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
);
127 static noinline
int cow_file_range(struct btrfs_inode
*inode
,
128 struct page
*locked_page
,
129 u64 start
, u64 end
, int *page_started
,
130 unsigned long *nr_written
, int unlock
,
132 static struct extent_map
*create_io_em(struct btrfs_inode
*inode
, u64 start
,
133 u64 len
, u64 orig_start
, u64 block_start
,
134 u64 block_len
, u64 orig_block_len
,
135 u64 ram_bytes
, int compress_type
,
138 static int data_reloc_print_warning_inode(u64 inum
, u64 offset
, u64 num_bytes
,
139 u64 root
, void *warn_ctx
)
141 struct data_reloc_warn
*warn
= warn_ctx
;
142 struct btrfs_fs_info
*fs_info
= warn
->fs_info
;
143 struct extent_buffer
*eb
;
144 struct btrfs_inode_item
*inode_item
;
145 struct inode_fs_paths
*ipath
= NULL
;
146 struct btrfs_root
*local_root
;
147 struct btrfs_key key
;
148 unsigned int nofs_flag
;
152 local_root
= btrfs_get_fs_root(fs_info
, root
, true);
153 if (IS_ERR(local_root
)) {
154 ret
= PTR_ERR(local_root
);
158 /* This makes the path point to (inum INODE_ITEM ioff). */
160 key
.type
= BTRFS_INODE_ITEM_KEY
;
163 ret
= btrfs_search_slot(NULL
, local_root
, &key
, &warn
->path
, 0, 0);
165 btrfs_put_root(local_root
);
166 btrfs_release_path(&warn
->path
);
170 eb
= warn
->path
.nodes
[0];
171 inode_item
= btrfs_item_ptr(eb
, warn
->path
.slots
[0], struct btrfs_inode_item
);
172 nlink
= btrfs_inode_nlink(eb
, inode_item
);
173 btrfs_release_path(&warn
->path
);
175 nofs_flag
= memalloc_nofs_save();
176 ipath
= init_ipath(4096, local_root
, &warn
->path
);
177 memalloc_nofs_restore(nofs_flag
);
179 btrfs_put_root(local_root
);
180 ret
= PTR_ERR(ipath
);
183 * -ENOMEM, not a critical error, just output an generic error
187 "checksum error at logical %llu mirror %u root %llu, inode %llu offset %llu",
188 warn
->logical
, warn
->mirror_num
, root
, inum
, offset
);
191 ret
= paths_from_inode(inum
, ipath
);
196 * We deliberately ignore the bit ipath might have been too small to
197 * hold all of the paths here
199 for (int i
= 0; i
< ipath
->fspath
->elem_cnt
; i
++) {
201 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu length %u links %u (path: %s)",
202 warn
->logical
, warn
->mirror_num
, root
, inum
, offset
,
203 fs_info
->sectorsize
, nlink
,
204 (char *)(unsigned long)ipath
->fspath
->val
[i
]);
207 btrfs_put_root(local_root
);
213 "checksum error at logical %llu mirror %u root %llu inode %llu offset %llu, path resolving failed with ret=%d",
214 warn
->logical
, warn
->mirror_num
, root
, inum
, offset
, ret
);
221 * Do extra user-friendly error output (e.g. lookup all the affected files).
223 * Return true if we succeeded doing the backref lookup.
224 * Return false if such lookup failed, and has to fallback to the old error message.
226 static void print_data_reloc_error(const struct btrfs_inode
*inode
, u64 file_off
,
227 const u8
*csum
, const u8
*csum_expected
,
230 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
231 struct btrfs_path path
= { 0 };
232 struct btrfs_key found_key
= { 0 };
233 struct extent_buffer
*eb
;
234 struct btrfs_extent_item
*ei
;
235 const u32 csum_size
= fs_info
->csum_size
;
241 mutex_lock(&fs_info
->reloc_mutex
);
242 logical
= btrfs_get_reloc_bg_bytenr(fs_info
);
243 mutex_unlock(&fs_info
->reloc_mutex
);
245 if (logical
== U64_MAX
) {
246 btrfs_warn_rl(fs_info
, "has data reloc tree but no running relocation");
247 btrfs_warn_rl(fs_info
,
248 "csum failed root %lld ino %llu off %llu csum " CSUM_FMT
" expected csum " CSUM_FMT
" mirror %d",
249 inode
->root
->root_key
.objectid
, btrfs_ino(inode
), file_off
,
250 CSUM_FMT_VALUE(csum_size
, csum
),
251 CSUM_FMT_VALUE(csum_size
, csum_expected
),
257 btrfs_warn_rl(fs_info
,
258 "csum failed root %lld ino %llu off %llu logical %llu csum " CSUM_FMT
" expected csum " CSUM_FMT
" mirror %d",
259 inode
->root
->root_key
.objectid
,
260 btrfs_ino(inode
), file_off
, logical
,
261 CSUM_FMT_VALUE(csum_size
, csum
),
262 CSUM_FMT_VALUE(csum_size
, csum_expected
),
265 ret
= extent_from_logical(fs_info
, logical
, &path
, &found_key
, &flags
);
267 btrfs_err_rl(fs_info
, "failed to lookup extent item for logical %llu: %d",
272 ei
= btrfs_item_ptr(eb
, path
.slots
[0], struct btrfs_extent_item
);
273 item_size
= btrfs_item_size(eb
, path
.slots
[0]);
274 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
275 unsigned long ptr
= 0;
280 ret
= tree_backref_for_extent(&ptr
, eb
, &found_key
, ei
,
281 item_size
, &ref_root
,
284 btrfs_warn_rl(fs_info
,
285 "failed to resolve tree backref for logical %llu: %d",
292 btrfs_warn_rl(fs_info
,
293 "csum error at logical %llu mirror %u: metadata %s (level %d) in tree %llu",
295 (ref_level
? "node" : "leaf"),
296 ref_level
, ref_root
);
298 btrfs_release_path(&path
);
300 struct btrfs_backref_walk_ctx ctx
= { 0 };
301 struct data_reloc_warn reloc_warn
= { 0 };
303 btrfs_release_path(&path
);
305 ctx
.bytenr
= found_key
.objectid
;
306 ctx
.extent_item_pos
= logical
- found_key
.objectid
;
307 ctx
.fs_info
= fs_info
;
309 reloc_warn
.logical
= logical
;
310 reloc_warn
.extent_item_size
= found_key
.offset
;
311 reloc_warn
.mirror_num
= mirror_num
;
312 reloc_warn
.fs_info
= fs_info
;
314 iterate_extent_inodes(&ctx
, true,
315 data_reloc_print_warning_inode
, &reloc_warn
);
319 static void __cold
btrfs_print_data_csum_error(struct btrfs_inode
*inode
,
320 u64 logical_start
, u8
*csum
, u8
*csum_expected
, int mirror_num
)
322 struct btrfs_root
*root
= inode
->root
;
323 const u32 csum_size
= root
->fs_info
->csum_size
;
325 /* For data reloc tree, it's better to do a backref lookup instead. */
326 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
327 return print_data_reloc_error(inode
, logical_start
, csum
,
328 csum_expected
, mirror_num
);
330 /* Output without objectid, which is more meaningful */
331 if (root
->root_key
.objectid
>= BTRFS_LAST_FREE_OBJECTID
) {
332 btrfs_warn_rl(root
->fs_info
,
333 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT
" expected csum " CSUM_FMT
" mirror %d",
334 root
->root_key
.objectid
, btrfs_ino(inode
),
336 CSUM_FMT_VALUE(csum_size
, csum
),
337 CSUM_FMT_VALUE(csum_size
, csum_expected
),
340 btrfs_warn_rl(root
->fs_info
,
341 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT
" expected csum " CSUM_FMT
" mirror %d",
342 root
->root_key
.objectid
, btrfs_ino(inode
),
344 CSUM_FMT_VALUE(csum_size
, csum
),
345 CSUM_FMT_VALUE(csum_size
, csum_expected
),
351 * btrfs_inode_lock - lock inode i_rwsem based on arguments passed
353 * ilock_flags can have the following bit set:
355 * BTRFS_ILOCK_SHARED - acquire a shared lock on the inode
356 * BTRFS_ILOCK_TRY - try to acquire the lock, if fails on first attempt
358 * BTRFS_ILOCK_MMAP - acquire a write lock on the i_mmap_lock
360 int btrfs_inode_lock(struct btrfs_inode
*inode
, unsigned int ilock_flags
)
362 if (ilock_flags
& BTRFS_ILOCK_SHARED
) {
363 if (ilock_flags
& BTRFS_ILOCK_TRY
) {
364 if (!inode_trylock_shared(&inode
->vfs_inode
))
369 inode_lock_shared(&inode
->vfs_inode
);
371 if (ilock_flags
& BTRFS_ILOCK_TRY
) {
372 if (!inode_trylock(&inode
->vfs_inode
))
377 inode_lock(&inode
->vfs_inode
);
379 if (ilock_flags
& BTRFS_ILOCK_MMAP
)
380 down_write(&inode
->i_mmap_lock
);
385 * btrfs_inode_unlock - unock inode i_rwsem
387 * ilock_flags should contain the same bits set as passed to btrfs_inode_lock()
388 * to decide whether the lock acquired is shared or exclusive.
390 void btrfs_inode_unlock(struct btrfs_inode
*inode
, unsigned int ilock_flags
)
392 if (ilock_flags
& BTRFS_ILOCK_MMAP
)
393 up_write(&inode
->i_mmap_lock
);
394 if (ilock_flags
& BTRFS_ILOCK_SHARED
)
395 inode_unlock_shared(&inode
->vfs_inode
);
397 inode_unlock(&inode
->vfs_inode
);
401 * Cleanup all submitted ordered extents in specified range to handle errors
402 * from the btrfs_run_delalloc_range() callback.
404 * NOTE: caller must ensure that when an error happens, it can not call
405 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
406 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
407 * to be released, which we want to happen only when finishing the ordered
408 * extent (btrfs_finish_ordered_io()).
410 static inline void btrfs_cleanup_ordered_extents(struct btrfs_inode
*inode
,
411 struct page
*locked_page
,
412 u64 offset
, u64 bytes
)
414 unsigned long index
= offset
>> PAGE_SHIFT
;
415 unsigned long end_index
= (offset
+ bytes
- 1) >> PAGE_SHIFT
;
416 u64 page_start
= 0, page_end
= 0;
420 page_start
= page_offset(locked_page
);
421 page_end
= page_start
+ PAGE_SIZE
- 1;
424 while (index
<= end_index
) {
426 * For locked page, we will call end_extent_writepage() on it
427 * in run_delalloc_range() for the error handling. That
428 * end_extent_writepage() function will call
429 * btrfs_mark_ordered_io_finished() to clear page Ordered and
430 * 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 u64 blocksize
= fs_info
->sectorsize
;
839 u64 start
= async_chunk
->start
;
840 u64 end
= async_chunk
->end
;
844 struct page
**pages
= NULL
;
845 unsigned long nr_pages
;
846 unsigned long total_compressed
= 0;
847 unsigned long total_in
= 0;
850 int compress_type
= fs_info
->compress_type
;
851 int compressed_extents
= 0;
854 inode_should_defrag(inode
, start
, end
, end
- start
+ 1, SZ_16K
);
857 * We need to save i_size before now because it could change in between
858 * us evaluating the size and assigning it. This is because we lock and
859 * unlock the page in truncate and fallocate, and then modify the i_size
862 * The barriers are to emulate READ_ONCE, remove that once i_size_read
866 i_size
= i_size_read(&inode
->vfs_inode
);
868 actual_end
= min_t(u64
, i_size
, end
+ 1);
871 nr_pages
= (end
>> PAGE_SHIFT
) - (start
>> PAGE_SHIFT
) + 1;
872 nr_pages
= min_t(unsigned long, nr_pages
, BTRFS_MAX_COMPRESSED_PAGES
);
875 * we don't want to send crud past the end of i_size through
876 * compression, that's just a waste of CPU time. So, if the
877 * end of the file is before the start of our current
878 * requested range of bytes, we bail out to the uncompressed
879 * cleanup code that can deal with all of this.
881 * It isn't really the fastest way to fix things, but this is a
882 * very uncommon corner.
884 if (actual_end
<= start
)
885 goto cleanup_and_bail_uncompressed
;
887 total_compressed
= actual_end
- start
;
890 * Skip compression for a small file range(<=blocksize) that
891 * isn't an inline extent, since it doesn't save disk space at all.
893 if (total_compressed
<= blocksize
&&
894 (start
> 0 || end
+ 1 < inode
->disk_i_size
))
895 goto cleanup_and_bail_uncompressed
;
898 * For subpage case, we require full page alignment for the sector
900 * Thus we must also check against @actual_end, not just @end.
902 if (blocksize
< PAGE_SIZE
) {
903 if (!PAGE_ALIGNED(start
) ||
904 !PAGE_ALIGNED(round_up(actual_end
, blocksize
)))
905 goto cleanup_and_bail_uncompressed
;
908 total_compressed
= min_t(unsigned long, total_compressed
,
909 BTRFS_MAX_UNCOMPRESSED
);
914 * we do compression for mount -o compress and when the
915 * inode has not been flagged as nocompress. This flag can
916 * change at any time if we discover bad compression ratios.
918 if (inode_need_compress(inode
, start
, end
)) {
920 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
922 /* just bail out to the uncompressed code */
927 if (inode
->defrag_compress
)
928 compress_type
= inode
->defrag_compress
;
929 else if (inode
->prop_compress
)
930 compress_type
= inode
->prop_compress
;
933 * we need to call clear_page_dirty_for_io on each
934 * page in the range. Otherwise applications with the file
935 * mmap'd can wander in and change the page contents while
936 * we are compressing them.
938 * If the compression fails for any reason, we set the pages
939 * dirty again later on.
941 * Note that the remaining part is redirtied, the start pointer
942 * has moved, the end is the original one.
945 extent_range_clear_dirty_for_io(&inode
->vfs_inode
, start
, end
);
949 /* Compression level is applied here and only here */
950 ret
= btrfs_compress_pages(
951 compress_type
| (fs_info
->compress_level
<< 4),
952 inode
->vfs_inode
.i_mapping
, start
,
959 unsigned long offset
= offset_in_page(total_compressed
);
960 struct page
*page
= pages
[nr_pages
- 1];
962 /* zero the tail end of the last page, we might be
963 * sending it down to disk
966 memzero_page(page
, offset
, PAGE_SIZE
- offset
);
972 * Check cow_file_range() for why we don't even try to create inline
973 * extent for subpage case.
975 if (start
== 0 && fs_info
->sectorsize
== PAGE_SIZE
) {
976 /* lets try to make an inline extent */
977 if (ret
|| total_in
< actual_end
) {
978 /* we didn't compress the entire range, try
979 * to make an uncompressed inline extent.
981 ret
= cow_file_range_inline(inode
, actual_end
,
982 0, BTRFS_COMPRESS_NONE
,
985 /* try making a compressed inline extent */
986 ret
= cow_file_range_inline(inode
, actual_end
,
988 compress_type
, pages
,
992 unsigned long clear_flags
= EXTENT_DELALLOC
|
993 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
994 EXTENT_DO_ACCOUNTING
;
995 unsigned long page_error_op
;
997 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
1000 * inline extent creation worked or returned error,
1001 * we don't need to create any more async work items.
1002 * Unlock and free up our temp pages.
1004 * We use DO_ACCOUNTING here because we need the
1005 * delalloc_release_metadata to be done _after_ we drop
1006 * our outstanding extent for clearing delalloc for this
1009 extent_clear_unlock_delalloc(inode
, start
, end
,
1013 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;
1138 * Call cow_file_range() to run the delalloc range directly, since we
1139 * won't go to NOCOW or async path again.
1141 * Also we call cow_file_range() with @unlock_page == 0, so that we
1142 * can directly submit them without interruption.
1144 ret
= cow_file_range(inode
, locked_page
, start
, end
, &page_started
,
1145 &nr_written
, 0, NULL
);
1146 /* Inline extent inserted, page gets unlocked and everything is done */
1151 btrfs_cleanup_ordered_extents(inode
, locked_page
, start
, end
- start
+ 1);
1153 const u64 page_start
= page_offset(locked_page
);
1154 const u64 page_end
= page_start
+ PAGE_SIZE
- 1;
1156 btrfs_page_set_error(inode
->root
->fs_info
, locked_page
,
1157 page_start
, PAGE_SIZE
);
1158 set_page_writeback(locked_page
);
1159 end_page_writeback(locked_page
);
1160 end_extent_writepage(locked_page
, ret
, page_start
, page_end
);
1161 unlock_page(locked_page
);
1166 /* All pages will be unlocked, including @locked_page */
1167 return extent_write_locked_range(&inode
->vfs_inode
, start
, end
);
1170 static int submit_one_async_extent(struct btrfs_inode
*inode
,
1171 struct async_chunk
*async_chunk
,
1172 struct async_extent
*async_extent
,
1175 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
1176 struct btrfs_root
*root
= inode
->root
;
1177 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1178 struct btrfs_key ins
;
1179 struct page
*locked_page
= NULL
;
1180 struct extent_map
*em
;
1182 u64 start
= async_extent
->start
;
1183 u64 end
= async_extent
->start
+ async_extent
->ram_size
- 1;
1185 if (async_chunk
->blkcg_css
)
1186 kthread_associate_blkcg(async_chunk
->blkcg_css
);
1189 * If async_chunk->locked_page is in the async_extent range, we need to
1192 if (async_chunk
->locked_page
) {
1193 u64 locked_page_start
= page_offset(async_chunk
->locked_page
);
1194 u64 locked_page_end
= locked_page_start
+ PAGE_SIZE
- 1;
1196 if (!(start
>= locked_page_end
|| end
<= locked_page_start
))
1197 locked_page
= async_chunk
->locked_page
;
1199 lock_extent(io_tree
, start
, end
, NULL
);
1201 /* We have fall back to uncompressed write */
1202 if (!async_extent
->pages
) {
1203 ret
= submit_uncompressed_range(inode
, async_extent
, locked_page
);
1207 ret
= btrfs_reserve_extent(root
, async_extent
->ram_size
,
1208 async_extent
->compressed_size
,
1209 async_extent
->compressed_size
,
1210 0, *alloc_hint
, &ins
, 1, 1);
1212 free_async_extent_pages(async_extent
);
1214 * Here we used to try again by going back to non-compressed
1215 * path for ENOSPC. But we can't reserve space even for
1216 * compressed size, how could it work for uncompressed size
1217 * which requires larger size? So here we directly go error
1223 /* Here we're doing allocation and writeback of the compressed pages */
1224 em
= create_io_em(inode
, start
,
1225 async_extent
->ram_size
, /* len */
1226 start
, /* orig_start */
1227 ins
.objectid
, /* block_start */
1228 ins
.offset
, /* block_len */
1229 ins
.offset
, /* orig_block_len */
1230 async_extent
->ram_size
, /* ram_bytes */
1231 async_extent
->compress_type
,
1232 BTRFS_ORDERED_COMPRESSED
);
1235 goto out_free_reserve
;
1237 free_extent_map(em
);
1239 ret
= btrfs_add_ordered_extent(inode
, start
, /* file_offset */
1240 async_extent
->ram_size
, /* num_bytes */
1241 async_extent
->ram_size
, /* ram_bytes */
1242 ins
.objectid
, /* disk_bytenr */
1243 ins
.offset
, /* disk_num_bytes */
1245 1 << BTRFS_ORDERED_COMPRESSED
,
1246 async_extent
->compress_type
);
1248 btrfs_drop_extent_map_range(inode
, start
, end
, false);
1249 goto out_free_reserve
;
1251 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1253 /* Clear dirty, set writeback and unlock the pages. */
1254 extent_clear_unlock_delalloc(inode
, start
, end
,
1255 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
1256 PAGE_UNLOCK
| PAGE_START_WRITEBACK
);
1258 btrfs_submit_compressed_write(inode
, start
, /* file_offset */
1259 async_extent
->ram_size
, /* num_bytes */
1260 ins
.objectid
, /* disk_bytenr */
1261 ins
.offset
, /* compressed_len */
1262 async_extent
->pages
, /* compressed_pages */
1263 async_extent
->nr_pages
,
1264 async_chunk
->write_flags
, true);
1265 *alloc_hint
= ins
.objectid
+ ins
.offset
;
1267 if (async_chunk
->blkcg_css
)
1268 kthread_associate_blkcg(NULL
);
1269 kfree(async_extent
);
1273 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1274 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
1276 extent_clear_unlock_delalloc(inode
, start
, end
,
1277 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
1278 EXTENT_DELALLOC_NEW
|
1279 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
1280 PAGE_UNLOCK
| PAGE_START_WRITEBACK
|
1281 PAGE_END_WRITEBACK
| PAGE_SET_ERROR
);
1282 free_async_extent_pages(async_extent
);
1287 * Phase two of compressed writeback. This is the ordered portion of the code,
1288 * which only gets called in the order the work was queued. We walk all the
1289 * async extents created by compress_file_range and send them down to the disk.
1291 static noinline
void submit_compressed_extents(struct async_chunk
*async_chunk
)
1293 struct btrfs_inode
*inode
= async_chunk
->inode
;
1294 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
1295 struct async_extent
*async_extent
;
1299 while (!list_empty(&async_chunk
->extents
)) {
1303 async_extent
= list_entry(async_chunk
->extents
.next
,
1304 struct async_extent
, list
);
1305 list_del(&async_extent
->list
);
1306 extent_start
= async_extent
->start
;
1307 ram_size
= async_extent
->ram_size
;
1309 ret
= submit_one_async_extent(inode
, async_chunk
, async_extent
,
1311 btrfs_debug(fs_info
,
1312 "async extent submission failed root=%lld inode=%llu start=%llu len=%llu ret=%d",
1313 inode
->root
->root_key
.objectid
,
1314 btrfs_ino(inode
), extent_start
, ram_size
, ret
);
1318 static u64
get_extent_allocation_hint(struct btrfs_inode
*inode
, u64 start
,
1321 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
1322 struct extent_map
*em
;
1325 read_lock(&em_tree
->lock
);
1326 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
1329 * if block start isn't an actual block number then find the
1330 * first block in this inode and use that as a hint. If that
1331 * block is also bogus then just don't worry about it.
1333 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
1334 free_extent_map(em
);
1335 em
= search_extent_mapping(em_tree
, 0, 0);
1336 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
1337 alloc_hint
= em
->block_start
;
1339 free_extent_map(em
);
1341 alloc_hint
= em
->block_start
;
1342 free_extent_map(em
);
1345 read_unlock(&em_tree
->lock
);
1351 * when extent_io.c finds a delayed allocation range in the file,
1352 * the call backs end up in this code. The basic idea is to
1353 * allocate extents on disk for the range, and create ordered data structs
1354 * in ram to track those extents.
1356 * locked_page is the page that writepage had locked already. We use
1357 * it to make sure we don't do extra locks or unlocks.
1359 * *page_started is set to one if we unlock locked_page and do everything
1360 * required to start IO on it. It may be clean and already done with
1361 * IO when we return.
1363 * When unlock == 1, we unlock the pages in successfully allocated regions.
1364 * When unlock == 0, we leave them locked for writing them out.
1366 * However, we unlock all the pages except @locked_page in case of failure.
1368 * In summary, page locking state will be as follow:
1370 * - page_started == 1 (return value)
1371 * - All the pages are unlocked. IO is started.
1372 * - Note that this can happen only on success
1374 * - All the pages except @locked_page are unlocked in any case
1376 * - On success, all the pages are locked for writing out them
1377 * - On failure, all the pages except @locked_page are unlocked
1379 * When a failure happens in the second or later iteration of the
1380 * while-loop, the ordered extents created in previous iterations are kept
1381 * intact. So, the caller must clean them up by calling
1382 * btrfs_cleanup_ordered_extents(). See btrfs_run_delalloc_range() for
1385 static noinline
int cow_file_range(struct btrfs_inode
*inode
,
1386 struct page
*locked_page
,
1387 u64 start
, u64 end
, int *page_started
,
1388 unsigned long *nr_written
, int unlock
,
1391 struct btrfs_root
*root
= inode
->root
;
1392 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1394 u64 orig_start
= start
;
1396 unsigned long ram_size
;
1397 u64 cur_alloc_size
= 0;
1399 u64 blocksize
= fs_info
->sectorsize
;
1400 struct btrfs_key ins
;
1401 struct extent_map
*em
;
1402 unsigned clear_bits
;
1403 unsigned long page_ops
;
1404 bool extent_reserved
= false;
1407 if (btrfs_is_free_space_inode(inode
)) {
1412 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
1413 num_bytes
= max(blocksize
, num_bytes
);
1414 ASSERT(num_bytes
<= btrfs_super_total_bytes(fs_info
->super_copy
));
1416 inode_should_defrag(inode
, start
, end
, num_bytes
, SZ_64K
);
1419 * Due to the page size limit, for subpage we can only trigger the
1420 * writeback for the dirty sectors of page, that means data writeback
1421 * is doing more writeback than what we want.
1423 * This is especially unexpected for some call sites like fallocate,
1424 * where we only increase i_size after everything is done.
1425 * This means we can trigger inline extent even if we didn't want to.
1426 * So here we skip inline extent creation completely.
1428 if (start
== 0 && fs_info
->sectorsize
== PAGE_SIZE
) {
1429 u64 actual_end
= min_t(u64
, i_size_read(&inode
->vfs_inode
),
1432 /* lets try to make an inline extent */
1433 ret
= cow_file_range_inline(inode
, actual_end
, 0,
1434 BTRFS_COMPRESS_NONE
, NULL
, false);
1437 * We use DO_ACCOUNTING here because we need the
1438 * delalloc_release_metadata to be run _after_ we drop
1439 * our outstanding extent for clearing delalloc for this
1442 extent_clear_unlock_delalloc(inode
, start
, end
,
1444 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1445 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
1446 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1447 PAGE_START_WRITEBACK
| PAGE_END_WRITEBACK
);
1448 *nr_written
= *nr_written
+
1449 (end
- start
+ PAGE_SIZE
) / PAGE_SIZE
;
1452 * locked_page is locked by the caller of
1453 * writepage_delalloc(), not locked by
1454 * __process_pages_contig().
1456 * We can't let __process_pages_contig() to unlock it,
1457 * as it doesn't have any subpage::writers recorded.
1459 * Here we manually unlock the page, since the caller
1460 * can't use page_started to determine if it's an
1461 * inline extent or a compressed extent.
1463 unlock_page(locked_page
);
1465 } else if (ret
< 0) {
1470 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
1473 * Relocation relies on the relocated extents to have exactly the same
1474 * size as the original extents. Normally writeback for relocation data
1475 * extents follows a NOCOW path because relocation preallocates the
1476 * extents. However, due to an operation such as scrub turning a block
1477 * group to RO mode, it may fallback to COW mode, so we must make sure
1478 * an extent allocated during COW has exactly the requested size and can
1479 * not be split into smaller extents, otherwise relocation breaks and
1480 * fails during the stage where it updates the bytenr of file extent
1483 if (btrfs_is_data_reloc_root(root
))
1484 min_alloc_size
= num_bytes
;
1486 min_alloc_size
= fs_info
->sectorsize
;
1488 while (num_bytes
> 0) {
1489 cur_alloc_size
= num_bytes
;
1490 ret
= btrfs_reserve_extent(root
, cur_alloc_size
, cur_alloc_size
,
1491 min_alloc_size
, 0, alloc_hint
,
1495 cur_alloc_size
= ins
.offset
;
1496 extent_reserved
= true;
1498 ram_size
= ins
.offset
;
1499 em
= create_io_em(inode
, start
, ins
.offset
, /* len */
1500 start
, /* orig_start */
1501 ins
.objectid
, /* block_start */
1502 ins
.offset
, /* block_len */
1503 ins
.offset
, /* orig_block_len */
1504 ram_size
, /* ram_bytes */
1505 BTRFS_COMPRESS_NONE
, /* compress_type */
1506 BTRFS_ORDERED_REGULAR
/* type */);
1511 free_extent_map(em
);
1513 ret
= btrfs_add_ordered_extent(inode
, start
, ram_size
, ram_size
,
1514 ins
.objectid
, cur_alloc_size
, 0,
1515 1 << BTRFS_ORDERED_REGULAR
,
1516 BTRFS_COMPRESS_NONE
);
1518 goto out_drop_extent_cache
;
1520 if (btrfs_is_data_reloc_root(root
)) {
1521 ret
= btrfs_reloc_clone_csums(inode
, start
,
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,
1540 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1543 * We're not doing compressed IO, don't unlock the first page
1544 * (which the caller expects to stay locked), don't clear any
1545 * dirty bits and don't set any writeback bits
1547 * Do set the Ordered (Private2) bit so we know this page was
1548 * properly setup for writepage.
1550 page_ops
= unlock
? PAGE_UNLOCK
: 0;
1551 page_ops
|= PAGE_SET_ORDERED
;
1553 extent_clear_unlock_delalloc(inode
, start
, start
+ ram_size
- 1,
1555 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1557 if (num_bytes
< cur_alloc_size
)
1560 num_bytes
-= cur_alloc_size
;
1561 alloc_hint
= ins
.objectid
+ ins
.offset
;
1562 start
+= cur_alloc_size
;
1563 extent_reserved
= false;
1566 * btrfs_reloc_clone_csums() error, since start is increased
1567 * extent_clear_unlock_delalloc() at out_unlock label won't
1568 * free metadata of current ordered extent, we're OK to exit.
1576 out_drop_extent_cache
:
1577 btrfs_drop_extent_map_range(inode
, start
, start
+ ram_size
- 1, false);
1579 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
1580 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
1583 * If done_offset is non-NULL and ret == -EAGAIN, we expect the
1584 * caller to write out the successfully allocated region and retry.
1586 if (done_offset
&& ret
== -EAGAIN
) {
1587 if (orig_start
< start
)
1588 *done_offset
= start
- 1;
1590 *done_offset
= start
;
1592 } else if (ret
== -EAGAIN
) {
1593 /* Convert to -ENOSPC since the caller cannot retry. */
1598 * Now, we have three regions to clean up:
1600 * |-------(1)----|---(2)---|-------------(3)----------|
1601 * `- orig_start `- start `- start + cur_alloc_size `- end
1603 * We process each region below.
1606 clear_bits
= EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DELALLOC_NEW
|
1607 EXTENT_DEFRAG
| EXTENT_CLEAR_META_RESV
;
1608 page_ops
= PAGE_UNLOCK
| PAGE_START_WRITEBACK
| PAGE_END_WRITEBACK
;
1611 * For the range (1). We have already instantiated the ordered extents
1612 * for this region. They are cleaned up by
1613 * btrfs_cleanup_ordered_extents() in e.g,
1614 * btrfs_run_delalloc_range(). EXTENT_LOCKED | EXTENT_DELALLOC are
1615 * already cleared in the above loop. And, EXTENT_DELALLOC_NEW |
1616 * EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV are handled by the cleanup
1619 * However, in case of unlock == 0, we still need to unlock the pages
1620 * (except @locked_page) to ensure all the pages are unlocked.
1622 if (!unlock
&& orig_start
< start
) {
1624 mapping_set_error(inode
->vfs_inode
.i_mapping
, ret
);
1625 extent_clear_unlock_delalloc(inode
, orig_start
, start
- 1,
1626 locked_page
, 0, page_ops
);
1630 * For the range (2). If we reserved an extent for our delalloc range
1631 * (or a subrange) and failed to create the respective ordered extent,
1632 * then it means that when we reserved the extent we decremented the
1633 * extent's size from the data space_info's bytes_may_use counter and
1634 * incremented the space_info's bytes_reserved counter by the same
1635 * amount. We must make sure extent_clear_unlock_delalloc() does not try
1636 * to decrement again the data space_info's bytes_may_use counter,
1637 * therefore we do not pass it the flag EXTENT_CLEAR_DATA_RESV.
1639 if (extent_reserved
) {
1640 extent_clear_unlock_delalloc(inode
, start
,
1641 start
+ cur_alloc_size
- 1,
1645 start
+= cur_alloc_size
;
1651 * For the range (3). We never touched the region. In addition to the
1652 * clear_bits above, we add EXTENT_CLEAR_DATA_RESV to release the data
1653 * space_info's bytes_may_use counter, reserved in
1654 * btrfs_check_data_free_space().
1656 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1657 clear_bits
| EXTENT_CLEAR_DATA_RESV
,
1663 * work queue call back to started compression on a file and pages
1665 static noinline
void async_cow_start(struct btrfs_work
*work
)
1667 struct async_chunk
*async_chunk
;
1668 int compressed_extents
;
1670 async_chunk
= container_of(work
, struct async_chunk
, work
);
1672 compressed_extents
= compress_file_range(async_chunk
);
1673 if (compressed_extents
== 0) {
1674 btrfs_add_delayed_iput(async_chunk
->inode
);
1675 async_chunk
->inode
= NULL
;
1680 * work queue call back to submit previously compressed pages
1682 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1684 struct async_chunk
*async_chunk
= container_of(work
, struct async_chunk
,
1686 struct btrfs_fs_info
*fs_info
= btrfs_work_owner(work
);
1687 unsigned long nr_pages
;
1689 nr_pages
= (async_chunk
->end
- async_chunk
->start
+ PAGE_SIZE
) >>
1693 * ->inode could be NULL if async_chunk_start has failed to compress,
1694 * in which case we don't have anything to submit, yet we need to
1695 * always adjust ->async_delalloc_pages as its paired with the init
1696 * happening in cow_file_range_async
1698 if (async_chunk
->inode
)
1699 submit_compressed_extents(async_chunk
);
1701 /* atomic_sub_return implies a barrier */
1702 if (atomic_sub_return(nr_pages
, &fs_info
->async_delalloc_pages
) <
1704 cond_wake_up_nomb(&fs_info
->async_submit_wait
);
1707 static noinline
void async_cow_free(struct btrfs_work
*work
)
1709 struct async_chunk
*async_chunk
;
1710 struct async_cow
*async_cow
;
1712 async_chunk
= container_of(work
, struct async_chunk
, work
);
1713 if (async_chunk
->inode
)
1714 btrfs_add_delayed_iput(async_chunk
->inode
);
1715 if (async_chunk
->blkcg_css
)
1716 css_put(async_chunk
->blkcg_css
);
1718 async_cow
= async_chunk
->async_cow
;
1719 if (atomic_dec_and_test(&async_cow
->num_chunks
))
1723 static int cow_file_range_async(struct btrfs_inode
*inode
,
1724 struct writeback_control
*wbc
,
1725 struct page
*locked_page
,
1726 u64 start
, u64 end
, int *page_started
,
1727 unsigned long *nr_written
)
1729 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
1730 struct cgroup_subsys_state
*blkcg_css
= wbc_blkcg_css(wbc
);
1731 struct async_cow
*ctx
;
1732 struct async_chunk
*async_chunk
;
1733 unsigned long nr_pages
;
1735 u64 num_chunks
= DIV_ROUND_UP(end
- start
, SZ_512K
);
1737 bool should_compress
;
1739 const blk_opf_t write_flags
= wbc_to_write_flags(wbc
);
1741 unlock_extent(&inode
->io_tree
, start
, end
, NULL
);
1743 if (inode
->flags
& BTRFS_INODE_NOCOMPRESS
&&
1744 !btrfs_test_opt(fs_info
, FORCE_COMPRESS
)) {
1746 should_compress
= false;
1748 should_compress
= true;
1751 nofs_flag
= memalloc_nofs_save();
1752 ctx
= kvmalloc(struct_size(ctx
, chunks
, num_chunks
), GFP_KERNEL
);
1753 memalloc_nofs_restore(nofs_flag
);
1756 unsigned clear_bits
= EXTENT_LOCKED
| EXTENT_DELALLOC
|
1757 EXTENT_DELALLOC_NEW
| EXTENT_DEFRAG
|
1758 EXTENT_DO_ACCOUNTING
;
1759 unsigned long page_ops
= PAGE_UNLOCK
| PAGE_START_WRITEBACK
|
1760 PAGE_END_WRITEBACK
| PAGE_SET_ERROR
;
1762 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1763 clear_bits
, page_ops
);
1767 async_chunk
= ctx
->chunks
;
1768 atomic_set(&ctx
->num_chunks
, num_chunks
);
1770 for (i
= 0; i
< num_chunks
; i
++) {
1771 if (should_compress
)
1772 cur_end
= min(end
, start
+ SZ_512K
- 1);
1777 * igrab is called higher up in the call chain, take only the
1778 * lightweight reference for the callback lifetime
1780 ihold(&inode
->vfs_inode
);
1781 async_chunk
[i
].async_cow
= ctx
;
1782 async_chunk
[i
].inode
= inode
;
1783 async_chunk
[i
].start
= start
;
1784 async_chunk
[i
].end
= cur_end
;
1785 async_chunk
[i
].write_flags
= write_flags
;
1786 INIT_LIST_HEAD(&async_chunk
[i
].extents
);
1789 * The locked_page comes all the way from writepage and its
1790 * the original page we were actually given. As we spread
1791 * this large delalloc region across multiple async_chunk
1792 * structs, only the first struct needs a pointer to locked_page
1794 * This way we don't need racey decisions about who is supposed
1799 * Depending on the compressibility, the pages might or
1800 * might not go through async. We want all of them to
1801 * be accounted against wbc once. Let's do it here
1802 * before the paths diverge. wbc accounting is used
1803 * only for foreign writeback detection and doesn't
1804 * need full accuracy. Just account the whole thing
1805 * against the first page.
1807 wbc_account_cgroup_owner(wbc
, locked_page
,
1809 async_chunk
[i
].locked_page
= locked_page
;
1812 async_chunk
[i
].locked_page
= NULL
;
1815 if (blkcg_css
!= blkcg_root_css
) {
1817 async_chunk
[i
].blkcg_css
= blkcg_css
;
1818 async_chunk
[i
].write_flags
|= REQ_BTRFS_CGROUP_PUNT
;
1820 async_chunk
[i
].blkcg_css
= NULL
;
1823 btrfs_init_work(&async_chunk
[i
].work
, async_cow_start
,
1824 async_cow_submit
, async_cow_free
);
1826 nr_pages
= DIV_ROUND_UP(cur_end
- start
, PAGE_SIZE
);
1827 atomic_add(nr_pages
, &fs_info
->async_delalloc_pages
);
1829 btrfs_queue_work(fs_info
->delalloc_workers
, &async_chunk
[i
].work
);
1831 *nr_written
+= nr_pages
;
1832 start
= cur_end
+ 1;
1838 static noinline
int run_delalloc_zoned(struct btrfs_inode
*inode
,
1839 struct page
*locked_page
, u64 start
,
1840 u64 end
, int *page_started
,
1841 unsigned long *nr_written
)
1843 u64 done_offset
= end
;
1845 bool locked_page_done
= false;
1847 while (start
<= end
) {
1848 ret
= cow_file_range(inode
, locked_page
, start
, end
, page_started
,
1849 nr_written
, 0, &done_offset
);
1850 if (ret
&& ret
!= -EAGAIN
)
1853 if (*page_started
) {
1861 if (done_offset
== start
) {
1862 wait_on_bit_io(&inode
->root
->fs_info
->flags
,
1863 BTRFS_FS_NEED_ZONE_FINISH
,
1864 TASK_UNINTERRUPTIBLE
);
1868 if (!locked_page_done
) {
1869 __set_page_dirty_nobuffers(locked_page
);
1870 account_page_redirty(locked_page
);
1872 locked_page_done
= true;
1873 extent_write_locked_range(&inode
->vfs_inode
, start
, done_offset
);
1875 start
= done_offset
+ 1;
1883 static noinline
int csum_exist_in_range(struct btrfs_fs_info
*fs_info
,
1884 u64 bytenr
, u64 num_bytes
, bool nowait
)
1886 struct btrfs_root
*csum_root
= btrfs_csum_root(fs_info
, bytenr
);
1887 struct btrfs_ordered_sum
*sums
;
1891 ret
= btrfs_lookup_csums_list(csum_root
, bytenr
, bytenr
+ num_bytes
- 1,
1893 if (ret
== 0 && list_empty(&list
))
1896 while (!list_empty(&list
)) {
1897 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1898 list_del(&sums
->list
);
1906 static int fallback_to_cow(struct btrfs_inode
*inode
, struct page
*locked_page
,
1907 const u64 start
, const u64 end
,
1908 int *page_started
, unsigned long *nr_written
)
1910 const bool is_space_ino
= btrfs_is_free_space_inode(inode
);
1911 const bool is_reloc_ino
= btrfs_is_data_reloc_root(inode
->root
);
1912 const u64 range_bytes
= end
+ 1 - start
;
1913 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
1914 u64 range_start
= start
;
1918 * If EXTENT_NORESERVE is set it means that when the buffered write was
1919 * made we had not enough available data space and therefore we did not
1920 * reserve data space for it, since we though we could do NOCOW for the
1921 * respective file range (either there is prealloc extent or the inode
1922 * has the NOCOW bit set).
1924 * However when we need to fallback to COW mode (because for example the
1925 * block group for the corresponding extent was turned to RO mode by a
1926 * scrub or relocation) we need to do the following:
1928 * 1) We increment the bytes_may_use counter of the data space info.
1929 * If COW succeeds, it allocates a new data extent and after doing
1930 * that it decrements the space info's bytes_may_use counter and
1931 * increments its bytes_reserved counter by the same amount (we do
1932 * this at btrfs_add_reserved_bytes()). So we need to increment the
1933 * bytes_may_use counter to compensate (when space is reserved at
1934 * buffered write time, the bytes_may_use counter is incremented);
1936 * 2) We clear the EXTENT_NORESERVE bit from the range. We do this so
1937 * that if the COW path fails for any reason, it decrements (through
1938 * extent_clear_unlock_delalloc()) the bytes_may_use counter of the
1939 * data space info, which we incremented in the step above.
1941 * If we need to fallback to cow and the inode corresponds to a free
1942 * space cache inode or an inode of the data relocation tree, we must
1943 * also increment bytes_may_use of the data space_info for the same
1944 * reason. Space caches and relocated data extents always get a prealloc
1945 * extent for them, however scrub or balance may have set the block
1946 * group that contains that extent to RO mode and therefore force COW
1947 * when starting writeback.
1949 count
= count_range_bits(io_tree
, &range_start
, end
, range_bytes
,
1950 EXTENT_NORESERVE
, 0, NULL
);
1951 if (count
> 0 || is_space_ino
|| is_reloc_ino
) {
1953 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
1954 struct btrfs_space_info
*sinfo
= fs_info
->data_sinfo
;
1956 if (is_space_ino
|| is_reloc_ino
)
1957 bytes
= range_bytes
;
1959 spin_lock(&sinfo
->lock
);
1960 btrfs_space_info_update_bytes_may_use(fs_info
, sinfo
, bytes
);
1961 spin_unlock(&sinfo
->lock
);
1964 clear_extent_bit(io_tree
, start
, end
, EXTENT_NORESERVE
,
1968 return cow_file_range(inode
, locked_page
, start
, end
, page_started
,
1969 nr_written
, 1, NULL
);
1972 struct can_nocow_file_extent_args
{
1975 /* Start file offset of the range we want to NOCOW. */
1977 /* End file offset (inclusive) of the range we want to NOCOW. */
1979 bool writeback_path
;
1982 * Free the path passed to can_nocow_file_extent() once it's not needed
1987 /* Output fields. Only set when can_nocow_file_extent() returns 1. */
1992 /* Number of bytes that can be written to in NOCOW mode. */
1997 * Check if we can NOCOW the file extent that the path points to.
1998 * This function may return with the path released, so the caller should check
1999 * if path->nodes[0] is NULL or not if it needs to use the path afterwards.
2001 * Returns: < 0 on error
2002 * 0 if we can not NOCOW
2005 static int can_nocow_file_extent(struct btrfs_path
*path
,
2006 struct btrfs_key
*key
,
2007 struct btrfs_inode
*inode
,
2008 struct can_nocow_file_extent_args
*args
)
2010 const bool is_freespace_inode
= btrfs_is_free_space_inode(inode
);
2011 struct extent_buffer
*leaf
= path
->nodes
[0];
2012 struct btrfs_root
*root
= inode
->root
;
2013 struct btrfs_file_extent_item
*fi
;
2018 bool nowait
= path
->nowait
;
2020 fi
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_file_extent_item
);
2021 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2023 if (extent_type
== BTRFS_FILE_EXTENT_INLINE
)
2026 /* Can't access these fields unless we know it's not an inline extent. */
2027 args
->disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2028 args
->disk_num_bytes
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
2029 args
->extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
2031 if (!(inode
->flags
& BTRFS_INODE_NODATACOW
) &&
2032 extent_type
== BTRFS_FILE_EXTENT_REG
)
2036 * If the extent was created before the generation where the last snapshot
2037 * for its subvolume was created, then this implies the extent is shared,
2038 * hence we must COW.
2040 if (!args
->strict
&&
2041 btrfs_file_extent_generation(leaf
, fi
) <=
2042 btrfs_root_last_snapshot(&root
->root_item
))
2045 /* An explicit hole, must COW. */
2046 if (args
->disk_bytenr
== 0)
2049 /* Compressed/encrypted/encoded extents must be COWed. */
2050 if (btrfs_file_extent_compression(leaf
, fi
) ||
2051 btrfs_file_extent_encryption(leaf
, fi
) ||
2052 btrfs_file_extent_other_encoding(leaf
, fi
))
2055 extent_end
= btrfs_file_extent_end(path
);
2058 * The following checks can be expensive, as they need to take other
2059 * locks and do btree or rbtree searches, so release the path to avoid
2060 * blocking other tasks for too long.
2062 btrfs_release_path(path
);
2064 ret
= btrfs_cross_ref_exist(root
, btrfs_ino(inode
),
2065 key
->offset
- args
->extent_offset
,
2066 args
->disk_bytenr
, args
->strict
, path
);
2067 WARN_ON_ONCE(ret
> 0 && is_freespace_inode
);
2071 if (args
->free_path
) {
2073 * We don't need the path anymore, plus through the
2074 * csum_exist_in_range() call below we will end up allocating
2075 * another path. So free the path to avoid unnecessary extra
2078 btrfs_free_path(path
);
2082 /* If there are pending snapshots for this root, we must COW. */
2083 if (args
->writeback_path
&& !is_freespace_inode
&&
2084 atomic_read(&root
->snapshot_force_cow
))
2087 args
->disk_bytenr
+= args
->extent_offset
;
2088 args
->disk_bytenr
+= args
->start
- key
->offset
;
2089 args
->num_bytes
= min(args
->end
+ 1, extent_end
) - args
->start
;
2092 * Force COW if csums exist in the range. This ensures that csums for a
2093 * given extent are either valid or do not exist.
2095 ret
= csum_exist_in_range(root
->fs_info
, args
->disk_bytenr
, args
->num_bytes
,
2097 WARN_ON_ONCE(ret
> 0 && is_freespace_inode
);
2103 if (args
->free_path
&& path
)
2104 btrfs_free_path(path
);
2106 return ret
< 0 ? ret
: can_nocow
;
2110 * when nowcow writeback call back. This checks for snapshots or COW copies
2111 * of the extents that exist in the file, and COWs the file as required.
2113 * If no cow copies or snapshots exist, we write directly to the existing
2116 static noinline
int run_delalloc_nocow(struct btrfs_inode
*inode
,
2117 struct page
*locked_page
,
2118 const u64 start
, const u64 end
,
2120 unsigned long *nr_written
)
2122 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2123 struct btrfs_root
*root
= inode
->root
;
2124 struct btrfs_path
*path
;
2125 u64 cow_start
= (u64
)-1;
2126 u64 cur_offset
= start
;
2128 bool check_prev
= true;
2129 u64 ino
= btrfs_ino(inode
);
2130 struct btrfs_block_group
*bg
;
2132 struct can_nocow_file_extent_args nocow_args
= { 0 };
2134 path
= btrfs_alloc_path();
2136 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
2137 EXTENT_LOCKED
| EXTENT_DELALLOC
|
2138 EXTENT_DO_ACCOUNTING
|
2139 EXTENT_DEFRAG
, PAGE_UNLOCK
|
2140 PAGE_START_WRITEBACK
|
2141 PAGE_END_WRITEBACK
);
2145 nocow_args
.end
= end
;
2146 nocow_args
.writeback_path
= true;
2149 struct btrfs_key found_key
;
2150 struct btrfs_file_extent_item
*fi
;
2151 struct extent_buffer
*leaf
;
2159 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, ino
,
2165 * If there is no extent for our range when doing the initial
2166 * search, then go back to the previous slot as it will be the
2167 * one containing the search offset
2169 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
2170 leaf
= path
->nodes
[0];
2171 btrfs_item_key_to_cpu(leaf
, &found_key
,
2172 path
->slots
[0] - 1);
2173 if (found_key
.objectid
== ino
&&
2174 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
2179 /* Go to next leaf if we have exhausted the current one */
2180 leaf
= path
->nodes
[0];
2181 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2182 ret
= btrfs_next_leaf(root
, path
);
2184 if (cow_start
!= (u64
)-1)
2185 cur_offset
= cow_start
;
2190 leaf
= path
->nodes
[0];
2193 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2195 /* Didn't find anything for our INO */
2196 if (found_key
.objectid
> ino
)
2199 * Keep searching until we find an EXTENT_ITEM or there are no
2200 * more extents for this inode
2202 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
2203 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
2208 /* Found key is not EXTENT_DATA_KEY or starts after req range */
2209 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
2210 found_key
.offset
> end
)
2214 * If the found extent starts after requested offset, then
2215 * adjust extent_end to be right before this extent begins
2217 if (found_key
.offset
> cur_offset
) {
2218 extent_end
= found_key
.offset
;
2224 * Found extent which begins before our range and potentially
2227 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2228 struct btrfs_file_extent_item
);
2229 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2230 /* If this is triggered then we have a memory corruption. */
2231 ASSERT(extent_type
< BTRFS_NR_FILE_EXTENT_TYPES
);
2232 if (WARN_ON(extent_type
>= BTRFS_NR_FILE_EXTENT_TYPES
)) {
2236 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
2237 extent_end
= btrfs_file_extent_end(path
);
2240 * If the extent we got ends before our current offset, skip to
2243 if (extent_end
<= cur_offset
) {
2248 nocow_args
.start
= cur_offset
;
2249 ret
= can_nocow_file_extent(path
, &found_key
, inode
, &nocow_args
);
2251 if (cow_start
!= (u64
)-1)
2252 cur_offset
= cow_start
;
2254 } else if (ret
== 0) {
2259 bg
= btrfs_inc_nocow_writers(fs_info
, nocow_args
.disk_bytenr
);
2264 * If nocow is false then record the beginning of the range
2265 * that needs to be COWed
2268 if (cow_start
== (u64
)-1)
2269 cow_start
= cur_offset
;
2270 cur_offset
= extent_end
;
2271 if (cur_offset
> end
)
2273 if (!path
->nodes
[0])
2280 * COW range from cow_start to found_key.offset - 1. As the key
2281 * will contain the beginning of the first extent that can be
2282 * NOCOW, following one which needs to be COW'ed
2284 if (cow_start
!= (u64
)-1) {
2285 ret
= fallback_to_cow(inode
, locked_page
,
2286 cow_start
, found_key
.offset
- 1,
2287 page_started
, nr_written
);
2290 cow_start
= (u64
)-1;
2293 nocow_end
= cur_offset
+ nocow_args
.num_bytes
- 1;
2295 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
2296 u64 orig_start
= found_key
.offset
- nocow_args
.extent_offset
;
2297 struct extent_map
*em
;
2299 em
= create_io_em(inode
, cur_offset
, nocow_args
.num_bytes
,
2301 nocow_args
.disk_bytenr
, /* block_start */
2302 nocow_args
.num_bytes
, /* block_len */
2303 nocow_args
.disk_num_bytes
, /* orig_block_len */
2304 ram_bytes
, BTRFS_COMPRESS_NONE
,
2305 BTRFS_ORDERED_PREALLOC
);
2310 free_extent_map(em
);
2311 ret
= btrfs_add_ordered_extent(inode
,
2312 cur_offset
, nocow_args
.num_bytes
,
2313 nocow_args
.num_bytes
,
2314 nocow_args
.disk_bytenr
,
2315 nocow_args
.num_bytes
, 0,
2316 1 << BTRFS_ORDERED_PREALLOC
,
2317 BTRFS_COMPRESS_NONE
);
2319 btrfs_drop_extent_map_range(inode
, cur_offset
,
2324 ret
= btrfs_add_ordered_extent(inode
, cur_offset
,
2325 nocow_args
.num_bytes
,
2326 nocow_args
.num_bytes
,
2327 nocow_args
.disk_bytenr
,
2328 nocow_args
.num_bytes
,
2330 1 << BTRFS_ORDERED_NOCOW
,
2331 BTRFS_COMPRESS_NONE
);
2337 btrfs_dec_nocow_writers(bg
);
2341 if (btrfs_is_data_reloc_root(root
))
2343 * Error handled later, as we must prevent
2344 * extent_clear_unlock_delalloc() in error handler
2345 * from freeing metadata of created ordered extent.
2347 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
2348 nocow_args
.num_bytes
);
2350 extent_clear_unlock_delalloc(inode
, cur_offset
, nocow_end
,
2351 locked_page
, EXTENT_LOCKED
|
2353 EXTENT_CLEAR_DATA_RESV
,
2354 PAGE_UNLOCK
| PAGE_SET_ORDERED
);
2356 cur_offset
= extent_end
;
2359 * btrfs_reloc_clone_csums() error, now we're OK to call error
2360 * handler, as metadata for created ordered extent will only
2361 * be freed by btrfs_finish_ordered_io().
2365 if (cur_offset
> end
)
2368 btrfs_release_path(path
);
2370 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
2371 cow_start
= cur_offset
;
2373 if (cow_start
!= (u64
)-1) {
2375 ret
= fallback_to_cow(inode
, locked_page
, cow_start
, end
,
2376 page_started
, nr_written
);
2383 btrfs_dec_nocow_writers(bg
);
2385 if (ret
&& cur_offset
< end
)
2386 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
2387 locked_page
, EXTENT_LOCKED
|
2388 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
2389 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
2390 PAGE_START_WRITEBACK
|
2391 PAGE_END_WRITEBACK
);
2392 btrfs_free_path(path
);
2396 static bool should_nocow(struct btrfs_inode
*inode
, u64 start
, u64 end
)
2398 if (inode
->flags
& (BTRFS_INODE_NODATACOW
| BTRFS_INODE_PREALLOC
)) {
2399 if (inode
->defrag_bytes
&&
2400 test_range_bit(&inode
->io_tree
, start
, end
, EXTENT_DEFRAG
,
2409 * Function to process delayed allocation (create CoW) for ranges which are
2410 * being touched for the first time.
2412 int btrfs_run_delalloc_range(struct btrfs_inode
*inode
, struct page
*locked_page
,
2413 u64 start
, u64 end
, int *page_started
, unsigned long *nr_written
,
2414 struct writeback_control
*wbc
)
2417 const bool zoned
= btrfs_is_zoned(inode
->root
->fs_info
);
2420 * The range must cover part of the @locked_page, or the returned
2421 * @page_started can confuse the caller.
2423 ASSERT(!(end
<= page_offset(locked_page
) ||
2424 start
>= page_offset(locked_page
) + PAGE_SIZE
));
2426 if (should_nocow(inode
, start
, end
)) {
2428 * Normally on a zoned device we're only doing COW writes, but
2429 * in case of relocation on a zoned filesystem we have taken
2430 * precaution, that we're only writing sequentially. It's safe
2431 * to use run_delalloc_nocow() here, like for regular
2432 * preallocated inodes.
2434 ASSERT(!zoned
|| btrfs_is_data_reloc_root(inode
->root
));
2435 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
2436 page_started
, nr_written
);
2437 } else if (!btrfs_inode_can_compress(inode
) ||
2438 !inode_need_compress(inode
, start
, end
)) {
2440 ret
= run_delalloc_zoned(inode
, locked_page
, start
, end
,
2441 page_started
, nr_written
);
2443 ret
= cow_file_range(inode
, locked_page
, start
, end
,
2444 page_started
, nr_written
, 1, NULL
);
2446 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
, &inode
->runtime_flags
);
2447 ret
= cow_file_range_async(inode
, wbc
, locked_page
, start
, end
,
2448 page_started
, nr_written
);
2452 btrfs_cleanup_ordered_extents(inode
, locked_page
, start
,
2457 void btrfs_split_delalloc_extent(struct btrfs_inode
*inode
,
2458 struct extent_state
*orig
, u64 split
)
2460 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2463 /* not delalloc, ignore it */
2464 if (!(orig
->state
& EXTENT_DELALLOC
))
2467 size
= orig
->end
- orig
->start
+ 1;
2468 if (size
> fs_info
->max_extent_size
) {
2473 * See the explanation in btrfs_merge_delalloc_extent, the same
2474 * applies here, just in reverse.
2476 new_size
= orig
->end
- split
+ 1;
2477 num_extents
= count_max_extents(fs_info
, new_size
);
2478 new_size
= split
- orig
->start
;
2479 num_extents
+= count_max_extents(fs_info
, new_size
);
2480 if (count_max_extents(fs_info
, size
) >= num_extents
)
2484 spin_lock(&inode
->lock
);
2485 btrfs_mod_outstanding_extents(inode
, 1);
2486 spin_unlock(&inode
->lock
);
2490 * Handle merged delayed allocation extents so we can keep track of new extents
2491 * that are just merged onto old extents, such as when we are doing sequential
2492 * writes, so we can properly account for the metadata space we'll need.
2494 void btrfs_merge_delalloc_extent(struct btrfs_inode
*inode
, struct extent_state
*new,
2495 struct extent_state
*other
)
2497 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2498 u64 new_size
, old_size
;
2501 /* not delalloc, ignore it */
2502 if (!(other
->state
& EXTENT_DELALLOC
))
2505 if (new->start
> other
->start
)
2506 new_size
= new->end
- other
->start
+ 1;
2508 new_size
= other
->end
- new->start
+ 1;
2510 /* we're not bigger than the max, unreserve the space and go */
2511 if (new_size
<= fs_info
->max_extent_size
) {
2512 spin_lock(&inode
->lock
);
2513 btrfs_mod_outstanding_extents(inode
, -1);
2514 spin_unlock(&inode
->lock
);
2519 * We have to add up either side to figure out how many extents were
2520 * accounted for before we merged into one big extent. If the number of
2521 * extents we accounted for is <= the amount we need for the new range
2522 * then we can return, otherwise drop. Think of it like this
2526 * So we've grown the extent by a MAX_SIZE extent, this would mean we
2527 * need 2 outstanding extents, on one side we have 1 and the other side
2528 * we have 1 so they are == and we can return. But in this case
2530 * [MAX_SIZE+4k][MAX_SIZE+4k]
2532 * Each range on their own accounts for 2 extents, but merged together
2533 * they are only 3 extents worth of accounting, so we need to drop in
2536 old_size
= other
->end
- other
->start
+ 1;
2537 num_extents
= count_max_extents(fs_info
, old_size
);
2538 old_size
= new->end
- new->start
+ 1;
2539 num_extents
+= count_max_extents(fs_info
, old_size
);
2540 if (count_max_extents(fs_info
, new_size
) >= num_extents
)
2543 spin_lock(&inode
->lock
);
2544 btrfs_mod_outstanding_extents(inode
, -1);
2545 spin_unlock(&inode
->lock
);
2548 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
2549 struct btrfs_inode
*inode
)
2551 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2553 spin_lock(&root
->delalloc_lock
);
2554 if (list_empty(&inode
->delalloc_inodes
)) {
2555 list_add_tail(&inode
->delalloc_inodes
, &root
->delalloc_inodes
);
2556 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
, &inode
->runtime_flags
);
2557 root
->nr_delalloc_inodes
++;
2558 if (root
->nr_delalloc_inodes
== 1) {
2559 spin_lock(&fs_info
->delalloc_root_lock
);
2560 BUG_ON(!list_empty(&root
->delalloc_root
));
2561 list_add_tail(&root
->delalloc_root
,
2562 &fs_info
->delalloc_roots
);
2563 spin_unlock(&fs_info
->delalloc_root_lock
);
2566 spin_unlock(&root
->delalloc_lock
);
2569 void __btrfs_del_delalloc_inode(struct btrfs_root
*root
,
2570 struct btrfs_inode
*inode
)
2572 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2574 if (!list_empty(&inode
->delalloc_inodes
)) {
2575 list_del_init(&inode
->delalloc_inodes
);
2576 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
2577 &inode
->runtime_flags
);
2578 root
->nr_delalloc_inodes
--;
2579 if (!root
->nr_delalloc_inodes
) {
2580 ASSERT(list_empty(&root
->delalloc_inodes
));
2581 spin_lock(&fs_info
->delalloc_root_lock
);
2582 BUG_ON(list_empty(&root
->delalloc_root
));
2583 list_del_init(&root
->delalloc_root
);
2584 spin_unlock(&fs_info
->delalloc_root_lock
);
2589 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
2590 struct btrfs_inode
*inode
)
2592 spin_lock(&root
->delalloc_lock
);
2593 __btrfs_del_delalloc_inode(root
, inode
);
2594 spin_unlock(&root
->delalloc_lock
);
2598 * Properly track delayed allocation bytes in the inode and to maintain the
2599 * list of inodes that have pending delalloc work to be done.
2601 void btrfs_set_delalloc_extent(struct btrfs_inode
*inode
, struct extent_state
*state
,
2604 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2606 if ((bits
& EXTENT_DEFRAG
) && !(bits
& EXTENT_DELALLOC
))
2609 * set_bit and clear bit hooks normally require _irqsave/restore
2610 * but in this case, we are only testing for the DELALLOC
2611 * bit, which is only set or cleared with irqs on
2613 if (!(state
->state
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
2614 struct btrfs_root
*root
= inode
->root
;
2615 u64 len
= state
->end
+ 1 - state
->start
;
2616 u32 num_extents
= count_max_extents(fs_info
, len
);
2617 bool do_list
= !btrfs_is_free_space_inode(inode
);
2619 spin_lock(&inode
->lock
);
2620 btrfs_mod_outstanding_extents(inode
, num_extents
);
2621 spin_unlock(&inode
->lock
);
2623 /* For sanity tests */
2624 if (btrfs_is_testing(fs_info
))
2627 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, len
,
2628 fs_info
->delalloc_batch
);
2629 spin_lock(&inode
->lock
);
2630 inode
->delalloc_bytes
+= len
;
2631 if (bits
& EXTENT_DEFRAG
)
2632 inode
->defrag_bytes
+= len
;
2633 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
2634 &inode
->runtime_flags
))
2635 btrfs_add_delalloc_inodes(root
, inode
);
2636 spin_unlock(&inode
->lock
);
2639 if (!(state
->state
& EXTENT_DELALLOC_NEW
) &&
2640 (bits
& EXTENT_DELALLOC_NEW
)) {
2641 spin_lock(&inode
->lock
);
2642 inode
->new_delalloc_bytes
+= state
->end
+ 1 - state
->start
;
2643 spin_unlock(&inode
->lock
);
2648 * Once a range is no longer delalloc this function ensures that proper
2649 * accounting happens.
2651 void btrfs_clear_delalloc_extent(struct btrfs_inode
*inode
,
2652 struct extent_state
*state
, u32 bits
)
2654 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
2655 u64 len
= state
->end
+ 1 - state
->start
;
2656 u32 num_extents
= count_max_extents(fs_info
, len
);
2658 if ((state
->state
& EXTENT_DEFRAG
) && (bits
& EXTENT_DEFRAG
)) {
2659 spin_lock(&inode
->lock
);
2660 inode
->defrag_bytes
-= len
;
2661 spin_unlock(&inode
->lock
);
2665 * set_bit and clear bit hooks normally require _irqsave/restore
2666 * but in this case, we are only testing for the DELALLOC
2667 * bit, which is only set or cleared with irqs on
2669 if ((state
->state
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
2670 struct btrfs_root
*root
= inode
->root
;
2671 bool do_list
= !btrfs_is_free_space_inode(inode
);
2673 spin_lock(&inode
->lock
);
2674 btrfs_mod_outstanding_extents(inode
, -num_extents
);
2675 spin_unlock(&inode
->lock
);
2678 * We don't reserve metadata space for space cache inodes so we
2679 * don't need to call delalloc_release_metadata if there is an
2682 if (bits
& EXTENT_CLEAR_META_RESV
&&
2683 root
!= fs_info
->tree_root
)
2684 btrfs_delalloc_release_metadata(inode
, len
, false);
2686 /* For sanity tests. */
2687 if (btrfs_is_testing(fs_info
))
2690 if (!btrfs_is_data_reloc_root(root
) &&
2691 do_list
&& !(state
->state
& EXTENT_NORESERVE
) &&
2692 (bits
& EXTENT_CLEAR_DATA_RESV
))
2693 btrfs_free_reserved_data_space_noquota(fs_info
, len
);
2695 percpu_counter_add_batch(&fs_info
->delalloc_bytes
, -len
,
2696 fs_info
->delalloc_batch
);
2697 spin_lock(&inode
->lock
);
2698 inode
->delalloc_bytes
-= len
;
2699 if (do_list
&& inode
->delalloc_bytes
== 0 &&
2700 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
2701 &inode
->runtime_flags
))
2702 btrfs_del_delalloc_inode(root
, inode
);
2703 spin_unlock(&inode
->lock
);
2706 if ((state
->state
& EXTENT_DELALLOC_NEW
) &&
2707 (bits
& EXTENT_DELALLOC_NEW
)) {
2708 spin_lock(&inode
->lock
);
2709 ASSERT(inode
->new_delalloc_bytes
>= len
);
2710 inode
->new_delalloc_bytes
-= len
;
2711 if (bits
& EXTENT_ADD_INODE_BYTES
)
2712 inode_add_bytes(&inode
->vfs_inode
, len
);
2713 spin_unlock(&inode
->lock
);
2717 static int btrfs_extract_ordered_extent(struct btrfs_bio
*bbio
,
2718 struct btrfs_ordered_extent
*ordered
)
2720 u64 start
= (u64
)bbio
->bio
.bi_iter
.bi_sector
<< SECTOR_SHIFT
;
2721 u64 len
= bbio
->bio
.bi_iter
.bi_size
;
2722 struct btrfs_ordered_extent
*new;
2725 /* Must always be called for the beginning of an ordered extent. */
2726 if (WARN_ON_ONCE(start
!= ordered
->disk_bytenr
))
2729 /* No need to split if the ordered extent covers the entire bio. */
2730 if (ordered
->disk_num_bytes
== len
)
2734 * Don't split the extent_map for NOCOW extents, as we're writing into
2735 * a pre-existing one.
2737 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
2738 ret
= split_extent_map(bbio
->inode
, bbio
->file_offset
,
2739 ordered
->num_bytes
, len
);
2744 new = btrfs_split_ordered_extent(ordered
, len
);
2746 return PTR_ERR(new);
2747 btrfs_put_ordered_extent(new);
2753 * given a list of ordered sums record them in the inode. This happens
2754 * at IO completion time based on sums calculated at bio submission time.
2756 static int add_pending_csums(struct btrfs_trans_handle
*trans
,
2757 struct list_head
*list
)
2759 struct btrfs_ordered_sum
*sum
;
2760 struct btrfs_root
*csum_root
= NULL
;
2763 list_for_each_entry(sum
, list
, list
) {
2764 trans
->adding_csums
= true;
2766 csum_root
= btrfs_csum_root(trans
->fs_info
,
2768 ret
= btrfs_csum_file_blocks(trans
, csum_root
, sum
);
2769 trans
->adding_csums
= false;
2776 static int btrfs_find_new_delalloc_bytes(struct btrfs_inode
*inode
,
2779 struct extent_state
**cached_state
)
2781 u64 search_start
= start
;
2782 const u64 end
= start
+ len
- 1;
2784 while (search_start
< end
) {
2785 const u64 search_len
= end
- search_start
+ 1;
2786 struct extent_map
*em
;
2790 em
= btrfs_get_extent(inode
, NULL
, 0, search_start
, search_len
);
2794 if (em
->block_start
!= EXTENT_MAP_HOLE
)
2798 if (em
->start
< search_start
)
2799 em_len
-= search_start
- em
->start
;
2800 if (em_len
> search_len
)
2801 em_len
= search_len
;
2803 ret
= set_extent_bit(&inode
->io_tree
, search_start
,
2804 search_start
+ em_len
- 1,
2805 EXTENT_DELALLOC_NEW
, cached_state
);
2807 search_start
= extent_map_end(em
);
2808 free_extent_map(em
);
2815 int btrfs_set_extent_delalloc(struct btrfs_inode
*inode
, u64 start
, u64 end
,
2816 unsigned int extra_bits
,
2817 struct extent_state
**cached_state
)
2819 WARN_ON(PAGE_ALIGNED(end
));
2821 if (start
>= i_size_read(&inode
->vfs_inode
) &&
2822 !(inode
->flags
& BTRFS_INODE_PREALLOC
)) {
2824 * There can't be any extents following eof in this case so just
2825 * set the delalloc new bit for the range directly.
2827 extra_bits
|= EXTENT_DELALLOC_NEW
;
2831 ret
= btrfs_find_new_delalloc_bytes(inode
, start
,
2838 return set_extent_bit(&inode
->io_tree
, start
, end
,
2839 EXTENT_DELALLOC
| extra_bits
, cached_state
);
2842 /* see btrfs_writepage_start_hook for details on why this is required */
2843 struct btrfs_writepage_fixup
{
2845 struct btrfs_inode
*inode
;
2846 struct btrfs_work work
;
2849 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
2851 struct btrfs_writepage_fixup
*fixup
;
2852 struct btrfs_ordered_extent
*ordered
;
2853 struct extent_state
*cached_state
= NULL
;
2854 struct extent_changeset
*data_reserved
= NULL
;
2856 struct btrfs_inode
*inode
;
2860 bool free_delalloc_space
= true;
2862 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
2864 inode
= fixup
->inode
;
2865 page_start
= page_offset(page
);
2866 page_end
= page_offset(page
) + PAGE_SIZE
- 1;
2869 * This is similar to page_mkwrite, we need to reserve the space before
2870 * we take the page lock.
2872 ret
= btrfs_delalloc_reserve_space(inode
, &data_reserved
, page_start
,
2878 * Before we queued this fixup, we took a reference on the page.
2879 * page->mapping may go NULL, but it shouldn't be moved to a different
2882 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
2884 * Unfortunately this is a little tricky, either
2886 * 1) We got here and our page had already been dealt with and
2887 * we reserved our space, thus ret == 0, so we need to just
2888 * drop our space reservation and bail. This can happen the
2889 * first time we come into the fixup worker, or could happen
2890 * while waiting for the ordered extent.
2891 * 2) Our page was already dealt with, but we happened to get an
2892 * ENOSPC above from the btrfs_delalloc_reserve_space. In
2893 * this case we obviously don't have anything to release, but
2894 * because the page was already dealt with we don't want to
2895 * mark the page with an error, so make sure we're resetting
2896 * ret to 0. This is why we have this check _before_ the ret
2897 * check, because we do not want to have a surprise ENOSPC
2898 * when the page was already properly dealt with.
2901 btrfs_delalloc_release_extents(inode
, PAGE_SIZE
);
2902 btrfs_delalloc_release_space(inode
, data_reserved
,
2903 page_start
, PAGE_SIZE
,
2911 * We can't mess with the page state unless it is locked, so now that
2912 * it is locked bail if we failed to make our space reservation.
2917 lock_extent(&inode
->io_tree
, page_start
, page_end
, &cached_state
);
2919 /* already ordered? We're done */
2920 if (PageOrdered(page
))
2923 ordered
= btrfs_lookup_ordered_range(inode
, page_start
, PAGE_SIZE
);
2925 unlock_extent(&inode
->io_tree
, page_start
, page_end
,
2928 btrfs_start_ordered_extent(ordered
);
2929 btrfs_put_ordered_extent(ordered
);
2933 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
, 0,
2939 * Everything went as planned, we're now the owner of a dirty page with
2940 * delayed allocation bits set and space reserved for our COW
2943 * The page was dirty when we started, nothing should have cleaned it.
2945 BUG_ON(!PageDirty(page
));
2946 free_delalloc_space
= false;
2948 btrfs_delalloc_release_extents(inode
, PAGE_SIZE
);
2949 if (free_delalloc_space
)
2950 btrfs_delalloc_release_space(inode
, data_reserved
, page_start
,
2952 unlock_extent(&inode
->io_tree
, page_start
, page_end
, &cached_state
);
2956 * We hit ENOSPC or other errors. Update the mapping and page
2957 * to reflect the errors and clean the page.
2959 mapping_set_error(page
->mapping
, ret
);
2960 end_extent_writepage(page
, ret
, page_start
, page_end
);
2961 clear_page_dirty_for_io(page
);
2964 btrfs_page_clear_checked(inode
->root
->fs_info
, page
, page_start
, PAGE_SIZE
);
2968 extent_changeset_free(data_reserved
);
2970 * As a precaution, do a delayed iput in case it would be the last iput
2971 * that could need flushing space. Recursing back to fixup worker would
2974 btrfs_add_delayed_iput(inode
);
2978 * There are a few paths in the higher layers of the kernel that directly
2979 * set the page dirty bit without asking the filesystem if it is a
2980 * good idea. This causes problems because we want to make sure COW
2981 * properly happens and the data=ordered rules are followed.
2983 * In our case any range that doesn't have the ORDERED bit set
2984 * hasn't been properly setup for IO. We kick off an async process
2985 * to fix it up. The async helper will wait for ordered extents, set
2986 * the delalloc bit and make it safe to write the page.
2988 int btrfs_writepage_cow_fixup(struct page
*page
)
2990 struct inode
*inode
= page
->mapping
->host
;
2991 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
2992 struct btrfs_writepage_fixup
*fixup
;
2994 /* This page has ordered extent covering it already */
2995 if (PageOrdered(page
))
2999 * PageChecked is set below when we create a fixup worker for this page,
3000 * don't try to create another one if we're already PageChecked()
3002 * The extent_io writepage code will redirty the page if we send back
3005 if (PageChecked(page
))
3008 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
3013 * We are already holding a reference to this inode from
3014 * write_cache_pages. We need to hold it because the space reservation
3015 * takes place outside of the page lock, and we can't trust
3016 * page->mapping outside of the page lock.
3019 btrfs_page_set_checked(fs_info
, page
, page_offset(page
), PAGE_SIZE
);
3021 btrfs_init_work(&fixup
->work
, btrfs_writepage_fixup_worker
, NULL
, NULL
);
3023 fixup
->inode
= BTRFS_I(inode
);
3024 btrfs_queue_work(fs_info
->fixup_workers
, &fixup
->work
);
3029 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
3030 struct btrfs_inode
*inode
, u64 file_pos
,
3031 struct btrfs_file_extent_item
*stack_fi
,
3032 const bool update_inode_bytes
,
3033 u64 qgroup_reserved
)
3035 struct btrfs_root
*root
= inode
->root
;
3036 const u64 sectorsize
= root
->fs_info
->sectorsize
;
3037 struct btrfs_path
*path
;
3038 struct extent_buffer
*leaf
;
3039 struct btrfs_key ins
;
3040 u64 disk_num_bytes
= btrfs_stack_file_extent_disk_num_bytes(stack_fi
);
3041 u64 disk_bytenr
= btrfs_stack_file_extent_disk_bytenr(stack_fi
);
3042 u64 offset
= btrfs_stack_file_extent_offset(stack_fi
);
3043 u64 num_bytes
= btrfs_stack_file_extent_num_bytes(stack_fi
);
3044 u64 ram_bytes
= btrfs_stack_file_extent_ram_bytes(stack_fi
);
3045 struct btrfs_drop_extents_args drop_args
= { 0 };
3048 path
= btrfs_alloc_path();
3053 * we may be replacing one extent in the tree with another.
3054 * The new extent is pinned in the extent map, and we don't want
3055 * to drop it from the cache until it is completely in the btree.
3057 * So, tell btrfs_drop_extents to leave this extent in the cache.
3058 * the caller is expected to unpin it and allow it to be merged
3061 drop_args
.path
= path
;
3062 drop_args
.start
= file_pos
;
3063 drop_args
.end
= file_pos
+ num_bytes
;
3064 drop_args
.replace_extent
= true;
3065 drop_args
.extent_item_size
= sizeof(*stack_fi
);
3066 ret
= btrfs_drop_extents(trans
, root
, inode
, &drop_args
);
3070 if (!drop_args
.extent_inserted
) {
3071 ins
.objectid
= btrfs_ino(inode
);
3072 ins
.offset
= file_pos
;
3073 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
3075 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
3080 leaf
= path
->nodes
[0];
3081 btrfs_set_stack_file_extent_generation(stack_fi
, trans
->transid
);
3082 write_extent_buffer(leaf
, stack_fi
,
3083 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
3084 sizeof(struct btrfs_file_extent_item
));
3086 btrfs_mark_buffer_dirty(leaf
);
3087 btrfs_release_path(path
);
3090 * If we dropped an inline extent here, we know the range where it is
3091 * was not marked with the EXTENT_DELALLOC_NEW bit, so we update the
3092 * number of bytes only for that range containing the inline extent.
3093 * The remaining of the range will be processed when clearning the
3094 * EXTENT_DELALLOC_BIT bit through the ordered extent completion.
3096 if (file_pos
== 0 && !IS_ALIGNED(drop_args
.bytes_found
, sectorsize
)) {
3097 u64 inline_size
= round_down(drop_args
.bytes_found
, sectorsize
);
3099 inline_size
= drop_args
.bytes_found
- inline_size
;
3100 btrfs_update_inode_bytes(inode
, sectorsize
, inline_size
);
3101 drop_args
.bytes_found
-= inline_size
;
3102 num_bytes
-= sectorsize
;
3105 if (update_inode_bytes
)
3106 btrfs_update_inode_bytes(inode
, num_bytes
, drop_args
.bytes_found
);
3108 ins
.objectid
= disk_bytenr
;
3109 ins
.offset
= disk_num_bytes
;
3110 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
3112 ret
= btrfs_inode_set_file_extent_range(inode
, file_pos
, ram_bytes
);
3116 ret
= btrfs_alloc_reserved_file_extent(trans
, root
, btrfs_ino(inode
),
3118 qgroup_reserved
, &ins
);
3120 btrfs_free_path(path
);
3125 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info
*fs_info
,
3128 struct btrfs_block_group
*cache
;
3130 cache
= btrfs_lookup_block_group(fs_info
, start
);
3133 spin_lock(&cache
->lock
);
3134 cache
->delalloc_bytes
-= len
;
3135 spin_unlock(&cache
->lock
);
3137 btrfs_put_block_group(cache
);
3140 static int insert_ordered_extent_file_extent(struct btrfs_trans_handle
*trans
,
3141 struct btrfs_ordered_extent
*oe
)
3143 struct btrfs_file_extent_item stack_fi
;
3144 bool update_inode_bytes
;
3145 u64 num_bytes
= oe
->num_bytes
;
3146 u64 ram_bytes
= oe
->ram_bytes
;
3148 memset(&stack_fi
, 0, sizeof(stack_fi
));
3149 btrfs_set_stack_file_extent_type(&stack_fi
, BTRFS_FILE_EXTENT_REG
);
3150 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi
, oe
->disk_bytenr
);
3151 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi
,
3152 oe
->disk_num_bytes
);
3153 btrfs_set_stack_file_extent_offset(&stack_fi
, oe
->offset
);
3154 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &oe
->flags
)) {
3155 num_bytes
= oe
->truncated_len
;
3156 ram_bytes
= num_bytes
;
3158 btrfs_set_stack_file_extent_num_bytes(&stack_fi
, num_bytes
);
3159 btrfs_set_stack_file_extent_ram_bytes(&stack_fi
, ram_bytes
);
3160 btrfs_set_stack_file_extent_compression(&stack_fi
, oe
->compress_type
);
3161 /* Encryption and other encoding is reserved and all 0 */
3164 * For delalloc, when completing an ordered extent we update the inode's
3165 * bytes when clearing the range in the inode's io tree, so pass false
3166 * as the argument 'update_inode_bytes' to insert_reserved_file_extent(),
3167 * except if the ordered extent was truncated.
3169 update_inode_bytes
= test_bit(BTRFS_ORDERED_DIRECT
, &oe
->flags
) ||
3170 test_bit(BTRFS_ORDERED_ENCODED
, &oe
->flags
) ||
3171 test_bit(BTRFS_ORDERED_TRUNCATED
, &oe
->flags
);
3173 return insert_reserved_file_extent(trans
, BTRFS_I(oe
->inode
),
3174 oe
->file_offset
, &stack_fi
,
3175 update_inode_bytes
, oe
->qgroup_rsv
);
3179 * As ordered data IO finishes, this gets called so we can finish
3180 * an ordered extent if the range of bytes in the file it covers are
3183 int btrfs_finish_one_ordered(struct btrfs_ordered_extent
*ordered_extent
)
3185 struct btrfs_inode
*inode
= BTRFS_I(ordered_extent
->inode
);
3186 struct btrfs_root
*root
= inode
->root
;
3187 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3188 struct btrfs_trans_handle
*trans
= NULL
;
3189 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
3190 struct extent_state
*cached_state
= NULL
;
3192 int compress_type
= 0;
3194 u64 logical_len
= ordered_extent
->num_bytes
;
3195 bool freespace_inode
;
3196 bool truncated
= false;
3197 bool clear_reserved_extent
= true;
3198 unsigned int clear_bits
= EXTENT_DEFRAG
;
3200 start
= ordered_extent
->file_offset
;
3201 end
= start
+ ordered_extent
->num_bytes
- 1;
3203 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3204 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
) &&
3205 !test_bit(BTRFS_ORDERED_DIRECT
, &ordered_extent
->flags
) &&
3206 !test_bit(BTRFS_ORDERED_ENCODED
, &ordered_extent
->flags
))
3207 clear_bits
|= EXTENT_DELALLOC_NEW
;
3209 freespace_inode
= btrfs_is_free_space_inode(inode
);
3210 if (!freespace_inode
)
3211 btrfs_lockdep_acquire(fs_info
, btrfs_ordered_extent
);
3213 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
3218 if (btrfs_is_zoned(fs_info
))
3219 btrfs_zone_finish_endio(fs_info
, ordered_extent
->disk_bytenr
,
3220 ordered_extent
->disk_num_bytes
);
3222 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
3224 logical_len
= ordered_extent
->truncated_len
;
3225 /* Truncated the entire extent, don't bother adding */
3230 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
3231 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
3233 btrfs_inode_safe_disk_i_size_write(inode
, 0);
3234 if (freespace_inode
)
3235 trans
= btrfs_join_transaction_spacecache(root
);
3237 trans
= btrfs_join_transaction(root
);
3238 if (IS_ERR(trans
)) {
3239 ret
= PTR_ERR(trans
);
3243 trans
->block_rsv
= &inode
->block_rsv
;
3244 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3245 if (ret
) /* -ENOMEM or corruption */
3246 btrfs_abort_transaction(trans
, ret
);
3250 clear_bits
|= EXTENT_LOCKED
;
3251 lock_extent(io_tree
, start
, end
, &cached_state
);
3253 if (freespace_inode
)
3254 trans
= btrfs_join_transaction_spacecache(root
);
3256 trans
= btrfs_join_transaction(root
);
3257 if (IS_ERR(trans
)) {
3258 ret
= PTR_ERR(trans
);
3263 trans
->block_rsv
= &inode
->block_rsv
;
3265 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
3266 compress_type
= ordered_extent
->compress_type
;
3267 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
3268 BUG_ON(compress_type
);
3269 ret
= btrfs_mark_extent_written(trans
, inode
,
3270 ordered_extent
->file_offset
,
3271 ordered_extent
->file_offset
+
3273 btrfs_zoned_release_data_reloc_bg(fs_info
, ordered_extent
->disk_bytenr
,
3274 ordered_extent
->disk_num_bytes
);
3276 BUG_ON(root
== fs_info
->tree_root
);
3277 ret
= insert_ordered_extent_file_extent(trans
, ordered_extent
);
3279 clear_reserved_extent
= false;
3280 btrfs_release_delalloc_bytes(fs_info
,
3281 ordered_extent
->disk_bytenr
,
3282 ordered_extent
->disk_num_bytes
);
3285 unpin_extent_cache(&inode
->extent_tree
, ordered_extent
->file_offset
,
3286 ordered_extent
->num_bytes
, trans
->transid
);
3288 btrfs_abort_transaction(trans
, ret
);
3292 ret
= add_pending_csums(trans
, &ordered_extent
->list
);
3294 btrfs_abort_transaction(trans
, ret
);
3299 * If this is a new delalloc range, clear its new delalloc flag to
3300 * update the inode's number of bytes. This needs to be done first
3301 * before updating the inode item.
3303 if ((clear_bits
& EXTENT_DELALLOC_NEW
) &&
3304 !test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
))
3305 clear_extent_bit(&inode
->io_tree
, start
, end
,
3306 EXTENT_DELALLOC_NEW
| EXTENT_ADD_INODE_BYTES
,
3309 btrfs_inode_safe_disk_i_size_write(inode
, 0);
3310 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
3311 if (ret
) { /* -ENOMEM or corruption */
3312 btrfs_abort_transaction(trans
, ret
);
3317 clear_extent_bit(&inode
->io_tree
, start
, end
, clear_bits
,
3321 btrfs_end_transaction(trans
);
3323 if (ret
|| truncated
) {
3324 u64 unwritten_start
= start
;
3327 * If we failed to finish this ordered extent for any reason we
3328 * need to make sure BTRFS_ORDERED_IOERR is set on the ordered
3329 * extent, and mark the inode with the error if it wasn't
3330 * already set. Any error during writeback would have already
3331 * set the mapping error, so we need to set it if we're the ones
3332 * marking this ordered extent as failed.
3334 if (ret
&& !test_and_set_bit(BTRFS_ORDERED_IOERR
,
3335 &ordered_extent
->flags
))
3336 mapping_set_error(ordered_extent
->inode
->i_mapping
, -EIO
);
3339 unwritten_start
+= logical_len
;
3340 clear_extent_uptodate(io_tree
, unwritten_start
, end
, NULL
);
3342 /* Drop extent maps for the part of the extent we didn't write. */
3343 btrfs_drop_extent_map_range(inode
, unwritten_start
, end
, false);
3346 * If the ordered extent had an IOERR or something else went
3347 * wrong we need to return the space for this ordered extent
3348 * back to the allocator. We only free the extent in the
3349 * truncated case if we didn't write out the extent at all.
3351 * If we made it past insert_reserved_file_extent before we
3352 * errored out then we don't need to do this as the accounting
3353 * has already been done.
3355 if ((ret
|| !logical_len
) &&
3356 clear_reserved_extent
&&
3357 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
3358 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
3360 * Discard the range before returning it back to the
3363 if (ret
&& btrfs_test_opt(fs_info
, DISCARD_SYNC
))
3364 btrfs_discard_extent(fs_info
,
3365 ordered_extent
->disk_bytenr
,
3366 ordered_extent
->disk_num_bytes
,
3368 btrfs_free_reserved_extent(fs_info
,
3369 ordered_extent
->disk_bytenr
,
3370 ordered_extent
->disk_num_bytes
, 1);
3375 * This needs to be done to make sure anybody waiting knows we are done
3376 * updating everything for this ordered extent.
3378 btrfs_remove_ordered_extent(inode
, ordered_extent
);
3381 btrfs_put_ordered_extent(ordered_extent
);
3382 /* once for the tree */
3383 btrfs_put_ordered_extent(ordered_extent
);
3388 int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered
)
3390 if (btrfs_is_zoned(btrfs_sb(ordered
->inode
->i_sb
)) &&
3391 !test_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
))
3392 btrfs_finish_ordered_zoned(ordered
);
3393 return btrfs_finish_one_ordered(ordered
);
3396 void btrfs_writepage_endio_finish_ordered(struct btrfs_inode
*inode
,
3397 struct page
*page
, u64 start
,
3398 u64 end
, bool uptodate
)
3400 trace_btrfs_writepage_end_io_hook(inode
, start
, end
, uptodate
);
3402 btrfs_mark_ordered_io_finished(inode
, page
, start
, end
+ 1 - start
, uptodate
);
3406 * Verify the checksum for a single sector without any extra action that depend
3407 * on the type of I/O.
3409 int btrfs_check_sector_csum(struct btrfs_fs_info
*fs_info
, struct page
*page
,
3410 u32 pgoff
, u8
*csum
, const u8
* const csum_expected
)
3412 SHASH_DESC_ON_STACK(shash
, fs_info
->csum_shash
);
3415 ASSERT(pgoff
+ fs_info
->sectorsize
<= PAGE_SIZE
);
3417 shash
->tfm
= fs_info
->csum_shash
;
3419 kaddr
= kmap_local_page(page
) + pgoff
;
3420 crypto_shash_digest(shash
, kaddr
, fs_info
->sectorsize
, csum
);
3421 kunmap_local(kaddr
);
3423 if (memcmp(csum
, csum_expected
, fs_info
->csum_size
))
3429 * Verify the checksum of a single data sector.
3431 * @bbio: btrfs_io_bio which contains the csum
3432 * @dev: device the sector is on
3433 * @bio_offset: offset to the beginning of the bio (in bytes)
3434 * @bv: bio_vec to check
3436 * Check if the checksum on a data block is valid. When a checksum mismatch is
3437 * detected, report the error and fill the corrupted range with zero.
3439 * Return %true if the sector is ok or had no checksum to start with, else %false.
3441 bool btrfs_data_csum_ok(struct btrfs_bio
*bbio
, struct btrfs_device
*dev
,
3442 u32 bio_offset
, struct bio_vec
*bv
)
3444 struct btrfs_inode
*inode
= bbio
->inode
;
3445 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
3446 u64 file_offset
= bbio
->file_offset
+ bio_offset
;
3447 u64 end
= file_offset
+ bv
->bv_len
- 1;
3449 u8 csum
[BTRFS_CSUM_SIZE
];
3451 ASSERT(bv
->bv_len
== fs_info
->sectorsize
);
3456 if (btrfs_is_data_reloc_root(inode
->root
) &&
3457 test_range_bit(&inode
->io_tree
, file_offset
, end
, EXTENT_NODATASUM
,
3459 /* Skip the range without csum for data reloc inode */
3460 clear_extent_bits(&inode
->io_tree
, file_offset
, end
,
3465 csum_expected
= bbio
->csum
+ (bio_offset
>> fs_info
->sectorsize_bits
) *
3467 if (btrfs_check_sector_csum(fs_info
, bv
->bv_page
, bv
->bv_offset
, csum
,
3473 btrfs_print_data_csum_error(inode
, file_offset
, csum
, csum_expected
,
3476 btrfs_dev_stat_inc_and_print(dev
, BTRFS_DEV_STAT_CORRUPTION_ERRS
);
3482 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3484 * @inode: The inode we want to perform iput on
3486 * This function uses the generic vfs_inode::i_count to track whether we should
3487 * just decrement it (in case it's > 1) or if this is the last iput then link
3488 * the inode to the delayed iput machinery. Delayed iputs are processed at
3489 * transaction commit time/superblock commit/cleaner kthread.
3491 void btrfs_add_delayed_iput(struct btrfs_inode
*inode
)
3493 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
3495 if (atomic_add_unless(&inode
->vfs_inode
.i_count
, -1, 1))
3498 atomic_inc(&fs_info
->nr_delayed_iputs
);
3499 spin_lock(&fs_info
->delayed_iput_lock
);
3500 ASSERT(list_empty(&inode
->delayed_iput
));
3501 list_add_tail(&inode
->delayed_iput
, &fs_info
->delayed_iputs
);
3502 spin_unlock(&fs_info
->delayed_iput_lock
);
3503 if (!test_bit(BTRFS_FS_CLEANER_RUNNING
, &fs_info
->flags
))
3504 wake_up_process(fs_info
->cleaner_kthread
);
3507 static void run_delayed_iput_locked(struct btrfs_fs_info
*fs_info
,
3508 struct btrfs_inode
*inode
)
3510 list_del_init(&inode
->delayed_iput
);
3511 spin_unlock(&fs_info
->delayed_iput_lock
);
3512 iput(&inode
->vfs_inode
);
3513 if (atomic_dec_and_test(&fs_info
->nr_delayed_iputs
))
3514 wake_up(&fs_info
->delayed_iputs_wait
);
3515 spin_lock(&fs_info
->delayed_iput_lock
);
3518 static void btrfs_run_delayed_iput(struct btrfs_fs_info
*fs_info
,
3519 struct btrfs_inode
*inode
)
3521 if (!list_empty(&inode
->delayed_iput
)) {
3522 spin_lock(&fs_info
->delayed_iput_lock
);
3523 if (!list_empty(&inode
->delayed_iput
))
3524 run_delayed_iput_locked(fs_info
, inode
);
3525 spin_unlock(&fs_info
->delayed_iput_lock
);
3529 void btrfs_run_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3532 spin_lock(&fs_info
->delayed_iput_lock
);
3533 while (!list_empty(&fs_info
->delayed_iputs
)) {
3534 struct btrfs_inode
*inode
;
3536 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3537 struct btrfs_inode
, delayed_iput
);
3538 run_delayed_iput_locked(fs_info
, inode
);
3539 cond_resched_lock(&fs_info
->delayed_iput_lock
);
3541 spin_unlock(&fs_info
->delayed_iput_lock
);
3545 * Wait for flushing all delayed iputs
3547 * @fs_info: the filesystem
3549 * This will wait on any delayed iputs that are currently running with KILLABLE
3550 * set. Once they are all done running we will return, unless we are killed in
3551 * which case we return EINTR. This helps in user operations like fallocate etc
3552 * that might get blocked on the iputs.
3554 * Return EINTR if we were killed, 0 if nothing's pending
3556 int btrfs_wait_on_delayed_iputs(struct btrfs_fs_info
*fs_info
)
3558 int ret
= wait_event_killable(fs_info
->delayed_iputs_wait
,
3559 atomic_read(&fs_info
->nr_delayed_iputs
) == 0);
3566 * This creates an orphan entry for the given inode in case something goes wrong
3567 * in the middle of an unlink.
3569 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
,
3570 struct btrfs_inode
*inode
)
3574 ret
= btrfs_insert_orphan_item(trans
, inode
->root
, btrfs_ino(inode
));
3575 if (ret
&& ret
!= -EEXIST
) {
3576 btrfs_abort_transaction(trans
, ret
);
3584 * We have done the delete so we can go ahead and remove the orphan item for
3585 * this particular inode.
3587 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3588 struct btrfs_inode
*inode
)
3590 return btrfs_del_orphan_item(trans
, inode
->root
, btrfs_ino(inode
));
3594 * this cleans up any orphans that may be left on the list from the last use
3597 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3599 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3600 struct btrfs_path
*path
;
3601 struct extent_buffer
*leaf
;
3602 struct btrfs_key key
, found_key
;
3603 struct btrfs_trans_handle
*trans
;
3604 struct inode
*inode
;
3605 u64 last_objectid
= 0;
3606 int ret
= 0, nr_unlink
= 0;
3608 if (test_and_set_bit(BTRFS_ROOT_ORPHAN_CLEANUP
, &root
->state
))
3611 path
= btrfs_alloc_path();
3616 path
->reada
= READA_BACK
;
3618 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3619 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3620 key
.offset
= (u64
)-1;
3623 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3628 * if ret == 0 means we found what we were searching for, which
3629 * is weird, but possible, so only screw with path if we didn't
3630 * find the key and see if we have stuff that matches
3634 if (path
->slots
[0] == 0)
3639 /* pull out the item */
3640 leaf
= path
->nodes
[0];
3641 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3643 /* make sure the item matches what we want */
3644 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3646 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3649 /* release the path since we're done with it */
3650 btrfs_release_path(path
);
3653 * this is where we are basically btrfs_lookup, without the
3654 * crossing root thing. we store the inode number in the
3655 * offset of the orphan item.
3658 if (found_key
.offset
== last_objectid
) {
3660 "Error removing orphan entry, stopping orphan cleanup");
3665 last_objectid
= found_key
.offset
;
3667 found_key
.objectid
= found_key
.offset
;
3668 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3669 found_key
.offset
= 0;
3670 inode
= btrfs_iget(fs_info
->sb
, last_objectid
, root
);
3671 ret
= PTR_ERR_OR_ZERO(inode
);
3672 if (ret
&& ret
!= -ENOENT
)
3675 if (ret
== -ENOENT
&& root
== fs_info
->tree_root
) {
3676 struct btrfs_root
*dead_root
;
3677 int is_dead_root
= 0;
3680 * This is an orphan in the tree root. Currently these
3681 * could come from 2 sources:
3682 * a) a root (snapshot/subvolume) deletion in progress
3683 * b) a free space cache inode
3684 * We need to distinguish those two, as the orphan item
3685 * for a root must not get deleted before the deletion
3686 * of the snapshot/subvolume's tree completes.
3688 * btrfs_find_orphan_roots() ran before us, which has
3689 * found all deleted roots and loaded them into
3690 * fs_info->fs_roots_radix. So here we can find if an
3691 * orphan item corresponds to a deleted root by looking
3692 * up the root from that radix tree.
3695 spin_lock(&fs_info
->fs_roots_radix_lock
);
3696 dead_root
= radix_tree_lookup(&fs_info
->fs_roots_radix
,
3697 (unsigned long)found_key
.objectid
);
3698 if (dead_root
&& btrfs_root_refs(&dead_root
->root_item
) == 0)
3700 spin_unlock(&fs_info
->fs_roots_radix_lock
);
3703 /* prevent this orphan from being found again */
3704 key
.offset
= found_key
.objectid
- 1;
3711 * If we have an inode with links, there are a couple of
3714 * 1. We were halfway through creating fsverity metadata for the
3715 * file. In that case, the orphan item represents incomplete
3716 * fsverity metadata which must be cleaned up with
3717 * btrfs_drop_verity_items and deleting the orphan item.
3719 * 2. Old kernels (before v3.12) used to create an
3720 * orphan item for truncate indicating that there were possibly
3721 * extent items past i_size that needed to be deleted. In v3.12,
3722 * truncate was changed to update i_size in sync with the extent
3723 * items, but the (useless) orphan item was still created. Since
3724 * v4.18, we don't create the orphan item for truncate at all.
3726 * So, this item could mean that we need to do a truncate, but
3727 * only if this filesystem was last used on a pre-v3.12 kernel
3728 * and was not cleanly unmounted. The odds of that are quite
3729 * slim, and it's a pain to do the truncate now, so just delete
3732 * It's also possible that this orphan item was supposed to be
3733 * deleted but wasn't. The inode number may have been reused,
3734 * but either way, we can delete the orphan item.
3736 if (ret
== -ENOENT
|| inode
->i_nlink
) {
3738 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
);
3749 btrfs_debug(fs_info
, "auto deleting %Lu",
3750 found_key
.objectid
);
3751 ret
= btrfs_del_orphan_item(trans
, root
,
3752 found_key
.objectid
);
3753 btrfs_end_transaction(trans
);
3763 /* this will do delete_inode and everything for us */
3766 /* release the path since we're done with it */
3767 btrfs_release_path(path
);
3769 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3770 trans
= btrfs_join_transaction(root
);
3772 btrfs_end_transaction(trans
);
3776 btrfs_debug(fs_info
, "unlinked %d orphans", nr_unlink
);
3780 btrfs_err(fs_info
, "could not do orphan cleanup %d", ret
);
3781 btrfs_free_path(path
);
3786 * very simple check to peek ahead in the leaf looking for xattrs. If we
3787 * don't find any xattrs, we know there can't be any acls.
3789 * slot is the slot the inode is in, objectid is the objectid of the inode
3791 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3792 int slot
, u64 objectid
,
3793 int *first_xattr_slot
)
3795 u32 nritems
= btrfs_header_nritems(leaf
);
3796 struct btrfs_key found_key
;
3797 static u64 xattr_access
= 0;
3798 static u64 xattr_default
= 0;
3801 if (!xattr_access
) {
3802 xattr_access
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS
,
3803 strlen(XATTR_NAME_POSIX_ACL_ACCESS
));
3804 xattr_default
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT
,
3805 strlen(XATTR_NAME_POSIX_ACL_DEFAULT
));
3809 *first_xattr_slot
= -1;
3810 while (slot
< nritems
) {
3811 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3813 /* we found a different objectid, there must not be acls */
3814 if (found_key
.objectid
!= objectid
)
3817 /* we found an xattr, assume we've got an acl */
3818 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3819 if (*first_xattr_slot
== -1)
3820 *first_xattr_slot
= slot
;
3821 if (found_key
.offset
== xattr_access
||
3822 found_key
.offset
== xattr_default
)
3827 * we found a key greater than an xattr key, there can't
3828 * be any acls later on
3830 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3837 * it goes inode, inode backrefs, xattrs, extents,
3838 * so if there are a ton of hard links to an inode there can
3839 * be a lot of backrefs. Don't waste time searching too hard,
3840 * this is just an optimization
3845 /* we hit the end of the leaf before we found an xattr or
3846 * something larger than an xattr. We have to assume the inode
3849 if (*first_xattr_slot
== -1)
3850 *first_xattr_slot
= slot
;
3855 * read an inode from the btree into the in-memory inode
3857 static int btrfs_read_locked_inode(struct inode
*inode
,
3858 struct btrfs_path
*in_path
)
3860 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
3861 struct btrfs_path
*path
= in_path
;
3862 struct extent_buffer
*leaf
;
3863 struct btrfs_inode_item
*inode_item
;
3864 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3865 struct btrfs_key location
;
3870 bool filled
= false;
3871 int first_xattr_slot
;
3873 ret
= btrfs_fill_inode(inode
, &rdev
);
3878 path
= btrfs_alloc_path();
3883 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3885 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3887 if (path
!= in_path
)
3888 btrfs_free_path(path
);
3892 leaf
= path
->nodes
[0];
3897 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3898 struct btrfs_inode_item
);
3899 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3900 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3901 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3902 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3903 btrfs_i_size_write(BTRFS_I(inode
), btrfs_inode_size(leaf
, inode_item
));
3904 btrfs_inode_set_file_extent_range(BTRFS_I(inode
), 0,
3905 round_up(i_size_read(inode
), fs_info
->sectorsize
));
3907 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3908 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3910 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3911 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3913 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3914 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3916 BTRFS_I(inode
)->i_otime
.tv_sec
=
3917 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3918 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3919 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3921 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3922 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3923 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3925 inode_set_iversion_queried(inode
,
3926 btrfs_inode_sequence(leaf
, inode_item
));
3927 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3929 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3931 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3932 btrfs_inode_split_flags(btrfs_inode_flags(leaf
, inode_item
),
3933 &BTRFS_I(inode
)->flags
, &BTRFS_I(inode
)->ro_flags
);
3937 * If we were modified in the current generation and evicted from memory
3938 * and then re-read we need to do a full sync since we don't have any
3939 * idea about which extents were modified before we were evicted from
3942 * This is required for both inode re-read from disk and delayed inode
3943 * in delayed_nodes_tree.
3945 if (BTRFS_I(inode
)->last_trans
== fs_info
->generation
)
3946 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3947 &BTRFS_I(inode
)->runtime_flags
);
3950 * We don't persist the id of the transaction where an unlink operation
3951 * against the inode was last made. So here we assume the inode might
3952 * have been evicted, and therefore the exact value of last_unlink_trans
3953 * lost, and set it to last_trans to avoid metadata inconsistencies
3954 * between the inode and its parent if the inode is fsync'ed and the log
3955 * replayed. For example, in the scenario:
3958 * ln mydir/foo mydir/bar
3961 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3962 * xfs_io -c fsync mydir/foo
3964 * mount fs, triggers fsync log replay
3966 * We must make sure that when we fsync our inode foo we also log its
3967 * parent inode, otherwise after log replay the parent still has the
3968 * dentry with the "bar" name but our inode foo has a link count of 1
3969 * and doesn't have an inode ref with the name "bar" anymore.
3971 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3972 * but it guarantees correctness at the expense of occasional full
3973 * transaction commits on fsync if our inode is a directory, or if our
3974 * inode is not a directory, logging its parent unnecessarily.
3976 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3979 * Same logic as for last_unlink_trans. We don't persist the generation
3980 * of the last transaction where this inode was used for a reflink
3981 * operation, so after eviction and reloading the inode we must be
3982 * pessimistic and assume the last transaction that modified the inode.
3984 BTRFS_I(inode
)->last_reflink_trans
= BTRFS_I(inode
)->last_trans
;
3987 if (inode
->i_nlink
!= 1 ||
3988 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3991 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3992 if (location
.objectid
!= btrfs_ino(BTRFS_I(inode
)))
3995 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3996 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3997 struct btrfs_inode_ref
*ref
;
3999 ref
= (struct btrfs_inode_ref
*)ptr
;
4000 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
4001 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
4002 struct btrfs_inode_extref
*extref
;
4004 extref
= (struct btrfs_inode_extref
*)ptr
;
4005 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
4010 * try to precache a NULL acl entry for files that don't have
4011 * any xattrs or acls
4013 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
4014 btrfs_ino(BTRFS_I(inode
)), &first_xattr_slot
);
4015 if (first_xattr_slot
!= -1) {
4016 path
->slots
[0] = first_xattr_slot
;
4017 ret
= btrfs_load_inode_props(inode
, path
);
4020 "error loading props for ino %llu (root %llu): %d",
4021 btrfs_ino(BTRFS_I(inode
)),
4022 root
->root_key
.objectid
, ret
);
4024 if (path
!= in_path
)
4025 btrfs_free_path(path
);
4028 cache_no_acl(inode
);
4030 switch (inode
->i_mode
& S_IFMT
) {
4032 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4033 inode
->i_fop
= &btrfs_file_operations
;
4034 inode
->i_op
= &btrfs_file_inode_operations
;
4037 inode
->i_fop
= &btrfs_dir_file_operations
;
4038 inode
->i_op
= &btrfs_dir_inode_operations
;
4041 inode
->i_op
= &btrfs_symlink_inode_operations
;
4042 inode_nohighmem(inode
);
4043 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4046 inode
->i_op
= &btrfs_special_inode_operations
;
4047 init_special_inode(inode
, inode
->i_mode
, rdev
);
4051 btrfs_sync_inode_flags_to_i_flags(inode
);
4056 * given a leaf and an inode, copy the inode fields into the leaf
4058 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
4059 struct extent_buffer
*leaf
,
4060 struct btrfs_inode_item
*item
,
4061 struct inode
*inode
)
4063 struct btrfs_map_token token
;
4066 btrfs_init_map_token(&token
, leaf
);
4068 btrfs_set_token_inode_uid(&token
, item
, i_uid_read(inode
));
4069 btrfs_set_token_inode_gid(&token
, item
, i_gid_read(inode
));
4070 btrfs_set_token_inode_size(&token
, item
, BTRFS_I(inode
)->disk_i_size
);
4071 btrfs_set_token_inode_mode(&token
, item
, inode
->i_mode
);
4072 btrfs_set_token_inode_nlink(&token
, item
, inode
->i_nlink
);
4074 btrfs_set_token_timespec_sec(&token
, &item
->atime
,
4075 inode
->i_atime
.tv_sec
);
4076 btrfs_set_token_timespec_nsec(&token
, &item
->atime
,
4077 inode
->i_atime
.tv_nsec
);
4079 btrfs_set_token_timespec_sec(&token
, &item
->mtime
,
4080 inode
->i_mtime
.tv_sec
);
4081 btrfs_set_token_timespec_nsec(&token
, &item
->mtime
,
4082 inode
->i_mtime
.tv_nsec
);
4084 btrfs_set_token_timespec_sec(&token
, &item
->ctime
,
4085 inode
->i_ctime
.tv_sec
);
4086 btrfs_set_token_timespec_nsec(&token
, &item
->ctime
,
4087 inode
->i_ctime
.tv_nsec
);
4089 btrfs_set_token_timespec_sec(&token
, &item
->otime
,
4090 BTRFS_I(inode
)->i_otime
.tv_sec
);
4091 btrfs_set_token_timespec_nsec(&token
, &item
->otime
,
4092 BTRFS_I(inode
)->i_otime
.tv_nsec
);
4094 btrfs_set_token_inode_nbytes(&token
, item
, inode_get_bytes(inode
));
4095 btrfs_set_token_inode_generation(&token
, item
,
4096 BTRFS_I(inode
)->generation
);
4097 btrfs_set_token_inode_sequence(&token
, item
, inode_peek_iversion(inode
));
4098 btrfs_set_token_inode_transid(&token
, item
, trans
->transid
);
4099 btrfs_set_token_inode_rdev(&token
, item
, inode
->i_rdev
);
4100 flags
= btrfs_inode_combine_flags(BTRFS_I(inode
)->flags
,
4101 BTRFS_I(inode
)->ro_flags
);
4102 btrfs_set_token_inode_flags(&token
, item
, flags
);
4103 btrfs_set_token_inode_block_group(&token
, item
, 0);
4107 * copy everything in the in-memory inode into the btree.
4109 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
4110 struct btrfs_root
*root
,
4111 struct btrfs_inode
*inode
)
4113 struct btrfs_inode_item
*inode_item
;
4114 struct btrfs_path
*path
;
4115 struct extent_buffer
*leaf
;
4118 path
= btrfs_alloc_path();
4122 ret
= btrfs_lookup_inode(trans
, root
, path
, &inode
->location
, 1);
4129 leaf
= path
->nodes
[0];
4130 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4131 struct btrfs_inode_item
);
4133 fill_inode_item(trans
, leaf
, inode_item
, &inode
->vfs_inode
);
4134 btrfs_mark_buffer_dirty(leaf
);
4135 btrfs_set_inode_last_trans(trans
, inode
);
4138 btrfs_free_path(path
);
4143 * copy everything in the in-memory inode into the btree.
4145 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
4146 struct btrfs_root
*root
,
4147 struct btrfs_inode
*inode
)
4149 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4153 * If the inode is a free space inode, we can deadlock during commit
4154 * if we put it into the delayed code.
4156 * The data relocation inode should also be directly updated
4159 if (!btrfs_is_free_space_inode(inode
)
4160 && !btrfs_is_data_reloc_root(root
)
4161 && !test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
)) {
4162 btrfs_update_root_times(trans
, root
);
4164 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
4166 btrfs_set_inode_last_trans(trans
, inode
);
4170 return btrfs_update_inode_item(trans
, root
, inode
);
4173 int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
4174 struct btrfs_root
*root
, struct btrfs_inode
*inode
)
4178 ret
= btrfs_update_inode(trans
, root
, inode
);
4180 return btrfs_update_inode_item(trans
, root
, inode
);
4185 * unlink helper that gets used here in inode.c and in the tree logging
4186 * recovery code. It remove a link in a directory with a given name, and
4187 * also drops the back refs in the inode to the directory
4189 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4190 struct btrfs_inode
*dir
,
4191 struct btrfs_inode
*inode
,
4192 const struct fscrypt_str
*name
,
4193 struct btrfs_rename_ctx
*rename_ctx
)
4195 struct btrfs_root
*root
= dir
->root
;
4196 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4197 struct btrfs_path
*path
;
4199 struct btrfs_dir_item
*di
;
4201 u64 ino
= btrfs_ino(inode
);
4202 u64 dir_ino
= btrfs_ino(dir
);
4204 path
= btrfs_alloc_path();
4210 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
, name
, -1);
4211 if (IS_ERR_OR_NULL(di
)) {
4212 ret
= di
? PTR_ERR(di
) : -ENOENT
;
4215 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4218 btrfs_release_path(path
);
4221 * If we don't have dir index, we have to get it by looking up
4222 * the inode ref, since we get the inode ref, remove it directly,
4223 * it is unnecessary to do delayed deletion.
4225 * But if we have dir index, needn't search inode ref to get it.
4226 * Since the inode ref is close to the inode item, it is better
4227 * that we delay to delete it, and just do this deletion when
4228 * we update the inode item.
4230 if (inode
->dir_index
) {
4231 ret
= btrfs_delayed_delete_inode_ref(inode
);
4233 index
= inode
->dir_index
;
4238 ret
= btrfs_del_inode_ref(trans
, root
, name
, ino
, dir_ino
, &index
);
4241 "failed to delete reference to %.*s, inode %llu parent %llu",
4242 name
->len
, name
->name
, ino
, dir_ino
);
4243 btrfs_abort_transaction(trans
, ret
);
4248 rename_ctx
->index
= index
;
4250 ret
= btrfs_delete_delayed_dir_index(trans
, dir
, index
);
4252 btrfs_abort_transaction(trans
, ret
);
4257 * If we are in a rename context, we don't need to update anything in the
4258 * log. That will be done later during the rename by btrfs_log_new_name().
4259 * Besides that, doing it here would only cause extra unnecessary btree
4260 * operations on the log tree, increasing latency for applications.
4263 btrfs_del_inode_ref_in_log(trans
, root
, name
, inode
, dir_ino
);
4264 btrfs_del_dir_entries_in_log(trans
, root
, name
, dir
, index
);
4268 * If we have a pending delayed iput we could end up with the final iput
4269 * being run in btrfs-cleaner context. If we have enough of these built
4270 * up we can end up burning a lot of time in btrfs-cleaner without any
4271 * way to throttle the unlinks. Since we're currently holding a ref on
4272 * the inode we can run the delayed iput here without any issues as the
4273 * final iput won't be done until after we drop the ref we're currently
4276 btrfs_run_delayed_iput(fs_info
, inode
);
4278 btrfs_free_path(path
);
4282 btrfs_i_size_write(dir
, dir
->vfs_inode
.i_size
- name
->len
* 2);
4283 inode_inc_iversion(&inode
->vfs_inode
);
4284 inode_inc_iversion(&dir
->vfs_inode
);
4285 inode
->vfs_inode
.i_ctime
= current_time(&inode
->vfs_inode
);
4286 dir
->vfs_inode
.i_mtime
= inode
->vfs_inode
.i_ctime
;
4287 dir
->vfs_inode
.i_ctime
= inode
->vfs_inode
.i_ctime
;
4288 ret
= btrfs_update_inode(trans
, root
, dir
);
4293 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4294 struct btrfs_inode
*dir
, struct btrfs_inode
*inode
,
4295 const struct fscrypt_str
*name
)
4299 ret
= __btrfs_unlink_inode(trans
, dir
, inode
, name
, NULL
);
4301 drop_nlink(&inode
->vfs_inode
);
4302 ret
= btrfs_update_inode(trans
, inode
->root
, inode
);
4308 * helper to start transaction for unlink and rmdir.
4310 * unlink and rmdir are special in btrfs, they do not always free space, so
4311 * if we cannot make our reservations the normal way try and see if there is
4312 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4313 * allow the unlink to occur.
4315 static struct btrfs_trans_handle
*__unlink_start_trans(struct btrfs_inode
*dir
)
4317 struct btrfs_root
*root
= dir
->root
;
4319 return btrfs_start_transaction_fallback_global_rsv(root
,
4320 BTRFS_UNLINK_METADATA_UNITS
);
4323 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4325 struct btrfs_trans_handle
*trans
;
4326 struct inode
*inode
= d_inode(dentry
);
4328 struct fscrypt_name fname
;
4330 ret
= fscrypt_setup_filename(dir
, &dentry
->d_name
, 1, &fname
);
4334 /* This needs to handle no-key deletions later on */
4336 trans
= __unlink_start_trans(BTRFS_I(dir
));
4337 if (IS_ERR(trans
)) {
4338 ret
= PTR_ERR(trans
);
4342 btrfs_record_unlink_dir(trans
, BTRFS_I(dir
), BTRFS_I(d_inode(dentry
)),
4345 ret
= btrfs_unlink_inode(trans
, BTRFS_I(dir
), BTRFS_I(d_inode(dentry
)),
4350 if (inode
->i_nlink
== 0) {
4351 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4357 btrfs_end_transaction(trans
);
4358 btrfs_btree_balance_dirty(BTRFS_I(dir
)->root
->fs_info
);
4360 fscrypt_free_filename(&fname
);
4364 static int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4365 struct btrfs_inode
*dir
, struct dentry
*dentry
)
4367 struct btrfs_root
*root
= dir
->root
;
4368 struct btrfs_inode
*inode
= BTRFS_I(d_inode(dentry
));
4369 struct btrfs_path
*path
;
4370 struct extent_buffer
*leaf
;
4371 struct btrfs_dir_item
*di
;
4372 struct btrfs_key key
;
4376 u64 dir_ino
= btrfs_ino(dir
);
4377 struct fscrypt_name fname
;
4379 ret
= fscrypt_setup_filename(&dir
->vfs_inode
, &dentry
->d_name
, 1, &fname
);
4383 /* This needs to handle no-key deletions later on */
4385 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
) {
4386 objectid
= inode
->root
->root_key
.objectid
;
4387 } else if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
) {
4388 objectid
= inode
->location
.objectid
;
4391 fscrypt_free_filename(&fname
);
4395 path
= btrfs_alloc_path();
4401 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4402 &fname
.disk_name
, -1);
4403 if (IS_ERR_OR_NULL(di
)) {
4404 ret
= di
? PTR_ERR(di
) : -ENOENT
;
4408 leaf
= path
->nodes
[0];
4409 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4410 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4411 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4413 btrfs_abort_transaction(trans
, ret
);
4416 btrfs_release_path(path
);
4419 * This is a placeholder inode for a subvolume we didn't have a
4420 * reference to at the time of the snapshot creation. In the meantime
4421 * we could have renamed the real subvol link into our snapshot, so
4422 * depending on btrfs_del_root_ref to return -ENOENT here is incorrect.
4423 * Instead simply lookup the dir_index_item for this entry so we can
4424 * remove it. Otherwise we know we have a ref to the root and we can
4425 * call btrfs_del_root_ref, and it _shouldn't_ fail.
4427 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
) {
4428 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
, &fname
.disk_name
);
4429 if (IS_ERR_OR_NULL(di
)) {
4434 btrfs_abort_transaction(trans
, ret
);
4438 leaf
= path
->nodes
[0];
4439 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4441 btrfs_release_path(path
);
4443 ret
= btrfs_del_root_ref(trans
, objectid
,
4444 root
->root_key
.objectid
, dir_ino
,
4445 &index
, &fname
.disk_name
);
4447 btrfs_abort_transaction(trans
, ret
);
4452 ret
= btrfs_delete_delayed_dir_index(trans
, dir
, index
);
4454 btrfs_abort_transaction(trans
, ret
);
4458 btrfs_i_size_write(dir
, dir
->vfs_inode
.i_size
- fname
.disk_name
.len
* 2);
4459 inode_inc_iversion(&dir
->vfs_inode
);
4460 dir
->vfs_inode
.i_mtime
= current_time(&dir
->vfs_inode
);
4461 dir
->vfs_inode
.i_ctime
= dir
->vfs_inode
.i_mtime
;
4462 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4464 btrfs_abort_transaction(trans
, ret
);
4466 btrfs_free_path(path
);
4467 fscrypt_free_filename(&fname
);
4472 * Helper to check if the subvolume references other subvolumes or if it's
4475 static noinline
int may_destroy_subvol(struct btrfs_root
*root
)
4477 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4478 struct btrfs_path
*path
;
4479 struct btrfs_dir_item
*di
;
4480 struct btrfs_key key
;
4481 struct fscrypt_str name
= FSTR_INIT("default", 7);
4485 path
= btrfs_alloc_path();
4489 /* Make sure this root isn't set as the default subvol */
4490 dir_id
= btrfs_super_root_dir(fs_info
->super_copy
);
4491 di
= btrfs_lookup_dir_item(NULL
, fs_info
->tree_root
, path
,
4493 if (di
&& !IS_ERR(di
)) {
4494 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, &key
);
4495 if (key
.objectid
== root
->root_key
.objectid
) {
4498 "deleting default subvolume %llu is not allowed",
4502 btrfs_release_path(path
);
4505 key
.objectid
= root
->root_key
.objectid
;
4506 key
.type
= BTRFS_ROOT_REF_KEY
;
4507 key
.offset
= (u64
)-1;
4509 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
4515 if (path
->slots
[0] > 0) {
4517 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, path
->slots
[0]);
4518 if (key
.objectid
== root
->root_key
.objectid
&&
4519 key
.type
== BTRFS_ROOT_REF_KEY
)
4523 btrfs_free_path(path
);
4527 /* Delete all dentries for inodes belonging to the root */
4528 static void btrfs_prune_dentries(struct btrfs_root
*root
)
4530 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4531 struct rb_node
*node
;
4532 struct rb_node
*prev
;
4533 struct btrfs_inode
*entry
;
4534 struct inode
*inode
;
4537 if (!BTRFS_FS_ERROR(fs_info
))
4538 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4540 spin_lock(&root
->inode_lock
);
4542 node
= root
->inode_tree
.rb_node
;
4546 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4548 if (objectid
< btrfs_ino(entry
))
4549 node
= node
->rb_left
;
4550 else if (objectid
> btrfs_ino(entry
))
4551 node
= node
->rb_right
;
4557 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4558 if (objectid
<= btrfs_ino(entry
)) {
4562 prev
= rb_next(prev
);
4566 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4567 objectid
= btrfs_ino(entry
) + 1;
4568 inode
= igrab(&entry
->vfs_inode
);
4570 spin_unlock(&root
->inode_lock
);
4571 if (atomic_read(&inode
->i_count
) > 1)
4572 d_prune_aliases(inode
);
4574 * btrfs_drop_inode will have it removed from the inode
4575 * cache when its usage count hits zero.
4579 spin_lock(&root
->inode_lock
);
4583 if (cond_resched_lock(&root
->inode_lock
))
4586 node
= rb_next(node
);
4588 spin_unlock(&root
->inode_lock
);
4591 int btrfs_delete_subvolume(struct btrfs_inode
*dir
, struct dentry
*dentry
)
4593 struct btrfs_fs_info
*fs_info
= btrfs_sb(dentry
->d_sb
);
4594 struct btrfs_root
*root
= dir
->root
;
4595 struct inode
*inode
= d_inode(dentry
);
4596 struct btrfs_root
*dest
= BTRFS_I(inode
)->root
;
4597 struct btrfs_trans_handle
*trans
;
4598 struct btrfs_block_rsv block_rsv
;
4603 * Don't allow to delete a subvolume with send in progress. This is
4604 * inside the inode lock so the error handling that has to drop the bit
4605 * again is not run concurrently.
4607 spin_lock(&dest
->root_item_lock
);
4608 if (dest
->send_in_progress
) {
4609 spin_unlock(&dest
->root_item_lock
);
4611 "attempt to delete subvolume %llu during send",
4612 dest
->root_key
.objectid
);
4615 if (atomic_read(&dest
->nr_swapfiles
)) {
4616 spin_unlock(&dest
->root_item_lock
);
4618 "attempt to delete subvolume %llu with active swapfile",
4619 root
->root_key
.objectid
);
4622 root_flags
= btrfs_root_flags(&dest
->root_item
);
4623 btrfs_set_root_flags(&dest
->root_item
,
4624 root_flags
| BTRFS_ROOT_SUBVOL_DEAD
);
4625 spin_unlock(&dest
->root_item_lock
);
4627 down_write(&fs_info
->subvol_sem
);
4629 ret
= may_destroy_subvol(dest
);
4633 btrfs_init_block_rsv(&block_rsv
, BTRFS_BLOCK_RSV_TEMP
);
4635 * One for dir inode,
4636 * two for dir entries,
4637 * two for root ref/backref.
4639 ret
= btrfs_subvolume_reserve_metadata(root
, &block_rsv
, 5, true);
4643 trans
= btrfs_start_transaction(root
, 0);
4644 if (IS_ERR(trans
)) {
4645 ret
= PTR_ERR(trans
);
4648 trans
->block_rsv
= &block_rsv
;
4649 trans
->bytes_reserved
= block_rsv
.size
;
4651 btrfs_record_snapshot_destroy(trans
, dir
);
4653 ret
= btrfs_unlink_subvol(trans
, dir
, dentry
);
4655 btrfs_abort_transaction(trans
, ret
);
4659 ret
= btrfs_record_root_in_trans(trans
, dest
);
4661 btrfs_abort_transaction(trans
, ret
);
4665 memset(&dest
->root_item
.drop_progress
, 0,
4666 sizeof(dest
->root_item
.drop_progress
));
4667 btrfs_set_root_drop_level(&dest
->root_item
, 0);
4668 btrfs_set_root_refs(&dest
->root_item
, 0);
4670 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &dest
->state
)) {
4671 ret
= btrfs_insert_orphan_item(trans
,
4673 dest
->root_key
.objectid
);
4675 btrfs_abort_transaction(trans
, ret
);
4680 ret
= btrfs_uuid_tree_remove(trans
, dest
->root_item
.uuid
,
4681 BTRFS_UUID_KEY_SUBVOL
,
4682 dest
->root_key
.objectid
);
4683 if (ret
&& ret
!= -ENOENT
) {
4684 btrfs_abort_transaction(trans
, ret
);
4687 if (!btrfs_is_empty_uuid(dest
->root_item
.received_uuid
)) {
4688 ret
= btrfs_uuid_tree_remove(trans
,
4689 dest
->root_item
.received_uuid
,
4690 BTRFS_UUID_KEY_RECEIVED_SUBVOL
,
4691 dest
->root_key
.objectid
);
4692 if (ret
&& ret
!= -ENOENT
) {
4693 btrfs_abort_transaction(trans
, ret
);
4698 free_anon_bdev(dest
->anon_dev
);
4701 trans
->block_rsv
= NULL
;
4702 trans
->bytes_reserved
= 0;
4703 ret
= btrfs_end_transaction(trans
);
4704 inode
->i_flags
|= S_DEAD
;
4706 btrfs_subvolume_release_metadata(root
, &block_rsv
);
4708 up_write(&fs_info
->subvol_sem
);
4710 spin_lock(&dest
->root_item_lock
);
4711 root_flags
= btrfs_root_flags(&dest
->root_item
);
4712 btrfs_set_root_flags(&dest
->root_item
,
4713 root_flags
& ~BTRFS_ROOT_SUBVOL_DEAD
);
4714 spin_unlock(&dest
->root_item_lock
);
4716 d_invalidate(dentry
);
4717 btrfs_prune_dentries(dest
);
4718 ASSERT(dest
->send_in_progress
== 0);
4724 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4726 struct inode
*inode
= d_inode(dentry
);
4727 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
4729 struct btrfs_trans_handle
*trans
;
4730 u64 last_unlink_trans
;
4731 struct fscrypt_name fname
;
4733 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4735 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_FIRST_FREE_OBJECTID
) {
4736 if (unlikely(btrfs_fs_incompat(fs_info
, EXTENT_TREE_V2
))) {
4738 "extent tree v2 doesn't support snapshot deletion yet");
4741 return btrfs_delete_subvolume(BTRFS_I(dir
), dentry
);
4744 err
= fscrypt_setup_filename(dir
, &dentry
->d_name
, 1, &fname
);
4748 /* This needs to handle no-key deletions later on */
4750 trans
= __unlink_start_trans(BTRFS_I(dir
));
4751 if (IS_ERR(trans
)) {
4752 err
= PTR_ERR(trans
);
4756 if (unlikely(btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4757 err
= btrfs_unlink_subvol(trans
, BTRFS_I(dir
), dentry
);
4761 err
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
4765 last_unlink_trans
= BTRFS_I(inode
)->last_unlink_trans
;
4767 /* now the directory is empty */
4768 err
= btrfs_unlink_inode(trans
, BTRFS_I(dir
), BTRFS_I(d_inode(dentry
)),
4771 btrfs_i_size_write(BTRFS_I(inode
), 0);
4773 * Propagate the last_unlink_trans value of the deleted dir to
4774 * its parent directory. This is to prevent an unrecoverable
4775 * log tree in the case we do something like this:
4777 * 2) create snapshot under dir foo
4778 * 3) delete the snapshot
4781 * 6) fsync foo or some file inside foo
4783 if (last_unlink_trans
>= trans
->transid
)
4784 BTRFS_I(dir
)->last_unlink_trans
= last_unlink_trans
;
4787 btrfs_end_transaction(trans
);
4789 btrfs_btree_balance_dirty(fs_info
);
4790 fscrypt_free_filename(&fname
);
4796 * btrfs_truncate_block - read, zero a chunk and write a block
4797 * @inode - inode that we're zeroing
4798 * @from - the offset to start zeroing
4799 * @len - the length to zero, 0 to zero the entire range respective to the
4801 * @front - zero up to the offset instead of from the offset on
4803 * This will find the block for the "from" offset and cow the block and zero the
4804 * part we want to zero. This is used with truncate and hole punching.
4806 int btrfs_truncate_block(struct btrfs_inode
*inode
, loff_t from
, loff_t len
,
4809 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
4810 struct address_space
*mapping
= inode
->vfs_inode
.i_mapping
;
4811 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
4812 struct btrfs_ordered_extent
*ordered
;
4813 struct extent_state
*cached_state
= NULL
;
4814 struct extent_changeset
*data_reserved
= NULL
;
4815 bool only_release_metadata
= false;
4816 u32 blocksize
= fs_info
->sectorsize
;
4817 pgoff_t index
= from
>> PAGE_SHIFT
;
4818 unsigned offset
= from
& (blocksize
- 1);
4820 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4821 size_t write_bytes
= blocksize
;
4826 if (IS_ALIGNED(offset
, blocksize
) &&
4827 (!len
|| IS_ALIGNED(len
, blocksize
)))
4830 block_start
= round_down(from
, blocksize
);
4831 block_end
= block_start
+ blocksize
- 1;
4833 ret
= btrfs_check_data_free_space(inode
, &data_reserved
, block_start
,
4836 if (btrfs_check_nocow_lock(inode
, block_start
, &write_bytes
, false) > 0) {
4837 /* For nocow case, no need to reserve data space */
4838 only_release_metadata
= true;
4843 ret
= btrfs_delalloc_reserve_metadata(inode
, blocksize
, blocksize
, false);
4845 if (!only_release_metadata
)
4846 btrfs_free_reserved_data_space(inode
, data_reserved
,
4847 block_start
, blocksize
);
4851 page
= find_or_create_page(mapping
, index
, mask
);
4853 btrfs_delalloc_release_space(inode
, data_reserved
, block_start
,
4855 btrfs_delalloc_release_extents(inode
, blocksize
);
4859 ret
= set_page_extent_mapped(page
);
4863 if (!PageUptodate(page
)) {
4864 ret
= btrfs_read_folio(NULL
, page_folio(page
));
4866 if (page
->mapping
!= mapping
) {
4871 if (!PageUptodate(page
)) {
4876 wait_on_page_writeback(page
);
4878 lock_extent(io_tree
, block_start
, block_end
, &cached_state
);
4880 ordered
= btrfs_lookup_ordered_extent(inode
, block_start
);
4882 unlock_extent(io_tree
, block_start
, block_end
, &cached_state
);
4885 btrfs_start_ordered_extent(ordered
);
4886 btrfs_put_ordered_extent(ordered
);
4890 clear_extent_bit(&inode
->io_tree
, block_start
, block_end
,
4891 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4894 ret
= btrfs_set_extent_delalloc(inode
, block_start
, block_end
, 0,
4897 unlock_extent(io_tree
, block_start
, block_end
, &cached_state
);
4901 if (offset
!= blocksize
) {
4903 len
= blocksize
- offset
;
4905 memzero_page(page
, (block_start
- page_offset(page
)),
4908 memzero_page(page
, (block_start
- page_offset(page
)) + offset
,
4911 btrfs_page_clear_checked(fs_info
, page
, block_start
,
4912 block_end
+ 1 - block_start
);
4913 btrfs_page_set_dirty(fs_info
, page
, block_start
, block_end
+ 1 - block_start
);
4914 unlock_extent(io_tree
, block_start
, block_end
, &cached_state
);
4916 if (only_release_metadata
)
4917 set_extent_bit(&inode
->io_tree
, block_start
, block_end
,
4918 EXTENT_NORESERVE
, NULL
);
4922 if (only_release_metadata
)
4923 btrfs_delalloc_release_metadata(inode
, blocksize
, true);
4925 btrfs_delalloc_release_space(inode
, data_reserved
,
4926 block_start
, blocksize
, true);
4928 btrfs_delalloc_release_extents(inode
, blocksize
);
4932 if (only_release_metadata
)
4933 btrfs_check_nocow_unlock(inode
);
4934 extent_changeset_free(data_reserved
);
4938 static int maybe_insert_hole(struct btrfs_root
*root
, struct btrfs_inode
*inode
,
4939 u64 offset
, u64 len
)
4941 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4942 struct btrfs_trans_handle
*trans
;
4943 struct btrfs_drop_extents_args drop_args
= { 0 };
4947 * If NO_HOLES is enabled, we don't need to do anything.
4948 * Later, up in the call chain, either btrfs_set_inode_last_sub_trans()
4949 * or btrfs_update_inode() will be called, which guarantee that the next
4950 * fsync will know this inode was changed and needs to be logged.
4952 if (btrfs_fs_incompat(fs_info
, NO_HOLES
))
4956 * 1 - for the one we're dropping
4957 * 1 - for the one we're adding
4958 * 1 - for updating the inode.
4960 trans
= btrfs_start_transaction(root
, 3);
4962 return PTR_ERR(trans
);
4964 drop_args
.start
= offset
;
4965 drop_args
.end
= offset
+ len
;
4966 drop_args
.drop_cache
= true;
4968 ret
= btrfs_drop_extents(trans
, root
, inode
, &drop_args
);
4970 btrfs_abort_transaction(trans
, ret
);
4971 btrfs_end_transaction(trans
);
4975 ret
= btrfs_insert_hole_extent(trans
, root
, btrfs_ino(inode
), offset
, len
);
4977 btrfs_abort_transaction(trans
, ret
);
4979 btrfs_update_inode_bytes(inode
, 0, drop_args
.bytes_found
);
4980 btrfs_update_inode(trans
, root
, inode
);
4982 btrfs_end_transaction(trans
);
4987 * This function puts in dummy file extents for the area we're creating a hole
4988 * for. So if we are truncating this file to a larger size we need to insert
4989 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4990 * the range between oldsize and size
4992 int btrfs_cont_expand(struct btrfs_inode
*inode
, loff_t oldsize
, loff_t size
)
4994 struct btrfs_root
*root
= inode
->root
;
4995 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4996 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
4997 struct extent_map
*em
= NULL
;
4998 struct extent_state
*cached_state
= NULL
;
4999 u64 hole_start
= ALIGN(oldsize
, fs_info
->sectorsize
);
5000 u64 block_end
= ALIGN(size
, fs_info
->sectorsize
);
5007 * If our size started in the middle of a block we need to zero out the
5008 * rest of the block before we expand the i_size, otherwise we could
5009 * expose stale data.
5011 err
= btrfs_truncate_block(inode
, oldsize
, 0, 0);
5015 if (size
<= hole_start
)
5018 btrfs_lock_and_flush_ordered_range(inode
, hole_start
, block_end
- 1,
5020 cur_offset
= hole_start
;
5022 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
5023 block_end
- cur_offset
);
5029 last_byte
= min(extent_map_end(em
), block_end
);
5030 last_byte
= ALIGN(last_byte
, fs_info
->sectorsize
);
5031 hole_size
= last_byte
- cur_offset
;
5033 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
5034 struct extent_map
*hole_em
;
5036 err
= maybe_insert_hole(root
, inode
, cur_offset
,
5041 err
= btrfs_inode_set_file_extent_range(inode
,
5042 cur_offset
, hole_size
);
5046 hole_em
= alloc_extent_map();
5048 btrfs_drop_extent_map_range(inode
, cur_offset
,
5049 cur_offset
+ hole_size
- 1,
5051 btrfs_set_inode_full_sync(inode
);
5054 hole_em
->start
= cur_offset
;
5055 hole_em
->len
= hole_size
;
5056 hole_em
->orig_start
= cur_offset
;
5058 hole_em
->block_start
= EXTENT_MAP_HOLE
;
5059 hole_em
->block_len
= 0;
5060 hole_em
->orig_block_len
= 0;
5061 hole_em
->ram_bytes
= hole_size
;
5062 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
5063 hole_em
->generation
= fs_info
->generation
;
5065 err
= btrfs_replace_extent_map_range(inode
, hole_em
, true);
5066 free_extent_map(hole_em
);
5068 err
= btrfs_inode_set_file_extent_range(inode
,
5069 cur_offset
, hole_size
);
5074 free_extent_map(em
);
5076 cur_offset
= last_byte
;
5077 if (cur_offset
>= block_end
)
5080 free_extent_map(em
);
5081 unlock_extent(io_tree
, hole_start
, block_end
- 1, &cached_state
);
5085 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
5087 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5088 struct btrfs_trans_handle
*trans
;
5089 loff_t oldsize
= i_size_read(inode
);
5090 loff_t newsize
= attr
->ia_size
;
5091 int mask
= attr
->ia_valid
;
5095 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5096 * special case where we need to update the times despite not having
5097 * these flags set. For all other operations the VFS set these flags
5098 * explicitly if it wants a timestamp update.
5100 if (newsize
!= oldsize
) {
5101 inode_inc_iversion(inode
);
5102 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
))) {
5103 inode
->i_mtime
= current_time(inode
);
5104 inode
->i_ctime
= inode
->i_mtime
;
5108 if (newsize
> oldsize
) {
5110 * Don't do an expanding truncate while snapshotting is ongoing.
5111 * This is to ensure the snapshot captures a fully consistent
5112 * state of this file - if the snapshot captures this expanding
5113 * truncation, it must capture all writes that happened before
5116 btrfs_drew_write_lock(&root
->snapshot_lock
);
5117 ret
= btrfs_cont_expand(BTRFS_I(inode
), oldsize
, newsize
);
5119 btrfs_drew_write_unlock(&root
->snapshot_lock
);
5123 trans
= btrfs_start_transaction(root
, 1);
5124 if (IS_ERR(trans
)) {
5125 btrfs_drew_write_unlock(&root
->snapshot_lock
);
5126 return PTR_ERR(trans
);
5129 i_size_write(inode
, newsize
);
5130 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode
), 0);
5131 pagecache_isize_extended(inode
, oldsize
, newsize
);
5132 ret
= btrfs_update_inode(trans
, root
, BTRFS_I(inode
));
5133 btrfs_drew_write_unlock(&root
->snapshot_lock
);
5134 btrfs_end_transaction(trans
);
5136 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5138 if (btrfs_is_zoned(fs_info
)) {
5139 ret
= btrfs_wait_ordered_range(inode
,
5140 ALIGN(newsize
, fs_info
->sectorsize
),
5147 * We're truncating a file that used to have good data down to
5148 * zero. Make sure any new writes to the file get on disk
5152 set_bit(BTRFS_INODE_FLUSH_ON_CLOSE
,
5153 &BTRFS_I(inode
)->runtime_flags
);
5155 truncate_setsize(inode
, newsize
);
5157 inode_dio_wait(inode
);
5159 ret
= btrfs_truncate(BTRFS_I(inode
), newsize
== oldsize
);
5160 if (ret
&& inode
->i_nlink
) {
5164 * Truncate failed, so fix up the in-memory size. We
5165 * adjusted disk_i_size down as we removed extents, so
5166 * wait for disk_i_size to be stable and then update the
5167 * in-memory size to match.
5169 err
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5172 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
5179 static int btrfs_setattr(struct mnt_idmap
*idmap
, struct dentry
*dentry
,
5182 struct inode
*inode
= d_inode(dentry
);
5183 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5186 if (btrfs_root_readonly(root
))
5189 err
= setattr_prepare(idmap
, dentry
, attr
);
5193 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5194 err
= btrfs_setsize(inode
, attr
);
5199 if (attr
->ia_valid
) {
5200 setattr_copy(idmap
, inode
, attr
);
5201 inode_inc_iversion(inode
);
5202 err
= btrfs_dirty_inode(BTRFS_I(inode
));
5204 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5205 err
= posix_acl_chmod(idmap
, dentry
, inode
->i_mode
);
5212 * While truncating the inode pages during eviction, we get the VFS
5213 * calling btrfs_invalidate_folio() against each folio of the inode. This
5214 * is slow because the calls to btrfs_invalidate_folio() result in a
5215 * huge amount of calls to lock_extent() and clear_extent_bit(),
5216 * which keep merging and splitting extent_state structures over and over,
5217 * wasting lots of time.
5219 * Therefore if the inode is being evicted, let btrfs_invalidate_folio()
5220 * skip all those expensive operations on a per folio basis and do only
5221 * the ordered io finishing, while we release here the extent_map and
5222 * extent_state structures, without the excessive merging and splitting.
5224 static void evict_inode_truncate_pages(struct inode
*inode
)
5226 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5227 struct rb_node
*node
;
5229 ASSERT(inode
->i_state
& I_FREEING
);
5230 truncate_inode_pages_final(&inode
->i_data
);
5232 btrfs_drop_extent_map_range(BTRFS_I(inode
), 0, (u64
)-1, false);
5235 * Keep looping until we have no more ranges in the io tree.
5236 * We can have ongoing bios started by readahead that have
5237 * their endio callback (extent_io.c:end_bio_extent_readpage)
5238 * still in progress (unlocked the pages in the bio but did not yet
5239 * unlocked the ranges in the io tree). Therefore this means some
5240 * ranges can still be locked and eviction started because before
5241 * submitting those bios, which are executed by a separate task (work
5242 * queue kthread), inode references (inode->i_count) were not taken
5243 * (which would be dropped in the end io callback of each bio).
5244 * Therefore here we effectively end up waiting for those bios and
5245 * anyone else holding locked ranges without having bumped the inode's
5246 * reference count - if we don't do it, when they access the inode's
5247 * io_tree to unlock a range it may be too late, leading to an
5248 * use-after-free issue.
5250 spin_lock(&io_tree
->lock
);
5251 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5252 struct extent_state
*state
;
5253 struct extent_state
*cached_state
= NULL
;
5256 unsigned state_flags
;
5258 node
= rb_first(&io_tree
->state
);
5259 state
= rb_entry(node
, struct extent_state
, rb_node
);
5260 start
= state
->start
;
5262 state_flags
= state
->state
;
5263 spin_unlock(&io_tree
->lock
);
5265 lock_extent(io_tree
, start
, end
, &cached_state
);
5268 * If still has DELALLOC flag, the extent didn't reach disk,
5269 * and its reserved space won't be freed by delayed_ref.
5270 * So we need to free its reserved space here.
5271 * (Refer to comment in btrfs_invalidate_folio, case 2)
5273 * Note, end is the bytenr of last byte, so we need + 1 here.
5275 if (state_flags
& EXTENT_DELALLOC
)
5276 btrfs_qgroup_free_data(BTRFS_I(inode
), NULL
, start
,
5279 clear_extent_bit(io_tree
, start
, end
,
5280 EXTENT_CLEAR_ALL_BITS
| EXTENT_DO_ACCOUNTING
,
5284 spin_lock(&io_tree
->lock
);
5286 spin_unlock(&io_tree
->lock
);
5289 static struct btrfs_trans_handle
*evict_refill_and_join(struct btrfs_root
*root
,
5290 struct btrfs_block_rsv
*rsv
)
5292 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5293 struct btrfs_trans_handle
*trans
;
5294 u64 delayed_refs_extra
= btrfs_calc_delayed_ref_bytes(fs_info
, 1);
5298 * Eviction should be taking place at some place safe because of our
5299 * delayed iputs. However the normal flushing code will run delayed
5300 * iputs, so we cannot use FLUSH_ALL otherwise we'll deadlock.
5302 * We reserve the delayed_refs_extra here again because we can't use
5303 * btrfs_start_transaction(root, 0) for the same deadlocky reason as
5304 * above. We reserve our extra bit here because we generate a ton of
5305 * delayed refs activity by truncating.
5307 * BTRFS_RESERVE_FLUSH_EVICT will steal from the global_rsv if it can,
5308 * if we fail to make this reservation we can re-try without the
5309 * delayed_refs_extra so we can make some forward progress.
5311 ret
= btrfs_block_rsv_refill(fs_info
, rsv
, rsv
->size
+ delayed_refs_extra
,
5312 BTRFS_RESERVE_FLUSH_EVICT
);
5314 ret
= btrfs_block_rsv_refill(fs_info
, rsv
, rsv
->size
,
5315 BTRFS_RESERVE_FLUSH_EVICT
);
5318 "could not allocate space for delete; will truncate on mount");
5319 return ERR_PTR(-ENOSPC
);
5321 delayed_refs_extra
= 0;
5324 trans
= btrfs_join_transaction(root
);
5328 if (delayed_refs_extra
) {
5329 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5330 trans
->bytes_reserved
= delayed_refs_extra
;
5331 btrfs_block_rsv_migrate(rsv
, trans
->block_rsv
,
5332 delayed_refs_extra
, true);
5337 void btrfs_evict_inode(struct inode
*inode
)
5339 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
5340 struct btrfs_trans_handle
*trans
;
5341 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5342 struct btrfs_block_rsv
*rsv
= NULL
;
5345 trace_btrfs_inode_evict(inode
);
5348 fsverity_cleanup_inode(inode
);
5353 evict_inode_truncate_pages(inode
);
5355 if (inode
->i_nlink
&&
5356 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5357 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5358 btrfs_is_free_space_inode(BTRFS_I(inode
))))
5361 if (is_bad_inode(inode
))
5364 if (test_bit(BTRFS_FS_LOG_RECOVERING
, &fs_info
->flags
))
5367 if (inode
->i_nlink
> 0) {
5368 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5369 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5374 * This makes sure the inode item in tree is uptodate and the space for
5375 * the inode update is released.
5377 ret
= btrfs_commit_inode_delayed_inode(BTRFS_I(inode
));
5382 * This drops any pending insert or delete operations we have for this
5383 * inode. We could have a delayed dir index deletion queued up, but
5384 * we're removing the inode completely so that'll be taken care of in
5387 btrfs_kill_delayed_inode_items(BTRFS_I(inode
));
5389 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
5392 rsv
->size
= btrfs_calc_metadata_size(fs_info
, 1);
5393 rsv
->failfast
= true;
5395 btrfs_i_size_write(BTRFS_I(inode
), 0);
5398 struct btrfs_truncate_control control
= {
5399 .inode
= BTRFS_I(inode
),
5400 .ino
= btrfs_ino(BTRFS_I(inode
)),
5405 trans
= evict_refill_and_join(root
, rsv
);
5409 trans
->block_rsv
= rsv
;
5411 ret
= btrfs_truncate_inode_items(trans
, root
, &control
);
5412 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5413 btrfs_end_transaction(trans
);
5415 * We have not added new delayed items for our inode after we
5416 * have flushed its delayed items, so no need to throttle on
5417 * delayed items. However we have modified extent buffers.
5419 btrfs_btree_balance_dirty_nodelay(fs_info
);
5420 if (ret
&& ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5427 * Errors here aren't a big deal, it just means we leave orphan items in
5428 * the tree. They will be cleaned up on the next mount. If the inode
5429 * number gets reused, cleanup deletes the orphan item without doing
5430 * anything, and unlink reuses the existing orphan item.
5432 * If it turns out that we are dropping too many of these, we might want
5433 * to add a mechanism for retrying these after a commit.
5435 trans
= evict_refill_and_join(root
, rsv
);
5436 if (!IS_ERR(trans
)) {
5437 trans
->block_rsv
= rsv
;
5438 btrfs_orphan_del(trans
, BTRFS_I(inode
));
5439 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
5440 btrfs_end_transaction(trans
);
5444 btrfs_free_block_rsv(fs_info
, rsv
);
5446 * If we didn't successfully delete, the orphan item will still be in
5447 * the tree and we'll retry on the next mount. Again, we might also want
5448 * to retry these periodically in the future.
5450 btrfs_remove_delayed_node(BTRFS_I(inode
));
5451 fsverity_cleanup_inode(inode
);
5456 * Return the key found in the dir entry in the location pointer, fill @type
5457 * with BTRFS_FT_*, and return 0.
5459 * If no dir entries were found, returns -ENOENT.
5460 * If found a corrupted location in dir entry, returns -EUCLEAN.
5462 static int btrfs_inode_by_name(struct btrfs_inode
*dir
, struct dentry
*dentry
,
5463 struct btrfs_key
*location
, u8
*type
)
5465 struct btrfs_dir_item
*di
;
5466 struct btrfs_path
*path
;
5467 struct btrfs_root
*root
= dir
->root
;
5469 struct fscrypt_name fname
;
5471 path
= btrfs_alloc_path();
5475 ret
= fscrypt_setup_filename(&dir
->vfs_inode
, &dentry
->d_name
, 1, &fname
);
5479 * fscrypt_setup_filename() should never return a positive value, but
5480 * gcc on sparc/parisc thinks it can, so assert that doesn't happen.
5484 /* This needs to handle no-key deletions later on */
5486 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
),
5487 &fname
.disk_name
, 0);
5488 if (IS_ERR_OR_NULL(di
)) {
5489 ret
= di
? PTR_ERR(di
) : -ENOENT
;
5493 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5494 if (location
->type
!= BTRFS_INODE_ITEM_KEY
&&
5495 location
->type
!= BTRFS_ROOT_ITEM_KEY
) {
5497 btrfs_warn(root
->fs_info
,
5498 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5499 __func__
, fname
.disk_name
.name
, btrfs_ino(dir
),
5500 location
->objectid
, location
->type
, location
->offset
);
5503 *type
= btrfs_dir_ftype(path
->nodes
[0], di
);
5505 fscrypt_free_filename(&fname
);
5506 btrfs_free_path(path
);
5511 * when we hit a tree root in a directory, the btrfs part of the inode
5512 * needs to be changed to reflect the root directory of the tree root. This
5513 * is kind of like crossing a mount point.
5515 static int fixup_tree_root_location(struct btrfs_fs_info
*fs_info
,
5516 struct btrfs_inode
*dir
,
5517 struct dentry
*dentry
,
5518 struct btrfs_key
*location
,
5519 struct btrfs_root
**sub_root
)
5521 struct btrfs_path
*path
;
5522 struct btrfs_root
*new_root
;
5523 struct btrfs_root_ref
*ref
;
5524 struct extent_buffer
*leaf
;
5525 struct btrfs_key key
;
5528 struct fscrypt_name fname
;
5530 ret
= fscrypt_setup_filename(&dir
->vfs_inode
, &dentry
->d_name
, 0, &fname
);
5534 path
= btrfs_alloc_path();
5541 key
.objectid
= dir
->root
->root_key
.objectid
;
5542 key
.type
= BTRFS_ROOT_REF_KEY
;
5543 key
.offset
= location
->objectid
;
5545 ret
= btrfs_search_slot(NULL
, fs_info
->tree_root
, &key
, path
, 0, 0);
5552 leaf
= path
->nodes
[0];
5553 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5554 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
5555 btrfs_root_ref_name_len(leaf
, ref
) != fname
.disk_name
.len
)
5558 ret
= memcmp_extent_buffer(leaf
, fname
.disk_name
.name
,
5559 (unsigned long)(ref
+ 1), fname
.disk_name
.len
);
5563 btrfs_release_path(path
);
5565 new_root
= btrfs_get_fs_root(fs_info
, location
->objectid
, true);
5566 if (IS_ERR(new_root
)) {
5567 err
= PTR_ERR(new_root
);
5571 *sub_root
= new_root
;
5572 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5573 location
->type
= BTRFS_INODE_ITEM_KEY
;
5574 location
->offset
= 0;
5577 btrfs_free_path(path
);
5578 fscrypt_free_filename(&fname
);
5582 static void inode_tree_add(struct btrfs_inode
*inode
)
5584 struct btrfs_root
*root
= inode
->root
;
5585 struct btrfs_inode
*entry
;
5587 struct rb_node
*parent
;
5588 struct rb_node
*new = &inode
->rb_node
;
5589 u64 ino
= btrfs_ino(inode
);
5591 if (inode_unhashed(&inode
->vfs_inode
))
5594 spin_lock(&root
->inode_lock
);
5595 p
= &root
->inode_tree
.rb_node
;
5598 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5600 if (ino
< btrfs_ino(entry
))
5601 p
= &parent
->rb_left
;
5602 else if (ino
> btrfs_ino(entry
))
5603 p
= &parent
->rb_right
;
5605 WARN_ON(!(entry
->vfs_inode
.i_state
&
5606 (I_WILL_FREE
| I_FREEING
)));
5607 rb_replace_node(parent
, new, &root
->inode_tree
);
5608 RB_CLEAR_NODE(parent
);
5609 spin_unlock(&root
->inode_lock
);
5613 rb_link_node(new, parent
, p
);
5614 rb_insert_color(new, &root
->inode_tree
);
5615 spin_unlock(&root
->inode_lock
);
5618 static void inode_tree_del(struct btrfs_inode
*inode
)
5620 struct btrfs_root
*root
= inode
->root
;
5623 spin_lock(&root
->inode_lock
);
5624 if (!RB_EMPTY_NODE(&inode
->rb_node
)) {
5625 rb_erase(&inode
->rb_node
, &root
->inode_tree
);
5626 RB_CLEAR_NODE(&inode
->rb_node
);
5627 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5629 spin_unlock(&root
->inode_lock
);
5631 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5632 spin_lock(&root
->inode_lock
);
5633 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5634 spin_unlock(&root
->inode_lock
);
5636 btrfs_add_dead_root(root
);
5641 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5643 struct btrfs_iget_args
*args
= p
;
5645 inode
->i_ino
= args
->ino
;
5646 BTRFS_I(inode
)->location
.objectid
= args
->ino
;
5647 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
5648 BTRFS_I(inode
)->location
.offset
= 0;
5649 BTRFS_I(inode
)->root
= btrfs_grab_root(args
->root
);
5650 BUG_ON(args
->root
&& !BTRFS_I(inode
)->root
);
5652 if (args
->root
&& args
->root
== args
->root
->fs_info
->tree_root
&&
5653 args
->ino
!= BTRFS_BTREE_INODE_OBJECTID
)
5654 set_bit(BTRFS_INODE_FREE_SPACE_INODE
,
5655 &BTRFS_I(inode
)->runtime_flags
);
5659 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5661 struct btrfs_iget_args
*args
= opaque
;
5663 return args
->ino
== BTRFS_I(inode
)->location
.objectid
&&
5664 args
->root
== BTRFS_I(inode
)->root
;
5667 static struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 ino
,
5668 struct btrfs_root
*root
)
5670 struct inode
*inode
;
5671 struct btrfs_iget_args args
;
5672 unsigned long hashval
= btrfs_inode_hash(ino
, root
);
5677 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5678 btrfs_init_locked_inode
,
5684 * Get an inode object given its inode number and corresponding root.
5685 * Path can be preallocated to prevent recursing back to iget through
5686 * allocator. NULL is also valid but may require an additional allocation
5689 struct inode
*btrfs_iget_path(struct super_block
*s
, u64 ino
,
5690 struct btrfs_root
*root
, struct btrfs_path
*path
)
5692 struct inode
*inode
;
5694 inode
= btrfs_iget_locked(s
, ino
, root
);
5696 return ERR_PTR(-ENOMEM
);
5698 if (inode
->i_state
& I_NEW
) {
5701 ret
= btrfs_read_locked_inode(inode
, path
);
5703 inode_tree_add(BTRFS_I(inode
));
5704 unlock_new_inode(inode
);
5708 * ret > 0 can come from btrfs_search_slot called by
5709 * btrfs_read_locked_inode, this means the inode item
5714 inode
= ERR_PTR(ret
);
5721 struct inode
*btrfs_iget(struct super_block
*s
, u64 ino
, struct btrfs_root
*root
)
5723 return btrfs_iget_path(s
, ino
, root
, NULL
);
5726 static struct inode
*new_simple_dir(struct super_block
*s
,
5727 struct btrfs_key
*key
,
5728 struct btrfs_root
*root
)
5730 struct inode
*inode
= new_inode(s
);
5733 return ERR_PTR(-ENOMEM
);
5735 BTRFS_I(inode
)->root
= btrfs_grab_root(root
);
5736 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5737 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5739 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5741 * We only need lookup, the rest is read-only and there's no inode
5742 * associated with the dentry
5744 inode
->i_op
= &simple_dir_inode_operations
;
5745 inode
->i_opflags
&= ~IOP_XATTR
;
5746 inode
->i_fop
= &simple_dir_operations
;
5747 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5748 inode
->i_mtime
= current_time(inode
);
5749 inode
->i_atime
= inode
->i_mtime
;
5750 inode
->i_ctime
= inode
->i_mtime
;
5751 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5756 static_assert(BTRFS_FT_UNKNOWN
== FT_UNKNOWN
);
5757 static_assert(BTRFS_FT_REG_FILE
== FT_REG_FILE
);
5758 static_assert(BTRFS_FT_DIR
== FT_DIR
);
5759 static_assert(BTRFS_FT_CHRDEV
== FT_CHRDEV
);
5760 static_assert(BTRFS_FT_BLKDEV
== FT_BLKDEV
);
5761 static_assert(BTRFS_FT_FIFO
== FT_FIFO
);
5762 static_assert(BTRFS_FT_SOCK
== FT_SOCK
);
5763 static_assert(BTRFS_FT_SYMLINK
== FT_SYMLINK
);
5765 static inline u8
btrfs_inode_type(struct inode
*inode
)
5767 return fs_umode_to_ftype(inode
->i_mode
);
5770 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5772 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
5773 struct inode
*inode
;
5774 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5775 struct btrfs_root
*sub_root
= root
;
5776 struct btrfs_key location
;
5780 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5781 return ERR_PTR(-ENAMETOOLONG
);
5783 ret
= btrfs_inode_by_name(BTRFS_I(dir
), dentry
, &location
, &di_type
);
5785 return ERR_PTR(ret
);
5787 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5788 inode
= btrfs_iget(dir
->i_sb
, location
.objectid
, root
);
5792 /* Do extra check against inode mode with di_type */
5793 if (btrfs_inode_type(inode
) != di_type
) {
5795 "inode mode mismatch with dir: inode mode=0%o btrfs type=%u dir type=%u",
5796 inode
->i_mode
, btrfs_inode_type(inode
),
5799 return ERR_PTR(-EUCLEAN
);
5804 ret
= fixup_tree_root_location(fs_info
, BTRFS_I(dir
), dentry
,
5805 &location
, &sub_root
);
5808 inode
= ERR_PTR(ret
);
5810 inode
= new_simple_dir(dir
->i_sb
, &location
, root
);
5812 inode
= btrfs_iget(dir
->i_sb
, location
.objectid
, sub_root
);
5813 btrfs_put_root(sub_root
);
5818 down_read(&fs_info
->cleanup_work_sem
);
5819 if (!sb_rdonly(inode
->i_sb
))
5820 ret
= btrfs_orphan_cleanup(sub_root
);
5821 up_read(&fs_info
->cleanup_work_sem
);
5824 inode
= ERR_PTR(ret
);
5831 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5833 struct btrfs_root
*root
;
5834 struct inode
*inode
= d_inode(dentry
);
5836 if (!inode
&& !IS_ROOT(dentry
))
5837 inode
= d_inode(dentry
->d_parent
);
5840 root
= BTRFS_I(inode
)->root
;
5841 if (btrfs_root_refs(&root
->root_item
) == 0)
5844 if (btrfs_ino(BTRFS_I(inode
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5850 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5853 struct inode
*inode
= btrfs_lookup_dentry(dir
, dentry
);
5855 if (inode
== ERR_PTR(-ENOENT
))
5857 return d_splice_alias(inode
, dentry
);
5861 * All this infrastructure exists because dir_emit can fault, and we are holding
5862 * the tree lock when doing readdir. For now just allocate a buffer and copy
5863 * our information into that, and then dir_emit from the buffer. This is
5864 * similar to what NFS does, only we don't keep the buffer around in pagecache
5865 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5866 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5869 static int btrfs_opendir(struct inode
*inode
, struct file
*file
)
5871 struct btrfs_file_private
*private;
5873 private = kzalloc(sizeof(struct btrfs_file_private
), GFP_KERNEL
);
5876 private->filldir_buf
= kzalloc(PAGE_SIZE
, GFP_KERNEL
);
5877 if (!private->filldir_buf
) {
5881 file
->private_data
= private;
5892 static int btrfs_filldir(void *addr
, int entries
, struct dir_context
*ctx
)
5895 struct dir_entry
*entry
= addr
;
5896 char *name
= (char *)(entry
+ 1);
5898 ctx
->pos
= get_unaligned(&entry
->offset
);
5899 if (!dir_emit(ctx
, name
, get_unaligned(&entry
->name_len
),
5900 get_unaligned(&entry
->ino
),
5901 get_unaligned(&entry
->type
)))
5903 addr
+= sizeof(struct dir_entry
) +
5904 get_unaligned(&entry
->name_len
);
5910 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5912 struct inode
*inode
= file_inode(file
);
5913 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5914 struct btrfs_file_private
*private = file
->private_data
;
5915 struct btrfs_dir_item
*di
;
5916 struct btrfs_key key
;
5917 struct btrfs_key found_key
;
5918 struct btrfs_path
*path
;
5920 struct list_head ins_list
;
5921 struct list_head del_list
;
5928 struct btrfs_key location
;
5930 if (!dir_emit_dots(file
, ctx
))
5933 path
= btrfs_alloc_path();
5937 addr
= private->filldir_buf
;
5938 path
->reada
= READA_FORWARD
;
5940 INIT_LIST_HEAD(&ins_list
);
5941 INIT_LIST_HEAD(&del_list
);
5942 put
= btrfs_readdir_get_delayed_items(inode
, &ins_list
, &del_list
);
5945 key
.type
= BTRFS_DIR_INDEX_KEY
;
5946 key
.offset
= ctx
->pos
;
5947 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
5949 btrfs_for_each_slot(root
, &key
, &found_key
, path
, ret
) {
5950 struct dir_entry
*entry
;
5951 struct extent_buffer
*leaf
= path
->nodes
[0];
5954 if (found_key
.objectid
!= key
.objectid
)
5956 if (found_key
.type
!= BTRFS_DIR_INDEX_KEY
)
5958 if (found_key
.offset
< ctx
->pos
)
5960 if (btrfs_should_delete_dir_index(&del_list
, found_key
.offset
))
5962 di
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dir_item
);
5963 name_len
= btrfs_dir_name_len(leaf
, di
);
5964 if ((total_len
+ sizeof(struct dir_entry
) + name_len
) >=
5966 btrfs_release_path(path
);
5967 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
5970 addr
= private->filldir_buf
;
5976 ftype
= btrfs_dir_flags_to_ftype(btrfs_dir_flags(leaf
, di
));
5978 name_ptr
= (char *)(entry
+ 1);
5979 read_extent_buffer(leaf
, name_ptr
,
5980 (unsigned long)(di
+ 1), name_len
);
5981 put_unaligned(name_len
, &entry
->name_len
);
5982 put_unaligned(fs_ftype_to_dtype(ftype
), &entry
->type
);
5983 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5984 put_unaligned(location
.objectid
, &entry
->ino
);
5985 put_unaligned(found_key
.offset
, &entry
->offset
);
5987 addr
+= sizeof(struct dir_entry
) + name_len
;
5988 total_len
+= sizeof(struct dir_entry
) + name_len
;
5990 /* Catch error encountered during iteration */
5994 btrfs_release_path(path
);
5996 ret
= btrfs_filldir(private->filldir_buf
, entries
, ctx
);
6000 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
6005 * Stop new entries from being returned after we return the last
6008 * New directory entries are assigned a strictly increasing
6009 * offset. This means that new entries created during readdir
6010 * are *guaranteed* to be seen in the future by that readdir.
6011 * This has broken buggy programs which operate on names as
6012 * they're returned by readdir. Until we re-use freed offsets
6013 * we have this hack to stop new entries from being returned
6014 * under the assumption that they'll never reach this huge
6017 * This is being careful not to overflow 32bit loff_t unless the
6018 * last entry requires it because doing so has broken 32bit apps
6021 if (ctx
->pos
>= INT_MAX
)
6022 ctx
->pos
= LLONG_MAX
;
6029 btrfs_readdir_put_delayed_items(inode
, &ins_list
, &del_list
);
6030 btrfs_free_path(path
);
6035 * This is somewhat expensive, updating the tree every time the
6036 * inode changes. But, it is most likely to find the inode in cache.
6037 * FIXME, needs more benchmarking...there are no reasons other than performance
6038 * to keep or drop this code.
6040 static int btrfs_dirty_inode(struct btrfs_inode
*inode
)
6042 struct btrfs_root
*root
= inode
->root
;
6043 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
6044 struct btrfs_trans_handle
*trans
;
6047 if (test_bit(BTRFS_INODE_DUMMY
, &inode
->runtime_flags
))
6050 trans
= btrfs_join_transaction(root
);
6052 return PTR_ERR(trans
);
6054 ret
= btrfs_update_inode(trans
, root
, inode
);
6055 if (ret
&& (ret
== -ENOSPC
|| ret
== -EDQUOT
)) {
6056 /* whoops, lets try again with the full transaction */
6057 btrfs_end_transaction(trans
);
6058 trans
= btrfs_start_transaction(root
, 1);
6060 return PTR_ERR(trans
);
6062 ret
= btrfs_update_inode(trans
, root
, inode
);
6064 btrfs_end_transaction(trans
);
6065 if (inode
->delayed_node
)
6066 btrfs_balance_delayed_items(fs_info
);
6072 * This is a copy of file_update_time. We need this so we can return error on
6073 * ENOSPC for updating the inode in the case of file write and mmap writes.
6075 static int btrfs_update_time(struct inode
*inode
, struct timespec64
*now
,
6078 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6079 bool dirty
= flags
& ~S_VERSION
;
6081 if (btrfs_root_readonly(root
))
6084 if (flags
& S_VERSION
)
6085 dirty
|= inode_maybe_inc_iversion(inode
, dirty
);
6086 if (flags
& S_CTIME
)
6087 inode
->i_ctime
= *now
;
6088 if (flags
& S_MTIME
)
6089 inode
->i_mtime
= *now
;
6090 if (flags
& S_ATIME
)
6091 inode
->i_atime
= *now
;
6092 return dirty
? btrfs_dirty_inode(BTRFS_I(inode
)) : 0;
6096 * find the highest existing sequence number in a directory
6097 * and then set the in-memory index_cnt variable to reflect
6098 * free sequence numbers
6100 static int btrfs_set_inode_index_count(struct btrfs_inode
*inode
)
6102 struct btrfs_root
*root
= inode
->root
;
6103 struct btrfs_key key
, found_key
;
6104 struct btrfs_path
*path
;
6105 struct extent_buffer
*leaf
;
6108 key
.objectid
= btrfs_ino(inode
);
6109 key
.type
= BTRFS_DIR_INDEX_KEY
;
6110 key
.offset
= (u64
)-1;
6112 path
= btrfs_alloc_path();
6116 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6119 /* FIXME: we should be able to handle this */
6124 if (path
->slots
[0] == 0) {
6125 inode
->index_cnt
= BTRFS_DIR_START_INDEX
;
6131 leaf
= path
->nodes
[0];
6132 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6134 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6135 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6136 inode
->index_cnt
= BTRFS_DIR_START_INDEX
;
6140 inode
->index_cnt
= found_key
.offset
+ 1;
6142 btrfs_free_path(path
);
6147 * helper to find a free sequence number in a given directory. This current
6148 * code is very simple, later versions will do smarter things in the btree
6150 int btrfs_set_inode_index(struct btrfs_inode
*dir
, u64
*index
)
6154 if (dir
->index_cnt
== (u64
)-1) {
6155 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6157 ret
= btrfs_set_inode_index_count(dir
);
6163 *index
= dir
->index_cnt
;
6169 static int btrfs_insert_inode_locked(struct inode
*inode
)
6171 struct btrfs_iget_args args
;
6173 args
.ino
= BTRFS_I(inode
)->location
.objectid
;
6174 args
.root
= BTRFS_I(inode
)->root
;
6176 return insert_inode_locked4(inode
,
6177 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6178 btrfs_find_actor
, &args
);
6181 int btrfs_new_inode_prepare(struct btrfs_new_inode_args
*args
,
6182 unsigned int *trans_num_items
)
6184 struct inode
*dir
= args
->dir
;
6185 struct inode
*inode
= args
->inode
;
6188 if (!args
->orphan
) {
6189 ret
= fscrypt_setup_filename(dir
, &args
->dentry
->d_name
, 0,
6195 ret
= posix_acl_create(dir
, &inode
->i_mode
, &args
->default_acl
, &args
->acl
);
6197 fscrypt_free_filename(&args
->fname
);
6201 /* 1 to add inode item */
6202 *trans_num_items
= 1;
6203 /* 1 to add compression property */
6204 if (BTRFS_I(dir
)->prop_compress
)
6205 (*trans_num_items
)++;
6206 /* 1 to add default ACL xattr */
6207 if (args
->default_acl
)
6208 (*trans_num_items
)++;
6209 /* 1 to add access ACL xattr */
6211 (*trans_num_items
)++;
6212 #ifdef CONFIG_SECURITY
6213 /* 1 to add LSM xattr */
6214 if (dir
->i_security
)
6215 (*trans_num_items
)++;
6218 /* 1 to add orphan item */
6219 (*trans_num_items
)++;
6223 * 1 to add dir index
6224 * 1 to update parent inode item
6226 * No need for 1 unit for the inode ref item because it is
6227 * inserted in a batch together with the inode item at
6228 * btrfs_create_new_inode().
6230 *trans_num_items
+= 3;
6235 void btrfs_new_inode_args_destroy(struct btrfs_new_inode_args
*args
)
6237 posix_acl_release(args
->acl
);
6238 posix_acl_release(args
->default_acl
);
6239 fscrypt_free_filename(&args
->fname
);
6243 * Inherit flags from the parent inode.
6245 * Currently only the compression flags and the cow flags are inherited.
6247 static void btrfs_inherit_iflags(struct btrfs_inode
*inode
, struct btrfs_inode
*dir
)
6253 if (flags
& BTRFS_INODE_NOCOMPRESS
) {
6254 inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6255 inode
->flags
|= BTRFS_INODE_NOCOMPRESS
;
6256 } else if (flags
& BTRFS_INODE_COMPRESS
) {
6257 inode
->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
6258 inode
->flags
|= BTRFS_INODE_COMPRESS
;
6261 if (flags
& BTRFS_INODE_NODATACOW
) {
6262 inode
->flags
|= BTRFS_INODE_NODATACOW
;
6263 if (S_ISREG(inode
->vfs_inode
.i_mode
))
6264 inode
->flags
|= BTRFS_INODE_NODATASUM
;
6267 btrfs_sync_inode_flags_to_i_flags(&inode
->vfs_inode
);
6270 int btrfs_create_new_inode(struct btrfs_trans_handle
*trans
,
6271 struct btrfs_new_inode_args
*args
)
6273 struct inode
*dir
= args
->dir
;
6274 struct inode
*inode
= args
->inode
;
6275 const struct fscrypt_str
*name
= args
->orphan
? NULL
: &args
->fname
.disk_name
;
6276 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6277 struct btrfs_root
*root
;
6278 struct btrfs_inode_item
*inode_item
;
6279 struct btrfs_key
*location
;
6280 struct btrfs_path
*path
;
6282 struct btrfs_inode_ref
*ref
;
6283 struct btrfs_key key
[2];
6285 struct btrfs_item_batch batch
;
6289 path
= btrfs_alloc_path();
6294 BTRFS_I(inode
)->root
= btrfs_grab_root(BTRFS_I(dir
)->root
);
6295 root
= BTRFS_I(inode
)->root
;
6297 ret
= btrfs_get_free_objectid(root
, &objectid
);
6300 inode
->i_ino
= objectid
;
6304 * O_TMPFILE, set link count to 0, so that after this point, we
6305 * fill in an inode item with the correct link count.
6307 set_nlink(inode
, 0);
6309 trace_btrfs_inode_request(dir
);
6311 ret
= btrfs_set_inode_index(BTRFS_I(dir
), &BTRFS_I(inode
)->dir_index
);
6315 /* index_cnt is ignored for everything but a dir. */
6316 BTRFS_I(inode
)->index_cnt
= BTRFS_DIR_START_INDEX
;
6317 BTRFS_I(inode
)->generation
= trans
->transid
;
6318 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6321 * Subvolumes don't inherit flags from their parent directory.
6322 * Originally this was probably by accident, but we probably can't
6323 * change it now without compatibility issues.
6326 btrfs_inherit_iflags(BTRFS_I(inode
), BTRFS_I(dir
));
6328 if (S_ISREG(inode
->i_mode
)) {
6329 if (btrfs_test_opt(fs_info
, NODATASUM
))
6330 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6331 if (btrfs_test_opt(fs_info
, NODATACOW
))
6332 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6333 BTRFS_INODE_NODATASUM
;
6336 location
= &BTRFS_I(inode
)->location
;
6337 location
->objectid
= objectid
;
6338 location
->offset
= 0;
6339 location
->type
= BTRFS_INODE_ITEM_KEY
;
6341 ret
= btrfs_insert_inode_locked(inode
);
6344 BTRFS_I(dir
)->index_cnt
--;
6349 * We could have gotten an inode number from somebody who was fsynced
6350 * and then removed in this same transaction, so let's just set full
6351 * sync since it will be a full sync anyway and this will blow away the
6352 * old info in the log.
6354 btrfs_set_inode_full_sync(BTRFS_I(inode
));
6356 key
[0].objectid
= objectid
;
6357 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6360 sizes
[0] = sizeof(struct btrfs_inode_item
);
6362 if (!args
->orphan
) {
6364 * Start new inodes with an inode_ref. This is slightly more
6365 * efficient for small numbers of hard links since they will
6366 * be packed into one item. Extended refs will kick in if we
6367 * add more hard links than can fit in the ref item.
6369 key
[1].objectid
= objectid
;
6370 key
[1].type
= BTRFS_INODE_REF_KEY
;
6372 key
[1].offset
= objectid
;
6373 sizes
[1] = 2 + sizeof(*ref
);
6375 key
[1].offset
= btrfs_ino(BTRFS_I(dir
));
6376 sizes
[1] = name
->len
+ sizeof(*ref
);
6380 batch
.keys
= &key
[0];
6381 batch
.data_sizes
= &sizes
[0];
6382 batch
.total_data_size
= sizes
[0] + (args
->orphan
? 0 : sizes
[1]);
6383 batch
.nr
= args
->orphan
? 1 : 2;
6384 ret
= btrfs_insert_empty_items(trans
, root
, path
, &batch
);
6386 btrfs_abort_transaction(trans
, ret
);
6390 inode
->i_mtime
= current_time(inode
);
6391 inode
->i_atime
= inode
->i_mtime
;
6392 inode
->i_ctime
= inode
->i_mtime
;
6393 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6396 * We're going to fill the inode item now, so at this point the inode
6397 * must be fully initialized.
6400 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6401 struct btrfs_inode_item
);
6402 memzero_extent_buffer(path
->nodes
[0], (unsigned long)inode_item
,
6403 sizeof(*inode_item
));
6404 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6406 if (!args
->orphan
) {
6407 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6408 struct btrfs_inode_ref
);
6409 ptr
= (unsigned long)(ref
+ 1);
6411 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, 2);
6412 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, 0);
6413 write_extent_buffer(path
->nodes
[0], "..", ptr
, 2);
6415 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
,
6417 btrfs_set_inode_ref_index(path
->nodes
[0], ref
,
6418 BTRFS_I(inode
)->dir_index
);
6419 write_extent_buffer(path
->nodes
[0], name
->name
, ptr
,
6424 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6426 * We don't need the path anymore, plus inheriting properties, adding
6427 * ACLs, security xattrs, orphan item or adding the link, will result in
6428 * allocating yet another path. So just free our path.
6430 btrfs_free_path(path
);
6434 struct inode
*parent
;
6437 * Subvolumes inherit properties from their parent subvolume,
6438 * not the directory they were created in.
6440 parent
= btrfs_iget(fs_info
->sb
, BTRFS_FIRST_FREE_OBJECTID
,
6441 BTRFS_I(dir
)->root
);
6442 if (IS_ERR(parent
)) {
6443 ret
= PTR_ERR(parent
);
6445 ret
= btrfs_inode_inherit_props(trans
, inode
, parent
);
6449 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6453 "error inheriting props for ino %llu (root %llu): %d",
6454 btrfs_ino(BTRFS_I(inode
)), root
->root_key
.objectid
,
6459 * Subvolumes don't inherit ACLs or get passed to the LSM. This is
6462 if (!args
->subvol
) {
6463 ret
= btrfs_init_inode_security(trans
, args
);
6465 btrfs_abort_transaction(trans
, ret
);
6470 inode_tree_add(BTRFS_I(inode
));
6472 trace_btrfs_inode_new(inode
);
6473 btrfs_set_inode_last_trans(trans
, BTRFS_I(inode
));
6475 btrfs_update_root_times(trans
, root
);
6478 ret
= btrfs_orphan_add(trans
, BTRFS_I(inode
));
6480 ret
= btrfs_add_link(trans
, BTRFS_I(dir
), BTRFS_I(inode
), name
,
6481 0, BTRFS_I(inode
)->dir_index
);
6484 btrfs_abort_transaction(trans
, ret
);
6492 * discard_new_inode() calls iput(), but the caller owns the reference
6496 discard_new_inode(inode
);
6498 btrfs_free_path(path
);
6503 * utility function to add 'inode' into 'parent_inode' with
6504 * a give name and a given sequence number.
6505 * if 'add_backref' is true, also insert a backref from the
6506 * inode to the parent directory.
6508 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6509 struct btrfs_inode
*parent_inode
, struct btrfs_inode
*inode
,
6510 const struct fscrypt_str
*name
, int add_backref
, u64 index
)
6513 struct btrfs_key key
;
6514 struct btrfs_root
*root
= parent_inode
->root
;
6515 u64 ino
= btrfs_ino(inode
);
6516 u64 parent_ino
= btrfs_ino(parent_inode
);
6518 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6519 memcpy(&key
, &inode
->root
->root_key
, sizeof(key
));
6522 key
.type
= BTRFS_INODE_ITEM_KEY
;
6526 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6527 ret
= btrfs_add_root_ref(trans
, key
.objectid
,
6528 root
->root_key
.objectid
, parent_ino
,
6530 } else if (add_backref
) {
6531 ret
= btrfs_insert_inode_ref(trans
, root
, name
,
6532 ino
, parent_ino
, index
);
6535 /* Nothing to clean up yet */
6539 ret
= btrfs_insert_dir_item(trans
, name
, parent_inode
, &key
,
6540 btrfs_inode_type(&inode
->vfs_inode
), index
);
6541 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6544 btrfs_abort_transaction(trans
, ret
);
6548 btrfs_i_size_write(parent_inode
, parent_inode
->vfs_inode
.i_size
+
6550 inode_inc_iversion(&parent_inode
->vfs_inode
);
6552 * If we are replaying a log tree, we do not want to update the mtime
6553 * and ctime of the parent directory with the current time, since the
6554 * log replay procedure is responsible for setting them to their correct
6555 * values (the ones it had when the fsync was done).
6557 if (!test_bit(BTRFS_FS_LOG_RECOVERING
, &root
->fs_info
->flags
)) {
6558 struct timespec64 now
= current_time(&parent_inode
->vfs_inode
);
6560 parent_inode
->vfs_inode
.i_mtime
= now
;
6561 parent_inode
->vfs_inode
.i_ctime
= now
;
6563 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
6565 btrfs_abort_transaction(trans
, ret
);
6569 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6572 err
= btrfs_del_root_ref(trans
, key
.objectid
,
6573 root
->root_key
.objectid
, parent_ino
,
6574 &local_index
, name
);
6576 btrfs_abort_transaction(trans
, err
);
6577 } else if (add_backref
) {
6581 err
= btrfs_del_inode_ref(trans
, root
, name
, ino
, parent_ino
,
6584 btrfs_abort_transaction(trans
, err
);
6587 /* Return the original error code */
6591 static int btrfs_create_common(struct inode
*dir
, struct dentry
*dentry
,
6592 struct inode
*inode
)
6594 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
6595 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6596 struct btrfs_new_inode_args new_inode_args
= {
6601 unsigned int trans_num_items
;
6602 struct btrfs_trans_handle
*trans
;
6605 err
= btrfs_new_inode_prepare(&new_inode_args
, &trans_num_items
);
6609 trans
= btrfs_start_transaction(root
, trans_num_items
);
6610 if (IS_ERR(trans
)) {
6611 err
= PTR_ERR(trans
);
6612 goto out_new_inode_args
;
6615 err
= btrfs_create_new_inode(trans
, &new_inode_args
);
6617 d_instantiate_new(dentry
, inode
);
6619 btrfs_end_transaction(trans
);
6620 btrfs_btree_balance_dirty(fs_info
);
6622 btrfs_new_inode_args_destroy(&new_inode_args
);
6629 static int btrfs_mknod(struct mnt_idmap
*idmap
, struct inode
*dir
,
6630 struct dentry
*dentry
, umode_t mode
, dev_t rdev
)
6632 struct inode
*inode
;
6634 inode
= new_inode(dir
->i_sb
);
6637 inode_init_owner(idmap
, inode
, dir
, mode
);
6638 inode
->i_op
= &btrfs_special_inode_operations
;
6639 init_special_inode(inode
, inode
->i_mode
, rdev
);
6640 return btrfs_create_common(dir
, dentry
, inode
);
6643 static int btrfs_create(struct mnt_idmap
*idmap
, struct inode
*dir
,
6644 struct dentry
*dentry
, umode_t mode
, bool excl
)
6646 struct inode
*inode
;
6648 inode
= new_inode(dir
->i_sb
);
6651 inode_init_owner(idmap
, inode
, dir
, mode
);
6652 inode
->i_fop
= &btrfs_file_operations
;
6653 inode
->i_op
= &btrfs_file_inode_operations
;
6654 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6655 return btrfs_create_common(dir
, dentry
, inode
);
6658 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6659 struct dentry
*dentry
)
6661 struct btrfs_trans_handle
*trans
= NULL
;
6662 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6663 struct inode
*inode
= d_inode(old_dentry
);
6664 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
6665 struct fscrypt_name fname
;
6670 /* do not allow sys_link's with other subvols of the same device */
6671 if (root
->root_key
.objectid
!= BTRFS_I(inode
)->root
->root_key
.objectid
)
6674 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6677 err
= fscrypt_setup_filename(dir
, &dentry
->d_name
, 0, &fname
);
6681 err
= btrfs_set_inode_index(BTRFS_I(dir
), &index
);
6686 * 2 items for inode and inode ref
6687 * 2 items for dir items
6688 * 1 item for parent inode
6689 * 1 item for orphan item deletion if O_TMPFILE
6691 trans
= btrfs_start_transaction(root
, inode
->i_nlink
? 5 : 6);
6692 if (IS_ERR(trans
)) {
6693 err
= PTR_ERR(trans
);
6698 /* There are several dir indexes for this inode, clear the cache. */
6699 BTRFS_I(inode
)->dir_index
= 0ULL;
6701 inode_inc_iversion(inode
);
6702 inode
->i_ctime
= current_time(inode
);
6704 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6706 err
= btrfs_add_link(trans
, BTRFS_I(dir
), BTRFS_I(inode
),
6707 &fname
.disk_name
, 1, index
);
6712 struct dentry
*parent
= dentry
->d_parent
;
6714 err
= btrfs_update_inode(trans
, root
, BTRFS_I(inode
));
6717 if (inode
->i_nlink
== 1) {
6719 * If new hard link count is 1, it's a file created
6720 * with open(2) O_TMPFILE flag.
6722 err
= btrfs_orphan_del(trans
, BTRFS_I(inode
));
6726 d_instantiate(dentry
, inode
);
6727 btrfs_log_new_name(trans
, old_dentry
, NULL
, 0, parent
);
6731 fscrypt_free_filename(&fname
);
6733 btrfs_end_transaction(trans
);
6735 inode_dec_link_count(inode
);
6738 btrfs_btree_balance_dirty(fs_info
);
6742 static int btrfs_mkdir(struct mnt_idmap
*idmap
, struct inode
*dir
,
6743 struct dentry
*dentry
, umode_t mode
)
6745 struct inode
*inode
;
6747 inode
= new_inode(dir
->i_sb
);
6750 inode_init_owner(idmap
, inode
, dir
, S_IFDIR
| mode
);
6751 inode
->i_op
= &btrfs_dir_inode_operations
;
6752 inode
->i_fop
= &btrfs_dir_file_operations
;
6753 return btrfs_create_common(dir
, dentry
, inode
);
6756 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6758 struct btrfs_file_extent_item
*item
)
6761 struct extent_buffer
*leaf
= path
->nodes
[0];
6764 unsigned long inline_size
;
6768 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6769 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6770 inline_size
= btrfs_file_extent_inline_item_len(leaf
, path
->slots
[0]);
6771 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6774 ptr
= btrfs_file_extent_inline_start(item
);
6776 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6778 max_size
= min_t(unsigned long, PAGE_SIZE
, max_size
);
6779 ret
= btrfs_decompress(compress_type
, tmp
, page
, 0, inline_size
, max_size
);
6782 * decompression code contains a memset to fill in any space between the end
6783 * of the uncompressed data and the end of max_size in case the decompressed
6784 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6785 * the end of an inline extent and the beginning of the next block, so we
6786 * cover that region here.
6789 if (max_size
< PAGE_SIZE
)
6790 memzero_page(page
, max_size
, PAGE_SIZE
- max_size
);
6795 static int read_inline_extent(struct btrfs_inode
*inode
, struct btrfs_path
*path
,
6798 struct btrfs_file_extent_item
*fi
;
6802 if (!page
|| PageUptodate(page
))
6805 ASSERT(page_offset(page
) == 0);
6807 fi
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6808 struct btrfs_file_extent_item
);
6809 if (btrfs_file_extent_compression(path
->nodes
[0], fi
) != BTRFS_COMPRESS_NONE
)
6810 return uncompress_inline(path
, page
, fi
);
6812 copy_size
= min_t(u64
, PAGE_SIZE
,
6813 btrfs_file_extent_ram_bytes(path
->nodes
[0], fi
));
6814 kaddr
= kmap_local_page(page
);
6815 read_extent_buffer(path
->nodes
[0], kaddr
,
6816 btrfs_file_extent_inline_start(fi
), copy_size
);
6817 kunmap_local(kaddr
);
6818 if (copy_size
< PAGE_SIZE
)
6819 memzero_page(page
, copy_size
, PAGE_SIZE
- copy_size
);
6824 * Lookup the first extent overlapping a range in a file.
6826 * @inode: file to search in
6827 * @page: page to read extent data into if the extent is inline
6828 * @pg_offset: offset into @page to copy to
6829 * @start: file offset
6830 * @len: length of range starting at @start
6832 * Return the first &struct extent_map which overlaps the given range, reading
6833 * it from the B-tree and caching it if necessary. Note that there may be more
6834 * extents which overlap the given range after the returned extent_map.
6836 * If @page is not NULL and the extent is inline, this also reads the extent
6837 * data directly into the page and marks the extent up to date in the io_tree.
6839 * Return: ERR_PTR on error, non-NULL extent_map on success.
6841 struct extent_map
*btrfs_get_extent(struct btrfs_inode
*inode
,
6842 struct page
*page
, size_t pg_offset
,
6845 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
6847 u64 extent_start
= 0;
6849 u64 objectid
= btrfs_ino(inode
);
6850 int extent_type
= -1;
6851 struct btrfs_path
*path
= NULL
;
6852 struct btrfs_root
*root
= inode
->root
;
6853 struct btrfs_file_extent_item
*item
;
6854 struct extent_buffer
*leaf
;
6855 struct btrfs_key found_key
;
6856 struct extent_map
*em
= NULL
;
6857 struct extent_map_tree
*em_tree
= &inode
->extent_tree
;
6859 read_lock(&em_tree
->lock
);
6860 em
= lookup_extent_mapping(em_tree
, start
, len
);
6861 read_unlock(&em_tree
->lock
);
6864 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6865 free_extent_map(em
);
6866 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6867 free_extent_map(em
);
6871 em
= alloc_extent_map();
6876 em
->start
= EXTENT_MAP_HOLE
;
6877 em
->orig_start
= EXTENT_MAP_HOLE
;
6879 em
->block_len
= (u64
)-1;
6881 path
= btrfs_alloc_path();
6887 /* Chances are we'll be called again, so go ahead and do readahead */
6888 path
->reada
= READA_FORWARD
;
6891 * The same explanation in load_free_space_cache applies here as well,
6892 * we only read when we're loading the free space cache, and at that
6893 * point the commit_root has everything we need.
6895 if (btrfs_is_free_space_inode(inode
)) {
6896 path
->search_commit_root
= 1;
6897 path
->skip_locking
= 1;
6900 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, objectid
, start
, 0);
6903 } else if (ret
> 0) {
6904 if (path
->slots
[0] == 0)
6910 leaf
= path
->nodes
[0];
6911 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6912 struct btrfs_file_extent_item
);
6913 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6914 if (found_key
.objectid
!= objectid
||
6915 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6917 * If we backup past the first extent we want to move forward
6918 * and see if there is an extent in front of us, otherwise we'll
6919 * say there is a hole for our whole search range which can
6926 extent_type
= btrfs_file_extent_type(leaf
, item
);
6927 extent_start
= found_key
.offset
;
6928 extent_end
= btrfs_file_extent_end(path
);
6929 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
6930 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6931 /* Only regular file could have regular/prealloc extent */
6932 if (!S_ISREG(inode
->vfs_inode
.i_mode
)) {
6935 "regular/prealloc extent found for non-regular inode %llu",
6939 trace_btrfs_get_extent_show_fi_regular(inode
, leaf
, item
,
6941 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
6942 trace_btrfs_get_extent_show_fi_inline(inode
, leaf
, item
,
6947 if (start
>= extent_end
) {
6949 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6950 ret
= btrfs_next_leaf(root
, path
);
6956 leaf
= path
->nodes
[0];
6958 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6959 if (found_key
.objectid
!= objectid
||
6960 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6962 if (start
+ len
<= found_key
.offset
)
6964 if (start
> found_key
.offset
)
6967 /* New extent overlaps with existing one */
6969 em
->orig_start
= start
;
6970 em
->len
= found_key
.offset
- start
;
6971 em
->block_start
= EXTENT_MAP_HOLE
;
6975 btrfs_extent_item_to_extent_map(inode
, path
, item
, em
);
6977 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
6978 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6980 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
6982 * Inline extent can only exist at file offset 0. This is
6983 * ensured by tree-checker and inline extent creation path.
6984 * Thus all members representing file offsets should be zero.
6986 ASSERT(pg_offset
== 0);
6987 ASSERT(extent_start
== 0);
6988 ASSERT(em
->start
== 0);
6991 * btrfs_extent_item_to_extent_map() should have properly
6992 * initialized em members already.
6994 * Other members are not utilized for inline extents.
6996 ASSERT(em
->block_start
== EXTENT_MAP_INLINE
);
6997 ASSERT(em
->len
== fs_info
->sectorsize
);
6999 ret
= read_inline_extent(inode
, path
, page
);
7006 em
->orig_start
= start
;
7008 em
->block_start
= EXTENT_MAP_HOLE
;
7011 btrfs_release_path(path
);
7012 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
7014 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7015 em
->start
, em
->len
, start
, len
);
7020 write_lock(&em_tree
->lock
);
7021 ret
= btrfs_add_extent_mapping(fs_info
, em_tree
, &em
, start
, len
);
7022 write_unlock(&em_tree
->lock
);
7024 btrfs_free_path(path
);
7026 trace_btrfs_get_extent(root
, inode
, em
);
7029 free_extent_map(em
);
7030 return ERR_PTR(ret
);
7035 static struct extent_map
*btrfs_create_dio_extent(struct btrfs_inode
*inode
,
7036 struct btrfs_dio_data
*dio_data
,
7039 const u64 orig_start
,
7040 const u64 block_start
,
7041 const u64 block_len
,
7042 const u64 orig_block_len
,
7043 const u64 ram_bytes
,
7046 struct extent_map
*em
= NULL
;
7047 struct btrfs_ordered_extent
*ordered
;
7049 if (type
!= BTRFS_ORDERED_NOCOW
) {
7050 em
= create_io_em(inode
, start
, len
, orig_start
, block_start
,
7051 block_len
, orig_block_len
, ram_bytes
,
7052 BTRFS_COMPRESS_NONE
, /* compress_type */
7057 ordered
= btrfs_alloc_ordered_extent(inode
, start
, len
, len
,
7058 block_start
, block_len
, 0,
7060 (1 << BTRFS_ORDERED_DIRECT
),
7061 BTRFS_COMPRESS_NONE
);
7062 if (IS_ERR(ordered
)) {
7064 free_extent_map(em
);
7065 btrfs_drop_extent_map_range(inode
, start
,
7066 start
+ len
- 1, false);
7068 em
= ERR_CAST(ordered
);
7070 ASSERT(!dio_data
->ordered
);
7071 dio_data
->ordered
= ordered
;
7078 static struct extent_map
*btrfs_new_extent_direct(struct btrfs_inode
*inode
,
7079 struct btrfs_dio_data
*dio_data
,
7082 struct btrfs_root
*root
= inode
->root
;
7083 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7084 struct extent_map
*em
;
7085 struct btrfs_key ins
;
7089 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7090 ret
= btrfs_reserve_extent(root
, len
, len
, fs_info
->sectorsize
,
7091 0, alloc_hint
, &ins
, 1, 1);
7093 return ERR_PTR(ret
);
7095 em
= btrfs_create_dio_extent(inode
, dio_data
, start
, ins
.offset
, start
,
7096 ins
.objectid
, ins
.offset
, ins
.offset
,
7097 ins
.offset
, BTRFS_ORDERED_REGULAR
);
7098 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
7100 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
,
7106 static bool btrfs_extent_readonly(struct btrfs_fs_info
*fs_info
, u64 bytenr
)
7108 struct btrfs_block_group
*block_group
;
7109 bool readonly
= false;
7111 block_group
= btrfs_lookup_block_group(fs_info
, bytenr
);
7112 if (!block_group
|| block_group
->ro
)
7115 btrfs_put_block_group(block_group
);
7120 * Check if we can do nocow write into the range [@offset, @offset + @len)
7122 * @offset: File offset
7123 * @len: The length to write, will be updated to the nocow writeable
7125 * @orig_start: (optional) Return the original file offset of the file extent
7126 * @orig_len: (optional) Return the original on-disk length of the file extent
7127 * @ram_bytes: (optional) Return the ram_bytes of the file extent
7128 * @strict: if true, omit optimizations that might force us into unnecessary
7129 * cow. e.g., don't trust generation number.
7132 * >0 and update @len if we can do nocow write
7133 * 0 if we can't do nocow write
7134 * <0 if error happened
7136 * NOTE: This only checks the file extents, caller is responsible to wait for
7137 * any ordered extents.
7139 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7140 u64
*orig_start
, u64
*orig_block_len
,
7141 u64
*ram_bytes
, bool nowait
, bool strict
)
7143 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7144 struct can_nocow_file_extent_args nocow_args
= { 0 };
7145 struct btrfs_path
*path
;
7147 struct extent_buffer
*leaf
;
7148 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7149 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7150 struct btrfs_file_extent_item
*fi
;
7151 struct btrfs_key key
;
7154 path
= btrfs_alloc_path();
7157 path
->nowait
= nowait
;
7159 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
7160 btrfs_ino(BTRFS_I(inode
)), offset
, 0);
7165 if (path
->slots
[0] == 0) {
7166 /* can't find the item, must cow */
7173 leaf
= path
->nodes
[0];
7174 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
7175 if (key
.objectid
!= btrfs_ino(BTRFS_I(inode
)) ||
7176 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7177 /* not our file or wrong item type, must cow */
7181 if (key
.offset
> offset
) {
7182 /* Wrong offset, must cow */
7186 if (btrfs_file_extent_end(path
) <= offset
)
7189 fi
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_file_extent_item
);
7190 found_type
= btrfs_file_extent_type(leaf
, fi
);
7192 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7194 nocow_args
.start
= offset
;
7195 nocow_args
.end
= offset
+ *len
- 1;
7196 nocow_args
.strict
= strict
;
7197 nocow_args
.free_path
= true;
7199 ret
= can_nocow_file_extent(path
, &key
, BTRFS_I(inode
), &nocow_args
);
7200 /* can_nocow_file_extent() has freed the path. */
7204 /* Treat errors as not being able to NOCOW. */
7210 if (btrfs_extent_readonly(fs_info
, nocow_args
.disk_bytenr
))
7213 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7214 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7217 range_end
= round_up(offset
+ nocow_args
.num_bytes
,
7218 root
->fs_info
->sectorsize
) - 1;
7219 ret
= test_range_bit(io_tree
, offset
, range_end
,
7220 EXTENT_DELALLOC
, 0, NULL
);
7228 *orig_start
= key
.offset
- nocow_args
.extent_offset
;
7230 *orig_block_len
= nocow_args
.disk_num_bytes
;
7232 *len
= nocow_args
.num_bytes
;
7235 btrfs_free_path(path
);
7239 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7240 struct extent_state
**cached_state
,
7241 unsigned int iomap_flags
)
7243 const bool writing
= (iomap_flags
& IOMAP_WRITE
);
7244 const bool nowait
= (iomap_flags
& IOMAP_NOWAIT
);
7245 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7246 struct btrfs_ordered_extent
*ordered
;
7251 if (!try_lock_extent(io_tree
, lockstart
, lockend
,
7255 lock_extent(io_tree
, lockstart
, lockend
, cached_state
);
7258 * We're concerned with the entire range that we're going to be
7259 * doing DIO to, so we need to make sure there's no ordered
7260 * extents in this range.
7262 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), lockstart
,
7263 lockend
- lockstart
+ 1);
7266 * We need to make sure there are no buffered pages in this
7267 * range either, we could have raced between the invalidate in
7268 * generic_file_direct_write and locking the extent. The
7269 * invalidate needs to happen so that reads after a write do not
7273 (!writing
|| !filemap_range_has_page(inode
->i_mapping
,
7274 lockstart
, lockend
)))
7277 unlock_extent(io_tree
, lockstart
, lockend
, cached_state
);
7281 btrfs_put_ordered_extent(ordered
);
7286 * If we are doing a DIO read and the ordered extent we
7287 * found is for a buffered write, we can not wait for it
7288 * to complete and retry, because if we do so we can
7289 * deadlock with concurrent buffered writes on page
7290 * locks. This happens only if our DIO read covers more
7291 * than one extent map, if at this point has already
7292 * created an ordered extent for a previous extent map
7293 * and locked its range in the inode's io tree, and a
7294 * concurrent write against that previous extent map's
7295 * range and this range started (we unlock the ranges
7296 * in the io tree only when the bios complete and
7297 * buffered writes always lock pages before attempting
7298 * to lock range in the io tree).
7301 test_bit(BTRFS_ORDERED_DIRECT
, &ordered
->flags
))
7302 btrfs_start_ordered_extent(ordered
);
7304 ret
= nowait
? -EAGAIN
: -ENOTBLK
;
7305 btrfs_put_ordered_extent(ordered
);
7308 * We could trigger writeback for this range (and wait
7309 * for it to complete) and then invalidate the pages for
7310 * this range (through invalidate_inode_pages2_range()),
7311 * but that can lead us to a deadlock with a concurrent
7312 * call to readahead (a buffered read or a defrag call
7313 * triggered a readahead) on a page lock due to an
7314 * ordered dio extent we created before but did not have
7315 * yet a corresponding bio submitted (whence it can not
7316 * complete), which makes readahead wait for that
7317 * ordered extent to complete while holding a lock on
7320 ret
= nowait
? -EAGAIN
: -ENOTBLK
;
7332 /* The callers of this must take lock_extent() */
7333 static struct extent_map
*create_io_em(struct btrfs_inode
*inode
, u64 start
,
7334 u64 len
, u64 orig_start
, u64 block_start
,
7335 u64 block_len
, u64 orig_block_len
,
7336 u64 ram_bytes
, int compress_type
,
7339 struct extent_map
*em
;
7342 ASSERT(type
== BTRFS_ORDERED_PREALLOC
||
7343 type
== BTRFS_ORDERED_COMPRESSED
||
7344 type
== BTRFS_ORDERED_NOCOW
||
7345 type
== BTRFS_ORDERED_REGULAR
);
7347 em
= alloc_extent_map();
7349 return ERR_PTR(-ENOMEM
);
7352 em
->orig_start
= orig_start
;
7354 em
->block_len
= block_len
;
7355 em
->block_start
= block_start
;
7356 em
->orig_block_len
= orig_block_len
;
7357 em
->ram_bytes
= ram_bytes
;
7358 em
->generation
= -1;
7359 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7360 if (type
== BTRFS_ORDERED_PREALLOC
) {
7361 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7362 } else if (type
== BTRFS_ORDERED_COMPRESSED
) {
7363 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
7364 em
->compress_type
= compress_type
;
7367 ret
= btrfs_replace_extent_map_range(inode
, em
, true);
7369 free_extent_map(em
);
7370 return ERR_PTR(ret
);
7373 /* em got 2 refs now, callers needs to do free_extent_map once. */
7378 static int btrfs_get_blocks_direct_write(struct extent_map
**map
,
7379 struct inode
*inode
,
7380 struct btrfs_dio_data
*dio_data
,
7381 u64 start
, u64
*lenp
,
7382 unsigned int iomap_flags
)
7384 const bool nowait
= (iomap_flags
& IOMAP_NOWAIT
);
7385 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7386 struct extent_map
*em
= *map
;
7388 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7389 struct btrfs_block_group
*bg
;
7390 bool can_nocow
= false;
7391 bool space_reserved
= false;
7397 * We don't allocate a new extent in the following cases
7399 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7401 * 2) The extent is marked as PREALLOC. We're good to go here and can
7402 * just use the extent.
7405 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7406 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7407 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7408 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7409 type
= BTRFS_ORDERED_PREALLOC
;
7411 type
= BTRFS_ORDERED_NOCOW
;
7412 len
= min(len
, em
->len
- (start
- em
->start
));
7413 block_start
= em
->block_start
+ (start
- em
->start
);
7415 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7416 &orig_block_len
, &ram_bytes
, false, false) == 1) {
7417 bg
= btrfs_inc_nocow_writers(fs_info
, block_start
);
7425 struct extent_map
*em2
;
7427 /* We can NOCOW, so only need to reserve metadata space. */
7428 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
, len
,
7431 /* Our caller expects us to free the input extent map. */
7432 free_extent_map(em
);
7434 btrfs_dec_nocow_writers(bg
);
7435 if (nowait
&& (ret
== -ENOSPC
|| ret
== -EDQUOT
))
7439 space_reserved
= true;
7441 em2
= btrfs_create_dio_extent(BTRFS_I(inode
), dio_data
, start
, len
,
7442 orig_start
, block_start
,
7443 len
, orig_block_len
,
7445 btrfs_dec_nocow_writers(bg
);
7446 if (type
== BTRFS_ORDERED_PREALLOC
) {
7447 free_extent_map(em
);
7457 dio_data
->nocow_done
= true;
7459 /* Our caller expects us to free the input extent map. */
7460 free_extent_map(em
);
7469 * If we could not allocate data space before locking the file
7470 * range and we can't do a NOCOW write, then we have to fail.
7472 if (!dio_data
->data_space_reserved
) {
7478 * We have to COW and we have already reserved data space before,
7479 * so now we reserve only metadata.
7481 ret
= btrfs_delalloc_reserve_metadata(BTRFS_I(inode
), len
, len
,
7485 space_reserved
= true;
7487 em
= btrfs_new_extent_direct(BTRFS_I(inode
), dio_data
, start
, len
);
7493 len
= min(len
, em
->len
- (start
- em
->start
));
7495 btrfs_delalloc_release_metadata(BTRFS_I(inode
),
7496 prev_len
- len
, true);
7500 * We have created our ordered extent, so we can now release our reservation
7501 * for an outstanding extent.
7503 btrfs_delalloc_release_extents(BTRFS_I(inode
), prev_len
);
7506 * Need to update the i_size under the extent lock so buffered
7507 * readers will get the updated i_size when we unlock.
7509 if (start
+ len
> i_size_read(inode
))
7510 i_size_write(inode
, start
+ len
);
7512 if (ret
&& space_reserved
) {
7513 btrfs_delalloc_release_extents(BTRFS_I(inode
), len
);
7514 btrfs_delalloc_release_metadata(BTRFS_I(inode
), len
, true);
7520 static int btrfs_dio_iomap_begin(struct inode
*inode
, loff_t start
,
7521 loff_t length
, unsigned int flags
, struct iomap
*iomap
,
7522 struct iomap
*srcmap
)
7524 struct iomap_iter
*iter
= container_of(iomap
, struct iomap_iter
, iomap
);
7525 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
7526 struct extent_map
*em
;
7527 struct extent_state
*cached_state
= NULL
;
7528 struct btrfs_dio_data
*dio_data
= iter
->private;
7529 u64 lockstart
, lockend
;
7530 const bool write
= !!(flags
& IOMAP_WRITE
);
7533 const u64 data_alloc_len
= length
;
7534 bool unlock_extents
= false;
7537 * We could potentially fault if we have a buffer > PAGE_SIZE, and if
7538 * we're NOWAIT we may submit a bio for a partial range and return
7539 * EIOCBQUEUED, which would result in an errant short read.
7541 * The best way to handle this would be to allow for partial completions
7542 * of iocb's, so we could submit the partial bio, return and fault in
7543 * the rest of the pages, and then submit the io for the rest of the
7544 * range. However we don't have that currently, so simply return
7545 * -EAGAIN at this point so that the normal path is used.
7547 if (!write
&& (flags
& IOMAP_NOWAIT
) && length
> PAGE_SIZE
)
7551 * Cap the size of reads to that usually seen in buffered I/O as we need
7552 * to allocate a contiguous array for the checksums.
7555 len
= min_t(u64
, len
, fs_info
->sectorsize
* BTRFS_MAX_BIO_SECTORS
);
7558 lockend
= start
+ len
- 1;
7561 * iomap_dio_rw() only does filemap_write_and_wait_range(), which isn't
7562 * enough if we've written compressed pages to this area, so we need to
7563 * flush the dirty pages again to make absolutely sure that any
7564 * outstanding dirty pages are on disk - the first flush only starts
7565 * compression on the data, while keeping the pages locked, so by the
7566 * time the second flush returns we know bios for the compressed pages
7567 * were submitted and finished, and the pages no longer under writeback.
7569 * If we have a NOWAIT request and we have any pages in the range that
7570 * are locked, likely due to compression still in progress, we don't want
7571 * to block on page locks. We also don't want to block on pages marked as
7572 * dirty or under writeback (same as for the non-compression case).
7573 * iomap_dio_rw() did the same check, but after that and before we got
7574 * here, mmap'ed writes may have happened or buffered reads started
7575 * (readpage() and readahead(), which lock pages), as we haven't locked
7576 * the file range yet.
7578 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
7579 &BTRFS_I(inode
)->runtime_flags
)) {
7580 if (flags
& IOMAP_NOWAIT
) {
7581 if (filemap_range_needs_writeback(inode
->i_mapping
,
7582 lockstart
, lockend
))
7585 ret
= filemap_fdatawrite_range(inode
->i_mapping
, start
,
7586 start
+ length
- 1);
7592 memset(dio_data
, 0, sizeof(*dio_data
));
7595 * We always try to allocate data space and must do it before locking
7596 * the file range, to avoid deadlocks with concurrent writes to the same
7597 * range if the range has several extents and the writes don't expand the
7598 * current i_size (the inode lock is taken in shared mode). If we fail to
7599 * allocate data space here we continue and later, after locking the
7600 * file range, we fail with ENOSPC only if we figure out we can not do a
7603 if (write
&& !(flags
& IOMAP_NOWAIT
)) {
7604 ret
= btrfs_check_data_free_space(BTRFS_I(inode
),
7605 &dio_data
->data_reserved
,
7606 start
, data_alloc_len
, false);
7608 dio_data
->data_space_reserved
= true;
7609 else if (ret
&& !(BTRFS_I(inode
)->flags
&
7610 (BTRFS_INODE_NODATACOW
| BTRFS_INODE_PREALLOC
)))
7615 * If this errors out it's because we couldn't invalidate pagecache for
7616 * this range and we need to fallback to buffered IO, or we are doing a
7617 * NOWAIT read/write and we need to block.
7619 ret
= lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
, flags
);
7623 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
);
7630 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7631 * io. INLINE is special, and we could probably kludge it in here, but
7632 * it's still buffered so for safety lets just fall back to the generic
7635 * For COMPRESSED we _have_ to read the entire extent in so we can
7636 * decompress it, so there will be buffering required no matter what we
7637 * do, so go ahead and fallback to buffered.
7639 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7640 * to buffered IO. Don't blame me, this is the price we pay for using
7643 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7644 em
->block_start
== EXTENT_MAP_INLINE
) {
7645 free_extent_map(em
);
7647 * If we are in a NOWAIT context, return -EAGAIN in order to
7648 * fallback to buffered IO. This is not only because we can
7649 * block with buffered IO (no support for NOWAIT semantics at
7650 * the moment) but also to avoid returning short reads to user
7651 * space - this happens if we were able to read some data from
7652 * previous non-compressed extents and then when we fallback to
7653 * buffered IO, at btrfs_file_read_iter() by calling
7654 * filemap_read(), we fail to fault in pages for the read buffer,
7655 * in which case filemap_read() returns a short read (the number
7656 * of bytes previously read is > 0, so it does not return -EFAULT).
7658 ret
= (flags
& IOMAP_NOWAIT
) ? -EAGAIN
: -ENOTBLK
;
7662 len
= min(len
, em
->len
- (start
- em
->start
));
7665 * If we have a NOWAIT request and the range contains multiple extents
7666 * (or a mix of extents and holes), then we return -EAGAIN to make the
7667 * caller fallback to a context where it can do a blocking (without
7668 * NOWAIT) request. This way we avoid doing partial IO and returning
7669 * success to the caller, which is not optimal for writes and for reads
7670 * it can result in unexpected behaviour for an application.
7672 * When doing a read, because we use IOMAP_DIO_PARTIAL when calling
7673 * iomap_dio_rw(), we can end up returning less data then what the caller
7674 * asked for, resulting in an unexpected, and incorrect, short read.
7675 * That is, the caller asked to read N bytes and we return less than that,
7676 * which is wrong unless we are crossing EOF. This happens if we get a
7677 * page fault error when trying to fault in pages for the buffer that is
7678 * associated to the struct iov_iter passed to iomap_dio_rw(), and we
7679 * have previously submitted bios for other extents in the range, in
7680 * which case iomap_dio_rw() may return us EIOCBQUEUED if not all of
7681 * those bios have completed by the time we get the page fault error,
7682 * which we return back to our caller - we should only return EIOCBQUEUED
7683 * after we have submitted bios for all the extents in the range.
7685 if ((flags
& IOMAP_NOWAIT
) && len
< length
) {
7686 free_extent_map(em
);
7692 ret
= btrfs_get_blocks_direct_write(&em
, inode
, dio_data
,
7693 start
, &len
, flags
);
7696 unlock_extents
= true;
7697 /* Recalc len in case the new em is smaller than requested */
7698 len
= min(len
, em
->len
- (start
- em
->start
));
7699 if (dio_data
->data_space_reserved
) {
7701 u64 release_len
= 0;
7703 if (dio_data
->nocow_done
) {
7704 release_offset
= start
;
7705 release_len
= data_alloc_len
;
7706 } else if (len
< data_alloc_len
) {
7707 release_offset
= start
+ len
;
7708 release_len
= data_alloc_len
- len
;
7711 if (release_len
> 0)
7712 btrfs_free_reserved_data_space(BTRFS_I(inode
),
7713 dio_data
->data_reserved
,
7719 * We need to unlock only the end area that we aren't using.
7720 * The rest is going to be unlocked by the endio routine.
7722 lockstart
= start
+ len
;
7723 if (lockstart
< lockend
)
7724 unlock_extents
= true;
7728 unlock_extent(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7731 free_extent_state(cached_state
);
7734 * Translate extent map information to iomap.
7735 * We trim the extents (and move the addr) even though iomap code does
7736 * that, since we have locked only the parts we are performing I/O in.
7738 if ((em
->block_start
== EXTENT_MAP_HOLE
) ||
7739 (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) && !write
)) {
7740 iomap
->addr
= IOMAP_NULL_ADDR
;
7741 iomap
->type
= IOMAP_HOLE
;
7743 iomap
->addr
= em
->block_start
+ (start
- em
->start
);
7744 iomap
->type
= IOMAP_MAPPED
;
7746 iomap
->offset
= start
;
7747 iomap
->bdev
= fs_info
->fs_devices
->latest_dev
->bdev
;
7748 iomap
->length
= len
;
7749 free_extent_map(em
);
7754 unlock_extent(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7757 if (dio_data
->data_space_reserved
) {
7758 btrfs_free_reserved_data_space(BTRFS_I(inode
),
7759 dio_data
->data_reserved
,
7760 start
, data_alloc_len
);
7761 extent_changeset_free(dio_data
->data_reserved
);
7767 static int btrfs_dio_iomap_end(struct inode
*inode
, loff_t pos
, loff_t length
,
7768 ssize_t written
, unsigned int flags
, struct iomap
*iomap
)
7770 struct iomap_iter
*iter
= container_of(iomap
, struct iomap_iter
, iomap
);
7771 struct btrfs_dio_data
*dio_data
= iter
->private;
7772 size_t submitted
= dio_data
->submitted
;
7773 const bool write
= !!(flags
& IOMAP_WRITE
);
7776 if (!write
&& (iomap
->type
== IOMAP_HOLE
)) {
7777 /* If reading from a hole, unlock and return */
7778 unlock_extent(&BTRFS_I(inode
)->io_tree
, pos
, pos
+ length
- 1,
7783 if (submitted
< length
) {
7785 length
-= submitted
;
7787 btrfs_mark_ordered_io_finished(BTRFS_I(inode
), NULL
,
7788 pos
, length
, false);
7790 unlock_extent(&BTRFS_I(inode
)->io_tree
, pos
,
7791 pos
+ length
- 1, NULL
);
7795 btrfs_put_ordered_extent(dio_data
->ordered
);
7796 dio_data
->ordered
= NULL
;
7800 extent_changeset_free(dio_data
->data_reserved
);
7804 static void btrfs_dio_end_io(struct btrfs_bio
*bbio
)
7806 struct btrfs_dio_private
*dip
=
7807 container_of(bbio
, struct btrfs_dio_private
, bbio
);
7808 struct btrfs_inode
*inode
= bbio
->inode
;
7809 struct bio
*bio
= &bbio
->bio
;
7811 if (bio
->bi_status
) {
7812 btrfs_warn(inode
->root
->fs_info
,
7813 "direct IO failed ino %llu op 0x%0x offset %#llx len %u err no %d",
7814 btrfs_ino(inode
), bio
->bi_opf
,
7815 dip
->file_offset
, dip
->bytes
, bio
->bi_status
);
7818 if (btrfs_op(bio
) == BTRFS_MAP_WRITE
)
7819 btrfs_mark_ordered_io_finished(inode
, NULL
, dip
->file_offset
,
7820 dip
->bytes
, !bio
->bi_status
);
7822 unlock_extent(&inode
->io_tree
, dip
->file_offset
,
7823 dip
->file_offset
+ dip
->bytes
- 1, NULL
);
7825 bbio
->bio
.bi_private
= bbio
->private;
7826 iomap_dio_bio_end_io(bio
);
7829 static void btrfs_dio_submit_io(const struct iomap_iter
*iter
, struct bio
*bio
,
7832 struct btrfs_bio
*bbio
= btrfs_bio(bio
);
7833 struct btrfs_dio_private
*dip
=
7834 container_of(bbio
, struct btrfs_dio_private
, bbio
);
7835 struct btrfs_dio_data
*dio_data
= iter
->private;
7837 btrfs_bio_init(bbio
, BTRFS_I(iter
->inode
)->root
->fs_info
,
7838 btrfs_dio_end_io
, bio
->bi_private
);
7839 bbio
->inode
= BTRFS_I(iter
->inode
);
7840 bbio
->file_offset
= file_offset
;
7842 dip
->file_offset
= file_offset
;
7843 dip
->bytes
= bio
->bi_iter
.bi_size
;
7845 dio_data
->submitted
+= bio
->bi_iter
.bi_size
;
7848 * Check if we are doing a partial write. If we are, we need to split
7849 * the ordered extent to match the submitted bio. Hang on to the
7850 * remaining unfinishable ordered_extent in dio_data so that it can be
7851 * cancelled in iomap_end to avoid a deadlock wherein faulting the
7852 * remaining pages is blocked on the outstanding ordered extent.
7854 if (iter
->flags
& IOMAP_WRITE
) {
7857 ret
= btrfs_extract_ordered_extent(bbio
, dio_data
->ordered
);
7859 btrfs_bio_end_io(bbio
, errno_to_blk_status(ret
));
7864 btrfs_submit_bio(bbio
, 0);
7867 static const struct iomap_ops btrfs_dio_iomap_ops
= {
7868 .iomap_begin
= btrfs_dio_iomap_begin
,
7869 .iomap_end
= btrfs_dio_iomap_end
,
7872 static const struct iomap_dio_ops btrfs_dio_ops
= {
7873 .submit_io
= btrfs_dio_submit_io
,
7874 .bio_set
= &btrfs_dio_bioset
,
7877 ssize_t
btrfs_dio_read(struct kiocb
*iocb
, struct iov_iter
*iter
, size_t done_before
)
7879 struct btrfs_dio_data data
= { 0 };
7881 return iomap_dio_rw(iocb
, iter
, &btrfs_dio_iomap_ops
, &btrfs_dio_ops
,
7882 IOMAP_DIO_PARTIAL
, &data
, done_before
);
7885 struct iomap_dio
*btrfs_dio_write(struct kiocb
*iocb
, struct iov_iter
*iter
,
7888 struct btrfs_dio_data data
= { 0 };
7890 return __iomap_dio_rw(iocb
, iter
, &btrfs_dio_iomap_ops
, &btrfs_dio_ops
,
7891 IOMAP_DIO_PARTIAL
, &data
, done_before
);
7894 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
7899 ret
= fiemap_prep(inode
, fieinfo
, start
, &len
, 0);
7904 * fiemap_prep() called filemap_write_and_wait() for the whole possible
7905 * file range (0 to LLONG_MAX), but that is not enough if we have
7906 * compression enabled. The first filemap_fdatawrite_range() only kicks
7907 * in the compression of data (in an async thread) and will return
7908 * before the compression is done and writeback is started. A second
7909 * filemap_fdatawrite_range() is needed to wait for the compression to
7910 * complete and writeback to start. We also need to wait for ordered
7911 * extents to complete, because our fiemap implementation uses mainly
7912 * file extent items to list the extents, searching for extent maps
7913 * only for file ranges with holes or prealloc extents to figure out
7914 * if we have delalloc in those ranges.
7916 if (fieinfo
->fi_flags
& FIEMAP_FLAG_SYNC
) {
7917 ret
= btrfs_wait_ordered_range(inode
, 0, LLONG_MAX
);
7922 return extent_fiemap(BTRFS_I(inode
), fieinfo
, start
, len
);
7925 static int btrfs_writepages(struct address_space
*mapping
,
7926 struct writeback_control
*wbc
)
7928 return extent_writepages(mapping
, wbc
);
7931 static void btrfs_readahead(struct readahead_control
*rac
)
7933 extent_readahead(rac
);
7937 * For release_folio() and invalidate_folio() we have a race window where
7938 * folio_end_writeback() is called but the subpage spinlock is not yet released.
7939 * If we continue to release/invalidate the page, we could cause use-after-free
7940 * for subpage spinlock. So this function is to spin and wait for subpage
7943 static void wait_subpage_spinlock(struct page
*page
)
7945 struct btrfs_fs_info
*fs_info
= btrfs_sb(page
->mapping
->host
->i_sb
);
7946 struct btrfs_subpage
*subpage
;
7948 if (!btrfs_is_subpage(fs_info
, page
))
7951 ASSERT(PagePrivate(page
) && page
->private);
7952 subpage
= (struct btrfs_subpage
*)page
->private;
7955 * This may look insane as we just acquire the spinlock and release it,
7956 * without doing anything. But we just want to make sure no one is
7957 * still holding the subpage spinlock.
7958 * And since the page is not dirty nor writeback, and we have page
7959 * locked, the only possible way to hold a spinlock is from the endio
7960 * function to clear page writeback.
7962 * Here we just acquire the spinlock so that all existing callers
7963 * should exit and we're safe to release/invalidate the page.
7965 spin_lock_irq(&subpage
->lock
);
7966 spin_unlock_irq(&subpage
->lock
);
7969 static bool __btrfs_release_folio(struct folio
*folio
, gfp_t gfp_flags
)
7971 int ret
= try_release_extent_mapping(&folio
->page
, gfp_flags
);
7974 wait_subpage_spinlock(&folio
->page
);
7975 clear_page_extent_mapped(&folio
->page
);
7980 static bool btrfs_release_folio(struct folio
*folio
, gfp_t gfp_flags
)
7982 if (folio_test_writeback(folio
) || folio_test_dirty(folio
))
7984 return __btrfs_release_folio(folio
, gfp_flags
);
7987 #ifdef CONFIG_MIGRATION
7988 static int btrfs_migrate_folio(struct address_space
*mapping
,
7989 struct folio
*dst
, struct folio
*src
,
7990 enum migrate_mode mode
)
7992 int ret
= filemap_migrate_folio(mapping
, dst
, src
, mode
);
7994 if (ret
!= MIGRATEPAGE_SUCCESS
)
7997 if (folio_test_ordered(src
)) {
7998 folio_clear_ordered(src
);
7999 folio_set_ordered(dst
);
8002 return MIGRATEPAGE_SUCCESS
;
8005 #define btrfs_migrate_folio NULL
8008 static void btrfs_invalidate_folio(struct folio
*folio
, size_t offset
,
8011 struct btrfs_inode
*inode
= BTRFS_I(folio
->mapping
->host
);
8012 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
8013 struct extent_io_tree
*tree
= &inode
->io_tree
;
8014 struct extent_state
*cached_state
= NULL
;
8015 u64 page_start
= folio_pos(folio
);
8016 u64 page_end
= page_start
+ folio_size(folio
) - 1;
8018 int inode_evicting
= inode
->vfs_inode
.i_state
& I_FREEING
;
8021 * We have folio locked so no new ordered extent can be created on this
8022 * page, nor bio can be submitted for this folio.
8024 * But already submitted bio can still be finished on this folio.
8025 * Furthermore, endio function won't skip folio which has Ordered
8026 * (Private2) already cleared, so it's possible for endio and
8027 * invalidate_folio to do the same ordered extent accounting twice
8030 * So here we wait for any submitted bios to finish, so that we won't
8031 * do double ordered extent accounting on the same folio.
8033 folio_wait_writeback(folio
);
8034 wait_subpage_spinlock(&folio
->page
);
8037 * For subpage case, we have call sites like
8038 * btrfs_punch_hole_lock_range() which passes range not aligned to
8040 * If the range doesn't cover the full folio, we don't need to and
8041 * shouldn't clear page extent mapped, as folio->private can still
8042 * record subpage dirty bits for other part of the range.
8044 * For cases that invalidate the full folio even the range doesn't
8045 * cover the full folio, like invalidating the last folio, we're
8046 * still safe to wait for ordered extent to finish.
8048 if (!(offset
== 0 && length
== folio_size(folio
))) {
8049 btrfs_release_folio(folio
, GFP_NOFS
);
8053 if (!inode_evicting
)
8054 lock_extent(tree
, page_start
, page_end
, &cached_state
);
8057 while (cur
< page_end
) {
8058 struct btrfs_ordered_extent
*ordered
;
8061 u32 extra_flags
= 0;
8063 ordered
= btrfs_lookup_first_ordered_range(inode
, cur
,
8064 page_end
+ 1 - cur
);
8066 range_end
= page_end
;
8068 * No ordered extent covering this range, we are safe
8069 * to delete all extent states in the range.
8071 extra_flags
= EXTENT_CLEAR_ALL_BITS
;
8074 if (ordered
->file_offset
> cur
) {
8076 * There is a range between [cur, oe->file_offset) not
8077 * covered by any ordered extent.
8078 * We are safe to delete all extent states, and handle
8079 * the ordered extent in the next iteration.
8081 range_end
= ordered
->file_offset
- 1;
8082 extra_flags
= EXTENT_CLEAR_ALL_BITS
;
8086 range_end
= min(ordered
->file_offset
+ ordered
->num_bytes
- 1,
8088 ASSERT(range_end
+ 1 - cur
< U32_MAX
);
8089 range_len
= range_end
+ 1 - cur
;
8090 if (!btrfs_page_test_ordered(fs_info
, &folio
->page
, cur
, range_len
)) {
8092 * If Ordered (Private2) is cleared, it means endio has
8093 * already been executed for the range.
8094 * We can't delete the extent states as
8095 * btrfs_finish_ordered_io() may still use some of them.
8099 btrfs_page_clear_ordered(fs_info
, &folio
->page
, cur
, range_len
);
8102 * IO on this page will never be started, so we need to account
8103 * for any ordered extents now. Don't clear EXTENT_DELALLOC_NEW
8104 * here, must leave that up for the ordered extent completion.
8106 * This will also unlock the range for incoming
8107 * btrfs_finish_ordered_io().
8109 if (!inode_evicting
)
8110 clear_extent_bit(tree
, cur
, range_end
,
8112 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8113 EXTENT_DEFRAG
, &cached_state
);
8115 spin_lock_irq(&inode
->ordered_tree
.lock
);
8116 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8117 ordered
->truncated_len
= min(ordered
->truncated_len
,
8118 cur
- ordered
->file_offset
);
8119 spin_unlock_irq(&inode
->ordered_tree
.lock
);
8122 * If the ordered extent has finished, we're safe to delete all
8123 * the extent states of the range, otherwise
8124 * btrfs_finish_ordered_io() will get executed by endio for
8125 * other pages, so we can't delete extent states.
8127 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8128 cur
, range_end
+ 1 - cur
)) {
8129 btrfs_finish_ordered_io(ordered
);
8131 * The ordered extent has finished, now we're again
8132 * safe to delete all extent states of the range.
8134 extra_flags
= EXTENT_CLEAR_ALL_BITS
;
8138 btrfs_put_ordered_extent(ordered
);
8140 * Qgroup reserved space handler
8141 * Sector(s) here will be either:
8143 * 1) Already written to disk or bio already finished
8144 * Then its QGROUP_RESERVED bit in io_tree is already cleared.
8145 * Qgroup will be handled by its qgroup_record then.
8146 * btrfs_qgroup_free_data() call will do nothing here.
8148 * 2) Not written to disk yet
8149 * Then btrfs_qgroup_free_data() call will clear the
8150 * QGROUP_RESERVED bit of its io_tree, and free the qgroup
8151 * reserved data space.
8152 * Since the IO will never happen for this page.
8154 btrfs_qgroup_free_data(inode
, NULL
, cur
, range_end
+ 1 - cur
);
8155 if (!inode_evicting
) {
8156 clear_extent_bit(tree
, cur
, range_end
, EXTENT_LOCKED
|
8157 EXTENT_DELALLOC
| EXTENT_UPTODATE
|
8158 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
|
8159 extra_flags
, &cached_state
);
8161 cur
= range_end
+ 1;
8164 * We have iterated through all ordered extents of the page, the page
8165 * should not have Ordered (Private2) anymore, or the above iteration
8166 * did something wrong.
8168 ASSERT(!folio_test_ordered(folio
));
8169 btrfs_page_clear_checked(fs_info
, &folio
->page
, folio_pos(folio
), folio_size(folio
));
8170 if (!inode_evicting
)
8171 __btrfs_release_folio(folio
, GFP_NOFS
);
8172 clear_page_extent_mapped(&folio
->page
);
8176 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8177 * called from a page fault handler when a page is first dirtied. Hence we must
8178 * be careful to check for EOF conditions here. We set the page up correctly
8179 * for a written page which means we get ENOSPC checking when writing into
8180 * holes and correct delalloc and unwritten extent mapping on filesystems that
8181 * support these features.
8183 * We are not allowed to take the i_mutex here so we have to play games to
8184 * protect against truncate races as the page could now be beyond EOF. Because
8185 * truncate_setsize() writes the inode size before removing pages, once we have
8186 * the page lock we can determine safely if the page is beyond EOF. If it is not
8187 * beyond EOF, then the page is guaranteed safe against truncation until we
8190 vm_fault_t
btrfs_page_mkwrite(struct vm_fault
*vmf
)
8192 struct page
*page
= vmf
->page
;
8193 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
8194 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
8195 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8196 struct btrfs_ordered_extent
*ordered
;
8197 struct extent_state
*cached_state
= NULL
;
8198 struct extent_changeset
*data_reserved
= NULL
;
8199 unsigned long zero_start
;
8209 reserved_space
= PAGE_SIZE
;
8211 sb_start_pagefault(inode
->i_sb
);
8212 page_start
= page_offset(page
);
8213 page_end
= page_start
+ PAGE_SIZE
- 1;
8217 * Reserving delalloc space after obtaining the page lock can lead to
8218 * deadlock. For example, if a dirty page is locked by this function
8219 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8220 * dirty page write out, then the btrfs_writepages() function could
8221 * end up waiting indefinitely to get a lock on the page currently
8222 * being processed by btrfs_page_mkwrite() function.
8224 ret2
= btrfs_delalloc_reserve_space(BTRFS_I(inode
), &data_reserved
,
8225 page_start
, reserved_space
);
8227 ret2
= file_update_time(vmf
->vma
->vm_file
);
8231 ret
= vmf_error(ret2
);
8237 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8239 down_read(&BTRFS_I(inode
)->i_mmap_lock
);
8241 size
= i_size_read(inode
);
8243 if ((page
->mapping
!= inode
->i_mapping
) ||
8244 (page_start
>= size
)) {
8245 /* page got truncated out from underneath us */
8248 wait_on_page_writeback(page
);
8250 lock_extent(io_tree
, page_start
, page_end
, &cached_state
);
8251 ret2
= set_page_extent_mapped(page
);
8253 ret
= vmf_error(ret2
);
8254 unlock_extent(io_tree
, page_start
, page_end
, &cached_state
);
8259 * we can't set the delalloc bits if there are pending ordered
8260 * extents. Drop our locks and wait for them to finish
8262 ordered
= btrfs_lookup_ordered_range(BTRFS_I(inode
), page_start
,
8265 unlock_extent(io_tree
, page_start
, page_end
, &cached_state
);
8267 up_read(&BTRFS_I(inode
)->i_mmap_lock
);
8268 btrfs_start_ordered_extent(ordered
);
8269 btrfs_put_ordered_extent(ordered
);
8273 if (page
->index
== ((size
- 1) >> PAGE_SHIFT
)) {
8274 reserved_space
= round_up(size
- page_start
,
8275 fs_info
->sectorsize
);
8276 if (reserved_space
< PAGE_SIZE
) {
8277 end
= page_start
+ reserved_space
- 1;
8278 btrfs_delalloc_release_space(BTRFS_I(inode
),
8279 data_reserved
, page_start
,
8280 PAGE_SIZE
- reserved_space
, true);
8285 * page_mkwrite gets called when the page is firstly dirtied after it's
8286 * faulted in, but write(2) could also dirty a page and set delalloc
8287 * bits, thus in this case for space account reason, we still need to
8288 * clear any delalloc bits within this page range since we have to
8289 * reserve data&meta space before lock_page() (see above comments).
8291 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, end
,
8292 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
8293 EXTENT_DEFRAG
, &cached_state
);
8295 ret2
= btrfs_set_extent_delalloc(BTRFS_I(inode
), page_start
, end
, 0,
8298 unlock_extent(io_tree
, page_start
, page_end
, &cached_state
);
8299 ret
= VM_FAULT_SIGBUS
;
8303 /* page is wholly or partially inside EOF */
8304 if (page_start
+ PAGE_SIZE
> size
)
8305 zero_start
= offset_in_page(size
);
8307 zero_start
= PAGE_SIZE
;
8309 if (zero_start
!= PAGE_SIZE
)
8310 memzero_page(page
, zero_start
, PAGE_SIZE
- zero_start
);
8312 btrfs_page_clear_checked(fs_info
, page
, page_start
, PAGE_SIZE
);
8313 btrfs_page_set_dirty(fs_info
, page
, page_start
, end
+ 1 - page_start
);
8314 btrfs_page_set_uptodate(fs_info
, page
, page_start
, end
+ 1 - page_start
);
8316 btrfs_set_inode_last_sub_trans(BTRFS_I(inode
));
8318 unlock_extent(io_tree
, page_start
, page_end
, &cached_state
);
8319 up_read(&BTRFS_I(inode
)->i_mmap_lock
);
8321 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
);
8322 sb_end_pagefault(inode
->i_sb
);
8323 extent_changeset_free(data_reserved
);
8324 return VM_FAULT_LOCKED
;
8328 up_read(&BTRFS_I(inode
)->i_mmap_lock
);
8330 btrfs_delalloc_release_extents(BTRFS_I(inode
), PAGE_SIZE
);
8331 btrfs_delalloc_release_space(BTRFS_I(inode
), data_reserved
, page_start
,
8332 reserved_space
, (ret
!= 0));
8334 sb_end_pagefault(inode
->i_sb
);
8335 extent_changeset_free(data_reserved
);
8339 static int btrfs_truncate(struct btrfs_inode
*inode
, bool skip_writeback
)
8341 struct btrfs_truncate_control control
= {
8343 .ino
= btrfs_ino(inode
),
8344 .min_type
= BTRFS_EXTENT_DATA_KEY
,
8345 .clear_extent_range
= true,
8347 struct btrfs_root
*root
= inode
->root
;
8348 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
8349 struct btrfs_block_rsv
*rsv
;
8351 struct btrfs_trans_handle
*trans
;
8352 u64 mask
= fs_info
->sectorsize
- 1;
8353 u64 min_size
= btrfs_calc_metadata_size(fs_info
, 1);
8355 if (!skip_writeback
) {
8356 ret
= btrfs_wait_ordered_range(&inode
->vfs_inode
,
8357 inode
->vfs_inode
.i_size
& (~mask
),
8364 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8365 * things going on here:
8367 * 1) We need to reserve space to update our inode.
8369 * 2) We need to have something to cache all the space that is going to
8370 * be free'd up by the truncate operation, but also have some slack
8371 * space reserved in case it uses space during the truncate (thank you
8372 * very much snapshotting).
8374 * And we need these to be separate. The fact is we can use a lot of
8375 * space doing the truncate, and we have no earthly idea how much space
8376 * we will use, so we need the truncate reservation to be separate so it
8377 * doesn't end up using space reserved for updating the inode. We also
8378 * need to be able to stop the transaction and start a new one, which
8379 * means we need to be able to update the inode several times, and we
8380 * have no idea of knowing how many times that will be, so we can't just
8381 * reserve 1 item for the entirety of the operation, so that has to be
8382 * done separately as well.
8384 * So that leaves us with
8386 * 1) rsv - for the truncate reservation, which we will steal from the
8387 * transaction reservation.
8388 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8389 * updating the inode.
8391 rsv
= btrfs_alloc_block_rsv(fs_info
, BTRFS_BLOCK_RSV_TEMP
);
8394 rsv
->size
= min_size
;
8395 rsv
->failfast
= true;
8398 * 1 for the truncate slack space
8399 * 1 for updating the inode.
8401 trans
= btrfs_start_transaction(root
, 2);
8402 if (IS_ERR(trans
)) {
8403 ret
= PTR_ERR(trans
);
8407 /* Migrate the slack space for the truncate to our reserve */
8408 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
, rsv
,
8412 trans
->block_rsv
= rsv
;
8415 struct extent_state
*cached_state
= NULL
;
8416 const u64 new_size
= inode
->vfs_inode
.i_size
;
8417 const u64 lock_start
= ALIGN_DOWN(new_size
, fs_info
->sectorsize
);
8419 control
.new_size
= new_size
;
8420 lock_extent(&inode
->io_tree
, lock_start
, (u64
)-1, &cached_state
);
8422 * We want to drop from the next block forward in case this new
8423 * size is not block aligned since we will be keeping the last
8424 * block of the extent just the way it is.
8426 btrfs_drop_extent_map_range(inode
,
8427 ALIGN(new_size
, fs_info
->sectorsize
),
8430 ret
= btrfs_truncate_inode_items(trans
, root
, &control
);
8432 inode_sub_bytes(&inode
->vfs_inode
, control
.sub_bytes
);
8433 btrfs_inode_safe_disk_i_size_write(inode
, control
.last_size
);
8435 unlock_extent(&inode
->io_tree
, lock_start
, (u64
)-1, &cached_state
);
8437 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
8438 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
8441 ret
= btrfs_update_inode(trans
, root
, inode
);
8445 btrfs_end_transaction(trans
);
8446 btrfs_btree_balance_dirty(fs_info
);
8448 trans
= btrfs_start_transaction(root
, 2);
8449 if (IS_ERR(trans
)) {
8450 ret
= PTR_ERR(trans
);
8455 btrfs_block_rsv_release(fs_info
, rsv
, -1, NULL
);
8456 ret
= btrfs_block_rsv_migrate(&fs_info
->trans_block_rsv
,
8457 rsv
, min_size
, false);
8458 BUG_ON(ret
); /* shouldn't happen */
8459 trans
->block_rsv
= rsv
;
8463 * We can't call btrfs_truncate_block inside a trans handle as we could
8464 * deadlock with freeze, if we got BTRFS_NEED_TRUNCATE_BLOCK then we
8465 * know we've truncated everything except the last little bit, and can
8466 * do btrfs_truncate_block and then update the disk_i_size.
8468 if (ret
== BTRFS_NEED_TRUNCATE_BLOCK
) {
8469 btrfs_end_transaction(trans
);
8470 btrfs_btree_balance_dirty(fs_info
);
8472 ret
= btrfs_truncate_block(inode
, inode
->vfs_inode
.i_size
, 0, 0);
8475 trans
= btrfs_start_transaction(root
, 1);
8476 if (IS_ERR(trans
)) {
8477 ret
= PTR_ERR(trans
);
8480 btrfs_inode_safe_disk_i_size_write(inode
, 0);
8486 trans
->block_rsv
= &fs_info
->trans_block_rsv
;
8487 ret2
= btrfs_update_inode(trans
, root
, inode
);
8491 ret2
= btrfs_end_transaction(trans
);
8494 btrfs_btree_balance_dirty(fs_info
);
8497 btrfs_free_block_rsv(fs_info
, rsv
);
8499 * So if we truncate and then write and fsync we normally would just
8500 * write the extents that changed, which is a problem if we need to
8501 * first truncate that entire inode. So set this flag so we write out
8502 * all of the extents in the inode to the sync log so we're completely
8505 * If no extents were dropped or trimmed we don't need to force the next
8506 * fsync to truncate all the inode's items from the log and re-log them
8507 * all. This means the truncate operation did not change the file size,
8508 * or changed it to a smaller size but there was only an implicit hole
8509 * between the old i_size and the new i_size, and there were no prealloc
8510 * extents beyond i_size to drop.
8512 if (control
.extents_found
> 0)
8513 btrfs_set_inode_full_sync(inode
);
8518 struct inode
*btrfs_new_subvol_inode(struct mnt_idmap
*idmap
,
8521 struct inode
*inode
;
8523 inode
= new_inode(dir
->i_sb
);
8526 * Subvolumes don't inherit the sgid bit or the parent's gid if
8527 * the parent's sgid bit is set. This is probably a bug.
8529 inode_init_owner(idmap
, inode
, NULL
,
8530 S_IFDIR
| (~current_umask() & S_IRWXUGO
));
8531 inode
->i_op
= &btrfs_dir_inode_operations
;
8532 inode
->i_fop
= &btrfs_dir_file_operations
;
8537 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
8539 struct btrfs_fs_info
*fs_info
= btrfs_sb(sb
);
8540 struct btrfs_inode
*ei
;
8541 struct inode
*inode
;
8543 ei
= alloc_inode_sb(sb
, btrfs_inode_cachep
, GFP_KERNEL
);
8550 ei
->last_sub_trans
= 0;
8551 ei
->logged_trans
= 0;
8552 ei
->delalloc_bytes
= 0;
8553 ei
->new_delalloc_bytes
= 0;
8554 ei
->defrag_bytes
= 0;
8555 ei
->disk_i_size
= 0;
8559 ei
->index_cnt
= (u64
)-1;
8561 ei
->last_unlink_trans
= 0;
8562 ei
->last_reflink_trans
= 0;
8563 ei
->last_log_commit
= 0;
8565 spin_lock_init(&ei
->lock
);
8566 ei
->outstanding_extents
= 0;
8567 if (sb
->s_magic
!= BTRFS_TEST_MAGIC
)
8568 btrfs_init_metadata_block_rsv(fs_info
, &ei
->block_rsv
,
8569 BTRFS_BLOCK_RSV_DELALLOC
);
8570 ei
->runtime_flags
= 0;
8571 ei
->prop_compress
= BTRFS_COMPRESS_NONE
;
8572 ei
->defrag_compress
= BTRFS_COMPRESS_NONE
;
8574 ei
->delayed_node
= NULL
;
8576 ei
->i_otime
.tv_sec
= 0;
8577 ei
->i_otime
.tv_nsec
= 0;
8579 inode
= &ei
->vfs_inode
;
8580 extent_map_tree_init(&ei
->extent_tree
);
8581 extent_io_tree_init(fs_info
, &ei
->io_tree
, IO_TREE_INODE_IO
);
8582 ei
->io_tree
.inode
= ei
;
8583 extent_io_tree_init(fs_info
, &ei
->file_extent_tree
,
8584 IO_TREE_INODE_FILE_EXTENT
);
8585 mutex_init(&ei
->log_mutex
);
8586 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
8587 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
8588 INIT_LIST_HEAD(&ei
->delayed_iput
);
8589 RB_CLEAR_NODE(&ei
->rb_node
);
8590 init_rwsem(&ei
->i_mmap_lock
);
8595 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8596 void btrfs_test_destroy_inode(struct inode
*inode
)
8598 btrfs_drop_extent_map_range(BTRFS_I(inode
), 0, (u64
)-1, false);
8599 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
8603 void btrfs_free_inode(struct inode
*inode
)
8605 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
8608 void btrfs_destroy_inode(struct inode
*vfs_inode
)
8610 struct btrfs_ordered_extent
*ordered
;
8611 struct btrfs_inode
*inode
= BTRFS_I(vfs_inode
);
8612 struct btrfs_root
*root
= inode
->root
;
8613 bool freespace_inode
;
8615 WARN_ON(!hlist_empty(&vfs_inode
->i_dentry
));
8616 WARN_ON(vfs_inode
->i_data
.nrpages
);
8617 WARN_ON(inode
->block_rsv
.reserved
);
8618 WARN_ON(inode
->block_rsv
.size
);
8619 WARN_ON(inode
->outstanding_extents
);
8620 if (!S_ISDIR(vfs_inode
->i_mode
)) {
8621 WARN_ON(inode
->delalloc_bytes
);
8622 WARN_ON(inode
->new_delalloc_bytes
);
8624 WARN_ON(inode
->csum_bytes
);
8625 WARN_ON(inode
->defrag_bytes
);
8628 * This can happen where we create an inode, but somebody else also
8629 * created the same inode and we need to destroy the one we already
8636 * If this is a free space inode do not take the ordered extents lockdep
8639 freespace_inode
= btrfs_is_free_space_inode(inode
);
8642 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
8646 btrfs_err(root
->fs_info
,
8647 "found ordered extent %llu %llu on inode cleanup",
8648 ordered
->file_offset
, ordered
->num_bytes
);
8650 if (!freespace_inode
)
8651 btrfs_lockdep_acquire(root
->fs_info
, btrfs_ordered_extent
);
8653 btrfs_remove_ordered_extent(inode
, ordered
);
8654 btrfs_put_ordered_extent(ordered
);
8655 btrfs_put_ordered_extent(ordered
);
8658 btrfs_qgroup_check_reserved_leak(inode
);
8659 inode_tree_del(inode
);
8660 btrfs_drop_extent_map_range(inode
, 0, (u64
)-1, false);
8661 btrfs_inode_clear_file_extent_range(inode
, 0, (u64
)-1);
8662 btrfs_put_root(inode
->root
);
8665 int btrfs_drop_inode(struct inode
*inode
)
8667 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8672 /* the snap/subvol tree is on deleting */
8673 if (btrfs_root_refs(&root
->root_item
) == 0)
8676 return generic_drop_inode(inode
);
8679 static void init_once(void *foo
)
8681 struct btrfs_inode
*ei
= foo
;
8683 inode_init_once(&ei
->vfs_inode
);
8686 void __cold
btrfs_destroy_cachep(void)
8689 * Make sure all delayed rcu free inodes are flushed before we
8693 bioset_exit(&btrfs_dio_bioset
);
8694 kmem_cache_destroy(btrfs_inode_cachep
);
8697 int __init
btrfs_init_cachep(void)
8699 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
8700 sizeof(struct btrfs_inode
), 0,
8701 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
| SLAB_ACCOUNT
,
8703 if (!btrfs_inode_cachep
)
8706 if (bioset_init(&btrfs_dio_bioset
, BIO_POOL_SIZE
,
8707 offsetof(struct btrfs_dio_private
, bbio
.bio
),
8713 btrfs_destroy_cachep();
8717 static int btrfs_getattr(struct mnt_idmap
*idmap
,
8718 const struct path
*path
, struct kstat
*stat
,
8719 u32 request_mask
, unsigned int flags
)
8723 struct inode
*inode
= d_inode(path
->dentry
);
8724 u32 blocksize
= inode
->i_sb
->s_blocksize
;
8725 u32 bi_flags
= BTRFS_I(inode
)->flags
;
8726 u32 bi_ro_flags
= BTRFS_I(inode
)->ro_flags
;
8728 stat
->result_mask
|= STATX_BTIME
;
8729 stat
->btime
.tv_sec
= BTRFS_I(inode
)->i_otime
.tv_sec
;
8730 stat
->btime
.tv_nsec
= BTRFS_I(inode
)->i_otime
.tv_nsec
;
8731 if (bi_flags
& BTRFS_INODE_APPEND
)
8732 stat
->attributes
|= STATX_ATTR_APPEND
;
8733 if (bi_flags
& BTRFS_INODE_COMPRESS
)
8734 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
8735 if (bi_flags
& BTRFS_INODE_IMMUTABLE
)
8736 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
8737 if (bi_flags
& BTRFS_INODE_NODUMP
)
8738 stat
->attributes
|= STATX_ATTR_NODUMP
;
8739 if (bi_ro_flags
& BTRFS_INODE_RO_VERITY
)
8740 stat
->attributes
|= STATX_ATTR_VERITY
;
8742 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
8743 STATX_ATTR_COMPRESSED
|
8744 STATX_ATTR_IMMUTABLE
|
8747 generic_fillattr(idmap
, inode
, stat
);
8748 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
8750 spin_lock(&BTRFS_I(inode
)->lock
);
8751 delalloc_bytes
= BTRFS_I(inode
)->new_delalloc_bytes
;
8752 inode_bytes
= inode_get_bytes(inode
);
8753 spin_unlock(&BTRFS_I(inode
)->lock
);
8754 stat
->blocks
= (ALIGN(inode_bytes
, blocksize
) +
8755 ALIGN(delalloc_bytes
, blocksize
)) >> SECTOR_SHIFT
;
8759 static int btrfs_rename_exchange(struct inode
*old_dir
,
8760 struct dentry
*old_dentry
,
8761 struct inode
*new_dir
,
8762 struct dentry
*new_dentry
)
8764 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
8765 struct btrfs_trans_handle
*trans
;
8766 unsigned int trans_num_items
;
8767 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
8768 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
8769 struct inode
*new_inode
= new_dentry
->d_inode
;
8770 struct inode
*old_inode
= old_dentry
->d_inode
;
8771 struct timespec64 ctime
= current_time(old_inode
);
8772 struct btrfs_rename_ctx old_rename_ctx
;
8773 struct btrfs_rename_ctx new_rename_ctx
;
8774 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
8775 u64 new_ino
= btrfs_ino(BTRFS_I(new_inode
));
8780 bool need_abort
= false;
8781 struct fscrypt_name old_fname
, new_fname
;
8782 struct fscrypt_str
*old_name
, *new_name
;
8785 * For non-subvolumes allow exchange only within one subvolume, in the
8786 * same inode namespace. Two subvolumes (represented as directory) can
8787 * be exchanged as they're a logical link and have a fixed inode number.
8790 (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
||
8791 new_ino
!= BTRFS_FIRST_FREE_OBJECTID
))
8794 ret
= fscrypt_setup_filename(old_dir
, &old_dentry
->d_name
, 0, &old_fname
);
8798 ret
= fscrypt_setup_filename(new_dir
, &new_dentry
->d_name
, 0, &new_fname
);
8800 fscrypt_free_filename(&old_fname
);
8804 old_name
= &old_fname
.disk_name
;
8805 new_name
= &new_fname
.disk_name
;
8807 /* close the race window with snapshot create/destroy ioctl */
8808 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
||
8809 new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8810 down_read(&fs_info
->subvol_sem
);
8814 * 1 to remove old dir item
8815 * 1 to remove old dir index
8816 * 1 to add new dir item
8817 * 1 to add new dir index
8818 * 1 to update parent inode
8820 * If the parents are the same, we only need to account for one
8822 trans_num_items
= (old_dir
== new_dir
? 9 : 10);
8823 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
8825 * 1 to remove old root ref
8826 * 1 to remove old root backref
8827 * 1 to add new root ref
8828 * 1 to add new root backref
8830 trans_num_items
+= 4;
8833 * 1 to update inode item
8834 * 1 to remove old inode ref
8835 * 1 to add new inode ref
8837 trans_num_items
+= 3;
8839 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8840 trans_num_items
+= 4;
8842 trans_num_items
+= 3;
8843 trans
= btrfs_start_transaction(root
, trans_num_items
);
8844 if (IS_ERR(trans
)) {
8845 ret
= PTR_ERR(trans
);
8850 ret
= btrfs_record_root_in_trans(trans
, dest
);
8856 * We need to find a free sequence number both in the source and
8857 * in the destination directory for the exchange.
8859 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &old_idx
);
8862 ret
= btrfs_set_inode_index(BTRFS_I(old_dir
), &new_idx
);
8866 BTRFS_I(old_inode
)->dir_index
= 0ULL;
8867 BTRFS_I(new_inode
)->dir_index
= 0ULL;
8869 /* Reference for the source. */
8870 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
8871 /* force full log commit if subvolume involved. */
8872 btrfs_set_log_full_commit(trans
);
8874 ret
= btrfs_insert_inode_ref(trans
, dest
, new_name
, old_ino
,
8875 btrfs_ino(BTRFS_I(new_dir
)),
8882 /* And now for the dest. */
8883 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
8884 /* force full log commit if subvolume involved. */
8885 btrfs_set_log_full_commit(trans
);
8887 ret
= btrfs_insert_inode_ref(trans
, root
, old_name
, new_ino
,
8888 btrfs_ino(BTRFS_I(old_dir
)),
8892 btrfs_abort_transaction(trans
, ret
);
8897 /* Update inode version and ctime/mtime. */
8898 inode_inc_iversion(old_dir
);
8899 inode_inc_iversion(new_dir
);
8900 inode_inc_iversion(old_inode
);
8901 inode_inc_iversion(new_inode
);
8902 old_dir
->i_mtime
= ctime
;
8903 old_dir
->i_ctime
= ctime
;
8904 new_dir
->i_mtime
= ctime
;
8905 new_dir
->i_ctime
= ctime
;
8906 old_inode
->i_ctime
= ctime
;
8907 new_inode
->i_ctime
= ctime
;
8909 if (old_dentry
->d_parent
!= new_dentry
->d_parent
) {
8910 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
8911 BTRFS_I(old_inode
), true);
8912 btrfs_record_unlink_dir(trans
, BTRFS_I(new_dir
),
8913 BTRFS_I(new_inode
), true);
8916 /* src is a subvolume */
8917 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
8918 ret
= btrfs_unlink_subvol(trans
, BTRFS_I(old_dir
), old_dentry
);
8919 } else { /* src is an inode */
8920 ret
= __btrfs_unlink_inode(trans
, BTRFS_I(old_dir
),
8921 BTRFS_I(old_dentry
->d_inode
),
8922 old_name
, &old_rename_ctx
);
8924 ret
= btrfs_update_inode(trans
, root
, BTRFS_I(old_inode
));
8927 btrfs_abort_transaction(trans
, ret
);
8931 /* dest is a subvolume */
8932 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
8933 ret
= btrfs_unlink_subvol(trans
, BTRFS_I(new_dir
), new_dentry
);
8934 } else { /* dest is an inode */
8935 ret
= __btrfs_unlink_inode(trans
, BTRFS_I(new_dir
),
8936 BTRFS_I(new_dentry
->d_inode
),
8937 new_name
, &new_rename_ctx
);
8939 ret
= btrfs_update_inode(trans
, dest
, BTRFS_I(new_inode
));
8942 btrfs_abort_transaction(trans
, ret
);
8946 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
8947 new_name
, 0, old_idx
);
8949 btrfs_abort_transaction(trans
, ret
);
8953 ret
= btrfs_add_link(trans
, BTRFS_I(old_dir
), BTRFS_I(new_inode
),
8954 old_name
, 0, new_idx
);
8956 btrfs_abort_transaction(trans
, ret
);
8960 if (old_inode
->i_nlink
== 1)
8961 BTRFS_I(old_inode
)->dir_index
= old_idx
;
8962 if (new_inode
->i_nlink
== 1)
8963 BTRFS_I(new_inode
)->dir_index
= new_idx
;
8966 * Now pin the logs of the roots. We do it to ensure that no other task
8967 * can sync the logs while we are in progress with the rename, because
8968 * that could result in an inconsistency in case any of the inodes that
8969 * are part of this rename operation were logged before.
8971 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
8972 btrfs_pin_log_trans(root
);
8973 if (new_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
8974 btrfs_pin_log_trans(dest
);
8976 /* Do the log updates for all inodes. */
8977 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
8978 btrfs_log_new_name(trans
, old_dentry
, BTRFS_I(old_dir
),
8979 old_rename_ctx
.index
, new_dentry
->d_parent
);
8980 if (new_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
8981 btrfs_log_new_name(trans
, new_dentry
, BTRFS_I(new_dir
),
8982 new_rename_ctx
.index
, old_dentry
->d_parent
);
8984 /* Now unpin the logs. */
8985 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
8986 btrfs_end_log_trans(root
);
8987 if (new_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
8988 btrfs_end_log_trans(dest
);
8990 ret2
= btrfs_end_transaction(trans
);
8991 ret
= ret
? ret
: ret2
;
8993 if (new_ino
== BTRFS_FIRST_FREE_OBJECTID
||
8994 old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8995 up_read(&fs_info
->subvol_sem
);
8997 fscrypt_free_filename(&new_fname
);
8998 fscrypt_free_filename(&old_fname
);
9002 static struct inode
*new_whiteout_inode(struct mnt_idmap
*idmap
,
9005 struct inode
*inode
;
9007 inode
= new_inode(dir
->i_sb
);
9009 inode_init_owner(idmap
, inode
, dir
,
9010 S_IFCHR
| WHITEOUT_MODE
);
9011 inode
->i_op
= &btrfs_special_inode_operations
;
9012 init_special_inode(inode
, inode
->i_mode
, WHITEOUT_DEV
);
9017 static int btrfs_rename(struct mnt_idmap
*idmap
,
9018 struct inode
*old_dir
, struct dentry
*old_dentry
,
9019 struct inode
*new_dir
, struct dentry
*new_dentry
,
9022 struct btrfs_fs_info
*fs_info
= btrfs_sb(old_dir
->i_sb
);
9023 struct btrfs_new_inode_args whiteout_args
= {
9025 .dentry
= old_dentry
,
9027 struct btrfs_trans_handle
*trans
;
9028 unsigned int trans_num_items
;
9029 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9030 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9031 struct inode
*new_inode
= d_inode(new_dentry
);
9032 struct inode
*old_inode
= d_inode(old_dentry
);
9033 struct btrfs_rename_ctx rename_ctx
;
9037 u64 old_ino
= btrfs_ino(BTRFS_I(old_inode
));
9038 struct fscrypt_name old_fname
, new_fname
;
9040 if (btrfs_ino(BTRFS_I(new_dir
)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9043 /* we only allow rename subvolume link between subvolumes */
9044 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9047 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9048 (new_inode
&& btrfs_ino(BTRFS_I(new_inode
)) == BTRFS_FIRST_FREE_OBJECTID
))
9051 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9052 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9055 ret
= fscrypt_setup_filename(old_dir
, &old_dentry
->d_name
, 0, &old_fname
);
9059 ret
= fscrypt_setup_filename(new_dir
, &new_dentry
->d_name
, 0, &new_fname
);
9061 fscrypt_free_filename(&old_fname
);
9065 /* check for collisions, even if the name isn't there */
9066 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
, &new_fname
.disk_name
);
9068 if (ret
== -EEXIST
) {
9070 * eexist without a new_inode */
9071 if (WARN_ON(!new_inode
)) {
9072 goto out_fscrypt_names
;
9075 /* maybe -EOVERFLOW */
9076 goto out_fscrypt_names
;
9082 * we're using rename to replace one file with another. Start IO on it
9083 * now so we don't add too much work to the end of the transaction
9085 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9086 filemap_flush(old_inode
->i_mapping
);
9088 if (flags
& RENAME_WHITEOUT
) {
9089 whiteout_args
.inode
= new_whiteout_inode(idmap
, old_dir
);
9090 if (!whiteout_args
.inode
) {
9092 goto out_fscrypt_names
;
9094 ret
= btrfs_new_inode_prepare(&whiteout_args
, &trans_num_items
);
9096 goto out_whiteout_inode
;
9098 /* 1 to update the old parent inode. */
9099 trans_num_items
= 1;
9102 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
9103 /* Close the race window with snapshot create/destroy ioctl */
9104 down_read(&fs_info
->subvol_sem
);
9106 * 1 to remove old root ref
9107 * 1 to remove old root backref
9108 * 1 to add new root ref
9109 * 1 to add new root backref
9111 trans_num_items
+= 4;
9115 * 1 to remove old inode ref
9116 * 1 to add new inode ref
9118 trans_num_items
+= 3;
9121 * 1 to remove old dir item
9122 * 1 to remove old dir index
9123 * 1 to add new dir item
9124 * 1 to add new dir index
9126 trans_num_items
+= 4;
9127 /* 1 to update new parent inode if it's not the same as the old parent */
9128 if (new_dir
!= old_dir
)
9133 * 1 to remove inode ref
9134 * 1 to remove dir item
9135 * 1 to remove dir index
9136 * 1 to possibly add orphan item
9138 trans_num_items
+= 5;
9140 trans
= btrfs_start_transaction(root
, trans_num_items
);
9141 if (IS_ERR(trans
)) {
9142 ret
= PTR_ERR(trans
);
9147 ret
= btrfs_record_root_in_trans(trans
, dest
);
9152 ret
= btrfs_set_inode_index(BTRFS_I(new_dir
), &index
);
9156 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9157 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9158 /* force full log commit if subvolume involved. */
9159 btrfs_set_log_full_commit(trans
);
9161 ret
= btrfs_insert_inode_ref(trans
, dest
, &new_fname
.disk_name
,
9162 old_ino
, btrfs_ino(BTRFS_I(new_dir
)),
9168 inode_inc_iversion(old_dir
);
9169 inode_inc_iversion(new_dir
);
9170 inode_inc_iversion(old_inode
);
9171 old_dir
->i_mtime
= current_time(old_dir
);
9172 old_dir
->i_ctime
= old_dir
->i_mtime
;
9173 new_dir
->i_mtime
= old_dir
->i_mtime
;
9174 new_dir
->i_ctime
= old_dir
->i_mtime
;
9175 old_inode
->i_ctime
= old_dir
->i_mtime
;
9177 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9178 btrfs_record_unlink_dir(trans
, BTRFS_I(old_dir
),
9179 BTRFS_I(old_inode
), true);
9181 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9182 ret
= btrfs_unlink_subvol(trans
, BTRFS_I(old_dir
), old_dentry
);
9184 ret
= __btrfs_unlink_inode(trans
, BTRFS_I(old_dir
),
9185 BTRFS_I(d_inode(old_dentry
)),
9186 &old_fname
.disk_name
, &rename_ctx
);
9188 ret
= btrfs_update_inode(trans
, root
, BTRFS_I(old_inode
));
9191 btrfs_abort_transaction(trans
, ret
);
9196 inode_inc_iversion(new_inode
);
9197 new_inode
->i_ctime
= current_time(new_inode
);
9198 if (unlikely(btrfs_ino(BTRFS_I(new_inode
)) ==
9199 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9200 ret
= btrfs_unlink_subvol(trans
, BTRFS_I(new_dir
), new_dentry
);
9201 BUG_ON(new_inode
->i_nlink
== 0);
9203 ret
= btrfs_unlink_inode(trans
, BTRFS_I(new_dir
),
9204 BTRFS_I(d_inode(new_dentry
)),
9205 &new_fname
.disk_name
);
9207 if (!ret
&& new_inode
->i_nlink
== 0)
9208 ret
= btrfs_orphan_add(trans
,
9209 BTRFS_I(d_inode(new_dentry
)));
9211 btrfs_abort_transaction(trans
, ret
);
9216 ret
= btrfs_add_link(trans
, BTRFS_I(new_dir
), BTRFS_I(old_inode
),
9217 &new_fname
.disk_name
, 0, index
);
9219 btrfs_abort_transaction(trans
, ret
);
9223 if (old_inode
->i_nlink
== 1)
9224 BTRFS_I(old_inode
)->dir_index
= index
;
9226 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
)
9227 btrfs_log_new_name(trans
, old_dentry
, BTRFS_I(old_dir
),
9228 rename_ctx
.index
, new_dentry
->d_parent
);
9230 if (flags
& RENAME_WHITEOUT
) {
9231 ret
= btrfs_create_new_inode(trans
, &whiteout_args
);
9233 btrfs_abort_transaction(trans
, ret
);
9236 unlock_new_inode(whiteout_args
.inode
);
9237 iput(whiteout_args
.inode
);
9238 whiteout_args
.inode
= NULL
;
9242 ret2
= btrfs_end_transaction(trans
);
9243 ret
= ret
? ret
: ret2
;
9245 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9246 up_read(&fs_info
->subvol_sem
);
9247 if (flags
& RENAME_WHITEOUT
)
9248 btrfs_new_inode_args_destroy(&whiteout_args
);
9250 if (flags
& RENAME_WHITEOUT
)
9251 iput(whiteout_args
.inode
);
9253 fscrypt_free_filename(&old_fname
);
9254 fscrypt_free_filename(&new_fname
);
9258 static int btrfs_rename2(struct mnt_idmap
*idmap
, struct inode
*old_dir
,
9259 struct dentry
*old_dentry
, struct inode
*new_dir
,
9260 struct dentry
*new_dentry
, unsigned int flags
)
9264 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
9267 if (flags
& RENAME_EXCHANGE
)
9268 ret
= btrfs_rename_exchange(old_dir
, old_dentry
, new_dir
,
9271 ret
= btrfs_rename(idmap
, old_dir
, old_dentry
, new_dir
,
9274 btrfs_btree_balance_dirty(BTRFS_I(new_dir
)->root
->fs_info
);
9279 struct btrfs_delalloc_work
{
9280 struct inode
*inode
;
9281 struct completion completion
;
9282 struct list_head list
;
9283 struct btrfs_work work
;
9286 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9288 struct btrfs_delalloc_work
*delalloc_work
;
9289 struct inode
*inode
;
9291 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9293 inode
= delalloc_work
->inode
;
9294 filemap_flush(inode
->i_mapping
);
9295 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9296 &BTRFS_I(inode
)->runtime_flags
))
9297 filemap_flush(inode
->i_mapping
);
9300 complete(&delalloc_work
->completion
);
9303 static struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
)
9305 struct btrfs_delalloc_work
*work
;
9307 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
9311 init_completion(&work
->completion
);
9312 INIT_LIST_HEAD(&work
->list
);
9313 work
->inode
= inode
;
9314 btrfs_init_work(&work
->work
, btrfs_run_delalloc_work
, NULL
, NULL
);
9320 * some fairly slow code that needs optimization. This walks the list
9321 * of all the inodes with pending delalloc and forces them to disk.
9323 static int start_delalloc_inodes(struct btrfs_root
*root
,
9324 struct writeback_control
*wbc
, bool snapshot
,
9325 bool in_reclaim_context
)
9327 struct btrfs_inode
*binode
;
9328 struct inode
*inode
;
9329 struct btrfs_delalloc_work
*work
, *next
;
9330 struct list_head works
;
9331 struct list_head splice
;
9333 bool full_flush
= wbc
->nr_to_write
== LONG_MAX
;
9335 INIT_LIST_HEAD(&works
);
9336 INIT_LIST_HEAD(&splice
);
9338 mutex_lock(&root
->delalloc_mutex
);
9339 spin_lock(&root
->delalloc_lock
);
9340 list_splice_init(&root
->delalloc_inodes
, &splice
);
9341 while (!list_empty(&splice
)) {
9342 binode
= list_entry(splice
.next
, struct btrfs_inode
,
9345 list_move_tail(&binode
->delalloc_inodes
,
9346 &root
->delalloc_inodes
);
9348 if (in_reclaim_context
&&
9349 test_bit(BTRFS_INODE_NO_DELALLOC_FLUSH
, &binode
->runtime_flags
))
9352 inode
= igrab(&binode
->vfs_inode
);
9354 cond_resched_lock(&root
->delalloc_lock
);
9357 spin_unlock(&root
->delalloc_lock
);
9360 set_bit(BTRFS_INODE_SNAPSHOT_FLUSH
,
9361 &binode
->runtime_flags
);
9363 work
= btrfs_alloc_delalloc_work(inode
);
9369 list_add_tail(&work
->list
, &works
);
9370 btrfs_queue_work(root
->fs_info
->flush_workers
,
9373 ret
= filemap_fdatawrite_wbc(inode
->i_mapping
, wbc
);
9374 btrfs_add_delayed_iput(BTRFS_I(inode
));
9375 if (ret
|| wbc
->nr_to_write
<= 0)
9379 spin_lock(&root
->delalloc_lock
);
9381 spin_unlock(&root
->delalloc_lock
);
9384 list_for_each_entry_safe(work
, next
, &works
, list
) {
9385 list_del_init(&work
->list
);
9386 wait_for_completion(&work
->completion
);
9390 if (!list_empty(&splice
)) {
9391 spin_lock(&root
->delalloc_lock
);
9392 list_splice_tail(&splice
, &root
->delalloc_inodes
);
9393 spin_unlock(&root
->delalloc_lock
);
9395 mutex_unlock(&root
->delalloc_mutex
);
9399 int btrfs_start_delalloc_snapshot(struct btrfs_root
*root
, bool in_reclaim_context
)
9401 struct writeback_control wbc
= {
9402 .nr_to_write
= LONG_MAX
,
9403 .sync_mode
= WB_SYNC_NONE
,
9405 .range_end
= LLONG_MAX
,
9407 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9409 if (BTRFS_FS_ERROR(fs_info
))
9412 return start_delalloc_inodes(root
, &wbc
, true, in_reclaim_context
);
9415 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, long nr
,
9416 bool in_reclaim_context
)
9418 struct writeback_control wbc
= {
9420 .sync_mode
= WB_SYNC_NONE
,
9422 .range_end
= LLONG_MAX
,
9424 struct btrfs_root
*root
;
9425 struct list_head splice
;
9428 if (BTRFS_FS_ERROR(fs_info
))
9431 INIT_LIST_HEAD(&splice
);
9433 mutex_lock(&fs_info
->delalloc_root_mutex
);
9434 spin_lock(&fs_info
->delalloc_root_lock
);
9435 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
9436 while (!list_empty(&splice
)) {
9438 * Reset nr_to_write here so we know that we're doing a full
9442 wbc
.nr_to_write
= LONG_MAX
;
9444 root
= list_first_entry(&splice
, struct btrfs_root
,
9446 root
= btrfs_grab_root(root
);
9448 list_move_tail(&root
->delalloc_root
,
9449 &fs_info
->delalloc_roots
);
9450 spin_unlock(&fs_info
->delalloc_root_lock
);
9452 ret
= start_delalloc_inodes(root
, &wbc
, false, in_reclaim_context
);
9453 btrfs_put_root(root
);
9454 if (ret
< 0 || wbc
.nr_to_write
<= 0)
9456 spin_lock(&fs_info
->delalloc_root_lock
);
9458 spin_unlock(&fs_info
->delalloc_root_lock
);
9462 if (!list_empty(&splice
)) {
9463 spin_lock(&fs_info
->delalloc_root_lock
);
9464 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
9465 spin_unlock(&fs_info
->delalloc_root_lock
);
9467 mutex_unlock(&fs_info
->delalloc_root_mutex
);
9471 static int btrfs_symlink(struct mnt_idmap
*idmap
, struct inode
*dir
,
9472 struct dentry
*dentry
, const char *symname
)
9474 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
9475 struct btrfs_trans_handle
*trans
;
9476 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9477 struct btrfs_path
*path
;
9478 struct btrfs_key key
;
9479 struct inode
*inode
;
9480 struct btrfs_new_inode_args new_inode_args
= {
9484 unsigned int trans_num_items
;
9489 struct btrfs_file_extent_item
*ei
;
9490 struct extent_buffer
*leaf
;
9492 name_len
= strlen(symname
);
9493 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(fs_info
))
9494 return -ENAMETOOLONG
;
9496 inode
= new_inode(dir
->i_sb
);
9499 inode_init_owner(idmap
, inode
, dir
, S_IFLNK
| S_IRWXUGO
);
9500 inode
->i_op
= &btrfs_symlink_inode_operations
;
9501 inode_nohighmem(inode
);
9502 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9503 btrfs_i_size_write(BTRFS_I(inode
), name_len
);
9504 inode_set_bytes(inode
, name_len
);
9506 new_inode_args
.inode
= inode
;
9507 err
= btrfs_new_inode_prepare(&new_inode_args
, &trans_num_items
);
9510 /* 1 additional item for the inline extent */
9513 trans
= btrfs_start_transaction(root
, trans_num_items
);
9514 if (IS_ERR(trans
)) {
9515 err
= PTR_ERR(trans
);
9516 goto out_new_inode_args
;
9519 err
= btrfs_create_new_inode(trans
, &new_inode_args
);
9523 path
= btrfs_alloc_path();
9526 btrfs_abort_transaction(trans
, err
);
9527 discard_new_inode(inode
);
9531 key
.objectid
= btrfs_ino(BTRFS_I(inode
));
9533 key
.type
= BTRFS_EXTENT_DATA_KEY
;
9534 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
9535 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
9538 btrfs_abort_transaction(trans
, err
);
9539 btrfs_free_path(path
);
9540 discard_new_inode(inode
);
9544 leaf
= path
->nodes
[0];
9545 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
9546 struct btrfs_file_extent_item
);
9547 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
9548 btrfs_set_file_extent_type(leaf
, ei
,
9549 BTRFS_FILE_EXTENT_INLINE
);
9550 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
9551 btrfs_set_file_extent_compression(leaf
, ei
, 0);
9552 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
9553 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
9555 ptr
= btrfs_file_extent_inline_start(ei
);
9556 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
9557 btrfs_mark_buffer_dirty(leaf
);
9558 btrfs_free_path(path
);
9560 d_instantiate_new(dentry
, inode
);
9563 btrfs_end_transaction(trans
);
9564 btrfs_btree_balance_dirty(fs_info
);
9566 btrfs_new_inode_args_destroy(&new_inode_args
);
9573 static struct btrfs_trans_handle
*insert_prealloc_file_extent(
9574 struct btrfs_trans_handle
*trans_in
,
9575 struct btrfs_inode
*inode
,
9576 struct btrfs_key
*ins
,
9579 struct btrfs_file_extent_item stack_fi
;
9580 struct btrfs_replace_extent_info extent_info
;
9581 struct btrfs_trans_handle
*trans
= trans_in
;
9582 struct btrfs_path
*path
;
9583 u64 start
= ins
->objectid
;
9584 u64 len
= ins
->offset
;
9585 int qgroup_released
;
9588 memset(&stack_fi
, 0, sizeof(stack_fi
));
9590 btrfs_set_stack_file_extent_type(&stack_fi
, BTRFS_FILE_EXTENT_PREALLOC
);
9591 btrfs_set_stack_file_extent_disk_bytenr(&stack_fi
, start
);
9592 btrfs_set_stack_file_extent_disk_num_bytes(&stack_fi
, len
);
9593 btrfs_set_stack_file_extent_num_bytes(&stack_fi
, len
);
9594 btrfs_set_stack_file_extent_ram_bytes(&stack_fi
, len
);
9595 btrfs_set_stack_file_extent_compression(&stack_fi
, BTRFS_COMPRESS_NONE
);
9596 /* Encryption and other encoding is reserved and all 0 */
9598 qgroup_released
= btrfs_qgroup_release_data(inode
, file_offset
, len
);
9599 if (qgroup_released
< 0)
9600 return ERR_PTR(qgroup_released
);
9603 ret
= insert_reserved_file_extent(trans
, inode
,
9604 file_offset
, &stack_fi
,
9605 true, qgroup_released
);
9611 extent_info
.disk_offset
= start
;
9612 extent_info
.disk_len
= len
;
9613 extent_info
.data_offset
= 0;
9614 extent_info
.data_len
= len
;
9615 extent_info
.file_offset
= file_offset
;
9616 extent_info
.extent_buf
= (char *)&stack_fi
;
9617 extent_info
.is_new_extent
= true;
9618 extent_info
.update_times
= true;
9619 extent_info
.qgroup_reserved
= qgroup_released
;
9620 extent_info
.insertions
= 0;
9622 path
= btrfs_alloc_path();
9628 ret
= btrfs_replace_file_extents(inode
, path
, file_offset
,
9629 file_offset
+ len
- 1, &extent_info
,
9631 btrfs_free_path(path
);
9638 * We have released qgroup data range at the beginning of the function,
9639 * and normally qgroup_released bytes will be freed when committing
9641 * But if we error out early, we have to free what we have released
9642 * or we leak qgroup data reservation.
9644 btrfs_qgroup_free_refroot(inode
->root
->fs_info
,
9645 inode
->root
->root_key
.objectid
, qgroup_released
,
9646 BTRFS_QGROUP_RSV_DATA
);
9647 return ERR_PTR(ret
);
9650 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9651 u64 start
, u64 num_bytes
, u64 min_size
,
9652 loff_t actual_len
, u64
*alloc_hint
,
9653 struct btrfs_trans_handle
*trans
)
9655 struct btrfs_fs_info
*fs_info
= btrfs_sb(inode
->i_sb
);
9656 struct extent_map
*em
;
9657 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9658 struct btrfs_key ins
;
9659 u64 cur_offset
= start
;
9660 u64 clear_offset
= start
;
9663 u64 last_alloc
= (u64
)-1;
9665 bool own_trans
= true;
9666 u64 end
= start
+ num_bytes
- 1;
9670 while (num_bytes
> 0) {
9671 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
9672 cur_bytes
= max(cur_bytes
, min_size
);
9674 * If we are severely fragmented we could end up with really
9675 * small allocations, so if the allocator is returning small
9676 * chunks lets make its job easier by only searching for those
9679 cur_bytes
= min(cur_bytes
, last_alloc
);
9680 ret
= btrfs_reserve_extent(root
, cur_bytes
, cur_bytes
,
9681 min_size
, 0, *alloc_hint
, &ins
, 1, 0);
9686 * We've reserved this space, and thus converted it from
9687 * ->bytes_may_use to ->bytes_reserved. Any error that happens
9688 * from here on out we will only need to clear our reservation
9689 * for the remaining unreserved area, so advance our
9690 * clear_offset by our extent size.
9692 clear_offset
+= ins
.offset
;
9694 last_alloc
= ins
.offset
;
9695 trans
= insert_prealloc_file_extent(trans
, BTRFS_I(inode
),
9698 * Now that we inserted the prealloc extent we can finally
9699 * decrement the number of reservations in the block group.
9700 * If we did it before, we could race with relocation and have
9701 * relocation miss the reserved extent, making it fail later.
9703 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
9704 if (IS_ERR(trans
)) {
9705 ret
= PTR_ERR(trans
);
9706 btrfs_free_reserved_extent(fs_info
, ins
.objectid
,
9711 em
= alloc_extent_map();
9713 btrfs_drop_extent_map_range(BTRFS_I(inode
), cur_offset
,
9714 cur_offset
+ ins
.offset
- 1, false);
9715 btrfs_set_inode_full_sync(BTRFS_I(inode
));
9719 em
->start
= cur_offset
;
9720 em
->orig_start
= cur_offset
;
9721 em
->len
= ins
.offset
;
9722 em
->block_start
= ins
.objectid
;
9723 em
->block_len
= ins
.offset
;
9724 em
->orig_block_len
= ins
.offset
;
9725 em
->ram_bytes
= ins
.offset
;
9726 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
9727 em
->generation
= trans
->transid
;
9729 ret
= btrfs_replace_extent_map_range(BTRFS_I(inode
), em
, true);
9730 free_extent_map(em
);
9732 num_bytes
-= ins
.offset
;
9733 cur_offset
+= ins
.offset
;
9734 *alloc_hint
= ins
.objectid
+ ins
.offset
;
9736 inode_inc_iversion(inode
);
9737 inode
->i_ctime
= current_time(inode
);
9738 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
9739 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
9740 (actual_len
> inode
->i_size
) &&
9741 (cur_offset
> inode
->i_size
)) {
9742 if (cur_offset
> actual_len
)
9743 i_size
= actual_len
;
9745 i_size
= cur_offset
;
9746 i_size_write(inode
, i_size
);
9747 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode
), 0);
9750 ret
= btrfs_update_inode(trans
, root
, BTRFS_I(inode
));
9753 btrfs_abort_transaction(trans
, ret
);
9755 btrfs_end_transaction(trans
);
9760 btrfs_end_transaction(trans
);
9764 if (clear_offset
< end
)
9765 btrfs_free_reserved_data_space(BTRFS_I(inode
), NULL
, clear_offset
,
9766 end
- clear_offset
+ 1);
9770 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9771 u64 start
, u64 num_bytes
, u64 min_size
,
9772 loff_t actual_len
, u64
*alloc_hint
)
9774 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9775 min_size
, actual_len
, alloc_hint
,
9779 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
9780 struct btrfs_trans_handle
*trans
, int mode
,
9781 u64 start
, u64 num_bytes
, u64 min_size
,
9782 loff_t actual_len
, u64
*alloc_hint
)
9784 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9785 min_size
, actual_len
, alloc_hint
, trans
);
9788 static int btrfs_permission(struct mnt_idmap
*idmap
,
9789 struct inode
*inode
, int mask
)
9791 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9792 umode_t mode
= inode
->i_mode
;
9794 if (mask
& MAY_WRITE
&&
9795 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
9796 if (btrfs_root_readonly(root
))
9798 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
9801 return generic_permission(idmap
, inode
, mask
);
9804 static int btrfs_tmpfile(struct mnt_idmap
*idmap
, struct inode
*dir
,
9805 struct file
*file
, umode_t mode
)
9807 struct btrfs_fs_info
*fs_info
= btrfs_sb(dir
->i_sb
);
9808 struct btrfs_trans_handle
*trans
;
9809 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9810 struct inode
*inode
;
9811 struct btrfs_new_inode_args new_inode_args
= {
9813 .dentry
= file
->f_path
.dentry
,
9816 unsigned int trans_num_items
;
9819 inode
= new_inode(dir
->i_sb
);
9822 inode_init_owner(idmap
, inode
, dir
, mode
);
9823 inode
->i_fop
= &btrfs_file_operations
;
9824 inode
->i_op
= &btrfs_file_inode_operations
;
9825 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9827 new_inode_args
.inode
= inode
;
9828 ret
= btrfs_new_inode_prepare(&new_inode_args
, &trans_num_items
);
9832 trans
= btrfs_start_transaction(root
, trans_num_items
);
9833 if (IS_ERR(trans
)) {
9834 ret
= PTR_ERR(trans
);
9835 goto out_new_inode_args
;
9838 ret
= btrfs_create_new_inode(trans
, &new_inode_args
);
9841 * We set number of links to 0 in btrfs_create_new_inode(), and here we
9842 * set it to 1 because d_tmpfile() will issue a warning if the count is
9845 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9847 set_nlink(inode
, 1);
9850 d_tmpfile(file
, inode
);
9851 unlock_new_inode(inode
);
9852 mark_inode_dirty(inode
);
9855 btrfs_end_transaction(trans
);
9856 btrfs_btree_balance_dirty(fs_info
);
9858 btrfs_new_inode_args_destroy(&new_inode_args
);
9862 return finish_open_simple(file
, ret
);
9865 void btrfs_set_range_writeback(struct btrfs_inode
*inode
, u64 start
, u64 end
)
9867 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
9868 unsigned long index
= start
>> PAGE_SHIFT
;
9869 unsigned long end_index
= end
>> PAGE_SHIFT
;
9873 ASSERT(end
+ 1 - start
<= U32_MAX
);
9874 len
= end
+ 1 - start
;
9875 while (index
<= end_index
) {
9876 page
= find_get_page(inode
->vfs_inode
.i_mapping
, index
);
9877 ASSERT(page
); /* Pages should be in the extent_io_tree */
9879 btrfs_page_set_writeback(fs_info
, page
, start
, len
);
9885 int btrfs_encoded_io_compression_from_extent(struct btrfs_fs_info
*fs_info
,
9888 switch (compress_type
) {
9889 case BTRFS_COMPRESS_NONE
:
9890 return BTRFS_ENCODED_IO_COMPRESSION_NONE
;
9891 case BTRFS_COMPRESS_ZLIB
:
9892 return BTRFS_ENCODED_IO_COMPRESSION_ZLIB
;
9893 case BTRFS_COMPRESS_LZO
:
9895 * The LZO format depends on the sector size. 64K is the maximum
9896 * sector size that we support.
9898 if (fs_info
->sectorsize
< SZ_4K
|| fs_info
->sectorsize
> SZ_64K
)
9900 return BTRFS_ENCODED_IO_COMPRESSION_LZO_4K
+
9901 (fs_info
->sectorsize_bits
- 12);
9902 case BTRFS_COMPRESS_ZSTD
:
9903 return BTRFS_ENCODED_IO_COMPRESSION_ZSTD
;
9909 static ssize_t
btrfs_encoded_read_inline(
9911 struct iov_iter
*iter
, u64 start
,
9913 struct extent_state
**cached_state
,
9914 u64 extent_start
, size_t count
,
9915 struct btrfs_ioctl_encoded_io_args
*encoded
,
9918 struct btrfs_inode
*inode
= BTRFS_I(file_inode(iocb
->ki_filp
));
9919 struct btrfs_root
*root
= inode
->root
;
9920 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
9921 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
9922 struct btrfs_path
*path
;
9923 struct extent_buffer
*leaf
;
9924 struct btrfs_file_extent_item
*item
;
9930 path
= btrfs_alloc_path();
9935 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
9939 /* The extent item disappeared? */
9944 leaf
= path
->nodes
[0];
9945 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_file_extent_item
);
9947 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, item
);
9948 ptr
= btrfs_file_extent_inline_start(item
);
9950 encoded
->len
= min_t(u64
, extent_start
+ ram_bytes
,
9951 inode
->vfs_inode
.i_size
) - iocb
->ki_pos
;
9952 ret
= btrfs_encoded_io_compression_from_extent(fs_info
,
9953 btrfs_file_extent_compression(leaf
, item
));
9956 encoded
->compression
= ret
;
9957 if (encoded
->compression
) {
9960 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
9962 if (inline_size
> count
) {
9966 count
= inline_size
;
9967 encoded
->unencoded_len
= ram_bytes
;
9968 encoded
->unencoded_offset
= iocb
->ki_pos
- extent_start
;
9970 count
= min_t(u64
, count
, encoded
->len
);
9971 encoded
->len
= count
;
9972 encoded
->unencoded_len
= count
;
9973 ptr
+= iocb
->ki_pos
- extent_start
;
9976 tmp
= kmalloc(count
, GFP_NOFS
);
9981 read_extent_buffer(leaf
, tmp
, ptr
, count
);
9982 btrfs_release_path(path
);
9983 unlock_extent(io_tree
, start
, lockend
, cached_state
);
9984 btrfs_inode_unlock(inode
, BTRFS_ILOCK_SHARED
);
9987 ret
= copy_to_iter(tmp
, count
, iter
);
9992 btrfs_free_path(path
);
9996 struct btrfs_encoded_read_private
{
9997 wait_queue_head_t wait
;
9999 blk_status_t status
;
10002 static void btrfs_encoded_read_endio(struct btrfs_bio
*bbio
)
10004 struct btrfs_encoded_read_private
*priv
= bbio
->private;
10006 if (bbio
->bio
.bi_status
) {
10008 * The memory barrier implied by the atomic_dec_return() here
10009 * pairs with the memory barrier implied by the
10010 * atomic_dec_return() or io_wait_event() in
10011 * btrfs_encoded_read_regular_fill_pages() to ensure that this
10012 * write is observed before the load of status in
10013 * btrfs_encoded_read_regular_fill_pages().
10015 WRITE_ONCE(priv
->status
, bbio
->bio
.bi_status
);
10017 if (!atomic_dec_return(&priv
->pending
))
10018 wake_up(&priv
->wait
);
10019 bio_put(&bbio
->bio
);
10022 int btrfs_encoded_read_regular_fill_pages(struct btrfs_inode
*inode
,
10023 u64 file_offset
, u64 disk_bytenr
,
10024 u64 disk_io_size
, struct page
**pages
)
10026 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
10027 struct btrfs_encoded_read_private priv
= {
10028 .pending
= ATOMIC_INIT(1),
10030 unsigned long i
= 0;
10031 struct btrfs_bio
*bbio
;
10033 init_waitqueue_head(&priv
.wait
);
10035 bbio
= btrfs_bio_alloc(BIO_MAX_VECS
, REQ_OP_READ
, fs_info
,
10036 btrfs_encoded_read_endio
, &priv
);
10037 bbio
->bio
.bi_iter
.bi_sector
= disk_bytenr
>> SECTOR_SHIFT
;
10038 bbio
->inode
= inode
;
10041 size_t bytes
= min_t(u64
, disk_io_size
, PAGE_SIZE
);
10043 if (bio_add_page(&bbio
->bio
, pages
[i
], bytes
, 0) < bytes
) {
10044 atomic_inc(&priv
.pending
);
10045 btrfs_submit_bio(bbio
, 0);
10047 bbio
= btrfs_bio_alloc(BIO_MAX_VECS
, REQ_OP_READ
, fs_info
,
10048 btrfs_encoded_read_endio
, &priv
);
10049 bbio
->bio
.bi_iter
.bi_sector
= disk_bytenr
>> SECTOR_SHIFT
;
10050 bbio
->inode
= inode
;
10055 disk_bytenr
+= bytes
;
10056 disk_io_size
-= bytes
;
10057 } while (disk_io_size
);
10059 atomic_inc(&priv
.pending
);
10060 btrfs_submit_bio(bbio
, 0);
10062 if (atomic_dec_return(&priv
.pending
))
10063 io_wait_event(priv
.wait
, !atomic_read(&priv
.pending
));
10064 /* See btrfs_encoded_read_endio() for ordering. */
10065 return blk_status_to_errno(READ_ONCE(priv
.status
));
10068 static ssize_t
btrfs_encoded_read_regular(struct kiocb
*iocb
,
10069 struct iov_iter
*iter
,
10070 u64 start
, u64 lockend
,
10071 struct extent_state
**cached_state
,
10072 u64 disk_bytenr
, u64 disk_io_size
,
10073 size_t count
, bool compressed
,
10076 struct btrfs_inode
*inode
= BTRFS_I(file_inode(iocb
->ki_filp
));
10077 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
10078 struct page
**pages
;
10079 unsigned long nr_pages
, i
;
10081 size_t page_offset
;
10084 nr_pages
= DIV_ROUND_UP(disk_io_size
, PAGE_SIZE
);
10085 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
10088 ret
= btrfs_alloc_page_array(nr_pages
, pages
);
10094 ret
= btrfs_encoded_read_regular_fill_pages(inode
, start
, disk_bytenr
,
10095 disk_io_size
, pages
);
10099 unlock_extent(io_tree
, start
, lockend
, cached_state
);
10100 btrfs_inode_unlock(inode
, BTRFS_ILOCK_SHARED
);
10107 i
= (iocb
->ki_pos
- start
) >> PAGE_SHIFT
;
10108 page_offset
= (iocb
->ki_pos
- start
) & (PAGE_SIZE
- 1);
10111 while (cur
< count
) {
10112 size_t bytes
= min_t(size_t, count
- cur
,
10113 PAGE_SIZE
- page_offset
);
10115 if (copy_page_to_iter(pages
[i
], page_offset
, bytes
,
10126 for (i
= 0; i
< nr_pages
; i
++) {
10128 __free_page(pages
[i
]);
10134 ssize_t
btrfs_encoded_read(struct kiocb
*iocb
, struct iov_iter
*iter
,
10135 struct btrfs_ioctl_encoded_io_args
*encoded
)
10137 struct btrfs_inode
*inode
= BTRFS_I(file_inode(iocb
->ki_filp
));
10138 struct btrfs_fs_info
*fs_info
= inode
->root
->fs_info
;
10139 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
10141 size_t count
= iov_iter_count(iter
);
10142 u64 start
, lockend
, disk_bytenr
, disk_io_size
;
10143 struct extent_state
*cached_state
= NULL
;
10144 struct extent_map
*em
;
10145 bool unlocked
= false;
10147 file_accessed(iocb
->ki_filp
);
10149 btrfs_inode_lock(inode
, BTRFS_ILOCK_SHARED
);
10151 if (iocb
->ki_pos
>= inode
->vfs_inode
.i_size
) {
10152 btrfs_inode_unlock(inode
, BTRFS_ILOCK_SHARED
);
10155 start
= ALIGN_DOWN(iocb
->ki_pos
, fs_info
->sectorsize
);
10157 * We don't know how long the extent containing iocb->ki_pos is, but if
10158 * it's compressed we know that it won't be longer than this.
10160 lockend
= start
+ BTRFS_MAX_UNCOMPRESSED
- 1;
10163 struct btrfs_ordered_extent
*ordered
;
10165 ret
= btrfs_wait_ordered_range(&inode
->vfs_inode
, start
,
10166 lockend
- start
+ 1);
10168 goto out_unlock_inode
;
10169 lock_extent(io_tree
, start
, lockend
, &cached_state
);
10170 ordered
= btrfs_lookup_ordered_range(inode
, start
,
10171 lockend
- start
+ 1);
10174 btrfs_put_ordered_extent(ordered
);
10175 unlock_extent(io_tree
, start
, lockend
, &cached_state
);
10179 em
= btrfs_get_extent(inode
, NULL
, 0, start
, lockend
- start
+ 1);
10182 goto out_unlock_extent
;
10185 if (em
->block_start
== EXTENT_MAP_INLINE
) {
10186 u64 extent_start
= em
->start
;
10189 * For inline extents we get everything we need out of the
10192 free_extent_map(em
);
10194 ret
= btrfs_encoded_read_inline(iocb
, iter
, start
, lockend
,
10195 &cached_state
, extent_start
,
10196 count
, encoded
, &unlocked
);
10201 * We only want to return up to EOF even if the extent extends beyond
10204 encoded
->len
= min_t(u64
, extent_map_end(em
),
10205 inode
->vfs_inode
.i_size
) - iocb
->ki_pos
;
10206 if (em
->block_start
== EXTENT_MAP_HOLE
||
10207 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
10208 disk_bytenr
= EXTENT_MAP_HOLE
;
10209 count
= min_t(u64
, count
, encoded
->len
);
10210 encoded
->len
= count
;
10211 encoded
->unencoded_len
= count
;
10212 } else if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
10213 disk_bytenr
= em
->block_start
;
10215 * Bail if the buffer isn't large enough to return the whole
10216 * compressed extent.
10218 if (em
->block_len
> count
) {
10222 disk_io_size
= em
->block_len
;
10223 count
= em
->block_len
;
10224 encoded
->unencoded_len
= em
->ram_bytes
;
10225 encoded
->unencoded_offset
= iocb
->ki_pos
- em
->orig_start
;
10226 ret
= btrfs_encoded_io_compression_from_extent(fs_info
,
10227 em
->compress_type
);
10230 encoded
->compression
= ret
;
10232 disk_bytenr
= em
->block_start
+ (start
- em
->start
);
10233 if (encoded
->len
> count
)
10234 encoded
->len
= count
;
10236 * Don't read beyond what we locked. This also limits the page
10237 * allocations that we'll do.
10239 disk_io_size
= min(lockend
+ 1, iocb
->ki_pos
+ encoded
->len
) - start
;
10240 count
= start
+ disk_io_size
- iocb
->ki_pos
;
10241 encoded
->len
= count
;
10242 encoded
->unencoded_len
= count
;
10243 disk_io_size
= ALIGN(disk_io_size
, fs_info
->sectorsize
);
10245 free_extent_map(em
);
10248 if (disk_bytenr
== EXTENT_MAP_HOLE
) {
10249 unlock_extent(io_tree
, start
, lockend
, &cached_state
);
10250 btrfs_inode_unlock(inode
, BTRFS_ILOCK_SHARED
);
10252 ret
= iov_iter_zero(count
, iter
);
10256 ret
= btrfs_encoded_read_regular(iocb
, iter
, start
, lockend
,
10257 &cached_state
, disk_bytenr
,
10258 disk_io_size
, count
,
10259 encoded
->compression
,
10265 iocb
->ki_pos
+= encoded
->len
;
10267 free_extent_map(em
);
10270 unlock_extent(io_tree
, start
, lockend
, &cached_state
);
10273 btrfs_inode_unlock(inode
, BTRFS_ILOCK_SHARED
);
10277 ssize_t
btrfs_do_encoded_write(struct kiocb
*iocb
, struct iov_iter
*from
,
10278 const struct btrfs_ioctl_encoded_io_args
*encoded
)
10280 struct btrfs_inode
*inode
= BTRFS_I(file_inode(iocb
->ki_filp
));
10281 struct btrfs_root
*root
= inode
->root
;
10282 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10283 struct extent_io_tree
*io_tree
= &inode
->io_tree
;
10284 struct extent_changeset
*data_reserved
= NULL
;
10285 struct extent_state
*cached_state
= NULL
;
10289 u64 num_bytes
, ram_bytes
, disk_num_bytes
;
10290 unsigned long nr_pages
, i
;
10291 struct page
**pages
;
10292 struct btrfs_key ins
;
10293 bool extent_reserved
= false;
10294 struct extent_map
*em
;
10297 switch (encoded
->compression
) {
10298 case BTRFS_ENCODED_IO_COMPRESSION_ZLIB
:
10299 compression
= BTRFS_COMPRESS_ZLIB
;
10301 case BTRFS_ENCODED_IO_COMPRESSION_ZSTD
:
10302 compression
= BTRFS_COMPRESS_ZSTD
;
10304 case BTRFS_ENCODED_IO_COMPRESSION_LZO_4K
:
10305 case BTRFS_ENCODED_IO_COMPRESSION_LZO_8K
:
10306 case BTRFS_ENCODED_IO_COMPRESSION_LZO_16K
:
10307 case BTRFS_ENCODED_IO_COMPRESSION_LZO_32K
:
10308 case BTRFS_ENCODED_IO_COMPRESSION_LZO_64K
:
10309 /* The sector size must match for LZO. */
10310 if (encoded
->compression
-
10311 BTRFS_ENCODED_IO_COMPRESSION_LZO_4K
+ 12 !=
10312 fs_info
->sectorsize_bits
)
10314 compression
= BTRFS_COMPRESS_LZO
;
10319 if (encoded
->encryption
!= BTRFS_ENCODED_IO_ENCRYPTION_NONE
)
10322 orig_count
= iov_iter_count(from
);
10324 /* The extent size must be sane. */
10325 if (encoded
->unencoded_len
> BTRFS_MAX_UNCOMPRESSED
||
10326 orig_count
> BTRFS_MAX_COMPRESSED
|| orig_count
== 0)
10330 * The compressed data must be smaller than the decompressed data.
10332 * It's of course possible for data to compress to larger or the same
10333 * size, but the buffered I/O path falls back to no compression for such
10334 * data, and we don't want to break any assumptions by creating these
10337 * Note that this is less strict than the current check we have that the
10338 * compressed data must be at least one sector smaller than the
10339 * decompressed data. We only want to enforce the weaker requirement
10340 * from old kernels that it is at least one byte smaller.
10342 if (orig_count
>= encoded
->unencoded_len
)
10345 /* The extent must start on a sector boundary. */
10346 start
= iocb
->ki_pos
;
10347 if (!IS_ALIGNED(start
, fs_info
->sectorsize
))
10351 * The extent must end on a sector boundary. However, we allow a write
10352 * which ends at or extends i_size to have an unaligned length; we round
10353 * up the extent size and set i_size to the unaligned end.
10355 if (start
+ encoded
->len
< inode
->vfs_inode
.i_size
&&
10356 !IS_ALIGNED(start
+ encoded
->len
, fs_info
->sectorsize
))
10359 /* Finally, the offset in the unencoded data must be sector-aligned. */
10360 if (!IS_ALIGNED(encoded
->unencoded_offset
, fs_info
->sectorsize
))
10363 num_bytes
= ALIGN(encoded
->len
, fs_info
->sectorsize
);
10364 ram_bytes
= ALIGN(encoded
->unencoded_len
, fs_info
->sectorsize
);
10365 end
= start
+ num_bytes
- 1;
10368 * If the extent cannot be inline, the compressed data on disk must be
10369 * sector-aligned. For convenience, we extend it with zeroes if it
10372 disk_num_bytes
= ALIGN(orig_count
, fs_info
->sectorsize
);
10373 nr_pages
= DIV_ROUND_UP(disk_num_bytes
, PAGE_SIZE
);
10374 pages
= kvcalloc(nr_pages
, sizeof(struct page
*), GFP_KERNEL_ACCOUNT
);
10377 for (i
= 0; i
< nr_pages
; i
++) {
10378 size_t bytes
= min_t(size_t, PAGE_SIZE
, iov_iter_count(from
));
10381 pages
[i
] = alloc_page(GFP_KERNEL_ACCOUNT
);
10386 kaddr
= kmap_local_page(pages
[i
]);
10387 if (copy_from_iter(kaddr
, bytes
, from
) != bytes
) {
10388 kunmap_local(kaddr
);
10392 if (bytes
< PAGE_SIZE
)
10393 memset(kaddr
+ bytes
, 0, PAGE_SIZE
- bytes
);
10394 kunmap_local(kaddr
);
10398 struct btrfs_ordered_extent
*ordered
;
10400 ret
= btrfs_wait_ordered_range(&inode
->vfs_inode
, start
, num_bytes
);
10403 ret
= invalidate_inode_pages2_range(inode
->vfs_inode
.i_mapping
,
10404 start
>> PAGE_SHIFT
,
10405 end
>> PAGE_SHIFT
);
10408 lock_extent(io_tree
, start
, end
, &cached_state
);
10409 ordered
= btrfs_lookup_ordered_range(inode
, start
, num_bytes
);
10411 !filemap_range_has_page(inode
->vfs_inode
.i_mapping
, start
, end
))
10414 btrfs_put_ordered_extent(ordered
);
10415 unlock_extent(io_tree
, start
, end
, &cached_state
);
10420 * We don't use the higher-level delalloc space functions because our
10421 * num_bytes and disk_num_bytes are different.
10423 ret
= btrfs_alloc_data_chunk_ondemand(inode
, disk_num_bytes
);
10426 ret
= btrfs_qgroup_reserve_data(inode
, &data_reserved
, start
, num_bytes
);
10428 goto out_free_data_space
;
10429 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
, disk_num_bytes
,
10432 goto out_qgroup_free_data
;
10434 /* Try an inline extent first. */
10435 if (start
== 0 && encoded
->unencoded_len
== encoded
->len
&&
10436 encoded
->unencoded_offset
== 0) {
10437 ret
= cow_file_range_inline(inode
, encoded
->len
, orig_count
,
10438 compression
, pages
, true);
10442 goto out_delalloc_release
;
10446 ret
= btrfs_reserve_extent(root
, disk_num_bytes
, disk_num_bytes
,
10447 disk_num_bytes
, 0, 0, &ins
, 1, 1);
10449 goto out_delalloc_release
;
10450 extent_reserved
= true;
10452 em
= create_io_em(inode
, start
, num_bytes
,
10453 start
- encoded
->unencoded_offset
, ins
.objectid
,
10454 ins
.offset
, ins
.offset
, ram_bytes
, compression
,
10455 BTRFS_ORDERED_COMPRESSED
);
10458 goto out_free_reserved
;
10460 free_extent_map(em
);
10462 ret
= btrfs_add_ordered_extent(inode
, start
, num_bytes
, ram_bytes
,
10463 ins
.objectid
, ins
.offset
,
10464 encoded
->unencoded_offset
,
10465 (1 << BTRFS_ORDERED_ENCODED
) |
10466 (1 << BTRFS_ORDERED_COMPRESSED
),
10469 btrfs_drop_extent_map_range(inode
, start
, end
, false);
10470 goto out_free_reserved
;
10472 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
10474 if (start
+ encoded
->len
> inode
->vfs_inode
.i_size
)
10475 i_size_write(&inode
->vfs_inode
, start
+ encoded
->len
);
10477 unlock_extent(io_tree
, start
, end
, &cached_state
);
10479 btrfs_delalloc_release_extents(inode
, num_bytes
);
10481 btrfs_submit_compressed_write(inode
, start
, num_bytes
, ins
.objectid
,
10482 ins
.offset
, pages
, nr_pages
, 0, false);
10487 btrfs_dec_block_group_reservations(fs_info
, ins
.objectid
);
10488 btrfs_free_reserved_extent(fs_info
, ins
.objectid
, ins
.offset
, 1);
10489 out_delalloc_release
:
10490 btrfs_delalloc_release_extents(inode
, num_bytes
);
10491 btrfs_delalloc_release_metadata(inode
, disk_num_bytes
, ret
< 0);
10492 out_qgroup_free_data
:
10494 btrfs_qgroup_free_data(inode
, data_reserved
, start
, num_bytes
);
10495 out_free_data_space
:
10497 * If btrfs_reserve_extent() succeeded, then we already decremented
10500 if (!extent_reserved
)
10501 btrfs_free_reserved_data_space_noquota(fs_info
, disk_num_bytes
);
10503 unlock_extent(io_tree
, start
, end
, &cached_state
);
10505 for (i
= 0; i
< nr_pages
; i
++) {
10507 __free_page(pages
[i
]);
10512 iocb
->ki_pos
+= encoded
->len
;
10518 * Add an entry indicating a block group or device which is pinned by a
10519 * swapfile. Returns 0 on success, 1 if there is already an entry for it, or a
10520 * negative errno on failure.
10522 static int btrfs_add_swapfile_pin(struct inode
*inode
, void *ptr
,
10523 bool is_block_group
)
10525 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
10526 struct btrfs_swapfile_pin
*sp
, *entry
;
10527 struct rb_node
**p
;
10528 struct rb_node
*parent
= NULL
;
10530 sp
= kmalloc(sizeof(*sp
), GFP_NOFS
);
10535 sp
->is_block_group
= is_block_group
;
10536 sp
->bg_extent_count
= 1;
10538 spin_lock(&fs_info
->swapfile_pins_lock
);
10539 p
= &fs_info
->swapfile_pins
.rb_node
;
10542 entry
= rb_entry(parent
, struct btrfs_swapfile_pin
, node
);
10543 if (sp
->ptr
< entry
->ptr
||
10544 (sp
->ptr
== entry
->ptr
&& sp
->inode
< entry
->inode
)) {
10545 p
= &(*p
)->rb_left
;
10546 } else if (sp
->ptr
> entry
->ptr
||
10547 (sp
->ptr
== entry
->ptr
&& sp
->inode
> entry
->inode
)) {
10548 p
= &(*p
)->rb_right
;
10550 if (is_block_group
)
10551 entry
->bg_extent_count
++;
10552 spin_unlock(&fs_info
->swapfile_pins_lock
);
10557 rb_link_node(&sp
->node
, parent
, p
);
10558 rb_insert_color(&sp
->node
, &fs_info
->swapfile_pins
);
10559 spin_unlock(&fs_info
->swapfile_pins_lock
);
10563 /* Free all of the entries pinned by this swapfile. */
10564 static void btrfs_free_swapfile_pins(struct inode
*inode
)
10566 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
10567 struct btrfs_swapfile_pin
*sp
;
10568 struct rb_node
*node
, *next
;
10570 spin_lock(&fs_info
->swapfile_pins_lock
);
10571 node
= rb_first(&fs_info
->swapfile_pins
);
10573 next
= rb_next(node
);
10574 sp
= rb_entry(node
, struct btrfs_swapfile_pin
, node
);
10575 if (sp
->inode
== inode
) {
10576 rb_erase(&sp
->node
, &fs_info
->swapfile_pins
);
10577 if (sp
->is_block_group
) {
10578 btrfs_dec_block_group_swap_extents(sp
->ptr
,
10579 sp
->bg_extent_count
);
10580 btrfs_put_block_group(sp
->ptr
);
10586 spin_unlock(&fs_info
->swapfile_pins_lock
);
10589 struct btrfs_swap_info
{
10595 unsigned long nr_pages
;
10599 static int btrfs_add_swap_extent(struct swap_info_struct
*sis
,
10600 struct btrfs_swap_info
*bsi
)
10602 unsigned long nr_pages
;
10603 unsigned long max_pages
;
10604 u64 first_ppage
, first_ppage_reported
, next_ppage
;
10608 * Our swapfile may have had its size extended after the swap header was
10609 * written. In that case activating the swapfile should not go beyond
10610 * the max size set in the swap header.
10612 if (bsi
->nr_pages
>= sis
->max
)
10615 max_pages
= sis
->max
- bsi
->nr_pages
;
10616 first_ppage
= PAGE_ALIGN(bsi
->block_start
) >> PAGE_SHIFT
;
10617 next_ppage
= PAGE_ALIGN_DOWN(bsi
->block_start
+ bsi
->block_len
) >> PAGE_SHIFT
;
10619 if (first_ppage
>= next_ppage
)
10621 nr_pages
= next_ppage
- first_ppage
;
10622 nr_pages
= min(nr_pages
, max_pages
);
10624 first_ppage_reported
= first_ppage
;
10625 if (bsi
->start
== 0)
10626 first_ppage_reported
++;
10627 if (bsi
->lowest_ppage
> first_ppage_reported
)
10628 bsi
->lowest_ppage
= first_ppage_reported
;
10629 if (bsi
->highest_ppage
< (next_ppage
- 1))
10630 bsi
->highest_ppage
= next_ppage
- 1;
10632 ret
= add_swap_extent(sis
, bsi
->nr_pages
, nr_pages
, first_ppage
);
10635 bsi
->nr_extents
+= ret
;
10636 bsi
->nr_pages
+= nr_pages
;
10640 static void btrfs_swap_deactivate(struct file
*file
)
10642 struct inode
*inode
= file_inode(file
);
10644 btrfs_free_swapfile_pins(inode
);
10645 atomic_dec(&BTRFS_I(inode
)->root
->nr_swapfiles
);
10648 static int btrfs_swap_activate(struct swap_info_struct
*sis
, struct file
*file
,
10651 struct inode
*inode
= file_inode(file
);
10652 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
10653 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
10654 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
10655 struct extent_state
*cached_state
= NULL
;
10656 struct extent_map
*em
= NULL
;
10657 struct btrfs_device
*device
= NULL
;
10658 struct btrfs_swap_info bsi
= {
10659 .lowest_ppage
= (sector_t
)-1ULL,
10666 * If the swap file was just created, make sure delalloc is done. If the
10667 * file changes again after this, the user is doing something stupid and
10668 * we don't really care.
10670 ret
= btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
10675 * The inode is locked, so these flags won't change after we check them.
10677 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
) {
10678 btrfs_warn(fs_info
, "swapfile must not be compressed");
10681 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)) {
10682 btrfs_warn(fs_info
, "swapfile must not be copy-on-write");
10685 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
10686 btrfs_warn(fs_info
, "swapfile must not be checksummed");
10691 * Balance or device remove/replace/resize can move stuff around from
10692 * under us. The exclop protection makes sure they aren't running/won't
10693 * run concurrently while we are mapping the swap extents, and
10694 * fs_info->swapfile_pins prevents them from running while the swap
10695 * file is active and moving the extents. Note that this also prevents
10696 * a concurrent device add which isn't actually necessary, but it's not
10697 * really worth the trouble to allow it.
10699 if (!btrfs_exclop_start(fs_info
, BTRFS_EXCLOP_SWAP_ACTIVATE
)) {
10700 btrfs_warn(fs_info
,
10701 "cannot activate swapfile while exclusive operation is running");
10706 * Prevent snapshot creation while we are activating the swap file.
10707 * We do not want to race with snapshot creation. If snapshot creation
10708 * already started before we bumped nr_swapfiles from 0 to 1 and
10709 * completes before the first write into the swap file after it is
10710 * activated, than that write would fallback to COW.
10712 if (!btrfs_drew_try_write_lock(&root
->snapshot_lock
)) {
10713 btrfs_exclop_finish(fs_info
);
10714 btrfs_warn(fs_info
,
10715 "cannot activate swapfile because snapshot creation is in progress");
10719 * Snapshots can create extents which require COW even if NODATACOW is
10720 * set. We use this counter to prevent snapshots. We must increment it
10721 * before walking the extents because we don't want a concurrent
10722 * snapshot to run after we've already checked the extents.
10724 * It is possible that subvolume is marked for deletion but still not
10725 * removed yet. To prevent this race, we check the root status before
10726 * activating the swapfile.
10728 spin_lock(&root
->root_item_lock
);
10729 if (btrfs_root_dead(root
)) {
10730 spin_unlock(&root
->root_item_lock
);
10732 btrfs_exclop_finish(fs_info
);
10733 btrfs_warn(fs_info
,
10734 "cannot activate swapfile because subvolume %llu is being deleted",
10735 root
->root_key
.objectid
);
10738 atomic_inc(&root
->nr_swapfiles
);
10739 spin_unlock(&root
->root_item_lock
);
10741 isize
= ALIGN_DOWN(inode
->i_size
, fs_info
->sectorsize
);
10743 lock_extent(io_tree
, 0, isize
- 1, &cached_state
);
10745 while (start
< isize
) {
10746 u64 logical_block_start
, physical_block_start
;
10747 struct btrfs_block_group
*bg
;
10748 u64 len
= isize
- start
;
10750 em
= btrfs_get_extent(BTRFS_I(inode
), NULL
, 0, start
, len
);
10756 if (em
->block_start
== EXTENT_MAP_HOLE
) {
10757 btrfs_warn(fs_info
, "swapfile must not have holes");
10761 if (em
->block_start
== EXTENT_MAP_INLINE
) {
10763 * It's unlikely we'll ever actually find ourselves
10764 * here, as a file small enough to fit inline won't be
10765 * big enough to store more than the swap header, but in
10766 * case something changes in the future, let's catch it
10767 * here rather than later.
10769 btrfs_warn(fs_info
, "swapfile must not be inline");
10773 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
10774 btrfs_warn(fs_info
, "swapfile must not be compressed");
10779 logical_block_start
= em
->block_start
+ (start
- em
->start
);
10780 len
= min(len
, em
->len
- (start
- em
->start
));
10781 free_extent_map(em
);
10784 ret
= can_nocow_extent(inode
, start
, &len
, NULL
, NULL
, NULL
, false, true);
10790 btrfs_warn(fs_info
,
10791 "swapfile must not be copy-on-write");
10796 em
= btrfs_get_chunk_map(fs_info
, logical_block_start
, len
);
10802 if (em
->map_lookup
->type
& BTRFS_BLOCK_GROUP_PROFILE_MASK
) {
10803 btrfs_warn(fs_info
,
10804 "swapfile must have single data profile");
10809 if (device
== NULL
) {
10810 device
= em
->map_lookup
->stripes
[0].dev
;
10811 ret
= btrfs_add_swapfile_pin(inode
, device
, false);
10816 } else if (device
!= em
->map_lookup
->stripes
[0].dev
) {
10817 btrfs_warn(fs_info
, "swapfile must be on one device");
10822 physical_block_start
= (em
->map_lookup
->stripes
[0].physical
+
10823 (logical_block_start
- em
->start
));
10824 len
= min(len
, em
->len
- (logical_block_start
- em
->start
));
10825 free_extent_map(em
);
10828 bg
= btrfs_lookup_block_group(fs_info
, logical_block_start
);
10830 btrfs_warn(fs_info
,
10831 "could not find block group containing swapfile");
10836 if (!btrfs_inc_block_group_swap_extents(bg
)) {
10837 btrfs_warn(fs_info
,
10838 "block group for swapfile at %llu is read-only%s",
10840 atomic_read(&fs_info
->scrubs_running
) ?
10841 " (scrub running)" : "");
10842 btrfs_put_block_group(bg
);
10847 ret
= btrfs_add_swapfile_pin(inode
, bg
, true);
10849 btrfs_put_block_group(bg
);
10856 if (bsi
.block_len
&&
10857 bsi
.block_start
+ bsi
.block_len
== physical_block_start
) {
10858 bsi
.block_len
+= len
;
10860 if (bsi
.block_len
) {
10861 ret
= btrfs_add_swap_extent(sis
, &bsi
);
10866 bsi
.block_start
= physical_block_start
;
10867 bsi
.block_len
= len
;
10874 ret
= btrfs_add_swap_extent(sis
, &bsi
);
10877 if (!IS_ERR_OR_NULL(em
))
10878 free_extent_map(em
);
10880 unlock_extent(io_tree
, 0, isize
- 1, &cached_state
);
10883 btrfs_swap_deactivate(file
);
10885 btrfs_drew_write_unlock(&root
->snapshot_lock
);
10887 btrfs_exclop_finish(fs_info
);
10893 sis
->bdev
= device
->bdev
;
10894 *span
= bsi
.highest_ppage
- bsi
.lowest_ppage
+ 1;
10895 sis
->max
= bsi
.nr_pages
;
10896 sis
->pages
= bsi
.nr_pages
- 1;
10897 sis
->highest_bit
= bsi
.nr_pages
- 1;
10898 return bsi
.nr_extents
;
10901 static void btrfs_swap_deactivate(struct file
*file
)
10905 static int btrfs_swap_activate(struct swap_info_struct
*sis
, struct file
*file
,
10908 return -EOPNOTSUPP
;
10913 * Update the number of bytes used in the VFS' inode. When we replace extents in
10914 * a range (clone, dedupe, fallocate's zero range), we must update the number of
10915 * bytes used by the inode in an atomic manner, so that concurrent stat(2) calls
10916 * always get a correct value.
10918 void btrfs_update_inode_bytes(struct btrfs_inode
*inode
,
10919 const u64 add_bytes
,
10920 const u64 del_bytes
)
10922 if (add_bytes
== del_bytes
)
10925 spin_lock(&inode
->lock
);
10927 inode_sub_bytes(&inode
->vfs_inode
, del_bytes
);
10929 inode_add_bytes(&inode
->vfs_inode
, add_bytes
);
10930 spin_unlock(&inode
->lock
);
10934 * Verify that there are no ordered extents for a given file range.
10936 * @inode: The target inode.
10937 * @start: Start offset of the file range, should be sector size aligned.
10938 * @end: End offset (inclusive) of the file range, its value +1 should be
10939 * sector size aligned.
10941 * This should typically be used for cases where we locked an inode's VFS lock in
10942 * exclusive mode, we have also locked the inode's i_mmap_lock in exclusive mode,
10943 * we have flushed all delalloc in the range, we have waited for all ordered
10944 * extents in the range to complete and finally we have locked the file range in
10945 * the inode's io_tree.
10947 void btrfs_assert_inode_range_clean(struct btrfs_inode
*inode
, u64 start
, u64 end
)
10949 struct btrfs_root
*root
= inode
->root
;
10950 struct btrfs_ordered_extent
*ordered
;
10952 if (!IS_ENABLED(CONFIG_BTRFS_ASSERT
))
10955 ordered
= btrfs_lookup_first_ordered_range(inode
, start
, end
+ 1 - start
);
10957 btrfs_err(root
->fs_info
,
10958 "found unexpected ordered extent in file range [%llu, %llu] for inode %llu root %llu (ordered range [%llu, %llu])",
10959 start
, end
, btrfs_ino(inode
), root
->root_key
.objectid
,
10960 ordered
->file_offset
,
10961 ordered
->file_offset
+ ordered
->num_bytes
- 1);
10962 btrfs_put_ordered_extent(ordered
);
10965 ASSERT(ordered
== NULL
);
10968 static const struct inode_operations btrfs_dir_inode_operations
= {
10969 .getattr
= btrfs_getattr
,
10970 .lookup
= btrfs_lookup
,
10971 .create
= btrfs_create
,
10972 .unlink
= btrfs_unlink
,
10973 .link
= btrfs_link
,
10974 .mkdir
= btrfs_mkdir
,
10975 .rmdir
= btrfs_rmdir
,
10976 .rename
= btrfs_rename2
,
10977 .symlink
= btrfs_symlink
,
10978 .setattr
= btrfs_setattr
,
10979 .mknod
= btrfs_mknod
,
10980 .listxattr
= btrfs_listxattr
,
10981 .permission
= btrfs_permission
,
10982 .get_inode_acl
= btrfs_get_acl
,
10983 .set_acl
= btrfs_set_acl
,
10984 .update_time
= btrfs_update_time
,
10985 .tmpfile
= btrfs_tmpfile
,
10986 .fileattr_get
= btrfs_fileattr_get
,
10987 .fileattr_set
= btrfs_fileattr_set
,
10990 static const struct file_operations btrfs_dir_file_operations
= {
10991 .llseek
= generic_file_llseek
,
10992 .read
= generic_read_dir
,
10993 .iterate_shared
= btrfs_real_readdir
,
10994 .open
= btrfs_opendir
,
10995 .unlocked_ioctl
= btrfs_ioctl
,
10996 #ifdef CONFIG_COMPAT
10997 .compat_ioctl
= btrfs_compat_ioctl
,
10999 .release
= btrfs_release_file
,
11000 .fsync
= btrfs_sync_file
,
11004 * btrfs doesn't support the bmap operation because swapfiles
11005 * use bmap to make a mapping of extents in the file. They assume
11006 * these extents won't change over the life of the file and they
11007 * use the bmap result to do IO directly to the drive.
11009 * the btrfs bmap call would return logical addresses that aren't
11010 * suitable for IO and they also will change frequently as COW
11011 * operations happen. So, swapfile + btrfs == corruption.
11013 * For now we're avoiding this by dropping bmap.
11015 static const struct address_space_operations btrfs_aops
= {
11016 .read_folio
= btrfs_read_folio
,
11017 .writepages
= btrfs_writepages
,
11018 .readahead
= btrfs_readahead
,
11019 .direct_IO
= noop_direct_IO
,
11020 .invalidate_folio
= btrfs_invalidate_folio
,
11021 .release_folio
= btrfs_release_folio
,
11022 .migrate_folio
= btrfs_migrate_folio
,
11023 .dirty_folio
= filemap_dirty_folio
,
11024 .error_remove_page
= generic_error_remove_page
,
11025 .swap_activate
= btrfs_swap_activate
,
11026 .swap_deactivate
= btrfs_swap_deactivate
,
11029 static const struct inode_operations btrfs_file_inode_operations
= {
11030 .getattr
= btrfs_getattr
,
11031 .setattr
= btrfs_setattr
,
11032 .listxattr
= btrfs_listxattr
,
11033 .permission
= btrfs_permission
,
11034 .fiemap
= btrfs_fiemap
,
11035 .get_inode_acl
= btrfs_get_acl
,
11036 .set_acl
= btrfs_set_acl
,
11037 .update_time
= btrfs_update_time
,
11038 .fileattr_get
= btrfs_fileattr_get
,
11039 .fileattr_set
= btrfs_fileattr_set
,
11041 static const struct inode_operations btrfs_special_inode_operations
= {
11042 .getattr
= btrfs_getattr
,
11043 .setattr
= btrfs_setattr
,
11044 .permission
= btrfs_permission
,
11045 .listxattr
= btrfs_listxattr
,
11046 .get_inode_acl
= btrfs_get_acl
,
11047 .set_acl
= btrfs_set_acl
,
11048 .update_time
= btrfs_update_time
,
11050 static const struct inode_operations btrfs_symlink_inode_operations
= {
11051 .get_link
= page_get_link
,
11052 .getattr
= btrfs_getattr
,
11053 .setattr
= btrfs_setattr
,
11054 .permission
= btrfs_permission
,
11055 .listxattr
= btrfs_listxattr
,
11056 .update_time
= btrfs_update_time
,
11059 const struct dentry_operations btrfs_dentry_operations
= {
11060 .d_delete
= btrfs_dentry_delete
,