be logged.
4964 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4968 * 1 - for the one we're dropping
4969 * 1 - for the one we're adding
4970 * 1 - for updating the inode.
4972 trans = btrfs_start_transaction(root, 3);
4974 return PTR_ERR(trans);
4976 drop_args.start = offset;
4977 drop_args.end = offset + len;
4978 drop_args.drop_cache = true;
4980 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4982 btrfs_abort_transaction(trans, ret);
4983 btrfs_end_transaction(trans);
4987 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4989 btrfs_abort_transaction(trans, ret);
4991 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4992 btrfs_update_inode(trans, root, inode);
4994 btrfs_end_transaction(trans);
4999 * This function puts in dummy file extents for the area we're creating a hole
5000 * for. So if we are truncating this file to a larger size we need to insert
5001 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5002 * the range between oldsize and size
5004 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
5006 struct btrfs_root *root = inode->root;
5007 struct btrfs_fs_info *fs_info = root->fs_info;
5008 struct extent_io_tree *io_tree = &inode->io_tree;
5009 struct extent_map *em = NULL;
5010 struct extent_state *cached_state = NULL;
5011 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5012 u64 block_end = ALIGN(size, fs_info->sectorsize);
5019 * If our size started in the middle of a block we need to zero out the
5020 * rest of the block before we expand the i_size, otherwise we could
5021 * expose stale data.
5023 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5027 if (size <= hole_start)
5030 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5032 cur_offset = hole_start;
5034 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5035 block_end - cur_offset);
5041 last_byte = min(extent_map_end(em), block_end);
5042 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5043 hole_size = last_byte - cur_offset;
5045 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5046 struct extent_map *hole_em;
5048 err = maybe_insert_hole(root, inode, cur_offset,
5053 err = btrfs_inode_set_file_extent_range(inode,
5054 cur_offset, hole_size);
5058 hole_em = alloc_extent_map();
5060 btrfs_drop_extent_map_range(inode, cur_offset,
5061 cur_offset + hole_size - 1,
5063 btrfs_set_inode_full_sync(inode);
5066 hole_em->start = cur_offset;
5067 hole_em->len = hole_size;
5068 hole_em->orig_start = cur_offset;
5070 hole_em->block_start = EXTENT_MAP_HOLE;
5071 hole_em->block_len = 0;
5072 hole_em->orig_block_len = 0;
5073 hole_em->ram_bytes = hole_size;
5074 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5075 hole_em->generation = fs_info->generation;
5077 err = btrfs_replace_extent_map_range(inode, hole_em, true);
5078 free_extent_map(hole_em);
5080 err = btrfs_inode_set_file_extent_range(inode,
5081 cur_offset, hole_size);
5086 free_extent_map(em);
5088 cur_offset = last_byte;
5089 if (cur_offset >= block_end)
5092 free_extent_map(em);
5093 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5097 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5099 struct btrfs_root *root = BTRFS_I(inode)->root;
5100 struct btrfs_trans_handle *trans;
5101 loff_t oldsize = i_size_read(inode);
5102 loff_t newsize = attr->ia_size;
5103 int mask = attr->ia_valid;
5107 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5108 * special case where we need to update the times despite not having
5109 * these flags set. For all other operations the VFS set these flags
5110 * explicitly if it wants a timestamp update.
5112 if (newsize != oldsize) {
5113 inode_inc_iversion(inode);
5114 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5115 inode->i_mtime = current_time(inode);
5116 inode->i_ctime = inode->i_mtime;
5120 if (newsize > oldsize) {
5122 * Don't do an expanding truncate while snapshotting is ongoing.
5123 * This is to ensure the snapshot captures a fully consistent
5124 * state of this file - if the snapshot captures this expanding
5125 * truncation, it must capture all writes that happened before
5128 btrfs_drew_write_lock(&root->snapshot_lock);
5129 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5131 btrfs_drew_write_unlock(&root->snapshot_lock);
5135 trans = btrfs_start_transaction(root, 1);
5136 if (IS_ERR(trans)) {
5137 btrfs_drew_write_unlock(&root->snapshot_lock);
5138 return PTR_ERR(trans);
5141 i_size_write(inode, newsize);
5142 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5143 pagecache_isize_extended(inode, oldsize, newsize);
5144 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5145 btrfs_drew_write_unlock(&root->snapshot_lock);
5146 btrfs_end_transaction(trans);
5148 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5150 if (btrfs_is_zoned(fs_info)) {
5151 ret = btrfs_wait_ordered_range(inode,
5152 ALIGN(newsize, fs_info->sectorsize),
5159 * We're truncating a file that used to have good data down to
5160 * zero. Make sure any new writes to the file get on disk
5164 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5165 &BTRFS_I(inode)->runtime_flags);
5167 truncate_setsize(inode, newsize);
5169 inode_dio_wait(inode);
5171 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5172 if (ret && inode->i_nlink) {
5176 * Truncate failed, so fix up the in-memory size. We
5177 * adjusted disk_i_size down as we removed extents, so
5178 * wait for disk_i_size to be stable and then update the
5179 * in-memory size to match.
5181 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5184 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5191 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5194 struct inode *inode = d_inode(dentry);
5195 struct btrfs_root *root = BTRFS_I(inode)->root;
5198 if (btrfs_root_readonly(root))
5201 err = setattr_prepare(idmap, dentry, attr);
5205 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5206 err = btrfs_setsize(inode, attr);
5211 if (attr->ia_valid) {
5212 setattr_copy(idmap, inode, attr);
5213 inode_inc_iversion(inode);
5214 err = btrfs_dirty_inode(BTRFS_I(inode));
5216 if (!err && attr->ia_valid & ATTR_MODE)
5217 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5224 * While truncating the inode pages during eviction, we get the VFS
5225 * calling btrfs_invalidate_folio() against each folio of the inode. This
5226 * is slow because the calls to btrfs_invalidate_folio() result in a
5227 * huge amount of calls to lock_extent() and clear_extent_bit(),
5228 * which keep merging and splitting extent_state structures over and over,
5229 * wasting lots of time.
5231 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5232 * skip all those expensive operations on a per folio basis and do only
5233 * the ordered io finishing, while we release here the extent_map and
5234 * extent_state structures, without the excessive merging and splitting.
5236 static void evict_inode_truncate_pages(struct inode *inode)
5238 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5239 struct rb_node *node;
5241 ASSERT(inode->i_state & I_FREEING);
5242 truncate_inode_pages_final(&inode->i_data);
5244 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5247 * Keep looping until we have no more ranges in the io tree.
5248 * We can have ongoing bios started by readahead that have
5249 * their endio callback (extent_io.c:end_bio_extent_readpage)
5250 * still in progress (unlocked the pages in the bio but did not yet
5251 * unlocked the ranges in the io tree). Therefore this means some
5252 * ranges can still be locked and eviction started because before
5253 * submitting those bios, which are executed by a separate task (work
5254 * queue kthread), inode references (inode->i_count) were not taken
5255 * (which would be dropped in the end io callback of each bio).
5256 * Therefore here we effectively end up waiting for those bios and
5257 * anyone else holding locked ranges without having bumped the inode's
5258 * reference count - if we don't do it, when they access the inode's
5259 * io_tree to unlock a range it may be too late, leading to an
5260 * use-after-free issue.
5262 spin_lock(&io_tree->lock);
5263 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5264 struct extent_state *state;
5265 struct extent_state *cached_state = NULL;
5268 unsigned state_flags;
5270 node = rb_first(&io_tree->state);
5271 state = rb_entry(node, struct extent_state, rb_node);
5272 start = state->start;
5274 state_flags = state->state;
5275 spin_unlock(&io_tree->lock);
5277 lock_extent(io_tree, start, end, &cached_state);
5280 * If still has DELALLOC flag, the extent didn't reach disk,
5281 * and its reserved space won't be freed by delayed_ref.
5282 * So we need to free its reserved space here.
5283 * (Refer to comment in btrfs_invalidate_folio, case 2)
5285 * Note, end is the bytenr of last byte, so we need + 1 here.
5287 if (state_flags & EXTENT_DELALLOC)
5288 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5291 clear_extent_bit(io_tree, start, end,
5292 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5296 spin_lock(&io_tree->lock);
5298 spin_unlock(&io_tree->lock);
5301 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5302 struct btrfs_block_rsv *rsv)
5304 struct btrfs_fs_info *fs_info = root->fs_info;
5305 struct btrfs_trans_handle *trans;
5306 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5310 * Eviction should be taking place at some place safe because of our
5311 * delayed iputs. However the normal flushing code will run delayed
5312 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5314 * We reserve the delayed_refs_extra here again because we can't use
5315 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5316 * above. We reserve our extra bit here because we generate a ton of
5317 * delayed refs activity by truncating.
5319 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5320 * if we fail to make this reservation we can re-try without the
5321 * delayed_refs_extra so we can make some forward progress.
5323 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5324 BTRFS_RESERVE_FLUSH_EVICT);
5326 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5327 BTRFS_RESERVE_FLUSH_EVICT);
5330 "could not allocate space for delete; will truncate on mount");
5331 return ERR_PTR(-ENOSPC);
5333 delayed_refs_extra = 0;
5336 trans = btrfs_join_transaction(root);
5340 if (delayed_refs_extra) {
5341 trans->block_rsv = &fs_info->trans_block_rsv;
5342 trans->bytes_reserved = delayed_refs_extra;
5343 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5344 delayed_refs_extra, true);
5349 void btrfs_evict_inode(struct inode *inode)
5351 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5352 struct btrfs_trans_handle *trans;
5353 struct btrfs_root *root = BTRFS_I(inode)->root;
5354 struct btrfs_block_rsv *rsv = NULL;
5357 trace_btrfs_inode_evict(inode);
5360 fsverity_cleanup_inode(inode);
5365 evict_inode_truncate_pages(inode);
5367 if (inode->i_nlink &&
5368 ((btrfs_root_refs(&root->root_item) != 0 &&
5369 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5370 btrfs_is_free_space_inode(BTRFS_I(inode))))
5373 if (is_bad_inode(inode))
5376 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5379 if (inode->i_nlink > 0) {
5380 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5381 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5386 * This makes sure the inode item in tree is uptodate and the space for
5387 * the inode update is released.
5389 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5394 * This drops any pending insert or delete operations we have for this
5395 * inode. We could have a delayed dir index deletion queued up, but
5396 * we're removing the inode completely so that'll be taken care of in
5399 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5401 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5404 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5405 rsv->failfast = true;
5407 btrfs_i_size_write(BTRFS_I(inode), 0);
5410 struct btrfs_truncate_control control = {
5411 .inode = BTRFS_I(inode),
5412 .ino = btrfs_ino(BTRFS_I(inode)),
5417 trans = evict_refill_and_join(root, rsv);
5421 trans->block_rsv = rsv;
5423 ret = btrfs_truncate_inode_items(trans, root, &control);
5424 trans->block_rsv = &fs_info->trans_block_rsv;
5425 btrfs_end_transaction(trans);
5427 * We have not added new delayed items for our inode after we
5428 * have flushed its delayed items, so no need to throttle on
5429 * delayed items. However we have modified extent buffers.
5431 btrfs_btree_balance_dirty_nodelay(fs_info);
5432 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5439 * Errors here aren't a big deal, it just means we leave orphan items in
5440 * the tree. They will be cleaned up on the next mount. If the inode
5441 * number gets reused, cleanup deletes the orphan item without doing
5442 * anything, and unlink reuses the existing orphan item.
5444 * If it turns out that we are dropping too many of these, we might want
5445 * to add a mechanism for retrying these after a commit.
5447 trans = evict_refill_and_join(root, rsv);
5448 if (!IS_ERR(trans)) {
5449 trans->block_rsv = rsv;
5450 btrfs_orphan_del(trans, BTRFS_I(inode));
5451 trans->block_rsv = &fs_info->trans_block_rsv;
5452 btrfs_end_transaction(trans);
5456 btrfs_free_block_rsv(fs_info, rsv);
5458 * If we didn't successfully delete, the orphan item will still be in
5459 * the tree and we'll retry on the next mount. Again, we might also want
5460 * to retry these periodically in the future.
5462 btrfs_remove_delayed_node(BTRFS_I(inode));
5463 fsverity_cleanup_inode(inode);
5468 * Return the key found in the dir entry in the location pointer, fill @type
5469 * with BTRFS_FT_*, and return 0.
5471 * If no dir entries were found, returns -ENOENT.
5472 * If found a corrupted location in dir entry, returns -EUCLEAN.
5474 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5475 struct btrfs_key *location, u8 *type)
5477 struct btrfs_dir_item *di;
5478 struct btrfs_path *path;
5479 struct btrfs_root *root = dir->root;
5481 struct fscrypt_name fname;
5483 path = btrfs_alloc_path();
5487 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5491 * fscrypt_setup_filename() should never return a positive value, but
5492 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5496 /* This needs to handle no-key deletions later on */
5498 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5499 &fname.disk_name, 0);
5500 if (IS_ERR_OR_NULL(di)) {
5501 ret = di ? PTR_ERR(di) : -ENOENT;
5505 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5506 if (location->type != BTRFS_INODE_ITEM_KEY &&
5507 location->type != BTRFS_ROOT_ITEM_KEY) {
5509 btrfs_warn(root->fs_info,
5510 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5511 __func__, fname.disk_name.name, btrfs_ino(dir),
5512 location->objectid, location->type, location->offset);
5515 *type = btrfs_dir_ftype(path->nodes[0], di);
5517 fscrypt_free_filename(&fname);
5518 btrfs_free_path(path);
5523 * when we hit a tree root in a directory, the btrfs part of the inode
5524 * needs to be changed to reflect the root directory of the tree root. This
5525 * is kind of like crossing a mount point.
5527 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5528 struct btrfs_inode *dir,
5529 struct dentry *dentry,
5530 struct btrfs_key *location,
5531 struct btrfs_root **sub_root)
5533 struct btrfs_path *path;
5534 struct btrfs_root *new_root;
5535 struct btrfs_root_ref *ref;
5536 struct extent_buffer *leaf;
5537 struct btrfs_key key;
5540 struct fscrypt_name fname;
5542 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5546 path = btrfs_alloc_path();
5553 key.objectid = dir->root->root_key.objectid;
5554 key.type = BTRFS_ROOT_REF_KEY;
5555 key.offset = location->objectid;
5557 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5564 leaf = path->nodes[0];
5565 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5566 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5567 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5570 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5571 (unsigned long)(ref + 1), fname.disk_name.len);
5575 btrfs_release_path(path);
5577 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5578 if (IS_ERR(new_root)) {
5579 err = PTR_ERR(new_root);
5583 *sub_root = new_root;
5584 location->objectid = btrfs_root_dirid(&new_root->root_item);
5585 location->type = BTRFS_INODE_ITEM_KEY;
5586 location->offset = 0;
5589 btrfs_free_path(path);
5590 fscrypt_free_filename(&fname);
5594 static void inode_tree_add(struct btrfs_inode *inode)
5596 struct btrfs_root *root = inode->root;
5597 struct btrfs_inode *entry;
5599 struct rb_node *parent;
5600 struct rb_node *new = &inode->rb_node;
5601 u64 ino = btrfs_ino(inode);
5603 if (inode_unhashed(&inode->vfs_inode))
5606 spin_lock(&root->inode_lock);
5607 p = &root->inode_tree.rb_node;
5610 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5612 if (ino < btrfs_ino(entry))
5613 p = &parent->rb_left;
5614 else if (ino > btrfs_ino(entry))
5615 p = &parent->rb_right;
5617 WARN_ON(!(entry->vfs_inode.i_state &
5618 (I_WILL_FREE | I_FREEING)));
5619 rb_replace_node(parent, new, &root->inode_tree);
5620 RB_CLEAR_NODE(parent);
5621 spin_unlock(&root->inode_lock);
5625 rb_link_node(new, parent, p);
5626 rb_insert_color(new, &root->inode_tree);
5627 spin_unlock(&root->inode_lock);
5630 static void inode_tree_del(struct btrfs_inode *inode)
5632 struct btrfs_root *root = inode->root;
5635 spin_lock(&root->inode_lock);
5636 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5637 rb_erase(&inode->rb_node, &root->inode_tree);
5638 RB_CLEAR_NODE(&inode->rb_node);
5639 empty = RB_EMPTY_ROOT(&root->inode_tree);
5641 spin_unlock(&root->inode_lock);
5643 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5644 spin_lock(&root->inode_lock);
5645 empty = RB_EMPTY_ROOT(&root->inode_tree);
5646 spin_unlock(&root->inode_lock);
5648 btrfs_add_dead_root(root);
5653 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5655 struct btrfs_iget_args *args = p;
5657 inode->i_ino = args->ino;
5658 BTRFS_I(inode)->location.objectid = args->ino;
5659 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5660 BTRFS_I(inode)->location.offset = 0;
5661 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5662 BUG_ON(args->root && !BTRFS_I(inode)->root);
5664 if (args->root && args->root == args->root->fs_info->tree_root &&
5665 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5666 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5667 &BTRFS_I(inode)->runtime_flags);
5671 static int btrfs_find_actor(struct inode *inode, void *opaque)
5673 struct btrfs_iget_args *args = opaque;
5675 return args->ino == BTRFS_I(inode)->location.objectid &&
5676 args->root == BTRFS_I(inode)->root;
5679 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5680 struct btrfs_root *root)
5682 struct inode *inode;
5683 struct btrfs_iget_args args;
5684 unsigned long hashval = btrfs_inode_hash(ino, root);
5689 inode = iget5_locked(s, hashval, btrfs_find_actor,
5690 btrfs_init_locked_inode,
5696 * Get an inode object given its inode number and corresponding root.
5697 * Path can be preallocated to prevent recursing back to iget through
5698 * allocator. NULL is also valid but may require an additional allocation
5701 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5702 struct btrfs_root *root, struct btrfs_path *path)
5704 struct inode *inode;
5706 inode = btrfs_iget_locked(s, ino, root);
5708 return ERR_PTR(-ENOMEM);
5710 if (inode->i_state & I_NEW) {
5713 ret = btrfs_read_locked_inode(inode, path);
5715 inode_tree_add(BTRFS_I(inode));
5716 unlock_new_inode(inode);
5720 * ret > 0 can come from btrfs_search_slot called by
5721 * btrfs_read_locked_inode, this means the inode item
5726 inode = ERR_PTR(ret);
5733 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5735 return btrfs_iget_path(s, ino, root, NULL);
5738 static struct inode *new_simple_dir(struct super_block *s,
5739 struct btrfs_key *key,
5740 struct btrfs_root *root)
5742 struct inode *inode = new_inode(s);
5745 return ERR_PTR(-ENOMEM);
5747 BTRFS_I(inode)->root = btrfs_grab_root(root);
5748 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5749 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5751 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5753 * We only need lookup, the rest is read-only and there's no inode
5754 * associated with the dentry
5756 inode->i_op = &simple_dir_inode_operations;
5757 inode->i_opflags &= ~IOP_XATTR;
5758 inode->i_fop = &simple_dir_operations;
5759 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5760 inode->i_mtime = current_time(inode);
5761 inode->i_atime = inode->i_mtime;
5762 inode->i_ctime = inode->i_mtime;
5763 BTRFS_I(inode)->i_otime = inode->i_mtime;
5768 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5769 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5770 static_assert(BTRFS_FT_DIR == FT_DIR);
5771 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5772 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5773 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5774 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5775 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5777 static inline u8 btrfs_inode_type(struct inode *inode)
5779 return fs_umode_to_ftype(inode->i_mode);
5782 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5784 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5785 struct inode *inode;
5786 struct btrfs_root *root = BTRFS_I(dir)->root;
5787 struct btrfs_root *sub_root = root;
5788 struct btrfs_key location;
5792 if (dentry->d_name.len > BTRFS_NAME_LEN)
5793 return ERR_PTR(-ENAMETOOLONG);
5795 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5797 return ERR_PTR(ret);
5799 if (location.type == BTRFS_INODE_ITEM_KEY) {
5800 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5804 /* Do extra check against inode mode with di_type */
5805 if (btrfs_inode_type(inode) != di_type) {
5807 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5808 inode->i_mode, btrfs_inode_type(inode),
5811 return ERR_PTR(-EUCLEAN);
5816 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5817 &location, &sub_root);
5820 inode = ERR_PTR(ret);
5822 inode = new_simple_dir(dir->i_sb, &location, root);
5824 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5825 btrfs_put_root(sub_root);
5830 down_read(&fs_info->cleanup_work_sem);
5831 if (!sb_rdonly(inode->i_sb))
5832 ret = btrfs_orphan_cleanup(sub_root);
5833 up_read(&fs_info->cleanup_work_sem);
5836 inode = ERR_PTR(ret);
5843 static int btrfs_dentry_delete(const struct dentry *dentry)
5845 struct btrfs_root *root;
5846 struct inode *inode = d_inode(dentry);
5848 if (!inode && !IS_ROOT(dentry))
5849 inode = d_inode(dentry->d_parent);
5852 root = BTRFS_I(inode)->root;
5853 if (btrfs_root_refs(&root->root_item) == 0)
5856 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5862 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5865 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5867 if (inode == ERR_PTR(-ENOENT))
5869 return d_splice_alias(inode, dentry);
5873 * Find the highest existing sequence number in a directory and then set the
5874 * in-memory index_cnt variable to the first free sequence number.
5876 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5878 struct btrfs_root *root = inode->root;
5879 struct btrfs_key key, found_key;
5880 struct btrfs_path *path;
5881 struct extent_buffer *leaf;
5884 key.objectid = btrfs_ino(inode);
5885 key.type = BTRFS_DIR_INDEX_KEY;
5886 key.offset = (u64)-1;
5888 path = btrfs_alloc_path();
5892 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5895 /* FIXME: we should be able to handle this */
5900 if (path->slots[0] == 0) {
5901 inode->index_cnt = BTRFS_DIR_START_INDEX;
5907 leaf = path->nodes[0];
5908 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5910 if (found_key.objectid != btrfs_ino(inode) ||
5911 found_key.type != BTRFS_DIR_INDEX_KEY) {
5912 inode->index_cnt = BTRFS_DIR_START_INDEX;
5916 inode->index_cnt = found_key.offset + 1;
5918 btrfs_free_path(path);
5922 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5924 if (dir->index_cnt == (u64)-1) {
5927 ret = btrfs_inode_delayed_dir_index_count(dir);
5929 ret = btrfs_set_inode_index_count(dir);
5935 *index = dir->index_cnt;
5941 * All this infrastructure exists because dir_emit can fault, and we are holding
5942 * the tree lock when doing readdir. For now just allocate a buffer and copy
5943 * our information into that, and then dir_emit from the buffer. This is
5944 * similar to what NFS does, only we don't keep the buffer around in pagecache
5945 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5946 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5949 static int btrfs_opendir(struct inode *inode, struct file *file)
5951 struct btrfs_file_private *private;
5955 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5959 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5962 private->last_index = last_index;
5963 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5964 if (!private->filldir_buf) {
5968 file->private_data = private;
5979 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5982 struct dir_entry *entry = addr;
5983 char *name = (char *)(entry + 1);
5985 ctx->pos = get_unaligned(&entry->offset);
5986 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5987 get_unaligned(&entry->ino),
5988 get_unaligned(&entry->type)))
5990 addr += sizeof(struct dir_entry) +
5991 get_unaligned(&entry->name_len);
5997 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5999 struct inode *inode = file_inode(file);
6000 struct btrfs_root *root = BTRFS_I(inode)->root;
6001 struct btrfs_file_private *private = file->private_data;
6002 struct btrfs_dir_item *di;
6003 struct btrfs_key key;
6004 struct btrfs_key found_key;
6005 struct btrfs_path *path;
6007 struct list_head ins_list;
6008 struct list_head del_list;
6015 struct btrfs_key location;
6017 if (!dir_emit_dots(file, ctx))
6020 path = btrfs_alloc_path();
6024 addr = private->filldir_buf;
6025 path->reada = READA_FORWARD;
6027 INIT_LIST_HEAD(&ins_list);
6028 INIT_LIST_HEAD(&del_list);
6029 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
6030 &ins_list, &del_list);
6033 key.type = BTRFS_DIR_INDEX_KEY;
6034 key.offset = ctx->pos;
6035 key.objectid = btrfs_ino(BTRFS_I(inode));
6037 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
6038 struct dir_entry *entry;
6039 struct extent_buffer *leaf = path->nodes[0];
6042 if (found_key.objectid != key.objectid)
6044 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6046 if (found_key.offset < ctx->pos)
6048 if (found_key.offset > private->last_index)
6050 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6052 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
6053 name_len = btrfs_dir_name_len(leaf, di);
6054 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6056 btrfs_release_path(path);
6057 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6060 addr = private->filldir_buf;
6066 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
6068 name_ptr = (char *)(entry + 1);
6069 read_extent_buffer(leaf, name_ptr,
6070 (unsigned long)(di + 1), name_len);
6071 put_unaligned(name_len, &entry->name_len);
6072 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
6073 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6074 put_unaligned(location.objectid, &entry->ino);
6075 put_unaligned(found_key.offset, &entry->offset);
6077 addr += sizeof(struct dir_entry) + name_len;
6078 total_len += sizeof(struct dir_entry) + name_len;
6080 /* Catch error encountered during iteration */
6084 btrfs_release_path(path);
6086 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6090 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6095 * Stop new entries from being returned after we return the last
6098 * New directory entries are assigned a strictly increasing
6099 * offset. This means that new entries created during readdir
6100 * are *guaranteed* to be seen in the future by that readdir.
6101 * This has broken buggy programs which operate on names as
6102 * they're returned by readdir. Until we re-use freed offsets
6103 * we have this hack to stop new entries from being returned
6104 * under the assumption that they'll never reach this huge
6107 * This is being careful not to overflow 32bit loff_t unless the
6108 * last entry requires it because doing so has broken 32bit apps
6111 if (ctx->pos >= INT_MAX)
6112 ctx->pos = LLONG_MAX;
6119 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6120 btrfs_free_path(path);
6125 * This is somewhat expensive, updating the tree every time the
6126 * inode changes. But, it is most likely to find the inode in cache.
6127 * FIXME, needs more benchmarking...there are no reasons other than performance
6128 * to keep or drop this code.
6130 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6132 struct btrfs_root *root = inode->root;
6133 struct btrfs_fs_info *fs_info = root->fs_info;
6134 struct btrfs_trans_handle *trans;
6137 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6140 trans = btrfs_join_transaction(root);
6142 return PTR_ERR(trans);
6144 ret = btrfs_update_inode(trans, root, inode);
6145 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6146 /* whoops, lets try again with the full transaction */
6147 btrfs_end_transaction(trans);
6148 trans = btrfs_start_transaction(root, 1);
6150 return PTR_ERR(trans);
6152 ret = btrfs_update_inode(trans, root, inode);
6154 btrfs_end_transaction(trans);
6155 if (inode->delayed_node)
6156 btrfs_balance_delayed_items(fs_info);
6162 * This is a copy of file_update_time. We need this so we can return error on
6163 * ENOSPC for updating the inode in the case of file write and mmap writes.
6165 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6168 struct btrfs_root *root = BTRFS_I(inode)->root;
6169 bool dirty = flags & ~S_VERSION;
6171 if (btrfs_root_readonly(root))
6174 if (flags & S_VERSION)
6175 dirty |= inode_maybe_inc_iversion(inode, dirty);
6176 if (flags & S_CTIME)
6177 inode->i_ctime = *now;
6178 if (flags & S_MTIME)
6179 inode->i_mtime = *now;
6180 if (flags & S_ATIME)
6181 inode->i_atime = *now;
6182 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6186 * helper to find a free sequence number in a given directory. This current
6187 * code is very simple, later versions will do smarter things in the btree
6189 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6193 if (dir->index_cnt == (u64)-1) {
6194 ret = btrfs_inode_delayed_dir_index_count(dir);
6196 ret = btrfs_set_inode_index_count(dir);
6202 *index = dir->index_cnt;
6208 static int btrfs_insert_inode_locked(struct inode *inode)
6210 struct btrfs_iget_args args;
6212 args.ino = BTRFS_I(inode)->location.objectid;
6213 args.root = BTRFS_I(inode)->root;
6215 return insert_inode_locked4(inode,
6216 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6217 btrfs_find_actor, &args);
6220 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6221 unsigned int *trans_num_items)
6223 struct inode *dir = args->dir;
6224 struct inode *inode = args->inode;
6227 if (!args->orphan) {
6228 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6234 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6236 fscrypt_free_filename(&args->fname);
6240 /* 1 to add inode item */
6241 *trans_num_items = 1;
6242 /* 1 to add compression property */
6243 if (BTRFS_I(dir)->prop_compress)
6244 (*trans_num_items)++;
6245 /* 1 to add default ACL xattr */
6246 if (args->default_acl)
6247 (*trans_num_items)++;
6248 /* 1 to add access ACL xattr */
6250 (*trans_num_items)++;
6251 #ifdef CONFIG_SECURITY
6252 /* 1 to add LSM xattr */
6253 if (dir->i_security)
6254 (*trans_num_items)++;
6257 /* 1 to add orphan item */
6258 (*trans_num_items)++;
6262 * 1 to add dir index
6263 * 1 to update parent inode item
6265 * No need for 1 unit for the inode ref item because it is
6266 * inserted in a batch together with the inode item at
6267 * btrfs_create_new_inode().
6269 *trans_num_items += 3;
6274 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6276 posix_acl_release(args->acl);
6277 posix_acl_release(args->default_acl);
6278 fscrypt_free_filename(&args->fname);
6282 * Inherit flags from the parent inode.
6284 * Currently only the compression flags and the cow flags are inherited.
6286 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6292 if (flags & BTRFS_INODE_NOCOMPRESS) {
6293 inode->flags &= ~BTRFS_INODE_COMPRESS;
6294 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6295 } else if (flags & BTRFS_INODE_COMPRESS) {
6296 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6297 inode->flags |= BTRFS_INODE_COMPRESS;
6300 if (flags & BTRFS_INODE_NODATACOW) {
6301 inode->flags |= BTRFS_INODE_NODATACOW;
6302 if (S_ISREG(inode->vfs_inode.i_mode))
6303 inode->flags |= BTRFS_INODE_NODATASUM;
6306 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6309 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6310 struct btrfs_new_inode_args *args)
6312 struct inode *dir = args->dir;
6313 struct inode *inode = args->inode;
6314 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6315 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6316 struct btrfs_root *root;
6317 struct btrfs_inode_item *inode_item;
6318 struct btrfs_key *location;
6319 struct btrfs_path *path;
6321 struct btrfs_inode_ref *ref;
6322 struct btrfs_key key[2];
6324 struct btrfs_item_batch batch;
6328 path = btrfs_alloc_path();
6333 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6334 root = BTRFS_I(inode)->root;
6336 ret = btrfs_get_free_objectid(root, &objectid);
6339 inode->i_ino = objectid;
6343 * O_TMPFILE, set link count to 0, so that after this point, we
6344 * fill in an inode item with the correct link count.
6346 set_nlink(inode, 0);
6348 trace_btrfs_inode_request(dir);
6350 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6354 /* index_cnt is ignored for everything but a dir. */
6355 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6356 BTRFS_I(inode)->generation = trans->transid;
6357 inode->i_generation = BTRFS_I(inode)->generation;
6360 * Subvolumes don't inherit flags from their parent directory.
6361 * Originally this was probably by accident, but we probably can't
6362 * change it now without compatibility issues.
6365 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6367 if (S_ISREG(inode->i_mode)) {
6368 if (btrfs_test_opt(fs_info, NODATASUM))
6369 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6370 if (btrfs_test_opt(fs_info, NODATACOW))
6371 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6372 BTRFS_INODE_NODATASUM;
6375 location = &BTRFS_I(inode)->location;
6376 location->objectid = objectid;
6377 location->offset = 0;
6378 location->type = BTRFS_INODE_ITEM_KEY;
6380 ret = btrfs_insert_inode_locked(inode);
6383 BTRFS_I(dir)->index_cnt--;
6388 * We could have gotten an inode number from somebody who was fsynced
6389 * and then removed in this same transaction, so let's just set full
6390 * sync since it will be a full sync anyway and this will blow away the
6391 * old info in the log.
6393 btrfs_set_inode_full_sync(BTRFS_I(inode));
6395 key[0].objectid = objectid;
6396 key[0].type = BTRFS_INODE_ITEM_KEY;
6399 sizes[0] = sizeof(struct btrfs_inode_item);
6401 if (!args->orphan) {
6403 * Start new inodes with an inode_ref. This is slightly more
6404 * efficient for small numbers of hard links since they will
6405 * be packed into one item. Extended refs will kick in if we
6406 * add more hard links than can fit in the ref item.
6408 key[1].objectid = objectid;
6409 key[1].type = BTRFS_INODE_REF_KEY;
6411 key[1].offset = objectid;
6412 sizes[1] = 2 + sizeof(*ref);
6414 key[1].offset = btrfs_ino(BTRFS_I(dir));
6415 sizes[1] = name->len + sizeof(*ref);
6419 batch.keys = &key[0];
6420 batch.data_sizes = &sizes[0];
6421 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6422 batch.nr = args->orphan ? 1 : 2;
6423 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6425 btrfs_abort_transaction(trans, ret);
6429 inode->i_mtime = current_time(inode);
6430 inode->i_atime = inode->i_mtime;
6431 inode->i_ctime = inode->i_mtime;
6432 BTRFS_I(inode)->i_otime = inode->i_mtime;
6435 * We're going to fill the inode item now, so at this point the inode
6436 * must be fully initialized.
6439 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6440 struct btrfs_inode_item);
6441 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6442 sizeof(*inode_item));
6443 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6445 if (!args->orphan) {
6446 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6447 struct btrfs_inode_ref);
6448 ptr = (unsigned long)(ref + 1);
6450 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6451 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6452 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6454 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6456 btrfs_set_inode_ref_index(path->nodes[0], ref,
6457 BTRFS_I(inode)->dir_index);
6458 write_extent_buffer(path->nodes[0], name->name, ptr,
6463 btrfs_mark_buffer_dirty(path->nodes[0]);
6465 * We don't need the path anymore, plus inheriting properties, adding
6466 * ACLs, security xattrs, orphan item or adding the link, will result in
6467 * allocating yet another path. So just free our path.
6469 btrfs_free_path(path);
6473 struct inode *parent;
6476 * Subvolumes inherit properties from their parent subvolume,
6477 * not the directory they were created in.
6479 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6480 BTRFS_I(dir)->root);
6481 if (IS_ERR(parent)) {
6482 ret = PTR_ERR(parent);
6484 ret = btrfs_inode_inherit_props(trans, inode, parent);
6488 ret = btrfs_inode_inherit_props(trans, inode, dir);
6492 "error inheriting props for ino %llu (root %llu): %d",
6493 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6498 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6501 if (!args->subvol) {
6502 ret = btrfs_init_inode_security(trans, args);
6504 btrfs_abort_transaction(trans, ret);
6509 inode_tree_add(BTRFS_I(inode));
6511 trace_btrfs_inode_new(inode);
6512 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6514 btrfs_update_root_times(trans, root);
6517 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6519 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6520 0, BTRFS_I(inode)->dir_index);
6523 btrfs_abort_transaction(trans, ret);
6531 * discard_new_inode() calls iput(), but the caller owns the reference
6535 discard_new_inode(inode);
6537 btrfs_free_path(path);
6542 * utility function to add 'inode' into 'parent_inode' with
6543 * a give name and a given sequence number.
6544 * if 'add_backref' is true, also insert a backref from the
6545 * inode to the parent directory.
6547 int btrfs_add_link(struct btrfs_trans_handle *trans,
6548 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6549 const struct fscrypt_str *name, int add_backref, u64 index)
6552 struct btrfs_key key;
6553 struct btrfs_root *root = parent_inode->root;
6554 u64 ino = btrfs_ino(inode);
6555 u64 parent_ino = btrfs_ino(parent_inode);
6557 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6558 memcpy(&key, &inode->root->root_key, sizeof(key));
6561 key.type = BTRFS_INODE_ITEM_KEY;
6565 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6566 ret = btrfs_add_root_ref(trans, key.objectid,
6567 root->root_key.objectid, parent_ino,
6569 } else if (add_backref) {
6570 ret = btrfs_insert_inode_ref(trans, root, name,
6571 ino, parent_ino, index);
6574 /* Nothing to clean up yet */
6578 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6579 btrfs_inode_type(&inode->vfs_inode), index);
6580 if (ret == -EEXIST || ret == -EOVERFLOW)
6583 btrfs_abort_transaction(trans, ret);
6587 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6589 inode_inc_iversion(&parent_inode->vfs_inode);
6591 * If we are replaying a log tree, we do not want to update the mtime
6592 * and ctime of the parent directory with the current time, since the
6593 * log replay procedure is responsible for setting them to their correct
6594 * values (the ones it had when the fsync was done).
6596 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6597 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6599 parent_inode->vfs_inode.i_mtime = now;
6600 parent_inode->vfs_inode.i_ctime = now;
6602 ret = btrfs_update_inode(trans, root, parent_inode);
6604 btrfs_abort_transaction(trans, ret);
6608 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6611 err = btrfs_del_root_ref(trans, key.objectid,
6612 root->root_key.objectid, parent_ino,
6613 &local_index, name);
6615 btrfs_abort_transaction(trans, err);
6616 } else if (add_backref) {
6620 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6623 btrfs_abort_transaction(trans, err);
6626 /* Return the original error code */
6630 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6631 struct inode *inode)
6633 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6634 struct btrfs_root *root = BTRFS_I(dir)->root;
6635 struct btrfs_new_inode_args new_inode_args = {
6640 unsigned int trans_num_items;
6641 struct btrfs_trans_handle *trans;
6644 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6648 trans = btrfs_start_transaction(root, trans_num_items);
6649 if (IS_ERR(trans)) {
6650 err = PTR_ERR(trans);
6651 goto out_new_inode_args;
6654 err = btrfs_create_new_inode(trans, &new_inode_args);
6656 d_instantiate_new(dentry, inode);
6658 btrfs_end_transaction(trans);
6659 btrfs_btree_balance_dirty(fs_info);
6661 btrfs_new_inode_args_destroy(&new_inode_args);
6668 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6669 struct dentry *dentry, umode_t mode, dev_t rdev)
6671 struct inode *inode;
6673 inode = new_inode(dir->i_sb);
6676 inode_init_owner(idmap, inode, dir, mode);
6677 inode->i_op = &btrfs_special_inode_operations;
6678 init_special_inode(inode, inode->i_mode, rdev);
6679 return btrfs_create_common(dir, dentry, inode);
6682 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6683 struct dentry *dentry, umode_t mode, bool excl)
6685 struct inode *inode;
6687 inode = new_inode(dir->i_sb);
6690 inode_init_owner(idmap, inode, dir, mode);
6691 inode->i_fop = &btrfs_file_operations;
6692 inode->i_op = &btrfs_file_inode_operations;
6693 inode->i_mapping->a_ops = &btrfs_aops;
6694 return btrfs_create_common(dir, dentry, inode);
6697 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6698 struct dentry *dentry)
6700 struct btrfs_trans_handle *trans = NULL;
6701 struct btrfs_root *root = BTRFS_I(dir)->root;
6702 struct inode *inode = d_inode(old_dentry);
6703 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6704 struct fscrypt_name fname;
6709 /* do not allow sys_link's with other subvols of the same device */
6710 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6713 if (inode->i_nlink >= BTRFS_LINK_MAX)
6716 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6720 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6725 * 2 items for inode and inode ref
6726 * 2 items for dir items
6727 * 1 item for parent inode
6728 * 1 item for orphan item deletion if O_TMPFILE
6730 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6731 if (IS_ERR(trans)) {
6732 err = PTR_ERR(trans);
6737 /* There are several dir indexes for this inode, clear the cache. */
6738 BTRFS_I(inode)->dir_index = 0ULL;
6740 inode_inc_iversion(inode);
6741 inode->i_ctime = current_time(inode);
6743 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6745 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6746 &fname.disk_name, 1, index);
6751 struct dentry *parent = dentry->d_parent;
6753 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6756 if (inode->i_nlink == 1) {
6758 * If new hard link count is 1, it's a file created
6759 * with open(2) O_TMPFILE flag.
6761 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6765 d_instantiate(dentry, inode);
6766 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6770 fscrypt_free_filename(&fname);
6772 btrfs_end_transaction(trans);
6774 inode_dec_link_count(inode);
6777 btrfs_btree_balance_dirty(fs_info);
6781 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6782 struct dentry *dentry, umode_t mode)
6784 struct inode *inode;
6786 inode = new_inode(dir->i_sb);
6789 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6790 inode->i_op = &btrfs_dir_inode_operations;
6791 inode->i_fop = &btrfs_dir_file_operations;
6792 return btrfs_create_common(dir, dentry, inode);
6795 static noinline int uncompress_inline(struct btrfs_path *path,
6797 struct btrfs_file_extent_item *item)
6800 struct extent_buffer *leaf = path->nodes[0];
6803 unsigned long inline_size;
6807 compress_type = btrfs_file_extent_compression(leaf, item);
6808 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6809 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6810 tmp = kmalloc(inline_size, GFP_NOFS);
6813 ptr = btrfs_file_extent_inline_start(item);
6815 read_extent_buffer(leaf, tmp, ptr, inline_size);
6817 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6818 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6821 * decompression code contains a memset to fill in any space between the end
6822 * of the uncompressed data and the end of max_size in case the decompressed
6823 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6824 * the end of an inline extent and the beginning of the next block, so we
6825 * cover that region here.
6828 if (max_size < PAGE_SIZE)
6829 memzero_page(page, max_size, PAGE_SIZE - max_size);
6834 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6837 struct btrfs_file_extent_item *fi;
6841 if (!page || PageUptodate(page))
6844 ASSERT(page_offset(page) == 0);
6846 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6847 struct btrfs_file_extent_item);
6848 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6849 return uncompress_inline(path, page, fi);
6851 copy_size = min_t(u64, PAGE_SIZE,
6852 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6853 kaddr = kmap_local_page(page);
6854 read_extent_buffer(path->nodes[0], kaddr,
6855 btrfs_file_extent_inline_start(fi), copy_size);
6856 kunmap_local(kaddr);
6857 if (copy_size < PAGE_SIZE)
6858 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6863 * Lookup the first extent overlapping a range in a file.
6865 * @inode: file to search in
6866 * @page: page to read extent data into if the extent is inline
6867 * @pg_offset: offset into @page to copy to
6868 * @start: file offset
6869 * @len: length of range starting at @start
6871 * Return the first &struct extent_map which overlaps the given range, reading
6872 * it from the B-tree and caching it if necessary. Note that there may be more
6873 * extents which overlap the given range after the returned extent_map.
6875 * If @page is not NULL and the extent is inline, this also reads the extent
6876 * data directly into the page and marks the extent up to date in the io_tree.
6878 * Return: ERR_PTR on error, non-NULL extent_map on success.
6880 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6881 struct page *page, size_t pg_offset,
6884 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6886 u64 extent_start = 0;
6888 u64 objectid = btrfs_ino(inode);
6889 int extent_type = -1;
6890 struct btrfs_path *path = NULL;
6891 struct btrfs_root *root = inode->root;
6892 struct btrfs_file_extent_item *item;
6893 struct extent_buffer *leaf;
6894 struct btrfs_key found_key;
6895 struct extent_map *em = NULL;
6896 struct extent_map_tree *em_tree = &inode->extent_tree;
6898 read_lock(&em_tree->lock);
6899 em = lookup_extent_mapping(em_tree, start, len);
6900 read_unlock(&em_tree->lock);
6903 if (em->start > start || em->start + em->len <= start)
6904 free_extent_map(em);
6905 else if (em->block_start == EXTENT_MAP_INLINE && page)
6906 free_extent_map(em);
6910 em = alloc_extent_map();
6915 em->start = EXTENT_MAP_HOLE;
6916 em->orig_start = EXTENT_MAP_HOLE;
6918 em->block_len = (u64)-1;
6920 path = btrfs_alloc_path();
6926 /* Chances are we'll be called again, so go ahead and do readahead */
6927 path->reada = READA_FORWARD;
6930 * The same explanation in load_free_space_cache applies here as well,
6931 * we only read when we're loading the free space cache, and at that
6932 * point the commit_root has everything we need.
6934 if (btrfs_is_free_space_inode(inode)) {
6935 path->search_commit_root = 1;
6936 path->skip_locking = 1;
6939 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6942 } else if (ret > 0) {
6943 if (path->slots[0] == 0)
6949 leaf = path->nodes[0];
6950 item = btrfs_item_ptr(leaf, path->slots[0],
6951 struct btrfs_file_extent_item);
6952 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6953 if (found_key.objectid != objectid ||
6954 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6956 * If we backup past the first extent we want to move forward
6957 * and see if there is an extent in front of us, otherwise we'll
6958 * say there is a hole for our whole search range which can
6965 extent_type = btrfs_file_extent_type(leaf, item);
6966 extent_start = found_key.offset;
6967 extent_end = btrfs_file_extent_end(path);
6968 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6969 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6970 /* Only regular file could have regular/prealloc extent */
6971 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6974 "regular/prealloc extent found for non-regular inode %llu",
6978 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6980 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6981 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6986 if (start >= extent_end) {
6988 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6989 ret = btrfs_next_leaf(root, path);
6995 leaf = path->nodes[0];
6997 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6998 if (found_key.objectid != objectid ||
6999 found_key.type != BTRFS_EXTENT_DATA_KEY)
7001 if (start + len <= found_key.offset)
7003 if (start > found_key.offset)
7006 /* New extent overlaps with existing one */
7008 em->orig_start = start;
7009 em->len = found_key.offset - start;
7010 em->block_start = EXTENT_MAP_HOLE;
7014 btrfs_extent_item_to_extent_map(inode, path, item, em);
7016 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7017 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7019 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7021 * Inline extent can only exist at file offset 0. This is
7022 * ensured by tree-checker and inline extent creation path.
7023 * Thus all members representing file offsets should be zero.
7025 ASSERT(pg_offset == 0);
7026 ASSERT(extent_start == 0);
7027 ASSERT(em->start == 0);
7030 * btrfs_extent_item_to_extent_map() should have properly
7031 * initialized em members already.
7033 * Other members are not utilized for inline extents.
7035 ASSERT(em->block_start == EXTENT_MAP_INLINE);
7036 ASSERT(em->len == fs_info->sectorsize);
7038 ret = read_inline_extent(inode, path, page);
7045 em->orig_start = start;
7047 em->block_start = EXTENT_MAP_HOLE;
7050 btrfs_release_path(path);
7051 if (em->start > start || extent_map_end(em) <= start) {
7053 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7054 em->start, em->len, start, len);
7059 write_lock(&em_tree->lock);
7060 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7061 write_unlock(&em_tree->lock);
7063 btrfs_free_path(path);
7065 trace_btrfs_get_extent(root, inode, em);
7068 free_extent_map(em);
7069 return ERR_PTR(ret);
7074 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7075 struct btrfs_dio_data *dio_data,
7078 const u64 orig_start,
7079 const u64 block_start,
7080 const u64 block_len,
7081 const u64 orig_block_len,
7082 const u64 ram_bytes,
7085 struct extent_map *em = NULL;
7086 struct btrfs_ordered_extent *ordered;
7088 if (type != BTRFS_ORDERED_NOCOW) {
7089 em = create_io_em(inode, start, len, orig_start, block_start,
7090 block_len, orig_block_len, ram_bytes,
7091 BTRFS_COMPRESS_NONE, /* compress_type */
7096 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7097 block_start, block_len, 0,
7099 (1 << BTRFS_ORDERED_DIRECT),
7100 BTRFS_COMPRESS_NONE);
7101 if (IS_ERR(ordered)) {
7103 free_extent_map(em);
7104 btrfs_drop_extent_map_range(inode, start,
7105 start + len - 1, false);
7107 em = ERR_CAST(ordered);
7109 ASSERT(!dio_data->ordered);
7110 dio_data->ordered = ordered;
7117 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7118 struct btrfs_dio_data *dio_data,
7121 struct btrfs_root *root = inode->root;
7122 struct btrfs_fs_info *fs_info = root->fs_info;
7123 struct extent_map *em;
7124 struct btrfs_key ins;
7128 alloc_hint = get_extent_allocation_hint(inode, start, len);
7129 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7130 0, alloc_hint, &ins, 1, 1);
7132 return ERR_PTR(ret);
7134 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7135 ins.objectid, ins.offset, ins.offset,
7136 ins.offset, BTRFS_ORDERED_REGULAR);
7137 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7139 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7145 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7147 struct btrfs_block_group *block_group;
7148 bool readonly = false;
7150 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7151 if (!block_group || block_group->ro)
7154 btrfs_put_block_group(block_group);
7159 * Check if we can do nocow write into the range [@offset, @offset + @len)
7161 * @offset: File offset
7162 * @len: The length to write, will be updated to the nocow writeable
7164 * @orig_start: (optional) Return the original file offset of the file extent
7165 * @orig_len: (optional) Return the original on-disk length of the file extent
7166 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7167 * @strict: if true, omit optimizations that might force us into unnecessary
7168 * cow. e.g., don't trust generation number.
7171 * >0 and update @len if we can do nocow write
7172 * 0 if we can't do nocow write
7173 * <0 if error happened
7175 * NOTE: This only checks the file extents, caller is responsible to wait for
7176 * any ordered extents.
7178 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7179 u64 *orig_start, u64 *orig_block_len,
7180 u64 *ram_bytes, bool nowait, bool strict)
7182 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7183 struct can_nocow_file_extent_args nocow_args = { 0 };
7184 struct btrfs_path *path;
7186 struct extent_buffer *leaf;
7187 struct btrfs_root *root = BTRFS_I(inode)->root;
7188 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7189 struct btrfs_file_extent_item *fi;
7190 struct btrfs_key key;
7193 path = btrfs_alloc_path();
7196 path->nowait = nowait;
7198 ret = btrfs_lookup_file_extent(NULL, root, path,
7199 btrfs_ino(BTRFS_I(inode)), offset, 0);
7204 if (path->slots[0] == 0) {
7205 /* can't find the item, must cow */
7212 leaf = path->nodes[0];
7213 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7214 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7215 key.type != BTRFS_EXTENT_DATA_KEY) {
7216 /* not our file or wrong item type, must cow */
7220 if (key.offset > offset) {
7221 /* Wrong offset, must cow */
7225 if (btrfs_file_extent_end(path) <= offset)
7228 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7229 found_type = btrfs_file_extent_type(leaf, fi);
7231 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7233 nocow_args.start = offset;
7234 nocow_args.end = offset + *len - 1;
7235 nocow_args.strict = strict;
7236 nocow_args.free_path = true;
7238 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7239 /* can_nocow_file_extent() has freed the path. */
7243 /* Treat errors as not being able to NOCOW. */
7249 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7252 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7253 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7256 range_end = round_up(offset + nocow_args.num_bytes,
7257 root->fs_info->sectorsize) - 1;
7258 ret = test_range_bit(io_tree, offset, range_end,
7259 EXTENT_DELALLOC, 0, NULL);
7267 *orig_start = key.offset - nocow_args.extent_offset;
7269 *orig_block_len = nocow_args.disk_num_bytes;
7271 *len = nocow_args.num_bytes;
7274 btrfs_free_path(path);
7278 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7279 struct extent_state **cached_state,
7280 unsigned int iomap_flags)
7282 const bool writing = (iomap_flags & IOMAP_WRITE);
7283 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7284 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7285 struct btrfs_ordered_extent *ordered;
7290 if (!try_lock_extent(io_tree, lockstart, lockend,
7294 lock_extent(io_tree, lockstart, lockend, cached_state);
7297 * We're concerned with the entire range that we're going to be
7298 * doing DIO to, so we need to make sure there's no ordered
7299 * extents in this range.
7301 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7302 lockend - lockstart + 1);
7305 * We need to make sure there are no buffered pages in this
7306 * range either, we could have raced between the invalidate in
7307 * generic_file_direct_write and locking the extent. The
7308 * invalidate needs to happen so that reads after a write do not
7312 (!writing || !filemap_range_has_page(inode->i_mapping,
7313 lockstart, lockend)))
7316 unlock_extent(io_tree, lockstart, lockend, cached_state);
7320 btrfs_put_ordered_extent(ordered);
7325 * If we are doing a DIO read and the ordered extent we
7326 * found is for a buffered write, we can not wait for it
7327 * to complete and retry, because if we do so we can
7328 * deadlock with concurrent buffered writes on page
7329 * locks. This happens only if our DIO read covers more
7330 * than one extent map, if at this point has already
7331 * created an ordered extent for a previous extent map
7332 * and locked its range in the inode's io tree, and a
7333 * concurrent write against that previous extent map's
7334 * range and this range started (we unlock the ranges
7335 * in the io tree only when the bios complete and
7336 * buffered writes always lock pages before attempting
7337 * to lock range in the io tree).
7340 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7341 btrfs_start_ordered_extent(ordered);
7343 ret = nowait ? -EAGAIN : -ENOTBLK;
7344 btrfs_put_ordered_extent(ordered);
7347 * We could trigger writeback for this range (and wait
7348 * for it to complete) and then invalidate the pages for
7349 * this range (through invalidate_inode_pages2_range()),
7350 * but that can lead us to a deadlock with a concurrent
7351 * call to readahead (a buffered read or a defrag call
7352 * triggered a readahead) on a page lock due to an
7353 * ordered dio extent we created before but did not have
7354 * yet a corresponding bio submitted (whence it can not
7355 * complete), which makes readahead wait for that
7356 * ordered extent to complete while holding a lock on
7359 ret = nowait ? -EAGAIN : -ENOTBLK;
7371 /* The callers of this must take lock_extent() */
7372 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7373 u64 len, u64 orig_start, u64 block_start,
7374 u64 block_len, u64 orig_block_len,
7375 u64 ram_bytes, int compress_type,
7378 struct extent_map *em;
7381 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7382 type == BTRFS_ORDERED_COMPRESSED ||
7383 type == BTRFS_ORDERED_NOCOW ||
7384 type == BTRFS_ORDERED_REGULAR);
7386 em = alloc_extent_map();
7388 return ERR_PTR(-ENOMEM);
7391 em->orig_start = orig_start;
7393 em->block_len = block_len;
7394 em->block_start = block_start;
7395 em->orig_block_len = orig_block_len;
7396 em->ram_bytes = ram_bytes;
7397 em->generation = -1;
7398 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7399 if (type == BTRFS_ORDERED_PREALLOC) {
7400 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7401 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7402 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7403 em->compress_type = compress_type;
7406 ret = btrfs_replace_extent_map_range(inode, em, true);
7408 free_extent_map(em);
7409 return ERR_PTR(ret);
7412 /* em got 2 refs now, callers needs to do free_extent_map once. */
7417 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7418 struct inode *inode,
7419 struct btrfs_dio_data *dio_data,
7420 u64 start, u64 *lenp,
7421 unsigned int iomap_flags)
7423 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7424 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7425 struct extent_map *em = *map;
7427 u64 block_start, orig_start, orig_block_len, ram_bytes;
7428 struct btrfs_block_group *bg;
7429 bool can_nocow = false;
7430 bool space_reserved = false;
7436 * We don't allocate a new extent in the following cases
7438 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7440 * 2) The extent is marked as PREALLOC. We're good to go here and can
7441 * just use the extent.
7444 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7445 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7446 em->block_start != EXTENT_MAP_HOLE)) {
7447 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7448 type = BTRFS_ORDERED_PREALLOC;
7450 type = BTRFS_ORDERED_NOCOW;
7451 len = min(len, em->len - (start - em->start));
7452 block_start = em->block_start + (start - em->start);
7454 if (can_nocow_extent(inode, start, &len, &orig_start,
7455 &orig_block_len, &ram_bytes, false, false) == 1) {
7456 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7464 struct extent_map *em2;
7466 /* We can NOCOW, so only need to reserve metadata space. */
7467 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7470 /* Our caller expects us to free the input extent map. */
7471 free_extent_map(em);
7473 btrfs_dec_nocow_writers(bg);
7474 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7478 space_reserved = true;
7480 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7481 orig_start, block_start,
7482 len, orig_block_len,
7484 btrfs_dec_nocow_writers(bg);
7485 if (type == BTRFS_ORDERED_PREALLOC) {
7486 free_extent_map(em);
7496 dio_data->nocow_done = true;
7498 /* Our caller expects us to free the input extent map. */
7499 free_extent_map(em);
7508 * If we could not allocate data space before locking the file
7509 * range and we can't do a NOCOW write, then we have to fail.
7511 if (!dio_data->data_space_reserved) {
7517 * We have to COW and we have already reserved data space before,
7518 * so now we reserve only metadata.
7520 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7524 space_reserved = true;
7526 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7532 len = min(len, em->len - (start - em->start));
7534 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7535 prev_len - len, true);
7539 * We have created our ordered extent, so we can now release our reservation
7540 * for an outstanding extent.
7542 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7545 * Need to update the i_size under the extent lock so buffered
7546 * readers will get the updated i_size when we unlock.
7548 if (start + len > i_size_read(inode))
7549 i_size_write(inode, start + len);
7551 if (ret && space_reserved) {
7552 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7553 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7559 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7560 loff_t length, unsigned int flags, struct iomap *iomap,
7561 struct iomap *srcmap)
7563 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7564 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7565 struct extent_map *em;
7566 struct extent_state *cached_state = NULL;
7567 struct btrfs_dio_data *dio_data = iter->private;
7568 u64 lockstart, lockend;
7569 const bool write = !!(flags & IOMAP_WRITE);
7572 const u64 data_alloc_len = length;
7573 bool unlock_extents = false;
7576 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7577 * we're NOWAIT we may submit a bio for a partial range and return
7578 * EIOCBQUEUED, which would result in an errant short read.
7580 * The best way to handle this would be to allow for partial completions
7581 * of iocb's, so we could submit the partial bio, return and fault in
7582 * the rest of the pages, and then submit the io for the rest of the
7583 * range. However we don't have that currently, so simply return
7584 * -EAGAIN at this point so that the normal path is used.
7586 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7590 * Cap the size of reads to that usually seen in buffered I/O as we need
7591 * to allocate a contiguous array for the checksums.
7594 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7597 lockend = start + len - 1;
7600 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7601 * enough if we've written compressed pages to this area, so we need to
7602 * flush the dirty pages again to make absolutely sure that any
7603 * outstanding dirty pages are on disk - the first flush only starts
7604 * compression on the data, while keeping the pages locked, so by the
7605 * time the second flush returns we know bios for the compressed pages
7606 * were submitted and finished, and the pages no longer under writeback.
7608 * If we have a NOWAIT request and we have any pages in the range that
7609 * are locked, likely due to compression still in progress, we don't want
7610 * to block on page locks. We also don't want to block on pages marked as
7611 * dirty or under writeback (same as for the non-compression case).
7612 * iomap_dio_rw() did the same check, but after that and before we got
7613 * here, mmap'ed writes may have happened or buffered reads started
7614 * (readpage() and readahead(), which lock pages), as we haven't locked
7615 * the file range yet.
7617 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7618 &BTRFS_I(inode)->runtime_flags)) {
7619 if (flags & IOMAP_NOWAIT) {
7620 if (filemap_range_needs_writeback(inode->i_mapping,
7621 lockstart, lockend))
7624 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7625 start + length - 1);
7631 memset(dio_data, 0, sizeof(*dio_data));
7634 * We always try to allocate data space and must do it before locking
7635 * the file range, to avoid deadlocks with concurrent writes to the same
7636 * range if the range has several extents and the writes don't expand the
7637 * current i_size (the inode lock is taken in shared mode). If we fail to
7638 * allocate data space here we continue and later, after locking the
7639 * file range, we fail with ENOSPC only if we figure out we can not do a
7642 if (write && !(flags & IOMAP_NOWAIT)) {
7643 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7644 &dio_data->data_reserved,
7645 start, data_alloc_len, false);
7647 dio_data->data_space_reserved = true;
7648 else if (ret && !(BTRFS_I(inode)->flags &
7649 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7654 * If this errors out it's because we couldn't invalidate pagecache for
7655 * this range and we need to fallback to buffered IO, or we are doing a
7656 * NOWAIT read/write and we need to block.
7658 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7662 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7669 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7670 * io. INLINE is special, and we could probably kludge it in here, but
7671 * it's still buffered so for safety lets just fall back to the generic
7674 * For COMPRESSED we _have_ to read the entire extent in so we can
7675 * decompress it, so there will be buffering required no matter what we
7676 * do, so go ahead and fallback to buffered.
7678 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7679 * to buffered IO. Don't blame me, this is the price we pay for using
7682 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7683 em->block_start == EXTENT_MAP_INLINE) {
7684 free_extent_map(em);
7686 * If we are in a NOWAIT context, return -EAGAIN in order to
7687 * fallback to buffered IO. This is not only because we can
7688 * block with buffered IO (no support for NOWAIT semantics at
7689 * the moment) but also to avoid returning short reads to user
7690 * space - this happens if we were able to read some data from
7691 * previous non-compressed extents and then when we fallback to
7692 * buffered IO, at btrfs_file_read_iter() by calling
7693 * filemap_read(), we fail to fault in pages for the read buffer,
7694 * in which case filemap_read() returns a short read (the number
7695 * of bytes previously read is > 0, so it does not return -EFAULT).
7697 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7701 len = min(len, em->len - (start - em->start));
7704 * If we have a NOWAIT request and the range contains multiple extents
7705 * (or a mix of extents and holes), then we return -EAGAIN to make the
7706 * caller fallback to a context where it can do a blocking (without
7707 * NOWAIT) request. This way we avoid doing partial IO and returning
7708 * success to the caller, which is not optimal for writes and for reads
7709 * it can result in unexpected behaviour for an application.
7711 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7712 * iomap_dio_rw(), we can end up returning less data then what the caller
7713 * asked for, resulting in an unexpected, and incorrect, short read.
7714 * That is, the caller asked to read N bytes and we return less than that,
7715 * which is wrong unless we are crossing EOF. This happens if we get a
7716 * page fault error when trying to fault in pages for the buffer that is
7717 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7718 * have previously submitted bios for other extents in the range, in
7719 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7720 * those bios have completed by the time we get the page fault error,
7721 * which we return back to our caller - we should only return EIOCBQUEUED
7722 * after we have submitted bios for all the extents in the range.
7724 if ((flags & IOMAP_NOWAIT) && len < length) {
7725 free_extent_map(em);
7731 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7732 start, &len, flags);
7735 unlock_extents = true;
7736 /* Recalc len in case the new em is smaller than requested */
7737 len = min(len, em->len - (start - em->start));
7738 if (dio_data->data_space_reserved) {
7740 u64 release_len = 0;
7742 if (dio_data->nocow_done) {
7743 release_offset = start;
7744 release_len = data_alloc_len;
7745 } else if (len < data_alloc_len) {
7746 release_offset = start + len;
7747 release_len = data_alloc_len - len;
7750 if (release_len > 0)
7751 btrfs_free_reserved_data_space(BTRFS_I(inode),
7752 dio_data->data_reserved,
7758 * We need to unlock only the end area that we aren't using.
7759 * The rest is going to be unlocked by the endio routine.
7761 lockstart = start + len;
7762 if (lockstart < lockend)
7763 unlock_extents = true;
7767 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7770 free_extent_state(cached_state);
7773 * Translate extent map information to iomap.
7774 * We trim the extents (and move the addr) even though iomap code does
7775 * that, since we have locked only the parts we are performing I/O in.
7777 if ((em->block_start == EXTENT_MAP_HOLE) ||
7778 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7779 iomap->addr = IOMAP_NULL_ADDR;
7780 iomap->type = IOMAP_HOLE;
7782 iomap->addr = em->block_start + (start - em->start);
7783 iomap->type = IOMAP_MAPPED;
7785 iomap->offset = start;
7786 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7787 iomap->length = len;
7788 free_extent_map(em);
7793 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7796 if (dio_data->data_space_reserved) {
7797 btrfs_free_reserved_data_space(BTRFS_I(inode),
7798 dio_data->data_reserved,
7799 start, data_alloc_len);
7800 extent_changeset_free(dio_data->data_reserved);
7806 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7807 ssize_t written, unsigned int flags, struct iomap *iomap)
7809 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7810 struct btrfs_dio_data *dio_data = iter->private;
7811 size_t submitted = dio_data->submitted;
7812 const bool write = !!(flags & IOMAP_WRITE);
7815 if (!write && (iomap->type == IOMAP_HOLE)) {
7816 /* If reading from a hole, unlock and return */
7817 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7822 if (submitted < length) {
7824 length -= submitted;
7826 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7827 pos, length, false);
7829 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7830 pos + length - 1, NULL);
7834 btrfs_put_ordered_extent(dio_data->ordered);
7835 dio_data->ordered = NULL;
7839 extent_changeset_free(dio_data->data_reserved);
7843 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7845 struct btrfs_dio_private *dip =
7846 container_of(bbio, struct btrfs_dio_private, bbio);
7847 struct btrfs_inode *inode = bbio->inode;
7848 struct bio *bio = &bbio->bio;
7850 if (bio->bi_status) {
7851 btrfs_warn(inode->root->fs_info,
7852 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7853 btrfs_ino(inode), bio->bi_opf,
7854 dip->file_offset, dip->bytes, bio->bi_status);
7857 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7858 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7859 dip->file_offset, dip->bytes,
7862 unlock_extent(&inode->io_tree, dip->file_offset,
7863 dip->file_offset + dip->bytes - 1, NULL);
7866 bbio->bio.bi_private = bbio->private;
7867 iomap_dio_bio_end_io(bio);
7870 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7873 struct btrfs_bio *bbio = btrfs_bio(bio);
7874 struct btrfs_dio_private *dip =
7875 container_of(bbio, struct btrfs_dio_private, bbio);
7876 struct btrfs_dio_data *dio_data = iter->private;
7878 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7879 btrfs_dio_end_io, bio->bi_private);
7880 bbio->inode = BTRFS_I(iter->inode);
7881 bbio->file_offset = file_offset;
7883 dip->file_offset = file_offset;
7884 dip->bytes = bio->bi_iter.bi_size;
7886 dio_data->submitted += bio->bi_iter.bi_size;
7889 * Check if we are doing a partial write. If we are, we need to split
7890 * the ordered extent to match the submitted bio. Hang on to the
7891 * remaining unfinishable ordered_extent in dio_data so that it can be
7892 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7893 * remaining pages is blocked on the outstanding ordered extent.
7895 if (iter->flags & IOMAP_WRITE) {
7898 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7900 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7901 file_offset, dip->bytes,
7903 bio->bi_status = errno_to_blk_status(ret);
7904 iomap_dio_bio_end_io(bio);
7909 btrfs_submit_bio(bbio, 0);
7912 static const struct iomap_ops btrfs_dio_iomap_ops = {
7913 .iomap_begin = btrfs_dio_iomap_begin,
7914 .iomap_end = btrfs_dio_iomap_end,
7917 static const struct iomap_dio_ops btrfs_dio_ops = {
7918 .submit_io = btrfs_dio_submit_io,
7919 .bio_set = &btrfs_dio_bioset,
7922 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7924 struct btrfs_dio_data data = { 0 };
7926 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7927 IOMAP_DIO_PARTIAL, &data, done_before);
7930 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7933 struct btrfs_dio_data data = { 0 };
7935 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7936 IOMAP_DIO_PARTIAL, &data, done_before);
7939 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7944 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7949 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7950 * file range (0 to LLONG_MAX), but that is not enough if we have
7951 * compression enabled. The first filemap_fdatawrite_range() only kicks
7952 * in the compression of data (in an async thread) and will return
7953 * before the compression is done and writeback is started. A second
7954 * filemap_fdatawrite_range() is needed to wait for the compression to
7955 * complete and writeback to start. We also need to wait for ordered
7956 * extents to complete, because our fiemap implementation uses mainly
7957 * file extent items to list the extents, searching for extent maps
7958 * only for file ranges with holes or prealloc extents to figure out
7959 * if we have delalloc in those ranges.
7961 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7962 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7967 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7970 static int btrfs_writepages(struct address_space *mapping,
7971 struct writeback_control *wbc)
7973 return extent_writepages(mapping, wbc);
7976 static void btrfs_readahead(struct readahead_control *rac)
7978 extent_readahead(rac);
7982 * For release_folio() and invalidate_folio() we have a race window where
7983 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7984 * If we continue to release/invalidate the page, we could cause use-after-free
7985 * for subpage spinlock. So this function is to spin and wait for subpage
7988 static void wait_subpage_spinlock(struct page *page)
7990 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7991 struct btrfs_subpage *subpage;
7993 if (!btrfs_is_subpage(fs_info, page))
7996 ASSERT(PagePrivate(page) && page->private);
7997 subpage = (struct btrfs_subpage *)page->private;
8000 * This may look insane as we just acquire the spinlock and release it,
8001 * without doing anything. But we just want to make sure no one is
8002 * still holding the subpage spinlock.
8003 * And since the page is not dirty nor writeback, and we have page
8004 * locked, the only possible way to hold a spinlock is from the endio
8005 * function to clear page writeback.
8007 * Here we just acquire the spinlock so that all existing callers
8008 * should exit and we're safe to release/invalidate the page.
8010 spin_lock_irq(&subpage->lock);
8011 spin_unlock_irq(&subpage->lock);
8014 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8016 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
8019 wait_subpage_spinlock(&folio->page);
8020 clear_page_extent_mapped(&folio->page);
8025 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8027 if (folio_test_writeback(folio) || folio_test_dirty(folio))
8029 return __btrfs_release_folio(folio, gfp_flags);
8032 #ifdef CONFIG_MIGRATION
8033 static int btrfs_migrate_folio(struct address_space *mapping,
8034 struct folio *dst, struct folio *src,
8035 enum migrate_mode mode)
8037 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8039 if (ret != MIGRATEPAGE_SUCCESS)
8042 if (folio_test_ordered(src)) {
8043 folio_clear_ordered(src);
8044 folio_set_ordered(dst);
8047 return MIGRATEPAGE_SUCCESS;
8050 #define btrfs_migrate_folio NULL
8053 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8056 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8057 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8058 struct extent_io_tree *tree = &inode->io_tree;
8059 struct extent_state *cached_state = NULL;
8060 u64 page_start = folio_pos(folio);
8061 u64 page_end = page_start + folio_size(folio) - 1;
8063 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8066 * We have folio locked so no new ordered extent can be created on this
8067 * page, nor bio can be submitted for this folio.
8069 * But already submitted bio can still be finished on this folio.
8070 * Furthermore, endio function won't skip folio which has Ordered
8071 * (Private2) already cleared, so it's possible for endio and
8072 * invalidate_folio to do the same ordered extent accounting twice
8075 * So here we wait for any submitted bios to finish, so that we won't
8076 * do double ordered extent accounting on the same folio.
8078 folio_wait_writeback(folio);
8079 wait_subpage_spinlock(&folio->page);
8082 * For subpage case, we have call sites like
8083 * btrfs_punch_hole_lock_range() which passes range not aligned to
8085 * If the range doesn't cover the full folio, we don't need to and
8086 * shouldn't clear page extent mapped, as folio->private can still
8087 * record subpage dirty bits for other part of the range.
8089 * For cases that invalidate the full folio even the range doesn't
8090 * cover the full folio, like invalidating the last folio, we're
8091 * still safe to wait for ordered extent to finish.
8093 if (!(offset == 0 && length == folio_size(folio))) {
8094 btrfs_release_folio(folio, GFP_NOFS);
8098 if (!inode_evicting)
8099 lock_extent(tree, page_start, page_end, &cached_state);
8102 while (cur < page_end) {
8103 struct btrfs_ordered_extent *ordered;
8106 u32 extra_flags = 0;
8108 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8109 page_end + 1 - cur);
8111 range_end = page_end;
8113 * No ordered extent covering this range, we are safe
8114 * to delete all extent states in the range.
8116 extra_flags = EXTENT_CLEAR_ALL_BITS;
8119 if (ordered->file_offset > cur) {
8121 * There is a range between [cur, oe->file_offset) not
8122 * covered by any ordered extent.
8123 * We are safe to delete all extent states, and handle
8124 * the ordered extent in the next iteration.
8126 range_end = ordered->file_offset - 1;
8127 extra_flags = EXTENT_CLEAR_ALL_BITS;
8131 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8133 ASSERT(range_end + 1 - cur < U32_MAX);
8134 range_len = range_end + 1 - cur;
8135 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8137 * If Ordered (Private2) is cleared, it means endio has
8138 * already been executed for the range.
8139 * We can't delete the extent states as
8140 * btrfs_finish_ordered_io() may still use some of them.
8144 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8147 * IO on this page will never be started, so we need to account
8148 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8149 * here, must leave that up for the ordered extent completion.
8151 * This will also unlock the range for incoming
8152 * btrfs_finish_ordered_io().
8154 if (!inode_evicting)
8155 clear_extent_bit(tree, cur, range_end,
8157 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8158 EXTENT_DEFRAG, &cached_state);
8160 spin_lock_irq(&inode->ordered_tree.lock);
8161 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8162 ordered->truncated_len = min(ordered->truncated_len,
8163 cur - ordered->file_offset);
8164 spin_unlock_irq(&inode->ordered_tree.lock);
8167 * If the ordered extent has finished, we're safe to delete all
8168 * the extent states of the range, otherwise
8169 * btrfs_finish_ordered_io() will get executed by endio for
8170 * other pages, so we can't delete extent states.
8172 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8173 cur, range_end + 1 - cur)) {
8174 btrfs_finish_ordered_io(ordered);
8176 * The ordered extent has finished, now we're again
8177 * safe to delete all extent states of the range.
8179 extra_flags = EXTENT_CLEAR_ALL_BITS;
8183 btrfs_put_ordered_extent(ordered);
8185 * Qgroup reserved space handler
8186 * Sector(s) here will be either:
8188 * 1) Already written to disk or bio already finished
8189 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8190 * Qgroup will be handled by its qgroup_record then.
8191 * btrfs_qgroup_free_data() call will do nothing here.
8193 * 2) Not written to disk yet
8194 * Then btrfs_qgroup_free_data() call will clear the
8195 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8196 * reserved data space.
8197 * Since the IO will never happen for this page.
8199 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8200 if (!inode_evicting) {
8201 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8202 EXTENT_DELALLOC | EXTENT_UPTODATE |
8203 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8204 extra_flags, &cached_state);
8206 cur = range_end + 1;
8209 * We have iterated through all ordered extents of the page, the page
8210 * should not have Ordered (Private2) anymore, or the above iteration
8211 * did something wrong.
8213 ASSERT(!folio_test_ordered(folio));
8214 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8215 if (!inode_evicting)
8216 __btrfs_release_folio(folio, GFP_NOFS);
8217 clear_page_extent_mapped(&folio->page);
8221 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8222 * called from a page fault handler when a page is first dirtied. Hence we must
8223 * be careful to check for EOF conditions here. We set the page up correctly
8224 * for a written page which means we get ENOSPC checking when writing into
8225 * holes and correct delalloc and unwritten extent mapping on filesystems that
8226 * support these features.
8228 * We are not allowed to take the i_mutex here so we have to play games to
8229 * protect against truncate races as the page could now be beyond EOF. Because
8230 * truncate_setsize() writes the inode size before removing pages, once we have
8231 * the page lock we can determine safely if the page is beyond EOF. If it is not
8232 * beyond EOF, then the page is guaranteed safe against truncation until we
8235 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8237 struct page *page = vmf->page;
8238 struct inode *inode = file_inode(vmf->vma->vm_file);
8239 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8240 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8241 struct btrfs_ordered_extent *ordered;
8242 struct extent_state *cached_state = NULL;
8243 struct extent_changeset *data_reserved = NULL;
8244 unsigned long zero_start;
8254 reserved_space = PAGE_SIZE;
8256 sb_start_pagefault(inode->i_sb);
8257 page_start = page_offset(page);
8258 page_end = page_start + PAGE_SIZE - 1;
8262 * Reserving delalloc space after obtaining the page lock can lead to
8263 * deadlock. For example, if a dirty page is locked by this function
8264 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8265 * dirty page write out, then the btrfs_writepages() function could
8266 * end up waiting indefinitely to get a lock on the page currently
8267 * being processed by btrfs_page_mkwrite() function.
8269 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8270 page_start, reserved_space);
8272 ret2 = file_update_time(vmf->vma->vm_file);
8276 ret = vmf_error(ret2);
8282 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8284 down_read(&BTRFS_I(inode)->i_mmap_lock);
8286 size = i_size_read(inode);
8288 if ((page->mapping != inode->i_mapping) ||
8289 (page_start >= size)) {
8290 /* page got truncated out from underneath us */
8293 wait_on_page_writeback(page);
8295 lock_extent(io_tree, page_start, page_end, &cached_state);
8296 ret2 = set_page_extent_mapped(page);
8298 ret = vmf_error(ret2);
8299 unlock_extent(io_tree, page_start, page_end, &cached_state);
8304 * we can't set the delalloc bits if there are pending ordered
8305 * extents. Drop our locks and wait for them to finish
8307 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8310 unlock_extent(io_tree, page_start, page_end, &cached_state);
8312 up_read(&BTRFS_I(inode)->i_mmap_lock);
8313 btrfs_start_ordered_extent(ordered);
8314 btrfs_put_ordered_extent(ordered);
8318 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8319 reserved_space = round_up(size - page_start,
8320 fs_info->sectorsize);
8321 if (reserved_space < PAGE_SIZE) {
8322 end = page_start + reserved_space - 1;
8323 btrfs_delalloc_release_space(BTRFS_I(inode),
8324 data_reserved, page_start,
8325 PAGE_SIZE - reserved_space, true);
8330 * page_mkwrite gets called when the page is firstly dirtied after it's
8331 * faulted in, but write(2) could also dirty a page and set delalloc
8332 * bits, thus in this case for space account reason, we still need to
8333 * clear any delalloc bits within this page range since we have to
8334 * reserve data&meta space before lock_page() (see above comments).
8336 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8337 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8338 EXTENT_DEFRAG, &cached_state);
8340 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8343 unlock_extent(io_tree, page_start, page_end, &cached_state);
8344 ret = VM_FAULT_SIGBUS;
8348 /* page is wholly or partially inside EOF */
8349 if (page_start + PAGE_SIZE > size)
8350 zero_start = offset_in_page(size);
8352 zero_start = PAGE_SIZE;
8354 if (zero_start != PAGE_SIZE)
8355 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8357 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8358 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8359 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8361 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8363 unlock_extent(io_tree, page_start, page_end, &cached_state);
8364 up_read(&BTRFS_I(inode)->i_mmap_lock);
8366 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8367 sb_end_pagefault(inode->i_sb);
8368 extent_changeset_free(data_reserved);
8369 return VM_FAULT_LOCKED;
8373 up_read(&BTRFS_I(inode)->i_mmap_lock);
8375 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8376 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8377 reserved_space, (ret != 0));
8379 sb_end_pagefault(inode->i_sb);
8380 extent_changeset_free(data_reserved);
8384 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8386 struct btrfs_truncate_control control = {
8388 .ino = btrfs_ino(inode),
8389 .min_type = BTRFS_EXTENT_DATA_KEY,
8390 .clear_extent_range = true,
8392 struct btrfs_root *root = inode->root;
8393 struct btrfs_fs_info *fs_info = root->fs_info;
8394 struct btrfs_block_rsv *rsv;
8396 struct btrfs_trans_handle *trans;
8397 u64 mask = fs_info->sectorsize - 1;
8398 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8400 if (!skip_writeback) {
8401 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8402 inode->vfs_inode.i_size & (~mask),
8409 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8410 * things going on here:
8412 * 1) We need to reserve space to update our inode.
8414 * 2) We need to have something to cache all the space that is going to
8415 * be free'd up by the truncate operation, but also have some slack
8416 * space reserved in case it uses space during the truncate (thank you
8417 * very much snapshotting).
8419 * And we need these to be separate. The fact is we can use a lot of
8420 * space doing the truncate, and we have no earthly idea how much space
8421 * we will use, so we need the truncate reservation to be separate so it
8422 * doesn't end up using space reserved for updating the inode. We also
8423 * need to be able to stop the transaction and start a new one, which
8424 * means we need to be able to update the inode several times, and we
8425 * have no idea of knowing how many times that will be, so we can't just
8426 * reserve 1 item for the entirety of the operation, so that has to be
8427 * done separately as well.
8429 * So that leaves us with
8431 * 1) rsv - for the truncate reservation, which we will steal from the
8432 * transaction reservation.
8433 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8434 * updating the inode.
8436 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8439 rsv->size = min_size;
8440 rsv->failfast = true;
8443 * 1 for the truncate slack space
8444 * 1 for updating the inode.
8446 trans = btrfs_start_transaction(root, 2);
8447 if (IS_ERR(trans)) {
8448 ret = PTR_ERR(trans);
8452 /* Migrate the slack space for the truncate to our reserve */
8453 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8456 * We have reserved 2 metadata units when we started the transaction and
8457 * min_size matches 1 unit, so this should never fail, but if it does,
8458 * it's not critical we just fail truncation.
8461 btrfs_end_transaction(trans);
8465 trans->block_rsv = rsv;
8468 struct extent_state *cached_state = NULL;
8469 const u64 new_size = inode->vfs_inode.i_size;
8470 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8472 control.new_size = new_size;
8473 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8475 * We want to drop from the next block forward in case this new
8476 * size is not block aligned since we will be keeping the last
8477 * block of the extent just the way it is.
8479 btrfs_drop_extent_map_range(inode,
8480 ALIGN(new_size, fs_info->sectorsize),
8483 ret = btrfs_truncate_inode_items(trans, root, &control);
8485 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8486 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8488 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8490 trans->block_rsv = &fs_info->trans_block_rsv;
8491 if (ret != -ENOSPC && ret != -EAGAIN)
8494 ret = btrfs_update_inode(trans, root, inode);
8498 btrfs_end_transaction(trans);
8499 btrfs_btree_balance_dirty(fs_info);
8501 trans = btrfs_start_transaction(root, 2);
8502 if (IS_ERR(trans)) {
8503 ret = PTR_ERR(trans);
8508 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8509 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8510 rsv, min_size, false);
8512 * We have reserved 2 metadata units when we started the
8513 * transaction and min_size matches 1 unit, so this should never
8514 * fail, but if it does, it's not critical we just fail truncation.
8519 trans->block_rsv = rsv;
8523 * We can't call btrfs_truncate_block inside a trans handle as we could
8524 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8525 * know we've truncated everything except the last little bit, and can
8526 * do btrfs_truncate_block and then update the disk_i_size.
8528 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8529 btrfs_end_transaction(trans);
8530 btrfs_btree_balance_dirty(fs_info);
8532 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8535 trans = btrfs_start_transaction(root, 1);
8536 if (IS_ERR(trans)) {
8537 ret = PTR_ERR(trans);
8540 btrfs_inode_safe_disk_i_size_write(inode, 0);
8546 trans->block_rsv = &fs_info->trans_block_rsv;
8547 ret2 = btrfs_update_inode(trans, root, inode);
8551 ret2 = btrfs_end_transaction(trans);
8554 btrfs_btree_balance_dirty(fs_info);
8557 btrfs_free_block_rsv(fs_info, rsv);
8559 * So if we truncate and then write and fsync we normally would just
8560 * write the extents that changed, which is a problem if we need to
8561 * first truncate that entire inode. So set this flag so we write out
8562 * all of the extents in the inode to the sync log so we're completely
8565 * If no extents were dropped or trimmed we don't need to force the next
8566 * fsync to truncate all the inode's items from the log and re-log them
8567 * all. This means the truncate operation did not change the file size,
8568 * or changed it to a smaller size but there was only an implicit hole
8569 * between the old i_size and the new i_size, and there were no prealloc
8570 * extents beyond i_size to drop.
8572 if (control.extents_found > 0)
8573 btrfs_set_inode_full_sync(inode);
8578 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8581 struct inode *inode;
8583 inode = new_inode(dir->i_sb);
8586 * Subvolumes don't inherit the sgid bit or the parent's gid if
8587 * the parent's sgid bit is set. This is probably a bug.
8589 inode_init_owner(idmap, inode, NULL,
8590 S_IFDIR | (~current_umask() & S_IRWXUGO));
8591 inode->i_op = &btrfs_dir_inode_operations;
8592 inode->i_fop = &btrfs_dir_file_operations;
8597 struct inode *btrfs_alloc_inode(struct super_block *sb)
8599 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8600 struct btrfs_inode *ei;
8601 struct inode *inode;
8603 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8610 ei->last_sub_trans = 0;
8611 ei->logged_trans = 0;
8612 ei->delalloc_bytes = 0;
8613 ei->new_delalloc_bytes = 0;
8614 ei->defrag_bytes = 0;
8615 ei->disk_i_size = 0;
8619 ei->index_cnt = (u64)-1;
8621 ei->last_unlink_trans = 0;
8622 ei->last_reflink_trans = 0;
8623 ei->last_log_commit = 0;
8625 spin_lock_init(&ei->lock);
8626 ei->outstanding_extents = 0;
8627 if (sb->s_magic != BTRFS_TEST_MAGIC)
8628 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8629 BTRFS_BLOCK_RSV_DELALLOC);
8630 ei->runtime_flags = 0;
8631 ei->prop_compress = BTRFS_COMPRESS_NONE;
8632 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8634 ei->delayed_node = NULL;
8636 ei->i_otime.tv_sec = 0;
8637 ei->i_otime.tv_nsec = 0;
8639 inode = &ei->vfs_inode;
8640 extent_map_tree_init(&ei->extent_tree);
8641 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8642 ei->io_tree.inode = ei;
8643 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8644 IO_TREE_INODE_FILE_EXTENT);
8645 mutex_init(&ei->log_mutex);
8646 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8647 INIT_LIST_HEAD(&ei->delalloc_inodes);
8648 INIT_LIST_HEAD(&ei->delayed_iput);
8649 RB_CLEAR_NODE(&ei->rb_node);
8650 init_rwsem(&ei->i_mmap_lock);
8655 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8656 void btrfs_test_destroy_inode(struct inode *inode)
8658 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8659 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8663 void btrfs_free_inode(struct inode *inode)
8665 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8668 void btrfs_destroy_inode(struct inode *vfs_inode)
8670 struct btrfs_ordered_extent *ordered;
8671 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8672 struct btrfs_root *root = inode->root;
8673 bool freespace_inode;
8675 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8676 WARN_ON(vfs_inode->i_data.nrpages);
8677 WARN_ON(inode->block_rsv.reserved);
8678 WARN_ON(inode->block_rsv.size);
8679 WARN_ON(inode->outstanding_extents);
8680 if (!S_ISDIR(vfs_inode->i_mode)) {
8681 WARN_ON(inode->delalloc_bytes);
8682 WARN_ON(inode->new_delalloc_bytes);
8684 WARN_ON(inode->csum_bytes);
8685 WARN_ON(inode->defrag_bytes);
8688 * This can happen where we create an inode, but somebody else also
8689 * created the same inode and we need to destroy the one we already
8696 * If this is a free space inode do not take the ordered extents lockdep
8699 freespace_inode = btrfs_is_free_space_inode(inode);
8702 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8706 btrfs_err(root->fs_info,
8707 "found ordered extent %llu %llu on inode cleanup",
8708 ordered->file_offset, ordered->num_bytes);
8710 if (!freespace_inode)
8711 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8713 btrfs_remove_ordered_extent(inode, ordered);
8714 btrfs_put_ordered_extent(ordered);
8715 btrfs_put_ordered_extent(ordered);
8718 btrfs_qgroup_check_reserved_leak(inode);
8719 inode_tree_del(inode);
8720 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8721 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8722 btrfs_put_root(inode->root);
8725 int btrfs_drop_inode(struct inode *inode)
8727 struct btrfs_root *root = BTRFS_I(inode)->root;
8732 /* the snap/subvol tree is on deleting */
8733 if (btrfs_root_refs(&root->root_item) == 0)
8736 return generic_drop_inode(inode);
8739 static void init_once(void *foo)
8741 struct btrfs_inode *ei = foo;
8743 inode_init_once(&ei->vfs_inode);
8746 void __cold btrfs_destroy_cachep(void)
8749 * Make sure all delayed rcu free inodes are flushed before we
8753 bioset_exit(&btrfs_dio_bioset);
8754 kmem_cache_destroy(btrfs_inode_cachep);
8757 int __init btrfs_init_cachep(void)
8759 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8760 sizeof(struct btrfs_inode), 0,
8761 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8763 if (!btrfs_inode_cachep)
8766 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8767 offsetof(struct btrfs_dio_private, bbio.bio),
8773 btrfs_destroy_cachep();
8777 static int btrfs_getattr(struct mnt_idmap *idmap,
8778 const struct path *path, struct kstat *stat,
8779 u32 request_mask, unsigned int flags)
8783 struct inode *inode = d_inode(path->dentry);
8784 u32 blocksize = inode->i_sb->s_blocksize;
8785 u32 bi_flags = BTRFS_I(inode)->flags;
8786 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8788 stat->result_mask |= STATX_BTIME;
8789 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8790 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8791 if (bi_flags & BTRFS_INODE_APPEND)
8792 stat->attributes |= STATX_ATTR_APPEND;
8793 if (bi_flags & BTRFS_INODE_COMPRESS)
8794 stat->attributes |= STATX_ATTR_COMPRESSED;
8795 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8796 stat->attributes |= STATX_ATTR_IMMUTABLE;
8797 if (bi_flags & BTRFS_INODE_NODUMP)
8798 stat->attributes |= STATX_ATTR_NODUMP;
8799 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8800 stat->attributes |= STATX_ATTR_VERITY;
8802 stat->attributes_mask |= (STATX_ATTR_APPEND |
8803 STATX_ATTR_COMPRESSED |
8804 STATX_ATTR_IMMUTABLE |
8807 generic_fillattr(idmap, inode, stat);
8808 stat->dev = BTRFS_I(inode)->root->anon_dev;
8810 spin_lock(&BTRFS_I(inode)->lock);
8811 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8812 inode_bytes = inode_get_bytes(inode);
8813 spin_unlock(&BTRFS_I(inode)->lock);
8814 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8815 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8819 static int btrfs_rename_exchange(struct inode *old_dir,
8820 struct dentry *old_dentry,
8821 struct inode *new_dir,
8822 struct dentry *new_dentry)
8824 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8825 struct btrfs_trans_handle *trans;
8826 unsigned int trans_num_items;
8827 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8828 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8829 struct inode *new_inode = new_dentry->d_inode;
8830 struct inode *old_inode = old_dentry->d_inode;
8831 struct timespec64 ctime = current_time(old_inode);
8832 struct btrfs_rename_ctx old_rename_ctx;
8833 struct btrfs_rename_ctx new_rename_ctx;
8834 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8835 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8840 bool need_abort = false;
8841 struct fscrypt_name old_fname, new_fname;
8842 struct fscrypt_str *old_name, *new_name;
8845 * For non-subvolumes allow exchange only within one subvolume, in the
8846 * same inode namespace. Two subvolumes (represented as directory) can
8847 * be exchanged as they're a logical link and have a fixed inode number.
8850 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8851 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8854 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8858 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8860 fscrypt_free_filename(&old_fname);
8864 old_name = &old_fname.disk_name;
8865 new_name = &new_fname.disk_name;
8867 /* close the race window with snapshot create/destroy ioctl */
8868 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8869 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8870 down_read(&fs_info->subvol_sem);
8874 * 1 to remove old dir item
8875 * 1 to remove old dir index
8876 * 1 to add new dir item
8877 * 1 to add new dir index
8878 * 1 to update parent inode
8880 * If the parents are the same, we only need to account for one
8882 trans_num_items = (old_dir == new_dir ? 9 : 10);
8883 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8885 * 1 to remove old root ref
8886 * 1 to remove old root backref
8887 * 1 to add new root ref
8888 * 1 to add new root backref
8890 trans_num_items += 4;
8893 * 1 to update inode item
8894 * 1 to remove old inode ref
8895 * 1 to add new inode ref
8897 trans_num_items += 3;
8899 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8900 trans_num_items += 4;
8902 trans_num_items += 3;
8903 trans = btrfs_start_transaction(root, trans_num_items);
8904 if (IS_ERR(trans)) {
8905 ret = PTR_ERR(trans);
8910 ret = btrfs_record_root_in_trans(trans, dest);
8916 * We need to find a free sequence number both in the source and
8917 * in the destination directory for the exchange.
8919 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8922 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8926 BTRFS_I(old_inode)->dir_index = 0ULL;
8927 BTRFS_I(new_inode)->dir_index = 0ULL;
8929 /* Reference for the source. */
8930 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8931 /* force full log commit if subvolume involved. */
8932 btrfs_set_log_full_commit(trans);
8934 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8935 btrfs_ino(BTRFS_I(new_dir)),
8942 /* And now for the dest. */
8943 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8944 /* force full log commit if subvolume involved. */
8945 btrfs_set_log_full_commit(trans);
8947 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8948 btrfs_ino(BTRFS_I(old_dir)),
8952 btrfs_abort_transaction(trans, ret);
8957 /* Update inode version and ctime/mtime. */
8958 inode_inc_iversion(old_dir);
8959 inode_inc_iversion(new_dir);
8960 inode_inc_iversion(old_inode);
8961 inode_inc_iversion(new_inode);
8962 old_dir->i_mtime = ctime;
8963 old_dir->i_ctime = ctime;
8964 new_dir->i_mtime = ctime;
8965 new_dir->i_ctime = ctime;
8966 old_inode->i_ctime = ctime;
8967 new_inode->i_ctime = ctime;
8969 if (old_dentry->d_parent != new_dentry->d_parent) {
8970 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8971 BTRFS_I(old_inode), true);
8972 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8973 BTRFS_I(new_inode), true);
8976 /* src is a subvolume */
8977 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8978 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8979 } else { /* src is an inode */
8980 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8981 BTRFS_I(old_dentry->d_inode),
8982 old_name, &old_rename_ctx);
8984 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8987 btrfs_abort_transaction(trans, ret);
8991 /* dest is a subvolume */
8992 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8993 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8994 } else { /* dest is an inode */
8995 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8996 BTRFS_I(new_dentry->d_inode),
8997 new_name, &new_rename_ctx);
8999 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
9002 btrfs_abort_transaction(trans, ret);
9006 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9007 new_name, 0, old_idx);
9009 btrfs_abort_transaction(trans, ret);
9013 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9014 old_name, 0, new_idx);
9016 btrfs_abort_transaction(trans, ret);
9020 if (old_inode->i_nlink == 1)
9021 BTRFS_I(old_inode)->dir_index = old_idx;
9022 if (new_inode->i_nlink == 1)
9023 BTRFS_I(new_inode)->dir_index = new_idx;
9026 * Now pin the logs of the roots. We do it to ensure that no other task
9027 * can sync the logs while we are in progress with the rename, because
9028 * that could result in an inconsistency in case any of the inodes that
9029 * are part of this rename operation were logged before.
9031 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9032 btrfs_pin_log_trans(root);
9033 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9034 btrfs_pin_log_trans(dest);
9036 /* Do the log updates for all inodes. */
9037 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9038 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9039 old_rename_ctx.index, new_dentry->d_parent);
9040 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9041 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9042 new_rename_ctx.index, old_dentry->d_parent);
9044 /* Now unpin the logs. */
9045 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9046 btrfs_end_log_trans(root);
9047 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9048 btrfs_end_log_trans(dest);
9050 ret2 = btrfs_end_transaction(trans);
9051 ret = ret ? ret : ret2;
9053 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9054 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9055 up_read(&fs_info->subvol_sem);
9057 fscrypt_free_filename(&new_fname);
9058 fscrypt_free_filename(&old_fname);
9062 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9065 struct inode *inode;
9067 inode = new_inode(dir->i_sb);
9069 inode_init_owner(idmap, inode, dir,
9070 S_IFCHR | WHITEOUT_MODE);
9071 inode->i_op = &btrfs_special_inode_operations;
9072 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9077 static int btrfs_rename(struct mnt_idmap *idmap,
9078 struct inode *old_dir, struct dentry *old_dentry,
9079 struct inode *new_dir, struct dentry *new_dentry,
9082 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9083 struct btrfs_new_inode_args whiteout_args = {
9085 .dentry = old_dentry,
9087 struct btrfs_trans_handle *trans;
9088 unsigned int trans_num_items;
9089 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9090 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9091 struct inode *new_inode = d_inode(new_dentry);
9092 struct inode *old_inode = d_inode(old_dentry);
9093 struct btrfs_rename_ctx rename_ctx;
9097 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9098 struct fscrypt_name old_fname, new_fname;
9100 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9103 /* we only allow rename subvolume link between subvolumes */
9104 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9107 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9108 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9111 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9112 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9115 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9119 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9121 fscrypt_free_filename(&old_fname);
9125 /* check for collisions, even if the name isn't there */
9126 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9128 if (ret == -EEXIST) {
9130 * eexist without a new_inode */
9131 if (WARN_ON(!new_inode)) {
9132 goto out_fscrypt_names;
9135 /* maybe -EOVERFLOW */
9136 goto out_fscrypt_names;
9142 * we're using rename to replace one file with another. Start IO on it
9143 * now so we don't add too much work to the end of the transaction
9145 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9146 filemap_flush(old_inode->i_mapping);
9148 if (flags & RENAME_WHITEOUT) {
9149 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9150 if (!whiteout_args.inode) {
9152 goto out_fscrypt_names;
9154 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9156 goto out_whiteout_inode;
9158 /* 1 to update the old parent inode. */
9159 trans_num_items = 1;
9162 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9163 /* Close the race window with snapshot create/destroy ioctl */
9164 down_read(&fs_info->subvol_sem);
9166 * 1 to remove old root ref
9167 * 1 to remove old root backref
9168 * 1 to add new root ref
9169 * 1 to add new root backref
9171 trans_num_items += 4;
9175 * 1 to remove old inode ref
9176 * 1 to add new inode ref
9178 trans_num_items += 3;
9181 * 1 to remove old dir item
9182 * 1 to remove old dir index
9183 * 1 to add new dir item
9184 * 1 to add new dir index
9186 trans_num_items += 4;
9187 /* 1 to update new parent inode if it's not the same as the old parent */
9188 if (new_dir != old_dir)
9193 * 1 to remove inode ref
9194 * 1 to remove dir item
9195 * 1 to remove dir index
9196 * 1 to possibly add orphan item
9198 trans_num_items += 5;
9200 trans = btrfs_start_transaction(root, trans_num_items);
9201 if (IS_ERR(trans)) {
9202 ret = PTR_ERR(trans);
9207 ret = btrfs_record_root_in_trans(trans, dest);
9212 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9216 BTRFS_I(old_inode)->dir_index = 0ULL;
9217 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9218 /* force full log commit if subvolume involved. */
9219 btrfs_set_log_full_commit(trans);
9221 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9222 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9228 inode_inc_iversion(old_dir);
9229 inode_inc_iversion(new_dir);
9230 inode_inc_iversion(old_inode);
9231 old_dir->i_mtime = current_time(old_dir);
9232 old_dir->i_ctime = old_dir->i_mtime;
9233 new_dir->i_mtime = old_dir->i_mtime;
9234 new_dir->i_ctime = old_dir->i_mtime;
9235 old_inode->i_ctime = old_dir->i_mtime;
9237 if (old_dentry->d_parent != new_dentry->d_parent)
9238 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9239 BTRFS_I(old_inode), true);
9241 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9242 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9244 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9245 BTRFS_I(d_inode(old_dentry)),
9246 &old_fname.disk_name, &rename_ctx);
9248 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9251 btrfs_abort_transaction(trans, ret);
9256 inode_inc_iversion(new_inode);
9257 new_inode->i_ctime = current_time(new_inode);
9258 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9259 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9260 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9261 BUG_ON(new_inode->i_nlink == 0);
9263 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9264 BTRFS_I(d_inode(new_dentry)),
9265 &new_fname.disk_name);
9267 if (!ret && new_inode->i_nlink == 0)
9268 ret = btrfs_orphan_add(trans,
9269 BTRFS_I(d_inode(new_dentry)));
9271 btrfs_abort_transaction(trans, ret);
9276 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9277 &new_fname.disk_name, 0, index);
9279 btrfs_abort_transaction(trans, ret);
9283 if (old_inode->i_nlink == 1)
9284 BTRFS_I(old_inode)->dir_index = index;
9286 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9287 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9288 rename_ctx.index, new_dentry->d_parent);
9290 if (flags & RENAME_WHITEOUT) {
9291 ret = btrfs_create_new_inode(trans, &whiteout_args);
9293 btrfs_abort_transaction(trans, ret);
9296 unlock_new_inode(whiteout_args.inode);
9297 iput(whiteout_args.inode);
9298 whiteout_args.inode = NULL;
9302 ret2 = btrfs_end_transaction(trans);
9303 ret = ret ? ret : ret2;
9305 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9306 up_read(&fs_info->subvol_sem);
9307 if (flags & RENAME_WHITEOUT)
9308 btrfs_new_inode_args_destroy(&whiteout_args);
9310 if (flags & RENAME_WHITEOUT)
9311 iput(whiteout_args.inode);
9313 fscrypt_free_filename(&old_fname);
9314 fscrypt_free_filename(&new_fname);
9318 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9319 struct dentry *old_dentry, struct inode *new_dir,
9320 struct dentry *new_dentry, unsigned int flags)
9324 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9327 if (flags & RENAME_EXCHANGE)
9328 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9331 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9334 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9339 struct btrfs_delalloc_work {
9340 struct inode *inode;
9341 struct completion completion;
9342 struct list_head list;
9343 struct btrfs_work work;
9346 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9348 struct btrfs_delalloc_work *delalloc_work;
9349 struct inode *inode;
9351 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9353 inode = delalloc_work->inode;
9354 filemap_flush(inode->i_mapping);
9355 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9356 &BTRFS_I(inode)->runtime_flags))
9357 filemap_flush(inode->i_mapping);
9360 complete(&delalloc_work->completion);
9363 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9365 struct btrfs_delalloc_work *work;
9367 work = kmalloc(sizeof(*work), GFP_NOFS);
9371 init_completion(&work->completion);
9372 INIT_LIST_HEAD(&work->list);
9373 work->inode = inode;
9374 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9380 * some fairly slow code that needs optimization. This walks the list
9381 * of all the inodes with pending delalloc and forces them to disk.
9383 static int start_delalloc_inodes(struct btrfs_root *root,
9384 struct writeback_control *wbc, bool snapshot,
9385 bool in_reclaim_context)
9387 struct btrfs_inode *binode;
9388 struct inode *inode;
9389 struct btrfs_delalloc_work *work, *next;
9390 struct list_head works;
9391 struct list_head splice;
9393 bool full_flush = wbc->nr_to_write == LONG_MAX;
9395 INIT_LIST_HEAD(&works);
9396 INIT_LIST_HEAD(&splice);
9398 mutex_lock(&root->delalloc_mutex);
9399 spin_lock(&root->delalloc_lock);
9400 list_splice_init(&root->delalloc_inodes, &splice);
9401 while (!list_empty(&splice)) {
9402 binode = list_entry(splice.next, struct btrfs_inode,
9405 list_move_tail(&binode->delalloc_inodes,
9406 &root->delalloc_inodes);
9408 if (in_reclaim_context &&
9409 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9412 inode = igrab(&binode->vfs_inode);
9414 cond_resched_lock(&root->delalloc_lock);
9417 spin_unlock(&root->delalloc_lock);
9420 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9421 &binode->runtime_flags);
9423 work = btrfs_alloc_delalloc_work(inode);
9429 list_add_tail(&work->list, &works);
9430 btrfs_queue_work(root->fs_info->flush_workers,
9433 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9434 btrfs_add_delayed_iput(BTRFS_I(inode));
9435 if (ret || wbc->nr_to_write <= 0)
9439 spin_lock(&root->delalloc_lock);
9441 spin_unlock(&root->delalloc_lock);
9444 list_for_each_entry_safe(work, next, &works, list) {
9445 list_del_init(&work->list);
9446 wait_for_completion(&work->completion);
9450 if (!list_empty(&splice)) {
9451 spin_lock(&root->delalloc_lock);
9452 list_splice_tail(&splice, &root->delalloc_inodes);
9453 spin_unlock(&root->delalloc_lock);
9455 mutex_unlock(&root->delalloc_mutex);
9459 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9461 struct writeback_control wbc = {
9462 .nr_to_write = LONG_MAX,
9463 .sync_mode = WB_SYNC_NONE,
9465 .range_end = LLONG_MAX,
9467 struct btrfs_fs_info *fs_info = root->fs_info;
9469 if (BTRFS_FS_ERROR(fs_info))
9472 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9475 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9476 bool in_reclaim_context)
9478 struct writeback_control wbc = {
9480 .sync_mode = WB_SYNC_NONE,
9482 .range_end = LLONG_MAX,
9484 struct btrfs_root *root;
9485 struct list_head splice;
9488 if (BTRFS_FS_ERROR(fs_info))
9491 INIT_LIST_HEAD(&splice);
9493 mutex_lock(&fs_info->delalloc_root_mutex);
9494 spin_lock(&fs_info->delalloc_root_lock);
9495 list_splice_init(&fs_info->delalloc_roots, &splice);
9496 while (!list_empty(&splice)) {
9498 * Reset nr_to_write here so we know that we're doing a full
9502 wbc.nr_to_write = LONG_MAX;
9504 root = list_first_entry(&splice, struct btrfs_root,
9506 root = btrfs_grab_root(root);
9508 list_move_tail(&root->delalloc_root,
9509 &fs_info->delalloc_roots);
9510 spin_unlock(&fs_info->delalloc_root_lock);
9512 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9513 btrfs_put_root(root);
9514 if (ret < 0 || wbc.nr_to_write <= 0)
9516 spin_lock(&fs_info->delalloc_root_lock);
9518 spin_unlock(&fs_info->delalloc_root_lock);
9522 if (!list_empty(&splice)) {
9523 spin_lock(&fs_info->delalloc_root_lock);
9524 list_splice_tail(&splice, &fs_info->delalloc_roots);
9525 spin_unlock(&fs_info->delalloc_root_lock);
9527 mutex_unlock(&fs_info->delalloc_root_mutex);
9531 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9532 struct dentry *dentry, const char *symname)
9534 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9535 struct btrfs_trans_handle *trans;
9536 struct btrfs_root *root = BTRFS_I(dir)->root;
9537 struct btrfs_path *path;
9538 struct btrfs_key key;
9539 struct inode *inode;
9540 struct btrfs_new_inode_args new_inode_args = {
9544 unsigned int trans_num_items;
9549 struct btrfs_file_extent_item *ei;
9550 struct extent_buffer *leaf;
9552 name_len = strlen(symname);
9553 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9554 return -ENAMETOOLONG;
9556 inode = new_inode(dir->i_sb);
9559 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9560 inode->i_op = &btrfs_symlink_inode_operations;
9561 inode_nohighmem(inode);
9562 inode->i_mapping->a_ops = &btrfs_aops;
9563 btrfs_i_size_write(BTRFS_I(inode), name_len);
9564 inode_set_bytes(inode, name_len);
9566 new_inode_args.inode = inode;
9567 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9570 /* 1 additional item for the inline extent */
9573 trans = btrfs_start_transaction(root, trans_num_items);
9574 if (IS_ERR(trans)) {
9575 err = PTR_ERR(trans);
9576 goto out_new_inode_args;
9579 err = btrfs_create_new_inode(trans, &new_inode_args);
9583 path = btrfs_alloc_path();
9586 btrfs_abort_transaction(trans, err);
9587 discard_new_inode(inode);
9591 key.objectid = btrfs_ino(BTRFS_I(inode));
9593 key.type = BTRFS_EXTENT_DATA_KEY;
9594 datasize = btrfs_file_extent_calc_inline_size(name_len);
9595 err = btrfs_insert_empty_item(trans, root, path, &key,
9598 btrfs_abort_transaction(trans, err);
9599 btrfs_free_path(path);
9600 discard_new_inode(inode);
9604 leaf = path->nodes[0];
9605 ei = btrfs_item_ptr(leaf, path->slots[0],
9606 struct btrfs_file_extent_item);
9607 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9608 btrfs_set_file_extent_type(leaf, ei,
9609 BTRFS_FILE_EXTENT_INLINE);
9610 btrfs_set_file_extent_encryption(leaf, ei, 0);
9611 btrfs_set_file_extent_compression(leaf, ei, 0);
9612 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9613 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9615 ptr = btrfs_file_extent_inline_start(ei);
9616 write_extent_buffer(leaf, symname, ptr, name_len);
9617 btrfs_mark_buffer_dirty(leaf);
9618 btrfs_free_path(path);
9620 d_instantiate_new(dentry, inode);
9623 btrfs_end_transaction(trans);
9624 btrfs_btree_balance_dirty(fs_info);
9626 btrfs_new_inode_args_destroy(&new_inode_args);
9633 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9634 struct btrfs_trans_handle *trans_in,
9635 struct btrfs_inode *inode,
9636 struct btrfs_key *ins,
9639 struct btrfs_file_extent_item stack_fi;
9640 struct btrfs_replace_extent_info extent_info;
9641 struct btrfs_trans_handle *trans = trans_in;
9642 struct btrfs_path *path;
9643 u64 start = ins->objectid;
9644 u64 len = ins->offset;
9645 int qgroup_released;
9648 memset(&stack_fi, 0, sizeof(stack_fi));
9650 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9651 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9652 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9653 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9654 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9655 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9656 /* Encryption and other encoding is reserved and all 0 */
9658 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9659 if (qgroup_released < 0)
9660 return ERR_PTR(qgroup_released);
9663 ret = insert_reserved_file_extent(trans, inode,
9664 file_offset, &stack_fi,
9665 true, qgroup_released);
9671 extent_info.disk_offset = start;
9672 extent_info.disk_len = len;
9673 extent_info.data_offset = 0;
9674 extent_info.data_len = len;
9675 extent_info.file_offset = file_offset;
9676 extent_info.extent_buf = (char *)&stack_fi;
9677 extent_info.is_new_extent = true;
9678 extent_info.update_times = true;
9679 extent_info.qgroup_reserved = qgroup_released;
9680 extent_info.insertions = 0;
9682 path = btrfs_alloc_path();
9688 ret = btrfs_replace_file_extents(inode, path, file_offset,
9689 file_offset + len - 1, &extent_info,
9691 btrfs_free_path(path);
9698 * We have released qgroup data range at the beginning of the function,
9699 * and normally qgroup_released bytes will be freed when committing
9701 * But if we error out early, we have to free what we have released
9702 * or we leak qgroup data reservation.
9704 btrfs_qgroup_free_refroot(inode->root->fs_info,
9705 inode->root->root_key.objectid, qgroup_released,
9706 BTRFS_QGROUP_RSV_DATA);
9707 return ERR_PTR(ret);
9710 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9711 u64 start, u64 num_bytes, u64 min_size,
9712 loff_t actual_len, u64 *alloc_hint,
9713 struct btrfs_trans_handle *trans)
9715 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9716 struct extent_map *em;
9717 struct btrfs_root *root = BTRFS_I(inode)->root;
9718 struct btrfs_key ins;
9719 u64 cur_offset = start;
9720 u64 clear_offset = start;
9723 u64 last_alloc = (u64)-1;
9725 bool own_trans = true;
9726 u64 end = start + num_bytes - 1;
9730 while (num_bytes > 0) {
9731 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9732 cur_bytes = max(cur_bytes, min_size);
9734 * If we are severely fragmented we could end up with really
9735 * small allocations, so if the allocator is returning small
9736 * chunks lets make its job easier by only searching for those
9739 cur_bytes = min(cur_bytes, last_alloc);
9740 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9741 min_size, 0, *alloc_hint, &ins, 1, 0);
9746 * We've reserved this space, and thus converted it from
9747 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9748 * from here on out we will only need to clear our reservation
9749 * for the remaining unreserved area, so advance our
9750 * clear_offset by our extent size.
9752 clear_offset += ins.offset;
9754 last_alloc = ins.offset;
9755 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9758 * Now that we inserted the prealloc extent we can finally
9759 * decrement the number of reservations in the block group.
9760 * If we did it before, we could race with relocation and have
9761 * relocation miss the reserved extent, making it fail later.
9763 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9764 if (IS_ERR(trans)) {
9765 ret = PTR_ERR(trans);
9766 btrfs_free_reserved_extent(fs_info, ins.objectid,
9771 em = alloc_extent_map();
9773 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9774 cur_offset + ins.offset - 1, false);
9775 btrfs_set_inode_full_sync(BTRFS_I(inode));
9779 em->start = cur_offset;
9780 em->orig_start = cur_offset;
9781 em->len = ins.offset;
9782 em->block_start = ins.objectid;
9783 em->block_len = ins.offset;
9784 em->orig_block_len = ins.offset;
9785 em->ram_bytes = ins.offset;
9786 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9787 em->generation = trans->transid;
9789 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9790 free_extent_map(em);
9792 num_bytes -= ins.offset;
9793 cur_offset += ins.offset;
9794 *alloc_hint = ins.objectid + ins.offset;
9796 inode_inc_iversion(inode);
9797 inode->i_ctime = current_time(inode);
9798 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9799 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9800 (actual_len > inode->i_size) &&
9801 (cur_offset > inode->i_size)) {
9802 if (cur_offset > actual_len)
9803 i_size = actual_len;
9805 i_size = cur_offset;
9806 i_size_write(inode, i_size);
9807 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9810 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9813 btrfs_abort_transaction(trans, ret);
9815 btrfs_end_transaction(trans);
9820 btrfs_end_transaction(trans);
9824 if (clear_offset < end)
9825 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9826 end - clear_offset + 1);
9830 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9831 u64 start, u64 num_bytes, u64 min_size,
9832 loff_t actual_len, u64 *alloc_hint)
9834 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9835 min_size, actual_len, alloc_hint,
9839 int btrfs_prealloc_file_range_trans(struct inode *inode,
9840 struct btrfs_trans_handle *trans, int mode,
9841 u64 start, u64 num_bytes, u64 min_size,
9842 loff_t actual_len, u64 *alloc_hint)
9844 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9845 min_size, actual_len, alloc_hint, trans);
9848 static int btrfs_permission(struct mnt_idmap *idmap,
9849 struct inode *inode, int mask)
9851 struct btrfs_root *root = BTRFS_I(inode)->root;
9852 umode_t mode = inode->i_mode;
9854 if (mask & MAY_WRITE &&
9855 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9856 if (btrfs_root_readonly(root))
9858 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9861 return generic_permission(idmap, inode, mask);
9864 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9865 struct file *file, umode_t mode)
9867 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9868 struct btrfs_trans_handle *trans;
9869 struct btrfs_root *root = BTRFS_I(dir)->root;
9870 struct inode *inode;
9871 struct btrfs_new_inode_args new_inode_args = {
9873 .dentry = file->f_path.dentry,
9876 unsigned int trans_num_items;
9879 inode = new_inode(dir->i_sb);
9882 inode_init_owner(idmap, inode, dir, mode);
9883 inode->i_fop = &btrfs_file_operations;
9884 inode->i_op = &btrfs_file_inode_operations;
9885 inode->i_mapping->a_ops = &btrfs_aops;
9887 new_inode_args.inode = inode;
9888 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9892 trans = btrfs_start_transaction(root, trans_num_items);
9893 if (IS_ERR(trans)) {
9894 ret = PTR_ERR(trans);
9895 goto out_new_inode_args;
9898 ret = btrfs_create_new_inode(trans, &new_inode_args);
9901 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9902 * set it to 1 because d_tmpfile() will issue a warning if the count is
9905 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9907 set_nlink(inode, 1);
9910 d_tmpfile(file, inode);
9911 unlock_new_inode(inode);
9912 mark_inode_dirty(inode);
9915 btrfs_end_transaction(trans);
9916 btrfs_btree_balance_dirty(fs_info);
9918 btrfs_new_inode_args_destroy(&new_inode_args);
9922 return finish_open_simple(file, ret);
9925 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9927 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9928 unsigned long index = start >> PAGE_SHIFT;
9929 unsigned long end_index = end >> PAGE_SHIFT;
9933 ASSERT(end + 1 - start <= U32_MAX);
9934 len = end + 1 - start;
9935 while (index <= end_index) {
9936 page = find_get_page(inode->vfs_inode.i_mapping, index);
9937 ASSERT(page); /* Pages should be in the extent_io_tree */
9939 btrfs_page_set_writeback(fs_info, page, start, len);
9945 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9948 switch (compress_type) {
9949 case BTRFS_COMPRESS_NONE:
9950 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9951 case BTRFS_COMPRESS_ZLIB:
9952 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9953 case BTRFS_COMPRESS_LZO:
9955 * The LZO format depends on the sector size. 64K is the maximum
9956 * sector size that we support.
9958 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9960 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9961 (fs_info->sectorsize_bits - 12);
9962 case BTRFS_COMPRESS_ZSTD:
9963 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9969 static ssize_t btrfs_encoded_read_inline(
9971 struct iov_iter *iter, u64 start,
9973 struct extent_state **cached_state,
9974 u64 extent_start, size_t count,
9975 struct btrfs_ioctl_encoded_io_args *encoded,
9978 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9979 struct btrfs_root *root = inode->root;
9980 struct btrfs_fs_info *fs_info = root->fs_info;
9981 struct extent_io_tree *io_tree = &inode->io_tree;
9982 struct btrfs_path *path;
9983 struct extent_buffer *leaf;
9984 struct btrfs_file_extent_item *item;
9990 path = btrfs_alloc_path();
9995 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9999 /* The extent item disappeared? */
10004 leaf = path->nodes[0];
10005 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10007 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10008 ptr = btrfs_file_extent_inline_start(item);
10010 encoded->len = min_t(u64, extent_start + ram_bytes,
10011 inode->vfs_inode.i_size) - iocb->ki_pos;
10012 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10013 btrfs_file_extent_compression(leaf, item));
10016 encoded->compression = ret;
10017 if (encoded->compression) {
10018 size_t inline_size;
10020 inline_size = btrfs_file_extent_inline_item_len(leaf,
10022 if (inline_size > count) {
10026 count = inline_size;
10027 encoded->unencoded_len = ram_bytes;
10028 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10030 count = min_t(u64, count, encoded->len);
10031 encoded->len = count;
10032 encoded->unencoded_len = count;
10033 ptr += iocb->ki_pos - extent_start;
10036 tmp = kmalloc(count, GFP_NOFS);
10041 read_extent_buffer(leaf, tmp, ptr, count);
10042 btrfs_release_path(path);
10043 unlock_extent(io_tree, start, lockend, cached_state);
10044 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10047 ret = copy_to_iter(tmp, count, iter);
10052 btrfs_free_path(path);
10056 struct btrfs_encoded_read_private {
10057 wait_queue_head_t wait;
10059 blk_status_t status;
10062 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10064 struct btrfs_encoded_read_private *priv = bbio->private;
10066 if (bbio->bio.bi_status) {
10068 * The memory barrier implied by the atomic_dec_return() here
10069 * pairs with the memory barrier implied by the
10070 * atomic_dec_return() or io_wait_event() in
10071 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10072 * write is observed before the load of status in
10073 * btrfs_encoded_read_regular_fill_pages().
10075 WRITE_ONCE(priv->status, bbio->bio.bi_status);
10077 if (!atomic_dec_return(&priv->pending))
10078 wake_up(&priv->wait);
10079 bio_put(&bbio->bio);
10082 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10083 u64 file_offset, u64 disk_bytenr,
10084 u64 disk_io_size, struct page **pages)
10086 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10087 struct btrfs_encoded_read_private priv = {
10088 .pending = ATOMIC_INIT(1),
10090 unsigned long i = 0;
10091 struct btrfs_bio *bbio;
10093 init_waitqueue_head(&priv.wait);
10095 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10096 btrfs_encoded_read_endio, &priv);
10097 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10098 bbio->inode = inode;
10101 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10103 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10104 atomic_inc(&priv.pending);
10105 btrfs_submit_bio(bbio, 0);
10107 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10108 btrfs_encoded_read_endio, &priv);
10109 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10110 bbio->inode = inode;
10115 disk_bytenr += bytes;
10116 disk_io_size -= bytes;
10117 } while (disk_io_size);
10119 atomic_inc(&priv.pending);
10120 btrfs_submit_bio(bbio, 0);
10122 if (atomic_dec_return(&priv.pending))
10123 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10124 /* See btrfs_encoded_read_endio() for ordering. */
10125 return blk_status_to_errno(READ_ONCE(priv.status));
10128 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10129 struct iov_iter *iter,
10130 u64 start, u64 lockend,
10131 struct extent_state **cached_state,
10132 u64 disk_bytenr, u64 disk_io_size,
10133 size_t count, bool compressed,
10136 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10137 struct extent_io_tree *io_tree = &inode->io_tree;
10138 struct page **pages;
10139 unsigned long nr_pages, i;
10141 size_t page_offset;
10144 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10145 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10148 ret = btrfs_alloc_page_array(nr_pages, pages);
10154 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10155 disk_io_size, pages);
10159 unlock_extent(io_tree, start, lockend, cached_state);
10160 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10167 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10168 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10171 while (cur < count) {
10172 size_t bytes = min_t(size_t, count - cur,
10173 PAGE_SIZE - page_offset);
10175 if (copy_page_to_iter(pages[i], page_offset, bytes,
10186 for (i = 0; i < nr_pages; i++) {
10188 __free_page(pages[i]);
10194 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10195 struct btrfs_ioctl_encoded_io_args *encoded)
10197 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10198 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10199 struct extent_io_tree *io_tree = &inode->io_tree;
10201 size_t count = iov_iter_count(iter);
10202 u64 start, lockend, disk_bytenr, disk_io_size;
10203 struct extent_state *cached_state = NULL;
10204 struct extent_map *em;
10205 bool unlocked = false;
10207 file_accessed(iocb->ki_filp);
10209 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10211 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10212 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10215 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10217 * We don't know how long the extent containing iocb->ki_pos is, but if
10218 * it's compressed we know that it won't be longer than this.
10220 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10223 struct btrfs_ordered_extent *ordered;
10225 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10226 lockend - start + 1);
10228 goto out_unlock_inode;
10229 lock_extent(io_tree, start, lockend, &cached_state);
10230 ordered = btrfs_lookup_ordered_range(inode, start,
10231 lockend - start + 1);
10234 btrfs_put_ordered_extent(ordered);
10235 unlock_extent(io_tree, start, lockend, &cached_state);
10239 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10242 goto out_unlock_extent;
10245 if (em->block_start == EXTENT_MAP_INLINE) {
10246 u64 extent_start = em->start;
10249 * For inline extents we get everything we need out of the
10252 free_extent_map(em);
10254 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10255 &cached_state, extent_start,
10256 count, encoded, &unlocked);
10261 * We only want to return up to EOF even if the extent extends beyond
10264 encoded->len = min_t(u64, extent_map_end(em),
10265 inode->vfs_inode.i_size) - iocb->ki_pos;
10266 if (em->block_start == EXTENT_MAP_HOLE ||
10267 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10268 disk_bytenr = EXTENT_MAP_HOLE;
10269 count = min_t(u64, count, encoded->len);
10270 encoded->len = count;
10271 encoded->unencoded_len = count;
10272 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10273 disk_bytenr = em->block_start;
10275 * Bail if the buffer isn't large enough to return the whole
10276 * compressed extent.
10278 if (em->block_len > count) {
10282 disk_io_size = em->block_len;
10283 count = em->block_len;
10284 encoded->unencoded_len = em->ram_bytes;
10285 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10286 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10287 em->compress_type);
10290 encoded->compression = ret;
10292 disk_bytenr = em->block_start + (start - em->start);
10293 if (encoded->len > count)
10294 encoded->len = count;
10296 * Don't read beyond what we locked. This also limits the page
10297 * allocations that we'll do.
10299 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10300 count = start + disk_io_size - iocb->ki_pos;
10301 encoded->len = count;
10302 encoded->unencoded_len = count;
10303 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10305 free_extent_map(em);
10308 if (disk_bytenr == EXTENT_MAP_HOLE) {
10309 unlock_extent(io_tree, start, lockend, &cached_state);
10310 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10312 ret = iov_iter_zero(count, iter);
10316 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10317 &cached_state, disk_bytenr,
10318 disk_io_size, count,
10319 encoded->compression,
10325 iocb->ki_pos += encoded->len;
10327 free_extent_map(em);
10330 unlock_extent(io_tree, start, lockend, &cached_state);
10333 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10337 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10338 const struct btrfs_ioctl_encoded_io_args *encoded)
10340 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10341 struct btrfs_root *root = inode->root;
10342 struct btrfs_fs_info *fs_info = root->fs_info;
10343 struct extent_io_tree *io_tree = &inode->io_tree;
10344 struct extent_changeset *data_reserved = NULL;
10345 struct extent_state *cached_state = NULL;
10346 struct btrfs_ordered_extent *ordered;
10350 u64 num_bytes, ram_bytes, disk_num_bytes;
10351 unsigned long nr_pages, i;
10352 struct page **pages;
10353 struct btrfs_key ins;
10354 bool extent_reserved = false;
10355 struct extent_map *em;
10358 switch (encoded->compression) {
10359 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10360 compression = BTRFS_COMPRESS_ZLIB;
10362 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10363 compression = BTRFS_COMPRESS_ZSTD;
10365 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10366 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10367 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10368 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10369 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10370 /* The sector size must match for LZO. */
10371 if (encoded->compression -
10372 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10373 fs_info->sectorsize_bits)
10375 compression = BTRFS_COMPRESS_LZO;
10380 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10383 orig_count = iov_iter_count(from);
10385 /* The extent size must be sane. */
10386 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10387 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10391 * The compressed data must be smaller than the decompressed data.
10393 * It's of course possible for data to compress to larger or the same
10394 * size, but the buffered I/O path falls back to no compression for such
10395 * data, and we don't want to break any assumptions by creating these
10398 * Note that this is less strict than the current check we have that the
10399 * compressed data must be at least one sector smaller than the
10400 * decompressed data. We only want to enforce the weaker requirement
10401 * from old kernels that it is at least one byte smaller.
10403 if (orig_count >= encoded->unencoded_len)
10406 /* The extent must start on a sector boundary. */
10407 start = iocb->ki_pos;
10408 if (!IS_ALIGNED(start, fs_info->sectorsize))
10412 * The extent must end on a sector boundary. However, we allow a write
10413 * which ends at or extends i_size to have an unaligned length; we round
10414 * up the extent size and set i_size to the unaligned end.
10416 if (start + encoded->len < inode->vfs_inode.i_size &&
10417 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10420 /* Finally, the offset in the unencoded data must be sector-aligned. */
10421 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10424 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10425 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10426 end = start + num_bytes - 1;
10429 * If the extent cannot be inline, the compressed data on disk must be
10430 * sector-aligned. For convenience, we extend it with zeroes if it
10433 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10434 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10435 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10438 for (i = 0; i < nr_pages; i++) {
10439 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10442 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10447 kaddr = kmap_local_page(pages[i]);
10448 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10449 kunmap_local(kaddr);
10453 if (bytes < PAGE_SIZE)
10454 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10455 kunmap_local(kaddr);
10459 struct btrfs_ordered_extent *ordered;
10461 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10464 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10465 start >> PAGE_SHIFT,
10466 end >> PAGE_SHIFT);
10469 lock_extent(io_tree, start, end, &cached_state);
10470 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10472 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10475 btrfs_put_ordered_extent(ordered);
10476 unlock_extent(io_tree, start, end, &cached_state);
10481 * We don't use the higher-level delalloc space functions because our
10482 * num_bytes and disk_num_bytes are different.
10484 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10487 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10489 goto out_free_data_space;
10490 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10493 goto out_qgroup_free_data;
10495 /* Try an inline extent first. */
10496 if (start == 0 && encoded->unencoded_len == encoded->len &&
10497 encoded->unencoded_offset == 0) {
10498 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10499 compression, pages, true);
10503 goto out_delalloc_release;
10507 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10508 disk_num_bytes, 0, 0, &ins, 1, 1);
10510 goto out_delalloc_release;
10511 extent_reserved = true;
10513 em = create_io_em(inode, start, num_bytes,
10514 start - encoded->unencoded_offset, ins.objectid,
10515 ins.offset, ins.offset, ram_bytes, compression,
10516 BTRFS_ORDERED_COMPRESSED);
10519 goto out_free_reserved;
10521 free_extent_map(em);
10523 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10524 ins.objectid, ins.offset,
10525 encoded->unencoded_offset,
10526 (1 << BTRFS_ORDERED_ENCODED) |
10527 (1 << BTRFS_ORDERED_COMPRESSED),
10529 if (IS_ERR(ordered)) {
10530 btrfs_drop_extent_map_range(inode, start, end, false);
10531 ret = PTR_ERR(ordered);
10532 goto out_free_reserved;
10534 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10536 if (start + encoded->len > inode->vfs_inode.i_size)
10537 i_size_write(&inode->vfs_inode, start + encoded->len);
10539 unlock_extent(io_tree, start, end, &cached_state);
10541 btrfs_delalloc_release_extents(inode, num_bytes);
10543 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10548 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10549 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10550 out_delalloc_release:
10551 btrfs_delalloc_release_extents(inode, num_bytes);
10552 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10553 out_qgroup_free_data:
10555 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10556 out_free_data_space:
10558 * If btrfs_reserve_extent() succeeded, then we already decremented
10561 if (!extent_reserved)
10562 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10564 unlock_extent(io_tree, start, end, &cached_state);
10566 for (i = 0; i < nr_pages; i++) {
10568 __free_page(pages[i]);
10573 iocb->ki_pos += encoded->len;
10579 * Add an entry indicating a block group or device which is pinned by a
10580 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10581 * negative errno on failure.
10583 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10584 bool is_block_group)
10586 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10587 struct btrfs_swapfile_pin *sp, *entry;
10588 struct rb_node **p;
10589 struct rb_node *parent = NULL;
10591 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10596 sp->is_block_group = is_block_group;
10597 sp->bg_extent_count = 1;
10599 spin_lock(&fs_info->swapfile_pins_lock);
10600 p = &fs_info->swapfile_pins.rb_node;
10603 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10604 if (sp->ptr < entry->ptr ||
10605 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10606 p = &(*p)->rb_left;
10607 } else if (sp->ptr > entry->ptr ||
10608 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10609 p = &(*p)->rb_right;
10611 if (is_block_group)
10612 entry->bg_extent_count++;
10613 spin_unlock(&fs_info->swapfile_pins_lock);
10618 rb_link_node(&sp->node, parent, p);
10619 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10620 spin_unlock(&fs_info->swapfile_pins_lock);
10624 /* Free all of the entries pinned by this swapfile. */
10625 static void btrfs_free_swapfile_pins(struct inode *inode)
10627 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10628 struct btrfs_swapfile_pin *sp;
10629 struct rb_node *node, *next;
10631 spin_lock(&fs_info->swapfile_pins_lock);
10632 node = rb_first(&fs_info->swapfile_pins);
10634 next = rb_next(node);
10635 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10636 if (sp->inode == inode) {
10637 rb_erase(&sp->node, &fs_info->swapfile_pins);
10638 if (sp->is_block_group) {
10639 btrfs_dec_block_group_swap_extents(sp->ptr,
10640 sp->bg_extent_count);
10641 btrfs_put_block_group(sp->ptr);
10647 spin_unlock(&fs_info->swapfile_pins_lock);
10650 struct btrfs_swap_info {
10656 unsigned long nr_pages;
10660 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10661 struct btrfs_swap_info *bsi)
10663 unsigned long nr_pages;
10664 unsigned long max_pages;
10665 u64 first_ppage, first_ppage_reported, next_ppage;
10669 * Our swapfile may have had its size extended after the swap header was
10670 * written. In that case activating the swapfile should not go beyond
10671 * the max size set in the swap header.
10673 if (bsi->nr_pages >= sis->max)
10676 max_pages = sis->max - bsi->nr_pages;
10677 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10678 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10680 if (first_ppage >= next_ppage)
10682 nr_pages = next_ppage - first_ppage;
10683 nr_pages = min(nr_pages, max_pages);
10685 first_ppage_reported = first_ppage;
10686 if (bsi->start == 0)
10687 first_ppage_reported++;
10688 if (bsi->lowest_ppage > first_ppage_reported)
10689 bsi->lowest_ppage = first_ppage_reported;
10690 if (bsi->highest_ppage < (next_ppage - 1))
10691 bsi->highest_ppage = next_ppage - 1;
10693 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10696 bsi->nr_extents += ret;
10697 bsi->nr_pages += nr_pages;
10701 static void btrfs_swap_deactivate(struct file *file)
10703 struct inode *inode = file_inode(file);
10705 btrfs_free_swapfile_pins(inode);
10706 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10709 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10712 struct inode *inode = file_inode(file);
10713 struct btrfs_root *root = BTRFS_I(inode)->root;
10714 struct btrfs_fs_info *fs_info = root->fs_info;
10715 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10716 struct extent_state *cached_state = NULL;
10717 struct extent_map *em = NULL;
10718 struct btrfs_device *device = NULL;
10719 struct btrfs_swap_info bsi = {
10720 .lowest_ppage = (sector_t)-1ULL,
10727 * If the swap file was just created, make sure delalloc is done. If the
10728 * file changes again after this, the user is doing something stupid and
10729 * we don't really care.
10731 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10736 * The inode is locked, so these flags won't change after we check them.
10738 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10739 btrfs_warn(fs_info, "swapfile must not be compressed");
10742 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10743 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10746 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10747 btrfs_warn(fs_info, "swapfile must not be checksummed");
10752 * Balance or device remove/replace/resize can move stuff around from
10753 * under us. The exclop protection makes sure they aren't running/won't
10754 * run concurrently while we are mapping the swap extents, and
10755 * fs_info->swapfile_pins prevents them from running while the swap
10756 * file is active and moving the extents. Note that this also prevents
10757 * a concurrent device add which isn't actually necessary, but it's not
10758 * really worth the trouble to allow it.
10760 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10761 btrfs_warn(fs_info,
10762 "cannot activate swapfile while exclusive operation is running");
10767 * Prevent snapshot creation while we are activating the swap file.
10768 * We do not want to race with snapshot creation. If snapshot creation
10769 * already started before we bumped nr_swapfiles from 0 to 1 and
10770 * completes before the first write into the swap file after it is
10771 * activated, than that write would fallback to COW.
10773 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10774 btrfs_exclop_finish(fs_info);
10775 btrfs_warn(fs_info,
10776 "cannot activate swapfile because snapshot creation is in progress");
10780 * Snapshots can create extents which require COW even if NODATACOW is
10781 * set. We use this counter to prevent snapshots. We must increment it
10782 * before walking the extents because we don't want a concurrent
10783 * snapshot to run after we've already checked the extents.
10785 * It is possible that subvolume is marked for deletion but still not
10786 * removed yet. To prevent this race, we check the root status before
10787 * activating the swapfile.
10789 spin_lock(&root->root_item_lock);
10790 if (btrfs_root_dead(root)) {
10791 spin_unlock(&root->root_item_lock);
10793 btrfs_exclop_finish(fs_info);
10794 btrfs_warn(fs_info,
10795 "cannot activate swapfile because subvolume %llu is being deleted",
10796 root->root_key.objectid);
10799 atomic_inc(&root->nr_swapfiles);
10800 spin_unlock(&root->root_item_lock);
10802 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10804 lock_extent(io_tree, 0, isize - 1, &cached_state);
10806 while (start < isize) {
10807 u64 logical_block_start, physical_block_start;
10808 struct btrfs_block_group *bg;
10809 u64 len = isize - start;
10811 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10817 if (em->block_start == EXTENT_MAP_HOLE) {
10818 btrfs_warn(fs_info, "swapfile must not have holes");
10822 if (em->block_start == EXTENT_MAP_INLINE) {
10824 * It's unlikely we'll ever actually find ourselves
10825 * here, as a file small enough to fit inline won't be
10826 * big enough to store more than the swap header, but in
10827 * case something changes in the future, let's catch it
10828 * here rather than later.
10830 btrfs_warn(fs_info, "swapfile must not be inline");
10834 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10835 btrfs_warn(fs_info, "swapfile must not be compressed");
10840 logical_block_start = em->block_start + (start - em->start);
10841 len = min(len, em->len - (start - em->start));
10842 free_extent_map(em);
10845 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10851 btrfs_warn(fs_info,
10852 "swapfile must not be copy-on-write");
10857 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10863 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10864 btrfs_warn(fs_info,
10865 "swapfile must have single data profile");
10870 if (device == NULL) {
10871 device = em->map_lookup->stripes[0].dev;
10872 ret = btrfs_add_swapfile_pin(inode, device, false);
10877 } else if (device != em->map_lookup->stripes[0].dev) {
10878 btrfs_warn(fs_info, "swapfile must be on one device");
10883 physical_block_start = (em->map_lookup->stripes[0].physical +
10884 (logical_block_start - em->start));
10885 len = min(len, em->len - (logical_block_start - em->start));
10886 free_extent_map(em);
10889 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10891 btrfs_warn(fs_info,
10892 "could not find block group containing swapfile");
10897 if (!btrfs_inc_block_group_swap_extents(bg)) {
10898 btrfs_warn(fs_info,
10899 "block group for swapfile at %llu is read-only%s",
10901 atomic_read(&fs_info->scrubs_running) ?
10902 " (scrub running)" : "");
10903 btrfs_put_block_group(bg);
10908 ret = btrfs_add_swapfile_pin(inode, bg, true);
10910 btrfs_put_block_group(bg);
10917 if (bsi.block_len &&
10918 bsi.block_start + bsi.block_len == physical_block_start) {
10919 bsi.block_len += len;
10921 if (bsi.block_len) {
10922 ret = btrfs_add_swap_extent(sis, &bsi);
10927 bsi.block_start = physical_block_start;
10928 bsi.block_len = len;
10935 ret = btrfs_add_swap_extent(sis, &bsi);
10938 if (!IS_ERR_OR_NULL(em))
10939 free_extent_map(em);
10941 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10944 btrfs_swap_deactivate(file);
10946 btrfs_drew_write_unlock(&root->snapshot_lock);
10948 btrfs_exclop_finish(fs_info);
10954 sis->bdev = device->bdev;
10955 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10956 sis->max = bsi.nr_pages;
10957 sis->pages = bsi.nr_pages - 1;
10958 sis->highest_bit = bsi.nr_pages - 1;
10959 return bsi.nr_extents;
10962 static void btrfs_swap_deactivate(struct file *file)
10966 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10969 return -EOPNOTSUPP;
10974 * Update the number of bytes used in the VFS' inode. When we replace extents in
10975 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10976 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10977 * always get a correct value.
10979 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10980 const u64 add_bytes,
10981 const u64 del_bytes)
10983 if (add_bytes == del_bytes)
10986 spin_lock(&inode->lock);
10988 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10990 inode_add_bytes(&inode->vfs_inode, add_bytes);
10991 spin_unlock(&inode->lock);
10995 * Verify that there are no ordered extents for a given file range.
10997 * @inode: The target inode.
10998 * @start: Start offset of the file range, should be sector size aligned.
10999 * @end: End offset (inclusive) of the file range, its value +1 should be
11000 * sector size aligned.
11002 * This should typically be used for cases where we locked an inode's VFS lock in
11003 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
11004 * we have flushed all delalloc in the range, we have waited for all ordered
11005 * extents in the range to complete and finally we have locked the file range in
11006 * the inode's io_tree.
11008 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11010 struct btrfs_root *root = inode->root;
11011 struct btrfs_ordered_extent *ordered;
11013 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11016 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11018 btrfs_err(root->fs_info,
11019 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11020 start, end, btrfs_ino(inode), root->root_key.objectid,
11021 ordered->file_offset,
11022 ordered->file_offset + ordered->num_bytes - 1);
11023 btrfs_put_ordered_extent(ordered);
11026 ASSERT(ordered == NULL);
11029 static const struct inode_operations btrfs_dir_inode_operations = {
11030 .getattr = btrfs_getattr,
11031 .lookup = btrfs_lookup,
11032 .create = btrfs_create,
11033 .unlink = btrfs_unlink,
11034 .link = btrfs_link,
11035 .mkdir = btrfs_mkdir,
11036 .rmdir = btrfs_rmdir,
11037 .rename = btrfs_rename2,
11038 .symlink = btrfs_symlink,
11039 .setattr = btrfs_setattr,
11040 .mknod = btrfs_mknod,
11041 .listxattr = btrfs_listxattr,
11042 .permission = btrfs_permission,
11043 .get_inode_acl = btrfs_get_acl,
11044 .set_acl = btrfs_set_acl,
11045 .update_time = btrfs_update_time,
11046 .tmpfile = btrfs_tmpfile,
11047 .fileattr_get = btrfs_fileattr_get,
11048 .fileattr_set = btrfs_fileattr_set,
11051 static const struct file_operations btrfs_dir_file_operations = {
11052 .llseek = generic_file_llseek,
11053 .read = generic_read_dir,
11054 .iterate_shared = btrfs_real_readdir,
11055 .open = btrfs_opendir,
11056 .unlocked_ioctl = btrfs_ioctl,
11057 #ifdef CONFIG_COMPAT
11058 .compat_ioctl = btrfs_compat_ioctl,
11060 .release = btrfs_release_file,
11061 .fsync = btrfs_sync_file,
11065 * btrfs doesn't support the bmap operation because swapfiles
11066 * use bmap to make a mapping of extents in the file. They assume
11067 * these extents won't change over the life of the file and they
11068 * use the bmap result to do IO directly to the drive.
11070 * the btrfs bmap call would return logical addresses that aren't
11071 * suitable for IO and they also will change frequently as COW
11072 * operations happen. So, swapfile + btrfs == corruption.
11074 * For now we're avoiding this by dropping bmap.
11076 static const struct address_space_operations btrfs_aops = {
11077 .read_folio = btrfs_read_folio,
11078 .writepages = btrfs_writepages,
11079 .readahead = btrfs_readahead,
11080 .invalidate_folio = btrfs_invalidate_folio,
11081 .release_folio = btrfs_release_folio,
11082 .migrate_folio = btrfs_migrate_folio,
11083 .dirty_folio = filemap_dirty_folio,
11084 .error_remove_page = generic_error_remove_page,
11085 .swap_activate = btrfs_swap_activate,
11086 .swap_deactivate = btrfs_swap_deactivate,
11089 static const struct inode_operations btrfs_file_inode_operations = {
11090 .getattr = btrfs_getattr,
11091 .setattr = btrfs_setattr,
11092 .listxattr = btrfs_listxattr,
11093 .permission = btrfs_permission,
11094 .fiemap = btrfs_fiemap,
11095 .get_inode_acl = btrfs_get_acl,
11096 .set_acl = btrfs_set_acl,
11097 .update_time = btrfs_update_time,
11098 .fileattr_get = btrfs_fileattr_get,
11099 .fileattr_set = btrfs_fileattr_set,
11101 static const struct inode_operations btrfs_special_inode_operations = {
11102 .getattr = btrfs_getattr,
11103 .setattr = btrfs_setattr,
11104 .permission = btrfs_permission,
11105 .listxattr = btrfs_listxattr,
11106 .get_inode_acl = btrfs_get_acl,
11107 .set_acl = btrfs_set_acl,
11108 .update_time = btrfs_update_time,
11110 static const struct inode_operations btrfs_symlink_inode_operations = {
11111 .get_link = page_get_link,
11112 .getattr = btrfs_getattr,
11113 .setattr = btrfs_setattr,
11114 .permission = btrfs_permission,
11115 .listxattr = btrfs_listxattr,
11116 .update_time = btrfs_update_time,
11119 const struct dentry_operations btrfs_dentry_operations = {
11120 .d_delete = btrfs_dentry_delete,