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 void submit_one_async_extent(struct async_chunk *async_chunk,
1183 struct async_extent *async_extent,
1186 struct btrfs_inode *inode = async_chunk->inode;
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);
1293 if (async_chunk->blkcg_css)
1294 kthread_associate_blkcg(NULL);
1295 btrfs_debug(fs_info,
1296 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1297 root->root_key.objectid, btrfs_ino(inode), start,
1298 async_extent->ram_size, ret);
1299 kfree(async_extent);
1303 * Phase two of compressed writeback. This is the ordered portion of the code,
1304 * which only gets called in the order the work was queued. We walk all the
1305 * async extents created by compress_file_range and send them down to the disk.
1307 static noinline void submit_compressed_extents(struct async_chunk *async_chunk)
1309 struct async_extent *async_extent;
1312 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);
1317 submit_one_async_extent(async_chunk, async_extent, &alloc_hint);
1321 static u64 get_extent_allocation_hint(struct btrfs_inode *inode, u64 start,
1324 struct extent_map_tree *em_tree = &inode->extent_tree;
1325 struct extent_map *em;
1328 read_lock(&em_tree->lock);
1329 em = search_extent_mapping(em_tree, start, num_bytes);
1332 * if block start isn't an actual block number then find the
1333 * first block in this inode and use that as a hint. If that
1334 * block is also bogus then just don't worry about it.
1336 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1337 free_extent_map(em);
1338 em = search_extent_mapping(em_tree, 0, 0);
1339 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
1340 alloc_hint = em->block_start;
1342 free_extent_map(em);
1344 alloc_hint = em->block_start;
1345 free_extent_map(em);
1348 read_unlock(&em_tree->lock);
1354 * when extent_io.c finds a delayed allocation range in the file,
1355 * the call backs end up in this code. The basic idea is to
1356 * allocate extents on disk for the range, and create ordered data structs
1357 * in ram to track those extents.
1359 * locked_page is the page that writepage had locked already. We use
1360 * it to make sure we don't do extra locks or unlocks.
1362 * When this function fails, it unlocks all pages except @locked_page.
1364 * When this function successfully creates an inline extent, it sets page_started
1365 * to 1 and unlocks all pages including locked_page and starts I/O on them.
1366 * (In reality inline extents are limited to a single page, so locked_page is
1367 * the only page handled anyway).
1369 * When this function succeed and creates a normal extent, the page locking
1370 * status depends on the passed in flags:
1372 * - If @keep_locked is set, all pages are kept locked.
1373 * - Else all pages except for @locked_page are unlocked.
1375 * When a failure happens in the second or later iteration of the
1376 * while-loop, the ordered extents created in previous iterations are kept
1377 * intact. So, the caller must clean them up by calling
1378 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1381 static noinline int cow_file_range(struct btrfs_inode *inode,
1382 struct page *locked_page,
1383 u64 start, u64 end, int *page_started,
1384 unsigned long *nr_written, u64 *done_offset,
1385 bool keep_locked, bool no_inline)
1387 struct btrfs_root *root = inode->root;
1388 struct btrfs_fs_info *fs_info = root->fs_info;
1390 u64 orig_start = start;
1392 unsigned long ram_size;
1393 u64 cur_alloc_size = 0;
1395 u64 blocksize = fs_info->sectorsize;
1396 struct btrfs_key ins;
1397 struct extent_map *em;
1398 unsigned clear_bits;
1399 unsigned long page_ops;
1400 bool extent_reserved = false;
1403 if (btrfs_is_free_space_inode(inode)) {
1408 num_bytes = ALIGN(end - start + 1, blocksize);
1409 num_bytes = max(blocksize, num_bytes);
1410 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
1412 inode_should_defrag(inode, start, end, num_bytes, SZ_64K);
1415 * Due to the page size limit, for subpage we can only trigger the
1416 * writeback for the dirty sectors of page, that means data writeback
1417 * is doing more writeback than what we want.
1419 * This is especially unexpected for some call sites like fallocate,
1420 * where we only increase i_size after everything is done.
1421 * This means we can trigger inline extent even if we didn't want to.
1422 * So here we skip inline extent creation completely.
1424 if (start == 0 && fs_info->sectorsize == PAGE_SIZE && !no_inline) {
1425 u64 actual_end = min_t(u64, i_size_read(&inode->vfs_inode),
1428 /* lets try to make an inline extent */
1429 ret = cow_file_range_inline(inode, actual_end, 0,
1430 BTRFS_COMPRESS_NONE, NULL, false);
1433 * We use DO_ACCOUNTING here because we need the
1434 * delalloc_release_metadata to be run _after_ we drop
1435 * our outstanding extent for clearing delalloc for this
1438 extent_clear_unlock_delalloc(inode, start, end,
1440 EXTENT_LOCKED | EXTENT_DELALLOC |
1441 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
1442 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1443 PAGE_START_WRITEBACK | PAGE_END_WRITEBACK);
1444 *nr_written = *nr_written +
1445 (end - start + PAGE_SIZE) / PAGE_SIZE;
1448 * locked_page is locked by the caller of
1449 * writepage_delalloc(), not locked by
1450 * __process_pages_contig().
1452 * We can't let __process_pages_contig() to unlock it,
1453 * as it doesn't have any subpage::writers recorded.
1455 * Here we manually unlock the page, since the caller
1456 * can't use page_started to determine if it's an
1457 * inline extent or a compressed extent.
1459 unlock_page(locked_page);
1461 } else if (ret < 0) {
1466 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
1469 * Relocation relies on the relocated extents to have exactly the same
1470 * size as the original extents. Normally writeback for relocation data
1471 * extents follows a NOCOW path because relocation preallocates the
1472 * extents. However, due to an operation such as scrub turning a block
1473 * group to RO mode, it may fallback to COW mode, so we must make sure
1474 * an extent allocated during COW has exactly the requested size and can
1475 * not be split into smaller extents, otherwise relocation breaks and
1476 * fails during the stage where it updates the bytenr of file extent
1479 if (btrfs_is_data_reloc_root(root))
1480 min_alloc_size = num_bytes;
1482 min_alloc_size = fs_info->sectorsize;
1484 while (num_bytes > 0) {
1485 struct btrfs_ordered_extent *ordered;
1487 cur_alloc_size = num_bytes;
1488 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1489 min_alloc_size, 0, alloc_hint,
1493 cur_alloc_size = ins.offset;
1494 extent_reserved = true;
1496 ram_size = ins.offset;
1497 em = create_io_em(inode, start, ins.offset, /* len */
1498 start, /* orig_start */
1499 ins.objectid, /* block_start */
1500 ins.offset, /* block_len */
1501 ins.offset, /* orig_block_len */
1502 ram_size, /* ram_bytes */
1503 BTRFS_COMPRESS_NONE, /* compress_type */
1504 BTRFS_ORDERED_REGULAR /* type */);
1509 free_extent_map(em);
1511 ordered = btrfs_alloc_ordered_extent(inode, start, ram_size,
1512 ram_size, ins.objectid, cur_alloc_size,
1513 0, 1 << BTRFS_ORDERED_REGULAR,
1514 BTRFS_COMPRESS_NONE);
1515 if (IS_ERR(ordered)) {
1516 ret = PTR_ERR(ordered);
1517 goto out_drop_extent_cache;
1520 if (btrfs_is_data_reloc_root(root)) {
1521 ret = btrfs_reloc_clone_csums(ordered);
1524 * Only drop cache here, and process as normal.
1526 * We must not allow extent_clear_unlock_delalloc()
1527 * at out_unlock label to free meta of this ordered
1528 * extent, as its meta should be freed by
1529 * btrfs_finish_ordered_io().
1531 * So we must continue until @start is increased to
1532 * skip current ordered extent.
1535 btrfs_drop_extent_map_range(inode, start,
1536 start + ram_size - 1,
1539 btrfs_put_ordered_extent(ordered);
1541 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1544 * We're not doing compressed IO, don't unlock the first page
1545 * (which the caller expects to stay locked), don't clear any
1546 * dirty bits and don't set any writeback bits
1548 * Do set the Ordered (Private2) bit so we know this page was
1549 * properly setup for writepage.
1551 page_ops = (keep_locked ? 0 : PAGE_UNLOCK);
1552 page_ops |= PAGE_SET_ORDERED;
1554 extent_clear_unlock_delalloc(inode, start, start + ram_size - 1,
1556 EXTENT_LOCKED | EXTENT_DELALLOC,
1558 if (num_bytes < cur_alloc_size)
1561 num_bytes -= cur_alloc_size;
1562 alloc_hint = ins.objectid + ins.offset;
1563 start += cur_alloc_size;
1564 extent_reserved = false;
1567 * btrfs_reloc_clone_csums() error, since start is increased
1568 * extent_clear_unlock_delalloc() at out_unlock label won't
1569 * free metadata of current ordered extent, we're OK to exit.
1577 out_drop_extent_cache:
1578 btrfs_drop_extent_map_range(inode, start, start + ram_size - 1, false);
1580 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1581 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1584 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1585 * caller to write out the successfully allocated region and retry.
1587 if (done_offset && ret == -EAGAIN) {
1588 if (orig_start < start)
1589 *done_offset = start - 1;
1591 *done_offset = start;
1593 } else if (ret == -EAGAIN) {
1594 /* Convert to -ENOSPC since the caller cannot retry. */
1599 * Now, we have three regions to clean up:
1601 * |-------(1)----|---(2)---|-------------(3)----------|
1602 * `- orig_start `- start `- start + cur_alloc_size `- end
1604 * We process each region below.
1607 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1608 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1609 page_ops = PAGE_UNLOCK | PAGE_START_WRITEBACK | PAGE_END_WRITEBACK;
1612 * For the range (1). We have already instantiated the ordered extents
1613 * for this region. They are cleaned up by
1614 * btrfs_cleanup_ordered_extents() in e.g,
1615 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1616 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1617 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1620 * However, in case of @keep_locked, we still need to unlock the pages
1621 * (except @locked_page) to ensure all the pages are unlocked.
1623 if (keep_locked && orig_start < start) {
1625 mapping_set_error(inode->vfs_inode.i_mapping, ret);
1626 extent_clear_unlock_delalloc(inode, orig_start, start - 1,
1627 locked_page, 0, page_ops);
1631 * For the range (2). If we reserved an extent for our delalloc range
1632 * (or a subrange) and failed to create the respective ordered extent,
1633 * then it means that when we reserved the extent we decremented the
1634 * extent's size from the data space_info's bytes_may_use counter and
1635 * incremented the space_info's bytes_reserved counter by the same
1636 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1637 * to decrement again the data space_info's bytes_may_use counter,
1638 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1640 if (extent_reserved) {
1641 extent_clear_unlock_delalloc(inode, start,
1642 start + cur_alloc_size - 1,
1646 start += cur_alloc_size;
1650 * For the range (3). We never touched the region. In addition to the
1651 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1652 * space_info's bytes_may_use counter, reserved in
1653 * btrfs_check_data_free_space().
1656 clear_bits |= EXTENT_CLEAR_DATA_RESV;
1657 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1658 clear_bits, page_ops);
1664 * work queue call back to started compression on a file and pages
1666 static noinline void async_cow_start(struct btrfs_work *work)
1668 struct async_chunk *async_chunk;
1669 int compressed_extents;
1671 async_chunk = container_of(work, struct async_chunk, work);
1673 compressed_extents = compress_file_range(async_chunk);
1674 if (compressed_extents == 0) {
1675 btrfs_add_delayed_iput(async_chunk->inode);
1676 async_chunk->inode = NULL;
1681 * work queue call back to submit previously compressed pages
1683 static noinline void async_cow_submit(struct btrfs_work *work)
1685 struct async_chunk *async_chunk = container_of(work, struct async_chunk,
1687 struct btrfs_fs_info *fs_info = btrfs_work_owner(work);
1688 unsigned long nr_pages;
1690 nr_pages = (async_chunk->end - async_chunk->start + PAGE_SIZE) >>
1694 * ->inode could be NULL if async_chunk_start has failed to compress,
1695 * in which case we don't have anything to submit, yet we need to
1696 * always adjust ->async_delalloc_pages as its paired with the init
1697 * happening in run_delalloc_compressed
1699 if (async_chunk->inode)
1700 submit_compressed_extents(async_chunk);
1702 /* atomic_sub_return implies a barrier */
1703 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1705 cond_wake_up_nomb(&fs_info->async_submit_wait);
1708 static noinline void async_cow_free(struct btrfs_work *work)
1710 struct async_chunk *async_chunk;
1711 struct async_cow *async_cow;
1713 async_chunk = container_of(work, struct async_chunk, work);
1714 if (async_chunk->inode)
1715 btrfs_add_delayed_iput(async_chunk->inode);
1716 if (async_chunk->blkcg_css)
1717 css_put(async_chunk->blkcg_css);
1719 async_cow = async_chunk->async_cow;
1720 if (atomic_dec_and_test(&async_cow->num_chunks))
1724 static bool run_delalloc_compressed(struct btrfs_inode *inode,
1725 struct writeback_control *wbc,
1726 struct page *locked_page,
1727 u64 start, u64 end, int *page_started,
1728 unsigned long *nr_written)
1730 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1731 struct cgroup_subsys_state *blkcg_css = wbc_blkcg_css(wbc);
1732 struct async_cow *ctx;
1733 struct async_chunk *async_chunk;
1734 unsigned long nr_pages;
1735 u64 num_chunks = DIV_ROUND_UP(end - start, SZ_512K);
1738 const blk_opf_t write_flags = wbc_to_write_flags(wbc);
1740 nofs_flag = memalloc_nofs_save();
1741 ctx = kvmalloc(struct_size(ctx, chunks, num_chunks), GFP_KERNEL);
1742 memalloc_nofs_restore(nofs_flag);
1746 unlock_extent(&inode->io_tree, start, end, NULL);
1747 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, &inode->runtime_flags);
1749 async_chunk = ctx->chunks;
1750 atomic_set(&ctx->num_chunks, num_chunks);
1752 for (i = 0; i < num_chunks; i++) {
1753 u64 cur_end = min(end, start + SZ_512K - 1);
1756 * igrab is called higher up in the call chain, take only the
1757 * lightweight reference for the callback lifetime
1759 ihold(&inode->vfs_inode);
1760 async_chunk[i].async_cow = ctx;
1761 async_chunk[i].inode = inode;
1762 async_chunk[i].start = start;
1763 async_chunk[i].end = cur_end;
1764 async_chunk[i].write_flags = write_flags;
1765 INIT_LIST_HEAD(&async_chunk[i].extents);
1768 * The locked_page comes all the way from writepage and its
1769 * the original page we were actually given. As we spread
1770 * this large delalloc region across multiple async_chunk
1771 * structs, only the first struct needs a pointer to locked_page
1773 * This way we don't need racey decisions about who is supposed
1778 * Depending on the compressibility, the pages might or
1779 * might not go through async. We want all of them to
1780 * be accounted against wbc once. Let's do it here
1781 * before the paths diverge. wbc accounting is used
1782 * only for foreign writeback detection and doesn't
1783 * need full accuracy. Just account the whole thing
1784 * against the first page.
1786 wbc_account_cgroup_owner(wbc, locked_page,
1788 async_chunk[i].locked_page = locked_page;
1791 async_chunk[i].locked_page = NULL;
1794 if (blkcg_css != blkcg_root_css) {
1796 async_chunk[i].blkcg_css = blkcg_css;
1797 async_chunk[i].write_flags |= REQ_BTRFS_CGROUP_PUNT;
1799 async_chunk[i].blkcg_css = NULL;
1802 btrfs_init_work(&async_chunk[i].work, async_cow_start,
1803 async_cow_submit, async_cow_free);
1805 nr_pages = DIV_ROUND_UP(cur_end - start, PAGE_SIZE);
1806 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1808 btrfs_queue_work(fs_info->delalloc_workers, &async_chunk[i].work);
1810 *nr_written += nr_pages;
1811 start = cur_end + 1;
1817 static noinline int run_delalloc_zoned(struct btrfs_inode *inode,
1818 struct page *locked_page, u64 start,
1819 u64 end, int *page_started,
1820 unsigned long *nr_written,
1821 struct writeback_control *wbc)
1823 u64 done_offset = end;
1825 bool locked_page_done = false;
1827 while (start <= end) {
1828 ret = cow_file_range(inode, locked_page, start, end, page_started,
1829 nr_written, &done_offset, true, false);
1830 if (ret && ret != -EAGAIN)
1833 if (*page_started) {
1841 if (done_offset == start) {
1842 wait_on_bit_io(&inode->root->fs_info->flags,
1843 BTRFS_FS_NEED_ZONE_FINISH,
1844 TASK_UNINTERRUPTIBLE);
1848 if (!locked_page_done) {
1849 __set_page_dirty_nobuffers(locked_page);
1850 account_page_redirty(locked_page);
1852 locked_page_done = true;
1853 extent_write_locked_range(&inode->vfs_inode, start, done_offset,
1855 start = done_offset + 1;
1863 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1864 u64 bytenr, u64 num_bytes, bool nowait)
1866 struct btrfs_root *csum_root = btrfs_csum_root(fs_info, bytenr);
1867 struct btrfs_ordered_sum *sums;
1871 ret = btrfs_lookup_csums_list(csum_root, bytenr, bytenr + num_bytes - 1,
1873 if (ret == 0 && list_empty(&list))
1876 while (!list_empty(&list)) {
1877 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1878 list_del(&sums->list);
1886 static int fallback_to_cow(struct btrfs_inode *inode, struct page *locked_page,
1887 const u64 start, const u64 end)
1889 const bool is_space_ino = btrfs_is_free_space_inode(inode);
1890 const bool is_reloc_ino = btrfs_is_data_reloc_root(inode->root);
1891 const u64 range_bytes = end + 1 - start;
1892 struct extent_io_tree *io_tree = &inode->io_tree;
1893 int page_started = 0;
1894 unsigned long nr_written;
1895 u64 range_start = start;
1900 * If EXTENT_NORESERVE is set it means that when the buffered write was
1901 * made we had not enough available data space and therefore we did not
1902 * reserve data space for it, since we though we could do NOCOW for the
1903 * respective file range (either there is prealloc extent or the inode
1904 * has the NOCOW bit set).
1906 * However when we need to fallback to COW mode (because for example the
1907 * block group for the corresponding extent was turned to RO mode by a
1908 * scrub or relocation) we need to do the following:
1910 * 1) We increment the bytes_may_use counter of the data space info.
1911 * If COW succeeds, it allocates a new data extent and after doing
1912 * that it decrements the space info's bytes_may_use counter and
1913 * increments its bytes_reserved counter by the same amount (we do
1914 * this at btrfs_add_reserved_bytes()). So we need to increment the
1915 * bytes_may_use counter to compensate (when space is reserved at
1916 * buffered write time, the bytes_may_use counter is incremented);
1918 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1919 * that if the COW path fails for any reason, it decrements (through
1920 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1921 * data space info, which we incremented in the step above.
1923 * If we need to fallback to cow and the inode corresponds to a free
1924 * space cache inode or an inode of the data relocation tree, we must
1925 * also increment bytes_may_use of the data space_info for the same
1926 * reason. Space caches and relocated data extents always get a prealloc
1927 * extent for them, however scrub or balance may have set the block
1928 * group that contains that extent to RO mode and therefore force COW
1929 * when starting writeback.
1931 count = count_range_bits(io_tree, &range_start, end, range_bytes,
1932 EXTENT_NORESERVE, 0, NULL);
1933 if (count > 0 || is_space_ino || is_reloc_ino) {
1935 struct btrfs_fs_info *fs_info = inode->root->fs_info;
1936 struct btrfs_space_info *sinfo = fs_info->data_sinfo;
1938 if (is_space_ino || is_reloc_ino)
1939 bytes = range_bytes;
1941 spin_lock(&sinfo->lock);
1942 btrfs_space_info_update_bytes_may_use(fs_info, sinfo, bytes);
1943 spin_unlock(&sinfo->lock);
1946 clear_extent_bit(io_tree, start, end, EXTENT_NORESERVE,
1951 * Don't try to create inline extents, as a mix of inline extent that
1952 * is written out and unlocked directly and a normal NOCOW extent
1955 ret = cow_file_range(inode, locked_page, start, end, &page_started,
1956 &nr_written, NULL, false, true);
1957 ASSERT(!page_started);
1961 struct can_nocow_file_extent_args {
1964 /* Start file offset of the range we want to NOCOW. */
1966 /* End file offset (inclusive) of the range we want to NOCOW. */
1968 bool writeback_path;
1971 * Free the path passed to can_nocow_file_extent() once it's not needed
1976 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1981 /* Number of bytes that can be written to in NOCOW mode. */
1986 * Check if we can NOCOW the file extent that the path points to.
1987 * This function may return with the path released, so the caller should check
1988 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
1990 * Returns: < 0 on error
1991 * 0 if we can not NOCOW
1994 static int can_nocow_file_extent(struct btrfs_path *path,
1995 struct btrfs_key *key,
1996 struct btrfs_inode *inode,
1997 struct can_nocow_file_extent_args *args)
1999 const bool is_freespace_inode = btrfs_is_free_space_inode(inode);
2000 struct extent_buffer *leaf = path->nodes[0];
2001 struct btrfs_root *root = inode->root;
2002 struct btrfs_file_extent_item *fi;
2007 bool nowait = path->nowait;
2009 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
2010 extent_type = btrfs_file_extent_type(leaf, fi);
2012 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
2015 /* Can't access these fields unless we know it's not an inline extent. */
2016 args->disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2017 args->disk_num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
2018 args->extent_offset = btrfs_file_extent_offset(leaf, fi);
2020 if (!(inode->flags & BTRFS_INODE_NODATACOW) &&
2021 extent_type == BTRFS_FILE_EXTENT_REG)
2025 * If the extent was created before the generation where the last snapshot
2026 * for its subvolume was created, then this implies the extent is shared,
2027 * hence we must COW.
2029 if (!args->strict &&
2030 btrfs_file_extent_generation(leaf, fi) <=
2031 btrfs_root_last_snapshot(&root->root_item))
2034 /* An explicit hole, must COW. */
2035 if (args->disk_bytenr == 0)
2038 /* Compressed/encrypted/encoded extents must be COWed. */
2039 if (btrfs_file_extent_compression(leaf, fi) ||
2040 btrfs_file_extent_encryption(leaf, fi) ||
2041 btrfs_file_extent_other_encoding(leaf, fi))
2044 extent_end = btrfs_file_extent_end(path);
2047 * The following checks can be expensive, as they need to take other
2048 * locks and do btree or rbtree searches, so release the path to avoid
2049 * blocking other tasks for too long.
2051 btrfs_release_path(path);
2053 ret = btrfs_cross_ref_exist(root, btrfs_ino(inode),
2054 key->offset - args->extent_offset,
2055 args->disk_bytenr, args->strict, path);
2056 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2060 if (args->free_path) {
2062 * We don't need the path anymore, plus through the
2063 * csum_exist_in_range() call below we will end up allocating
2064 * another path. So free the path to avoid unnecessary extra
2067 btrfs_free_path(path);
2071 /* If there are pending snapshots for this root, we must COW. */
2072 if (args->writeback_path && !is_freespace_inode &&
2073 atomic_read(&root->snapshot_force_cow))
2076 args->disk_bytenr += args->extent_offset;
2077 args->disk_bytenr += args->start - key->offset;
2078 args->num_bytes = min(args->end + 1, extent_end) - args->start;
2081 * Force COW if csums exist in the range. This ensures that csums for a
2082 * given extent are either valid or do not exist.
2084 ret = csum_exist_in_range(root->fs_info, args->disk_bytenr, args->num_bytes,
2086 WARN_ON_ONCE(ret > 0 && is_freespace_inode);
2092 if (args->free_path && path)
2093 btrfs_free_path(path);
2095 return ret < 0 ? ret : can_nocow;
2099 * when nowcow writeback call back. This checks for snapshots or COW copies
2100 * of the extents that exist in the file, and COWs the file as required.
2102 * If no cow copies or snapshots exist, we write directly to the existing
2105 static noinline int run_delalloc_nocow(struct btrfs_inode *inode,
2106 struct page *locked_page,
2107 const u64 start, const u64 end)
2109 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2110 struct btrfs_root *root = inode->root;
2111 struct btrfs_path *path;
2112 u64 cow_start = (u64)-1;
2113 u64 cur_offset = start;
2115 bool check_prev = true;
2116 u64 ino = btrfs_ino(inode);
2117 struct btrfs_block_group *bg;
2119 struct can_nocow_file_extent_args nocow_args = { 0 };
2121 path = btrfs_alloc_path();
2123 extent_clear_unlock_delalloc(inode, start, end, locked_page,
2124 EXTENT_LOCKED | EXTENT_DELALLOC |
2125 EXTENT_DO_ACCOUNTING |
2126 EXTENT_DEFRAG, PAGE_UNLOCK |
2127 PAGE_START_WRITEBACK |
2128 PAGE_END_WRITEBACK);
2132 nocow_args.end = end;
2133 nocow_args.writeback_path = true;
2136 struct btrfs_ordered_extent *ordered;
2137 struct btrfs_key found_key;
2138 struct btrfs_file_extent_item *fi;
2139 struct extent_buffer *leaf;
2148 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
2154 * If there is no extent for our range when doing the initial
2155 * search, then go back to the previous slot as it will be the
2156 * one containing the search offset
2158 if (ret > 0 && path->slots[0] > 0 && check_prev) {
2159 leaf = path->nodes[0];
2160 btrfs_item_key_to_cpu(leaf, &found_key,
2161 path->slots[0] - 1);
2162 if (found_key.objectid == ino &&
2163 found_key.type == BTRFS_EXTENT_DATA_KEY)
2168 /* Go to next leaf if we have exhausted the current one */
2169 leaf = path->nodes[0];
2170 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2171 ret = btrfs_next_leaf(root, path);
2173 if (cow_start != (u64)-1)
2174 cur_offset = cow_start;
2179 leaf = path->nodes[0];
2182 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2184 /* Didn't find anything for our INO */
2185 if (found_key.objectid > ino)
2188 * Keep searching until we find an EXTENT_ITEM or there are no
2189 * more extents for this inode
2191 if (WARN_ON_ONCE(found_key.objectid < ino) ||
2192 found_key.type < BTRFS_EXTENT_DATA_KEY) {
2197 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2198 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
2199 found_key.offset > end)
2203 * If the found extent starts after requested offset, then
2204 * adjust extent_end to be right before this extent begins
2206 if (found_key.offset > cur_offset) {
2207 extent_end = found_key.offset;
2213 * Found extent which begins before our range and potentially
2216 fi = btrfs_item_ptr(leaf, path->slots[0],
2217 struct btrfs_file_extent_item);
2218 extent_type = btrfs_file_extent_type(leaf, fi);
2219 /* If this is triggered then we have a memory corruption. */
2220 ASSERT(extent_type < BTRFS_NR_FILE_EXTENT_TYPES);
2221 if (WARN_ON(extent_type >= BTRFS_NR_FILE_EXTENT_TYPES)) {
2225 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
2226 extent_end = btrfs_file_extent_end(path);
2229 * If the extent we got ends before our current offset, skip to
2232 if (extent_end <= cur_offset) {
2237 nocow_args.start = cur_offset;
2238 ret = can_nocow_file_extent(path, &found_key, inode, &nocow_args);
2240 if (cow_start != (u64)-1)
2241 cur_offset = cow_start;
2243 } else if (ret == 0) {
2248 bg = btrfs_inc_nocow_writers(fs_info, nocow_args.disk_bytenr);
2253 * If nocow is false then record the beginning of the range
2254 * that needs to be COWed
2257 if (cow_start == (u64)-1)
2258 cow_start = cur_offset;
2259 cur_offset = extent_end;
2260 if (cur_offset > end)
2262 if (!path->nodes[0])
2269 * COW range from cow_start to found_key.offset - 1. As the key
2270 * will contain the beginning of the first extent that can be
2271 * NOCOW, following one which needs to be COW'ed
2273 if (cow_start != (u64)-1) {
2274 ret = fallback_to_cow(inode, locked_page,
2275 cow_start, found_key.offset - 1);
2278 cow_start = (u64)-1;
2281 nocow_end = cur_offset + nocow_args.num_bytes - 1;
2282 is_prealloc = extent_type == BTRFS_FILE_EXTENT_PREALLOC;
2284 u64 orig_start = found_key.offset - nocow_args.extent_offset;
2285 struct extent_map *em;
2287 em = create_io_em(inode, cur_offset, nocow_args.num_bytes,
2289 nocow_args.disk_bytenr, /* block_start */
2290 nocow_args.num_bytes, /* block_len */
2291 nocow_args.disk_num_bytes, /* orig_block_len */
2292 ram_bytes, BTRFS_COMPRESS_NONE,
2293 BTRFS_ORDERED_PREALLOC);
2298 free_extent_map(em);
2301 ordered = btrfs_alloc_ordered_extent(inode, cur_offset,
2302 nocow_args.num_bytes, nocow_args.num_bytes,
2303 nocow_args.disk_bytenr, nocow_args.num_bytes, 0,
2305 ? (1 << BTRFS_ORDERED_PREALLOC)
2306 : (1 << BTRFS_ORDERED_NOCOW),
2307 BTRFS_COMPRESS_NONE);
2308 if (IS_ERR(ordered)) {
2310 btrfs_drop_extent_map_range(inode, cur_offset,
2313 ret = PTR_ERR(ordered);
2318 btrfs_dec_nocow_writers(bg);
2322 if (btrfs_is_data_reloc_root(root))
2324 * Error handled later, as we must prevent
2325 * extent_clear_unlock_delalloc() in error handler
2326 * from freeing metadata of created ordered extent.
2328 ret = btrfs_reloc_clone_csums(ordered);
2329 btrfs_put_ordered_extent(ordered);
2331 extent_clear_unlock_delalloc(inode, cur_offset, nocow_end,
2332 locked_page, EXTENT_LOCKED |
2334 EXTENT_CLEAR_DATA_RESV,
2335 PAGE_UNLOCK | PAGE_SET_ORDERED);
2337 cur_offset = extent_end;
2340 * btrfs_reloc_clone_csums() error, now we're OK to call error
2341 * handler, as metadata for created ordered extent will only
2342 * be freed by btrfs_finish_ordered_io().
2346 if (cur_offset > end)
2349 btrfs_release_path(path);
2351 if (cur_offset <= end && cow_start == (u64)-1)
2352 cow_start = cur_offset;
2354 if (cow_start != (u64)-1) {
2356 ret = fallback_to_cow(inode, locked_page, cow_start, end);
2363 btrfs_dec_nocow_writers(bg);
2365 if (ret && cur_offset < end)
2366 extent_clear_unlock_delalloc(inode, cur_offset, end,
2367 locked_page, EXTENT_LOCKED |
2368 EXTENT_DELALLOC | EXTENT_DEFRAG |
2369 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
2370 PAGE_START_WRITEBACK |
2371 PAGE_END_WRITEBACK);
2372 btrfs_free_path(path);
2376 static bool should_nocow(struct btrfs_inode *inode, u64 start, u64 end)
2378 if (inode->flags & (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)) {
2379 if (inode->defrag_bytes &&
2380 test_range_bit(&inode->io_tree, start, end, EXTENT_DEFRAG,
2389 * Function to process delayed allocation (create CoW) for ranges which are
2390 * being touched for the first time.
2392 int btrfs_run_delalloc_range(struct btrfs_inode *inode, struct page *locked_page,
2393 u64 start, u64 end, int *page_started, unsigned long *nr_written,
2394 struct writeback_control *wbc)
2397 const bool zoned = btrfs_is_zoned(inode->root->fs_info);
2400 * The range must cover part of the @locked_page, or the returned
2401 * @page_started can confuse the caller.
2403 ASSERT(!(end <= page_offset(locked_page) ||
2404 start >= page_offset(locked_page) + PAGE_SIZE));
2406 if (should_nocow(inode, start, end)) {
2408 * Normally on a zoned device we're only doing COW writes, but
2409 * in case of relocation on a zoned filesystem we have taken
2410 * precaution, that we're only writing sequentially. It's safe
2411 * to use run_delalloc_nocow() here, like for regular
2412 * preallocated inodes.
2414 ASSERT(!zoned || btrfs_is_data_reloc_root(inode->root));
2415 ret = run_delalloc_nocow(inode, locked_page, start, end);
2419 if (btrfs_inode_can_compress(inode) &&
2420 inode_need_compress(inode, start, end) &&
2421 run_delalloc_compressed(inode, wbc, locked_page, start,
2422 end, page_started, nr_written))
2426 ret = run_delalloc_zoned(inode, locked_page, start, end,
2427 page_started, nr_written, wbc);
2429 ret = cow_file_range(inode, locked_page, start, end,
2430 page_started, nr_written, NULL, false, false);
2435 btrfs_cleanup_ordered_extents(inode, locked_page, start,
2440 void btrfs_split_delalloc_extent(struct btrfs_inode *inode,
2441 struct extent_state *orig, u64 split)
2443 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2446 /* not delalloc, ignore it */
2447 if (!(orig->state & EXTENT_DELALLOC))
2450 size = orig->end - orig->start + 1;
2451 if (size > fs_info->max_extent_size) {
2456 * See the explanation in btrfs_merge_delalloc_extent, the same
2457 * applies here, just in reverse.
2459 new_size = orig->end - split + 1;
2460 num_extents = count_max_extents(fs_info, new_size);
2461 new_size = split - orig->start;
2462 num_extents += count_max_extents(fs_info, new_size);
2463 if (count_max_extents(fs_info, size) >= num_extents)
2467 spin_lock(&inode->lock);
2468 btrfs_mod_outstanding_extents(inode, 1);
2469 spin_unlock(&inode->lock);
2473 * Handle merged delayed allocation extents so we can keep track of new extents
2474 * that are just merged onto old extents, such as when we are doing sequential
2475 * writes, so we can properly account for the metadata space we'll need.
2477 void btrfs_merge_delalloc_extent(struct btrfs_inode *inode, struct extent_state *new,
2478 struct extent_state *other)
2480 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2481 u64 new_size, old_size;
2484 /* not delalloc, ignore it */
2485 if (!(other->state & EXTENT_DELALLOC))
2488 if (new->start > other->start)
2489 new_size = new->end - other->start + 1;
2491 new_size = other->end - new->start + 1;
2493 /* we're not bigger than the max, unreserve the space and go */
2494 if (new_size <= fs_info->max_extent_size) {
2495 spin_lock(&inode->lock);
2496 btrfs_mod_outstanding_extents(inode, -1);
2497 spin_unlock(&inode->lock);
2502 * We have to add up either side to figure out how many extents were
2503 * accounted for before we merged into one big extent. If the number of
2504 * extents we accounted for is <= the amount we need for the new range
2505 * then we can return, otherwise drop. Think of it like this
2509 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2510 * need 2 outstanding extents, on one side we have 1 and the other side
2511 * we have 1 so they are == and we can return. But in this case
2513 * [MAX_SIZE+4k][MAX_SIZE+4k]
2515 * Each range on their own accounts for 2 extents, but merged together
2516 * they are only 3 extents worth of accounting, so we need to drop in
2519 old_size = other->end - other->start + 1;
2520 num_extents = count_max_extents(fs_info, old_size);
2521 old_size = new->end - new->start + 1;
2522 num_extents += count_max_extents(fs_info, old_size);
2523 if (count_max_extents(fs_info, new_size) >= num_extents)
2526 spin_lock(&inode->lock);
2527 btrfs_mod_outstanding_extents(inode, -1);
2528 spin_unlock(&inode->lock);
2531 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
2532 struct btrfs_inode *inode)
2534 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2536 spin_lock(&root->delalloc_lock);
2537 if (list_empty(&inode->delalloc_inodes)) {
2538 list_add_tail(&inode->delalloc_inodes, &root->delalloc_inodes);
2539 set_bit(BTRFS_INODE_IN_DELALLOC_LIST, &inode->runtime_flags);
2540 root->nr_delalloc_inodes++;
2541 if (root->nr_delalloc_inodes == 1) {
2542 spin_lock(&fs_info->delalloc_root_lock);
2543 BUG_ON(!list_empty(&root->delalloc_root));
2544 list_add_tail(&root->delalloc_root,
2545 &fs_info->delalloc_roots);
2546 spin_unlock(&fs_info->delalloc_root_lock);
2549 spin_unlock(&root->delalloc_lock);
2552 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
2553 struct btrfs_inode *inode)
2555 struct btrfs_fs_info *fs_info = root->fs_info;
2557 if (!list_empty(&inode->delalloc_inodes)) {
2558 list_del_init(&inode->delalloc_inodes);
2559 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2560 &inode->runtime_flags);
2561 root->nr_delalloc_inodes--;
2562 if (!root->nr_delalloc_inodes) {
2563 ASSERT(list_empty(&root->delalloc_inodes));
2564 spin_lock(&fs_info->delalloc_root_lock);
2565 BUG_ON(list_empty(&root->delalloc_root));
2566 list_del_init(&root->delalloc_root);
2567 spin_unlock(&fs_info->delalloc_root_lock);
2572 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
2573 struct btrfs_inode *inode)
2575 spin_lock(&root->delalloc_lock);
2576 __btrfs_del_delalloc_inode(root, inode);
2577 spin_unlock(&root->delalloc_lock);
2581 * Properly track delayed allocation bytes in the inode and to maintain the
2582 * list of inodes that have pending delalloc work to be done.
2584 void btrfs_set_delalloc_extent(struct btrfs_inode *inode, struct extent_state *state,
2587 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2589 if ((bits & EXTENT_DEFRAG) && !(bits & EXTENT_DELALLOC))
2592 * set_bit and clear bit hooks normally require _irqsave/restore
2593 * but in this case, we are only testing for the DELALLOC
2594 * bit, which is only set or cleared with irqs on
2596 if (!(state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2597 struct btrfs_root *root = inode->root;
2598 u64 len = state->end + 1 - state->start;
2599 u32 num_extents = count_max_extents(fs_info, len);
2600 bool do_list = !btrfs_is_free_space_inode(inode);
2602 spin_lock(&inode->lock);
2603 btrfs_mod_outstanding_extents(inode, num_extents);
2604 spin_unlock(&inode->lock);
2606 /* For sanity tests */
2607 if (btrfs_is_testing(fs_info))
2610 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
2611 fs_info->delalloc_batch);
2612 spin_lock(&inode->lock);
2613 inode->delalloc_bytes += len;
2614 if (bits & EXTENT_DEFRAG)
2615 inode->defrag_bytes += len;
2616 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2617 &inode->runtime_flags))
2618 btrfs_add_delalloc_inodes(root, inode);
2619 spin_unlock(&inode->lock);
2622 if (!(state->state & EXTENT_DELALLOC_NEW) &&
2623 (bits & EXTENT_DELALLOC_NEW)) {
2624 spin_lock(&inode->lock);
2625 inode->new_delalloc_bytes += state->end + 1 - state->start;
2626 spin_unlock(&inode->lock);
2631 * Once a range is no longer delalloc this function ensures that proper
2632 * accounting happens.
2634 void btrfs_clear_delalloc_extent(struct btrfs_inode *inode,
2635 struct extent_state *state, u32 bits)
2637 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2638 u64 len = state->end + 1 - state->start;
2639 u32 num_extents = count_max_extents(fs_info, len);
2641 if ((state->state & EXTENT_DEFRAG) && (bits & EXTENT_DEFRAG)) {
2642 spin_lock(&inode->lock);
2643 inode->defrag_bytes -= len;
2644 spin_unlock(&inode->lock);
2648 * set_bit and clear bit hooks normally require _irqsave/restore
2649 * but in this case, we are only testing for the DELALLOC
2650 * bit, which is only set or cleared with irqs on
2652 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
2653 struct btrfs_root *root = inode->root;
2654 bool do_list = !btrfs_is_free_space_inode(inode);
2656 spin_lock(&inode->lock);
2657 btrfs_mod_outstanding_extents(inode, -num_extents);
2658 spin_unlock(&inode->lock);
2661 * We don't reserve metadata space for space cache inodes so we
2662 * don't need to call delalloc_release_metadata if there is an
2665 if (bits & EXTENT_CLEAR_META_RESV &&
2666 root != fs_info->tree_root)
2667 btrfs_delalloc_release_metadata(inode, len, false);
2669 /* For sanity tests. */
2670 if (btrfs_is_testing(fs_info))
2673 if (!btrfs_is_data_reloc_root(root) &&
2674 do_list && !(state->state & EXTENT_NORESERVE) &&
2675 (bits & EXTENT_CLEAR_DATA_RESV))
2676 btrfs_free_reserved_data_space_noquota(fs_info, len);
2678 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
2679 fs_info->delalloc_batch);
2680 spin_lock(&inode->lock);
2681 inode->delalloc_bytes -= len;
2682 if (do_list && inode->delalloc_bytes == 0 &&
2683 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
2684 &inode->runtime_flags))
2685 btrfs_del_delalloc_inode(root, inode);
2686 spin_unlock(&inode->lock);
2689 if ((state->state & EXTENT_DELALLOC_NEW) &&
2690 (bits & EXTENT_DELALLOC_NEW)) {
2691 spin_lock(&inode->lock);
2692 ASSERT(inode->new_delalloc_bytes >= len);
2693 inode->new_delalloc_bytes -= len;
2694 if (bits & EXTENT_ADD_INODE_BYTES)
2695 inode_add_bytes(&inode->vfs_inode, len);
2696 spin_unlock(&inode->lock);
2700 static int btrfs_extract_ordered_extent(struct btrfs_bio *bbio,
2701 struct btrfs_ordered_extent *ordered)
2703 u64 start = (u64)bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
2704 u64 len = bbio->bio.bi_iter.bi_size;
2705 struct btrfs_ordered_extent *new;
2708 /* Must always be called for the beginning of an ordered extent. */
2709 if (WARN_ON_ONCE(start != ordered->disk_bytenr))
2712 /* No need to split if the ordered extent covers the entire bio. */
2713 if (ordered->disk_num_bytes == len) {
2714 refcount_inc(&ordered->refs);
2715 bbio->ordered = ordered;
2720 * Don't split the extent_map for NOCOW extents, as we're writing into
2721 * a pre-existing one.
2723 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
2724 ret = split_extent_map(bbio->inode, bbio->file_offset,
2725 ordered->num_bytes, len,
2726 ordered->disk_bytenr);
2731 new = btrfs_split_ordered_extent(ordered, len);
2733 return PTR_ERR(new);
2734 bbio->ordered = new;
2739 * given a list of ordered sums record them in the inode. This happens
2740 * at IO completion time based on sums calculated at bio submission time.
2742 static int add_pending_csums(struct btrfs_trans_handle *trans,
2743 struct list_head *list)
2745 struct btrfs_ordered_sum *sum;
2746 struct btrfs_root *csum_root = NULL;
2749 list_for_each_entry(sum, list, list) {
2750 trans->adding_csums = true;
2752 csum_root = btrfs_csum_root(trans->fs_info,
2754 ret = btrfs_csum_file_blocks(trans, csum_root, sum);
2755 trans->adding_csums = false;
2762 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode *inode,
2765 struct extent_state **cached_state)
2767 u64 search_start = start;
2768 const u64 end = start + len - 1;
2770 while (search_start < end) {
2771 const u64 search_len = end - search_start + 1;
2772 struct extent_map *em;
2776 em = btrfs_get_extent(inode, NULL, 0, search_start, search_len);
2780 if (em->block_start != EXTENT_MAP_HOLE)
2784 if (em->start < search_start)
2785 em_len -= search_start - em->start;
2786 if (em_len > search_len)
2787 em_len = search_len;
2789 ret = set_extent_bit(&inode->io_tree, search_start,
2790 search_start + em_len - 1,
2791 EXTENT_DELALLOC_NEW, cached_state);
2793 search_start = extent_map_end(em);
2794 free_extent_map(em);
2801 int btrfs_set_extent_delalloc(struct btrfs_inode *inode, u64 start, u64 end,
2802 unsigned int extra_bits,
2803 struct extent_state **cached_state)
2805 WARN_ON(PAGE_ALIGNED(end));
2807 if (start >= i_size_read(&inode->vfs_inode) &&
2808 !(inode->flags & BTRFS_INODE_PREALLOC)) {
2810 * There can't be any extents following eof in this case so just
2811 * set the delalloc new bit for the range directly.
2813 extra_bits |= EXTENT_DELALLOC_NEW;
2817 ret = btrfs_find_new_delalloc_bytes(inode, start,
2824 return set_extent_bit(&inode->io_tree, start, end,
2825 EXTENT_DELALLOC | extra_bits, cached_state);
2828 /* see btrfs_writepage_start_hook for details on why this is required */
2829 struct btrfs_writepage_fixup {
2831 struct btrfs_inode *inode;
2832 struct btrfs_work work;
2835 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2837 struct btrfs_writepage_fixup *fixup =
2838 container_of(work, struct btrfs_writepage_fixup, work);
2839 struct btrfs_ordered_extent *ordered;
2840 struct extent_state *cached_state = NULL;
2841 struct extent_changeset *data_reserved = NULL;
2842 struct page *page = fixup->page;
2843 struct btrfs_inode *inode = fixup->inode;
2844 struct btrfs_fs_info *fs_info = inode->root->fs_info;
2845 u64 page_start = page_offset(page);
2846 u64 page_end = page_offset(page) + PAGE_SIZE - 1;
2848 bool free_delalloc_space = true;
2851 * This is similar to page_mkwrite, we need to reserve the space before
2852 * we take the page lock.
2854 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2860 * Before we queued this fixup, we took a reference on the page.
2861 * page->mapping may go NULL, but it shouldn't be moved to a different
2864 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2866 * Unfortunately this is a little tricky, either
2868 * 1) We got here and our page had already been dealt with and
2869 * we reserved our space, thus ret == 0, so we need to just
2870 * drop our space reservation and bail. This can happen the
2871 * first time we come into the fixup worker, or could happen
2872 * while waiting for the ordered extent.
2873 * 2) Our page was already dealt with, but we happened to get an
2874 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2875 * this case we obviously don't have anything to release, but
2876 * because the page was already dealt with we don't want to
2877 * mark the page with an error, so make sure we're resetting
2878 * ret to 0. This is why we have this check _before_ the ret
2879 * check, because we do not want to have a surprise ENOSPC
2880 * when the page was already properly dealt with.
2883 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2884 btrfs_delalloc_release_space(inode, data_reserved,
2885 page_start, PAGE_SIZE,
2893 * We can't mess with the page state unless it is locked, so now that
2894 * it is locked bail if we failed to make our space reservation.
2899 lock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2901 /* already ordered? We're done */
2902 if (PageOrdered(page))
2905 ordered = btrfs_lookup_ordered_range(inode, page_start, PAGE_SIZE);
2907 unlock_extent(&inode->io_tree, page_start, page_end,
2910 btrfs_start_ordered_extent(ordered);
2911 btrfs_put_ordered_extent(ordered);
2915 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2921 * Everything went as planned, we're now the owner of a dirty page with
2922 * delayed allocation bits set and space reserved for our COW
2925 * The page was dirty when we started, nothing should have cleaned it.
2927 BUG_ON(!PageDirty(page));
2928 free_delalloc_space = false;
2930 btrfs_delalloc_release_extents(inode, PAGE_SIZE);
2931 if (free_delalloc_space)
2932 btrfs_delalloc_release_space(inode, data_reserved, page_start,
2934 unlock_extent(&inode->io_tree, page_start, page_end, &cached_state);
2938 * We hit ENOSPC or other errors. Update the mapping and page
2939 * to reflect the errors and clean the page.
2941 mapping_set_error(page->mapping, ret);
2942 btrfs_mark_ordered_io_finished(inode, page, page_start,
2944 btrfs_page_clear_uptodate(fs_info, page, page_start, PAGE_SIZE);
2945 clear_page_dirty_for_io(page);
2947 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
2951 extent_changeset_free(data_reserved);
2953 * As a precaution, do a delayed iput in case it would be the last iput
2954 * that could need flushing space. Recursing back to fixup worker would
2957 btrfs_add_delayed_iput(inode);
2961 * There are a few paths in the higher layers of the kernel that directly
2962 * set the page dirty bit without asking the filesystem if it is a
2963 * good idea. This causes problems because we want to make sure COW
2964 * properly happens and the data=ordered rules are followed.
2966 * In our case any range that doesn't have the ORDERED bit set
2967 * hasn't been properly setup for IO. We kick off an async process
2968 * to fix it up. The async helper will wait for ordered extents, set
2969 * the delalloc bit and make it safe to write the page.
2971 int btrfs_writepage_cow_fixup(struct page *page)
2973 struct inode *inode = page->mapping->host;
2974 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2975 struct btrfs_writepage_fixup *fixup;
2977 /* This page has ordered extent covering it already */
2978 if (PageOrdered(page))
2982 * PageChecked is set below when we create a fixup worker for this page,
2983 * don't try to create another one if we're already PageChecked()
2985 * The extent_io writepage code will redirty the page if we send back
2988 if (PageChecked(page))
2991 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2996 * We are already holding a reference to this inode from
2997 * write_cache_pages. We need to hold it because the space reservation
2998 * takes place outside of the page lock, and we can't trust
2999 * page->mapping outside of the page lock.
3002 btrfs_page_set_checked(fs_info, page, page_offset(page), PAGE_SIZE);
3004 btrfs_init_work(&fixup->work, btrfs_writepage_fixup_worker, NULL, NULL);
3006 fixup->inode = BTRFS_I(inode);
3007 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
3012 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
3013 struct btrfs_inode *inode, u64 file_pos,
3014 struct btrfs_file_extent_item *stack_fi,
3015 const bool update_inode_bytes,
3016 u64 qgroup_reserved)
3018 struct btrfs_root *root = inode->root;
3019 const u64 sectorsize = root->fs_info->sectorsize;
3020 struct btrfs_path *path;
3021 struct extent_buffer *leaf;
3022 struct btrfs_key ins;
3023 u64 disk_num_bytes = btrfs_stack_file_extent_disk_num_bytes(stack_fi);
3024 u64 disk_bytenr = btrfs_stack_file_extent_disk_bytenr(stack_fi);
3025 u64 offset = btrfs_stack_file_extent_offset(stack_fi);
3026 u64 num_bytes = btrfs_stack_file_extent_num_bytes(stack_fi);
3027 u64 ram_bytes = btrfs_stack_file_extent_ram_bytes(stack_fi);
3028 struct btrfs_drop_extents_args drop_args = { 0 };
3031 path = btrfs_alloc_path();
3036 * we may be replacing one extent in the tree with another.
3037 * The new extent is pinned in the extent map, and we don't want
3038 * to drop it from the cache until it is completely in the btree.
3040 * So, tell btrfs_drop_extents to leave this extent in the cache.
3041 * the caller is expected to unpin it and allow it to be merged
3044 drop_args.path = path;
3045 drop_args.start = file_pos;
3046 drop_args.end = file_pos + num_bytes;
3047 drop_args.replace_extent = true;
3048 drop_args.extent_item_size = sizeof(*stack_fi);
3049 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
3053 if (!drop_args.extent_inserted) {
3054 ins.objectid = btrfs_ino(inode);
3055 ins.offset = file_pos;
3056 ins.type = BTRFS_EXTENT_DATA_KEY;
3058 ret = btrfs_insert_empty_item(trans, root, path, &ins,
3063 leaf = path->nodes[0];
3064 btrfs_set_stack_file_extent_generation(stack_fi, trans->transid);
3065 write_extent_buffer(leaf, stack_fi,
3066 btrfs_item_ptr_offset(leaf, path->slots[0]),
3067 sizeof(struct btrfs_file_extent_item));
3069 btrfs_mark_buffer_dirty(leaf);
3070 btrfs_release_path(path);
3073 * If we dropped an inline extent here, we know the range where it is
3074 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3075 * number of bytes only for that range containing the inline extent.
3076 * The remaining of the range will be processed when clearning the
3077 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3079 if (file_pos == 0 && !IS_ALIGNED(drop_args.bytes_found, sectorsize)) {
3080 u64 inline_size = round_down(drop_args.bytes_found, sectorsize);
3082 inline_size = drop_args.bytes_found - inline_size;
3083 btrfs_update_inode_bytes(inode, sectorsize, inline_size);
3084 drop_args.bytes_found -= inline_size;
3085 num_bytes -= sectorsize;
3088 if (update_inode_bytes)
3089 btrfs_update_inode_bytes(inode, num_bytes, drop_args.bytes_found);
3091 ins.objectid = disk_bytenr;
3092 ins.offset = disk_num_bytes;
3093 ins.type = BTRFS_EXTENT_ITEM_KEY;
3095 ret = btrfs_inode_set_file_extent_range(inode, file_pos, ram_bytes);
3099 ret = btrfs_alloc_reserved_file_extent(trans, root, btrfs_ino(inode),
3101 qgroup_reserved, &ins);
3103 btrfs_free_path(path);
3108 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
3111 struct btrfs_block_group *cache;
3113 cache = btrfs_lookup_block_group(fs_info, start);
3116 spin_lock(&cache->lock);
3117 cache->delalloc_bytes -= len;
3118 spin_unlock(&cache->lock);
3120 btrfs_put_block_group(cache);
3123 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle *trans,
3124 struct btrfs_ordered_extent *oe)
3126 struct btrfs_file_extent_item stack_fi;
3127 bool update_inode_bytes;
3128 u64 num_bytes = oe->num_bytes;
3129 u64 ram_bytes = oe->ram_bytes;
3131 memset(&stack_fi, 0, sizeof(stack_fi));
3132 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_REG);
3133 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, oe->disk_bytenr);
3134 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi,
3135 oe->disk_num_bytes);
3136 btrfs_set_stack_file_extent_offset(&stack_fi, oe->offset);
3137 if (test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags)) {
3138 num_bytes = oe->truncated_len;
3139 ram_bytes = num_bytes;
3141 btrfs_set_stack_file_extent_num_bytes(&stack_fi, num_bytes);
3142 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, ram_bytes);
3143 btrfs_set_stack_file_extent_compression(&stack_fi, oe->compress_type);
3144 /* Encryption and other encoding is reserved and all 0 */
3147 * For delalloc, when completing an ordered extent we update the inode's
3148 * bytes when clearing the range in the inode's io tree, so pass false
3149 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3150 * except if the ordered extent was truncated.
3152 update_inode_bytes = test_bit(BTRFS_ORDERED_DIRECT, &oe->flags) ||
3153 test_bit(BTRFS_ORDERED_ENCODED, &oe->flags) ||
3154 test_bit(BTRFS_ORDERED_TRUNCATED, &oe->flags);
3156 return insert_reserved_file_extent(trans, BTRFS_I(oe->inode),
3157 oe->file_offset, &stack_fi,
3158 update_inode_bytes, oe->qgroup_rsv);
3162 * As ordered data IO finishes, this gets called so we can finish
3163 * an ordered extent if the range of bytes in the file it covers are
3166 int btrfs_finish_one_ordered(struct btrfs_ordered_extent *ordered_extent)
3168 struct btrfs_inode *inode = BTRFS_I(ordered_extent->inode);
3169 struct btrfs_root *root = inode->root;
3170 struct btrfs_fs_info *fs_info = root->fs_info;
3171 struct btrfs_trans_handle *trans = NULL;
3172 struct extent_io_tree *io_tree = &inode->io_tree;
3173 struct extent_state *cached_state = NULL;
3175 int compress_type = 0;
3177 u64 logical_len = ordered_extent->num_bytes;
3178 bool freespace_inode;
3179 bool truncated = false;
3180 bool clear_reserved_extent = true;
3181 unsigned int clear_bits = EXTENT_DEFRAG;
3183 start = ordered_extent->file_offset;
3184 end = start + ordered_extent->num_bytes - 1;
3186 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3187 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
3188 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags) &&
3189 !test_bit(BTRFS_ORDERED_ENCODED, &ordered_extent->flags))
3190 clear_bits |= EXTENT_DELALLOC_NEW;
3192 freespace_inode = btrfs_is_free_space_inode(inode);
3193 if (!freespace_inode)
3194 btrfs_lockdep_acquire(fs_info, btrfs_ordered_extent);
3196 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
3201 if (btrfs_is_zoned(fs_info))
3202 btrfs_zone_finish_endio(fs_info, ordered_extent->disk_bytenr,
3203 ordered_extent->disk_num_bytes);
3205 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
3207 logical_len = ordered_extent->truncated_len;
3208 /* Truncated the entire extent, don't bother adding */
3213 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
3214 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3216 btrfs_inode_safe_disk_i_size_write(inode, 0);
3217 if (freespace_inode)
3218 trans = btrfs_join_transaction_spacecache(root);
3220 trans = btrfs_join_transaction(root);
3221 if (IS_ERR(trans)) {
3222 ret = PTR_ERR(trans);
3226 trans->block_rsv = &inode->block_rsv;
3227 ret = btrfs_update_inode_fallback(trans, root, inode);
3228 if (ret) /* -ENOMEM or corruption */
3229 btrfs_abort_transaction(trans, ret);
3233 clear_bits |= EXTENT_LOCKED;
3234 lock_extent(io_tree, start, end, &cached_state);
3236 if (freespace_inode)
3237 trans = btrfs_join_transaction_spacecache(root);
3239 trans = btrfs_join_transaction(root);
3240 if (IS_ERR(trans)) {
3241 ret = PTR_ERR(trans);
3246 trans->block_rsv = &inode->block_rsv;
3248 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3249 compress_type = ordered_extent->compress_type;
3250 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3251 BUG_ON(compress_type);
3252 ret = btrfs_mark_extent_written(trans, inode,
3253 ordered_extent->file_offset,
3254 ordered_extent->file_offset +
3256 btrfs_zoned_release_data_reloc_bg(fs_info, ordered_extent->disk_bytenr,
3257 ordered_extent->disk_num_bytes);
3259 BUG_ON(root == fs_info->tree_root);
3260 ret = insert_ordered_extent_file_extent(trans, ordered_extent);
3262 clear_reserved_extent = false;
3263 btrfs_release_delalloc_bytes(fs_info,
3264 ordered_extent->disk_bytenr,
3265 ordered_extent->disk_num_bytes);
3268 unpin_extent_cache(&inode->extent_tree, ordered_extent->file_offset,
3269 ordered_extent->num_bytes, trans->transid);
3271 btrfs_abort_transaction(trans, ret);
3275 ret = add_pending_csums(trans, &ordered_extent->list);
3277 btrfs_abort_transaction(trans, ret);
3282 * If this is a new delalloc range, clear its new delalloc flag to
3283 * update the inode's number of bytes. This needs to be done first
3284 * before updating the inode item.
3286 if ((clear_bits & EXTENT_DELALLOC_NEW) &&
3287 !test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags))
3288 clear_extent_bit(&inode->io_tree, start, end,
3289 EXTENT_DELALLOC_NEW | EXTENT_ADD_INODE_BYTES,
3292 btrfs_inode_safe_disk_i_size_write(inode, 0);
3293 ret = btrfs_update_inode_fallback(trans, root, inode);
3294 if (ret) { /* -ENOMEM or corruption */
3295 btrfs_abort_transaction(trans, ret);
3300 clear_extent_bit(&inode->io_tree, start, end, clear_bits,
3304 btrfs_end_transaction(trans);
3306 if (ret || truncated) {
3307 u64 unwritten_start = start;
3310 * If we failed to finish this ordered extent for any reason we
3311 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3312 * extent, and mark the inode with the error if it wasn't
3313 * already set. Any error during writeback would have already
3314 * set the mapping error, so we need to set it if we're the ones
3315 * marking this ordered extent as failed.
3317 if (ret && !test_and_set_bit(BTRFS_ORDERED_IOERR,
3318 &ordered_extent->flags))
3319 mapping_set_error(ordered_extent->inode->i_mapping, -EIO);
3322 unwritten_start += logical_len;
3323 clear_extent_uptodate(io_tree, unwritten_start, end, NULL);
3325 /* Drop extent maps for the part of the extent we didn't write. */
3326 btrfs_drop_extent_map_range(inode, unwritten_start, end, false);
3329 * If the ordered extent had an IOERR or something else went
3330 * wrong we need to return the space for this ordered extent
3331 * back to the allocator. We only free the extent in the
3332 * truncated case if we didn't write out the extent at all.
3334 * If we made it past insert_reserved_file_extent before we
3335 * errored out then we don't need to do this as the accounting
3336 * has already been done.
3338 if ((ret || !logical_len) &&
3339 clear_reserved_extent &&
3340 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3341 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3343 * Discard the range before returning it back to the
3346 if (ret && btrfs_test_opt(fs_info, DISCARD_SYNC))
3347 btrfs_discard_extent(fs_info,
3348 ordered_extent->disk_bytenr,
3349 ordered_extent->disk_num_bytes,
3351 btrfs_free_reserved_extent(fs_info,
3352 ordered_extent->disk_bytenr,
3353 ordered_extent->disk_num_bytes, 1);
3355 * Actually free the qgroup rsv which was released when
3356 * the ordered extent was created.
3358 btrfs_qgroup_free_refroot(fs_info, inode->root->root_key.objectid,
3359 ordered_extent->qgroup_rsv,
3360 BTRFS_QGROUP_RSV_DATA);
3365 * This needs to be done to make sure anybody waiting knows we are done
3366 * updating everything for this ordered extent.
3368 btrfs_remove_ordered_extent(inode, ordered_extent);
3371 btrfs_put_ordered_extent(ordered_extent);
3372 /* once for the tree */
3373 btrfs_put_ordered_extent(ordered_extent);
3378 int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered)
3380 if (btrfs_is_zoned(btrfs_sb(ordered->inode->i_sb)) &&
3381 !test_bit(BTRFS_ORDERED_IOERR, &ordered->flags))
3382 btrfs_finish_ordered_zoned(ordered);
3383 return btrfs_finish_one_ordered(ordered);
3387 * Verify the checksum for a single sector without any extra action that depend
3388 * on the type of I/O.
3390 int btrfs_check_sector_csum(struct btrfs_fs_info *fs_info, struct page *page,
3391 u32 pgoff, u8 *csum, const u8 * const csum_expected)
3393 SHASH_DESC_ON_STACK(shash, fs_info->csum_shash);
3396 ASSERT(pgoff + fs_info->sectorsize <= PAGE_SIZE);
3398 shash->tfm = fs_info->csum_shash;
3400 kaddr = kmap_local_page(page) + pgoff;
3401 crypto_shash_digest(shash, kaddr, fs_info->sectorsize, csum);
3402 kunmap_local(kaddr);
3404 if (memcmp(csum, csum_expected, fs_info->csum_size))
3410 * Verify the checksum of a single data sector.
3412 * @bbio: btrfs_io_bio which contains the csum
3413 * @dev: device the sector is on
3414 * @bio_offset: offset to the beginning of the bio (in bytes)
3415 * @bv: bio_vec to check
3417 * Check if the checksum on a data block is valid. When a checksum mismatch is
3418 * detected, report the error and fill the corrupted range with zero.
3420 * Return %true if the sector is ok or had no checksum to start with, else %false.
3422 bool btrfs_data_csum_ok(struct btrfs_bio *bbio, struct btrfs_device *dev,
3423 u32 bio_offset, struct bio_vec *bv)
3425 struct btrfs_inode *inode = bbio->inode;
3426 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3427 u64 file_offset = bbio->file_offset + bio_offset;
3428 u64 end = file_offset + bv->bv_len - 1;
3430 u8 csum[BTRFS_CSUM_SIZE];
3432 ASSERT(bv->bv_len == fs_info->sectorsize);
3437 if (btrfs_is_data_reloc_root(inode->root) &&
3438 test_range_bit(&inode->io_tree, file_offset, end, EXTENT_NODATASUM,
3440 /* Skip the range without csum for data reloc inode */
3441 clear_extent_bits(&inode->io_tree, file_offset, end,
3446 csum_expected = bbio->csum + (bio_offset >> fs_info->sectorsize_bits) *
3448 if (btrfs_check_sector_csum(fs_info, bv->bv_page, bv->bv_offset, csum,
3454 btrfs_print_data_csum_error(inode, file_offset, csum, csum_expected,
3457 btrfs_dev_stat_inc_and_print(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS);
3463 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3465 * @inode: The inode we want to perform iput on
3467 * This function uses the generic vfs_inode::i_count to track whether we should
3468 * just decrement it (in case it's > 1) or if this is the last iput then link
3469 * the inode to the delayed iput machinery. Delayed iputs are processed at
3470 * transaction commit time/superblock commit/cleaner kthread.
3472 void btrfs_add_delayed_iput(struct btrfs_inode *inode)
3474 struct btrfs_fs_info *fs_info = inode->root->fs_info;
3475 unsigned long flags;
3477 if (atomic_add_unless(&inode->vfs_inode.i_count, -1, 1))
3480 atomic_inc(&fs_info->nr_delayed_iputs);
3482 * Need to be irq safe here because we can be called from either an irq
3483 * context (see bio.c and btrfs_put_ordered_extent()) or a non-irq
3486 spin_lock_irqsave(&fs_info->delayed_iput_lock, flags);
3487 ASSERT(list_empty(&inode->delayed_iput));
3488 list_add_tail(&inode->delayed_iput, &fs_info->delayed_iputs);
3489 spin_unlock_irqrestore(&fs_info->delayed_iput_lock, flags);
3490 if (!test_bit(BTRFS_FS_CLEANER_RUNNING, &fs_info->flags))
3491 wake_up_process(fs_info->cleaner_kthread);
3494 static void run_delayed_iput_locked(struct btrfs_fs_info *fs_info,
3495 struct btrfs_inode *inode)
3497 list_del_init(&inode->delayed_iput);
3498 spin_unlock_irq(&fs_info->delayed_iput_lock);
3499 iput(&inode->vfs_inode);
3500 if (atomic_dec_and_test(&fs_info->nr_delayed_iputs))
3501 wake_up(&fs_info->delayed_iputs_wait);
3502 spin_lock_irq(&fs_info->delayed_iput_lock);
3505 static void btrfs_run_delayed_iput(struct btrfs_fs_info *fs_info,
3506 struct btrfs_inode *inode)
3508 if (!list_empty(&inode->delayed_iput)) {
3509 spin_lock_irq(&fs_info->delayed_iput_lock);
3510 if (!list_empty(&inode->delayed_iput))
3511 run_delayed_iput_locked(fs_info, inode);
3512 spin_unlock_irq(&fs_info->delayed_iput_lock);
3516 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3519 * btrfs_put_ordered_extent() can run in irq context (see bio.c), which
3520 * calls btrfs_add_delayed_iput() and that needs to lock
3521 * fs_info->delayed_iput_lock. So we need to disable irqs here to
3522 * prevent a deadlock.
3524 spin_lock_irq(&fs_info->delayed_iput_lock);
3525 while (!list_empty(&fs_info->delayed_iputs)) {
3526 struct btrfs_inode *inode;
3528 inode = list_first_entry(&fs_info->delayed_iputs,
3529 struct btrfs_inode, delayed_iput);
3530 run_delayed_iput_locked(fs_info, inode);
3531 if (need_resched()) {
3532 spin_unlock_irq(&fs_info->delayed_iput_lock);
3534 spin_lock_irq(&fs_info->delayed_iput_lock);
3537 spin_unlock_irq(&fs_info->delayed_iput_lock);
3541 * Wait for flushing all delayed iputs
3543 * @fs_info: the filesystem
3545 * This will wait on any delayed iputs that are currently running with KILLABLE
3546 * set. Once they are all done running we will return, unless we are killed in
3547 * which case we return EINTR. This helps in user operations like fallocate etc
3548 * that might get blocked on the iputs.
3550 * Return EINTR if we were killed, 0 if nothing's pending
3552 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info *fs_info)
3554 int ret = wait_event_killable(fs_info->delayed_iputs_wait,
3555 atomic_read(&fs_info->nr_delayed_iputs) == 0);
3562 * This creates an orphan entry for the given inode in case something goes wrong
3563 * in the middle of an unlink.
3565 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3566 struct btrfs_inode *inode)
3570 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3571 if (ret && ret != -EEXIST) {
3572 btrfs_abort_transaction(trans, ret);
3580 * We have done the delete so we can go ahead and remove the orphan item for
3581 * this particular inode.
3583 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3584 struct btrfs_inode *inode)
3586 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3590 * this cleans up any orphans that may be left on the list from the last use
3593 int btrfs_orphan_cleanup(struct btrfs_root *root)
3595 struct btrfs_fs_info *fs_info = root->fs_info;
3596 struct btrfs_path *path;
3597 struct extent_buffer *leaf;
3598 struct btrfs_key key, found_key;
3599 struct btrfs_trans_handle *trans;
3600 struct inode *inode;
3601 u64 last_objectid = 0;
3602 int ret = 0, nr_unlink = 0;
3604 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP, &root->state))
3607 path = btrfs_alloc_path();
3612 path->reada = READA_BACK;
3614 key.objectid = BTRFS_ORPHAN_OBJECTID;
3615 key.type = BTRFS_ORPHAN_ITEM_KEY;
3616 key.offset = (u64)-1;
3619 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3624 * if ret == 0 means we found what we were searching for, which
3625 * is weird, but possible, so only screw with path if we didn't
3626 * find the key and see if we have stuff that matches
3630 if (path->slots[0] == 0)
3635 /* pull out the item */
3636 leaf = path->nodes[0];
3637 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3639 /* make sure the item matches what we want */
3640 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3642 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3645 /* release the path since we're done with it */
3646 btrfs_release_path(path);
3649 * this is where we are basically btrfs_lookup, without the
3650 * crossing root thing. we store the inode number in the
3651 * offset of the orphan item.
3654 if (found_key.offset == last_objectid) {
3656 "Error removing orphan entry, stopping orphan cleanup");
3661 last_objectid = found_key.offset;
3663 found_key.objectid = found_key.offset;
3664 found_key.type = BTRFS_INODE_ITEM_KEY;
3665 found_key.offset = 0;
3666 inode = btrfs_iget(fs_info->sb, last_objectid, root);
3667 if (IS_ERR(inode)) {
3668 ret = PTR_ERR(inode);
3674 if (!inode && root == fs_info->tree_root) {
3675 struct btrfs_root *dead_root;
3676 int is_dead_root = 0;
3679 * This is an orphan in the tree root. Currently these
3680 * could come from 2 sources:
3681 * a) a root (snapshot/subvolume) deletion in progress
3682 * b) a free space cache inode
3683 * We need to distinguish those two, as the orphan item
3684 * for a root must not get deleted before the deletion
3685 * of the snapshot/subvolume's tree completes.
3687 * btrfs_find_orphan_roots() ran before us, which has
3688 * found all deleted roots and loaded them into
3689 * fs_info->fs_roots_radix. So here we can find if an
3690 * orphan item corresponds to a deleted root by looking
3691 * up the root from that radix tree.
3694 spin_lock(&fs_info->fs_roots_radix_lock);
3695 dead_root = radix_tree_lookup(&fs_info->fs_roots_radix,
3696 (unsigned long)found_key.objectid);
3697 if (dead_root && btrfs_root_refs(&dead_root->root_item) == 0)
3699 spin_unlock(&fs_info->fs_roots_radix_lock);
3702 /* prevent this orphan from being found again */
3703 key.offset = found_key.objectid - 1;
3710 * If we have an inode with links, there are a couple of
3713 * 1. We were halfway through creating fsverity metadata for the
3714 * file. In that case, the orphan item represents incomplete
3715 * fsverity metadata which must be cleaned up with
3716 * btrfs_drop_verity_items and deleting the orphan item.
3718 * 2. Old kernels (before v3.12) used to create an
3719 * orphan item for truncate indicating that there were possibly
3720 * extent items past i_size that needed to be deleted. In v3.12,
3721 * truncate was changed to update i_size in sync with the extent
3722 * items, but the (useless) orphan item was still created. Since
3723 * v4.18, we don't create the orphan item for truncate at all.
3725 * So, this item could mean that we need to do a truncate, but
3726 * only if this filesystem was last used on a pre-v3.12 kernel
3727 * and was not cleanly unmounted. The odds of that are quite
3728 * slim, and it's a pain to do the truncate now, so just delete
3731 * It's also possible that this orphan item was supposed to be
3732 * deleted but wasn't. The inode number may have been reused,
3733 * but either way, we can delete the orphan item.
3735 if (!inode || inode->i_nlink) {
3737 ret = btrfs_drop_verity_items(BTRFS_I(inode));
3743 trans = btrfs_start_transaction(root, 1);
3744 if (IS_ERR(trans)) {
3745 ret = PTR_ERR(trans);
3748 btrfs_debug(fs_info, "auto deleting %Lu",
3749 found_key.objectid);
3750 ret = btrfs_del_orphan_item(trans, root,
3751 found_key.objectid);
3752 btrfs_end_transaction(trans);
3760 /* this will do delete_inode and everything for us */
3763 /* release the path since we're done with it */
3764 btrfs_release_path(path);
3766 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3767 trans = btrfs_join_transaction(root);
3769 btrfs_end_transaction(trans);
3773 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3777 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3778 btrfs_free_path(path);
3783 * very simple check to peek ahead in the leaf looking for xattrs. If we
3784 * don't find any xattrs, we know there can't be any acls.
3786 * slot is the slot the inode is in, objectid is the objectid of the inode
3788 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3789 int slot, u64 objectid,
3790 int *first_xattr_slot)
3792 u32 nritems = btrfs_header_nritems(leaf);
3793 struct btrfs_key found_key;
3794 static u64 xattr_access = 0;
3795 static u64 xattr_default = 0;
3798 if (!xattr_access) {
3799 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3800 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3801 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3802 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3806 *first_xattr_slot = -1;
3807 while (slot < nritems) {
3808 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3810 /* we found a different objectid, there must not be acls */
3811 if (found_key.objectid != objectid)
3814 /* we found an xattr, assume we've got an acl */
3815 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3816 if (*first_xattr_slot == -1)
3817 *first_xattr_slot = slot;
3818 if (found_key.offset == xattr_access ||
3819 found_key.offset == xattr_default)
3824 * we found a key greater than an xattr key, there can't
3825 * be any acls later on
3827 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3834 * it goes inode, inode backrefs, xattrs, extents,
3835 * so if there are a ton of hard links to an inode there can
3836 * be a lot of backrefs. Don't waste time searching too hard,
3837 * this is just an optimization
3842 /* we hit the end of the leaf before we found an xattr or
3843 * something larger than an xattr. We have to assume the inode
3846 if (*first_xattr_slot == -1)
3847 *first_xattr_slot = slot;
3852 * read an inode from the btree into the in-memory inode
3854 static int btrfs_read_locked_inode(struct inode *inode,
3855 struct btrfs_path *in_path)
3857 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3858 struct btrfs_path *path = in_path;
3859 struct extent_buffer *leaf;
3860 struct btrfs_inode_item *inode_item;
3861 struct btrfs_root *root = BTRFS_I(inode)->root;
3862 struct btrfs_key location;
3867 bool filled = false;
3868 int first_xattr_slot;
3870 ret = btrfs_fill_inode(inode, &rdev);
3875 path = btrfs_alloc_path();
3880 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3882 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3884 if (path != in_path)
3885 btrfs_free_path(path);
3889 leaf = path->nodes[0];
3894 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3895 struct btrfs_inode_item);
3896 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3897 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3898 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3899 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3900 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3901 btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
3902 round_up(i_size_read(inode), fs_info->sectorsize));
3904 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3905 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3907 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3908 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3910 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3911 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3913 BTRFS_I(inode)->i_otime.tv_sec =
3914 btrfs_timespec_sec(leaf, &inode_item->otime);
3915 BTRFS_I(inode)->i_otime.tv_nsec =
3916 btrfs_timespec_nsec(leaf, &inode_item->otime);
3918 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3919 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3920 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3922 inode_set_iversion_queried(inode,
3923 btrfs_inode_sequence(leaf, inode_item));
3924 inode->i_generation = BTRFS_I(inode)->generation;
3926 rdev = btrfs_inode_rdev(leaf, inode_item);
3928 BTRFS_I(inode)->index_cnt = (u64)-1;
3929 btrfs_inode_split_flags(btrfs_inode_flags(leaf, inode_item),
3930 &BTRFS_I(inode)->flags, &BTRFS_I(inode)->ro_flags);
3934 * If we were modified in the current generation and evicted from memory
3935 * and then re-read we need to do a full sync since we don't have any
3936 * idea about which extents were modified before we were evicted from
3939 * This is required for both inode re-read from disk and delayed inode
3940 * in delayed_nodes_tree.
3942 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3943 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3944 &BTRFS_I(inode)->runtime_flags);
3947 * We don't persist the id of the transaction where an unlink operation
3948 * against the inode was last made. So here we assume the inode might
3949 * have been evicted, and therefore the exact value of last_unlink_trans
3950 * lost, and set it to last_trans to avoid metadata inconsistencies
3951 * between the inode and its parent if the inode is fsync'ed and the log
3952 * replayed. For example, in the scenario:
3955 * ln mydir/foo mydir/bar
3958 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3959 * xfs_io -c fsync mydir/foo
3961 * mount fs, triggers fsync log replay
3963 * We must make sure that when we fsync our inode foo we also log its
3964 * parent inode, otherwise after log replay the parent still has the
3965 * dentry with the "bar" name but our inode foo has a link count of 1
3966 * and doesn't have an inode ref with the name "bar" anymore.
3968 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3969 * but it guarantees correctness at the expense of occasional full
3970 * transaction commits on fsync if our inode is a directory, or if our
3971 * inode is not a directory, logging its parent unnecessarily.
3973 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3976 * Same logic as for last_unlink_trans. We don't persist the generation
3977 * of the last transaction where this inode was used for a reflink
3978 * operation, so after eviction and reloading the inode we must be
3979 * pessimistic and assume the last transaction that modified the inode.
3981 BTRFS_I(inode)->last_reflink_trans = BTRFS_I(inode)->last_trans;
3984 if (inode->i_nlink != 1 ||
3985 path->slots[0] >= btrfs_header_nritems(leaf))
3988 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3989 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3992 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3993 if (location.type == BTRFS_INODE_REF_KEY) {
3994 struct btrfs_inode_ref *ref;
3996 ref = (struct btrfs_inode_ref *)ptr;
3997 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3998 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3999 struct btrfs_inode_extref *extref;
4001 extref = (struct btrfs_inode_extref *)ptr;
4002 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
4007 * try to precache a NULL acl entry for files that don't have
4008 * any xattrs or acls
4010 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
4011 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
4012 if (first_xattr_slot != -1) {
4013 path->slots[0] = first_xattr_slot;
4014 ret = btrfs_load_inode_props(inode, path);
4017 "error loading props for ino %llu (root %llu): %d",
4018 btrfs_ino(BTRFS_I(inode)),
4019 root->root_key.objectid, ret);
4021 if (path != in_path)
4022 btrfs_free_path(path);
4025 cache_no_acl(inode);
4027 switch (inode->i_mode & S_IFMT) {
4029 inode->i_mapping->a_ops = &btrfs_aops;
4030 inode->i_fop = &btrfs_file_operations;
4031 inode->i_op = &btrfs_file_inode_operations;
4034 inode->i_fop = &btrfs_dir_file_operations;
4035 inode->i_op = &btrfs_dir_inode_operations;
4038 inode->i_op = &btrfs_symlink_inode_operations;
4039 inode_nohighmem(inode);
4040 inode->i_mapping->a_ops = &btrfs_aops;
4043 inode->i_op = &btrfs_special_inode_operations;
4044 init_special_inode(inode, inode->i_mode, rdev);
4048 btrfs_sync_inode_flags_to_i_flags(inode);
4053 * given a leaf and an inode, copy the inode fields into the leaf
4055 static void fill_inode_item(struct btrfs_trans_handle *trans,
4056 struct extent_buffer *leaf,
4057 struct btrfs_inode_item *item,
4058 struct inode *inode)
4060 struct btrfs_map_token token;
4063 btrfs_init_map_token(&token, leaf);
4065 btrfs_set_token_inode_uid(&token, item, i_uid_read(inode));
4066 btrfs_set_token_inode_gid(&token, item, i_gid_read(inode));
4067 btrfs_set_token_inode_size(&token, item, BTRFS_I(inode)->disk_i_size);
4068 btrfs_set_token_inode_mode(&token, item, inode->i_mode);
4069 btrfs_set_token_inode_nlink(&token, item, inode->i_nlink);
4071 btrfs_set_token_timespec_sec(&token, &item->atime,
4072 inode->i_atime.tv_sec);
4073 btrfs_set_token_timespec_nsec(&token, &item->atime,
4074 inode->i_atime.tv_nsec);
4076 btrfs_set_token_timespec_sec(&token, &item->mtime,
4077 inode->i_mtime.tv_sec);
4078 btrfs_set_token_timespec_nsec(&token, &item->mtime,
4079 inode->i_mtime.tv_nsec);
4081 btrfs_set_token_timespec_sec(&token, &item->ctime,
4082 inode->i_ctime.tv_sec);
4083 btrfs_set_token_timespec_nsec(&token, &item->ctime,
4084 inode->i_ctime.tv_nsec);
4086 btrfs_set_token_timespec_sec(&token, &item->otime,
4087 BTRFS_I(inode)->i_otime.tv_sec);
4088 btrfs_set_token_timespec_nsec(&token, &item->otime,
4089 BTRFS_I(inode)->i_otime.tv_nsec);
4091 btrfs_set_token_inode_nbytes(&token, item, inode_get_bytes(inode));
4092 btrfs_set_token_inode_generation(&token, item,
4093 BTRFS_I(inode)->generation);
4094 btrfs_set_token_inode_sequence(&token, item, inode_peek_iversion(inode));
4095 btrfs_set_token_inode_transid(&token, item, trans->transid);
4096 btrfs_set_token_inode_rdev(&token, item, inode->i_rdev);
4097 flags = btrfs_inode_combine_flags(BTRFS_I(inode)->flags,
4098 BTRFS_I(inode)->ro_flags);
4099 btrfs_set_token_inode_flags(&token, item, flags);
4100 btrfs_set_token_inode_block_group(&token, item, 0);
4104 * copy everything in the in-memory inode into the btree.
4106 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
4107 struct btrfs_root *root,
4108 struct btrfs_inode *inode)
4110 struct btrfs_inode_item *inode_item;
4111 struct btrfs_path *path;
4112 struct extent_buffer *leaf;
4115 path = btrfs_alloc_path();
4119 ret = btrfs_lookup_inode(trans, root, path, &inode->location, 1);
4126 leaf = path->nodes[0];
4127 inode_item = btrfs_item_ptr(leaf, path->slots[0],
4128 struct btrfs_inode_item);
4130 fill_inode_item(trans, leaf, inode_item, &inode->vfs_inode);
4131 btrfs_mark_buffer_dirty(leaf);
4132 btrfs_set_inode_last_trans(trans, inode);
4135 btrfs_free_path(path);
4140 * copy everything in the in-memory inode into the btree.
4142 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
4143 struct btrfs_root *root,
4144 struct btrfs_inode *inode)
4146 struct btrfs_fs_info *fs_info = root->fs_info;
4150 * If the inode is a free space inode, we can deadlock during commit
4151 * if we put it into the delayed code.
4153 * The data relocation inode should also be directly updated
4156 if (!btrfs_is_free_space_inode(inode)
4157 && !btrfs_is_data_reloc_root(root)
4158 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
4159 btrfs_update_root_times(trans, root);
4161 ret = btrfs_delayed_update_inode(trans, root, inode);
4163 btrfs_set_inode_last_trans(trans, inode);
4167 return btrfs_update_inode_item(trans, root, inode);
4170 int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4171 struct btrfs_root *root, struct btrfs_inode *inode)
4175 ret = btrfs_update_inode(trans, root, inode);
4177 return btrfs_update_inode_item(trans, root, inode);
4182 * unlink helper that gets used here in inode.c and in the tree logging
4183 * recovery code. It remove a link in a directory with a given name, and
4184 * also drops the back refs in the inode to the directory
4186 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4187 struct btrfs_inode *dir,
4188 struct btrfs_inode *inode,
4189 const struct fscrypt_str *name,
4190 struct btrfs_rename_ctx *rename_ctx)
4192 struct btrfs_root *root = dir->root;
4193 struct btrfs_fs_info *fs_info = root->fs_info;
4194 struct btrfs_path *path;
4196 struct btrfs_dir_item *di;
4198 u64 ino = btrfs_ino(inode);
4199 u64 dir_ino = btrfs_ino(dir);
4201 path = btrfs_alloc_path();
4207 di = btrfs_lookup_dir_item(trans, root, path, dir_ino, name, -1);
4208 if (IS_ERR_OR_NULL(di)) {
4209 ret = di ? PTR_ERR(di) : -ENOENT;
4212 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4215 btrfs_release_path(path);
4218 * If we don't have dir index, we have to get it by looking up
4219 * the inode ref, since we get the inode ref, remove it directly,
4220 * it is unnecessary to do delayed deletion.
4222 * But if we have dir index, needn't search inode ref to get it.
4223 * Since the inode ref is close to the inode item, it is better
4224 * that we delay to delete it, and just do this deletion when
4225 * we update the inode item.
4227 if (inode->dir_index) {
4228 ret = btrfs_delayed_delete_inode_ref(inode);
4230 index = inode->dir_index;
4235 ret = btrfs_del_inode_ref(trans, root, name, ino, dir_ino, &index);
4238 "failed to delete reference to %.*s, inode %llu parent %llu",
4239 name->len, name->name, ino, dir_ino);
4240 btrfs_abort_transaction(trans, ret);
4245 rename_ctx->index = index;
4247 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4249 btrfs_abort_transaction(trans, ret);
4254 * If we are in a rename context, we don't need to update anything in the
4255 * log. That will be done later during the rename by btrfs_log_new_name().
4256 * Besides that, doing it here would only cause extra unnecessary btree
4257 * operations on the log tree, increasing latency for applications.
4260 btrfs_del_inode_ref_in_log(trans, root, name, inode, dir_ino);
4261 btrfs_del_dir_entries_in_log(trans, root, name, dir, index);
4265 * If we have a pending delayed iput we could end up with the final iput
4266 * being run in btrfs-cleaner context. If we have enough of these built
4267 * up we can end up burning a lot of time in btrfs-cleaner without any
4268 * way to throttle the unlinks. Since we're currently holding a ref on
4269 * the inode we can run the delayed iput here without any issues as the
4270 * final iput won't be done until after we drop the ref we're currently
4273 btrfs_run_delayed_iput(fs_info, inode);
4275 btrfs_free_path(path);
4279 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name->len * 2);
4280 inode_inc_iversion(&inode->vfs_inode);
4281 inode_inc_iversion(&dir->vfs_inode);
4282 inode->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4283 dir->vfs_inode.i_mtime = inode->vfs_inode.i_ctime;
4284 dir->vfs_inode.i_ctime = inode->vfs_inode.i_ctime;
4285 ret = btrfs_update_inode(trans, root, dir);
4290 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4291 struct btrfs_inode *dir, struct btrfs_inode *inode,
4292 const struct fscrypt_str *name)
4296 ret = __btrfs_unlink_inode(trans, dir, inode, name, NULL);
4298 drop_nlink(&inode->vfs_inode);
4299 ret = btrfs_update_inode(trans, inode->root, inode);
4305 * helper to start transaction for unlink and rmdir.
4307 * unlink and rmdir are special in btrfs, they do not always free space, so
4308 * if we cannot make our reservations the normal way try and see if there is
4309 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4310 * allow the unlink to occur.
4312 static struct btrfs_trans_handle *__unlink_start_trans(struct btrfs_inode *dir)
4314 struct btrfs_root *root = dir->root;
4316 return btrfs_start_transaction_fallback_global_rsv(root,
4317 BTRFS_UNLINK_METADATA_UNITS);
4320 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4322 struct btrfs_trans_handle *trans;
4323 struct inode *inode = d_inode(dentry);
4325 struct fscrypt_name fname;
4327 ret = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4331 /* This needs to handle no-key deletions later on */
4333 trans = __unlink_start_trans(BTRFS_I(dir));
4334 if (IS_ERR(trans)) {
4335 ret = PTR_ERR(trans);
4339 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4342 ret = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4347 if (inode->i_nlink == 0) {
4348 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4354 btrfs_end_transaction(trans);
4355 btrfs_btree_balance_dirty(BTRFS_I(dir)->root->fs_info);
4357 fscrypt_free_filename(&fname);
4361 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4362 struct btrfs_inode *dir, struct dentry *dentry)
4364 struct btrfs_root *root = dir->root;
4365 struct btrfs_inode *inode = BTRFS_I(d_inode(dentry));
4366 struct btrfs_path *path;
4367 struct extent_buffer *leaf;
4368 struct btrfs_dir_item *di;
4369 struct btrfs_key key;
4373 u64 dir_ino = btrfs_ino(dir);
4374 struct fscrypt_name fname;
4376 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
4380 /* This needs to handle no-key deletions later on */
4382 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID) {
4383 objectid = inode->root->root_key.objectid;
4384 } else if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4385 objectid = inode->location.objectid;
4388 fscrypt_free_filename(&fname);
4392 path = btrfs_alloc_path();
4398 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4399 &fname.disk_name, -1);
4400 if (IS_ERR_OR_NULL(di)) {
4401 ret = di ? PTR_ERR(di) : -ENOENT;
4405 leaf = path->nodes[0];
4406 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4407 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4408 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4410 btrfs_abort_transaction(trans, ret);
4413 btrfs_release_path(path);
4416 * This is a placeholder inode for a subvolume we didn't have a
4417 * reference to at the time of the snapshot creation. In the meantime
4418 * we could have renamed the real subvol link into our snapshot, so
4419 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4420 * Instead simply lookup the dir_index_item for this entry so we can
4421 * remove it. Otherwise we know we have a ref to the root and we can
4422 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4424 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID) {
4425 di = btrfs_search_dir_index_item(root, path, dir_ino, &fname.disk_name);
4426 if (IS_ERR_OR_NULL(di)) {
4431 btrfs_abort_transaction(trans, ret);
4435 leaf = path->nodes[0];
4436 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4438 btrfs_release_path(path);
4440 ret = btrfs_del_root_ref(trans, objectid,
4441 root->root_key.objectid, dir_ino,
4442 &index, &fname.disk_name);
4444 btrfs_abort_transaction(trans, ret);
4449 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
4451 btrfs_abort_transaction(trans, ret);
4455 btrfs_i_size_write(dir, dir->vfs_inode.i_size - fname.disk_name.len * 2);
4456 inode_inc_iversion(&dir->vfs_inode);
4457 dir->vfs_inode.i_mtime = current_time(&dir->vfs_inode);
4458 dir->vfs_inode.i_ctime = dir->vfs_inode.i_mtime;
4459 ret = btrfs_update_inode_fallback(trans, root, dir);
4461 btrfs_abort_transaction(trans, ret);
4463 btrfs_free_path(path);
4464 fscrypt_free_filename(&fname);
4469 * Helper to check if the subvolume references other subvolumes or if it's
4472 static noinline int may_destroy_subvol(struct btrfs_root *root)
4474 struct btrfs_fs_info *fs_info = root->fs_info;
4475 struct btrfs_path *path;
4476 struct btrfs_dir_item *di;
4477 struct btrfs_key key;
4478 struct fscrypt_str name = FSTR_INIT("default", 7);
4482 path = btrfs_alloc_path();
4486 /* Make sure this root isn't set as the default subvol */
4487 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4488 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4490 if (di && !IS_ERR(di)) {
4491 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4492 if (key.objectid == root->root_key.objectid) {
4495 "deleting default subvolume %llu is not allowed",
4499 btrfs_release_path(path);
4502 key.objectid = root->root_key.objectid;
4503 key.type = BTRFS_ROOT_REF_KEY;
4504 key.offset = (u64)-1;
4506 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4512 if (path->slots[0] > 0) {
4514 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4515 if (key.objectid == root->root_key.objectid &&
4516 key.type == BTRFS_ROOT_REF_KEY)
4520 btrfs_free_path(path);
4524 /* Delete all dentries for inodes belonging to the root */
4525 static void btrfs_prune_dentries(struct btrfs_root *root)
4527 struct btrfs_fs_info *fs_info = root->fs_info;
4528 struct rb_node *node;
4529 struct rb_node *prev;
4530 struct btrfs_inode *entry;
4531 struct inode *inode;
4534 if (!BTRFS_FS_ERROR(fs_info))
4535 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4537 spin_lock(&root->inode_lock);
4539 node = root->inode_tree.rb_node;
4543 entry = rb_entry(node, struct btrfs_inode, rb_node);
4545 if (objectid < btrfs_ino(entry))
4546 node = node->rb_left;
4547 else if (objectid > btrfs_ino(entry))
4548 node = node->rb_right;
4554 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4555 if (objectid <= btrfs_ino(entry)) {
4559 prev = rb_next(prev);
4563 entry = rb_entry(node, struct btrfs_inode, rb_node);
4564 objectid = btrfs_ino(entry) + 1;
4565 inode = igrab(&entry->vfs_inode);
4567 spin_unlock(&root->inode_lock);
4568 if (atomic_read(&inode->i_count) > 1)
4569 d_prune_aliases(inode);
4571 * btrfs_drop_inode will have it removed from the inode
4572 * cache when its usage count hits zero.
4576 spin_lock(&root->inode_lock);
4580 if (cond_resched_lock(&root->inode_lock))
4583 node = rb_next(node);
4585 spin_unlock(&root->inode_lock);
4588 int btrfs_delete_subvolume(struct btrfs_inode *dir, struct dentry *dentry)
4590 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4591 struct btrfs_root *root = dir->root;
4592 struct inode *inode = d_inode(dentry);
4593 struct btrfs_root *dest = BTRFS_I(inode)->root;
4594 struct btrfs_trans_handle *trans;
4595 struct btrfs_block_rsv block_rsv;
4600 * Don't allow to delete a subvolume with send in progress. This is
4601 * inside the inode lock so the error handling that has to drop the bit
4602 * again is not run concurrently.
4604 spin_lock(&dest->root_item_lock);
4605 if (dest->send_in_progress) {
4606 spin_unlock(&dest->root_item_lock);
4608 "attempt to delete subvolume %llu during send",
4609 dest->root_key.objectid);
4612 if (atomic_read(&dest->nr_swapfiles)) {
4613 spin_unlock(&dest->root_item_lock);
4615 "attempt to delete subvolume %llu with active swapfile",
4616 root->root_key.objectid);
4619 root_flags = btrfs_root_flags(&dest->root_item);
4620 btrfs_set_root_flags(&dest->root_item,
4621 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4622 spin_unlock(&dest->root_item_lock);
4624 down_write(&fs_info->subvol_sem);
4626 ret = may_destroy_subvol(dest);
4630 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4632 * One for dir inode,
4633 * two for dir entries,
4634 * two for root ref/backref.
4636 ret = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4640 trans = btrfs_start_transaction(root, 0);
4641 if (IS_ERR(trans)) {
4642 ret = PTR_ERR(trans);
4645 trans->block_rsv = &block_rsv;
4646 trans->bytes_reserved = block_rsv.size;
4648 btrfs_record_snapshot_destroy(trans, dir);
4650 ret = btrfs_unlink_subvol(trans, dir, dentry);
4652 btrfs_abort_transaction(trans, ret);
4656 ret = btrfs_record_root_in_trans(trans, dest);
4658 btrfs_abort_transaction(trans, ret);
4662 memset(&dest->root_item.drop_progress, 0,
4663 sizeof(dest->root_item.drop_progress));
4664 btrfs_set_root_drop_level(&dest->root_item, 0);
4665 btrfs_set_root_refs(&dest->root_item, 0);
4667 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4668 ret = btrfs_insert_orphan_item(trans,
4670 dest->root_key.objectid);
4672 btrfs_abort_transaction(trans, ret);
4677 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4678 BTRFS_UUID_KEY_SUBVOL,
4679 dest->root_key.objectid);
4680 if (ret && ret != -ENOENT) {
4681 btrfs_abort_transaction(trans, ret);
4684 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4685 ret = btrfs_uuid_tree_remove(trans,
4686 dest->root_item.received_uuid,
4687 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4688 dest->root_key.objectid);
4689 if (ret && ret != -ENOENT) {
4690 btrfs_abort_transaction(trans, ret);
4695 free_anon_bdev(dest->anon_dev);
4698 trans->block_rsv = NULL;
4699 trans->bytes_reserved = 0;
4700 ret = btrfs_end_transaction(trans);
4701 inode->i_flags |= S_DEAD;
4703 btrfs_subvolume_release_metadata(root, &block_rsv);
4705 up_write(&fs_info->subvol_sem);
4707 spin_lock(&dest->root_item_lock);
4708 root_flags = btrfs_root_flags(&dest->root_item);
4709 btrfs_set_root_flags(&dest->root_item,
4710 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4711 spin_unlock(&dest->root_item_lock);
4713 d_invalidate(dentry);
4714 btrfs_prune_dentries(dest);
4715 ASSERT(dest->send_in_progress == 0);
4721 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4723 struct inode *inode = d_inode(dentry);
4724 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
4726 struct btrfs_trans_handle *trans;
4727 u64 last_unlink_trans;
4728 struct fscrypt_name fname;
4730 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4732 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID) {
4733 if (unlikely(btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))) {
4735 "extent tree v2 doesn't support snapshot deletion yet");
4738 return btrfs_delete_subvolume(BTRFS_I(dir), dentry);
4741 err = fscrypt_setup_filename(dir, &dentry->d_name, 1, &fname);
4745 /* This needs to handle no-key deletions later on */
4747 trans = __unlink_start_trans(BTRFS_I(dir));
4748 if (IS_ERR(trans)) {
4749 err = PTR_ERR(trans);
4753 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4754 err = btrfs_unlink_subvol(trans, BTRFS_I(dir), dentry);
4758 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4762 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4764 /* now the directory is empty */
4765 err = btrfs_unlink_inode(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4768 btrfs_i_size_write(BTRFS_I(inode), 0);
4770 * Propagate the last_unlink_trans value of the deleted dir to
4771 * its parent directory. This is to prevent an unrecoverable
4772 * log tree in the case we do something like this:
4774 * 2) create snapshot under dir foo
4775 * 3) delete the snapshot
4778 * 6) fsync foo or some file inside foo
4780 if (last_unlink_trans >= trans->transid)
4781 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4784 btrfs_end_transaction(trans);
4786 btrfs_btree_balance_dirty(fs_info);
4787 fscrypt_free_filename(&fname);
4793 * btrfs_truncate_block - read, zero a chunk and write a block
4794 * @inode - inode that we're zeroing
4795 * @from - the offset to start zeroing
4796 * @len - the length to zero, 0 to zero the entire range respective to the
4798 * @front - zero up to the offset instead of from the offset on
4800 * This will find the block for the "from" offset and cow the block and zero the
4801 * part we want to zero. This is used with truncate and hole punching.
4803 int btrfs_truncate_block(struct btrfs_inode *inode, loff_t from, loff_t len,
4806 struct btrfs_fs_info *fs_info = inode->root->fs_info;
4807 struct address_space *mapping = inode->vfs_inode.i_mapping;
4808 struct extent_io_tree *io_tree = &inode->io_tree;
4809 struct btrfs_ordered_extent *ordered;
4810 struct extent_state *cached_state = NULL;
4811 struct extent_changeset *data_reserved = NULL;
4812 bool only_release_metadata = false;
4813 u32 blocksize = fs_info->sectorsize;
4814 pgoff_t index = from >> PAGE_SHIFT;
4815 unsigned offset = from & (blocksize - 1);
4817 gfp_t mask = btrfs_alloc_write_mask(mapping);
4818 size_t write_bytes = blocksize;
4823 if (IS_ALIGNED(offset, blocksize) &&
4824 (!len || IS_ALIGNED(len, blocksize)))
4827 block_start = round_down(from, blocksize);
4828 block_end = block_start + blocksize - 1;
4830 ret = btrfs_check_data_free_space(inode, &data_reserved, block_start,
4833 if (btrfs_check_nocow_lock(inode, block_start, &write_bytes, false) > 0) {
4834 /* For nocow case, no need to reserve data space */
4835 only_release_metadata = true;
4840 ret = btrfs_delalloc_reserve_metadata(inode, blocksize, blocksize, false);
4842 if (!only_release_metadata)
4843 btrfs_free_reserved_data_space(inode, data_reserved,
4844 block_start, blocksize);
4848 page = find_or_create_page(mapping, index, mask);
4850 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4852 btrfs_delalloc_release_extents(inode, blocksize);
4857 if (!PageUptodate(page)) {
4858 ret = btrfs_read_folio(NULL, page_folio(page));
4860 if (page->mapping != mapping) {
4865 if (!PageUptodate(page)) {
4872 * We unlock the page after the io is completed and then re-lock it
4873 * above. release_folio() could have come in between that and cleared
4874 * PagePrivate(), but left the page in the mapping. Set the page mapped
4875 * here to make sure it's properly set for the subpage stuff.
4877 ret = set_page_extent_mapped(page);
4881 wait_on_page_writeback(page);
4883 lock_extent(io_tree, block_start, block_end, &cached_state);
4885 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4887 unlock_extent(io_tree, block_start, block_end, &cached_state);
4890 btrfs_start_ordered_extent(ordered);
4891 btrfs_put_ordered_extent(ordered);
4895 clear_extent_bit(&inode->io_tree, block_start, block_end,
4896 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4899 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4902 unlock_extent(io_tree, block_start, block_end, &cached_state);
4906 if (offset != blocksize) {
4908 len = blocksize - offset;
4910 memzero_page(page, (block_start - page_offset(page)),
4913 memzero_page(page, (block_start - page_offset(page)) + offset,
4916 btrfs_page_clear_checked(fs_info, page, block_start,
4917 block_end + 1 - block_start);
4918 btrfs_page_set_dirty(fs_info, page, block_start, block_end + 1 - block_start);
4919 unlock_extent(io_tree, block_start, block_end, &cached_state);
4921 if (only_release_metadata)
4922 set_extent_bit(&inode->io_tree, block_start, block_end,
4923 EXTENT_NORESERVE, NULL);
4927 if (only_release_metadata)
4928 btrfs_delalloc_release_metadata(inode, blocksize, true);
4930 btrfs_delalloc_release_space(inode, data_reserved,
4931 block_start, blocksize, true);
4933 btrfs_delalloc_release_extents(inode, blocksize);
4937 if (only_release_metadata)
4938 btrfs_check_nocow_unlock(inode);
4939 extent_changeset_free(data_reserved);
4943 static int maybe_insert_hole(struct btrfs_root *root, struct btrfs_inode *inode,
4944 u64 offset, u64 len)
4946 struct btrfs_fs_info *fs_info = root->fs_info;
4947 struct btrfs_trans_handle *trans;
4948 struct btrfs_drop_extents_args drop_args = { 0 };
4952 * If NO_HOLES is enabled, we don't need to do anything.
4953 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4954 * or btrfs_update_inode() will be called, which guarantee that the next
4955 * fsync will know this inode was changed and needs to be logged.
4957 if (btrfs_fs_incompat(fs_info, NO_HOLES))
4961 * 1 - for the one we're dropping
4962 * 1 - for the one we're adding
4963 * 1 - for updating the inode.
4965 trans = btrfs_start_transaction(root, 3);
4967 return PTR_ERR(trans);
4969 drop_args.start = offset;
4970 drop_args.end = offset + len;
4971 drop_args.drop_cache = true;
4973 ret = btrfs_drop_extents(trans, root, inode, &drop_args);
4975 btrfs_abort_transaction(trans, ret);
4976 btrfs_end_transaction(trans);
4980 ret = btrfs_insert_hole_extent(trans, root, btrfs_ino(inode), offset, len);
4982 btrfs_abort_transaction(trans, ret);
4984 btrfs_update_inode_bytes(inode, 0, drop_args.bytes_found);
4985 btrfs_update_inode(trans, root, inode);
4987 btrfs_end_transaction(trans);
4992 * This function puts in dummy file extents for the area we're creating a hole
4993 * for. So if we are truncating this file to a larger size we need to insert
4994 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4995 * the range between oldsize and size
4997 int btrfs_cont_expand(struct btrfs_inode *inode, loff_t oldsize, loff_t size)
4999 struct btrfs_root *root = inode->root;
5000 struct btrfs_fs_info *fs_info = root->fs_info;
5001 struct extent_io_tree *io_tree = &inode->io_tree;
5002 struct extent_map *em = NULL;
5003 struct extent_state *cached_state = NULL;
5004 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5005 u64 block_end = ALIGN(size, fs_info->sectorsize);
5012 * If our size started in the middle of a block we need to zero out the
5013 * rest of the block before we expand the i_size, otherwise we could
5014 * expose stale data.
5016 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5020 if (size <= hole_start)
5023 btrfs_lock_and_flush_ordered_range(inode, hole_start, block_end - 1,
5025 cur_offset = hole_start;
5027 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5028 block_end - cur_offset);
5034 last_byte = min(extent_map_end(em), block_end);
5035 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5036 hole_size = last_byte - cur_offset;
5038 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5039 struct extent_map *hole_em;
5041 err = maybe_insert_hole(root, inode, cur_offset,
5046 err = btrfs_inode_set_file_extent_range(inode,
5047 cur_offset, hole_size);
5051 hole_em = alloc_extent_map();
5053 btrfs_drop_extent_map_range(inode, cur_offset,
5054 cur_offset + hole_size - 1,
5056 btrfs_set_inode_full_sync(inode);
5059 hole_em->start = cur_offset;
5060 hole_em->len = hole_size;
5061 hole_em->orig_start = cur_offset;
5063 hole_em->block_start = EXTENT_MAP_HOLE;
5064 hole_em->block_len = 0;
5065 hole_em->orig_block_len = 0;
5066 hole_em->ram_bytes = hole_size;
5067 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5068 hole_em->generation = fs_info->generation;
5070 err = btrfs_replace_extent_map_range(inode, hole_em, true);
5071 free_extent_map(hole_em);
5073 err = btrfs_inode_set_file_extent_range(inode,
5074 cur_offset, hole_size);
5079 free_extent_map(em);
5081 cur_offset = last_byte;
5082 if (cur_offset >= block_end)
5085 free_extent_map(em);
5086 unlock_extent(io_tree, hole_start, block_end - 1, &cached_state);
5090 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5092 struct btrfs_root *root = BTRFS_I(inode)->root;
5093 struct btrfs_trans_handle *trans;
5094 loff_t oldsize = i_size_read(inode);
5095 loff_t newsize = attr->ia_size;
5096 int mask = attr->ia_valid;
5100 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5101 * special case where we need to update the times despite not having
5102 * these flags set. For all other operations the VFS set these flags
5103 * explicitly if it wants a timestamp update.
5105 if (newsize != oldsize) {
5106 inode_inc_iversion(inode);
5107 if (!(mask & (ATTR_CTIME | ATTR_MTIME))) {
5108 inode->i_mtime = current_time(inode);
5109 inode->i_ctime = inode->i_mtime;
5113 if (newsize > oldsize) {
5115 * Don't do an expanding truncate while snapshotting is ongoing.
5116 * This is to ensure the snapshot captures a fully consistent
5117 * state of this file - if the snapshot captures this expanding
5118 * truncation, it must capture all writes that happened before
5121 btrfs_drew_write_lock(&root->snapshot_lock);
5122 ret = btrfs_cont_expand(BTRFS_I(inode), oldsize, newsize);
5124 btrfs_drew_write_unlock(&root->snapshot_lock);
5128 trans = btrfs_start_transaction(root, 1);
5129 if (IS_ERR(trans)) {
5130 btrfs_drew_write_unlock(&root->snapshot_lock);
5131 return PTR_ERR(trans);
5134 i_size_write(inode, newsize);
5135 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
5136 pagecache_isize_extended(inode, oldsize, newsize);
5137 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
5138 btrfs_drew_write_unlock(&root->snapshot_lock);
5139 btrfs_end_transaction(trans);
5141 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5143 if (btrfs_is_zoned(fs_info)) {
5144 ret = btrfs_wait_ordered_range(inode,
5145 ALIGN(newsize, fs_info->sectorsize),
5152 * We're truncating a file that used to have good data down to
5153 * zero. Make sure any new writes to the file get on disk
5157 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE,
5158 &BTRFS_I(inode)->runtime_flags);
5160 truncate_setsize(inode, newsize);
5162 inode_dio_wait(inode);
5164 ret = btrfs_truncate(BTRFS_I(inode), newsize == oldsize);
5165 if (ret && inode->i_nlink) {
5169 * Truncate failed, so fix up the in-memory size. We
5170 * adjusted disk_i_size down as we removed extents, so
5171 * wait for disk_i_size to be stable and then update the
5172 * in-memory size to match.
5174 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5177 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5184 static int btrfs_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
5187 struct inode *inode = d_inode(dentry);
5188 struct btrfs_root *root = BTRFS_I(inode)->root;
5191 if (btrfs_root_readonly(root))
5194 err = setattr_prepare(idmap, dentry, attr);
5198 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5199 err = btrfs_setsize(inode, attr);
5204 if (attr->ia_valid) {
5205 setattr_copy(idmap, inode, attr);
5206 inode_inc_iversion(inode);
5207 err = btrfs_dirty_inode(BTRFS_I(inode));
5209 if (!err && attr->ia_valid & ATTR_MODE)
5210 err = posix_acl_chmod(idmap, dentry, inode->i_mode);
5217 * While truncating the inode pages during eviction, we get the VFS
5218 * calling btrfs_invalidate_folio() against each folio of the inode. This
5219 * is slow because the calls to btrfs_invalidate_folio() result in a
5220 * huge amount of calls to lock_extent() and clear_extent_bit(),
5221 * which keep merging and splitting extent_state structures over and over,
5222 * wasting lots of time.
5224 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5225 * skip all those expensive operations on a per folio basis and do only
5226 * the ordered io finishing, while we release here the extent_map and
5227 * extent_state structures, without the excessive merging and splitting.
5229 static void evict_inode_truncate_pages(struct inode *inode)
5231 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5232 struct rb_node *node;
5234 ASSERT(inode->i_state & I_FREEING);
5235 truncate_inode_pages_final(&inode->i_data);
5237 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
5240 * Keep looping until we have no more ranges in the io tree.
5241 * We can have ongoing bios started by readahead that have
5242 * their endio callback (extent_io.c:end_bio_extent_readpage)
5243 * still in progress (unlocked the pages in the bio but did not yet
5244 * unlocked the ranges in the io tree). Therefore this means some
5245 * ranges can still be locked and eviction started because before
5246 * submitting those bios, which are executed by a separate task (work
5247 * queue kthread), inode references (inode->i_count) were not taken
5248 * (which would be dropped in the end io callback of each bio).
5249 * Therefore here we effectively end up waiting for those bios and
5250 * anyone else holding locked ranges without having bumped the inode's
5251 * reference count - if we don't do it, when they access the inode's
5252 * io_tree to unlock a range it may be too late, leading to an
5253 * use-after-free issue.
5255 spin_lock(&io_tree->lock);
5256 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5257 struct extent_state *state;
5258 struct extent_state *cached_state = NULL;
5261 unsigned state_flags;
5263 node = rb_first(&io_tree->state);
5264 state = rb_entry(node, struct extent_state, rb_node);
5265 start = state->start;
5267 state_flags = state->state;
5268 spin_unlock(&io_tree->lock);
5270 lock_extent(io_tree, start, end, &cached_state);
5273 * If still has DELALLOC flag, the extent didn't reach disk,
5274 * and its reserved space won't be freed by delayed_ref.
5275 * So we need to free its reserved space here.
5276 * (Refer to comment in btrfs_invalidate_folio, case 2)
5278 * Note, end is the bytenr of last byte, so we need + 1 here.
5280 if (state_flags & EXTENT_DELALLOC)
5281 btrfs_qgroup_free_data(BTRFS_I(inode), NULL, start,
5284 clear_extent_bit(io_tree, start, end,
5285 EXTENT_CLEAR_ALL_BITS | EXTENT_DO_ACCOUNTING,
5289 spin_lock(&io_tree->lock);
5291 spin_unlock(&io_tree->lock);
5294 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5295 struct btrfs_block_rsv *rsv)
5297 struct btrfs_fs_info *fs_info = root->fs_info;
5298 struct btrfs_trans_handle *trans;
5299 u64 delayed_refs_extra = btrfs_calc_delayed_ref_bytes(fs_info, 1);
5303 * Eviction should be taking place at some place safe because of our
5304 * delayed iputs. However the normal flushing code will run delayed
5305 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5307 * We reserve the delayed_refs_extra here again because we can't use
5308 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5309 * above. We reserve our extra bit here because we generate a ton of
5310 * delayed refs activity by truncating.
5312 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5313 * if we fail to make this reservation we can re-try without the
5314 * delayed_refs_extra so we can make some forward progress.
5316 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size + delayed_refs_extra,
5317 BTRFS_RESERVE_FLUSH_EVICT);
5319 ret = btrfs_block_rsv_refill(fs_info, rsv, rsv->size,
5320 BTRFS_RESERVE_FLUSH_EVICT);
5323 "could not allocate space for delete; will truncate on mount");
5324 return ERR_PTR(-ENOSPC);
5326 delayed_refs_extra = 0;
5329 trans = btrfs_join_transaction(root);
5333 if (delayed_refs_extra) {
5334 trans->block_rsv = &fs_info->trans_block_rsv;
5335 trans->bytes_reserved = delayed_refs_extra;
5336 btrfs_block_rsv_migrate(rsv, trans->block_rsv,
5337 delayed_refs_extra, true);
5342 void btrfs_evict_inode(struct inode *inode)
5344 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5345 struct btrfs_trans_handle *trans;
5346 struct btrfs_root *root = BTRFS_I(inode)->root;
5347 struct btrfs_block_rsv *rsv = NULL;
5350 trace_btrfs_inode_evict(inode);
5353 fsverity_cleanup_inode(inode);
5358 evict_inode_truncate_pages(inode);
5360 if (inode->i_nlink &&
5361 ((btrfs_root_refs(&root->root_item) != 0 &&
5362 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5363 btrfs_is_free_space_inode(BTRFS_I(inode))))
5366 if (is_bad_inode(inode))
5369 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5372 if (inode->i_nlink > 0) {
5373 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5374 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5379 * This makes sure the inode item in tree is uptodate and the space for
5380 * the inode update is released.
5382 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5387 * This drops any pending insert or delete operations we have for this
5388 * inode. We could have a delayed dir index deletion queued up, but
5389 * we're removing the inode completely so that'll be taken care of in
5392 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
5394 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5397 rsv->size = btrfs_calc_metadata_size(fs_info, 1);
5398 rsv->failfast = true;
5400 btrfs_i_size_write(BTRFS_I(inode), 0);
5403 struct btrfs_truncate_control control = {
5404 .inode = BTRFS_I(inode),
5405 .ino = btrfs_ino(BTRFS_I(inode)),
5410 trans = evict_refill_and_join(root, rsv);
5414 trans->block_rsv = rsv;
5416 ret = btrfs_truncate_inode_items(trans, root, &control);
5417 trans->block_rsv = &fs_info->trans_block_rsv;
5418 btrfs_end_transaction(trans);
5420 * We have not added new delayed items for our inode after we
5421 * have flushed its delayed items, so no need to throttle on
5422 * delayed items. However we have modified extent buffers.
5424 btrfs_btree_balance_dirty_nodelay(fs_info);
5425 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5432 * Errors here aren't a big deal, it just means we leave orphan items in
5433 * the tree. They will be cleaned up on the next mount. If the inode
5434 * number gets reused, cleanup deletes the orphan item without doing
5435 * anything, and unlink reuses the existing orphan item.
5437 * If it turns out that we are dropping too many of these, we might want
5438 * to add a mechanism for retrying these after a commit.
5440 trans = evict_refill_and_join(root, rsv);
5441 if (!IS_ERR(trans)) {
5442 trans->block_rsv = rsv;
5443 btrfs_orphan_del(trans, BTRFS_I(inode));
5444 trans->block_rsv = &fs_info->trans_block_rsv;
5445 btrfs_end_transaction(trans);
5449 btrfs_free_block_rsv(fs_info, rsv);
5451 * If we didn't successfully delete, the orphan item will still be in
5452 * the tree and we'll retry on the next mount. Again, we might also want
5453 * to retry these periodically in the future.
5455 btrfs_remove_delayed_node(BTRFS_I(inode));
5456 fsverity_cleanup_inode(inode);
5461 * Return the key found in the dir entry in the location pointer, fill @type
5462 * with BTRFS_FT_*, and return 0.
5464 * If no dir entries were found, returns -ENOENT.
5465 * If found a corrupted location in dir entry, returns -EUCLEAN.
5467 static int btrfs_inode_by_name(struct btrfs_inode *dir, struct dentry *dentry,
5468 struct btrfs_key *location, u8 *type)
5470 struct btrfs_dir_item *di;
5471 struct btrfs_path *path;
5472 struct btrfs_root *root = dir->root;
5474 struct fscrypt_name fname;
5476 path = btrfs_alloc_path();
5480 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 1, &fname);
5484 * fscrypt_setup_filename() should never return a positive value, but
5485 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5489 /* This needs to handle no-key deletions later on */
5491 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir),
5492 &fname.disk_name, 0);
5493 if (IS_ERR_OR_NULL(di)) {
5494 ret = di ? PTR_ERR(di) : -ENOENT;
5498 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5499 if (location->type != BTRFS_INODE_ITEM_KEY &&
5500 location->type != BTRFS_ROOT_ITEM_KEY) {
5502 btrfs_warn(root->fs_info,
5503 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5504 __func__, fname.disk_name.name, btrfs_ino(dir),
5505 location->objectid, location->type, location->offset);
5508 *type = btrfs_dir_ftype(path->nodes[0], di);
5510 fscrypt_free_filename(&fname);
5511 btrfs_free_path(path);
5516 * when we hit a tree root in a directory, the btrfs part of the inode
5517 * needs to be changed to reflect the root directory of the tree root. This
5518 * is kind of like crossing a mount point.
5520 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5521 struct btrfs_inode *dir,
5522 struct dentry *dentry,
5523 struct btrfs_key *location,
5524 struct btrfs_root **sub_root)
5526 struct btrfs_path *path;
5527 struct btrfs_root *new_root;
5528 struct btrfs_root_ref *ref;
5529 struct extent_buffer *leaf;
5530 struct btrfs_key key;
5533 struct fscrypt_name fname;
5535 ret = fscrypt_setup_filename(&dir->vfs_inode, &dentry->d_name, 0, &fname);
5539 path = btrfs_alloc_path();
5546 key.objectid = dir->root->root_key.objectid;
5547 key.type = BTRFS_ROOT_REF_KEY;
5548 key.offset = location->objectid;
5550 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5557 leaf = path->nodes[0];
5558 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5559 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5560 btrfs_root_ref_name_len(leaf, ref) != fname.disk_name.len)
5563 ret = memcmp_extent_buffer(leaf, fname.disk_name.name,
5564 (unsigned long)(ref + 1), fname.disk_name.len);
5568 btrfs_release_path(path);
5570 new_root = btrfs_get_fs_root(fs_info, location->objectid, true);
5571 if (IS_ERR(new_root)) {
5572 err = PTR_ERR(new_root);
5576 *sub_root = new_root;
5577 location->objectid = btrfs_root_dirid(&new_root->root_item);
5578 location->type = BTRFS_INODE_ITEM_KEY;
5579 location->offset = 0;
5582 btrfs_free_path(path);
5583 fscrypt_free_filename(&fname);
5587 static void inode_tree_add(struct btrfs_inode *inode)
5589 struct btrfs_root *root = inode->root;
5590 struct btrfs_inode *entry;
5592 struct rb_node *parent;
5593 struct rb_node *new = &inode->rb_node;
5594 u64 ino = btrfs_ino(inode);
5596 if (inode_unhashed(&inode->vfs_inode))
5599 spin_lock(&root->inode_lock);
5600 p = &root->inode_tree.rb_node;
5603 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5605 if (ino < btrfs_ino(entry))
5606 p = &parent->rb_left;
5607 else if (ino > btrfs_ino(entry))
5608 p = &parent->rb_right;
5610 WARN_ON(!(entry->vfs_inode.i_state &
5611 (I_WILL_FREE | I_FREEING)));
5612 rb_replace_node(parent, new, &root->inode_tree);
5613 RB_CLEAR_NODE(parent);
5614 spin_unlock(&root->inode_lock);
5618 rb_link_node(new, parent, p);
5619 rb_insert_color(new, &root->inode_tree);
5620 spin_unlock(&root->inode_lock);
5623 static void inode_tree_del(struct btrfs_inode *inode)
5625 struct btrfs_root *root = inode->root;
5628 spin_lock(&root->inode_lock);
5629 if (!RB_EMPTY_NODE(&inode->rb_node)) {
5630 rb_erase(&inode->rb_node, &root->inode_tree);
5631 RB_CLEAR_NODE(&inode->rb_node);
5632 empty = RB_EMPTY_ROOT(&root->inode_tree);
5634 spin_unlock(&root->inode_lock);
5636 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5637 spin_lock(&root->inode_lock);
5638 empty = RB_EMPTY_ROOT(&root->inode_tree);
5639 spin_unlock(&root->inode_lock);
5641 btrfs_add_dead_root(root);
5646 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5648 struct btrfs_iget_args *args = p;
5650 inode->i_ino = args->ino;
5651 BTRFS_I(inode)->location.objectid = args->ino;
5652 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
5653 BTRFS_I(inode)->location.offset = 0;
5654 BTRFS_I(inode)->root = btrfs_grab_root(args->root);
5655 BUG_ON(args->root && !BTRFS_I(inode)->root);
5657 if (args->root && args->root == args->root->fs_info->tree_root &&
5658 args->ino != BTRFS_BTREE_INODE_OBJECTID)
5659 set_bit(BTRFS_INODE_FREE_SPACE_INODE,
5660 &BTRFS_I(inode)->runtime_flags);
5664 static int btrfs_find_actor(struct inode *inode, void *opaque)
5666 struct btrfs_iget_args *args = opaque;
5668 return args->ino == BTRFS_I(inode)->location.objectid &&
5669 args->root == BTRFS_I(inode)->root;
5672 static struct inode *btrfs_iget_locked(struct super_block *s, u64 ino,
5673 struct btrfs_root *root)
5675 struct inode *inode;
5676 struct btrfs_iget_args args;
5677 unsigned long hashval = btrfs_inode_hash(ino, root);
5682 inode = iget5_locked(s, hashval, btrfs_find_actor,
5683 btrfs_init_locked_inode,
5689 * Get an inode object given its inode number and corresponding root.
5690 * Path can be preallocated to prevent recursing back to iget through
5691 * allocator. NULL is also valid but may require an additional allocation
5694 struct inode *btrfs_iget_path(struct super_block *s, u64 ino,
5695 struct btrfs_root *root, struct btrfs_path *path)
5697 struct inode *inode;
5699 inode = btrfs_iget_locked(s, ino, root);
5701 return ERR_PTR(-ENOMEM);
5703 if (inode->i_state & I_NEW) {
5706 ret = btrfs_read_locked_inode(inode, path);
5708 inode_tree_add(BTRFS_I(inode));
5709 unlock_new_inode(inode);
5713 * ret > 0 can come from btrfs_search_slot called by
5714 * btrfs_read_locked_inode, this means the inode item
5719 inode = ERR_PTR(ret);
5726 struct inode *btrfs_iget(struct super_block *s, u64 ino, struct btrfs_root *root)
5728 return btrfs_iget_path(s, ino, root, NULL);
5731 static struct inode *new_simple_dir(struct super_block *s,
5732 struct btrfs_key *key,
5733 struct btrfs_root *root)
5735 struct inode *inode = new_inode(s);
5738 return ERR_PTR(-ENOMEM);
5740 BTRFS_I(inode)->root = btrfs_grab_root(root);
5741 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5742 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5744 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5746 * We only need lookup, the rest is read-only and there's no inode
5747 * associated with the dentry
5749 inode->i_op = &simple_dir_inode_operations;
5750 inode->i_opflags &= ~IOP_XATTR;
5751 inode->i_fop = &simple_dir_operations;
5752 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5753 inode->i_mtime = current_time(inode);
5754 inode->i_atime = inode->i_mtime;
5755 inode->i_ctime = inode->i_mtime;
5756 BTRFS_I(inode)->i_otime = inode->i_mtime;
5761 static_assert(BTRFS_FT_UNKNOWN == FT_UNKNOWN);
5762 static_assert(BTRFS_FT_REG_FILE == FT_REG_FILE);
5763 static_assert(BTRFS_FT_DIR == FT_DIR);
5764 static_assert(BTRFS_FT_CHRDEV == FT_CHRDEV);
5765 static_assert(BTRFS_FT_BLKDEV == FT_BLKDEV);
5766 static_assert(BTRFS_FT_FIFO == FT_FIFO);
5767 static_assert(BTRFS_FT_SOCK == FT_SOCK);
5768 static_assert(BTRFS_FT_SYMLINK == FT_SYMLINK);
5770 static inline u8 btrfs_inode_type(struct inode *inode)
5772 return fs_umode_to_ftype(inode->i_mode);
5775 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5777 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5778 struct inode *inode;
5779 struct btrfs_root *root = BTRFS_I(dir)->root;
5780 struct btrfs_root *sub_root = root;
5781 struct btrfs_key location;
5785 if (dentry->d_name.len > BTRFS_NAME_LEN)
5786 return ERR_PTR(-ENAMETOOLONG);
5788 ret = btrfs_inode_by_name(BTRFS_I(dir), dentry, &location, &di_type);
5790 return ERR_PTR(ret);
5792 if (location.type == BTRFS_INODE_ITEM_KEY) {
5793 inode = btrfs_iget(dir->i_sb, location.objectid, root);
5797 /* Do extra check against inode mode with di_type */
5798 if (btrfs_inode_type(inode) != di_type) {
5800 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5801 inode->i_mode, btrfs_inode_type(inode),
5804 return ERR_PTR(-EUCLEAN);
5809 ret = fixup_tree_root_location(fs_info, BTRFS_I(dir), dentry,
5810 &location, &sub_root);
5813 inode = ERR_PTR(ret);
5815 inode = new_simple_dir(dir->i_sb, &location, root);
5817 inode = btrfs_iget(dir->i_sb, location.objectid, sub_root);
5818 btrfs_put_root(sub_root);
5823 down_read(&fs_info->cleanup_work_sem);
5824 if (!sb_rdonly(inode->i_sb))
5825 ret = btrfs_orphan_cleanup(sub_root);
5826 up_read(&fs_info->cleanup_work_sem);
5829 inode = ERR_PTR(ret);
5836 static int btrfs_dentry_delete(const struct dentry *dentry)
5838 struct btrfs_root *root;
5839 struct inode *inode = d_inode(dentry);
5841 if (!inode && !IS_ROOT(dentry))
5842 inode = d_inode(dentry->d_parent);
5845 root = BTRFS_I(inode)->root;
5846 if (btrfs_root_refs(&root->root_item) == 0)
5849 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5855 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5858 struct inode *inode = btrfs_lookup_dentry(dir, dentry);
5860 if (inode == ERR_PTR(-ENOENT))
5862 return d_splice_alias(inode, dentry);
5866 * Find the highest existing sequence number in a directory and then set the
5867 * in-memory index_cnt variable to the first free sequence number.
5869 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5871 struct btrfs_root *root = inode->root;
5872 struct btrfs_key key, found_key;
5873 struct btrfs_path *path;
5874 struct extent_buffer *leaf;
5877 key.objectid = btrfs_ino(inode);
5878 key.type = BTRFS_DIR_INDEX_KEY;
5879 key.offset = (u64)-1;
5881 path = btrfs_alloc_path();
5885 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5888 /* FIXME: we should be able to handle this */
5893 if (path->slots[0] == 0) {
5894 inode->index_cnt = BTRFS_DIR_START_INDEX;
5900 leaf = path->nodes[0];
5901 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5903 if (found_key.objectid != btrfs_ino(inode) ||
5904 found_key.type != BTRFS_DIR_INDEX_KEY) {
5905 inode->index_cnt = BTRFS_DIR_START_INDEX;
5909 inode->index_cnt = found_key.offset + 1;
5911 btrfs_free_path(path);
5915 static int btrfs_get_dir_last_index(struct btrfs_inode *dir, u64 *index)
5917 if (dir->index_cnt == (u64)-1) {
5920 ret = btrfs_inode_delayed_dir_index_count(dir);
5922 ret = btrfs_set_inode_index_count(dir);
5928 *index = dir->index_cnt;
5934 * All this infrastructure exists because dir_emit can fault, and we are holding
5935 * the tree lock when doing readdir. For now just allocate a buffer and copy
5936 * our information into that, and then dir_emit from the buffer. This is
5937 * similar to what NFS does, only we don't keep the buffer around in pagecache
5938 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5939 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5942 static int btrfs_opendir(struct inode *inode, struct file *file)
5944 struct btrfs_file_private *private;
5948 ret = btrfs_get_dir_last_index(BTRFS_I(inode), &last_index);
5952 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5955 private->last_index = last_index;
5956 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5957 if (!private->filldir_buf) {
5961 file->private_data = private;
5972 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5975 struct dir_entry *entry = addr;
5976 char *name = (char *)(entry + 1);
5978 ctx->pos = get_unaligned(&entry->offset);
5979 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5980 get_unaligned(&entry->ino),
5981 get_unaligned(&entry->type)))
5983 addr += sizeof(struct dir_entry) +
5984 get_unaligned(&entry->name_len);
5990 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5992 struct inode *inode = file_inode(file);
5993 struct btrfs_root *root = BTRFS_I(inode)->root;
5994 struct btrfs_file_private *private = file->private_data;
5995 struct btrfs_dir_item *di;
5996 struct btrfs_key key;
5997 struct btrfs_key found_key;
5998 struct btrfs_path *path;
6000 struct list_head ins_list;
6001 struct list_head del_list;
6008 struct btrfs_key location;
6010 if (!dir_emit_dots(file, ctx))
6013 path = btrfs_alloc_path();
6017 addr = private->filldir_buf;
6018 path->reada = READA_FORWARD;
6020 INIT_LIST_HEAD(&ins_list);
6021 INIT_LIST_HEAD(&del_list);
6022 put = btrfs_readdir_get_delayed_items(inode, private->last_index,
6023 &ins_list, &del_list);
6026 key.type = BTRFS_DIR_INDEX_KEY;
6027 key.offset = ctx->pos;
6028 key.objectid = btrfs_ino(BTRFS_I(inode));
6030 btrfs_for_each_slot(root, &key, &found_key, path, ret) {
6031 struct dir_entry *entry;
6032 struct extent_buffer *leaf = path->nodes[0];
6035 if (found_key.objectid != key.objectid)
6037 if (found_key.type != BTRFS_DIR_INDEX_KEY)
6039 if (found_key.offset < ctx->pos)
6041 if (found_key.offset > private->last_index)
6043 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
6045 di = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dir_item);
6046 name_len = btrfs_dir_name_len(leaf, di);
6047 if ((total_len + sizeof(struct dir_entry) + name_len) >=
6049 btrfs_release_path(path);
6050 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6053 addr = private->filldir_buf;
6059 ftype = btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf, di));
6061 name_ptr = (char *)(entry + 1);
6062 read_extent_buffer(leaf, name_ptr,
6063 (unsigned long)(di + 1), name_len);
6064 put_unaligned(name_len, &entry->name_len);
6065 put_unaligned(fs_ftype_to_dtype(ftype), &entry->type);
6066 btrfs_dir_item_key_to_cpu(leaf, di, &location);
6067 put_unaligned(location.objectid, &entry->ino);
6068 put_unaligned(found_key.offset, &entry->offset);
6070 addr += sizeof(struct dir_entry) + name_len;
6071 total_len += sizeof(struct dir_entry) + name_len;
6073 /* Catch error encountered during iteration */
6077 btrfs_release_path(path);
6079 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
6083 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6088 * Stop new entries from being returned after we return the last
6091 * New directory entries are assigned a strictly increasing
6092 * offset. This means that new entries created during readdir
6093 * are *guaranteed* to be seen in the future by that readdir.
6094 * This has broken buggy programs which operate on names as
6095 * they're returned by readdir. Until we re-use freed offsets
6096 * we have this hack to stop new entries from being returned
6097 * under the assumption that they'll never reach this huge
6100 * This is being careful not to overflow 32bit loff_t unless the
6101 * last entry requires it because doing so has broken 32bit apps
6104 if (ctx->pos >= INT_MAX)
6105 ctx->pos = LLONG_MAX;
6112 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6113 btrfs_free_path(path);
6118 * This is somewhat expensive, updating the tree every time the
6119 * inode changes. But, it is most likely to find the inode in cache.
6120 * FIXME, needs more benchmarking...there are no reasons other than performance
6121 * to keep or drop this code.
6123 static int btrfs_dirty_inode(struct btrfs_inode *inode)
6125 struct btrfs_root *root = inode->root;
6126 struct btrfs_fs_info *fs_info = root->fs_info;
6127 struct btrfs_trans_handle *trans;
6130 if (test_bit(BTRFS_INODE_DUMMY, &inode->runtime_flags))
6133 trans = btrfs_join_transaction(root);
6135 return PTR_ERR(trans);
6137 ret = btrfs_update_inode(trans, root, inode);
6138 if (ret && (ret == -ENOSPC || ret == -EDQUOT)) {
6139 /* whoops, lets try again with the full transaction */
6140 btrfs_end_transaction(trans);
6141 trans = btrfs_start_transaction(root, 1);
6143 return PTR_ERR(trans);
6145 ret = btrfs_update_inode(trans, root, inode);
6147 btrfs_end_transaction(trans);
6148 if (inode->delayed_node)
6149 btrfs_balance_delayed_items(fs_info);
6155 * This is a copy of file_update_time. We need this so we can return error on
6156 * ENOSPC for updating the inode in the case of file write and mmap writes.
6158 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6161 struct btrfs_root *root = BTRFS_I(inode)->root;
6162 bool dirty = flags & ~S_VERSION;
6164 if (btrfs_root_readonly(root))
6167 if (flags & S_VERSION)
6168 dirty |= inode_maybe_inc_iversion(inode, dirty);
6169 if (flags & S_CTIME)
6170 inode->i_ctime = *now;
6171 if (flags & S_MTIME)
6172 inode->i_mtime = *now;
6173 if (flags & S_ATIME)
6174 inode->i_atime = *now;
6175 return dirty ? btrfs_dirty_inode(BTRFS_I(inode)) : 0;
6179 * helper to find a free sequence number in a given directory. This current
6180 * code is very simple, later versions will do smarter things in the btree
6182 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6186 if (dir->index_cnt == (u64)-1) {
6187 ret = btrfs_inode_delayed_dir_index_count(dir);
6189 ret = btrfs_set_inode_index_count(dir);
6195 *index = dir->index_cnt;
6201 static int btrfs_insert_inode_locked(struct inode *inode)
6203 struct btrfs_iget_args args;
6205 args.ino = BTRFS_I(inode)->location.objectid;
6206 args.root = BTRFS_I(inode)->root;
6208 return insert_inode_locked4(inode,
6209 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6210 btrfs_find_actor, &args);
6213 int btrfs_new_inode_prepare(struct btrfs_new_inode_args *args,
6214 unsigned int *trans_num_items)
6216 struct inode *dir = args->dir;
6217 struct inode *inode = args->inode;
6220 if (!args->orphan) {
6221 ret = fscrypt_setup_filename(dir, &args->dentry->d_name, 0,
6227 ret = posix_acl_create(dir, &inode->i_mode, &args->default_acl, &args->acl);
6229 fscrypt_free_filename(&args->fname);
6233 /* 1 to add inode item */
6234 *trans_num_items = 1;
6235 /* 1 to add compression property */
6236 if (BTRFS_I(dir)->prop_compress)
6237 (*trans_num_items)++;
6238 /* 1 to add default ACL xattr */
6239 if (args->default_acl)
6240 (*trans_num_items)++;
6241 /* 1 to add access ACL xattr */
6243 (*trans_num_items)++;
6244 #ifdef CONFIG_SECURITY
6245 /* 1 to add LSM xattr */
6246 if (dir->i_security)
6247 (*trans_num_items)++;
6250 /* 1 to add orphan item */
6251 (*trans_num_items)++;
6255 * 1 to add dir index
6256 * 1 to update parent inode item
6258 * No need for 1 unit for the inode ref item because it is
6259 * inserted in a batch together with the inode item at
6260 * btrfs_create_new_inode().
6262 *trans_num_items += 3;
6267 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args *args)
6269 posix_acl_release(args->acl);
6270 posix_acl_release(args->default_acl);
6271 fscrypt_free_filename(&args->fname);
6275 * Inherit flags from the parent inode.
6277 * Currently only the compression flags and the cow flags are inherited.
6279 static void btrfs_inherit_iflags(struct btrfs_inode *inode, struct btrfs_inode *dir)
6285 if (flags & BTRFS_INODE_NOCOMPRESS) {
6286 inode->flags &= ~BTRFS_INODE_COMPRESS;
6287 inode->flags |= BTRFS_INODE_NOCOMPRESS;
6288 } else if (flags & BTRFS_INODE_COMPRESS) {
6289 inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
6290 inode->flags |= BTRFS_INODE_COMPRESS;
6293 if (flags & BTRFS_INODE_NODATACOW) {
6294 inode->flags |= BTRFS_INODE_NODATACOW;
6295 if (S_ISREG(inode->vfs_inode.i_mode))
6296 inode->flags |= BTRFS_INODE_NODATASUM;
6299 btrfs_sync_inode_flags_to_i_flags(&inode->vfs_inode);
6302 int btrfs_create_new_inode(struct btrfs_trans_handle *trans,
6303 struct btrfs_new_inode_args *args)
6305 struct inode *dir = args->dir;
6306 struct inode *inode = args->inode;
6307 const struct fscrypt_str *name = args->orphan ? NULL : &args->fname.disk_name;
6308 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6309 struct btrfs_root *root;
6310 struct btrfs_inode_item *inode_item;
6311 struct btrfs_key *location;
6312 struct btrfs_path *path;
6314 struct btrfs_inode_ref *ref;
6315 struct btrfs_key key[2];
6317 struct btrfs_item_batch batch;
6321 path = btrfs_alloc_path();
6326 BTRFS_I(inode)->root = btrfs_grab_root(BTRFS_I(dir)->root);
6327 root = BTRFS_I(inode)->root;
6329 ret = btrfs_get_free_objectid(root, &objectid);
6332 inode->i_ino = objectid;
6336 * O_TMPFILE, set link count to 0, so that after this point, we
6337 * fill in an inode item with the correct link count.
6339 set_nlink(inode, 0);
6341 trace_btrfs_inode_request(dir);
6343 ret = btrfs_set_inode_index(BTRFS_I(dir), &BTRFS_I(inode)->dir_index);
6347 /* index_cnt is ignored for everything but a dir. */
6348 BTRFS_I(inode)->index_cnt = BTRFS_DIR_START_INDEX;
6349 BTRFS_I(inode)->generation = trans->transid;
6350 inode->i_generation = BTRFS_I(inode)->generation;
6353 * Subvolumes don't inherit flags from their parent directory.
6354 * Originally this was probably by accident, but we probably can't
6355 * change it now without compatibility issues.
6358 btrfs_inherit_iflags(BTRFS_I(inode), BTRFS_I(dir));
6360 if (S_ISREG(inode->i_mode)) {
6361 if (btrfs_test_opt(fs_info, NODATASUM))
6362 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6363 if (btrfs_test_opt(fs_info, NODATACOW))
6364 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6365 BTRFS_INODE_NODATASUM;
6368 location = &BTRFS_I(inode)->location;
6369 location->objectid = objectid;
6370 location->offset = 0;
6371 location->type = BTRFS_INODE_ITEM_KEY;
6373 ret = btrfs_insert_inode_locked(inode);
6376 BTRFS_I(dir)->index_cnt--;
6381 * We could have gotten an inode number from somebody who was fsynced
6382 * and then removed in this same transaction, so let's just set full
6383 * sync since it will be a full sync anyway and this will blow away the
6384 * old info in the log.
6386 btrfs_set_inode_full_sync(BTRFS_I(inode));
6388 key[0].objectid = objectid;
6389 key[0].type = BTRFS_INODE_ITEM_KEY;
6392 sizes[0] = sizeof(struct btrfs_inode_item);
6394 if (!args->orphan) {
6396 * Start new inodes with an inode_ref. This is slightly more
6397 * efficient for small numbers of hard links since they will
6398 * be packed into one item. Extended refs will kick in if we
6399 * add more hard links than can fit in the ref item.
6401 key[1].objectid = objectid;
6402 key[1].type = BTRFS_INODE_REF_KEY;
6404 key[1].offset = objectid;
6405 sizes[1] = 2 + sizeof(*ref);
6407 key[1].offset = btrfs_ino(BTRFS_I(dir));
6408 sizes[1] = name->len + sizeof(*ref);
6412 batch.keys = &key[0];
6413 batch.data_sizes = &sizes[0];
6414 batch.total_data_size = sizes[0] + (args->orphan ? 0 : sizes[1]);
6415 batch.nr = args->orphan ? 1 : 2;
6416 ret = btrfs_insert_empty_items(trans, root, path, &batch);
6418 btrfs_abort_transaction(trans, ret);
6422 inode->i_mtime = current_time(inode);
6423 inode->i_atime = inode->i_mtime;
6424 inode->i_ctime = inode->i_mtime;
6425 BTRFS_I(inode)->i_otime = inode->i_mtime;
6428 * We're going to fill the inode item now, so at this point the inode
6429 * must be fully initialized.
6432 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6433 struct btrfs_inode_item);
6434 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6435 sizeof(*inode_item));
6436 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6438 if (!args->orphan) {
6439 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6440 struct btrfs_inode_ref);
6441 ptr = (unsigned long)(ref + 1);
6443 btrfs_set_inode_ref_name_len(path->nodes[0], ref, 2);
6444 btrfs_set_inode_ref_index(path->nodes[0], ref, 0);
6445 write_extent_buffer(path->nodes[0], "..", ptr, 2);
6447 btrfs_set_inode_ref_name_len(path->nodes[0], ref,
6449 btrfs_set_inode_ref_index(path->nodes[0], ref,
6450 BTRFS_I(inode)->dir_index);
6451 write_extent_buffer(path->nodes[0], name->name, ptr,
6456 btrfs_mark_buffer_dirty(path->nodes[0]);
6458 * We don't need the path anymore, plus inheriting properties, adding
6459 * ACLs, security xattrs, orphan item or adding the link, will result in
6460 * allocating yet another path. So just free our path.
6462 btrfs_free_path(path);
6466 struct inode *parent;
6469 * Subvolumes inherit properties from their parent subvolume,
6470 * not the directory they were created in.
6472 parent = btrfs_iget(fs_info->sb, BTRFS_FIRST_FREE_OBJECTID,
6473 BTRFS_I(dir)->root);
6474 if (IS_ERR(parent)) {
6475 ret = PTR_ERR(parent);
6477 ret = btrfs_inode_inherit_props(trans, inode, parent);
6481 ret = btrfs_inode_inherit_props(trans, inode, dir);
6485 "error inheriting props for ino %llu (root %llu): %d",
6486 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid,
6491 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6494 if (!args->subvol) {
6495 ret = btrfs_init_inode_security(trans, args);
6497 btrfs_abort_transaction(trans, ret);
6502 inode_tree_add(BTRFS_I(inode));
6504 trace_btrfs_inode_new(inode);
6505 btrfs_set_inode_last_trans(trans, BTRFS_I(inode));
6507 btrfs_update_root_times(trans, root);
6510 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
6512 ret = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode), name,
6513 0, BTRFS_I(inode)->dir_index);
6516 btrfs_abort_transaction(trans, ret);
6524 * discard_new_inode() calls iput(), but the caller owns the reference
6528 discard_new_inode(inode);
6530 btrfs_free_path(path);
6535 * utility function to add 'inode' into 'parent_inode' with
6536 * a give name and a given sequence number.
6537 * if 'add_backref' is true, also insert a backref from the
6538 * inode to the parent directory.
6540 int btrfs_add_link(struct btrfs_trans_handle *trans,
6541 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6542 const struct fscrypt_str *name, int add_backref, u64 index)
6545 struct btrfs_key key;
6546 struct btrfs_root *root = parent_inode->root;
6547 u64 ino = btrfs_ino(inode);
6548 u64 parent_ino = btrfs_ino(parent_inode);
6550 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6551 memcpy(&key, &inode->root->root_key, sizeof(key));
6554 key.type = BTRFS_INODE_ITEM_KEY;
6558 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6559 ret = btrfs_add_root_ref(trans, key.objectid,
6560 root->root_key.objectid, parent_ino,
6562 } else if (add_backref) {
6563 ret = btrfs_insert_inode_ref(trans, root, name,
6564 ino, parent_ino, index);
6567 /* Nothing to clean up yet */
6571 ret = btrfs_insert_dir_item(trans, name, parent_inode, &key,
6572 btrfs_inode_type(&inode->vfs_inode), index);
6573 if (ret == -EEXIST || ret == -EOVERFLOW)
6576 btrfs_abort_transaction(trans, ret);
6580 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6582 inode_inc_iversion(&parent_inode->vfs_inode);
6584 * If we are replaying a log tree, we do not want to update the mtime
6585 * and ctime of the parent directory with the current time, since the
6586 * log replay procedure is responsible for setting them to their correct
6587 * values (the ones it had when the fsync was done).
6589 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &root->fs_info->flags)) {
6590 struct timespec64 now = current_time(&parent_inode->vfs_inode);
6592 parent_inode->vfs_inode.i_mtime = now;
6593 parent_inode->vfs_inode.i_ctime = now;
6595 ret = btrfs_update_inode(trans, root, parent_inode);
6597 btrfs_abort_transaction(trans, ret);
6601 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6604 err = btrfs_del_root_ref(trans, key.objectid,
6605 root->root_key.objectid, parent_ino,
6606 &local_index, name);
6608 btrfs_abort_transaction(trans, err);
6609 } else if (add_backref) {
6613 err = btrfs_del_inode_ref(trans, root, name, ino, parent_ino,
6616 btrfs_abort_transaction(trans, err);
6619 /* Return the original error code */
6623 static int btrfs_create_common(struct inode *dir, struct dentry *dentry,
6624 struct inode *inode)
6626 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6627 struct btrfs_root *root = BTRFS_I(dir)->root;
6628 struct btrfs_new_inode_args new_inode_args = {
6633 unsigned int trans_num_items;
6634 struct btrfs_trans_handle *trans;
6637 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
6641 trans = btrfs_start_transaction(root, trans_num_items);
6642 if (IS_ERR(trans)) {
6643 err = PTR_ERR(trans);
6644 goto out_new_inode_args;
6647 err = btrfs_create_new_inode(trans, &new_inode_args);
6649 d_instantiate_new(dentry, inode);
6651 btrfs_end_transaction(trans);
6652 btrfs_btree_balance_dirty(fs_info);
6654 btrfs_new_inode_args_destroy(&new_inode_args);
6661 static int btrfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
6662 struct dentry *dentry, umode_t mode, dev_t rdev)
6664 struct inode *inode;
6666 inode = new_inode(dir->i_sb);
6669 inode_init_owner(idmap, inode, dir, mode);
6670 inode->i_op = &btrfs_special_inode_operations;
6671 init_special_inode(inode, inode->i_mode, rdev);
6672 return btrfs_create_common(dir, dentry, inode);
6675 static int btrfs_create(struct mnt_idmap *idmap, struct inode *dir,
6676 struct dentry *dentry, umode_t mode, bool excl)
6678 struct inode *inode;
6680 inode = new_inode(dir->i_sb);
6683 inode_init_owner(idmap, inode, dir, mode);
6684 inode->i_fop = &btrfs_file_operations;
6685 inode->i_op = &btrfs_file_inode_operations;
6686 inode->i_mapping->a_ops = &btrfs_aops;
6687 return btrfs_create_common(dir, dentry, inode);
6690 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6691 struct dentry *dentry)
6693 struct btrfs_trans_handle *trans = NULL;
6694 struct btrfs_root *root = BTRFS_I(dir)->root;
6695 struct inode *inode = d_inode(old_dentry);
6696 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6697 struct fscrypt_name fname;
6702 /* do not allow sys_link's with other subvols of the same device */
6703 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6706 if (inode->i_nlink >= BTRFS_LINK_MAX)
6709 err = fscrypt_setup_filename(dir, &dentry->d_name, 0, &fname);
6713 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6718 * 2 items for inode and inode ref
6719 * 2 items for dir items
6720 * 1 item for parent inode
6721 * 1 item for orphan item deletion if O_TMPFILE
6723 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6724 if (IS_ERR(trans)) {
6725 err = PTR_ERR(trans);
6730 /* There are several dir indexes for this inode, clear the cache. */
6731 BTRFS_I(inode)->dir_index = 0ULL;
6733 inode_inc_iversion(inode);
6734 inode->i_ctime = current_time(inode);
6736 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6738 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6739 &fname.disk_name, 1, index);
6744 struct dentry *parent = dentry->d_parent;
6746 err = btrfs_update_inode(trans, root, BTRFS_I(inode));
6749 if (inode->i_nlink == 1) {
6751 * If new hard link count is 1, it's a file created
6752 * with open(2) O_TMPFILE flag.
6754 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6758 d_instantiate(dentry, inode);
6759 btrfs_log_new_name(trans, old_dentry, NULL, 0, parent);
6763 fscrypt_free_filename(&fname);
6765 btrfs_end_transaction(trans);
6767 inode_dec_link_count(inode);
6770 btrfs_btree_balance_dirty(fs_info);
6774 static int btrfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
6775 struct dentry *dentry, umode_t mode)
6777 struct inode *inode;
6779 inode = new_inode(dir->i_sb);
6782 inode_init_owner(idmap, inode, dir, S_IFDIR | mode);
6783 inode->i_op = &btrfs_dir_inode_operations;
6784 inode->i_fop = &btrfs_dir_file_operations;
6785 return btrfs_create_common(dir, dentry, inode);
6788 static noinline int uncompress_inline(struct btrfs_path *path,
6790 struct btrfs_file_extent_item *item)
6793 struct extent_buffer *leaf = path->nodes[0];
6796 unsigned long inline_size;
6800 compress_type = btrfs_file_extent_compression(leaf, item);
6801 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6802 inline_size = btrfs_file_extent_inline_item_len(leaf, path->slots[0]);
6803 tmp = kmalloc(inline_size, GFP_NOFS);
6806 ptr = btrfs_file_extent_inline_start(item);
6808 read_extent_buffer(leaf, tmp, ptr, inline_size);
6810 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6811 ret = btrfs_decompress(compress_type, tmp, page, 0, inline_size, max_size);
6814 * decompression code contains a memset to fill in any space between the end
6815 * of the uncompressed data and the end of max_size in case the decompressed
6816 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6817 * the end of an inline extent and the beginning of the next block, so we
6818 * cover that region here.
6821 if (max_size < PAGE_SIZE)
6822 memzero_page(page, max_size, PAGE_SIZE - max_size);
6827 static int read_inline_extent(struct btrfs_inode *inode, struct btrfs_path *path,
6830 struct btrfs_file_extent_item *fi;
6834 if (!page || PageUptodate(page))
6837 ASSERT(page_offset(page) == 0);
6839 fi = btrfs_item_ptr(path->nodes[0], path->slots[0],
6840 struct btrfs_file_extent_item);
6841 if (btrfs_file_extent_compression(path->nodes[0], fi) != BTRFS_COMPRESS_NONE)
6842 return uncompress_inline(path, page, fi);
6844 copy_size = min_t(u64, PAGE_SIZE,
6845 btrfs_file_extent_ram_bytes(path->nodes[0], fi));
6846 kaddr = kmap_local_page(page);
6847 read_extent_buffer(path->nodes[0], kaddr,
6848 btrfs_file_extent_inline_start(fi), copy_size);
6849 kunmap_local(kaddr);
6850 if (copy_size < PAGE_SIZE)
6851 memzero_page(page, copy_size, PAGE_SIZE - copy_size);
6856 * Lookup the first extent overlapping a range in a file.
6858 * @inode: file to search in
6859 * @page: page to read extent data into if the extent is inline
6860 * @pg_offset: offset into @page to copy to
6861 * @start: file offset
6862 * @len: length of range starting at @start
6864 * Return the first &struct extent_map which overlaps the given range, reading
6865 * it from the B-tree and caching it if necessary. Note that there may be more
6866 * extents which overlap the given range after the returned extent_map.
6868 * If @page is not NULL and the extent is inline, this also reads the extent
6869 * data directly into the page and marks the extent up to date in the io_tree.
6871 * Return: ERR_PTR on error, non-NULL extent_map on success.
6873 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6874 struct page *page, size_t pg_offset,
6877 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6879 u64 extent_start = 0;
6881 u64 objectid = btrfs_ino(inode);
6882 int extent_type = -1;
6883 struct btrfs_path *path = NULL;
6884 struct btrfs_root *root = inode->root;
6885 struct btrfs_file_extent_item *item;
6886 struct extent_buffer *leaf;
6887 struct btrfs_key found_key;
6888 struct extent_map *em = NULL;
6889 struct extent_map_tree *em_tree = &inode->extent_tree;
6891 read_lock(&em_tree->lock);
6892 em = lookup_extent_mapping(em_tree, start, len);
6893 read_unlock(&em_tree->lock);
6896 if (em->start > start || em->start + em->len <= start)
6897 free_extent_map(em);
6898 else if (em->block_start == EXTENT_MAP_INLINE && page)
6899 free_extent_map(em);
6903 em = alloc_extent_map();
6908 em->start = EXTENT_MAP_HOLE;
6909 em->orig_start = EXTENT_MAP_HOLE;
6911 em->block_len = (u64)-1;
6913 path = btrfs_alloc_path();
6919 /* Chances are we'll be called again, so go ahead and do readahead */
6920 path->reada = READA_FORWARD;
6923 * The same explanation in load_free_space_cache applies here as well,
6924 * we only read when we're loading the free space cache, and at that
6925 * point the commit_root has everything we need.
6927 if (btrfs_is_free_space_inode(inode)) {
6928 path->search_commit_root = 1;
6929 path->skip_locking = 1;
6932 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6935 } else if (ret > 0) {
6936 if (path->slots[0] == 0)
6942 leaf = path->nodes[0];
6943 item = btrfs_item_ptr(leaf, path->slots[0],
6944 struct btrfs_file_extent_item);
6945 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6946 if (found_key.objectid != objectid ||
6947 found_key.type != BTRFS_EXTENT_DATA_KEY) {
6949 * If we backup past the first extent we want to move forward
6950 * and see if there is an extent in front of us, otherwise we'll
6951 * say there is a hole for our whole search range which can
6958 extent_type = btrfs_file_extent_type(leaf, item);
6959 extent_start = found_key.offset;
6960 extent_end = btrfs_file_extent_end(path);
6961 if (extent_type == BTRFS_FILE_EXTENT_REG ||
6962 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
6963 /* Only regular file could have regular/prealloc extent */
6964 if (!S_ISREG(inode->vfs_inode.i_mode)) {
6967 "regular/prealloc extent found for non-regular inode %llu",
6971 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6973 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
6974 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6979 if (start >= extent_end) {
6981 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6982 ret = btrfs_next_leaf(root, path);
6988 leaf = path->nodes[0];
6990 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6991 if (found_key.objectid != objectid ||
6992 found_key.type != BTRFS_EXTENT_DATA_KEY)
6994 if (start + len <= found_key.offset)
6996 if (start > found_key.offset)
6999 /* New extent overlaps with existing one */
7001 em->orig_start = start;
7002 em->len = found_key.offset - start;
7003 em->block_start = EXTENT_MAP_HOLE;
7007 btrfs_extent_item_to_extent_map(inode, path, item, em);
7009 if (extent_type == BTRFS_FILE_EXTENT_REG ||
7010 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
7012 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
7014 * Inline extent can only exist at file offset 0. This is
7015 * ensured by tree-checker and inline extent creation path.
7016 * Thus all members representing file offsets should be zero.
7018 ASSERT(pg_offset == 0);
7019 ASSERT(extent_start == 0);
7020 ASSERT(em->start == 0);
7023 * btrfs_extent_item_to_extent_map() should have properly
7024 * initialized em members already.
7026 * Other members are not utilized for inline extents.
7028 ASSERT(em->block_start == EXTENT_MAP_INLINE);
7029 ASSERT(em->len == fs_info->sectorsize);
7031 ret = read_inline_extent(inode, path, page);
7038 em->orig_start = start;
7040 em->block_start = EXTENT_MAP_HOLE;
7043 btrfs_release_path(path);
7044 if (em->start > start || extent_map_end(em) <= start) {
7046 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7047 em->start, em->len, start, len);
7052 write_lock(&em_tree->lock);
7053 ret = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7054 write_unlock(&em_tree->lock);
7056 btrfs_free_path(path);
7058 trace_btrfs_get_extent(root, inode, em);
7061 free_extent_map(em);
7062 return ERR_PTR(ret);
7067 static struct extent_map *btrfs_create_dio_extent(struct btrfs_inode *inode,
7068 struct btrfs_dio_data *dio_data,
7071 const u64 orig_start,
7072 const u64 block_start,
7073 const u64 block_len,
7074 const u64 orig_block_len,
7075 const u64 ram_bytes,
7078 struct extent_map *em = NULL;
7079 struct btrfs_ordered_extent *ordered;
7081 if (type != BTRFS_ORDERED_NOCOW) {
7082 em = create_io_em(inode, start, len, orig_start, block_start,
7083 block_len, orig_block_len, ram_bytes,
7084 BTRFS_COMPRESS_NONE, /* compress_type */
7089 ordered = btrfs_alloc_ordered_extent(inode, start, len, len,
7090 block_start, block_len, 0,
7092 (1 << BTRFS_ORDERED_DIRECT),
7093 BTRFS_COMPRESS_NONE);
7094 if (IS_ERR(ordered)) {
7096 free_extent_map(em);
7097 btrfs_drop_extent_map_range(inode, start,
7098 start + len - 1, false);
7100 em = ERR_CAST(ordered);
7102 ASSERT(!dio_data->ordered);
7103 dio_data->ordered = ordered;
7110 static struct extent_map *btrfs_new_extent_direct(struct btrfs_inode *inode,
7111 struct btrfs_dio_data *dio_data,
7114 struct btrfs_root *root = inode->root;
7115 struct btrfs_fs_info *fs_info = root->fs_info;
7116 struct extent_map *em;
7117 struct btrfs_key ins;
7121 alloc_hint = get_extent_allocation_hint(inode, start, len);
7122 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7123 0, alloc_hint, &ins, 1, 1);
7125 return ERR_PTR(ret);
7127 em = btrfs_create_dio_extent(inode, dio_data, start, ins.offset, start,
7128 ins.objectid, ins.offset, ins.offset,
7129 ins.offset, BTRFS_ORDERED_REGULAR);
7130 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7132 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset,
7138 static bool btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
7140 struct btrfs_block_group *block_group;
7141 bool readonly = false;
7143 block_group = btrfs_lookup_block_group(fs_info, bytenr);
7144 if (!block_group || block_group->ro)
7147 btrfs_put_block_group(block_group);
7152 * Check if we can do nocow write into the range [@offset, @offset + @len)
7154 * @offset: File offset
7155 * @len: The length to write, will be updated to the nocow writeable
7157 * @orig_start: (optional) Return the original file offset of the file extent
7158 * @orig_len: (optional) Return the original on-disk length of the file extent
7159 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7160 * @strict: if true, omit optimizations that might force us into unnecessary
7161 * cow. e.g., don't trust generation number.
7164 * >0 and update @len if we can do nocow write
7165 * 0 if we can't do nocow write
7166 * <0 if error happened
7168 * NOTE: This only checks the file extents, caller is responsible to wait for
7169 * any ordered extents.
7171 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7172 u64 *orig_start, u64 *orig_block_len,
7173 u64 *ram_bytes, bool nowait, bool strict)
7175 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7176 struct can_nocow_file_extent_args nocow_args = { 0 };
7177 struct btrfs_path *path;
7179 struct extent_buffer *leaf;
7180 struct btrfs_root *root = BTRFS_I(inode)->root;
7181 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7182 struct btrfs_file_extent_item *fi;
7183 struct btrfs_key key;
7186 path = btrfs_alloc_path();
7189 path->nowait = nowait;
7191 ret = btrfs_lookup_file_extent(NULL, root, path,
7192 btrfs_ino(BTRFS_I(inode)), offset, 0);
7197 if (path->slots[0] == 0) {
7198 /* can't find the item, must cow */
7205 leaf = path->nodes[0];
7206 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
7207 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7208 key.type != BTRFS_EXTENT_DATA_KEY) {
7209 /* not our file or wrong item type, must cow */
7213 if (key.offset > offset) {
7214 /* Wrong offset, must cow */
7218 if (btrfs_file_extent_end(path) <= offset)
7221 fi = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
7222 found_type = btrfs_file_extent_type(leaf, fi);
7224 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7226 nocow_args.start = offset;
7227 nocow_args.end = offset + *len - 1;
7228 nocow_args.strict = strict;
7229 nocow_args.free_path = true;
7231 ret = can_nocow_file_extent(path, &key, BTRFS_I(inode), &nocow_args);
7232 /* can_nocow_file_extent() has freed the path. */
7236 /* Treat errors as not being able to NOCOW. */
7242 if (btrfs_extent_readonly(fs_info, nocow_args.disk_bytenr))
7245 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7246 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7249 range_end = round_up(offset + nocow_args.num_bytes,
7250 root->fs_info->sectorsize) - 1;
7251 ret = test_range_bit(io_tree, offset, range_end,
7252 EXTENT_DELALLOC, 0, NULL);
7260 *orig_start = key.offset - nocow_args.extent_offset;
7262 *orig_block_len = nocow_args.disk_num_bytes;
7264 *len = nocow_args.num_bytes;
7267 btrfs_free_path(path);
7271 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7272 struct extent_state **cached_state,
7273 unsigned int iomap_flags)
7275 const bool writing = (iomap_flags & IOMAP_WRITE);
7276 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7277 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7278 struct btrfs_ordered_extent *ordered;
7283 if (!try_lock_extent(io_tree, lockstart, lockend,
7287 lock_extent(io_tree, lockstart, lockend, cached_state);
7290 * We're concerned with the entire range that we're going to be
7291 * doing DIO to, so we need to make sure there's no ordered
7292 * extents in this range.
7294 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7295 lockend - lockstart + 1);
7298 * We need to make sure there are no buffered pages in this
7299 * range either, we could have raced between the invalidate in
7300 * generic_file_direct_write and locking the extent. The
7301 * invalidate needs to happen so that reads after a write do not
7305 (!writing || !filemap_range_has_page(inode->i_mapping,
7306 lockstart, lockend)))
7309 unlock_extent(io_tree, lockstart, lockend, cached_state);
7313 btrfs_put_ordered_extent(ordered);
7318 * If we are doing a DIO read and the ordered extent we
7319 * found is for a buffered write, we can not wait for it
7320 * to complete and retry, because if we do so we can
7321 * deadlock with concurrent buffered writes on page
7322 * locks. This happens only if our DIO read covers more
7323 * than one extent map, if at this point has already
7324 * created an ordered extent for a previous extent map
7325 * and locked its range in the inode's io tree, and a
7326 * concurrent write against that previous extent map's
7327 * range and this range started (we unlock the ranges
7328 * in the io tree only when the bios complete and
7329 * buffered writes always lock pages before attempting
7330 * to lock range in the io tree).
7333 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7334 btrfs_start_ordered_extent(ordered);
7336 ret = nowait ? -EAGAIN : -ENOTBLK;
7337 btrfs_put_ordered_extent(ordered);
7340 * We could trigger writeback for this range (and wait
7341 * for it to complete) and then invalidate the pages for
7342 * this range (through invalidate_inode_pages2_range()),
7343 * but that can lead us to a deadlock with a concurrent
7344 * call to readahead (a buffered read or a defrag call
7345 * triggered a readahead) on a page lock due to an
7346 * ordered dio extent we created before but did not have
7347 * yet a corresponding bio submitted (whence it can not
7348 * complete), which makes readahead wait for that
7349 * ordered extent to complete while holding a lock on
7352 ret = nowait ? -EAGAIN : -ENOTBLK;
7364 /* The callers of this must take lock_extent() */
7365 static struct extent_map *create_io_em(struct btrfs_inode *inode, u64 start,
7366 u64 len, u64 orig_start, u64 block_start,
7367 u64 block_len, u64 orig_block_len,
7368 u64 ram_bytes, int compress_type,
7371 struct extent_map *em;
7374 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7375 type == BTRFS_ORDERED_COMPRESSED ||
7376 type == BTRFS_ORDERED_NOCOW ||
7377 type == BTRFS_ORDERED_REGULAR);
7379 em = alloc_extent_map();
7381 return ERR_PTR(-ENOMEM);
7384 em->orig_start = orig_start;
7386 em->block_len = block_len;
7387 em->block_start = block_start;
7388 em->orig_block_len = orig_block_len;
7389 em->ram_bytes = ram_bytes;
7390 em->generation = -1;
7391 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7392 if (type == BTRFS_ORDERED_PREALLOC) {
7393 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7394 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7395 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7396 em->compress_type = compress_type;
7399 ret = btrfs_replace_extent_map_range(inode, em, true);
7401 free_extent_map(em);
7402 return ERR_PTR(ret);
7405 /* em got 2 refs now, callers needs to do free_extent_map once. */
7410 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7411 struct inode *inode,
7412 struct btrfs_dio_data *dio_data,
7413 u64 start, u64 *lenp,
7414 unsigned int iomap_flags)
7416 const bool nowait = (iomap_flags & IOMAP_NOWAIT);
7417 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7418 struct extent_map *em = *map;
7420 u64 block_start, orig_start, orig_block_len, ram_bytes;
7421 struct btrfs_block_group *bg;
7422 bool can_nocow = false;
7423 bool space_reserved = false;
7429 * We don't allocate a new extent in the following cases
7431 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7433 * 2) The extent is marked as PREALLOC. We're good to go here and can
7434 * just use the extent.
7437 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7438 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7439 em->block_start != EXTENT_MAP_HOLE)) {
7440 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7441 type = BTRFS_ORDERED_PREALLOC;
7443 type = BTRFS_ORDERED_NOCOW;
7444 len = min(len, em->len - (start - em->start));
7445 block_start = em->block_start + (start - em->start);
7447 if (can_nocow_extent(inode, start, &len, &orig_start,
7448 &orig_block_len, &ram_bytes, false, false) == 1) {
7449 bg = btrfs_inc_nocow_writers(fs_info, block_start);
7457 struct extent_map *em2;
7459 /* We can NOCOW, so only need to reserve metadata space. */
7460 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7463 /* Our caller expects us to free the input extent map. */
7464 free_extent_map(em);
7466 btrfs_dec_nocow_writers(bg);
7467 if (nowait && (ret == -ENOSPC || ret == -EDQUOT))
7471 space_reserved = true;
7473 em2 = btrfs_create_dio_extent(BTRFS_I(inode), dio_data, start, len,
7474 orig_start, block_start,
7475 len, orig_block_len,
7477 btrfs_dec_nocow_writers(bg);
7478 if (type == BTRFS_ORDERED_PREALLOC) {
7479 free_extent_map(em);
7489 dio_data->nocow_done = true;
7491 /* Our caller expects us to free the input extent map. */
7492 free_extent_map(em);
7501 * If we could not allocate data space before locking the file
7502 * range and we can't do a NOCOW write, then we have to fail.
7504 if (!dio_data->data_space_reserved) {
7510 * We have to COW and we have already reserved data space before,
7511 * so now we reserve only metadata.
7513 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len, len,
7517 space_reserved = true;
7519 em = btrfs_new_extent_direct(BTRFS_I(inode), dio_data, start, len);
7525 len = min(len, em->len - (start - em->start));
7527 btrfs_delalloc_release_metadata(BTRFS_I(inode),
7528 prev_len - len, true);
7532 * We have created our ordered extent, so we can now release our reservation
7533 * for an outstanding extent.
7535 btrfs_delalloc_release_extents(BTRFS_I(inode), prev_len);
7538 * Need to update the i_size under the extent lock so buffered
7539 * readers will get the updated i_size when we unlock.
7541 if (start + len > i_size_read(inode))
7542 i_size_write(inode, start + len);
7544 if (ret && space_reserved) {
7545 btrfs_delalloc_release_extents(BTRFS_I(inode), len);
7546 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, true);
7552 static int btrfs_dio_iomap_begin(struct inode *inode, loff_t start,
7553 loff_t length, unsigned int flags, struct iomap *iomap,
7554 struct iomap *srcmap)
7556 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7557 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7558 struct extent_map *em;
7559 struct extent_state *cached_state = NULL;
7560 struct btrfs_dio_data *dio_data = iter->private;
7561 u64 lockstart, lockend;
7562 const bool write = !!(flags & IOMAP_WRITE);
7565 const u64 data_alloc_len = length;
7566 bool unlock_extents = false;
7569 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7570 * we're NOWAIT we may submit a bio for a partial range and return
7571 * EIOCBQUEUED, which would result in an errant short read.
7573 * The best way to handle this would be to allow for partial completions
7574 * of iocb's, so we could submit the partial bio, return and fault in
7575 * the rest of the pages, and then submit the io for the rest of the
7576 * range. However we don't have that currently, so simply return
7577 * -EAGAIN at this point so that the normal path is used.
7579 if (!write && (flags & IOMAP_NOWAIT) && length > PAGE_SIZE)
7583 * Cap the size of reads to that usually seen in buffered I/O as we need
7584 * to allocate a contiguous array for the checksums.
7587 len = min_t(u64, len, fs_info->sectorsize * BTRFS_MAX_BIO_SECTORS);
7590 lockend = start + len - 1;
7593 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7594 * enough if we've written compressed pages to this area, so we need to
7595 * flush the dirty pages again to make absolutely sure that any
7596 * outstanding dirty pages are on disk - the first flush only starts
7597 * compression on the data, while keeping the pages locked, so by the
7598 * time the second flush returns we know bios for the compressed pages
7599 * were submitted and finished, and the pages no longer under writeback.
7601 * If we have a NOWAIT request and we have any pages in the range that
7602 * are locked, likely due to compression still in progress, we don't want
7603 * to block on page locks. We also don't want to block on pages marked as
7604 * dirty or under writeback (same as for the non-compression case).
7605 * iomap_dio_rw() did the same check, but after that and before we got
7606 * here, mmap'ed writes may have happened or buffered reads started
7607 * (readpage() and readahead(), which lock pages), as we haven't locked
7608 * the file range yet.
7610 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
7611 &BTRFS_I(inode)->runtime_flags)) {
7612 if (flags & IOMAP_NOWAIT) {
7613 if (filemap_range_needs_writeback(inode->i_mapping,
7614 lockstart, lockend))
7617 ret = filemap_fdatawrite_range(inode->i_mapping, start,
7618 start + length - 1);
7624 memset(dio_data, 0, sizeof(*dio_data));
7627 * We always try to allocate data space and must do it before locking
7628 * the file range, to avoid deadlocks with concurrent writes to the same
7629 * range if the range has several extents and the writes don't expand the
7630 * current i_size (the inode lock is taken in shared mode). If we fail to
7631 * allocate data space here we continue and later, after locking the
7632 * file range, we fail with ENOSPC only if we figure out we can not do a
7635 if (write && !(flags & IOMAP_NOWAIT)) {
7636 ret = btrfs_check_data_free_space(BTRFS_I(inode),
7637 &dio_data->data_reserved,
7638 start, data_alloc_len, false);
7640 dio_data->data_space_reserved = true;
7641 else if (ret && !(BTRFS_I(inode)->flags &
7642 (BTRFS_INODE_NODATACOW | BTRFS_INODE_PREALLOC)))
7647 * If this errors out it's because we couldn't invalidate pagecache for
7648 * this range and we need to fallback to buffered IO, or we are doing a
7649 * NOWAIT read/write and we need to block.
7651 ret = lock_extent_direct(inode, lockstart, lockend, &cached_state, flags);
7655 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
7662 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7663 * io. INLINE is special, and we could probably kludge it in here, but
7664 * it's still buffered so for safety lets just fall back to the generic
7667 * For COMPRESSED we _have_ to read the entire extent in so we can
7668 * decompress it, so there will be buffering required no matter what we
7669 * do, so go ahead and fallback to buffered.
7671 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7672 * to buffered IO. Don't blame me, this is the price we pay for using
7675 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7676 em->block_start == EXTENT_MAP_INLINE) {
7677 free_extent_map(em);
7679 * If we are in a NOWAIT context, return -EAGAIN in order to
7680 * fallback to buffered IO. This is not only because we can
7681 * block with buffered IO (no support for NOWAIT semantics at
7682 * the moment) but also to avoid returning short reads to user
7683 * space - this happens if we were able to read some data from
7684 * previous non-compressed extents and then when we fallback to
7685 * buffered IO, at btrfs_file_read_iter() by calling
7686 * filemap_read(), we fail to fault in pages for the read buffer,
7687 * in which case filemap_read() returns a short read (the number
7688 * of bytes previously read is > 0, so it does not return -EFAULT).
7690 ret = (flags & IOMAP_NOWAIT) ? -EAGAIN : -ENOTBLK;
7694 len = min(len, em->len - (start - em->start));
7697 * If we have a NOWAIT request and the range contains multiple extents
7698 * (or a mix of extents and holes), then we return -EAGAIN to make the
7699 * caller fallback to a context where it can do a blocking (without
7700 * NOWAIT) request. This way we avoid doing partial IO and returning
7701 * success to the caller, which is not optimal for writes and for reads
7702 * it can result in unexpected behaviour for an application.
7704 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7705 * iomap_dio_rw(), we can end up returning less data then what the caller
7706 * asked for, resulting in an unexpected, and incorrect, short read.
7707 * That is, the caller asked to read N bytes and we return less than that,
7708 * which is wrong unless we are crossing EOF. This happens if we get a
7709 * page fault error when trying to fault in pages for the buffer that is
7710 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7711 * have previously submitted bios for other extents in the range, in
7712 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7713 * those bios have completed by the time we get the page fault error,
7714 * which we return back to our caller - we should only return EIOCBQUEUED
7715 * after we have submitted bios for all the extents in the range.
7717 if ((flags & IOMAP_NOWAIT) && len < length) {
7718 free_extent_map(em);
7724 ret = btrfs_get_blocks_direct_write(&em, inode, dio_data,
7725 start, &len, flags);
7728 unlock_extents = true;
7729 /* Recalc len in case the new em is smaller than requested */
7730 len = min(len, em->len - (start - em->start));
7731 if (dio_data->data_space_reserved) {
7733 u64 release_len = 0;
7735 if (dio_data->nocow_done) {
7736 release_offset = start;
7737 release_len = data_alloc_len;
7738 } else if (len < data_alloc_len) {
7739 release_offset = start + len;
7740 release_len = data_alloc_len - len;
7743 if (release_len > 0)
7744 btrfs_free_reserved_data_space(BTRFS_I(inode),
7745 dio_data->data_reserved,
7751 * We need to unlock only the end area that we aren't using.
7752 * The rest is going to be unlocked by the endio routine.
7754 lockstart = start + len;
7755 if (lockstart < lockend)
7756 unlock_extents = true;
7760 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7763 free_extent_state(cached_state);
7766 * Translate extent map information to iomap.
7767 * We trim the extents (and move the addr) even though iomap code does
7768 * that, since we have locked only the parts we are performing I/O in.
7770 if ((em->block_start == EXTENT_MAP_HOLE) ||
7771 (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) && !write)) {
7772 iomap->addr = IOMAP_NULL_ADDR;
7773 iomap->type = IOMAP_HOLE;
7775 iomap->addr = em->block_start + (start - em->start);
7776 iomap->type = IOMAP_MAPPED;
7778 iomap->offset = start;
7779 iomap->bdev = fs_info->fs_devices->latest_dev->bdev;
7780 iomap->length = len;
7781 free_extent_map(em);
7786 unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7789 if (dio_data->data_space_reserved) {
7790 btrfs_free_reserved_data_space(BTRFS_I(inode),
7791 dio_data->data_reserved,
7792 start, data_alloc_len);
7793 extent_changeset_free(dio_data->data_reserved);
7799 static int btrfs_dio_iomap_end(struct inode *inode, loff_t pos, loff_t length,
7800 ssize_t written, unsigned int flags, struct iomap *iomap)
7802 struct iomap_iter *iter = container_of(iomap, struct iomap_iter, iomap);
7803 struct btrfs_dio_data *dio_data = iter->private;
7804 size_t submitted = dio_data->submitted;
7805 const bool write = !!(flags & IOMAP_WRITE);
7808 if (!write && (iomap->type == IOMAP_HOLE)) {
7809 /* If reading from a hole, unlock and return */
7810 unlock_extent(&BTRFS_I(inode)->io_tree, pos, pos + length - 1,
7815 if (submitted < length) {
7817 length -= submitted;
7819 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7820 pos, length, false);
7822 unlock_extent(&BTRFS_I(inode)->io_tree, pos,
7823 pos + length - 1, NULL);
7827 btrfs_put_ordered_extent(dio_data->ordered);
7828 dio_data->ordered = NULL;
7832 extent_changeset_free(dio_data->data_reserved);
7836 static void btrfs_dio_end_io(struct btrfs_bio *bbio)
7838 struct btrfs_dio_private *dip =
7839 container_of(bbio, struct btrfs_dio_private, bbio);
7840 struct btrfs_inode *inode = bbio->inode;
7841 struct bio *bio = &bbio->bio;
7843 if (bio->bi_status) {
7844 btrfs_warn(inode->root->fs_info,
7845 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7846 btrfs_ino(inode), bio->bi_opf,
7847 dip->file_offset, dip->bytes, bio->bi_status);
7850 if (btrfs_op(bio) == BTRFS_MAP_WRITE) {
7851 btrfs_finish_ordered_extent(bbio->ordered, NULL,
7852 dip->file_offset, dip->bytes,
7855 unlock_extent(&inode->io_tree, dip->file_offset,
7856 dip->file_offset + dip->bytes - 1, NULL);
7859 bbio->bio.bi_private = bbio->private;
7860 iomap_dio_bio_end_io(bio);
7863 static void btrfs_dio_submit_io(const struct iomap_iter *iter, struct bio *bio,
7866 struct btrfs_bio *bbio = btrfs_bio(bio);
7867 struct btrfs_dio_private *dip =
7868 container_of(bbio, struct btrfs_dio_private, bbio);
7869 struct btrfs_dio_data *dio_data = iter->private;
7871 btrfs_bio_init(bbio, BTRFS_I(iter->inode)->root->fs_info,
7872 btrfs_dio_end_io, bio->bi_private);
7873 bbio->inode = BTRFS_I(iter->inode);
7874 bbio->file_offset = file_offset;
7876 dip->file_offset = file_offset;
7877 dip->bytes = bio->bi_iter.bi_size;
7879 dio_data->submitted += bio->bi_iter.bi_size;
7882 * Check if we are doing a partial write. If we are, we need to split
7883 * the ordered extent to match the submitted bio. Hang on to the
7884 * remaining unfinishable ordered_extent in dio_data so that it can be
7885 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7886 * remaining pages is blocked on the outstanding ordered extent.
7888 if (iter->flags & IOMAP_WRITE) {
7891 ret = btrfs_extract_ordered_extent(bbio, dio_data->ordered);
7893 btrfs_finish_ordered_extent(dio_data->ordered, NULL,
7894 file_offset, dip->bytes,
7896 bio->bi_status = errno_to_blk_status(ret);
7897 iomap_dio_bio_end_io(bio);
7902 btrfs_submit_bio(bbio, 0);
7905 static const struct iomap_ops btrfs_dio_iomap_ops = {
7906 .iomap_begin = btrfs_dio_iomap_begin,
7907 .iomap_end = btrfs_dio_iomap_end,
7910 static const struct iomap_dio_ops btrfs_dio_ops = {
7911 .submit_io = btrfs_dio_submit_io,
7912 .bio_set = &btrfs_dio_bioset,
7915 ssize_t btrfs_dio_read(struct kiocb *iocb, struct iov_iter *iter, size_t done_before)
7917 struct btrfs_dio_data data = { 0 };
7919 return iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7920 IOMAP_DIO_PARTIAL, &data, done_before);
7923 struct iomap_dio *btrfs_dio_write(struct kiocb *iocb, struct iov_iter *iter,
7926 struct btrfs_dio_data data = { 0 };
7928 return __iomap_dio_rw(iocb, iter, &btrfs_dio_iomap_ops, &btrfs_dio_ops,
7929 IOMAP_DIO_PARTIAL, &data, done_before);
7932 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7937 ret = fiemap_prep(inode, fieinfo, start, &len, 0);
7942 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7943 * file range (0 to LLONG_MAX), but that is not enough if we have
7944 * compression enabled. The first filemap_fdatawrite_range() only kicks
7945 * in the compression of data (in an async thread) and will return
7946 * before the compression is done and writeback is started. A second
7947 * filemap_fdatawrite_range() is needed to wait for the compression to
7948 * complete and writeback to start. We also need to wait for ordered
7949 * extents to complete, because our fiemap implementation uses mainly
7950 * file extent items to list the extents, searching for extent maps
7951 * only for file ranges with holes or prealloc extents to figure out
7952 * if we have delalloc in those ranges.
7954 if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC) {
7955 ret = btrfs_wait_ordered_range(inode, 0, LLONG_MAX);
7960 return extent_fiemap(BTRFS_I(inode), fieinfo, start, len);
7963 static int btrfs_writepages(struct address_space *mapping,
7964 struct writeback_control *wbc)
7966 return extent_writepages(mapping, wbc);
7969 static void btrfs_readahead(struct readahead_control *rac)
7971 extent_readahead(rac);
7975 * For release_folio() and invalidate_folio() we have a race window where
7976 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7977 * If we continue to release/invalidate the page, we could cause use-after-free
7978 * for subpage spinlock. So this function is to spin and wait for subpage
7981 static void wait_subpage_spinlock(struct page *page)
7983 struct btrfs_fs_info *fs_info = btrfs_sb(page->mapping->host->i_sb);
7984 struct btrfs_subpage *subpage;
7986 if (!btrfs_is_subpage(fs_info, page))
7989 ASSERT(PagePrivate(page) && page->private);
7990 subpage = (struct btrfs_subpage *)page->private;
7993 * This may look insane as we just acquire the spinlock and release it,
7994 * without doing anything. But we just want to make sure no one is
7995 * still holding the subpage spinlock.
7996 * And since the page is not dirty nor writeback, and we have page
7997 * locked, the only possible way to hold a spinlock is from the endio
7998 * function to clear page writeback.
8000 * Here we just acquire the spinlock so that all existing callers
8001 * should exit and we're safe to release/invalidate the page.
8003 spin_lock_irq(&subpage->lock);
8004 spin_unlock_irq(&subpage->lock);
8007 static bool __btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8009 int ret = try_release_extent_mapping(&folio->page, gfp_flags);
8012 wait_subpage_spinlock(&folio->page);
8013 clear_page_extent_mapped(&folio->page);
8018 static bool btrfs_release_folio(struct folio *folio, gfp_t gfp_flags)
8020 if (folio_test_writeback(folio) || folio_test_dirty(folio))
8022 return __btrfs_release_folio(folio, gfp_flags);
8025 #ifdef CONFIG_MIGRATION
8026 static int btrfs_migrate_folio(struct address_space *mapping,
8027 struct folio *dst, struct folio *src,
8028 enum migrate_mode mode)
8030 int ret = filemap_migrate_folio(mapping, dst, src, mode);
8032 if (ret != MIGRATEPAGE_SUCCESS)
8035 if (folio_test_ordered(src)) {
8036 folio_clear_ordered(src);
8037 folio_set_ordered(dst);
8040 return MIGRATEPAGE_SUCCESS;
8043 #define btrfs_migrate_folio NULL
8046 static void btrfs_invalidate_folio(struct folio *folio, size_t offset,
8049 struct btrfs_inode *inode = BTRFS_I(folio->mapping->host);
8050 struct btrfs_fs_info *fs_info = inode->root->fs_info;
8051 struct extent_io_tree *tree = &inode->io_tree;
8052 struct extent_state *cached_state = NULL;
8053 u64 page_start = folio_pos(folio);
8054 u64 page_end = page_start + folio_size(folio) - 1;
8056 int inode_evicting = inode->vfs_inode.i_state & I_FREEING;
8059 * We have folio locked so no new ordered extent can be created on this
8060 * page, nor bio can be submitted for this folio.
8062 * But already submitted bio can still be finished on this folio.
8063 * Furthermore, endio function won't skip folio which has Ordered
8064 * (Private2) already cleared, so it's possible for endio and
8065 * invalidate_folio to do the same ordered extent accounting twice
8068 * So here we wait for any submitted bios to finish, so that we won't
8069 * do double ordered extent accounting on the same folio.
8071 folio_wait_writeback(folio);
8072 wait_subpage_spinlock(&folio->page);
8075 * For subpage case, we have call sites like
8076 * btrfs_punch_hole_lock_range() which passes range not aligned to
8078 * If the range doesn't cover the full folio, we don't need to and
8079 * shouldn't clear page extent mapped, as folio->private can still
8080 * record subpage dirty bits for other part of the range.
8082 * For cases that invalidate the full folio even the range doesn't
8083 * cover the full folio, like invalidating the last folio, we're
8084 * still safe to wait for ordered extent to finish.
8086 if (!(offset == 0 && length == folio_size(folio))) {
8087 btrfs_release_folio(folio, GFP_NOFS);
8091 if (!inode_evicting)
8092 lock_extent(tree, page_start, page_end, &cached_state);
8095 while (cur < page_end) {
8096 struct btrfs_ordered_extent *ordered;
8099 u32 extra_flags = 0;
8101 ordered = btrfs_lookup_first_ordered_range(inode, cur,
8102 page_end + 1 - cur);
8104 range_end = page_end;
8106 * No ordered extent covering this range, we are safe
8107 * to delete all extent states in the range.
8109 extra_flags = EXTENT_CLEAR_ALL_BITS;
8112 if (ordered->file_offset > cur) {
8114 * There is a range between [cur, oe->file_offset) not
8115 * covered by any ordered extent.
8116 * We are safe to delete all extent states, and handle
8117 * the ordered extent in the next iteration.
8119 range_end = ordered->file_offset - 1;
8120 extra_flags = EXTENT_CLEAR_ALL_BITS;
8124 range_end = min(ordered->file_offset + ordered->num_bytes - 1,
8126 ASSERT(range_end + 1 - cur < U32_MAX);
8127 range_len = range_end + 1 - cur;
8128 if (!btrfs_page_test_ordered(fs_info, &folio->page, cur, range_len)) {
8130 * If Ordered (Private2) is cleared, it means endio has
8131 * already been executed for the range.
8132 * We can't delete the extent states as
8133 * btrfs_finish_ordered_io() may still use some of them.
8137 btrfs_page_clear_ordered(fs_info, &folio->page, cur, range_len);
8140 * IO on this page will never be started, so we need to account
8141 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8142 * here, must leave that up for the ordered extent completion.
8144 * This will also unlock the range for incoming
8145 * btrfs_finish_ordered_io().
8147 if (!inode_evicting)
8148 clear_extent_bit(tree, cur, range_end,
8150 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8151 EXTENT_DEFRAG, &cached_state);
8153 spin_lock_irq(&inode->ordered_tree.lock);
8154 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8155 ordered->truncated_len = min(ordered->truncated_len,
8156 cur - ordered->file_offset);
8157 spin_unlock_irq(&inode->ordered_tree.lock);
8160 * If the ordered extent has finished, we're safe to delete all
8161 * the extent states of the range, otherwise
8162 * btrfs_finish_ordered_io() will get executed by endio for
8163 * other pages, so we can't delete extent states.
8165 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8166 cur, range_end + 1 - cur)) {
8167 btrfs_finish_ordered_io(ordered);
8169 * The ordered extent has finished, now we're again
8170 * safe to delete all extent states of the range.
8172 extra_flags = EXTENT_CLEAR_ALL_BITS;
8176 btrfs_put_ordered_extent(ordered);
8178 * Qgroup reserved space handler
8179 * Sector(s) here will be either:
8181 * 1) Already written to disk or bio already finished
8182 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8183 * Qgroup will be handled by its qgroup_record then.
8184 * btrfs_qgroup_free_data() call will do nothing here.
8186 * 2) Not written to disk yet
8187 * Then btrfs_qgroup_free_data() call will clear the
8188 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8189 * reserved data space.
8190 * Since the IO will never happen for this page.
8192 btrfs_qgroup_free_data(inode, NULL, cur, range_end + 1 - cur);
8193 if (!inode_evicting) {
8194 clear_extent_bit(tree, cur, range_end, EXTENT_LOCKED |
8195 EXTENT_DELALLOC | EXTENT_UPTODATE |
8196 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG |
8197 extra_flags, &cached_state);
8199 cur = range_end + 1;
8202 * We have iterated through all ordered extents of the page, the page
8203 * should not have Ordered (Private2) anymore, or the above iteration
8204 * did something wrong.
8206 ASSERT(!folio_test_ordered(folio));
8207 btrfs_page_clear_checked(fs_info, &folio->page, folio_pos(folio), folio_size(folio));
8208 if (!inode_evicting)
8209 __btrfs_release_folio(folio, GFP_NOFS);
8210 clear_page_extent_mapped(&folio->page);
8214 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8215 * called from a page fault handler when a page is first dirtied. Hence we must
8216 * be careful to check for EOF conditions here. We set the page up correctly
8217 * for a written page which means we get ENOSPC checking when writing into
8218 * holes and correct delalloc and unwritten extent mapping on filesystems that
8219 * support these features.
8221 * We are not allowed to take the i_mutex here so we have to play games to
8222 * protect against truncate races as the page could now be beyond EOF. Because
8223 * truncate_setsize() writes the inode size before removing pages, once we have
8224 * the page lock we can determine safely if the page is beyond EOF. If it is not
8225 * beyond EOF, then the page is guaranteed safe against truncation until we
8228 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8230 struct page *page = vmf->page;
8231 struct inode *inode = file_inode(vmf->vma->vm_file);
8232 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8233 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8234 struct btrfs_ordered_extent *ordered;
8235 struct extent_state *cached_state = NULL;
8236 struct extent_changeset *data_reserved = NULL;
8237 unsigned long zero_start;
8247 reserved_space = PAGE_SIZE;
8249 sb_start_pagefault(inode->i_sb);
8250 page_start = page_offset(page);
8251 page_end = page_start + PAGE_SIZE - 1;
8255 * Reserving delalloc space after obtaining the page lock can lead to
8256 * deadlock. For example, if a dirty page is locked by this function
8257 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8258 * dirty page write out, then the btrfs_writepages() function could
8259 * end up waiting indefinitely to get a lock on the page currently
8260 * being processed by btrfs_page_mkwrite() function.
8262 ret2 = btrfs_delalloc_reserve_space(BTRFS_I(inode), &data_reserved,
8263 page_start, reserved_space);
8265 ret2 = file_update_time(vmf->vma->vm_file);
8269 ret = vmf_error(ret2);
8275 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8277 down_read(&BTRFS_I(inode)->i_mmap_lock);
8279 size = i_size_read(inode);
8281 if ((page->mapping != inode->i_mapping) ||
8282 (page_start >= size)) {
8283 /* page got truncated out from underneath us */
8286 wait_on_page_writeback(page);
8288 lock_extent(io_tree, page_start, page_end, &cached_state);
8289 ret2 = set_page_extent_mapped(page);
8291 ret = vmf_error(ret2);
8292 unlock_extent(io_tree, page_start, page_end, &cached_state);
8297 * we can't set the delalloc bits if there are pending ordered
8298 * extents. Drop our locks and wait for them to finish
8300 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8303 unlock_extent(io_tree, page_start, page_end, &cached_state);
8305 up_read(&BTRFS_I(inode)->i_mmap_lock);
8306 btrfs_start_ordered_extent(ordered);
8307 btrfs_put_ordered_extent(ordered);
8311 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8312 reserved_space = round_up(size - page_start,
8313 fs_info->sectorsize);
8314 if (reserved_space < PAGE_SIZE) {
8315 end = page_start + reserved_space - 1;
8316 btrfs_delalloc_release_space(BTRFS_I(inode),
8317 data_reserved, page_start,
8318 PAGE_SIZE - reserved_space, true);
8323 * page_mkwrite gets called when the page is firstly dirtied after it's
8324 * faulted in, but write(2) could also dirty a page and set delalloc
8325 * bits, thus in this case for space account reason, we still need to
8326 * clear any delalloc bits within this page range since we have to
8327 * reserve data&meta space before lock_page() (see above comments).
8329 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8330 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8331 EXTENT_DEFRAG, &cached_state);
8333 ret2 = btrfs_set_extent_delalloc(BTRFS_I(inode), page_start, end, 0,
8336 unlock_extent(io_tree, page_start, page_end, &cached_state);
8337 ret = VM_FAULT_SIGBUS;
8341 /* page is wholly or partially inside EOF */
8342 if (page_start + PAGE_SIZE > size)
8343 zero_start = offset_in_page(size);
8345 zero_start = PAGE_SIZE;
8347 if (zero_start != PAGE_SIZE)
8348 memzero_page(page, zero_start, PAGE_SIZE - zero_start);
8350 btrfs_page_clear_checked(fs_info, page, page_start, PAGE_SIZE);
8351 btrfs_page_set_dirty(fs_info, page, page_start, end + 1 - page_start);
8352 btrfs_page_set_uptodate(fs_info, page, page_start, end + 1 - page_start);
8354 btrfs_set_inode_last_sub_trans(BTRFS_I(inode));
8356 unlock_extent(io_tree, page_start, page_end, &cached_state);
8357 up_read(&BTRFS_I(inode)->i_mmap_lock);
8359 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8360 sb_end_pagefault(inode->i_sb);
8361 extent_changeset_free(data_reserved);
8362 return VM_FAULT_LOCKED;
8366 up_read(&BTRFS_I(inode)->i_mmap_lock);
8368 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
8369 btrfs_delalloc_release_space(BTRFS_I(inode), data_reserved, page_start,
8370 reserved_space, (ret != 0));
8372 sb_end_pagefault(inode->i_sb);
8373 extent_changeset_free(data_reserved);
8377 static int btrfs_truncate(struct btrfs_inode *inode, bool skip_writeback)
8379 struct btrfs_truncate_control control = {
8381 .ino = btrfs_ino(inode),
8382 .min_type = BTRFS_EXTENT_DATA_KEY,
8383 .clear_extent_range = true,
8385 struct btrfs_root *root = inode->root;
8386 struct btrfs_fs_info *fs_info = root->fs_info;
8387 struct btrfs_block_rsv *rsv;
8389 struct btrfs_trans_handle *trans;
8390 u64 mask = fs_info->sectorsize - 1;
8391 const u64 min_size = btrfs_calc_metadata_size(fs_info, 1);
8393 if (!skip_writeback) {
8394 ret = btrfs_wait_ordered_range(&inode->vfs_inode,
8395 inode->vfs_inode.i_size & (~mask),
8402 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8403 * things going on here:
8405 * 1) We need to reserve space to update our inode.
8407 * 2) We need to have something to cache all the space that is going to
8408 * be free'd up by the truncate operation, but also have some slack
8409 * space reserved in case it uses space during the truncate (thank you
8410 * very much snapshotting).
8412 * And we need these to be separate. The fact is we can use a lot of
8413 * space doing the truncate, and we have no earthly idea how much space
8414 * we will use, so we need the truncate reservation to be separate so it
8415 * doesn't end up using space reserved for updating the inode. We also
8416 * need to be able to stop the transaction and start a new one, which
8417 * means we need to be able to update the inode several times, and we
8418 * have no idea of knowing how many times that will be, so we can't just
8419 * reserve 1 item for the entirety of the operation, so that has to be
8420 * done separately as well.
8422 * So that leaves us with
8424 * 1) rsv - for the truncate reservation, which we will steal from the
8425 * transaction reservation.
8426 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8427 * updating the inode.
8429 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8432 rsv->size = min_size;
8433 rsv->failfast = true;
8436 * 1 for the truncate slack space
8437 * 1 for updating the inode.
8439 trans = btrfs_start_transaction(root, 2);
8440 if (IS_ERR(trans)) {
8441 ret = PTR_ERR(trans);
8445 /* Migrate the slack space for the truncate to our reserve */
8446 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
8449 * We have reserved 2 metadata units when we started the transaction and
8450 * min_size matches 1 unit, so this should never fail, but if it does,
8451 * it's not critical we just fail truncation.
8454 btrfs_end_transaction(trans);
8458 trans->block_rsv = rsv;
8461 struct extent_state *cached_state = NULL;
8462 const u64 new_size = inode->vfs_inode.i_size;
8463 const u64 lock_start = ALIGN_DOWN(new_size, fs_info->sectorsize);
8465 control.new_size = new_size;
8466 lock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8468 * We want to drop from the next block forward in case this new
8469 * size is not block aligned since we will be keeping the last
8470 * block of the extent just the way it is.
8472 btrfs_drop_extent_map_range(inode,
8473 ALIGN(new_size, fs_info->sectorsize),
8476 ret = btrfs_truncate_inode_items(trans, root, &control);
8478 inode_sub_bytes(&inode->vfs_inode, control.sub_bytes);
8479 btrfs_inode_safe_disk_i_size_write(inode, control.last_size);
8481 unlock_extent(&inode->io_tree, lock_start, (u64)-1, &cached_state);
8483 trans->block_rsv = &fs_info->trans_block_rsv;
8484 if (ret != -ENOSPC && ret != -EAGAIN)
8487 ret = btrfs_update_inode(trans, root, inode);
8491 btrfs_end_transaction(trans);
8492 btrfs_btree_balance_dirty(fs_info);
8494 trans = btrfs_start_transaction(root, 2);
8495 if (IS_ERR(trans)) {
8496 ret = PTR_ERR(trans);
8501 btrfs_block_rsv_release(fs_info, rsv, -1, NULL);
8502 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
8503 rsv, min_size, false);
8505 * We have reserved 2 metadata units when we started the
8506 * transaction and min_size matches 1 unit, so this should never
8507 * fail, but if it does, it's not critical we just fail truncation.
8512 trans->block_rsv = rsv;
8516 * We can't call btrfs_truncate_block inside a trans handle as we could
8517 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8518 * know we've truncated everything except the last little bit, and can
8519 * do btrfs_truncate_block and then update the disk_i_size.
8521 if (ret == BTRFS_NEED_TRUNCATE_BLOCK) {
8522 btrfs_end_transaction(trans);
8523 btrfs_btree_balance_dirty(fs_info);
8525 ret = btrfs_truncate_block(inode, inode->vfs_inode.i_size, 0, 0);
8528 trans = btrfs_start_transaction(root, 1);
8529 if (IS_ERR(trans)) {
8530 ret = PTR_ERR(trans);
8533 btrfs_inode_safe_disk_i_size_write(inode, 0);
8539 trans->block_rsv = &fs_info->trans_block_rsv;
8540 ret2 = btrfs_update_inode(trans, root, inode);
8544 ret2 = btrfs_end_transaction(trans);
8547 btrfs_btree_balance_dirty(fs_info);
8550 btrfs_free_block_rsv(fs_info, rsv);
8552 * So if we truncate and then write and fsync we normally would just
8553 * write the extents that changed, which is a problem if we need to
8554 * first truncate that entire inode. So set this flag so we write out
8555 * all of the extents in the inode to the sync log so we're completely
8558 * If no extents were dropped or trimmed we don't need to force the next
8559 * fsync to truncate all the inode's items from the log and re-log them
8560 * all. This means the truncate operation did not change the file size,
8561 * or changed it to a smaller size but there was only an implicit hole
8562 * between the old i_size and the new i_size, and there were no prealloc
8563 * extents beyond i_size to drop.
8565 if (control.extents_found > 0)
8566 btrfs_set_inode_full_sync(inode);
8571 struct inode *btrfs_new_subvol_inode(struct mnt_idmap *idmap,
8574 struct inode *inode;
8576 inode = new_inode(dir->i_sb);
8579 * Subvolumes don't inherit the sgid bit or the parent's gid if
8580 * the parent's sgid bit is set. This is probably a bug.
8582 inode_init_owner(idmap, inode, NULL,
8583 S_IFDIR | (~current_umask() & S_IRWXUGO));
8584 inode->i_op = &btrfs_dir_inode_operations;
8585 inode->i_fop = &btrfs_dir_file_operations;
8590 struct inode *btrfs_alloc_inode(struct super_block *sb)
8592 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
8593 struct btrfs_inode *ei;
8594 struct inode *inode;
8596 ei = alloc_inode_sb(sb, btrfs_inode_cachep, GFP_KERNEL);
8603 ei->last_sub_trans = 0;
8604 ei->logged_trans = 0;
8605 ei->delalloc_bytes = 0;
8606 ei->new_delalloc_bytes = 0;
8607 ei->defrag_bytes = 0;
8608 ei->disk_i_size = 0;
8612 ei->index_cnt = (u64)-1;
8614 ei->last_unlink_trans = 0;
8615 ei->last_reflink_trans = 0;
8616 ei->last_log_commit = 0;
8618 spin_lock_init(&ei->lock);
8619 ei->outstanding_extents = 0;
8620 if (sb->s_magic != BTRFS_TEST_MAGIC)
8621 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
8622 BTRFS_BLOCK_RSV_DELALLOC);
8623 ei->runtime_flags = 0;
8624 ei->prop_compress = BTRFS_COMPRESS_NONE;
8625 ei->defrag_compress = BTRFS_COMPRESS_NONE;
8627 ei->delayed_node = NULL;
8629 ei->i_otime.tv_sec = 0;
8630 ei->i_otime.tv_nsec = 0;
8632 inode = &ei->vfs_inode;
8633 extent_map_tree_init(&ei->extent_tree);
8634 extent_io_tree_init(fs_info, &ei->io_tree, IO_TREE_INODE_IO);
8635 ei->io_tree.inode = ei;
8636 extent_io_tree_init(fs_info, &ei->file_extent_tree,
8637 IO_TREE_INODE_FILE_EXTENT);
8638 mutex_init(&ei->log_mutex);
8639 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8640 INIT_LIST_HEAD(&ei->delalloc_inodes);
8641 INIT_LIST_HEAD(&ei->delayed_iput);
8642 RB_CLEAR_NODE(&ei->rb_node);
8643 init_rwsem(&ei->i_mmap_lock);
8648 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8649 void btrfs_test_destroy_inode(struct inode *inode)
8651 btrfs_drop_extent_map_range(BTRFS_I(inode), 0, (u64)-1, false);
8652 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8656 void btrfs_free_inode(struct inode *inode)
8658 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8661 void btrfs_destroy_inode(struct inode *vfs_inode)
8663 struct btrfs_ordered_extent *ordered;
8664 struct btrfs_inode *inode = BTRFS_I(vfs_inode);
8665 struct btrfs_root *root = inode->root;
8666 bool freespace_inode;
8668 WARN_ON(!hlist_empty(&vfs_inode->i_dentry));
8669 WARN_ON(vfs_inode->i_data.nrpages);
8670 WARN_ON(inode->block_rsv.reserved);
8671 WARN_ON(inode->block_rsv.size);
8672 WARN_ON(inode->outstanding_extents);
8673 if (!S_ISDIR(vfs_inode->i_mode)) {
8674 WARN_ON(inode->delalloc_bytes);
8675 WARN_ON(inode->new_delalloc_bytes);
8677 WARN_ON(inode->csum_bytes);
8678 WARN_ON(inode->defrag_bytes);
8681 * This can happen where we create an inode, but somebody else also
8682 * created the same inode and we need to destroy the one we already
8689 * If this is a free space inode do not take the ordered extents lockdep
8692 freespace_inode = btrfs_is_free_space_inode(inode);
8695 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8699 btrfs_err(root->fs_info,
8700 "found ordered extent %llu %llu on inode cleanup",
8701 ordered->file_offset, ordered->num_bytes);
8703 if (!freespace_inode)
8704 btrfs_lockdep_acquire(root->fs_info, btrfs_ordered_extent);
8706 btrfs_remove_ordered_extent(inode, ordered);
8707 btrfs_put_ordered_extent(ordered);
8708 btrfs_put_ordered_extent(ordered);
8711 btrfs_qgroup_check_reserved_leak(inode);
8712 inode_tree_del(inode);
8713 btrfs_drop_extent_map_range(inode, 0, (u64)-1, false);
8714 btrfs_inode_clear_file_extent_range(inode, 0, (u64)-1);
8715 btrfs_put_root(inode->root);
8718 int btrfs_drop_inode(struct inode *inode)
8720 struct btrfs_root *root = BTRFS_I(inode)->root;
8725 /* the snap/subvol tree is on deleting */
8726 if (btrfs_root_refs(&root->root_item) == 0)
8729 return generic_drop_inode(inode);
8732 static void init_once(void *foo)
8734 struct btrfs_inode *ei = foo;
8736 inode_init_once(&ei->vfs_inode);
8739 void __cold btrfs_destroy_cachep(void)
8742 * Make sure all delayed rcu free inodes are flushed before we
8746 bioset_exit(&btrfs_dio_bioset);
8747 kmem_cache_destroy(btrfs_inode_cachep);
8750 int __init btrfs_init_cachep(void)
8752 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8753 sizeof(struct btrfs_inode), 0,
8754 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
8756 if (!btrfs_inode_cachep)
8759 if (bioset_init(&btrfs_dio_bioset, BIO_POOL_SIZE,
8760 offsetof(struct btrfs_dio_private, bbio.bio),
8766 btrfs_destroy_cachep();
8770 static int btrfs_getattr(struct mnt_idmap *idmap,
8771 const struct path *path, struct kstat *stat,
8772 u32 request_mask, unsigned int flags)
8776 struct inode *inode = d_inode(path->dentry);
8777 u32 blocksize = inode->i_sb->s_blocksize;
8778 u32 bi_flags = BTRFS_I(inode)->flags;
8779 u32 bi_ro_flags = BTRFS_I(inode)->ro_flags;
8781 stat->result_mask |= STATX_BTIME;
8782 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
8783 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
8784 if (bi_flags & BTRFS_INODE_APPEND)
8785 stat->attributes |= STATX_ATTR_APPEND;
8786 if (bi_flags & BTRFS_INODE_COMPRESS)
8787 stat->attributes |= STATX_ATTR_COMPRESSED;
8788 if (bi_flags & BTRFS_INODE_IMMUTABLE)
8789 stat->attributes |= STATX_ATTR_IMMUTABLE;
8790 if (bi_flags & BTRFS_INODE_NODUMP)
8791 stat->attributes |= STATX_ATTR_NODUMP;
8792 if (bi_ro_flags & BTRFS_INODE_RO_VERITY)
8793 stat->attributes |= STATX_ATTR_VERITY;
8795 stat->attributes_mask |= (STATX_ATTR_APPEND |
8796 STATX_ATTR_COMPRESSED |
8797 STATX_ATTR_IMMUTABLE |
8800 generic_fillattr(idmap, inode, stat);
8801 stat->dev = BTRFS_I(inode)->root->anon_dev;
8803 spin_lock(&BTRFS_I(inode)->lock);
8804 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
8805 inode_bytes = inode_get_bytes(inode);
8806 spin_unlock(&BTRFS_I(inode)->lock);
8807 stat->blocks = (ALIGN(inode_bytes, blocksize) +
8808 ALIGN(delalloc_bytes, blocksize)) >> SECTOR_SHIFT;
8812 static int btrfs_rename_exchange(struct inode *old_dir,
8813 struct dentry *old_dentry,
8814 struct inode *new_dir,
8815 struct dentry *new_dentry)
8817 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
8818 struct btrfs_trans_handle *trans;
8819 unsigned int trans_num_items;
8820 struct btrfs_root *root = BTRFS_I(old_dir)->root;
8821 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8822 struct inode *new_inode = new_dentry->d_inode;
8823 struct inode *old_inode = old_dentry->d_inode;
8824 struct timespec64 ctime = current_time(old_inode);
8825 struct btrfs_rename_ctx old_rename_ctx;
8826 struct btrfs_rename_ctx new_rename_ctx;
8827 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
8828 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
8833 bool need_abort = false;
8834 struct fscrypt_name old_fname, new_fname;
8835 struct fscrypt_str *old_name, *new_name;
8838 * For non-subvolumes allow exchange only within one subvolume, in the
8839 * same inode namespace. Two subvolumes (represented as directory) can
8840 * be exchanged as they're a logical link and have a fixed inode number.
8843 (old_ino != BTRFS_FIRST_FREE_OBJECTID ||
8844 new_ino != BTRFS_FIRST_FREE_OBJECTID))
8847 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
8851 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
8853 fscrypt_free_filename(&old_fname);
8857 old_name = &old_fname.disk_name;
8858 new_name = &new_fname.disk_name;
8860 /* close the race window with snapshot create/destroy ioctl */
8861 if (old_ino == BTRFS_FIRST_FREE_OBJECTID ||
8862 new_ino == BTRFS_FIRST_FREE_OBJECTID)
8863 down_read(&fs_info->subvol_sem);
8867 * 1 to remove old dir item
8868 * 1 to remove old dir index
8869 * 1 to add new dir item
8870 * 1 to add new dir index
8871 * 1 to update parent inode
8873 * If the parents are the same, we only need to account for one
8875 trans_num_items = (old_dir == new_dir ? 9 : 10);
8876 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8878 * 1 to remove old root ref
8879 * 1 to remove old root backref
8880 * 1 to add new root ref
8881 * 1 to add new root backref
8883 trans_num_items += 4;
8886 * 1 to update inode item
8887 * 1 to remove old inode ref
8888 * 1 to add new inode ref
8890 trans_num_items += 3;
8892 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
8893 trans_num_items += 4;
8895 trans_num_items += 3;
8896 trans = btrfs_start_transaction(root, trans_num_items);
8897 if (IS_ERR(trans)) {
8898 ret = PTR_ERR(trans);
8903 ret = btrfs_record_root_in_trans(trans, dest);
8909 * We need to find a free sequence number both in the source and
8910 * in the destination directory for the exchange.
8912 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
8915 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
8919 BTRFS_I(old_inode)->dir_index = 0ULL;
8920 BTRFS_I(new_inode)->dir_index = 0ULL;
8922 /* Reference for the source. */
8923 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8924 /* force full log commit if subvolume involved. */
8925 btrfs_set_log_full_commit(trans);
8927 ret = btrfs_insert_inode_ref(trans, dest, new_name, old_ino,
8928 btrfs_ino(BTRFS_I(new_dir)),
8935 /* And now for the dest. */
8936 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8937 /* force full log commit if subvolume involved. */
8938 btrfs_set_log_full_commit(trans);
8940 ret = btrfs_insert_inode_ref(trans, root, old_name, new_ino,
8941 btrfs_ino(BTRFS_I(old_dir)),
8945 btrfs_abort_transaction(trans, ret);
8950 /* Update inode version and ctime/mtime. */
8951 inode_inc_iversion(old_dir);
8952 inode_inc_iversion(new_dir);
8953 inode_inc_iversion(old_inode);
8954 inode_inc_iversion(new_inode);
8955 old_dir->i_mtime = ctime;
8956 old_dir->i_ctime = ctime;
8957 new_dir->i_mtime = ctime;
8958 new_dir->i_ctime = ctime;
8959 old_inode->i_ctime = ctime;
8960 new_inode->i_ctime = ctime;
8962 if (old_dentry->d_parent != new_dentry->d_parent) {
8963 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
8964 BTRFS_I(old_inode), true);
8965 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
8966 BTRFS_I(new_inode), true);
8969 /* src is a subvolume */
8970 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
8971 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
8972 } else { /* src is an inode */
8973 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
8974 BTRFS_I(old_dentry->d_inode),
8975 old_name, &old_rename_ctx);
8977 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
8980 btrfs_abort_transaction(trans, ret);
8984 /* dest is a subvolume */
8985 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
8986 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
8987 } else { /* dest is an inode */
8988 ret = __btrfs_unlink_inode(trans, BTRFS_I(new_dir),
8989 BTRFS_I(new_dentry->d_inode),
8990 new_name, &new_rename_ctx);
8992 ret = btrfs_update_inode(trans, dest, BTRFS_I(new_inode));
8995 btrfs_abort_transaction(trans, ret);
8999 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9000 new_name, 0, old_idx);
9002 btrfs_abort_transaction(trans, ret);
9006 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9007 old_name, 0, new_idx);
9009 btrfs_abort_transaction(trans, ret);
9013 if (old_inode->i_nlink == 1)
9014 BTRFS_I(old_inode)->dir_index = old_idx;
9015 if (new_inode->i_nlink == 1)
9016 BTRFS_I(new_inode)->dir_index = new_idx;
9019 * Now pin the logs of the roots. We do it to ensure that no other task
9020 * can sync the logs while we are in progress with the rename, because
9021 * that could result in an inconsistency in case any of the inodes that
9022 * are part of this rename operation were logged before.
9024 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9025 btrfs_pin_log_trans(root);
9026 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9027 btrfs_pin_log_trans(dest);
9029 /* Do the log updates for all inodes. */
9030 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9031 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9032 old_rename_ctx.index, new_dentry->d_parent);
9033 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9034 btrfs_log_new_name(trans, new_dentry, BTRFS_I(new_dir),
9035 new_rename_ctx.index, old_dentry->d_parent);
9037 /* Now unpin the logs. */
9038 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9039 btrfs_end_log_trans(root);
9040 if (new_ino != BTRFS_FIRST_FREE_OBJECTID)
9041 btrfs_end_log_trans(dest);
9043 ret2 = btrfs_end_transaction(trans);
9044 ret = ret ? ret : ret2;
9046 if (new_ino == BTRFS_FIRST_FREE_OBJECTID ||
9047 old_ino == BTRFS_FIRST_FREE_OBJECTID)
9048 up_read(&fs_info->subvol_sem);
9050 fscrypt_free_filename(&new_fname);
9051 fscrypt_free_filename(&old_fname);
9055 static struct inode *new_whiteout_inode(struct mnt_idmap *idmap,
9058 struct inode *inode;
9060 inode = new_inode(dir->i_sb);
9062 inode_init_owner(idmap, inode, dir,
9063 S_IFCHR | WHITEOUT_MODE);
9064 inode->i_op = &btrfs_special_inode_operations;
9065 init_special_inode(inode, inode->i_mode, WHITEOUT_DEV);
9070 static int btrfs_rename(struct mnt_idmap *idmap,
9071 struct inode *old_dir, struct dentry *old_dentry,
9072 struct inode *new_dir, struct dentry *new_dentry,
9075 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9076 struct btrfs_new_inode_args whiteout_args = {
9078 .dentry = old_dentry,
9080 struct btrfs_trans_handle *trans;
9081 unsigned int trans_num_items;
9082 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9083 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9084 struct inode *new_inode = d_inode(new_dentry);
9085 struct inode *old_inode = d_inode(old_dentry);
9086 struct btrfs_rename_ctx rename_ctx;
9090 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9091 struct fscrypt_name old_fname, new_fname;
9093 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9096 /* we only allow rename subvolume link between subvolumes */
9097 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9100 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9101 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9104 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9105 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9108 ret = fscrypt_setup_filename(old_dir, &old_dentry->d_name, 0, &old_fname);
9112 ret = fscrypt_setup_filename(new_dir, &new_dentry->d_name, 0, &new_fname);
9114 fscrypt_free_filename(&old_fname);
9118 /* check for collisions, even if the name isn't there */
9119 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino, &new_fname.disk_name);
9121 if (ret == -EEXIST) {
9123 * eexist without a new_inode */
9124 if (WARN_ON(!new_inode)) {
9125 goto out_fscrypt_names;
9128 /* maybe -EOVERFLOW */
9129 goto out_fscrypt_names;
9135 * we're using rename to replace one file with another. Start IO on it
9136 * now so we don't add too much work to the end of the transaction
9138 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9139 filemap_flush(old_inode->i_mapping);
9141 if (flags & RENAME_WHITEOUT) {
9142 whiteout_args.inode = new_whiteout_inode(idmap, old_dir);
9143 if (!whiteout_args.inode) {
9145 goto out_fscrypt_names;
9147 ret = btrfs_new_inode_prepare(&whiteout_args, &trans_num_items);
9149 goto out_whiteout_inode;
9151 /* 1 to update the old parent inode. */
9152 trans_num_items = 1;
9155 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9156 /* Close the race window with snapshot create/destroy ioctl */
9157 down_read(&fs_info->subvol_sem);
9159 * 1 to remove old root ref
9160 * 1 to remove old root backref
9161 * 1 to add new root ref
9162 * 1 to add new root backref
9164 trans_num_items += 4;
9168 * 1 to remove old inode ref
9169 * 1 to add new inode ref
9171 trans_num_items += 3;
9174 * 1 to remove old dir item
9175 * 1 to remove old dir index
9176 * 1 to add new dir item
9177 * 1 to add new dir index
9179 trans_num_items += 4;
9180 /* 1 to update new parent inode if it's not the same as the old parent */
9181 if (new_dir != old_dir)
9186 * 1 to remove inode ref
9187 * 1 to remove dir item
9188 * 1 to remove dir index
9189 * 1 to possibly add orphan item
9191 trans_num_items += 5;
9193 trans = btrfs_start_transaction(root, trans_num_items);
9194 if (IS_ERR(trans)) {
9195 ret = PTR_ERR(trans);
9200 ret = btrfs_record_root_in_trans(trans, dest);
9205 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9209 BTRFS_I(old_inode)->dir_index = 0ULL;
9210 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9211 /* force full log commit if subvolume involved. */
9212 btrfs_set_log_full_commit(trans);
9214 ret = btrfs_insert_inode_ref(trans, dest, &new_fname.disk_name,
9215 old_ino, btrfs_ino(BTRFS_I(new_dir)),
9221 inode_inc_iversion(old_dir);
9222 inode_inc_iversion(new_dir);
9223 inode_inc_iversion(old_inode);
9224 old_dir->i_mtime = current_time(old_dir);
9225 old_dir->i_ctime = old_dir->i_mtime;
9226 new_dir->i_mtime = old_dir->i_mtime;
9227 new_dir->i_ctime = old_dir->i_mtime;
9228 old_inode->i_ctime = old_dir->i_mtime;
9230 if (old_dentry->d_parent != new_dentry->d_parent)
9231 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9232 BTRFS_I(old_inode), true);
9234 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9235 ret = btrfs_unlink_subvol(trans, BTRFS_I(old_dir), old_dentry);
9237 ret = __btrfs_unlink_inode(trans, BTRFS_I(old_dir),
9238 BTRFS_I(d_inode(old_dentry)),
9239 &old_fname.disk_name, &rename_ctx);
9241 ret = btrfs_update_inode(trans, root, BTRFS_I(old_inode));
9244 btrfs_abort_transaction(trans, ret);
9249 inode_inc_iversion(new_inode);
9250 new_inode->i_ctime = current_time(new_inode);
9251 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9252 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9253 ret = btrfs_unlink_subvol(trans, BTRFS_I(new_dir), new_dentry);
9254 BUG_ON(new_inode->i_nlink == 0);
9256 ret = btrfs_unlink_inode(trans, BTRFS_I(new_dir),
9257 BTRFS_I(d_inode(new_dentry)),
9258 &new_fname.disk_name);
9260 if (!ret && new_inode->i_nlink == 0)
9261 ret = btrfs_orphan_add(trans,
9262 BTRFS_I(d_inode(new_dentry)));
9264 btrfs_abort_transaction(trans, ret);
9269 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9270 &new_fname.disk_name, 0, index);
9272 btrfs_abort_transaction(trans, ret);
9276 if (old_inode->i_nlink == 1)
9277 BTRFS_I(old_inode)->dir_index = index;
9279 if (old_ino != BTRFS_FIRST_FREE_OBJECTID)
9280 btrfs_log_new_name(trans, old_dentry, BTRFS_I(old_dir),
9281 rename_ctx.index, new_dentry->d_parent);
9283 if (flags & RENAME_WHITEOUT) {
9284 ret = btrfs_create_new_inode(trans, &whiteout_args);
9286 btrfs_abort_transaction(trans, ret);
9289 unlock_new_inode(whiteout_args.inode);
9290 iput(whiteout_args.inode);
9291 whiteout_args.inode = NULL;
9295 ret2 = btrfs_end_transaction(trans);
9296 ret = ret ? ret : ret2;
9298 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9299 up_read(&fs_info->subvol_sem);
9300 if (flags & RENAME_WHITEOUT)
9301 btrfs_new_inode_args_destroy(&whiteout_args);
9303 if (flags & RENAME_WHITEOUT)
9304 iput(whiteout_args.inode);
9306 fscrypt_free_filename(&old_fname);
9307 fscrypt_free_filename(&new_fname);
9311 static int btrfs_rename2(struct mnt_idmap *idmap, struct inode *old_dir,
9312 struct dentry *old_dentry, struct inode *new_dir,
9313 struct dentry *new_dentry, unsigned int flags)
9317 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9320 if (flags & RENAME_EXCHANGE)
9321 ret = btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9324 ret = btrfs_rename(idmap, old_dir, old_dentry, new_dir,
9327 btrfs_btree_balance_dirty(BTRFS_I(new_dir)->root->fs_info);
9332 struct btrfs_delalloc_work {
9333 struct inode *inode;
9334 struct completion completion;
9335 struct list_head list;
9336 struct btrfs_work work;
9339 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9341 struct btrfs_delalloc_work *delalloc_work;
9342 struct inode *inode;
9344 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9346 inode = delalloc_work->inode;
9347 filemap_flush(inode->i_mapping);
9348 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9349 &BTRFS_I(inode)->runtime_flags))
9350 filemap_flush(inode->i_mapping);
9353 complete(&delalloc_work->completion);
9356 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9358 struct btrfs_delalloc_work *work;
9360 work = kmalloc(sizeof(*work), GFP_NOFS);
9364 init_completion(&work->completion);
9365 INIT_LIST_HEAD(&work->list);
9366 work->inode = inode;
9367 btrfs_init_work(&work->work, btrfs_run_delalloc_work, NULL, NULL);
9373 * some fairly slow code that needs optimization. This walks the list
9374 * of all the inodes with pending delalloc and forces them to disk.
9376 static int start_delalloc_inodes(struct btrfs_root *root,
9377 struct writeback_control *wbc, bool snapshot,
9378 bool in_reclaim_context)
9380 struct btrfs_inode *binode;
9381 struct inode *inode;
9382 struct btrfs_delalloc_work *work, *next;
9383 struct list_head works;
9384 struct list_head splice;
9386 bool full_flush = wbc->nr_to_write == LONG_MAX;
9388 INIT_LIST_HEAD(&works);
9389 INIT_LIST_HEAD(&splice);
9391 mutex_lock(&root->delalloc_mutex);
9392 spin_lock(&root->delalloc_lock);
9393 list_splice_init(&root->delalloc_inodes, &splice);
9394 while (!list_empty(&splice)) {
9395 binode = list_entry(splice.next, struct btrfs_inode,
9398 list_move_tail(&binode->delalloc_inodes,
9399 &root->delalloc_inodes);
9401 if (in_reclaim_context &&
9402 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &binode->runtime_flags))
9405 inode = igrab(&binode->vfs_inode);
9407 cond_resched_lock(&root->delalloc_lock);
9410 spin_unlock(&root->delalloc_lock);
9413 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH,
9414 &binode->runtime_flags);
9416 work = btrfs_alloc_delalloc_work(inode);
9422 list_add_tail(&work->list, &works);
9423 btrfs_queue_work(root->fs_info->flush_workers,
9426 ret = filemap_fdatawrite_wbc(inode->i_mapping, wbc);
9427 btrfs_add_delayed_iput(BTRFS_I(inode));
9428 if (ret || wbc->nr_to_write <= 0)
9432 spin_lock(&root->delalloc_lock);
9434 spin_unlock(&root->delalloc_lock);
9437 list_for_each_entry_safe(work, next, &works, list) {
9438 list_del_init(&work->list);
9439 wait_for_completion(&work->completion);
9443 if (!list_empty(&splice)) {
9444 spin_lock(&root->delalloc_lock);
9445 list_splice_tail(&splice, &root->delalloc_inodes);
9446 spin_unlock(&root->delalloc_lock);
9448 mutex_unlock(&root->delalloc_mutex);
9452 int btrfs_start_delalloc_snapshot(struct btrfs_root *root, bool in_reclaim_context)
9454 struct writeback_control wbc = {
9455 .nr_to_write = LONG_MAX,
9456 .sync_mode = WB_SYNC_NONE,
9458 .range_end = LLONG_MAX,
9460 struct btrfs_fs_info *fs_info = root->fs_info;
9462 if (BTRFS_FS_ERROR(fs_info))
9465 return start_delalloc_inodes(root, &wbc, true, in_reclaim_context);
9468 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, long nr,
9469 bool in_reclaim_context)
9471 struct writeback_control wbc = {
9473 .sync_mode = WB_SYNC_NONE,
9475 .range_end = LLONG_MAX,
9477 struct btrfs_root *root;
9478 struct list_head splice;
9481 if (BTRFS_FS_ERROR(fs_info))
9484 INIT_LIST_HEAD(&splice);
9486 mutex_lock(&fs_info->delalloc_root_mutex);
9487 spin_lock(&fs_info->delalloc_root_lock);
9488 list_splice_init(&fs_info->delalloc_roots, &splice);
9489 while (!list_empty(&splice)) {
9491 * Reset nr_to_write here so we know that we're doing a full
9495 wbc.nr_to_write = LONG_MAX;
9497 root = list_first_entry(&splice, struct btrfs_root,
9499 root = btrfs_grab_root(root);
9501 list_move_tail(&root->delalloc_root,
9502 &fs_info->delalloc_roots);
9503 spin_unlock(&fs_info->delalloc_root_lock);
9505 ret = start_delalloc_inodes(root, &wbc, false, in_reclaim_context);
9506 btrfs_put_root(root);
9507 if (ret < 0 || wbc.nr_to_write <= 0)
9509 spin_lock(&fs_info->delalloc_root_lock);
9511 spin_unlock(&fs_info->delalloc_root_lock);
9515 if (!list_empty(&splice)) {
9516 spin_lock(&fs_info->delalloc_root_lock);
9517 list_splice_tail(&splice, &fs_info->delalloc_roots);
9518 spin_unlock(&fs_info->delalloc_root_lock);
9520 mutex_unlock(&fs_info->delalloc_root_mutex);
9524 static int btrfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
9525 struct dentry *dentry, const char *symname)
9527 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9528 struct btrfs_trans_handle *trans;
9529 struct btrfs_root *root = BTRFS_I(dir)->root;
9530 struct btrfs_path *path;
9531 struct btrfs_key key;
9532 struct inode *inode;
9533 struct btrfs_new_inode_args new_inode_args = {
9537 unsigned int trans_num_items;
9542 struct btrfs_file_extent_item *ei;
9543 struct extent_buffer *leaf;
9545 name_len = strlen(symname);
9546 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
9547 return -ENAMETOOLONG;
9549 inode = new_inode(dir->i_sb);
9552 inode_init_owner(idmap, inode, dir, S_IFLNK | S_IRWXUGO);
9553 inode->i_op = &btrfs_symlink_inode_operations;
9554 inode_nohighmem(inode);
9555 inode->i_mapping->a_ops = &btrfs_aops;
9556 btrfs_i_size_write(BTRFS_I(inode), name_len);
9557 inode_set_bytes(inode, name_len);
9559 new_inode_args.inode = inode;
9560 err = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9563 /* 1 additional item for the inline extent */
9566 trans = btrfs_start_transaction(root, trans_num_items);
9567 if (IS_ERR(trans)) {
9568 err = PTR_ERR(trans);
9569 goto out_new_inode_args;
9572 err = btrfs_create_new_inode(trans, &new_inode_args);
9576 path = btrfs_alloc_path();
9579 btrfs_abort_transaction(trans, err);
9580 discard_new_inode(inode);
9584 key.objectid = btrfs_ino(BTRFS_I(inode));
9586 key.type = BTRFS_EXTENT_DATA_KEY;
9587 datasize = btrfs_file_extent_calc_inline_size(name_len);
9588 err = btrfs_insert_empty_item(trans, root, path, &key,
9591 btrfs_abort_transaction(trans, err);
9592 btrfs_free_path(path);
9593 discard_new_inode(inode);
9597 leaf = path->nodes[0];
9598 ei = btrfs_item_ptr(leaf, path->slots[0],
9599 struct btrfs_file_extent_item);
9600 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9601 btrfs_set_file_extent_type(leaf, ei,
9602 BTRFS_FILE_EXTENT_INLINE);
9603 btrfs_set_file_extent_encryption(leaf, ei, 0);
9604 btrfs_set_file_extent_compression(leaf, ei, 0);
9605 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9606 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9608 ptr = btrfs_file_extent_inline_start(ei);
9609 write_extent_buffer(leaf, symname, ptr, name_len);
9610 btrfs_mark_buffer_dirty(leaf);
9611 btrfs_free_path(path);
9613 d_instantiate_new(dentry, inode);
9616 btrfs_end_transaction(trans);
9617 btrfs_btree_balance_dirty(fs_info);
9619 btrfs_new_inode_args_destroy(&new_inode_args);
9626 static struct btrfs_trans_handle *insert_prealloc_file_extent(
9627 struct btrfs_trans_handle *trans_in,
9628 struct btrfs_inode *inode,
9629 struct btrfs_key *ins,
9632 struct btrfs_file_extent_item stack_fi;
9633 struct btrfs_replace_extent_info extent_info;
9634 struct btrfs_trans_handle *trans = trans_in;
9635 struct btrfs_path *path;
9636 u64 start = ins->objectid;
9637 u64 len = ins->offset;
9638 int qgroup_released;
9641 memset(&stack_fi, 0, sizeof(stack_fi));
9643 btrfs_set_stack_file_extent_type(&stack_fi, BTRFS_FILE_EXTENT_PREALLOC);
9644 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi, start);
9645 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi, len);
9646 btrfs_set_stack_file_extent_num_bytes(&stack_fi, len);
9647 btrfs_set_stack_file_extent_ram_bytes(&stack_fi, len);
9648 btrfs_set_stack_file_extent_compression(&stack_fi, BTRFS_COMPRESS_NONE);
9649 /* Encryption and other encoding is reserved and all 0 */
9651 qgroup_released = btrfs_qgroup_release_data(inode, file_offset, len);
9652 if (qgroup_released < 0)
9653 return ERR_PTR(qgroup_released);
9656 ret = insert_reserved_file_extent(trans, inode,
9657 file_offset, &stack_fi,
9658 true, qgroup_released);
9664 extent_info.disk_offset = start;
9665 extent_info.disk_len = len;
9666 extent_info.data_offset = 0;
9667 extent_info.data_len = len;
9668 extent_info.file_offset = file_offset;
9669 extent_info.extent_buf = (char *)&stack_fi;
9670 extent_info.is_new_extent = true;
9671 extent_info.update_times = true;
9672 extent_info.qgroup_reserved = qgroup_released;
9673 extent_info.insertions = 0;
9675 path = btrfs_alloc_path();
9681 ret = btrfs_replace_file_extents(inode, path, file_offset,
9682 file_offset + len - 1, &extent_info,
9684 btrfs_free_path(path);
9691 * We have released qgroup data range at the beginning of the function,
9692 * and normally qgroup_released bytes will be freed when committing
9694 * But if we error out early, we have to free what we have released
9695 * or we leak qgroup data reservation.
9697 btrfs_qgroup_free_refroot(inode->root->fs_info,
9698 inode->root->root_key.objectid, qgroup_released,
9699 BTRFS_QGROUP_RSV_DATA);
9700 return ERR_PTR(ret);
9703 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9704 u64 start, u64 num_bytes, u64 min_size,
9705 loff_t actual_len, u64 *alloc_hint,
9706 struct btrfs_trans_handle *trans)
9708 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9709 struct extent_map *em;
9710 struct btrfs_root *root = BTRFS_I(inode)->root;
9711 struct btrfs_key ins;
9712 u64 cur_offset = start;
9713 u64 clear_offset = start;
9716 u64 last_alloc = (u64)-1;
9718 bool own_trans = true;
9719 u64 end = start + num_bytes - 1;
9723 while (num_bytes > 0) {
9724 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9725 cur_bytes = max(cur_bytes, min_size);
9727 * If we are severely fragmented we could end up with really
9728 * small allocations, so if the allocator is returning small
9729 * chunks lets make its job easier by only searching for those
9732 cur_bytes = min(cur_bytes, last_alloc);
9733 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
9734 min_size, 0, *alloc_hint, &ins, 1, 0);
9739 * We've reserved this space, and thus converted it from
9740 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9741 * from here on out we will only need to clear our reservation
9742 * for the remaining unreserved area, so advance our
9743 * clear_offset by our extent size.
9745 clear_offset += ins.offset;
9747 last_alloc = ins.offset;
9748 trans = insert_prealloc_file_extent(trans, BTRFS_I(inode),
9751 * Now that we inserted the prealloc extent we can finally
9752 * decrement the number of reservations in the block group.
9753 * If we did it before, we could race with relocation and have
9754 * relocation miss the reserved extent, making it fail later.
9756 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
9757 if (IS_ERR(trans)) {
9758 ret = PTR_ERR(trans);
9759 btrfs_free_reserved_extent(fs_info, ins.objectid,
9764 em = alloc_extent_map();
9766 btrfs_drop_extent_map_range(BTRFS_I(inode), cur_offset,
9767 cur_offset + ins.offset - 1, false);
9768 btrfs_set_inode_full_sync(BTRFS_I(inode));
9772 em->start = cur_offset;
9773 em->orig_start = cur_offset;
9774 em->len = ins.offset;
9775 em->block_start = ins.objectid;
9776 em->block_len = ins.offset;
9777 em->orig_block_len = ins.offset;
9778 em->ram_bytes = ins.offset;
9779 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9780 em->generation = trans->transid;
9782 ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, true);
9783 free_extent_map(em);
9785 num_bytes -= ins.offset;
9786 cur_offset += ins.offset;
9787 *alloc_hint = ins.objectid + ins.offset;
9789 inode_inc_iversion(inode);
9790 inode->i_ctime = current_time(inode);
9791 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9792 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9793 (actual_len > inode->i_size) &&
9794 (cur_offset > inode->i_size)) {
9795 if (cur_offset > actual_len)
9796 i_size = actual_len;
9798 i_size = cur_offset;
9799 i_size_write(inode, i_size);
9800 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
9803 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
9806 btrfs_abort_transaction(trans, ret);
9808 btrfs_end_transaction(trans);
9813 btrfs_end_transaction(trans);
9817 if (clear_offset < end)
9818 btrfs_free_reserved_data_space(BTRFS_I(inode), NULL, clear_offset,
9819 end - clear_offset + 1);
9823 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9824 u64 start, u64 num_bytes, u64 min_size,
9825 loff_t actual_len, u64 *alloc_hint)
9827 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9828 min_size, actual_len, alloc_hint,
9832 int btrfs_prealloc_file_range_trans(struct inode *inode,
9833 struct btrfs_trans_handle *trans, int mode,
9834 u64 start, u64 num_bytes, u64 min_size,
9835 loff_t actual_len, u64 *alloc_hint)
9837 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9838 min_size, actual_len, alloc_hint, trans);
9841 static int btrfs_permission(struct mnt_idmap *idmap,
9842 struct inode *inode, int mask)
9844 struct btrfs_root *root = BTRFS_I(inode)->root;
9845 umode_t mode = inode->i_mode;
9847 if (mask & MAY_WRITE &&
9848 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9849 if (btrfs_root_readonly(root))
9851 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9854 return generic_permission(idmap, inode, mask);
9857 static int btrfs_tmpfile(struct mnt_idmap *idmap, struct inode *dir,
9858 struct file *file, umode_t mode)
9860 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
9861 struct btrfs_trans_handle *trans;
9862 struct btrfs_root *root = BTRFS_I(dir)->root;
9863 struct inode *inode;
9864 struct btrfs_new_inode_args new_inode_args = {
9866 .dentry = file->f_path.dentry,
9869 unsigned int trans_num_items;
9872 inode = new_inode(dir->i_sb);
9875 inode_init_owner(idmap, inode, dir, mode);
9876 inode->i_fop = &btrfs_file_operations;
9877 inode->i_op = &btrfs_file_inode_operations;
9878 inode->i_mapping->a_ops = &btrfs_aops;
9880 new_inode_args.inode = inode;
9881 ret = btrfs_new_inode_prepare(&new_inode_args, &trans_num_items);
9885 trans = btrfs_start_transaction(root, trans_num_items);
9886 if (IS_ERR(trans)) {
9887 ret = PTR_ERR(trans);
9888 goto out_new_inode_args;
9891 ret = btrfs_create_new_inode(trans, &new_inode_args);
9894 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9895 * set it to 1 because d_tmpfile() will issue a warning if the count is
9898 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9900 set_nlink(inode, 1);
9903 d_tmpfile(file, inode);
9904 unlock_new_inode(inode);
9905 mark_inode_dirty(inode);
9908 btrfs_end_transaction(trans);
9909 btrfs_btree_balance_dirty(fs_info);
9911 btrfs_new_inode_args_destroy(&new_inode_args);
9915 return finish_open_simple(file, ret);
9918 void btrfs_set_range_writeback(struct btrfs_inode *inode, u64 start, u64 end)
9920 struct btrfs_fs_info *fs_info = inode->root->fs_info;
9921 unsigned long index = start >> PAGE_SHIFT;
9922 unsigned long end_index = end >> PAGE_SHIFT;
9926 ASSERT(end + 1 - start <= U32_MAX);
9927 len = end + 1 - start;
9928 while (index <= end_index) {
9929 page = find_get_page(inode->vfs_inode.i_mapping, index);
9930 ASSERT(page); /* Pages should be in the extent_io_tree */
9932 btrfs_page_set_writeback(fs_info, page, start, len);
9938 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info *fs_info,
9941 switch (compress_type) {
9942 case BTRFS_COMPRESS_NONE:
9943 return BTRFS_ENCODED_IO_COMPRESSION_NONE;
9944 case BTRFS_COMPRESS_ZLIB:
9945 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB;
9946 case BTRFS_COMPRESS_LZO:
9948 * The LZO format depends on the sector size. 64K is the maximum
9949 * sector size that we support.
9951 if (fs_info->sectorsize < SZ_4K || fs_info->sectorsize > SZ_64K)
9953 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K +
9954 (fs_info->sectorsize_bits - 12);
9955 case BTRFS_COMPRESS_ZSTD:
9956 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD;
9962 static ssize_t btrfs_encoded_read_inline(
9964 struct iov_iter *iter, u64 start,
9966 struct extent_state **cached_state,
9967 u64 extent_start, size_t count,
9968 struct btrfs_ioctl_encoded_io_args *encoded,
9971 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
9972 struct btrfs_root *root = inode->root;
9973 struct btrfs_fs_info *fs_info = root->fs_info;
9974 struct extent_io_tree *io_tree = &inode->io_tree;
9975 struct btrfs_path *path;
9976 struct extent_buffer *leaf;
9977 struct btrfs_file_extent_item *item;
9983 path = btrfs_alloc_path();
9988 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
9992 /* The extent item disappeared? */
9997 leaf = path->nodes[0];
9998 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_file_extent_item);
10000 ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
10001 ptr = btrfs_file_extent_inline_start(item);
10003 encoded->len = min_t(u64, extent_start + ram_bytes,
10004 inode->vfs_inode.i_size) - iocb->ki_pos;
10005 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10006 btrfs_file_extent_compression(leaf, item));
10009 encoded->compression = ret;
10010 if (encoded->compression) {
10011 size_t inline_size;
10013 inline_size = btrfs_file_extent_inline_item_len(leaf,
10015 if (inline_size > count) {
10019 count = inline_size;
10020 encoded->unencoded_len = ram_bytes;
10021 encoded->unencoded_offset = iocb->ki_pos - extent_start;
10023 count = min_t(u64, count, encoded->len);
10024 encoded->len = count;
10025 encoded->unencoded_len = count;
10026 ptr += iocb->ki_pos - extent_start;
10029 tmp = kmalloc(count, GFP_NOFS);
10034 read_extent_buffer(leaf, tmp, ptr, count);
10035 btrfs_release_path(path);
10036 unlock_extent(io_tree, start, lockend, cached_state);
10037 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10040 ret = copy_to_iter(tmp, count, iter);
10045 btrfs_free_path(path);
10049 struct btrfs_encoded_read_private {
10050 wait_queue_head_t wait;
10052 blk_status_t status;
10055 static void btrfs_encoded_read_endio(struct btrfs_bio *bbio)
10057 struct btrfs_encoded_read_private *priv = bbio->private;
10059 if (bbio->bio.bi_status) {
10061 * The memory barrier implied by the atomic_dec_return() here
10062 * pairs with the memory barrier implied by the
10063 * atomic_dec_return() or io_wait_event() in
10064 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10065 * write is observed before the load of status in
10066 * btrfs_encoded_read_regular_fill_pages().
10068 WRITE_ONCE(priv->status, bbio->bio.bi_status);
10070 if (!atomic_dec_return(&priv->pending))
10071 wake_up(&priv->wait);
10072 bio_put(&bbio->bio);
10075 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode *inode,
10076 u64 file_offset, u64 disk_bytenr,
10077 u64 disk_io_size, struct page **pages)
10079 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10080 struct btrfs_encoded_read_private priv = {
10081 .pending = ATOMIC_INIT(1),
10083 unsigned long i = 0;
10084 struct btrfs_bio *bbio;
10086 init_waitqueue_head(&priv.wait);
10088 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10089 btrfs_encoded_read_endio, &priv);
10090 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10091 bbio->inode = inode;
10094 size_t bytes = min_t(u64, disk_io_size, PAGE_SIZE);
10096 if (bio_add_page(&bbio->bio, pages[i], bytes, 0) < bytes) {
10097 atomic_inc(&priv.pending);
10098 btrfs_submit_bio(bbio, 0);
10100 bbio = btrfs_bio_alloc(BIO_MAX_VECS, REQ_OP_READ, fs_info,
10101 btrfs_encoded_read_endio, &priv);
10102 bbio->bio.bi_iter.bi_sector = disk_bytenr >> SECTOR_SHIFT;
10103 bbio->inode = inode;
10108 disk_bytenr += bytes;
10109 disk_io_size -= bytes;
10110 } while (disk_io_size);
10112 atomic_inc(&priv.pending);
10113 btrfs_submit_bio(bbio, 0);
10115 if (atomic_dec_return(&priv.pending))
10116 io_wait_event(priv.wait, !atomic_read(&priv.pending));
10117 /* See btrfs_encoded_read_endio() for ordering. */
10118 return blk_status_to_errno(READ_ONCE(priv.status));
10121 static ssize_t btrfs_encoded_read_regular(struct kiocb *iocb,
10122 struct iov_iter *iter,
10123 u64 start, u64 lockend,
10124 struct extent_state **cached_state,
10125 u64 disk_bytenr, u64 disk_io_size,
10126 size_t count, bool compressed,
10129 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10130 struct extent_io_tree *io_tree = &inode->io_tree;
10131 struct page **pages;
10132 unsigned long nr_pages, i;
10134 size_t page_offset;
10137 nr_pages = DIV_ROUND_UP(disk_io_size, PAGE_SIZE);
10138 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
10141 ret = btrfs_alloc_page_array(nr_pages, pages);
10147 ret = btrfs_encoded_read_regular_fill_pages(inode, start, disk_bytenr,
10148 disk_io_size, pages);
10152 unlock_extent(io_tree, start, lockend, cached_state);
10153 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10160 i = (iocb->ki_pos - start) >> PAGE_SHIFT;
10161 page_offset = (iocb->ki_pos - start) & (PAGE_SIZE - 1);
10164 while (cur < count) {
10165 size_t bytes = min_t(size_t, count - cur,
10166 PAGE_SIZE - page_offset);
10168 if (copy_page_to_iter(pages[i], page_offset, bytes,
10179 for (i = 0; i < nr_pages; i++) {
10181 __free_page(pages[i]);
10187 ssize_t btrfs_encoded_read(struct kiocb *iocb, struct iov_iter *iter,
10188 struct btrfs_ioctl_encoded_io_args *encoded)
10190 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10191 struct btrfs_fs_info *fs_info = inode->root->fs_info;
10192 struct extent_io_tree *io_tree = &inode->io_tree;
10194 size_t count = iov_iter_count(iter);
10195 u64 start, lockend, disk_bytenr, disk_io_size;
10196 struct extent_state *cached_state = NULL;
10197 struct extent_map *em;
10198 bool unlocked = false;
10200 file_accessed(iocb->ki_filp);
10202 btrfs_inode_lock(inode, BTRFS_ILOCK_SHARED);
10204 if (iocb->ki_pos >= inode->vfs_inode.i_size) {
10205 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10208 start = ALIGN_DOWN(iocb->ki_pos, fs_info->sectorsize);
10210 * We don't know how long the extent containing iocb->ki_pos is, but if
10211 * it's compressed we know that it won't be longer than this.
10213 lockend = start + BTRFS_MAX_UNCOMPRESSED - 1;
10216 struct btrfs_ordered_extent *ordered;
10218 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start,
10219 lockend - start + 1);
10221 goto out_unlock_inode;
10222 lock_extent(io_tree, start, lockend, &cached_state);
10223 ordered = btrfs_lookup_ordered_range(inode, start,
10224 lockend - start + 1);
10227 btrfs_put_ordered_extent(ordered);
10228 unlock_extent(io_tree, start, lockend, &cached_state);
10232 em = btrfs_get_extent(inode, NULL, 0, start, lockend - start + 1);
10235 goto out_unlock_extent;
10238 if (em->block_start == EXTENT_MAP_INLINE) {
10239 u64 extent_start = em->start;
10242 * For inline extents we get everything we need out of the
10245 free_extent_map(em);
10247 ret = btrfs_encoded_read_inline(iocb, iter, start, lockend,
10248 &cached_state, extent_start,
10249 count, encoded, &unlocked);
10254 * We only want to return up to EOF even if the extent extends beyond
10257 encoded->len = min_t(u64, extent_map_end(em),
10258 inode->vfs_inode.i_size) - iocb->ki_pos;
10259 if (em->block_start == EXTENT_MAP_HOLE ||
10260 test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
10261 disk_bytenr = EXTENT_MAP_HOLE;
10262 count = min_t(u64, count, encoded->len);
10263 encoded->len = count;
10264 encoded->unencoded_len = count;
10265 } else if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10266 disk_bytenr = em->block_start;
10268 * Bail if the buffer isn't large enough to return the whole
10269 * compressed extent.
10271 if (em->block_len > count) {
10275 disk_io_size = em->block_len;
10276 count = em->block_len;
10277 encoded->unencoded_len = em->ram_bytes;
10278 encoded->unencoded_offset = iocb->ki_pos - em->orig_start;
10279 ret = btrfs_encoded_io_compression_from_extent(fs_info,
10280 em->compress_type);
10283 encoded->compression = ret;
10285 disk_bytenr = em->block_start + (start - em->start);
10286 if (encoded->len > count)
10287 encoded->len = count;
10289 * Don't read beyond what we locked. This also limits the page
10290 * allocations that we'll do.
10292 disk_io_size = min(lockend + 1, iocb->ki_pos + encoded->len) - start;
10293 count = start + disk_io_size - iocb->ki_pos;
10294 encoded->len = count;
10295 encoded->unencoded_len = count;
10296 disk_io_size = ALIGN(disk_io_size, fs_info->sectorsize);
10298 free_extent_map(em);
10301 if (disk_bytenr == EXTENT_MAP_HOLE) {
10302 unlock_extent(io_tree, start, lockend, &cached_state);
10303 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10305 ret = iov_iter_zero(count, iter);
10309 ret = btrfs_encoded_read_regular(iocb, iter, start, lockend,
10310 &cached_state, disk_bytenr,
10311 disk_io_size, count,
10312 encoded->compression,
10318 iocb->ki_pos += encoded->len;
10320 free_extent_map(em);
10323 unlock_extent(io_tree, start, lockend, &cached_state);
10326 btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
10330 ssize_t btrfs_do_encoded_write(struct kiocb *iocb, struct iov_iter *from,
10331 const struct btrfs_ioctl_encoded_io_args *encoded)
10333 struct btrfs_inode *inode = BTRFS_I(file_inode(iocb->ki_filp));
10334 struct btrfs_root *root = inode->root;
10335 struct btrfs_fs_info *fs_info = root->fs_info;
10336 struct extent_io_tree *io_tree = &inode->io_tree;
10337 struct extent_changeset *data_reserved = NULL;
10338 struct extent_state *cached_state = NULL;
10339 struct btrfs_ordered_extent *ordered;
10343 u64 num_bytes, ram_bytes, disk_num_bytes;
10344 unsigned long nr_pages, i;
10345 struct page **pages;
10346 struct btrfs_key ins;
10347 bool extent_reserved = false;
10348 struct extent_map *em;
10351 switch (encoded->compression) {
10352 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB:
10353 compression = BTRFS_COMPRESS_ZLIB;
10355 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD:
10356 compression = BTRFS_COMPRESS_ZSTD;
10358 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K:
10359 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K:
10360 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K:
10361 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K:
10362 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K:
10363 /* The sector size must match for LZO. */
10364 if (encoded->compression -
10365 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K + 12 !=
10366 fs_info->sectorsize_bits)
10368 compression = BTRFS_COMPRESS_LZO;
10373 if (encoded->encryption != BTRFS_ENCODED_IO_ENCRYPTION_NONE)
10376 orig_count = iov_iter_count(from);
10378 /* The extent size must be sane. */
10379 if (encoded->unencoded_len > BTRFS_MAX_UNCOMPRESSED ||
10380 orig_count > BTRFS_MAX_COMPRESSED || orig_count == 0)
10384 * The compressed data must be smaller than the decompressed data.
10386 * It's of course possible for data to compress to larger or the same
10387 * size, but the buffered I/O path falls back to no compression for such
10388 * data, and we don't want to break any assumptions by creating these
10391 * Note that this is less strict than the current check we have that the
10392 * compressed data must be at least one sector smaller than the
10393 * decompressed data. We only want to enforce the weaker requirement
10394 * from old kernels that it is at least one byte smaller.
10396 if (orig_count >= encoded->unencoded_len)
10399 /* The extent must start on a sector boundary. */
10400 start = iocb->ki_pos;
10401 if (!IS_ALIGNED(start, fs_info->sectorsize))
10405 * The extent must end on a sector boundary. However, we allow a write
10406 * which ends at or extends i_size to have an unaligned length; we round
10407 * up the extent size and set i_size to the unaligned end.
10409 if (start + encoded->len < inode->vfs_inode.i_size &&
10410 !IS_ALIGNED(start + encoded->len, fs_info->sectorsize))
10413 /* Finally, the offset in the unencoded data must be sector-aligned. */
10414 if (!IS_ALIGNED(encoded->unencoded_offset, fs_info->sectorsize))
10417 num_bytes = ALIGN(encoded->len, fs_info->sectorsize);
10418 ram_bytes = ALIGN(encoded->unencoded_len, fs_info->sectorsize);
10419 end = start + num_bytes - 1;
10422 * If the extent cannot be inline, the compressed data on disk must be
10423 * sector-aligned. For convenience, we extend it with zeroes if it
10426 disk_num_bytes = ALIGN(orig_count, fs_info->sectorsize);
10427 nr_pages = DIV_ROUND_UP(disk_num_bytes, PAGE_SIZE);
10428 pages = kvcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL_ACCOUNT);
10431 for (i = 0; i < nr_pages; i++) {
10432 size_t bytes = min_t(size_t, PAGE_SIZE, iov_iter_count(from));
10435 pages[i] = alloc_page(GFP_KERNEL_ACCOUNT);
10440 kaddr = kmap_local_page(pages[i]);
10441 if (copy_from_iter(kaddr, bytes, from) != bytes) {
10442 kunmap_local(kaddr);
10446 if (bytes < PAGE_SIZE)
10447 memset(kaddr + bytes, 0, PAGE_SIZE - bytes);
10448 kunmap_local(kaddr);
10452 struct btrfs_ordered_extent *ordered;
10454 ret = btrfs_wait_ordered_range(&inode->vfs_inode, start, num_bytes);
10457 ret = invalidate_inode_pages2_range(inode->vfs_inode.i_mapping,
10458 start >> PAGE_SHIFT,
10459 end >> PAGE_SHIFT);
10462 lock_extent(io_tree, start, end, &cached_state);
10463 ordered = btrfs_lookup_ordered_range(inode, start, num_bytes);
10465 !filemap_range_has_page(inode->vfs_inode.i_mapping, start, end))
10468 btrfs_put_ordered_extent(ordered);
10469 unlock_extent(io_tree, start, end, &cached_state);
10474 * We don't use the higher-level delalloc space functions because our
10475 * num_bytes and disk_num_bytes are different.
10477 ret = btrfs_alloc_data_chunk_ondemand(inode, disk_num_bytes);
10480 ret = btrfs_qgroup_reserve_data(inode, &data_reserved, start, num_bytes);
10482 goto out_free_data_space;
10483 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes, disk_num_bytes,
10486 goto out_qgroup_free_data;
10488 /* Try an inline extent first. */
10489 if (start == 0 && encoded->unencoded_len == encoded->len &&
10490 encoded->unencoded_offset == 0) {
10491 ret = cow_file_range_inline(inode, encoded->len, orig_count,
10492 compression, pages, true);
10496 goto out_delalloc_release;
10500 ret = btrfs_reserve_extent(root, disk_num_bytes, disk_num_bytes,
10501 disk_num_bytes, 0, 0, &ins, 1, 1);
10503 goto out_delalloc_release;
10504 extent_reserved = true;
10506 em = create_io_em(inode, start, num_bytes,
10507 start - encoded->unencoded_offset, ins.objectid,
10508 ins.offset, ins.offset, ram_bytes, compression,
10509 BTRFS_ORDERED_COMPRESSED);
10512 goto out_free_reserved;
10514 free_extent_map(em);
10516 ordered = btrfs_alloc_ordered_extent(inode, start, num_bytes, ram_bytes,
10517 ins.objectid, ins.offset,
10518 encoded->unencoded_offset,
10519 (1 << BTRFS_ORDERED_ENCODED) |
10520 (1 << BTRFS_ORDERED_COMPRESSED),
10522 if (IS_ERR(ordered)) {
10523 btrfs_drop_extent_map_range(inode, start, end, false);
10524 ret = PTR_ERR(ordered);
10525 goto out_free_reserved;
10527 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10529 if (start + encoded->len > inode->vfs_inode.i_size)
10530 i_size_write(&inode->vfs_inode, start + encoded->len);
10532 unlock_extent(io_tree, start, end, &cached_state);
10534 btrfs_delalloc_release_extents(inode, num_bytes);
10536 btrfs_submit_compressed_write(ordered, pages, nr_pages, 0, false);
10541 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10542 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
10543 out_delalloc_release:
10544 btrfs_delalloc_release_extents(inode, num_bytes);
10545 btrfs_delalloc_release_metadata(inode, disk_num_bytes, ret < 0);
10546 out_qgroup_free_data:
10548 btrfs_qgroup_free_data(inode, data_reserved, start, num_bytes);
10549 out_free_data_space:
10551 * If btrfs_reserve_extent() succeeded, then we already decremented
10554 if (!extent_reserved)
10555 btrfs_free_reserved_data_space_noquota(fs_info, disk_num_bytes);
10557 unlock_extent(io_tree, start, end, &cached_state);
10559 for (i = 0; i < nr_pages; i++) {
10561 __free_page(pages[i]);
10566 iocb->ki_pos += encoded->len;
10572 * Add an entry indicating a block group or device which is pinned by a
10573 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10574 * negative errno on failure.
10576 static int btrfs_add_swapfile_pin(struct inode *inode, void *ptr,
10577 bool is_block_group)
10579 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10580 struct btrfs_swapfile_pin *sp, *entry;
10581 struct rb_node **p;
10582 struct rb_node *parent = NULL;
10584 sp = kmalloc(sizeof(*sp), GFP_NOFS);
10589 sp->is_block_group = is_block_group;
10590 sp->bg_extent_count = 1;
10592 spin_lock(&fs_info->swapfile_pins_lock);
10593 p = &fs_info->swapfile_pins.rb_node;
10596 entry = rb_entry(parent, struct btrfs_swapfile_pin, node);
10597 if (sp->ptr < entry->ptr ||
10598 (sp->ptr == entry->ptr && sp->inode < entry->inode)) {
10599 p = &(*p)->rb_left;
10600 } else if (sp->ptr > entry->ptr ||
10601 (sp->ptr == entry->ptr && sp->inode > entry->inode)) {
10602 p = &(*p)->rb_right;
10604 if (is_block_group)
10605 entry->bg_extent_count++;
10606 spin_unlock(&fs_info->swapfile_pins_lock);
10611 rb_link_node(&sp->node, parent, p);
10612 rb_insert_color(&sp->node, &fs_info->swapfile_pins);
10613 spin_unlock(&fs_info->swapfile_pins_lock);
10617 /* Free all of the entries pinned by this swapfile. */
10618 static void btrfs_free_swapfile_pins(struct inode *inode)
10620 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
10621 struct btrfs_swapfile_pin *sp;
10622 struct rb_node *node, *next;
10624 spin_lock(&fs_info->swapfile_pins_lock);
10625 node = rb_first(&fs_info->swapfile_pins);
10627 next = rb_next(node);
10628 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
10629 if (sp->inode == inode) {
10630 rb_erase(&sp->node, &fs_info->swapfile_pins);
10631 if (sp->is_block_group) {
10632 btrfs_dec_block_group_swap_extents(sp->ptr,
10633 sp->bg_extent_count);
10634 btrfs_put_block_group(sp->ptr);
10640 spin_unlock(&fs_info->swapfile_pins_lock);
10643 struct btrfs_swap_info {
10649 unsigned long nr_pages;
10653 static int btrfs_add_swap_extent(struct swap_info_struct *sis,
10654 struct btrfs_swap_info *bsi)
10656 unsigned long nr_pages;
10657 unsigned long max_pages;
10658 u64 first_ppage, first_ppage_reported, next_ppage;
10662 * Our swapfile may have had its size extended after the swap header was
10663 * written. In that case activating the swapfile should not go beyond
10664 * the max size set in the swap header.
10666 if (bsi->nr_pages >= sis->max)
10669 max_pages = sis->max - bsi->nr_pages;
10670 first_ppage = PAGE_ALIGN(bsi->block_start) >> PAGE_SHIFT;
10671 next_ppage = PAGE_ALIGN_DOWN(bsi->block_start + bsi->block_len) >> PAGE_SHIFT;
10673 if (first_ppage >= next_ppage)
10675 nr_pages = next_ppage - first_ppage;
10676 nr_pages = min(nr_pages, max_pages);
10678 first_ppage_reported = first_ppage;
10679 if (bsi->start == 0)
10680 first_ppage_reported++;
10681 if (bsi->lowest_ppage > first_ppage_reported)
10682 bsi->lowest_ppage = first_ppage_reported;
10683 if (bsi->highest_ppage < (next_ppage - 1))
10684 bsi->highest_ppage = next_ppage - 1;
10686 ret = add_swap_extent(sis, bsi->nr_pages, nr_pages, first_ppage);
10689 bsi->nr_extents += ret;
10690 bsi->nr_pages += nr_pages;
10694 static void btrfs_swap_deactivate(struct file *file)
10696 struct inode *inode = file_inode(file);
10698 btrfs_free_swapfile_pins(inode);
10699 atomic_dec(&BTRFS_I(inode)->root->nr_swapfiles);
10702 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10705 struct inode *inode = file_inode(file);
10706 struct btrfs_root *root = BTRFS_I(inode)->root;
10707 struct btrfs_fs_info *fs_info = root->fs_info;
10708 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
10709 struct extent_state *cached_state = NULL;
10710 struct extent_map *em = NULL;
10711 struct btrfs_device *device = NULL;
10712 struct btrfs_swap_info bsi = {
10713 .lowest_ppage = (sector_t)-1ULL,
10720 * If the swap file was just created, make sure delalloc is done. If the
10721 * file changes again after this, the user is doing something stupid and
10722 * we don't really care.
10724 ret = btrfs_wait_ordered_range(inode, 0, (u64)-1);
10729 * The inode is locked, so these flags won't change after we check them.
10731 if (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS) {
10732 btrfs_warn(fs_info, "swapfile must not be compressed");
10735 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)) {
10736 btrfs_warn(fs_info, "swapfile must not be copy-on-write");
10739 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
10740 btrfs_warn(fs_info, "swapfile must not be checksummed");
10745 * Balance or device remove/replace/resize can move stuff around from
10746 * under us. The exclop protection makes sure they aren't running/won't
10747 * run concurrently while we are mapping the swap extents, and
10748 * fs_info->swapfile_pins prevents them from running while the swap
10749 * file is active and moving the extents. Note that this also prevents
10750 * a concurrent device add which isn't actually necessary, but it's not
10751 * really worth the trouble to allow it.
10753 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_SWAP_ACTIVATE)) {
10754 btrfs_warn(fs_info,
10755 "cannot activate swapfile while exclusive operation is running");
10760 * Prevent snapshot creation while we are activating the swap file.
10761 * We do not want to race with snapshot creation. If snapshot creation
10762 * already started before we bumped nr_swapfiles from 0 to 1 and
10763 * completes before the first write into the swap file after it is
10764 * activated, than that write would fallback to COW.
10766 if (!btrfs_drew_try_write_lock(&root->snapshot_lock)) {
10767 btrfs_exclop_finish(fs_info);
10768 btrfs_warn(fs_info,
10769 "cannot activate swapfile because snapshot creation is in progress");
10773 * Snapshots can create extents which require COW even if NODATACOW is
10774 * set. We use this counter to prevent snapshots. We must increment it
10775 * before walking the extents because we don't want a concurrent
10776 * snapshot to run after we've already checked the extents.
10778 * It is possible that subvolume is marked for deletion but still not
10779 * removed yet. To prevent this race, we check the root status before
10780 * activating the swapfile.
10782 spin_lock(&root->root_item_lock);
10783 if (btrfs_root_dead(root)) {
10784 spin_unlock(&root->root_item_lock);
10786 btrfs_exclop_finish(fs_info);
10787 btrfs_warn(fs_info,
10788 "cannot activate swapfile because subvolume %llu is being deleted",
10789 root->root_key.objectid);
10792 atomic_inc(&root->nr_swapfiles);
10793 spin_unlock(&root->root_item_lock);
10795 isize = ALIGN_DOWN(inode->i_size, fs_info->sectorsize);
10797 lock_extent(io_tree, 0, isize - 1, &cached_state);
10799 while (start < isize) {
10800 u64 logical_block_start, physical_block_start;
10801 struct btrfs_block_group *bg;
10802 u64 len = isize - start;
10804 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len);
10810 if (em->block_start == EXTENT_MAP_HOLE) {
10811 btrfs_warn(fs_info, "swapfile must not have holes");
10815 if (em->block_start == EXTENT_MAP_INLINE) {
10817 * It's unlikely we'll ever actually find ourselves
10818 * here, as a file small enough to fit inline won't be
10819 * big enough to store more than the swap header, but in
10820 * case something changes in the future, let's catch it
10821 * here rather than later.
10823 btrfs_warn(fs_info, "swapfile must not be inline");
10827 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
10828 btrfs_warn(fs_info, "swapfile must not be compressed");
10833 logical_block_start = em->block_start + (start - em->start);
10834 len = min(len, em->len - (start - em->start));
10835 free_extent_map(em);
10838 ret = can_nocow_extent(inode, start, &len, NULL, NULL, NULL, false, true);
10844 btrfs_warn(fs_info,
10845 "swapfile must not be copy-on-write");
10850 em = btrfs_get_chunk_map(fs_info, logical_block_start, len);
10856 if (em->map_lookup->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
10857 btrfs_warn(fs_info,
10858 "swapfile must have single data profile");
10863 if (device == NULL) {
10864 device = em->map_lookup->stripes[0].dev;
10865 ret = btrfs_add_swapfile_pin(inode, device, false);
10870 } else if (device != em->map_lookup->stripes[0].dev) {
10871 btrfs_warn(fs_info, "swapfile must be on one device");
10876 physical_block_start = (em->map_lookup->stripes[0].physical +
10877 (logical_block_start - em->start));
10878 len = min(len, em->len - (logical_block_start - em->start));
10879 free_extent_map(em);
10882 bg = btrfs_lookup_block_group(fs_info, logical_block_start);
10884 btrfs_warn(fs_info,
10885 "could not find block group containing swapfile");
10890 if (!btrfs_inc_block_group_swap_extents(bg)) {
10891 btrfs_warn(fs_info,
10892 "block group for swapfile at %llu is read-only%s",
10894 atomic_read(&fs_info->scrubs_running) ?
10895 " (scrub running)" : "");
10896 btrfs_put_block_group(bg);
10901 ret = btrfs_add_swapfile_pin(inode, bg, true);
10903 btrfs_put_block_group(bg);
10910 if (bsi.block_len &&
10911 bsi.block_start + bsi.block_len == physical_block_start) {
10912 bsi.block_len += len;
10914 if (bsi.block_len) {
10915 ret = btrfs_add_swap_extent(sis, &bsi);
10920 bsi.block_start = physical_block_start;
10921 bsi.block_len = len;
10928 ret = btrfs_add_swap_extent(sis, &bsi);
10931 if (!IS_ERR_OR_NULL(em))
10932 free_extent_map(em);
10934 unlock_extent(io_tree, 0, isize - 1, &cached_state);
10937 btrfs_swap_deactivate(file);
10939 btrfs_drew_write_unlock(&root->snapshot_lock);
10941 btrfs_exclop_finish(fs_info);
10947 sis->bdev = device->bdev;
10948 *span = bsi.highest_ppage - bsi.lowest_ppage + 1;
10949 sis->max = bsi.nr_pages;
10950 sis->pages = bsi.nr_pages - 1;
10951 sis->highest_bit = bsi.nr_pages - 1;
10952 return bsi.nr_extents;
10955 static void btrfs_swap_deactivate(struct file *file)
10959 static int btrfs_swap_activate(struct swap_info_struct *sis, struct file *file,
10962 return -EOPNOTSUPP;
10967 * Update the number of bytes used in the VFS' inode. When we replace extents in
10968 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10969 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10970 * always get a correct value.
10972 void btrfs_update_inode_bytes(struct btrfs_inode *inode,
10973 const u64 add_bytes,
10974 const u64 del_bytes)
10976 if (add_bytes == del_bytes)
10979 spin_lock(&inode->lock);
10981 inode_sub_bytes(&inode->vfs_inode, del_bytes);
10983 inode_add_bytes(&inode->vfs_inode, add_bytes);
10984 spin_unlock(&inode->lock);
10988 * Verify that there are no ordered extents for a given file range.
10990 * @inode: The target inode.
10991 * @start: Start offset of the file range, should be sector size aligned.
10992 * @end: End offset (inclusive) of the file range, its value +1 should be
10993 * sector size aligned.
10995 * This should typically be used for cases where we locked an inode's VFS lock in
10996 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10997 * we have flushed all delalloc in the range, we have waited for all ordered
10998 * extents in the range to complete and finally we have locked the file range in
10999 * the inode's io_tree.
11001 void btrfs_assert_inode_range_clean(struct btrfs_inode *inode, u64 start, u64 end)
11003 struct btrfs_root *root = inode->root;
11004 struct btrfs_ordered_extent *ordered;
11006 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT))
11009 ordered = btrfs_lookup_first_ordered_range(inode, start, end + 1 - start);
11011 btrfs_err(root->fs_info,
11012 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
11013 start, end, btrfs_ino(inode), root->root_key.objectid,
11014 ordered->file_offset,
11015 ordered->file_offset + ordered->num_bytes - 1);
11016 btrfs_put_ordered_extent(ordered);
11019 ASSERT(ordered == NULL);
11022 static const struct inode_operations btrfs_dir_inode_operations = {
11023 .getattr = btrfs_getattr,
11024 .lookup = btrfs_lookup,
11025 .create = btrfs_create,
11026 .unlink = btrfs_unlink,
11027 .link = btrfs_link,
11028 .mkdir = btrfs_mkdir,
11029 .rmdir = btrfs_rmdir,
11030 .rename = btrfs_rename2,
11031 .symlink = btrfs_symlink,
11032 .setattr = btrfs_setattr,
11033 .mknod = btrfs_mknod,
11034 .listxattr = btrfs_listxattr,
11035 .permission = btrfs_permission,
11036 .get_inode_acl = btrfs_get_acl,
11037 .set_acl = btrfs_set_acl,
11038 .update_time = btrfs_update_time,
11039 .tmpfile = btrfs_tmpfile,
11040 .fileattr_get = btrfs_fileattr_get,
11041 .fileattr_set = btrfs_fileattr_set,
11044 static const struct file_operations btrfs_dir_file_operations = {
11045 .llseek = generic_file_llseek,
11046 .read = generic_read_dir,
11047 .iterate_shared = btrfs_real_readdir,
11048 .open = btrfs_opendir,
11049 .unlocked_ioctl = btrfs_ioctl,
11050 #ifdef CONFIG_COMPAT
11051 .compat_ioctl = btrfs_compat_ioctl,
11053 .release = btrfs_release_file,
11054 .fsync = btrfs_sync_file,
11058 * btrfs doesn't support the bmap operation because swapfiles
11059 * use bmap to make a mapping of extents in the file. They assume
11060 * these extents won't change over the life of the file and they
11061 * use the bmap result to do IO directly to the drive.
11063 * the btrfs bmap call would return logical addresses that aren't
11064 * suitable for IO and they also will change frequently as COW
11065 * operations happen. So, swapfile + btrfs == corruption.
11067 * For now we're avoiding this by dropping bmap.
11069 static const struct address_space_operations btrfs_aops = {
11070 .read_folio = btrfs_read_folio,
11071 .writepages = btrfs_writepages,
11072 .readahead = btrfs_readahead,
11073 .invalidate_folio = btrfs_invalidate_folio,
11074 .release_folio = btrfs_release_folio,
11075 .migrate_folio = btrfs_migrate_folio,
11076 .dirty_folio = filemap_dirty_folio,
11077 .error_remove_page = generic_error_remove_page,
11078 .swap_activate = btrfs_swap_activate,
11079 .swap_deactivate = btrfs_swap_deactivate,
11082 static const struct inode_operations btrfs_file_inode_operations = {
11083 .getattr = btrfs_getattr,
11084 .setattr = btrfs_setattr,
11085 .listxattr = btrfs_listxattr,
11086 .permission = btrfs_permission,
11087 .fiemap = btrfs_fiemap,
11088 .get_inode_acl = btrfs_get_acl,
11089 .set_acl = btrfs_set_acl,
11090 .update_time = btrfs_update_time,
11091 .fileattr_get = btrfs_fileattr_get,
11092 .fileattr_set = btrfs_fileattr_set,
11094 static const struct inode_operations btrfs_special_inode_operations = {
11095 .getattr = btrfs_getattr,
11096 .setattr = btrfs_setattr,
11097 .permission = btrfs_permission,
11098 .listxattr = btrfs_listxattr,
11099 .get_inode_acl = btrfs_get_acl,
11100 .set_acl = btrfs_set_acl,
11101 .update_time = btrfs_update_time,
11103 static const struct inode_operations btrfs_symlink_inode_operations = {
11104 .get_link = page_get_link,
11105 .getattr = btrfs_getattr,
11106 .setattr = btrfs_setattr,
11107 .permission = btrfs_permission,
11108 .listxattr = btrfs_listxattr,
11109 .update_time = btrfs_update_time,
11112 const struct dentry_operations btrfs_dentry_operations = {
11113 .d_delete = btrfs_dentry_delete,