2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
64 struct btrfs_iget_args
{
65 struct btrfs_key
*location
;
66 struct btrfs_root
*root
;
69 struct btrfs_dio_data
{
70 u64 outstanding_extents
;
72 u64 unsubmitted_oe_range_start
;
73 u64 unsubmitted_oe_range_end
;
76 static const struct inode_operations btrfs_dir_inode_operations
;
77 static const struct inode_operations btrfs_symlink_inode_operations
;
78 static const struct inode_operations btrfs_dir_ro_inode_operations
;
79 static const struct inode_operations btrfs_special_inode_operations
;
80 static const struct inode_operations btrfs_file_inode_operations
;
81 static const struct address_space_operations btrfs_aops
;
82 static const struct address_space_operations btrfs_symlink_aops
;
83 static const struct file_operations btrfs_dir_file_operations
;
84 static const struct extent_io_ops btrfs_extent_io_ops
;
86 static struct kmem_cache
*btrfs_inode_cachep
;
87 struct kmem_cache
*btrfs_trans_handle_cachep
;
88 struct kmem_cache
*btrfs_transaction_cachep
;
89 struct kmem_cache
*btrfs_path_cachep
;
90 struct kmem_cache
*btrfs_free_space_cachep
;
93 static const unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
94 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
95 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
96 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
97 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
98 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
99 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
100 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
103 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
104 static int btrfs_truncate(struct inode
*inode
);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
106 static noinline
int cow_file_range(struct inode
*inode
,
107 struct page
*locked_page
,
108 u64 start
, u64 end
, int *page_started
,
109 unsigned long *nr_written
, int unlock
);
110 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
111 u64 len
, u64 orig_start
,
112 u64 block_start
, u64 block_len
,
113 u64 orig_block_len
, u64 ram_bytes
,
116 static int btrfs_dirty_inode(struct inode
*inode
);
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode
*inode
)
121 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
125 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
126 struct inode
*inode
, struct inode
*dir
,
127 const struct qstr
*qstr
)
131 err
= btrfs_init_acl(trans
, inode
, dir
);
133 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
142 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
143 struct btrfs_path
*path
, int extent_inserted
,
144 struct btrfs_root
*root
, struct inode
*inode
,
145 u64 start
, size_t size
, size_t compressed_size
,
147 struct page
**compressed_pages
)
149 struct extent_buffer
*leaf
;
150 struct page
*page
= NULL
;
153 struct btrfs_file_extent_item
*ei
;
156 size_t cur_size
= size
;
157 unsigned long offset
;
159 if (compressed_size
&& compressed_pages
)
160 cur_size
= compressed_size
;
162 inode_add_bytes(inode
, size
);
164 if (!extent_inserted
) {
165 struct btrfs_key key
;
168 key
.objectid
= btrfs_ino(inode
);
170 key
.type
= BTRFS_EXTENT_DATA_KEY
;
172 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
173 path
->leave_spinning
= 1;
174 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
181 leaf
= path
->nodes
[0];
182 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
183 struct btrfs_file_extent_item
);
184 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
185 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
186 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
187 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
188 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
189 ptr
= btrfs_file_extent_inline_start(ei
);
191 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
194 while (compressed_size
> 0) {
195 cpage
= compressed_pages
[i
];
196 cur_size
= min_t(unsigned long, compressed_size
,
199 kaddr
= kmap_atomic(cpage
);
200 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
201 kunmap_atomic(kaddr
);
205 compressed_size
-= cur_size
;
207 btrfs_set_file_extent_compression(leaf
, ei
,
210 page
= find_get_page(inode
->i_mapping
,
211 start
>> PAGE_CACHE_SHIFT
);
212 btrfs_set_file_extent_compression(leaf
, ei
, 0);
213 kaddr
= kmap_atomic(page
);
214 offset
= start
& (PAGE_CACHE_SIZE
- 1);
215 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
216 kunmap_atomic(kaddr
);
217 page_cache_release(page
);
219 btrfs_mark_buffer_dirty(leaf
);
220 btrfs_release_path(path
);
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
231 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
232 ret
= btrfs_update_inode(trans
, root
, inode
);
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
245 static noinline
int cow_file_range_inline(struct btrfs_root
*root
,
246 struct inode
*inode
, u64 start
,
247 u64 end
, size_t compressed_size
,
249 struct page
**compressed_pages
)
251 struct btrfs_trans_handle
*trans
;
252 u64 isize
= i_size_read(inode
);
253 u64 actual_end
= min(end
+ 1, isize
);
254 u64 inline_len
= actual_end
- start
;
255 u64 aligned_end
= ALIGN(end
, root
->sectorsize
);
256 u64 data_len
= inline_len
;
258 struct btrfs_path
*path
;
259 int extent_inserted
= 0;
260 u32 extent_item_size
;
263 data_len
= compressed_size
;
266 actual_end
> PAGE_CACHE_SIZE
||
267 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
269 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
271 data_len
> root
->fs_info
->max_inline
) {
275 path
= btrfs_alloc_path();
279 trans
= btrfs_join_transaction(root
);
281 btrfs_free_path(path
);
282 return PTR_ERR(trans
);
284 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
286 if (compressed_size
&& compressed_pages
)
287 extent_item_size
= btrfs_file_extent_calc_inline_size(
290 extent_item_size
= btrfs_file_extent_calc_inline_size(
293 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
294 start
, aligned_end
, NULL
,
295 1, 1, extent_item_size
, &extent_inserted
);
297 btrfs_abort_transaction(trans
, root
, ret
);
301 if (isize
> actual_end
)
302 inline_len
= min_t(u64
, isize
, actual_end
);
303 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
305 inline_len
, compressed_size
,
306 compress_type
, compressed_pages
);
307 if (ret
&& ret
!= -ENOSPC
) {
308 btrfs_abort_transaction(trans
, root
, ret
);
310 } else if (ret
== -ENOSPC
) {
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
316 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
317 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
325 btrfs_qgroup_free_data(inode
, 0, PAGE_CACHE_SIZE
);
326 btrfs_free_path(path
);
327 btrfs_end_transaction(trans
, root
);
331 struct async_extent
{
336 unsigned long nr_pages
;
338 struct list_head list
;
343 struct btrfs_root
*root
;
344 struct page
*locked_page
;
347 struct list_head extents
;
348 struct btrfs_work work
;
351 static noinline
int add_async_extent(struct async_cow
*cow
,
352 u64 start
, u64 ram_size
,
355 unsigned long nr_pages
,
358 struct async_extent
*async_extent
;
360 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
361 BUG_ON(!async_extent
); /* -ENOMEM */
362 async_extent
->start
= start
;
363 async_extent
->ram_size
= ram_size
;
364 async_extent
->compressed_size
= compressed_size
;
365 async_extent
->pages
= pages
;
366 async_extent
->nr_pages
= nr_pages
;
367 async_extent
->compress_type
= compress_type
;
368 list_add_tail(&async_extent
->list
, &cow
->extents
);
372 static inline int inode_need_compress(struct inode
*inode
)
374 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
377 if (btrfs_test_opt(root
, FORCE_COMPRESS
))
379 /* bad compression ratios */
380 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
382 if (btrfs_test_opt(root
, COMPRESS
) ||
383 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
384 BTRFS_I(inode
)->force_compress
)
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
406 static noinline
void compress_file_range(struct inode
*inode
,
407 struct page
*locked_page
,
409 struct async_cow
*async_cow
,
412 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
414 u64 blocksize
= root
->sectorsize
;
416 u64 isize
= i_size_read(inode
);
418 struct page
**pages
= NULL
;
419 unsigned long nr_pages
;
420 unsigned long nr_pages_ret
= 0;
421 unsigned long total_compressed
= 0;
422 unsigned long total_in
= 0;
423 unsigned long max_compressed
= SZ_128K
;
424 unsigned long max_uncompressed
= SZ_128K
;
427 int compress_type
= root
->fs_info
->compress_type
;
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end
- start
+ 1) < SZ_16K
&&
432 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
433 btrfs_add_inode_defrag(NULL
, inode
);
435 actual_end
= min_t(u64
, isize
, end
+ 1);
438 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
439 nr_pages
= min_t(unsigned long, nr_pages
, SZ_128K
/ PAGE_CACHE_SIZE
);
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
451 if (actual_end
<= start
)
452 goto cleanup_and_bail_uncompressed
;
454 total_compressed
= actual_end
- start
;
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it dosen't save disk space at all.
460 if (total_compressed
<= blocksize
&&
461 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
462 goto cleanup_and_bail_uncompressed
;
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
474 total_compressed
= min(total_compressed
, max_uncompressed
);
475 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
476 num_bytes
= max(blocksize
, num_bytes
);
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
485 if (inode_need_compress(inode
)) {
487 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
489 /* just bail out to the uncompressed code */
493 if (BTRFS_I(inode
)->force_compress
)
494 compress_type
= BTRFS_I(inode
)->force_compress
;
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
505 extent_range_clear_dirty_for_io(inode
, start
, end
);
507 ret
= btrfs_compress_pages(compress_type
,
508 inode
->i_mapping
, start
,
509 total_compressed
, pages
,
510 nr_pages
, &nr_pages_ret
,
516 unsigned long offset
= total_compressed
&
517 (PAGE_CACHE_SIZE
- 1);
518 struct page
*page
= pages
[nr_pages_ret
- 1];
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
525 kaddr
= kmap_atomic(page
);
526 memset(kaddr
+ offset
, 0,
527 PAGE_CACHE_SIZE
- offset
);
528 kunmap_atomic(kaddr
);
535 /* lets try to make an inline extent */
536 if (ret
|| total_in
< (actual_end
- start
)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
540 ret
= cow_file_range_inline(root
, inode
, start
, end
,
543 /* try making a compressed inline extent */
544 ret
= cow_file_range_inline(root
, inode
, start
, end
,
546 compress_type
, pages
);
549 unsigned long clear_flags
= EXTENT_DELALLOC
|
551 unsigned long page_error_op
;
553 clear_flags
|= (ret
< 0) ? EXTENT_DO_ACCOUNTING
: 0;
554 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
561 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
562 clear_flags
, PAGE_UNLOCK
|
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
577 total_compressed
= ALIGN(total_compressed
, blocksize
);
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
583 total_in
= ALIGN(total_in
, PAGE_CACHE_SIZE
);
584 if (total_compressed
>= total_in
) {
587 num_bytes
= total_in
;
590 if (!will_compress
&& pages
) {
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
595 for (i
= 0; i
< nr_pages_ret
; i
++) {
596 WARN_ON(pages
[i
]->mapping
);
597 page_cache_release(pages
[i
]);
601 total_compressed
= 0;
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
606 !(BTRFS_I(inode
)->force_compress
)) {
607 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
617 add_async_extent(async_cow
, start
, num_bytes
,
618 total_compressed
, pages
, nr_pages_ret
,
621 if (start
+ num_bytes
< end
) {
628 cleanup_and_bail_uncompressed
:
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
636 if (page_offset(locked_page
) >= start
&&
637 page_offset(locked_page
) <= end
) {
638 __set_page_dirty_nobuffers(locked_page
);
639 /* unlocked later on in the async handlers */
642 extent_range_redirty_for_io(inode
, start
, end
);
643 add_async_extent(async_cow
, start
, end
- start
+ 1,
644 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
651 for (i
= 0; i
< nr_pages_ret
; i
++) {
652 WARN_ON(pages
[i
]->mapping
);
653 page_cache_release(pages
[i
]);
658 static void free_async_extent_pages(struct async_extent
*async_extent
)
662 if (!async_extent
->pages
)
665 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
666 WARN_ON(async_extent
->pages
[i
]->mapping
);
667 page_cache_release(async_extent
->pages
[i
]);
669 kfree(async_extent
->pages
);
670 async_extent
->nr_pages
= 0;
671 async_extent
->pages
= NULL
;
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
680 static noinline
void submit_compressed_extents(struct inode
*inode
,
681 struct async_cow
*async_cow
)
683 struct async_extent
*async_extent
;
685 struct btrfs_key ins
;
686 struct extent_map
*em
;
687 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
688 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
689 struct extent_io_tree
*io_tree
;
693 while (!list_empty(&async_cow
->extents
)) {
694 async_extent
= list_entry(async_cow
->extents
.next
,
695 struct async_extent
, list
);
696 list_del(&async_extent
->list
);
698 io_tree
= &BTRFS_I(inode
)->io_tree
;
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent
->pages
) {
703 int page_started
= 0;
704 unsigned long nr_written
= 0;
706 lock_extent(io_tree
, async_extent
->start
,
707 async_extent
->start
+
708 async_extent
->ram_size
- 1);
710 /* allocate blocks */
711 ret
= cow_file_range(inode
, async_cow
->locked_page
,
713 async_extent
->start
+
714 async_extent
->ram_size
- 1,
715 &page_started
, &nr_written
, 0);
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
725 if (!page_started
&& !ret
)
726 extent_write_locked_range(io_tree
,
727 inode
, async_extent
->start
,
728 async_extent
->start
+
729 async_extent
->ram_size
- 1,
733 unlock_page(async_cow
->locked_page
);
739 lock_extent(io_tree
, async_extent
->start
,
740 async_extent
->start
+ async_extent
->ram_size
- 1);
742 ret
= btrfs_reserve_extent(root
,
743 async_extent
->compressed_size
,
744 async_extent
->compressed_size
,
745 0, alloc_hint
, &ins
, 1, 1);
747 free_async_extent_pages(async_extent
);
749 if (ret
== -ENOSPC
) {
750 unlock_extent(io_tree
, async_extent
->start
,
751 async_extent
->start
+
752 async_extent
->ram_size
- 1);
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
760 extent_range_redirty_for_io(inode
,
762 async_extent
->start
+
763 async_extent
->ram_size
- 1);
770 * here we're doing allocation and writeback of the
773 btrfs_drop_extent_cache(inode
, async_extent
->start
,
774 async_extent
->start
+
775 async_extent
->ram_size
- 1, 0);
777 em
= alloc_extent_map();
780 goto out_free_reserve
;
782 em
->start
= async_extent
->start
;
783 em
->len
= async_extent
->ram_size
;
784 em
->orig_start
= em
->start
;
785 em
->mod_start
= em
->start
;
786 em
->mod_len
= em
->len
;
788 em
->block_start
= ins
.objectid
;
789 em
->block_len
= ins
.offset
;
790 em
->orig_block_len
= ins
.offset
;
791 em
->ram_bytes
= async_extent
->ram_size
;
792 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
793 em
->compress_type
= async_extent
->compress_type
;
794 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
795 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
799 write_lock(&em_tree
->lock
);
800 ret
= add_extent_mapping(em_tree
, em
, 1);
801 write_unlock(&em_tree
->lock
);
802 if (ret
!= -EEXIST
) {
806 btrfs_drop_extent_cache(inode
, async_extent
->start
,
807 async_extent
->start
+
808 async_extent
->ram_size
- 1, 0);
812 goto out_free_reserve
;
814 ret
= btrfs_add_ordered_extent_compress(inode
,
817 async_extent
->ram_size
,
819 BTRFS_ORDERED_COMPRESSED
,
820 async_extent
->compress_type
);
822 btrfs_drop_extent_cache(inode
, async_extent
->start
,
823 async_extent
->start
+
824 async_extent
->ram_size
- 1, 0);
825 goto out_free_reserve
;
829 * clear dirty, set writeback and unlock the pages.
831 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
832 async_extent
->start
+
833 async_extent
->ram_size
- 1,
834 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
835 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
837 ret
= btrfs_submit_compressed_write(inode
,
839 async_extent
->ram_size
,
841 ins
.offset
, async_extent
->pages
,
842 async_extent
->nr_pages
);
844 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
845 struct page
*p
= async_extent
->pages
[0];
846 const u64 start
= async_extent
->start
;
847 const u64 end
= start
+ async_extent
->ram_size
- 1;
849 p
->mapping
= inode
->i_mapping
;
850 tree
->ops
->writepage_end_io_hook(p
, start
, end
,
853 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
, 0,
856 free_async_extent_pages(async_extent
);
858 alloc_hint
= ins
.objectid
+ ins
.offset
;
864 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
866 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
867 async_extent
->start
+
868 async_extent
->ram_size
- 1,
869 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
870 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
871 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
872 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
874 free_async_extent_pages(async_extent
);
879 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
882 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
883 struct extent_map
*em
;
886 read_lock(&em_tree
->lock
);
887 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
894 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
896 em
= search_extent_mapping(em_tree
, 0, 0);
897 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
898 alloc_hint
= em
->block_start
;
902 alloc_hint
= em
->block_start
;
906 read_unlock(&em_tree
->lock
);
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
924 static noinline
int cow_file_range(struct inode
*inode
,
925 struct page
*locked_page
,
926 u64 start
, u64 end
, int *page_started
,
927 unsigned long *nr_written
,
930 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
933 unsigned long ram_size
;
936 u64 blocksize
= root
->sectorsize
;
937 struct btrfs_key ins
;
938 struct extent_map
*em
;
939 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
942 if (btrfs_is_free_space_inode(inode
)) {
948 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
949 num_bytes
= max(blocksize
, num_bytes
);
950 disk_num_bytes
= num_bytes
;
952 /* if this is a small write inside eof, kick off defrag */
953 if (num_bytes
< SZ_64K
&&
954 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
955 btrfs_add_inode_defrag(NULL
, inode
);
958 /* lets try to make an inline extent */
959 ret
= cow_file_range_inline(root
, inode
, start
, end
, 0, 0,
962 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
963 EXTENT_LOCKED
| EXTENT_DELALLOC
|
964 EXTENT_DEFRAG
, PAGE_UNLOCK
|
965 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
968 *nr_written
= *nr_written
+
969 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
972 } else if (ret
< 0) {
977 BUG_ON(disk_num_bytes
>
978 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
980 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
981 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
983 while (disk_num_bytes
> 0) {
986 cur_alloc_size
= disk_num_bytes
;
987 ret
= btrfs_reserve_extent(root
, cur_alloc_size
,
988 root
->sectorsize
, 0, alloc_hint
,
993 em
= alloc_extent_map();
999 em
->orig_start
= em
->start
;
1000 ram_size
= ins
.offset
;
1001 em
->len
= ins
.offset
;
1002 em
->mod_start
= em
->start
;
1003 em
->mod_len
= em
->len
;
1005 em
->block_start
= ins
.objectid
;
1006 em
->block_len
= ins
.offset
;
1007 em
->orig_block_len
= ins
.offset
;
1008 em
->ram_bytes
= ram_size
;
1009 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1010 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1011 em
->generation
= -1;
1014 write_lock(&em_tree
->lock
);
1015 ret
= add_extent_mapping(em_tree
, em
, 1);
1016 write_unlock(&em_tree
->lock
);
1017 if (ret
!= -EEXIST
) {
1018 free_extent_map(em
);
1021 btrfs_drop_extent_cache(inode
, start
,
1022 start
+ ram_size
- 1, 0);
1027 cur_alloc_size
= ins
.offset
;
1028 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1029 ram_size
, cur_alloc_size
, 0);
1031 goto out_drop_extent_cache
;
1033 if (root
->root_key
.objectid
==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1035 ret
= btrfs_reloc_clone_csums(inode
, start
,
1038 goto out_drop_extent_cache
;
1041 if (disk_num_bytes
< cur_alloc_size
)
1044 /* we're not doing compressed IO, don't unlock the first
1045 * page (which the caller expects to stay locked), don't
1046 * clear any dirty bits and don't set any writeback bits
1048 * Do set the Private2 bit so we know this page was properly
1049 * setup for writepage
1051 op
= unlock
? PAGE_UNLOCK
: 0;
1052 op
|= PAGE_SET_PRIVATE2
;
1054 extent_clear_unlock_delalloc(inode
, start
,
1055 start
+ ram_size
- 1, locked_page
,
1056 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1058 disk_num_bytes
-= cur_alloc_size
;
1059 num_bytes
-= cur_alloc_size
;
1060 alloc_hint
= ins
.objectid
+ ins
.offset
;
1061 start
+= cur_alloc_size
;
1066 out_drop_extent_cache
:
1067 btrfs_drop_extent_cache(inode
, start
, start
+ ram_size
- 1, 0);
1069 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
1071 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1072 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
1073 EXTENT_DELALLOC
| EXTENT_DEFRAG
,
1074 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
1075 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
);
1080 * work queue call back to started compression on a file and pages
1082 static noinline
void async_cow_start(struct btrfs_work
*work
)
1084 struct async_cow
*async_cow
;
1086 async_cow
= container_of(work
, struct async_cow
, work
);
1088 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1089 async_cow
->start
, async_cow
->end
, async_cow
,
1091 if (num_added
== 0) {
1092 btrfs_add_delayed_iput(async_cow
->inode
);
1093 async_cow
->inode
= NULL
;
1098 * work queue call back to submit previously compressed pages
1100 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1102 struct async_cow
*async_cow
;
1103 struct btrfs_root
*root
;
1104 unsigned long nr_pages
;
1106 async_cow
= container_of(work
, struct async_cow
, work
);
1108 root
= async_cow
->root
;
1109 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1113 * atomic_sub_return implies a barrier for waitqueue_active
1115 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1117 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1118 wake_up(&root
->fs_info
->async_submit_wait
);
1120 if (async_cow
->inode
)
1121 submit_compressed_extents(async_cow
->inode
, async_cow
);
1124 static noinline
void async_cow_free(struct btrfs_work
*work
)
1126 struct async_cow
*async_cow
;
1127 async_cow
= container_of(work
, struct async_cow
, work
);
1128 if (async_cow
->inode
)
1129 btrfs_add_delayed_iput(async_cow
->inode
);
1133 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1134 u64 start
, u64 end
, int *page_started
,
1135 unsigned long *nr_written
)
1137 struct async_cow
*async_cow
;
1138 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1139 unsigned long nr_pages
;
1141 int limit
= 10 * SZ_1M
;
1143 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1144 1, 0, NULL
, GFP_NOFS
);
1145 while (start
< end
) {
1146 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1147 BUG_ON(!async_cow
); /* -ENOMEM */
1148 async_cow
->inode
= igrab(inode
);
1149 async_cow
->root
= root
;
1150 async_cow
->locked_page
= locked_page
;
1151 async_cow
->start
= start
;
1153 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1154 !btrfs_test_opt(root
, FORCE_COMPRESS
))
1157 cur_end
= min(end
, start
+ SZ_512K
- 1);
1159 async_cow
->end
= cur_end
;
1160 INIT_LIST_HEAD(&async_cow
->extents
);
1162 btrfs_init_work(&async_cow
->work
,
1163 btrfs_delalloc_helper
,
1164 async_cow_start
, async_cow_submit
,
1167 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1169 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1171 btrfs_queue_work(root
->fs_info
->delalloc_workers
,
1174 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1175 wait_event(root
->fs_info
->async_submit_wait
,
1176 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1180 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1181 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1182 wait_event(root
->fs_info
->async_submit_wait
,
1183 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1187 *nr_written
+= nr_pages
;
1188 start
= cur_end
+ 1;
1194 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1195 u64 bytenr
, u64 num_bytes
)
1198 struct btrfs_ordered_sum
*sums
;
1201 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1202 bytenr
+ num_bytes
- 1, &list
, 0);
1203 if (ret
== 0 && list_empty(&list
))
1206 while (!list_empty(&list
)) {
1207 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1208 list_del(&sums
->list
);
1215 * when nowcow writeback call back. This checks for snapshots or COW copies
1216 * of the extents that exist in the file, and COWs the file as required.
1218 * If no cow copies or snapshots exist, we write directly to the existing
1221 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1222 struct page
*locked_page
,
1223 u64 start
, u64 end
, int *page_started
, int force
,
1224 unsigned long *nr_written
)
1226 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1227 struct btrfs_trans_handle
*trans
;
1228 struct extent_buffer
*leaf
;
1229 struct btrfs_path
*path
;
1230 struct btrfs_file_extent_item
*fi
;
1231 struct btrfs_key found_key
;
1246 u64 ino
= btrfs_ino(inode
);
1248 path
= btrfs_alloc_path();
1250 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1251 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1252 EXTENT_DO_ACCOUNTING
|
1253 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1255 PAGE_SET_WRITEBACK
|
1256 PAGE_END_WRITEBACK
);
1260 nolock
= btrfs_is_free_space_inode(inode
);
1263 trans
= btrfs_join_transaction_nolock(root
);
1265 trans
= btrfs_join_transaction(root
);
1267 if (IS_ERR(trans
)) {
1268 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1269 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1270 EXTENT_DO_ACCOUNTING
|
1271 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1273 PAGE_SET_WRITEBACK
|
1274 PAGE_END_WRITEBACK
);
1275 btrfs_free_path(path
);
1276 return PTR_ERR(trans
);
1279 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1281 cow_start
= (u64
)-1;
1284 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1288 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1289 leaf
= path
->nodes
[0];
1290 btrfs_item_key_to_cpu(leaf
, &found_key
,
1291 path
->slots
[0] - 1);
1292 if (found_key
.objectid
== ino
&&
1293 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1298 leaf
= path
->nodes
[0];
1299 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1300 ret
= btrfs_next_leaf(root
, path
);
1305 leaf
= path
->nodes
[0];
1311 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1313 if (found_key
.objectid
> ino
)
1315 if (WARN_ON_ONCE(found_key
.objectid
< ino
) ||
1316 found_key
.type
< BTRFS_EXTENT_DATA_KEY
) {
1320 if (found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1321 found_key
.offset
> end
)
1324 if (found_key
.offset
> cur_offset
) {
1325 extent_end
= found_key
.offset
;
1330 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1331 struct btrfs_file_extent_item
);
1332 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1334 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1335 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1336 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1337 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1338 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1339 extent_end
= found_key
.offset
+
1340 btrfs_file_extent_num_bytes(leaf
, fi
);
1342 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1343 if (extent_end
<= start
) {
1347 if (disk_bytenr
== 0)
1349 if (btrfs_file_extent_compression(leaf
, fi
) ||
1350 btrfs_file_extent_encryption(leaf
, fi
) ||
1351 btrfs_file_extent_other_encoding(leaf
, fi
))
1353 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1355 if (btrfs_extent_readonly(root
, disk_bytenr
))
1357 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1359 extent_offset
, disk_bytenr
))
1361 disk_bytenr
+= extent_offset
;
1362 disk_bytenr
+= cur_offset
- found_key
.offset
;
1363 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1365 * if there are pending snapshots for this root,
1366 * we fall into common COW way.
1369 err
= btrfs_start_write_no_snapshoting(root
);
1374 * force cow if csum exists in the range.
1375 * this ensure that csum for a given extent are
1376 * either valid or do not exist.
1378 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1381 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1382 extent_end
= found_key
.offset
+
1383 btrfs_file_extent_inline_len(leaf
,
1384 path
->slots
[0], fi
);
1385 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1390 if (extent_end
<= start
) {
1392 if (!nolock
&& nocow
)
1393 btrfs_end_write_no_snapshoting(root
);
1397 if (cow_start
== (u64
)-1)
1398 cow_start
= cur_offset
;
1399 cur_offset
= extent_end
;
1400 if (cur_offset
> end
)
1406 btrfs_release_path(path
);
1407 if (cow_start
!= (u64
)-1) {
1408 ret
= cow_file_range(inode
, locked_page
,
1409 cow_start
, found_key
.offset
- 1,
1410 page_started
, nr_written
, 1);
1412 if (!nolock
&& nocow
)
1413 btrfs_end_write_no_snapshoting(root
);
1416 cow_start
= (u64
)-1;
1419 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1420 struct extent_map
*em
;
1421 struct extent_map_tree
*em_tree
;
1422 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1423 em
= alloc_extent_map();
1424 BUG_ON(!em
); /* -ENOMEM */
1425 em
->start
= cur_offset
;
1426 em
->orig_start
= found_key
.offset
- extent_offset
;
1427 em
->len
= num_bytes
;
1428 em
->block_len
= num_bytes
;
1429 em
->block_start
= disk_bytenr
;
1430 em
->orig_block_len
= disk_num_bytes
;
1431 em
->ram_bytes
= ram_bytes
;
1432 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1433 em
->mod_start
= em
->start
;
1434 em
->mod_len
= em
->len
;
1435 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1436 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1437 em
->generation
= -1;
1439 write_lock(&em_tree
->lock
);
1440 ret
= add_extent_mapping(em_tree
, em
, 1);
1441 write_unlock(&em_tree
->lock
);
1442 if (ret
!= -EEXIST
) {
1443 free_extent_map(em
);
1446 btrfs_drop_extent_cache(inode
, em
->start
,
1447 em
->start
+ em
->len
- 1, 0);
1449 type
= BTRFS_ORDERED_PREALLOC
;
1451 type
= BTRFS_ORDERED_NOCOW
;
1454 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1455 num_bytes
, num_bytes
, type
);
1456 BUG_ON(ret
); /* -ENOMEM */
1458 if (root
->root_key
.objectid
==
1459 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1460 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1463 if (!nolock
&& nocow
)
1464 btrfs_end_write_no_snapshoting(root
);
1469 extent_clear_unlock_delalloc(inode
, cur_offset
,
1470 cur_offset
+ num_bytes
- 1,
1471 locked_page
, EXTENT_LOCKED
|
1472 EXTENT_DELALLOC
, PAGE_UNLOCK
|
1474 if (!nolock
&& nocow
)
1475 btrfs_end_write_no_snapshoting(root
);
1476 cur_offset
= extent_end
;
1477 if (cur_offset
> end
)
1480 btrfs_release_path(path
);
1482 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1483 cow_start
= cur_offset
;
1487 if (cow_start
!= (u64
)-1) {
1488 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1489 page_started
, nr_written
, 1);
1495 err
= btrfs_end_transaction(trans
, root
);
1499 if (ret
&& cur_offset
< end
)
1500 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1501 locked_page
, EXTENT_LOCKED
|
1502 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1503 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1505 PAGE_SET_WRITEBACK
|
1506 PAGE_END_WRITEBACK
);
1507 btrfs_free_path(path
);
1511 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1514 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1515 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1519 * @defrag_bytes is a hint value, no spinlock held here,
1520 * if is not zero, it means the file is defragging.
1521 * Force cow if given extent needs to be defragged.
1523 if (BTRFS_I(inode
)->defrag_bytes
&&
1524 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1525 EXTENT_DEFRAG
, 0, NULL
))
1532 * extent_io.c call back to do delayed allocation processing
1534 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1535 u64 start
, u64 end
, int *page_started
,
1536 unsigned long *nr_written
)
1539 int force_cow
= need_force_cow(inode
, start
, end
);
1541 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1542 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1543 page_started
, 1, nr_written
);
1544 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1545 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1546 page_started
, 0, nr_written
);
1547 } else if (!inode_need_compress(inode
)) {
1548 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1549 page_started
, nr_written
, 1);
1551 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1552 &BTRFS_I(inode
)->runtime_flags
);
1553 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1554 page_started
, nr_written
);
1559 static void btrfs_split_extent_hook(struct inode
*inode
,
1560 struct extent_state
*orig
, u64 split
)
1564 /* not delalloc, ignore it */
1565 if (!(orig
->state
& EXTENT_DELALLOC
))
1568 size
= orig
->end
- orig
->start
+ 1;
1569 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1574 * See the explanation in btrfs_merge_extent_hook, the same
1575 * applies here, just in reverse.
1577 new_size
= orig
->end
- split
+ 1;
1578 num_extents
= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1579 BTRFS_MAX_EXTENT_SIZE
);
1580 new_size
= split
- orig
->start
;
1581 num_extents
+= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1582 BTRFS_MAX_EXTENT_SIZE
);
1583 if (div64_u64(size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1584 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1588 spin_lock(&BTRFS_I(inode
)->lock
);
1589 BTRFS_I(inode
)->outstanding_extents
++;
1590 spin_unlock(&BTRFS_I(inode
)->lock
);
1594 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595 * extents so we can keep track of new extents that are just merged onto old
1596 * extents, such as when we are doing sequential writes, so we can properly
1597 * account for the metadata space we'll need.
1599 static void btrfs_merge_extent_hook(struct inode
*inode
,
1600 struct extent_state
*new,
1601 struct extent_state
*other
)
1603 u64 new_size
, old_size
;
1606 /* not delalloc, ignore it */
1607 if (!(other
->state
& EXTENT_DELALLOC
))
1610 if (new->start
> other
->start
)
1611 new_size
= new->end
- other
->start
+ 1;
1613 new_size
= other
->end
- new->start
+ 1;
1615 /* we're not bigger than the max, unreserve the space and go */
1616 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1617 spin_lock(&BTRFS_I(inode
)->lock
);
1618 BTRFS_I(inode
)->outstanding_extents
--;
1619 spin_unlock(&BTRFS_I(inode
)->lock
);
1624 * We have to add up either side to figure out how many extents were
1625 * accounted for before we merged into one big extent. If the number of
1626 * extents we accounted for is <= the amount we need for the new range
1627 * then we can return, otherwise drop. Think of it like this
1631 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632 * need 2 outstanding extents, on one side we have 1 and the other side
1633 * we have 1 so they are == and we can return. But in this case
1635 * [MAX_SIZE+4k][MAX_SIZE+4k]
1637 * Each range on their own accounts for 2 extents, but merged together
1638 * they are only 3 extents worth of accounting, so we need to drop in
1641 old_size
= other
->end
- other
->start
+ 1;
1642 num_extents
= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1643 BTRFS_MAX_EXTENT_SIZE
);
1644 old_size
= new->end
- new->start
+ 1;
1645 num_extents
+= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1646 BTRFS_MAX_EXTENT_SIZE
);
1648 if (div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1649 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1652 spin_lock(&BTRFS_I(inode
)->lock
);
1653 BTRFS_I(inode
)->outstanding_extents
--;
1654 spin_unlock(&BTRFS_I(inode
)->lock
);
1657 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1658 struct inode
*inode
)
1660 spin_lock(&root
->delalloc_lock
);
1661 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1662 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1663 &root
->delalloc_inodes
);
1664 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1665 &BTRFS_I(inode
)->runtime_flags
);
1666 root
->nr_delalloc_inodes
++;
1667 if (root
->nr_delalloc_inodes
== 1) {
1668 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1669 BUG_ON(!list_empty(&root
->delalloc_root
));
1670 list_add_tail(&root
->delalloc_root
,
1671 &root
->fs_info
->delalloc_roots
);
1672 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1675 spin_unlock(&root
->delalloc_lock
);
1678 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1679 struct inode
*inode
)
1681 spin_lock(&root
->delalloc_lock
);
1682 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1683 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1684 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1685 &BTRFS_I(inode
)->runtime_flags
);
1686 root
->nr_delalloc_inodes
--;
1687 if (!root
->nr_delalloc_inodes
) {
1688 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1689 BUG_ON(list_empty(&root
->delalloc_root
));
1690 list_del_init(&root
->delalloc_root
);
1691 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1694 spin_unlock(&root
->delalloc_lock
);
1698 * extent_io.c set_bit_hook, used to track delayed allocation
1699 * bytes in this file, and to maintain the list of inodes that
1700 * have pending delalloc work to be done.
1702 static void btrfs_set_bit_hook(struct inode
*inode
,
1703 struct extent_state
*state
, unsigned *bits
)
1706 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1709 * set_bit and clear bit hooks normally require _irqsave/restore
1710 * but in this case, we are only testing for the DELALLOC
1711 * bit, which is only set or cleared with irqs on
1713 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1714 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1715 u64 len
= state
->end
+ 1 - state
->start
;
1716 bool do_list
= !btrfs_is_free_space_inode(inode
);
1718 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1719 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1721 spin_lock(&BTRFS_I(inode
)->lock
);
1722 BTRFS_I(inode
)->outstanding_extents
++;
1723 spin_unlock(&BTRFS_I(inode
)->lock
);
1726 /* For sanity tests */
1727 if (btrfs_test_is_dummy_root(root
))
1730 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1731 root
->fs_info
->delalloc_batch
);
1732 spin_lock(&BTRFS_I(inode
)->lock
);
1733 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1734 if (*bits
& EXTENT_DEFRAG
)
1735 BTRFS_I(inode
)->defrag_bytes
+= len
;
1736 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1737 &BTRFS_I(inode
)->runtime_flags
))
1738 btrfs_add_delalloc_inodes(root
, inode
);
1739 spin_unlock(&BTRFS_I(inode
)->lock
);
1744 * extent_io.c clear_bit_hook, see set_bit_hook for why
1746 static void btrfs_clear_bit_hook(struct inode
*inode
,
1747 struct extent_state
*state
,
1750 u64 len
= state
->end
+ 1 - state
->start
;
1751 u64 num_extents
= div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
-1,
1752 BTRFS_MAX_EXTENT_SIZE
);
1754 spin_lock(&BTRFS_I(inode
)->lock
);
1755 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
))
1756 BTRFS_I(inode
)->defrag_bytes
-= len
;
1757 spin_unlock(&BTRFS_I(inode
)->lock
);
1760 * set_bit and clear bit hooks normally require _irqsave/restore
1761 * but in this case, we are only testing for the DELALLOC
1762 * bit, which is only set or cleared with irqs on
1764 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1765 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1766 bool do_list
= !btrfs_is_free_space_inode(inode
);
1768 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1769 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1770 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1771 spin_lock(&BTRFS_I(inode
)->lock
);
1772 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
1773 spin_unlock(&BTRFS_I(inode
)->lock
);
1777 * We don't reserve metadata space for space cache inodes so we
1778 * don't need to call dellalloc_release_metadata if there is an
1781 if (*bits
& EXTENT_DO_ACCOUNTING
&&
1782 root
!= root
->fs_info
->tree_root
)
1783 btrfs_delalloc_release_metadata(inode
, len
);
1785 /* For sanity tests. */
1786 if (btrfs_test_is_dummy_root(root
))
1789 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1790 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1791 btrfs_free_reserved_data_space_noquota(inode
,
1794 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1795 root
->fs_info
->delalloc_batch
);
1796 spin_lock(&BTRFS_I(inode
)->lock
);
1797 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1798 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1799 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1800 &BTRFS_I(inode
)->runtime_flags
))
1801 btrfs_del_delalloc_inode(root
, inode
);
1802 spin_unlock(&BTRFS_I(inode
)->lock
);
1807 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808 * we don't create bios that span stripes or chunks
1810 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1811 size_t size
, struct bio
*bio
,
1812 unsigned long bio_flags
)
1814 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1815 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1820 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1823 length
= bio
->bi_iter
.bi_size
;
1824 map_length
= length
;
1825 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1826 &map_length
, NULL
, 0);
1827 /* Will always return 0 with map_multi == NULL */
1829 if (map_length
< length
+ size
)
1835 * in order to insert checksums into the metadata in large chunks,
1836 * we wait until bio submission time. All the pages in the bio are
1837 * checksummed and sums are attached onto the ordered extent record.
1839 * At IO completion time the cums attached on the ordered extent record
1840 * are inserted into the btree
1842 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1843 struct bio
*bio
, int mirror_num
,
1844 unsigned long bio_flags
,
1847 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1850 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1851 BUG_ON(ret
); /* -ENOMEM */
1856 * in order to insert checksums into the metadata in large chunks,
1857 * we wait until bio submission time. All the pages in the bio are
1858 * checksummed and sums are attached onto the ordered extent record.
1860 * At IO completion time the cums attached on the ordered extent record
1861 * are inserted into the btree
1863 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1864 int mirror_num
, unsigned long bio_flags
,
1867 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1870 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1872 bio
->bi_error
= ret
;
1879 * extent_io.c submission hook. This does the right thing for csum calculation
1880 * on write, or reading the csums from the tree before a read
1882 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1883 int mirror_num
, unsigned long bio_flags
,
1886 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1887 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
1890 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1892 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1894 if (btrfs_is_free_space_inode(inode
))
1895 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
1897 if (!(rw
& REQ_WRITE
)) {
1898 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1902 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1903 ret
= btrfs_submit_compressed_read(inode
, bio
,
1907 } else if (!skip_sum
) {
1908 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1913 } else if (async
&& !skip_sum
) {
1914 /* csum items have already been cloned */
1915 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1917 /* we're doing a write, do the async checksumming */
1918 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1919 inode
, rw
, bio
, mirror_num
,
1920 bio_flags
, bio_offset
,
1921 __btrfs_submit_bio_start
,
1922 __btrfs_submit_bio_done
);
1924 } else if (!skip_sum
) {
1925 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1931 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1935 bio
->bi_error
= ret
;
1942 * given a list of ordered sums record them in the inode. This happens
1943 * at IO completion time based on sums calculated at bio submission time.
1945 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1946 struct inode
*inode
, u64 file_offset
,
1947 struct list_head
*list
)
1949 struct btrfs_ordered_sum
*sum
;
1951 list_for_each_entry(sum
, list
, list
) {
1952 trans
->adding_csums
= 1;
1953 btrfs_csum_file_blocks(trans
,
1954 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1955 trans
->adding_csums
= 0;
1960 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1961 struct extent_state
**cached_state
)
1963 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1964 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1965 cached_state
, GFP_NOFS
);
1968 /* see btrfs_writepage_start_hook for details on why this is required */
1969 struct btrfs_writepage_fixup
{
1971 struct btrfs_work work
;
1974 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1976 struct btrfs_writepage_fixup
*fixup
;
1977 struct btrfs_ordered_extent
*ordered
;
1978 struct extent_state
*cached_state
= NULL
;
1980 struct inode
*inode
;
1985 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1989 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1990 ClearPageChecked(page
);
1994 inode
= page
->mapping
->host
;
1995 page_start
= page_offset(page
);
1996 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1998 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2001 /* already ordered? We're done */
2002 if (PagePrivate2(page
))
2005 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2007 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
2008 page_end
, &cached_state
, GFP_NOFS
);
2010 btrfs_start_ordered_extent(inode
, ordered
, 1);
2011 btrfs_put_ordered_extent(ordered
);
2015 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
2018 mapping_set_error(page
->mapping
, ret
);
2019 end_extent_writepage(page
, ret
, page_start
, page_end
);
2020 ClearPageChecked(page
);
2024 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
2025 ClearPageChecked(page
);
2026 set_page_dirty(page
);
2028 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2029 &cached_state
, GFP_NOFS
);
2032 page_cache_release(page
);
2037 * There are a few paths in the higher layers of the kernel that directly
2038 * set the page dirty bit without asking the filesystem if it is a
2039 * good idea. This causes problems because we want to make sure COW
2040 * properly happens and the data=ordered rules are followed.
2042 * In our case any range that doesn't have the ORDERED bit set
2043 * hasn't been properly setup for IO. We kick off an async process
2044 * to fix it up. The async helper will wait for ordered extents, set
2045 * the delalloc bit and make it safe to write the page.
2047 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2049 struct inode
*inode
= page
->mapping
->host
;
2050 struct btrfs_writepage_fixup
*fixup
;
2051 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2053 /* this page is properly in the ordered list */
2054 if (TestClearPagePrivate2(page
))
2057 if (PageChecked(page
))
2060 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2064 SetPageChecked(page
);
2065 page_cache_get(page
);
2066 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2067 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2069 btrfs_queue_work(root
->fs_info
->fixup_workers
, &fixup
->work
);
2073 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2074 struct inode
*inode
, u64 file_pos
,
2075 u64 disk_bytenr
, u64 disk_num_bytes
,
2076 u64 num_bytes
, u64 ram_bytes
,
2077 u8 compression
, u8 encryption
,
2078 u16 other_encoding
, int extent_type
)
2080 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2081 struct btrfs_file_extent_item
*fi
;
2082 struct btrfs_path
*path
;
2083 struct extent_buffer
*leaf
;
2084 struct btrfs_key ins
;
2085 int extent_inserted
= 0;
2088 path
= btrfs_alloc_path();
2093 * we may be replacing one extent in the tree with another.
2094 * The new extent is pinned in the extent map, and we don't want
2095 * to drop it from the cache until it is completely in the btree.
2097 * So, tell btrfs_drop_extents to leave this extent in the cache.
2098 * the caller is expected to unpin it and allow it to be merged
2101 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2102 file_pos
+ num_bytes
, NULL
, 0,
2103 1, sizeof(*fi
), &extent_inserted
);
2107 if (!extent_inserted
) {
2108 ins
.objectid
= btrfs_ino(inode
);
2109 ins
.offset
= file_pos
;
2110 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2112 path
->leave_spinning
= 1;
2113 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2118 leaf
= path
->nodes
[0];
2119 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2120 struct btrfs_file_extent_item
);
2121 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2122 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2123 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2124 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2125 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2126 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2127 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2128 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2129 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2130 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2132 btrfs_mark_buffer_dirty(leaf
);
2133 btrfs_release_path(path
);
2135 inode_add_bytes(inode
, num_bytes
);
2137 ins
.objectid
= disk_bytenr
;
2138 ins
.offset
= disk_num_bytes
;
2139 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2140 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2141 root
->root_key
.objectid
,
2142 btrfs_ino(inode
), file_pos
,
2145 * Release the reserved range from inode dirty range map, as it is
2146 * already moved into delayed_ref_head
2148 btrfs_qgroup_release_data(inode
, file_pos
, ram_bytes
);
2150 btrfs_free_path(path
);
2155 /* snapshot-aware defrag */
2156 struct sa_defrag_extent_backref
{
2157 struct rb_node node
;
2158 struct old_sa_defrag_extent
*old
;
2167 struct old_sa_defrag_extent
{
2168 struct list_head list
;
2169 struct new_sa_defrag_extent
*new;
2178 struct new_sa_defrag_extent
{
2179 struct rb_root root
;
2180 struct list_head head
;
2181 struct btrfs_path
*path
;
2182 struct inode
*inode
;
2190 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2191 struct sa_defrag_extent_backref
*b2
)
2193 if (b1
->root_id
< b2
->root_id
)
2195 else if (b1
->root_id
> b2
->root_id
)
2198 if (b1
->inum
< b2
->inum
)
2200 else if (b1
->inum
> b2
->inum
)
2203 if (b1
->file_pos
< b2
->file_pos
)
2205 else if (b1
->file_pos
> b2
->file_pos
)
2209 * [------------------------------] ===> (a range of space)
2210 * |<--->| |<---->| =============> (fs/file tree A)
2211 * |<---------------------------->| ===> (fs/file tree B)
2213 * A range of space can refer to two file extents in one tree while
2214 * refer to only one file extent in another tree.
2216 * So we may process a disk offset more than one time(two extents in A)
2217 * and locate at the same extent(one extent in B), then insert two same
2218 * backrefs(both refer to the extent in B).
2223 static void backref_insert(struct rb_root
*root
,
2224 struct sa_defrag_extent_backref
*backref
)
2226 struct rb_node
**p
= &root
->rb_node
;
2227 struct rb_node
*parent
= NULL
;
2228 struct sa_defrag_extent_backref
*entry
;
2233 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2235 ret
= backref_comp(backref
, entry
);
2239 p
= &(*p
)->rb_right
;
2242 rb_link_node(&backref
->node
, parent
, p
);
2243 rb_insert_color(&backref
->node
, root
);
2247 * Note the backref might has changed, and in this case we just return 0.
2249 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2252 struct btrfs_file_extent_item
*extent
;
2253 struct btrfs_fs_info
*fs_info
;
2254 struct old_sa_defrag_extent
*old
= ctx
;
2255 struct new_sa_defrag_extent
*new = old
->new;
2256 struct btrfs_path
*path
= new->path
;
2257 struct btrfs_key key
;
2258 struct btrfs_root
*root
;
2259 struct sa_defrag_extent_backref
*backref
;
2260 struct extent_buffer
*leaf
;
2261 struct inode
*inode
= new->inode
;
2267 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2268 inum
== btrfs_ino(inode
))
2271 key
.objectid
= root_id
;
2272 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2273 key
.offset
= (u64
)-1;
2275 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2276 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2278 if (PTR_ERR(root
) == -ENOENT
)
2281 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2282 inum
, offset
, root_id
);
2283 return PTR_ERR(root
);
2286 key
.objectid
= inum
;
2287 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2288 if (offset
> (u64
)-1 << 32)
2291 key
.offset
= offset
;
2293 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2294 if (WARN_ON(ret
< 0))
2301 leaf
= path
->nodes
[0];
2302 slot
= path
->slots
[0];
2304 if (slot
>= btrfs_header_nritems(leaf
)) {
2305 ret
= btrfs_next_leaf(root
, path
);
2308 } else if (ret
> 0) {
2317 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2319 if (key
.objectid
> inum
)
2322 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2325 extent
= btrfs_item_ptr(leaf
, slot
,
2326 struct btrfs_file_extent_item
);
2328 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2332 * 'offset' refers to the exact key.offset,
2333 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2334 * (key.offset - extent_offset).
2336 if (key
.offset
!= offset
)
2339 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2340 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2342 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2343 old
->len
|| extent_offset
+ num_bytes
<=
2344 old
->extent_offset
+ old
->offset
)
2349 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2355 backref
->root_id
= root_id
;
2356 backref
->inum
= inum
;
2357 backref
->file_pos
= offset
;
2358 backref
->num_bytes
= num_bytes
;
2359 backref
->extent_offset
= extent_offset
;
2360 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2362 backref_insert(&new->root
, backref
);
2365 btrfs_release_path(path
);
2370 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2371 struct new_sa_defrag_extent
*new)
2373 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2374 struct old_sa_defrag_extent
*old
, *tmp
;
2379 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2380 ret
= iterate_inodes_from_logical(old
->bytenr
+
2381 old
->extent_offset
, fs_info
,
2382 path
, record_one_backref
,
2384 if (ret
< 0 && ret
!= -ENOENT
)
2387 /* no backref to be processed for this extent */
2389 list_del(&old
->list
);
2394 if (list_empty(&new->head
))
2400 static int relink_is_mergable(struct extent_buffer
*leaf
,
2401 struct btrfs_file_extent_item
*fi
,
2402 struct new_sa_defrag_extent
*new)
2404 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2407 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2410 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2413 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2414 btrfs_file_extent_other_encoding(leaf
, fi
))
2421 * Note the backref might has changed, and in this case we just return 0.
2423 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2424 struct sa_defrag_extent_backref
*prev
,
2425 struct sa_defrag_extent_backref
*backref
)
2427 struct btrfs_file_extent_item
*extent
;
2428 struct btrfs_file_extent_item
*item
;
2429 struct btrfs_ordered_extent
*ordered
;
2430 struct btrfs_trans_handle
*trans
;
2431 struct btrfs_fs_info
*fs_info
;
2432 struct btrfs_root
*root
;
2433 struct btrfs_key key
;
2434 struct extent_buffer
*leaf
;
2435 struct old_sa_defrag_extent
*old
= backref
->old
;
2436 struct new_sa_defrag_extent
*new = old
->new;
2437 struct inode
*src_inode
= new->inode
;
2438 struct inode
*inode
;
2439 struct extent_state
*cached
= NULL
;
2448 if (prev
&& prev
->root_id
== backref
->root_id
&&
2449 prev
->inum
== backref
->inum
&&
2450 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2453 /* step 1: get root */
2454 key
.objectid
= backref
->root_id
;
2455 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2456 key
.offset
= (u64
)-1;
2458 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2459 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2461 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2463 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2464 if (PTR_ERR(root
) == -ENOENT
)
2466 return PTR_ERR(root
);
2469 if (btrfs_root_readonly(root
)) {
2470 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2474 /* step 2: get inode */
2475 key
.objectid
= backref
->inum
;
2476 key
.type
= BTRFS_INODE_ITEM_KEY
;
2479 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2480 if (IS_ERR(inode
)) {
2481 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2485 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2487 /* step 3: relink backref */
2488 lock_start
= backref
->file_pos
;
2489 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2490 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2493 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2495 btrfs_put_ordered_extent(ordered
);
2499 trans
= btrfs_join_transaction(root
);
2500 if (IS_ERR(trans
)) {
2501 ret
= PTR_ERR(trans
);
2505 key
.objectid
= backref
->inum
;
2506 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2507 key
.offset
= backref
->file_pos
;
2509 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2512 } else if (ret
> 0) {
2517 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2518 struct btrfs_file_extent_item
);
2520 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2521 backref
->generation
)
2524 btrfs_release_path(path
);
2526 start
= backref
->file_pos
;
2527 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2528 start
+= old
->extent_offset
+ old
->offset
-
2529 backref
->extent_offset
;
2531 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2532 old
->extent_offset
+ old
->offset
+ old
->len
);
2533 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2535 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2540 key
.objectid
= btrfs_ino(inode
);
2541 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2544 path
->leave_spinning
= 1;
2546 struct btrfs_file_extent_item
*fi
;
2548 struct btrfs_key found_key
;
2550 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2555 leaf
= path
->nodes
[0];
2556 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2558 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2559 struct btrfs_file_extent_item
);
2560 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2562 if (extent_len
+ found_key
.offset
== start
&&
2563 relink_is_mergable(leaf
, fi
, new)) {
2564 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2566 btrfs_mark_buffer_dirty(leaf
);
2567 inode_add_bytes(inode
, len
);
2573 btrfs_release_path(path
);
2578 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2581 btrfs_abort_transaction(trans
, root
, ret
);
2585 leaf
= path
->nodes
[0];
2586 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2587 struct btrfs_file_extent_item
);
2588 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2589 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2590 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2591 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2592 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2593 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2594 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2595 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2596 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2597 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2599 btrfs_mark_buffer_dirty(leaf
);
2600 inode_add_bytes(inode
, len
);
2601 btrfs_release_path(path
);
2603 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2605 backref
->root_id
, backref
->inum
,
2606 new->file_pos
); /* start - extent_offset */
2608 btrfs_abort_transaction(trans
, root
, ret
);
2614 btrfs_release_path(path
);
2615 path
->leave_spinning
= 0;
2616 btrfs_end_transaction(trans
, root
);
2618 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2624 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2626 struct old_sa_defrag_extent
*old
, *tmp
;
2631 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2637 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2639 struct btrfs_path
*path
;
2640 struct sa_defrag_extent_backref
*backref
;
2641 struct sa_defrag_extent_backref
*prev
= NULL
;
2642 struct inode
*inode
;
2643 struct btrfs_root
*root
;
2644 struct rb_node
*node
;
2648 root
= BTRFS_I(inode
)->root
;
2650 path
= btrfs_alloc_path();
2654 if (!record_extent_backrefs(path
, new)) {
2655 btrfs_free_path(path
);
2658 btrfs_release_path(path
);
2661 node
= rb_first(&new->root
);
2664 rb_erase(node
, &new->root
);
2666 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2668 ret
= relink_extent_backref(path
, prev
, backref
);
2681 btrfs_free_path(path
);
2683 free_sa_defrag_extent(new);
2685 atomic_dec(&root
->fs_info
->defrag_running
);
2686 wake_up(&root
->fs_info
->transaction_wait
);
2689 static struct new_sa_defrag_extent
*
2690 record_old_file_extents(struct inode
*inode
,
2691 struct btrfs_ordered_extent
*ordered
)
2693 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2694 struct btrfs_path
*path
;
2695 struct btrfs_key key
;
2696 struct old_sa_defrag_extent
*old
;
2697 struct new_sa_defrag_extent
*new;
2700 new = kmalloc(sizeof(*new), GFP_NOFS
);
2705 new->file_pos
= ordered
->file_offset
;
2706 new->len
= ordered
->len
;
2707 new->bytenr
= ordered
->start
;
2708 new->disk_len
= ordered
->disk_len
;
2709 new->compress_type
= ordered
->compress_type
;
2710 new->root
= RB_ROOT
;
2711 INIT_LIST_HEAD(&new->head
);
2713 path
= btrfs_alloc_path();
2717 key
.objectid
= btrfs_ino(inode
);
2718 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2719 key
.offset
= new->file_pos
;
2721 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2724 if (ret
> 0 && path
->slots
[0] > 0)
2727 /* find out all the old extents for the file range */
2729 struct btrfs_file_extent_item
*extent
;
2730 struct extent_buffer
*l
;
2739 slot
= path
->slots
[0];
2741 if (slot
>= btrfs_header_nritems(l
)) {
2742 ret
= btrfs_next_leaf(root
, path
);
2750 btrfs_item_key_to_cpu(l
, &key
, slot
);
2752 if (key
.objectid
!= btrfs_ino(inode
))
2754 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2756 if (key
.offset
>= new->file_pos
+ new->len
)
2759 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2761 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2762 if (key
.offset
+ num_bytes
< new->file_pos
)
2765 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2769 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2771 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2775 offset
= max(new->file_pos
, key
.offset
);
2776 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2778 old
->bytenr
= disk_bytenr
;
2779 old
->extent_offset
= extent_offset
;
2780 old
->offset
= offset
- key
.offset
;
2781 old
->len
= end
- offset
;
2784 list_add_tail(&old
->list
, &new->head
);
2790 btrfs_free_path(path
);
2791 atomic_inc(&root
->fs_info
->defrag_running
);
2796 btrfs_free_path(path
);
2798 free_sa_defrag_extent(new);
2802 static void btrfs_release_delalloc_bytes(struct btrfs_root
*root
,
2805 struct btrfs_block_group_cache
*cache
;
2807 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2810 spin_lock(&cache
->lock
);
2811 cache
->delalloc_bytes
-= len
;
2812 spin_unlock(&cache
->lock
);
2814 btrfs_put_block_group(cache
);
2817 /* as ordered data IO finishes, this gets called so we can finish
2818 * an ordered extent if the range of bytes in the file it covers are
2821 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2823 struct inode
*inode
= ordered_extent
->inode
;
2824 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2825 struct btrfs_trans_handle
*trans
= NULL
;
2826 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2827 struct extent_state
*cached_state
= NULL
;
2828 struct new_sa_defrag_extent
*new = NULL
;
2829 int compress_type
= 0;
2831 u64 logical_len
= ordered_extent
->len
;
2833 bool truncated
= false;
2835 nolock
= btrfs_is_free_space_inode(inode
);
2837 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2842 btrfs_free_io_failure_record(inode
, ordered_extent
->file_offset
,
2843 ordered_extent
->file_offset
+
2844 ordered_extent
->len
- 1);
2846 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2848 logical_len
= ordered_extent
->truncated_len
;
2849 /* Truncated the entire extent, don't bother adding */
2854 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2855 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2858 * For mwrite(mmap + memset to write) case, we still reserve
2859 * space for NOCOW range.
2860 * As NOCOW won't cause a new delayed ref, just free the space
2862 btrfs_qgroup_free_data(inode
, ordered_extent
->file_offset
,
2863 ordered_extent
->len
);
2864 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2866 trans
= btrfs_join_transaction_nolock(root
);
2868 trans
= btrfs_join_transaction(root
);
2869 if (IS_ERR(trans
)) {
2870 ret
= PTR_ERR(trans
);
2874 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2875 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2876 if (ret
) /* -ENOMEM or corruption */
2877 btrfs_abort_transaction(trans
, root
, ret
);
2881 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2882 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2885 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2886 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2887 EXTENT_DEFRAG
, 1, cached_state
);
2889 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2890 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2891 /* the inode is shared */
2892 new = record_old_file_extents(inode
, ordered_extent
);
2894 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2895 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2896 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2900 trans
= btrfs_join_transaction_nolock(root
);
2902 trans
= btrfs_join_transaction(root
);
2903 if (IS_ERR(trans
)) {
2904 ret
= PTR_ERR(trans
);
2909 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2911 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2912 compress_type
= ordered_extent
->compress_type
;
2913 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2914 BUG_ON(compress_type
);
2915 ret
= btrfs_mark_extent_written(trans
, inode
,
2916 ordered_extent
->file_offset
,
2917 ordered_extent
->file_offset
+
2920 BUG_ON(root
== root
->fs_info
->tree_root
);
2921 ret
= insert_reserved_file_extent(trans
, inode
,
2922 ordered_extent
->file_offset
,
2923 ordered_extent
->start
,
2924 ordered_extent
->disk_len
,
2925 logical_len
, logical_len
,
2926 compress_type
, 0, 0,
2927 BTRFS_FILE_EXTENT_REG
);
2929 btrfs_release_delalloc_bytes(root
,
2930 ordered_extent
->start
,
2931 ordered_extent
->disk_len
);
2933 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2934 ordered_extent
->file_offset
, ordered_extent
->len
,
2937 btrfs_abort_transaction(trans
, root
, ret
);
2941 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2942 &ordered_extent
->list
);
2944 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2945 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2946 if (ret
) { /* -ENOMEM or corruption */
2947 btrfs_abort_transaction(trans
, root
, ret
);
2952 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2953 ordered_extent
->file_offset
+
2954 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2956 if (root
!= root
->fs_info
->tree_root
)
2957 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2959 btrfs_end_transaction(trans
, root
);
2961 if (ret
|| truncated
) {
2965 start
= ordered_extent
->file_offset
+ logical_len
;
2967 start
= ordered_extent
->file_offset
;
2968 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
2969 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
2971 /* Drop the cache for the part of the extent we didn't write. */
2972 btrfs_drop_extent_cache(inode
, start
, end
, 0);
2975 * If the ordered extent had an IOERR or something else went
2976 * wrong we need to return the space for this ordered extent
2977 * back to the allocator. We only free the extent in the
2978 * truncated case if we didn't write out the extent at all.
2980 if ((ret
|| !logical_len
) &&
2981 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2982 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2983 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
2984 ordered_extent
->disk_len
, 1);
2989 * This needs to be done to make sure anybody waiting knows we are done
2990 * updating everything for this ordered extent.
2992 btrfs_remove_ordered_extent(inode
, ordered_extent
);
2994 /* for snapshot-aware defrag */
2997 free_sa_defrag_extent(new);
2998 atomic_dec(&root
->fs_info
->defrag_running
);
3000 relink_file_extents(new);
3005 btrfs_put_ordered_extent(ordered_extent
);
3006 /* once for the tree */
3007 btrfs_put_ordered_extent(ordered_extent
);
3012 static void finish_ordered_fn(struct btrfs_work
*work
)
3014 struct btrfs_ordered_extent
*ordered_extent
;
3015 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
3016 btrfs_finish_ordered_io(ordered_extent
);
3019 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
3020 struct extent_state
*state
, int uptodate
)
3022 struct inode
*inode
= page
->mapping
->host
;
3023 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3024 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
3025 struct btrfs_workqueue
*wq
;
3026 btrfs_work_func_t func
;
3028 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
3030 ClearPagePrivate2(page
);
3031 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
3032 end
- start
+ 1, uptodate
))
3035 if (btrfs_is_free_space_inode(inode
)) {
3036 wq
= root
->fs_info
->endio_freespace_worker
;
3037 func
= btrfs_freespace_write_helper
;
3039 wq
= root
->fs_info
->endio_write_workers
;
3040 func
= btrfs_endio_write_helper
;
3043 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3045 btrfs_queue_work(wq
, &ordered_extent
->work
);
3050 static int __readpage_endio_check(struct inode
*inode
,
3051 struct btrfs_io_bio
*io_bio
,
3052 int icsum
, struct page
*page
,
3053 int pgoff
, u64 start
, size_t len
)
3059 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3061 kaddr
= kmap_atomic(page
);
3062 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3063 btrfs_csum_final(csum
, (char *)&csum
);
3064 if (csum
!= csum_expected
)
3067 kunmap_atomic(kaddr
);
3070 btrfs_warn_rl(BTRFS_I(inode
)->root
->fs_info
,
3071 "csum failed ino %llu off %llu csum %u expected csum %u",
3072 btrfs_ino(inode
), start
, csum
, csum_expected
);
3073 memset(kaddr
+ pgoff
, 1, len
);
3074 flush_dcache_page(page
);
3075 kunmap_atomic(kaddr
);
3076 if (csum_expected
== 0)
3082 * when reads are done, we need to check csums to verify the data is correct
3083 * if there's a match, we allow the bio to finish. If not, the code in
3084 * extent_io.c will try to find good copies for us.
3086 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3087 u64 phy_offset
, struct page
*page
,
3088 u64 start
, u64 end
, int mirror
)
3090 size_t offset
= start
- page_offset(page
);
3091 struct inode
*inode
= page
->mapping
->host
;
3092 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3093 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3095 if (PageChecked(page
)) {
3096 ClearPageChecked(page
);
3100 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3103 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3104 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3105 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
3110 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3111 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3112 start
, (size_t)(end
- start
+ 1));
3115 void btrfs_add_delayed_iput(struct inode
*inode
)
3117 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
3118 struct btrfs_inode
*binode
= BTRFS_I(inode
);
3120 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3123 spin_lock(&fs_info
->delayed_iput_lock
);
3124 if (binode
->delayed_iput_count
== 0) {
3125 ASSERT(list_empty(&binode
->delayed_iput
));
3126 list_add_tail(&binode
->delayed_iput
, &fs_info
->delayed_iputs
);
3128 binode
->delayed_iput_count
++;
3130 spin_unlock(&fs_info
->delayed_iput_lock
);
3133 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
3135 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3137 spin_lock(&fs_info
->delayed_iput_lock
);
3138 while (!list_empty(&fs_info
->delayed_iputs
)) {
3139 struct btrfs_inode
*inode
;
3141 inode
= list_first_entry(&fs_info
->delayed_iputs
,
3142 struct btrfs_inode
, delayed_iput
);
3143 if (inode
->delayed_iput_count
) {
3144 inode
->delayed_iput_count
--;
3145 list_move_tail(&inode
->delayed_iput
,
3146 &fs_info
->delayed_iputs
);
3148 list_del_init(&inode
->delayed_iput
);
3150 spin_unlock(&fs_info
->delayed_iput_lock
);
3151 iput(&inode
->vfs_inode
);
3152 spin_lock(&fs_info
->delayed_iput_lock
);
3154 spin_unlock(&fs_info
->delayed_iput_lock
);
3158 * This is called in transaction commit time. If there are no orphan
3159 * files in the subvolume, it removes orphan item and frees block_rsv
3162 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3163 struct btrfs_root
*root
)
3165 struct btrfs_block_rsv
*block_rsv
;
3168 if (atomic_read(&root
->orphan_inodes
) ||
3169 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3172 spin_lock(&root
->orphan_lock
);
3173 if (atomic_read(&root
->orphan_inodes
)) {
3174 spin_unlock(&root
->orphan_lock
);
3178 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3179 spin_unlock(&root
->orphan_lock
);
3183 block_rsv
= root
->orphan_block_rsv
;
3184 root
->orphan_block_rsv
= NULL
;
3185 spin_unlock(&root
->orphan_lock
);
3187 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3188 btrfs_root_refs(&root
->root_item
) > 0) {
3189 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
3190 root
->root_key
.objectid
);
3192 btrfs_abort_transaction(trans
, root
, ret
);
3194 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3199 WARN_ON(block_rsv
->size
> 0);
3200 btrfs_free_block_rsv(root
, block_rsv
);
3205 * This creates an orphan entry for the given inode in case something goes
3206 * wrong in the middle of an unlink/truncate.
3208 * NOTE: caller of this function should reserve 5 units of metadata for
3211 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
3213 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3214 struct btrfs_block_rsv
*block_rsv
= NULL
;
3219 if (!root
->orphan_block_rsv
) {
3220 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3225 spin_lock(&root
->orphan_lock
);
3226 if (!root
->orphan_block_rsv
) {
3227 root
->orphan_block_rsv
= block_rsv
;
3228 } else if (block_rsv
) {
3229 btrfs_free_block_rsv(root
, block_rsv
);
3233 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3234 &BTRFS_I(inode
)->runtime_flags
)) {
3237 * For proper ENOSPC handling, we should do orphan
3238 * cleanup when mounting. But this introduces backward
3239 * compatibility issue.
3241 if (!xchg(&root
->orphan_item_inserted
, 1))
3247 atomic_inc(&root
->orphan_inodes
);
3250 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3251 &BTRFS_I(inode
)->runtime_flags
))
3253 spin_unlock(&root
->orphan_lock
);
3255 /* grab metadata reservation from transaction handle */
3257 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3258 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
3261 /* insert an orphan item to track this unlinked/truncated file */
3263 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3265 atomic_dec(&root
->orphan_inodes
);
3267 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3268 &BTRFS_I(inode
)->runtime_flags
);
3269 btrfs_orphan_release_metadata(inode
);
3271 if (ret
!= -EEXIST
) {
3272 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3273 &BTRFS_I(inode
)->runtime_flags
);
3274 btrfs_abort_transaction(trans
, root
, ret
);
3281 /* insert an orphan item to track subvolume contains orphan files */
3283 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3284 root
->root_key
.objectid
);
3285 if (ret
&& ret
!= -EEXIST
) {
3286 btrfs_abort_transaction(trans
, root
, ret
);
3294 * We have done the truncate/delete so we can go ahead and remove the orphan
3295 * item for this particular inode.
3297 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3298 struct inode
*inode
)
3300 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3301 int delete_item
= 0;
3302 int release_rsv
= 0;
3305 spin_lock(&root
->orphan_lock
);
3306 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3307 &BTRFS_I(inode
)->runtime_flags
))
3310 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3311 &BTRFS_I(inode
)->runtime_flags
))
3313 spin_unlock(&root
->orphan_lock
);
3316 atomic_dec(&root
->orphan_inodes
);
3318 ret
= btrfs_del_orphan_item(trans
, root
,
3323 btrfs_orphan_release_metadata(inode
);
3329 * this cleans up any orphans that may be left on the list from the last use
3332 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3334 struct btrfs_path
*path
;
3335 struct extent_buffer
*leaf
;
3336 struct btrfs_key key
, found_key
;
3337 struct btrfs_trans_handle
*trans
;
3338 struct inode
*inode
;
3339 u64 last_objectid
= 0;
3340 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3342 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3345 path
= btrfs_alloc_path();
3350 path
->reada
= READA_BACK
;
3352 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3353 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3354 key
.offset
= (u64
)-1;
3357 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3362 * if ret == 0 means we found what we were searching for, which
3363 * is weird, but possible, so only screw with path if we didn't
3364 * find the key and see if we have stuff that matches
3368 if (path
->slots
[0] == 0)
3373 /* pull out the item */
3374 leaf
= path
->nodes
[0];
3375 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3377 /* make sure the item matches what we want */
3378 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3380 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3383 /* release the path since we're done with it */
3384 btrfs_release_path(path
);
3387 * this is where we are basically btrfs_lookup, without the
3388 * crossing root thing. we store the inode number in the
3389 * offset of the orphan item.
3392 if (found_key
.offset
== last_objectid
) {
3393 btrfs_err(root
->fs_info
,
3394 "Error removing orphan entry, stopping orphan cleanup");
3399 last_objectid
= found_key
.offset
;
3401 found_key
.objectid
= found_key
.offset
;
3402 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3403 found_key
.offset
= 0;
3404 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3405 ret
= PTR_ERR_OR_ZERO(inode
);
3406 if (ret
&& ret
!= -ESTALE
)
3409 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3410 struct btrfs_root
*dead_root
;
3411 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3412 int is_dead_root
= 0;
3415 * this is an orphan in the tree root. Currently these
3416 * could come from 2 sources:
3417 * a) a snapshot deletion in progress
3418 * b) a free space cache inode
3419 * We need to distinguish those two, as the snapshot
3420 * orphan must not get deleted.
3421 * find_dead_roots already ran before us, so if this
3422 * is a snapshot deletion, we should find the root
3423 * in the dead_roots list
3425 spin_lock(&fs_info
->trans_lock
);
3426 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3428 if (dead_root
->root_key
.objectid
==
3429 found_key
.objectid
) {
3434 spin_unlock(&fs_info
->trans_lock
);
3436 /* prevent this orphan from being found again */
3437 key
.offset
= found_key
.objectid
- 1;
3442 * Inode is already gone but the orphan item is still there,
3443 * kill the orphan item.
3445 if (ret
== -ESTALE
) {
3446 trans
= btrfs_start_transaction(root
, 1);
3447 if (IS_ERR(trans
)) {
3448 ret
= PTR_ERR(trans
);
3451 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3452 found_key
.objectid
);
3453 ret
= btrfs_del_orphan_item(trans
, root
,
3454 found_key
.objectid
);
3455 btrfs_end_transaction(trans
, root
);
3462 * add this inode to the orphan list so btrfs_orphan_del does
3463 * the proper thing when we hit it
3465 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3466 &BTRFS_I(inode
)->runtime_flags
);
3467 atomic_inc(&root
->orphan_inodes
);
3469 /* if we have links, this was a truncate, lets do that */
3470 if (inode
->i_nlink
) {
3471 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3477 /* 1 for the orphan item deletion. */
3478 trans
= btrfs_start_transaction(root
, 1);
3479 if (IS_ERR(trans
)) {
3481 ret
= PTR_ERR(trans
);
3484 ret
= btrfs_orphan_add(trans
, inode
);
3485 btrfs_end_transaction(trans
, root
);
3491 ret
= btrfs_truncate(inode
);
3493 btrfs_orphan_del(NULL
, inode
);
3498 /* this will do delete_inode and everything for us */
3503 /* release the path since we're done with it */
3504 btrfs_release_path(path
);
3506 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3508 if (root
->orphan_block_rsv
)
3509 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3512 if (root
->orphan_block_rsv
||
3513 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3514 trans
= btrfs_join_transaction(root
);
3516 btrfs_end_transaction(trans
, root
);
3520 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3522 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3526 btrfs_err(root
->fs_info
,
3527 "could not do orphan cleanup %d", ret
);
3528 btrfs_free_path(path
);
3533 * very simple check to peek ahead in the leaf looking for xattrs. If we
3534 * don't find any xattrs, we know there can't be any acls.
3536 * slot is the slot the inode is in, objectid is the objectid of the inode
3538 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3539 int slot
, u64 objectid
,
3540 int *first_xattr_slot
)
3542 u32 nritems
= btrfs_header_nritems(leaf
);
3543 struct btrfs_key found_key
;
3544 static u64 xattr_access
= 0;
3545 static u64 xattr_default
= 0;
3548 if (!xattr_access
) {
3549 xattr_access
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS
,
3550 strlen(XATTR_NAME_POSIX_ACL_ACCESS
));
3551 xattr_default
= btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT
,
3552 strlen(XATTR_NAME_POSIX_ACL_DEFAULT
));
3556 *first_xattr_slot
= -1;
3557 while (slot
< nritems
) {
3558 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3560 /* we found a different objectid, there must not be acls */
3561 if (found_key
.objectid
!= objectid
)
3564 /* we found an xattr, assume we've got an acl */
3565 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3566 if (*first_xattr_slot
== -1)
3567 *first_xattr_slot
= slot
;
3568 if (found_key
.offset
== xattr_access
||
3569 found_key
.offset
== xattr_default
)
3574 * we found a key greater than an xattr key, there can't
3575 * be any acls later on
3577 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3584 * it goes inode, inode backrefs, xattrs, extents,
3585 * so if there are a ton of hard links to an inode there can
3586 * be a lot of backrefs. Don't waste time searching too hard,
3587 * this is just an optimization
3592 /* we hit the end of the leaf before we found an xattr or
3593 * something larger than an xattr. We have to assume the inode
3596 if (*first_xattr_slot
== -1)
3597 *first_xattr_slot
= slot
;
3602 * read an inode from the btree into the in-memory inode
3604 static void btrfs_read_locked_inode(struct inode
*inode
)
3606 struct btrfs_path
*path
;
3607 struct extent_buffer
*leaf
;
3608 struct btrfs_inode_item
*inode_item
;
3609 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3610 struct btrfs_key location
;
3615 bool filled
= false;
3616 int first_xattr_slot
;
3618 ret
= btrfs_fill_inode(inode
, &rdev
);
3622 path
= btrfs_alloc_path();
3626 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3628 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3632 leaf
= path
->nodes
[0];
3637 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3638 struct btrfs_inode_item
);
3639 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3640 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3641 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3642 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3643 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3645 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3646 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3648 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3649 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3651 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3652 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3654 BTRFS_I(inode
)->i_otime
.tv_sec
=
3655 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3656 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3657 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3659 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3660 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3661 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3663 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3664 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3666 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3668 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3669 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3673 * If we were modified in the current generation and evicted from memory
3674 * and then re-read we need to do a full sync since we don't have any
3675 * idea about which extents were modified before we were evicted from
3678 * This is required for both inode re-read from disk and delayed inode
3679 * in delayed_nodes_tree.
3681 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3682 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3683 &BTRFS_I(inode
)->runtime_flags
);
3686 * We don't persist the id of the transaction where an unlink operation
3687 * against the inode was last made. So here we assume the inode might
3688 * have been evicted, and therefore the exact value of last_unlink_trans
3689 * lost, and set it to last_trans to avoid metadata inconsistencies
3690 * between the inode and its parent if the inode is fsync'ed and the log
3691 * replayed. For example, in the scenario:
3694 * ln mydir/foo mydir/bar
3697 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3698 * xfs_io -c fsync mydir/foo
3700 * mount fs, triggers fsync log replay
3702 * We must make sure that when we fsync our inode foo we also log its
3703 * parent inode, otherwise after log replay the parent still has the
3704 * dentry with the "bar" name but our inode foo has a link count of 1
3705 * and doesn't have an inode ref with the name "bar" anymore.
3707 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3708 * but it guarantees correctness at the expense of ocassional full
3709 * transaction commits on fsync if our inode is a directory, or if our
3710 * inode is not a directory, logging its parent unnecessarily.
3712 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3715 if (inode
->i_nlink
!= 1 ||
3716 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3719 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3720 if (location
.objectid
!= btrfs_ino(inode
))
3723 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3724 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3725 struct btrfs_inode_ref
*ref
;
3727 ref
= (struct btrfs_inode_ref
*)ptr
;
3728 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3729 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3730 struct btrfs_inode_extref
*extref
;
3732 extref
= (struct btrfs_inode_extref
*)ptr
;
3733 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3738 * try to precache a NULL acl entry for files that don't have
3739 * any xattrs or acls
3741 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3742 btrfs_ino(inode
), &first_xattr_slot
);
3743 if (first_xattr_slot
!= -1) {
3744 path
->slots
[0] = first_xattr_slot
;
3745 ret
= btrfs_load_inode_props(inode
, path
);
3747 btrfs_err(root
->fs_info
,
3748 "error loading props for ino %llu (root %llu): %d",
3750 root
->root_key
.objectid
, ret
);
3752 btrfs_free_path(path
);
3755 cache_no_acl(inode
);
3757 switch (inode
->i_mode
& S_IFMT
) {
3759 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3760 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3761 inode
->i_fop
= &btrfs_file_operations
;
3762 inode
->i_op
= &btrfs_file_inode_operations
;
3765 inode
->i_fop
= &btrfs_dir_file_operations
;
3766 if (root
== root
->fs_info
->tree_root
)
3767 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3769 inode
->i_op
= &btrfs_dir_inode_operations
;
3772 inode
->i_op
= &btrfs_symlink_inode_operations
;
3773 inode_nohighmem(inode
);
3774 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3777 inode
->i_op
= &btrfs_special_inode_operations
;
3778 init_special_inode(inode
, inode
->i_mode
, rdev
);
3782 btrfs_update_iflags(inode
);
3786 btrfs_free_path(path
);
3787 make_bad_inode(inode
);
3791 * given a leaf and an inode, copy the inode fields into the leaf
3793 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3794 struct extent_buffer
*leaf
,
3795 struct btrfs_inode_item
*item
,
3796 struct inode
*inode
)
3798 struct btrfs_map_token token
;
3800 btrfs_init_map_token(&token
);
3802 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3803 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3804 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3806 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3807 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3809 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3810 inode
->i_atime
.tv_sec
, &token
);
3811 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3812 inode
->i_atime
.tv_nsec
, &token
);
3814 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3815 inode
->i_mtime
.tv_sec
, &token
);
3816 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3817 inode
->i_mtime
.tv_nsec
, &token
);
3819 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3820 inode
->i_ctime
.tv_sec
, &token
);
3821 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3822 inode
->i_ctime
.tv_nsec
, &token
);
3824 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3825 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3826 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3827 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3829 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3831 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3833 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3834 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3835 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3836 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3837 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3841 * copy everything in the in-memory inode into the btree.
3843 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3844 struct btrfs_root
*root
, struct inode
*inode
)
3846 struct btrfs_inode_item
*inode_item
;
3847 struct btrfs_path
*path
;
3848 struct extent_buffer
*leaf
;
3851 path
= btrfs_alloc_path();
3855 path
->leave_spinning
= 1;
3856 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3864 leaf
= path
->nodes
[0];
3865 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3866 struct btrfs_inode_item
);
3868 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3869 btrfs_mark_buffer_dirty(leaf
);
3870 btrfs_set_inode_last_trans(trans
, inode
);
3873 btrfs_free_path(path
);
3878 * copy everything in the in-memory inode into the btree.
3880 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3881 struct btrfs_root
*root
, struct inode
*inode
)
3886 * If the inode is a free space inode, we can deadlock during commit
3887 * if we put it into the delayed code.
3889 * The data relocation inode should also be directly updated
3892 if (!btrfs_is_free_space_inode(inode
)
3893 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3894 && !root
->fs_info
->log_root_recovering
) {
3895 btrfs_update_root_times(trans
, root
);
3897 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3899 btrfs_set_inode_last_trans(trans
, inode
);
3903 return btrfs_update_inode_item(trans
, root
, inode
);
3906 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3907 struct btrfs_root
*root
,
3908 struct inode
*inode
)
3912 ret
= btrfs_update_inode(trans
, root
, inode
);
3914 return btrfs_update_inode_item(trans
, root
, inode
);
3919 * unlink helper that gets used here in inode.c and in the tree logging
3920 * recovery code. It remove a link in a directory with a given name, and
3921 * also drops the back refs in the inode to the directory
3923 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3924 struct btrfs_root
*root
,
3925 struct inode
*dir
, struct inode
*inode
,
3926 const char *name
, int name_len
)
3928 struct btrfs_path
*path
;
3930 struct extent_buffer
*leaf
;
3931 struct btrfs_dir_item
*di
;
3932 struct btrfs_key key
;
3934 u64 ino
= btrfs_ino(inode
);
3935 u64 dir_ino
= btrfs_ino(dir
);
3937 path
= btrfs_alloc_path();
3943 path
->leave_spinning
= 1;
3944 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3945 name
, name_len
, -1);
3954 leaf
= path
->nodes
[0];
3955 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3956 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3959 btrfs_release_path(path
);
3962 * If we don't have dir index, we have to get it by looking up
3963 * the inode ref, since we get the inode ref, remove it directly,
3964 * it is unnecessary to do delayed deletion.
3966 * But if we have dir index, needn't search inode ref to get it.
3967 * Since the inode ref is close to the inode item, it is better
3968 * that we delay to delete it, and just do this deletion when
3969 * we update the inode item.
3971 if (BTRFS_I(inode
)->dir_index
) {
3972 ret
= btrfs_delayed_delete_inode_ref(inode
);
3974 index
= BTRFS_I(inode
)->dir_index
;
3979 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3982 btrfs_info(root
->fs_info
,
3983 "failed to delete reference to %.*s, inode %llu parent %llu",
3984 name_len
, name
, ino
, dir_ino
);
3985 btrfs_abort_transaction(trans
, root
, ret
);
3989 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3991 btrfs_abort_transaction(trans
, root
, ret
);
3995 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
3997 if (ret
!= 0 && ret
!= -ENOENT
) {
3998 btrfs_abort_transaction(trans
, root
, ret
);
4002 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
4007 btrfs_abort_transaction(trans
, root
, ret
);
4009 btrfs_free_path(path
);
4013 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4014 inode_inc_iversion(inode
);
4015 inode_inc_iversion(dir
);
4016 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4017 ret
= btrfs_update_inode(trans
, root
, dir
);
4022 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
4023 struct btrfs_root
*root
,
4024 struct inode
*dir
, struct inode
*inode
,
4025 const char *name
, int name_len
)
4028 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4031 ret
= btrfs_update_inode(trans
, root
, inode
);
4037 * helper to start transaction for unlink and rmdir.
4039 * unlink and rmdir are special in btrfs, they do not always free space, so
4040 * if we cannot make our reservations the normal way try and see if there is
4041 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4042 * allow the unlink to occur.
4044 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4046 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4049 * 1 for the possible orphan item
4050 * 1 for the dir item
4051 * 1 for the dir index
4052 * 1 for the inode ref
4055 return btrfs_start_transaction_fallback_global_rsv(root
, 5, 5);
4058 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4060 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4061 struct btrfs_trans_handle
*trans
;
4062 struct inode
*inode
= d_inode(dentry
);
4065 trans
= __unlink_start_trans(dir
);
4067 return PTR_ERR(trans
);
4069 btrfs_record_unlink_dir(trans
, dir
, d_inode(dentry
), 0);
4071 ret
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4072 dentry
->d_name
.name
, dentry
->d_name
.len
);
4076 if (inode
->i_nlink
== 0) {
4077 ret
= btrfs_orphan_add(trans
, inode
);
4083 btrfs_end_transaction(trans
, root
);
4084 btrfs_btree_balance_dirty(root
);
4088 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4089 struct btrfs_root
*root
,
4090 struct inode
*dir
, u64 objectid
,
4091 const char *name
, int name_len
)
4093 struct btrfs_path
*path
;
4094 struct extent_buffer
*leaf
;
4095 struct btrfs_dir_item
*di
;
4096 struct btrfs_key key
;
4099 u64 dir_ino
= btrfs_ino(dir
);
4101 path
= btrfs_alloc_path();
4105 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4106 name
, name_len
, -1);
4107 if (IS_ERR_OR_NULL(di
)) {
4115 leaf
= path
->nodes
[0];
4116 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4117 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4118 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4120 btrfs_abort_transaction(trans
, root
, ret
);
4123 btrfs_release_path(path
);
4125 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4126 objectid
, root
->root_key
.objectid
,
4127 dir_ino
, &index
, name
, name_len
);
4129 if (ret
!= -ENOENT
) {
4130 btrfs_abort_transaction(trans
, root
, ret
);
4133 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4135 if (IS_ERR_OR_NULL(di
)) {
4140 btrfs_abort_transaction(trans
, root
, ret
);
4144 leaf
= path
->nodes
[0];
4145 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4146 btrfs_release_path(path
);
4149 btrfs_release_path(path
);
4151 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4153 btrfs_abort_transaction(trans
, root
, ret
);
4157 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4158 inode_inc_iversion(dir
);
4159 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4160 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4162 btrfs_abort_transaction(trans
, root
, ret
);
4164 btrfs_free_path(path
);
4168 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4170 struct inode
*inode
= d_inode(dentry
);
4172 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4173 struct btrfs_trans_handle
*trans
;
4175 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4177 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
4180 trans
= __unlink_start_trans(dir
);
4182 return PTR_ERR(trans
);
4184 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4185 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4186 BTRFS_I(inode
)->location
.objectid
,
4187 dentry
->d_name
.name
,
4188 dentry
->d_name
.len
);
4192 err
= btrfs_orphan_add(trans
, inode
);
4196 /* now the directory is empty */
4197 err
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4198 dentry
->d_name
.name
, dentry
->d_name
.len
);
4200 btrfs_i_size_write(inode
, 0);
4202 btrfs_end_transaction(trans
, root
);
4203 btrfs_btree_balance_dirty(root
);
4208 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4209 struct btrfs_root
*root
,
4214 bytes_deleted
= btrfs_csum_bytes_to_leaves(root
, bytes_deleted
);
4215 ret
= btrfs_block_rsv_add(root
, &root
->fs_info
->trans_block_rsv
,
4216 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4218 trans
->bytes_reserved
+= bytes_deleted
;
4223 static int truncate_inline_extent(struct inode
*inode
,
4224 struct btrfs_path
*path
,
4225 struct btrfs_key
*found_key
,
4229 struct extent_buffer
*leaf
= path
->nodes
[0];
4230 int slot
= path
->slots
[0];
4231 struct btrfs_file_extent_item
*fi
;
4232 u32 size
= (u32
)(new_size
- found_key
->offset
);
4233 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4235 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
4237 if (btrfs_file_extent_compression(leaf
, fi
) != BTRFS_COMPRESS_NONE
) {
4238 loff_t offset
= new_size
;
4239 loff_t page_end
= ALIGN(offset
, PAGE_CACHE_SIZE
);
4242 * Zero out the remaining of the last page of our inline extent,
4243 * instead of directly truncating our inline extent here - that
4244 * would be much more complex (decompressing all the data, then
4245 * compressing the truncated data, which might be bigger than
4246 * the size of the inline extent, resize the extent, etc).
4247 * We release the path because to get the page we might need to
4248 * read the extent item from disk (data not in the page cache).
4250 btrfs_release_path(path
);
4251 return btrfs_truncate_page(inode
, offset
, page_end
- offset
, 0);
4254 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4255 size
= btrfs_file_extent_calc_inline_size(size
);
4256 btrfs_truncate_item(root
, path
, size
, 1);
4258 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4259 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4265 * this can truncate away extent items, csum items and directory items.
4266 * It starts at a high offset and removes keys until it can't find
4267 * any higher than new_size
4269 * csum items that cross the new i_size are truncated to the new size
4272 * min_type is the minimum key type to truncate down to. If set to 0, this
4273 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4275 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4276 struct btrfs_root
*root
,
4277 struct inode
*inode
,
4278 u64 new_size
, u32 min_type
)
4280 struct btrfs_path
*path
;
4281 struct extent_buffer
*leaf
;
4282 struct btrfs_file_extent_item
*fi
;
4283 struct btrfs_key key
;
4284 struct btrfs_key found_key
;
4285 u64 extent_start
= 0;
4286 u64 extent_num_bytes
= 0;
4287 u64 extent_offset
= 0;
4289 u64 last_size
= new_size
;
4290 u32 found_type
= (u8
)-1;
4293 int pending_del_nr
= 0;
4294 int pending_del_slot
= 0;
4295 int extent_type
= -1;
4298 u64 ino
= btrfs_ino(inode
);
4299 u64 bytes_deleted
= 0;
4301 bool should_throttle
= 0;
4302 bool should_end
= 0;
4304 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4307 * for non-free space inodes and ref cows, we want to back off from
4310 if (!btrfs_is_free_space_inode(inode
) &&
4311 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4314 path
= btrfs_alloc_path();
4317 path
->reada
= READA_BACK
;
4320 * We want to drop from the next block forward in case this new size is
4321 * not block aligned since we will be keeping the last block of the
4322 * extent just the way it is.
4324 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4325 root
== root
->fs_info
->tree_root
)
4326 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
4327 root
->sectorsize
), (u64
)-1, 0);
4330 * This function is also used to drop the items in the log tree before
4331 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4332 * it is used to drop the loged items. So we shouldn't kill the delayed
4335 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4336 btrfs_kill_delayed_inode_items(inode
);
4339 key
.offset
= (u64
)-1;
4344 * with a 16K leaf size and 128MB extents, you can actually queue
4345 * up a huge file in a single leaf. Most of the time that
4346 * bytes_deleted is > 0, it will be huge by the time we get here
4348 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4349 if (btrfs_should_end_transaction(trans
, root
)) {
4356 path
->leave_spinning
= 1;
4357 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4364 /* there are no items in the tree for us to truncate, we're
4367 if (path
->slots
[0] == 0)
4374 leaf
= path
->nodes
[0];
4375 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4376 found_type
= found_key
.type
;
4378 if (found_key
.objectid
!= ino
)
4381 if (found_type
< min_type
)
4384 item_end
= found_key
.offset
;
4385 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4386 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4387 struct btrfs_file_extent_item
);
4388 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4389 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4391 btrfs_file_extent_num_bytes(leaf
, fi
);
4392 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4393 item_end
+= btrfs_file_extent_inline_len(leaf
,
4394 path
->slots
[0], fi
);
4398 if (found_type
> min_type
) {
4401 if (item_end
< new_size
)
4403 if (found_key
.offset
>= new_size
)
4409 /* FIXME, shrink the extent if the ref count is only 1 */
4410 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4414 last_size
= found_key
.offset
;
4416 last_size
= new_size
;
4418 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4420 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4422 u64 orig_num_bytes
=
4423 btrfs_file_extent_num_bytes(leaf
, fi
);
4424 extent_num_bytes
= ALIGN(new_size
-
4427 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4429 num_dec
= (orig_num_bytes
-
4431 if (test_bit(BTRFS_ROOT_REF_COWS
,
4434 inode_sub_bytes(inode
, num_dec
);
4435 btrfs_mark_buffer_dirty(leaf
);
4438 btrfs_file_extent_disk_num_bytes(leaf
,
4440 extent_offset
= found_key
.offset
-
4441 btrfs_file_extent_offset(leaf
, fi
);
4443 /* FIXME blocksize != 4096 */
4444 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4445 if (extent_start
!= 0) {
4447 if (test_bit(BTRFS_ROOT_REF_COWS
,
4449 inode_sub_bytes(inode
, num_dec
);
4452 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4454 * we can't truncate inline items that have had
4458 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4459 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4462 * Need to release path in order to truncate a
4463 * compressed extent. So delete any accumulated
4464 * extent items so far.
4466 if (btrfs_file_extent_compression(leaf
, fi
) !=
4467 BTRFS_COMPRESS_NONE
&& pending_del_nr
) {
4468 err
= btrfs_del_items(trans
, root
, path
,
4472 btrfs_abort_transaction(trans
,
4480 err
= truncate_inline_extent(inode
, path
,
4485 btrfs_abort_transaction(trans
,
4489 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4491 inode_sub_bytes(inode
, item_end
+ 1 - new_size
);
4496 if (!pending_del_nr
) {
4497 /* no pending yet, add ourselves */
4498 pending_del_slot
= path
->slots
[0];
4500 } else if (pending_del_nr
&&
4501 path
->slots
[0] + 1 == pending_del_slot
) {
4502 /* hop on the pending chunk */
4504 pending_del_slot
= path
->slots
[0];
4511 should_throttle
= 0;
4514 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4515 root
== root
->fs_info
->tree_root
)) {
4516 btrfs_set_path_blocking(path
);
4517 bytes_deleted
+= extent_num_bytes
;
4518 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4519 extent_num_bytes
, 0,
4520 btrfs_header_owner(leaf
),
4521 ino
, extent_offset
);
4523 if (btrfs_should_throttle_delayed_refs(trans
, root
))
4524 btrfs_async_run_delayed_refs(root
,
4525 trans
->delayed_ref_updates
* 2, 0);
4527 if (truncate_space_check(trans
, root
,
4528 extent_num_bytes
)) {
4531 if (btrfs_should_throttle_delayed_refs(trans
,
4533 should_throttle
= 1;
4538 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4541 if (path
->slots
[0] == 0 ||
4542 path
->slots
[0] != pending_del_slot
||
4543 should_throttle
|| should_end
) {
4544 if (pending_del_nr
) {
4545 ret
= btrfs_del_items(trans
, root
, path
,
4549 btrfs_abort_transaction(trans
,
4555 btrfs_release_path(path
);
4556 if (should_throttle
) {
4557 unsigned long updates
= trans
->delayed_ref_updates
;
4559 trans
->delayed_ref_updates
= 0;
4560 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4566 * if we failed to refill our space rsv, bail out
4567 * and let the transaction restart
4579 if (pending_del_nr
) {
4580 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4583 btrfs_abort_transaction(trans
, root
, ret
);
4586 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4587 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4589 btrfs_free_path(path
);
4591 if (be_nice
&& bytes_deleted
> SZ_32M
) {
4592 unsigned long updates
= trans
->delayed_ref_updates
;
4594 trans
->delayed_ref_updates
= 0;
4595 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4604 * btrfs_truncate_page - read, zero a chunk and write a page
4605 * @inode - inode that we're zeroing
4606 * @from - the offset to start zeroing
4607 * @len - the length to zero, 0 to zero the entire range respective to the
4609 * @front - zero up to the offset instead of from the offset on
4611 * This will find the page for the "from" offset and cow the page and zero the
4612 * part we want to zero. This is used with truncate and hole punching.
4614 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
4617 struct address_space
*mapping
= inode
->i_mapping
;
4618 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4619 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4620 struct btrfs_ordered_extent
*ordered
;
4621 struct extent_state
*cached_state
= NULL
;
4623 u32 blocksize
= root
->sectorsize
;
4624 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4625 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4627 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4632 if ((offset
& (blocksize
- 1)) == 0 &&
4633 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4635 ret
= btrfs_delalloc_reserve_space(inode
,
4636 round_down(from
, PAGE_CACHE_SIZE
), PAGE_CACHE_SIZE
);
4641 page
= find_or_create_page(mapping
, index
, mask
);
4643 btrfs_delalloc_release_space(inode
,
4644 round_down(from
, PAGE_CACHE_SIZE
),
4650 page_start
= page_offset(page
);
4651 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4653 if (!PageUptodate(page
)) {
4654 ret
= btrfs_readpage(NULL
, page
);
4656 if (page
->mapping
!= mapping
) {
4658 page_cache_release(page
);
4661 if (!PageUptodate(page
)) {
4666 wait_on_page_writeback(page
);
4668 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
4669 set_page_extent_mapped(page
);
4671 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4673 unlock_extent_cached(io_tree
, page_start
, page_end
,
4674 &cached_state
, GFP_NOFS
);
4676 page_cache_release(page
);
4677 btrfs_start_ordered_extent(inode
, ordered
, 1);
4678 btrfs_put_ordered_extent(ordered
);
4682 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
4683 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4684 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4685 0, 0, &cached_state
, GFP_NOFS
);
4687 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
4690 unlock_extent_cached(io_tree
, page_start
, page_end
,
4691 &cached_state
, GFP_NOFS
);
4695 if (offset
!= PAGE_CACHE_SIZE
) {
4697 len
= PAGE_CACHE_SIZE
- offset
;
4700 memset(kaddr
, 0, offset
);
4702 memset(kaddr
+ offset
, 0, len
);
4703 flush_dcache_page(page
);
4706 ClearPageChecked(page
);
4707 set_page_dirty(page
);
4708 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
4713 btrfs_delalloc_release_space(inode
, page_start
,
4716 page_cache_release(page
);
4721 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4722 u64 offset
, u64 len
)
4724 struct btrfs_trans_handle
*trans
;
4728 * Still need to make sure the inode looks like it's been updated so
4729 * that any holes get logged if we fsync.
4731 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
)) {
4732 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
4733 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4734 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4739 * 1 - for the one we're dropping
4740 * 1 - for the one we're adding
4741 * 1 - for updating the inode.
4743 trans
= btrfs_start_transaction(root
, 3);
4745 return PTR_ERR(trans
);
4747 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4749 btrfs_abort_transaction(trans
, root
, ret
);
4750 btrfs_end_transaction(trans
, root
);
4754 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
4755 0, 0, len
, 0, len
, 0, 0, 0);
4757 btrfs_abort_transaction(trans
, root
, ret
);
4759 btrfs_update_inode(trans
, root
, inode
);
4760 btrfs_end_transaction(trans
, root
);
4765 * This function puts in dummy file extents for the area we're creating a hole
4766 * for. So if we are truncating this file to a larger size we need to insert
4767 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4768 * the range between oldsize and size
4770 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4772 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4773 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4774 struct extent_map
*em
= NULL
;
4775 struct extent_state
*cached_state
= NULL
;
4776 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4777 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4778 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4785 * If our size started in the middle of a page we need to zero out the
4786 * rest of the page before we expand the i_size, otherwise we could
4787 * expose stale data.
4789 err
= btrfs_truncate_page(inode
, oldsize
, 0, 0);
4793 if (size
<= hole_start
)
4797 struct btrfs_ordered_extent
*ordered
;
4799 lock_extent_bits(io_tree
, hole_start
, block_end
- 1,
4801 ordered
= btrfs_lookup_ordered_range(inode
, hole_start
,
4802 block_end
- hole_start
);
4805 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4806 &cached_state
, GFP_NOFS
);
4807 btrfs_start_ordered_extent(inode
, ordered
, 1);
4808 btrfs_put_ordered_extent(ordered
);
4811 cur_offset
= hole_start
;
4813 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4814 block_end
- cur_offset
, 0);
4820 last_byte
= min(extent_map_end(em
), block_end
);
4821 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4822 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4823 struct extent_map
*hole_em
;
4824 hole_size
= last_byte
- cur_offset
;
4826 err
= maybe_insert_hole(root
, inode
, cur_offset
,
4830 btrfs_drop_extent_cache(inode
, cur_offset
,
4831 cur_offset
+ hole_size
- 1, 0);
4832 hole_em
= alloc_extent_map();
4834 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4835 &BTRFS_I(inode
)->runtime_flags
);
4838 hole_em
->start
= cur_offset
;
4839 hole_em
->len
= hole_size
;
4840 hole_em
->orig_start
= cur_offset
;
4842 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4843 hole_em
->block_len
= 0;
4844 hole_em
->orig_block_len
= 0;
4845 hole_em
->ram_bytes
= hole_size
;
4846 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4847 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4848 hole_em
->generation
= root
->fs_info
->generation
;
4851 write_lock(&em_tree
->lock
);
4852 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4853 write_unlock(&em_tree
->lock
);
4856 btrfs_drop_extent_cache(inode
, cur_offset
,
4860 free_extent_map(hole_em
);
4863 free_extent_map(em
);
4865 cur_offset
= last_byte
;
4866 if (cur_offset
>= block_end
)
4869 free_extent_map(em
);
4870 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4875 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4877 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4878 struct btrfs_trans_handle
*trans
;
4879 loff_t oldsize
= i_size_read(inode
);
4880 loff_t newsize
= attr
->ia_size
;
4881 int mask
= attr
->ia_valid
;
4885 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4886 * special case where we need to update the times despite not having
4887 * these flags set. For all other operations the VFS set these flags
4888 * explicitly if it wants a timestamp update.
4890 if (newsize
!= oldsize
) {
4891 inode_inc_iversion(inode
);
4892 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
4893 inode
->i_ctime
= inode
->i_mtime
=
4894 current_fs_time(inode
->i_sb
);
4897 if (newsize
> oldsize
) {
4898 truncate_pagecache(inode
, newsize
);
4900 * Don't do an expanding truncate while snapshoting is ongoing.
4901 * This is to ensure the snapshot captures a fully consistent
4902 * state of this file - if the snapshot captures this expanding
4903 * truncation, it must capture all writes that happened before
4906 btrfs_wait_for_snapshot_creation(root
);
4907 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4909 btrfs_end_write_no_snapshoting(root
);
4913 trans
= btrfs_start_transaction(root
, 1);
4914 if (IS_ERR(trans
)) {
4915 btrfs_end_write_no_snapshoting(root
);
4916 return PTR_ERR(trans
);
4919 i_size_write(inode
, newsize
);
4920 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4921 ret
= btrfs_update_inode(trans
, root
, inode
);
4922 btrfs_end_write_no_snapshoting(root
);
4923 btrfs_end_transaction(trans
, root
);
4927 * We're truncating a file that used to have good data down to
4928 * zero. Make sure it gets into the ordered flush list so that
4929 * any new writes get down to disk quickly.
4932 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4933 &BTRFS_I(inode
)->runtime_flags
);
4936 * 1 for the orphan item we're going to add
4937 * 1 for the orphan item deletion.
4939 trans
= btrfs_start_transaction(root
, 2);
4941 return PTR_ERR(trans
);
4944 * We need to do this in case we fail at _any_ point during the
4945 * actual truncate. Once we do the truncate_setsize we could
4946 * invalidate pages which forces any outstanding ordered io to
4947 * be instantly completed which will give us extents that need
4948 * to be truncated. If we fail to get an orphan inode down we
4949 * could have left over extents that were never meant to live,
4950 * so we need to garuntee from this point on that everything
4951 * will be consistent.
4953 ret
= btrfs_orphan_add(trans
, inode
);
4954 btrfs_end_transaction(trans
, root
);
4958 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4959 truncate_setsize(inode
, newsize
);
4961 /* Disable nonlocked read DIO to avoid the end less truncate */
4962 btrfs_inode_block_unlocked_dio(inode
);
4963 inode_dio_wait(inode
);
4964 btrfs_inode_resume_unlocked_dio(inode
);
4966 ret
= btrfs_truncate(inode
);
4967 if (ret
&& inode
->i_nlink
) {
4971 * failed to truncate, disk_i_size is only adjusted down
4972 * as we remove extents, so it should represent the true
4973 * size of the inode, so reset the in memory size and
4974 * delete our orphan entry.
4976 trans
= btrfs_join_transaction(root
);
4977 if (IS_ERR(trans
)) {
4978 btrfs_orphan_del(NULL
, inode
);
4981 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
4982 err
= btrfs_orphan_del(trans
, inode
);
4984 btrfs_abort_transaction(trans
, root
, err
);
4985 btrfs_end_transaction(trans
, root
);
4992 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4994 struct inode
*inode
= d_inode(dentry
);
4995 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4998 if (btrfs_root_readonly(root
))
5001 err
= inode_change_ok(inode
, attr
);
5005 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
5006 err
= btrfs_setsize(inode
, attr
);
5011 if (attr
->ia_valid
) {
5012 setattr_copy(inode
, attr
);
5013 inode_inc_iversion(inode
);
5014 err
= btrfs_dirty_inode(inode
);
5016 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
5017 err
= posix_acl_chmod(inode
, inode
->i_mode
);
5024 * While truncating the inode pages during eviction, we get the VFS calling
5025 * btrfs_invalidatepage() against each page of the inode. This is slow because
5026 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5027 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5028 * extent_state structures over and over, wasting lots of time.
5030 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5031 * those expensive operations on a per page basis and do only the ordered io
5032 * finishing, while we release here the extent_map and extent_state structures,
5033 * without the excessive merging and splitting.
5035 static void evict_inode_truncate_pages(struct inode
*inode
)
5037 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5038 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
5039 struct rb_node
*node
;
5041 ASSERT(inode
->i_state
& I_FREEING
);
5042 truncate_inode_pages_final(&inode
->i_data
);
5044 write_lock(&map_tree
->lock
);
5045 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5046 struct extent_map
*em
;
5048 node
= rb_first(&map_tree
->map
);
5049 em
= rb_entry(node
, struct extent_map
, rb_node
);
5050 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5051 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5052 remove_extent_mapping(map_tree
, em
);
5053 free_extent_map(em
);
5054 if (need_resched()) {
5055 write_unlock(&map_tree
->lock
);
5057 write_lock(&map_tree
->lock
);
5060 write_unlock(&map_tree
->lock
);
5063 * Keep looping until we have no more ranges in the io tree.
5064 * We can have ongoing bios started by readpages (called from readahead)
5065 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5066 * still in progress (unlocked the pages in the bio but did not yet
5067 * unlocked the ranges in the io tree). Therefore this means some
5068 * ranges can still be locked and eviction started because before
5069 * submitting those bios, which are executed by a separate task (work
5070 * queue kthread), inode references (inode->i_count) were not taken
5071 * (which would be dropped in the end io callback of each bio).
5072 * Therefore here we effectively end up waiting for those bios and
5073 * anyone else holding locked ranges without having bumped the inode's
5074 * reference count - if we don't do it, when they access the inode's
5075 * io_tree to unlock a range it may be too late, leading to an
5076 * use-after-free issue.
5078 spin_lock(&io_tree
->lock
);
5079 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5080 struct extent_state
*state
;
5081 struct extent_state
*cached_state
= NULL
;
5085 node
= rb_first(&io_tree
->state
);
5086 state
= rb_entry(node
, struct extent_state
, rb_node
);
5087 start
= state
->start
;
5089 spin_unlock(&io_tree
->lock
);
5091 lock_extent_bits(io_tree
, start
, end
, &cached_state
);
5094 * If still has DELALLOC flag, the extent didn't reach disk,
5095 * and its reserved space won't be freed by delayed_ref.
5096 * So we need to free its reserved space here.
5097 * (Refer to comment in btrfs_invalidatepage, case 2)
5099 * Note, end is the bytenr of last byte, so we need + 1 here.
5101 if (state
->state
& EXTENT_DELALLOC
)
5102 btrfs_qgroup_free_data(inode
, start
, end
- start
+ 1);
5104 clear_extent_bit(io_tree
, start
, end
,
5105 EXTENT_LOCKED
| EXTENT_DIRTY
|
5106 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5107 EXTENT_DEFRAG
, 1, 1,
5108 &cached_state
, GFP_NOFS
);
5111 spin_lock(&io_tree
->lock
);
5113 spin_unlock(&io_tree
->lock
);
5116 void btrfs_evict_inode(struct inode
*inode
)
5118 struct btrfs_trans_handle
*trans
;
5119 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5120 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5121 int steal_from_global
= 0;
5122 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
5125 trace_btrfs_inode_evict(inode
);
5127 evict_inode_truncate_pages(inode
);
5129 if (inode
->i_nlink
&&
5130 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5131 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5132 btrfs_is_free_space_inode(inode
)))
5135 if (is_bad_inode(inode
)) {
5136 btrfs_orphan_del(NULL
, inode
);
5139 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5140 if (!special_file(inode
->i_mode
))
5141 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5143 btrfs_free_io_failure_record(inode
, 0, (u64
)-1);
5145 if (root
->fs_info
->log_root_recovering
) {
5146 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5147 &BTRFS_I(inode
)->runtime_flags
));
5151 if (inode
->i_nlink
> 0) {
5152 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5153 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5157 ret
= btrfs_commit_inode_delayed_inode(inode
);
5159 btrfs_orphan_del(NULL
, inode
);
5163 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
5165 btrfs_orphan_del(NULL
, inode
);
5168 rsv
->size
= min_size
;
5170 global_rsv
= &root
->fs_info
->global_block_rsv
;
5172 btrfs_i_size_write(inode
, 0);
5175 * This is a bit simpler than btrfs_truncate since we've already
5176 * reserved our space for our orphan item in the unlink, so we just
5177 * need to reserve some slack space in case we add bytes and update
5178 * inode item when doing the truncate.
5181 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5182 BTRFS_RESERVE_FLUSH_LIMIT
);
5185 * Try and steal from the global reserve since we will
5186 * likely not use this space anyway, we want to try as
5187 * hard as possible to get this to work.
5190 steal_from_global
++;
5192 steal_from_global
= 0;
5196 * steal_from_global == 0: we reserved stuff, hooray!
5197 * steal_from_global == 1: we didn't reserve stuff, boo!
5198 * steal_from_global == 2: we've committed, still not a lot of
5199 * room but maybe we'll have room in the global reserve this
5201 * steal_from_global == 3: abandon all hope!
5203 if (steal_from_global
> 2) {
5204 btrfs_warn(root
->fs_info
,
5205 "Could not get space for a delete, will truncate on mount %d",
5207 btrfs_orphan_del(NULL
, inode
);
5208 btrfs_free_block_rsv(root
, rsv
);
5212 trans
= btrfs_join_transaction(root
);
5213 if (IS_ERR(trans
)) {
5214 btrfs_orphan_del(NULL
, inode
);
5215 btrfs_free_block_rsv(root
, rsv
);
5220 * We can't just steal from the global reserve, we need tomake
5221 * sure there is room to do it, if not we need to commit and try
5224 if (steal_from_global
) {
5225 if (!btrfs_check_space_for_delayed_refs(trans
, root
))
5226 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5233 * Couldn't steal from the global reserve, we have too much
5234 * pending stuff built up, commit the transaction and try it
5238 ret
= btrfs_commit_transaction(trans
, root
);
5240 btrfs_orphan_del(NULL
, inode
);
5241 btrfs_free_block_rsv(root
, rsv
);
5246 steal_from_global
= 0;
5249 trans
->block_rsv
= rsv
;
5251 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5252 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5255 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5256 btrfs_end_transaction(trans
, root
);
5258 btrfs_btree_balance_dirty(root
);
5261 btrfs_free_block_rsv(root
, rsv
);
5264 * Errors here aren't a big deal, it just means we leave orphan items
5265 * in the tree. They will be cleaned up on the next mount.
5268 trans
->block_rsv
= root
->orphan_block_rsv
;
5269 btrfs_orphan_del(trans
, inode
);
5271 btrfs_orphan_del(NULL
, inode
);
5274 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5275 if (!(root
== root
->fs_info
->tree_root
||
5276 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5277 btrfs_return_ino(root
, btrfs_ino(inode
));
5279 btrfs_end_transaction(trans
, root
);
5280 btrfs_btree_balance_dirty(root
);
5282 btrfs_remove_delayed_node(inode
);
5287 * this returns the key found in the dir entry in the location pointer.
5288 * If no dir entries were found, location->objectid is 0.
5290 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5291 struct btrfs_key
*location
)
5293 const char *name
= dentry
->d_name
.name
;
5294 int namelen
= dentry
->d_name
.len
;
5295 struct btrfs_dir_item
*di
;
5296 struct btrfs_path
*path
;
5297 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5300 path
= btrfs_alloc_path();
5304 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
5309 if (IS_ERR_OR_NULL(di
))
5312 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5314 btrfs_free_path(path
);
5317 location
->objectid
= 0;
5322 * when we hit a tree root in a directory, the btrfs part of the inode
5323 * needs to be changed to reflect the root directory of the tree root. This
5324 * is kind of like crossing a mount point.
5326 static int fixup_tree_root_location(struct btrfs_root
*root
,
5328 struct dentry
*dentry
,
5329 struct btrfs_key
*location
,
5330 struct btrfs_root
**sub_root
)
5332 struct btrfs_path
*path
;
5333 struct btrfs_root
*new_root
;
5334 struct btrfs_root_ref
*ref
;
5335 struct extent_buffer
*leaf
;
5336 struct btrfs_key key
;
5340 path
= btrfs_alloc_path();
5347 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5348 key
.type
= BTRFS_ROOT_REF_KEY
;
5349 key
.offset
= location
->objectid
;
5351 ret
= btrfs_search_slot(NULL
, root
->fs_info
->tree_root
, &key
, path
,
5359 leaf
= path
->nodes
[0];
5360 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5361 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
5362 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5365 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5366 (unsigned long)(ref
+ 1),
5367 dentry
->d_name
.len
);
5371 btrfs_release_path(path
);
5373 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
5374 if (IS_ERR(new_root
)) {
5375 err
= PTR_ERR(new_root
);
5379 *sub_root
= new_root
;
5380 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5381 location
->type
= BTRFS_INODE_ITEM_KEY
;
5382 location
->offset
= 0;
5385 btrfs_free_path(path
);
5389 static void inode_tree_add(struct inode
*inode
)
5391 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5392 struct btrfs_inode
*entry
;
5394 struct rb_node
*parent
;
5395 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5396 u64 ino
= btrfs_ino(inode
);
5398 if (inode_unhashed(inode
))
5401 spin_lock(&root
->inode_lock
);
5402 p
= &root
->inode_tree
.rb_node
;
5405 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5407 if (ino
< btrfs_ino(&entry
->vfs_inode
))
5408 p
= &parent
->rb_left
;
5409 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
5410 p
= &parent
->rb_right
;
5412 WARN_ON(!(entry
->vfs_inode
.i_state
&
5413 (I_WILL_FREE
| I_FREEING
)));
5414 rb_replace_node(parent
, new, &root
->inode_tree
);
5415 RB_CLEAR_NODE(parent
);
5416 spin_unlock(&root
->inode_lock
);
5420 rb_link_node(new, parent
, p
);
5421 rb_insert_color(new, &root
->inode_tree
);
5422 spin_unlock(&root
->inode_lock
);
5425 static void inode_tree_del(struct inode
*inode
)
5427 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5430 spin_lock(&root
->inode_lock
);
5431 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5432 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5433 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5434 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5436 spin_unlock(&root
->inode_lock
);
5438 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5439 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
5440 spin_lock(&root
->inode_lock
);
5441 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5442 spin_unlock(&root
->inode_lock
);
5444 btrfs_add_dead_root(root
);
5448 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5450 struct rb_node
*node
;
5451 struct rb_node
*prev
;
5452 struct btrfs_inode
*entry
;
5453 struct inode
*inode
;
5456 if (!test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
5457 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5459 spin_lock(&root
->inode_lock
);
5461 node
= root
->inode_tree
.rb_node
;
5465 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5467 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
5468 node
= node
->rb_left
;
5469 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
5470 node
= node
->rb_right
;
5476 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5477 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
5481 prev
= rb_next(prev
);
5485 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5486 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
5487 inode
= igrab(&entry
->vfs_inode
);
5489 spin_unlock(&root
->inode_lock
);
5490 if (atomic_read(&inode
->i_count
) > 1)
5491 d_prune_aliases(inode
);
5493 * btrfs_drop_inode will have it removed from
5494 * the inode cache when its usage count
5499 spin_lock(&root
->inode_lock
);
5503 if (cond_resched_lock(&root
->inode_lock
))
5506 node
= rb_next(node
);
5508 spin_unlock(&root
->inode_lock
);
5511 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5513 struct btrfs_iget_args
*args
= p
;
5514 inode
->i_ino
= args
->location
->objectid
;
5515 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5516 sizeof(*args
->location
));
5517 BTRFS_I(inode
)->root
= args
->root
;
5521 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5523 struct btrfs_iget_args
*args
= opaque
;
5524 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5525 args
->root
== BTRFS_I(inode
)->root
;
5528 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5529 struct btrfs_key
*location
,
5530 struct btrfs_root
*root
)
5532 struct inode
*inode
;
5533 struct btrfs_iget_args args
;
5534 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5536 args
.location
= location
;
5539 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5540 btrfs_init_locked_inode
,
5545 /* Get an inode object given its location and corresponding root.
5546 * Returns in *is_new if the inode was read from disk
5548 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5549 struct btrfs_root
*root
, int *new)
5551 struct inode
*inode
;
5553 inode
= btrfs_iget_locked(s
, location
, root
);
5555 return ERR_PTR(-ENOMEM
);
5557 if (inode
->i_state
& I_NEW
) {
5558 btrfs_read_locked_inode(inode
);
5559 if (!is_bad_inode(inode
)) {
5560 inode_tree_add(inode
);
5561 unlock_new_inode(inode
);
5565 unlock_new_inode(inode
);
5567 inode
= ERR_PTR(-ESTALE
);
5574 static struct inode
*new_simple_dir(struct super_block
*s
,
5575 struct btrfs_key
*key
,
5576 struct btrfs_root
*root
)
5578 struct inode
*inode
= new_inode(s
);
5581 return ERR_PTR(-ENOMEM
);
5583 BTRFS_I(inode
)->root
= root
;
5584 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5585 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5587 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5588 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5589 inode
->i_fop
= &simple_dir_operations
;
5590 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5591 inode
->i_mtime
= CURRENT_TIME
;
5592 inode
->i_atime
= inode
->i_mtime
;
5593 inode
->i_ctime
= inode
->i_mtime
;
5594 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5599 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5601 struct inode
*inode
;
5602 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5603 struct btrfs_root
*sub_root
= root
;
5604 struct btrfs_key location
;
5608 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5609 return ERR_PTR(-ENAMETOOLONG
);
5611 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5613 return ERR_PTR(ret
);
5615 if (location
.objectid
== 0)
5616 return ERR_PTR(-ENOENT
);
5618 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5619 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5623 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5625 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5626 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5627 &location
, &sub_root
);
5630 inode
= ERR_PTR(ret
);
5632 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5634 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5636 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5638 if (!IS_ERR(inode
) && root
!= sub_root
) {
5639 down_read(&root
->fs_info
->cleanup_work_sem
);
5640 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5641 ret
= btrfs_orphan_cleanup(sub_root
);
5642 up_read(&root
->fs_info
->cleanup_work_sem
);
5645 inode
= ERR_PTR(ret
);
5652 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5654 struct btrfs_root
*root
;
5655 struct inode
*inode
= d_inode(dentry
);
5657 if (!inode
&& !IS_ROOT(dentry
))
5658 inode
= d_inode(dentry
->d_parent
);
5661 root
= BTRFS_I(inode
)->root
;
5662 if (btrfs_root_refs(&root
->root_item
) == 0)
5665 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5671 static void btrfs_dentry_release(struct dentry
*dentry
)
5673 kfree(dentry
->d_fsdata
);
5676 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5679 struct inode
*inode
;
5681 inode
= btrfs_lookup_dentry(dir
, dentry
);
5682 if (IS_ERR(inode
)) {
5683 if (PTR_ERR(inode
) == -ENOENT
)
5686 return ERR_CAST(inode
);
5689 return d_splice_alias(inode
, dentry
);
5692 unsigned char btrfs_filetype_table
[] = {
5693 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5696 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5698 struct inode
*inode
= file_inode(file
);
5699 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5700 struct btrfs_item
*item
;
5701 struct btrfs_dir_item
*di
;
5702 struct btrfs_key key
;
5703 struct btrfs_key found_key
;
5704 struct btrfs_path
*path
;
5705 struct list_head ins_list
;
5706 struct list_head del_list
;
5708 struct extent_buffer
*leaf
;
5710 unsigned char d_type
;
5715 int key_type
= BTRFS_DIR_INDEX_KEY
;
5719 int is_curr
= 0; /* ctx->pos points to the current index? */
5721 /* FIXME, use a real flag for deciding about the key type */
5722 if (root
->fs_info
->tree_root
== root
)
5723 key_type
= BTRFS_DIR_ITEM_KEY
;
5725 if (!dir_emit_dots(file
, ctx
))
5728 path
= btrfs_alloc_path();
5732 path
->reada
= READA_FORWARD
;
5734 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5735 INIT_LIST_HEAD(&ins_list
);
5736 INIT_LIST_HEAD(&del_list
);
5737 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5740 key
.type
= key_type
;
5741 key
.offset
= ctx
->pos
;
5742 key
.objectid
= btrfs_ino(inode
);
5744 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5749 leaf
= path
->nodes
[0];
5750 slot
= path
->slots
[0];
5751 if (slot
>= btrfs_header_nritems(leaf
)) {
5752 ret
= btrfs_next_leaf(root
, path
);
5760 item
= btrfs_item_nr(slot
);
5761 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5763 if (found_key
.objectid
!= key
.objectid
)
5765 if (found_key
.type
!= key_type
)
5767 if (found_key
.offset
< ctx
->pos
)
5769 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5770 btrfs_should_delete_dir_index(&del_list
,
5774 ctx
->pos
= found_key
.offset
;
5777 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5779 di_total
= btrfs_item_size(leaf
, item
);
5781 while (di_cur
< di_total
) {
5782 struct btrfs_key location
;
5784 if (verify_dir_item(root
, leaf
, di
))
5787 name_len
= btrfs_dir_name_len(leaf
, di
);
5788 if (name_len
<= sizeof(tmp_name
)) {
5789 name_ptr
= tmp_name
;
5791 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5797 read_extent_buffer(leaf
, name_ptr
,
5798 (unsigned long)(di
+ 1), name_len
);
5800 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5801 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5804 /* is this a reference to our own snapshot? If so
5807 * In contrast to old kernels, we insert the snapshot's
5808 * dir item and dir index after it has been created, so
5809 * we won't find a reference to our own snapshot. We
5810 * still keep the following code for backward
5813 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5814 location
.objectid
== root
->root_key
.objectid
) {
5818 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5819 location
.objectid
, d_type
);
5822 if (name_ptr
!= tmp_name
)
5827 di_len
= btrfs_dir_name_len(leaf
, di
) +
5828 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5830 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5836 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5839 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
5844 /* Reached end of directory/root. Bump pos past the last item. */
5848 * Stop new entries from being returned after we return the last
5851 * New directory entries are assigned a strictly increasing
5852 * offset. This means that new entries created during readdir
5853 * are *guaranteed* to be seen in the future by that readdir.
5854 * This has broken buggy programs which operate on names as
5855 * they're returned by readdir. Until we re-use freed offsets
5856 * we have this hack to stop new entries from being returned
5857 * under the assumption that they'll never reach this huge
5860 * This is being careful not to overflow 32bit loff_t unless the
5861 * last entry requires it because doing so has broken 32bit apps
5864 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5865 if (ctx
->pos
>= INT_MAX
)
5866 ctx
->pos
= LLONG_MAX
;
5873 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5874 btrfs_put_delayed_items(&ins_list
, &del_list
);
5875 btrfs_free_path(path
);
5879 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5881 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5882 struct btrfs_trans_handle
*trans
;
5884 bool nolock
= false;
5886 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5889 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5892 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5894 trans
= btrfs_join_transaction_nolock(root
);
5896 trans
= btrfs_join_transaction(root
);
5898 return PTR_ERR(trans
);
5899 ret
= btrfs_commit_transaction(trans
, root
);
5905 * This is somewhat expensive, updating the tree every time the
5906 * inode changes. But, it is most likely to find the inode in cache.
5907 * FIXME, needs more benchmarking...there are no reasons other than performance
5908 * to keep or drop this code.
5910 static int btrfs_dirty_inode(struct inode
*inode
)
5912 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5913 struct btrfs_trans_handle
*trans
;
5916 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5919 trans
= btrfs_join_transaction(root
);
5921 return PTR_ERR(trans
);
5923 ret
= btrfs_update_inode(trans
, root
, inode
);
5924 if (ret
&& ret
== -ENOSPC
) {
5925 /* whoops, lets try again with the full transaction */
5926 btrfs_end_transaction(trans
, root
);
5927 trans
= btrfs_start_transaction(root
, 1);
5929 return PTR_ERR(trans
);
5931 ret
= btrfs_update_inode(trans
, root
, inode
);
5933 btrfs_end_transaction(trans
, root
);
5934 if (BTRFS_I(inode
)->delayed_node
)
5935 btrfs_balance_delayed_items(root
);
5941 * This is a copy of file_update_time. We need this so we can return error on
5942 * ENOSPC for updating the inode in the case of file write and mmap writes.
5944 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
5947 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5949 if (btrfs_root_readonly(root
))
5952 if (flags
& S_VERSION
)
5953 inode_inc_iversion(inode
);
5954 if (flags
& S_CTIME
)
5955 inode
->i_ctime
= *now
;
5956 if (flags
& S_MTIME
)
5957 inode
->i_mtime
= *now
;
5958 if (flags
& S_ATIME
)
5959 inode
->i_atime
= *now
;
5960 return btrfs_dirty_inode(inode
);
5964 * find the highest existing sequence number in a directory
5965 * and then set the in-memory index_cnt variable to reflect
5966 * free sequence numbers
5968 static int btrfs_set_inode_index_count(struct inode
*inode
)
5970 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5971 struct btrfs_key key
, found_key
;
5972 struct btrfs_path
*path
;
5973 struct extent_buffer
*leaf
;
5976 key
.objectid
= btrfs_ino(inode
);
5977 key
.type
= BTRFS_DIR_INDEX_KEY
;
5978 key
.offset
= (u64
)-1;
5980 path
= btrfs_alloc_path();
5984 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5987 /* FIXME: we should be able to handle this */
5993 * MAGIC NUMBER EXPLANATION:
5994 * since we search a directory based on f_pos we have to start at 2
5995 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5996 * else has to start at 2
5998 if (path
->slots
[0] == 0) {
5999 BTRFS_I(inode
)->index_cnt
= 2;
6005 leaf
= path
->nodes
[0];
6006 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6008 if (found_key
.objectid
!= btrfs_ino(inode
) ||
6009 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
6010 BTRFS_I(inode
)->index_cnt
= 2;
6014 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
6016 btrfs_free_path(path
);
6021 * helper to find a free sequence number in a given directory. This current
6022 * code is very simple, later versions will do smarter things in the btree
6024 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
6028 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
6029 ret
= btrfs_inode_delayed_dir_index_count(dir
);
6031 ret
= btrfs_set_inode_index_count(dir
);
6037 *index
= BTRFS_I(dir
)->index_cnt
;
6038 BTRFS_I(dir
)->index_cnt
++;
6043 static int btrfs_insert_inode_locked(struct inode
*inode
)
6045 struct btrfs_iget_args args
;
6046 args
.location
= &BTRFS_I(inode
)->location
;
6047 args
.root
= BTRFS_I(inode
)->root
;
6049 return insert_inode_locked4(inode
,
6050 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
6051 btrfs_find_actor
, &args
);
6054 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6055 struct btrfs_root
*root
,
6057 const char *name
, int name_len
,
6058 u64 ref_objectid
, u64 objectid
,
6059 umode_t mode
, u64
*index
)
6061 struct inode
*inode
;
6062 struct btrfs_inode_item
*inode_item
;
6063 struct btrfs_key
*location
;
6064 struct btrfs_path
*path
;
6065 struct btrfs_inode_ref
*ref
;
6066 struct btrfs_key key
[2];
6068 int nitems
= name
? 2 : 1;
6072 path
= btrfs_alloc_path();
6074 return ERR_PTR(-ENOMEM
);
6076 inode
= new_inode(root
->fs_info
->sb
);
6078 btrfs_free_path(path
);
6079 return ERR_PTR(-ENOMEM
);
6083 * O_TMPFILE, set link count to 0, so that after this point,
6084 * we fill in an inode item with the correct link count.
6087 set_nlink(inode
, 0);
6090 * we have to initialize this early, so we can reclaim the inode
6091 * number if we fail afterwards in this function.
6093 inode
->i_ino
= objectid
;
6096 trace_btrfs_inode_request(dir
);
6098 ret
= btrfs_set_inode_index(dir
, index
);
6100 btrfs_free_path(path
);
6102 return ERR_PTR(ret
);
6108 * index_cnt is ignored for everything but a dir,
6109 * btrfs_get_inode_index_count has an explanation for the magic
6112 BTRFS_I(inode
)->index_cnt
= 2;
6113 BTRFS_I(inode
)->dir_index
= *index
;
6114 BTRFS_I(inode
)->root
= root
;
6115 BTRFS_I(inode
)->generation
= trans
->transid
;
6116 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6119 * We could have gotten an inode number from somebody who was fsynced
6120 * and then removed in this same transaction, so let's just set full
6121 * sync since it will be a full sync anyway and this will blow away the
6122 * old info in the log.
6124 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6126 key
[0].objectid
= objectid
;
6127 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6130 sizes
[0] = sizeof(struct btrfs_inode_item
);
6134 * Start new inodes with an inode_ref. This is slightly more
6135 * efficient for small numbers of hard links since they will
6136 * be packed into one item. Extended refs will kick in if we
6137 * add more hard links than can fit in the ref item.
6139 key
[1].objectid
= objectid
;
6140 key
[1].type
= BTRFS_INODE_REF_KEY
;
6141 key
[1].offset
= ref_objectid
;
6143 sizes
[1] = name_len
+ sizeof(*ref
);
6146 location
= &BTRFS_I(inode
)->location
;
6147 location
->objectid
= objectid
;
6148 location
->offset
= 0;
6149 location
->type
= BTRFS_INODE_ITEM_KEY
;
6151 ret
= btrfs_insert_inode_locked(inode
);
6155 path
->leave_spinning
= 1;
6156 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6160 inode_init_owner(inode
, dir
, mode
);
6161 inode_set_bytes(inode
, 0);
6163 inode
->i_mtime
= CURRENT_TIME
;
6164 inode
->i_atime
= inode
->i_mtime
;
6165 inode
->i_ctime
= inode
->i_mtime
;
6166 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6168 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6169 struct btrfs_inode_item
);
6170 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
6171 sizeof(*inode_item
));
6172 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6175 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6176 struct btrfs_inode_ref
);
6177 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6178 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6179 ptr
= (unsigned long)(ref
+ 1);
6180 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6183 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6184 btrfs_free_path(path
);
6186 btrfs_inherit_iflags(inode
, dir
);
6188 if (S_ISREG(mode
)) {
6189 if (btrfs_test_opt(root
, NODATASUM
))
6190 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6191 if (btrfs_test_opt(root
, NODATACOW
))
6192 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6193 BTRFS_INODE_NODATASUM
;
6196 inode_tree_add(inode
);
6198 trace_btrfs_inode_new(inode
);
6199 btrfs_set_inode_last_trans(trans
, inode
);
6201 btrfs_update_root_times(trans
, root
);
6203 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6205 btrfs_err(root
->fs_info
,
6206 "error inheriting props for ino %llu (root %llu): %d",
6207 btrfs_ino(inode
), root
->root_key
.objectid
, ret
);
6212 unlock_new_inode(inode
);
6215 BTRFS_I(dir
)->index_cnt
--;
6216 btrfs_free_path(path
);
6218 return ERR_PTR(ret
);
6221 static inline u8
btrfs_inode_type(struct inode
*inode
)
6223 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6227 * utility function to add 'inode' into 'parent_inode' with
6228 * a give name and a given sequence number.
6229 * if 'add_backref' is true, also insert a backref from the
6230 * inode to the parent directory.
6232 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6233 struct inode
*parent_inode
, struct inode
*inode
,
6234 const char *name
, int name_len
, int add_backref
, u64 index
)
6237 struct btrfs_key key
;
6238 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
6239 u64 ino
= btrfs_ino(inode
);
6240 u64 parent_ino
= btrfs_ino(parent_inode
);
6242 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6243 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
6246 key
.type
= BTRFS_INODE_ITEM_KEY
;
6250 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6251 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
6252 key
.objectid
, root
->root_key
.objectid
,
6253 parent_ino
, index
, name
, name_len
);
6254 } else if (add_backref
) {
6255 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6259 /* Nothing to clean up yet */
6263 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6265 btrfs_inode_type(inode
), index
);
6266 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6269 btrfs_abort_transaction(trans
, root
, ret
);
6273 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
6275 inode_inc_iversion(parent_inode
);
6276 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
6277 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
6279 btrfs_abort_transaction(trans
, root
, ret
);
6283 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6286 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
6287 key
.objectid
, root
->root_key
.objectid
,
6288 parent_ino
, &local_index
, name
, name_len
);
6290 } else if (add_backref
) {
6294 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6295 ino
, parent_ino
, &local_index
);
6300 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6301 struct inode
*dir
, struct dentry
*dentry
,
6302 struct inode
*inode
, int backref
, u64 index
)
6304 int err
= btrfs_add_link(trans
, dir
, inode
,
6305 dentry
->d_name
.name
, dentry
->d_name
.len
,
6312 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6313 umode_t mode
, dev_t rdev
)
6315 struct btrfs_trans_handle
*trans
;
6316 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6317 struct inode
*inode
= NULL
;
6324 * 2 for inode item and ref
6326 * 1 for xattr if selinux is on
6328 trans
= btrfs_start_transaction(root
, 5);
6330 return PTR_ERR(trans
);
6332 err
= btrfs_find_free_ino(root
, &objectid
);
6336 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6337 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6339 if (IS_ERR(inode
)) {
6340 err
= PTR_ERR(inode
);
6345 * If the active LSM wants to access the inode during
6346 * d_instantiate it needs these. Smack checks to see
6347 * if the filesystem supports xattrs by looking at the
6350 inode
->i_op
= &btrfs_special_inode_operations
;
6351 init_special_inode(inode
, inode
->i_mode
, rdev
);
6353 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6355 goto out_unlock_inode
;
6357 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6359 goto out_unlock_inode
;
6361 btrfs_update_inode(trans
, root
, inode
);
6362 unlock_new_inode(inode
);
6363 d_instantiate(dentry
, inode
);
6367 btrfs_end_transaction(trans
, root
);
6368 btrfs_balance_delayed_items(root
);
6369 btrfs_btree_balance_dirty(root
);
6371 inode_dec_link_count(inode
);
6378 unlock_new_inode(inode
);
6383 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6384 umode_t mode
, bool excl
)
6386 struct btrfs_trans_handle
*trans
;
6387 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6388 struct inode
*inode
= NULL
;
6389 int drop_inode_on_err
= 0;
6395 * 2 for inode item and ref
6397 * 1 for xattr if selinux is on
6399 trans
= btrfs_start_transaction(root
, 5);
6401 return PTR_ERR(trans
);
6403 err
= btrfs_find_free_ino(root
, &objectid
);
6407 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6408 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6410 if (IS_ERR(inode
)) {
6411 err
= PTR_ERR(inode
);
6414 drop_inode_on_err
= 1;
6416 * If the active LSM wants to access the inode during
6417 * d_instantiate it needs these. Smack checks to see
6418 * if the filesystem supports xattrs by looking at the
6421 inode
->i_fop
= &btrfs_file_operations
;
6422 inode
->i_op
= &btrfs_file_inode_operations
;
6423 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6425 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6427 goto out_unlock_inode
;
6429 err
= btrfs_update_inode(trans
, root
, inode
);
6431 goto out_unlock_inode
;
6433 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6435 goto out_unlock_inode
;
6437 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6438 unlock_new_inode(inode
);
6439 d_instantiate(dentry
, inode
);
6442 btrfs_end_transaction(trans
, root
);
6443 if (err
&& drop_inode_on_err
) {
6444 inode_dec_link_count(inode
);
6447 btrfs_balance_delayed_items(root
);
6448 btrfs_btree_balance_dirty(root
);
6452 unlock_new_inode(inode
);
6457 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6458 struct dentry
*dentry
)
6460 struct btrfs_trans_handle
*trans
= NULL
;
6461 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6462 struct inode
*inode
= d_inode(old_dentry
);
6467 /* do not allow sys_link's with other subvols of the same device */
6468 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6471 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6474 err
= btrfs_set_inode_index(dir
, &index
);
6479 * 2 items for inode and inode ref
6480 * 2 items for dir items
6481 * 1 item for parent inode
6483 trans
= btrfs_start_transaction(root
, 5);
6484 if (IS_ERR(trans
)) {
6485 err
= PTR_ERR(trans
);
6490 /* There are several dir indexes for this inode, clear the cache. */
6491 BTRFS_I(inode
)->dir_index
= 0ULL;
6493 inode_inc_iversion(inode
);
6494 inode
->i_ctime
= CURRENT_TIME
;
6496 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6498 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
6503 struct dentry
*parent
= dentry
->d_parent
;
6504 err
= btrfs_update_inode(trans
, root
, inode
);
6507 if (inode
->i_nlink
== 1) {
6509 * If new hard link count is 1, it's a file created
6510 * with open(2) O_TMPFILE flag.
6512 err
= btrfs_orphan_del(trans
, inode
);
6516 d_instantiate(dentry
, inode
);
6517 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
6520 btrfs_balance_delayed_items(root
);
6523 btrfs_end_transaction(trans
, root
);
6525 inode_dec_link_count(inode
);
6528 btrfs_btree_balance_dirty(root
);
6532 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6534 struct inode
*inode
= NULL
;
6535 struct btrfs_trans_handle
*trans
;
6536 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6538 int drop_on_err
= 0;
6543 * 2 items for inode and ref
6544 * 2 items for dir items
6545 * 1 for xattr if selinux is on
6547 trans
= btrfs_start_transaction(root
, 5);
6549 return PTR_ERR(trans
);
6551 err
= btrfs_find_free_ino(root
, &objectid
);
6555 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6556 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6557 S_IFDIR
| mode
, &index
);
6558 if (IS_ERR(inode
)) {
6559 err
= PTR_ERR(inode
);
6564 /* these must be set before we unlock the inode */
6565 inode
->i_op
= &btrfs_dir_inode_operations
;
6566 inode
->i_fop
= &btrfs_dir_file_operations
;
6568 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6570 goto out_fail_inode
;
6572 btrfs_i_size_write(inode
, 0);
6573 err
= btrfs_update_inode(trans
, root
, inode
);
6575 goto out_fail_inode
;
6577 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
6578 dentry
->d_name
.len
, 0, index
);
6580 goto out_fail_inode
;
6582 d_instantiate(dentry
, inode
);
6584 * mkdir is special. We're unlocking after we call d_instantiate
6585 * to avoid a race with nfsd calling d_instantiate.
6587 unlock_new_inode(inode
);
6591 btrfs_end_transaction(trans
, root
);
6593 inode_dec_link_count(inode
);
6596 btrfs_balance_delayed_items(root
);
6597 btrfs_btree_balance_dirty(root
);
6601 unlock_new_inode(inode
);
6605 /* Find next extent map of a given extent map, caller needs to ensure locks */
6606 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6608 struct rb_node
*next
;
6610 next
= rb_next(&em
->rb_node
);
6613 return container_of(next
, struct extent_map
, rb_node
);
6616 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6618 struct rb_node
*prev
;
6620 prev
= rb_prev(&em
->rb_node
);
6623 return container_of(prev
, struct extent_map
, rb_node
);
6626 /* helper for btfs_get_extent. Given an existing extent in the tree,
6627 * the existing extent is the nearest extent to map_start,
6628 * and an extent that you want to insert, deal with overlap and insert
6629 * the best fitted new extent into the tree.
6631 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6632 struct extent_map
*existing
,
6633 struct extent_map
*em
,
6636 struct extent_map
*prev
;
6637 struct extent_map
*next
;
6642 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6644 if (existing
->start
> map_start
) {
6646 prev
= prev_extent_map(next
);
6649 next
= next_extent_map(prev
);
6652 start
= prev
? extent_map_end(prev
) : em
->start
;
6653 start
= max_t(u64
, start
, em
->start
);
6654 end
= next
? next
->start
: extent_map_end(em
);
6655 end
= min_t(u64
, end
, extent_map_end(em
));
6656 start_diff
= start
- em
->start
;
6658 em
->len
= end
- start
;
6659 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6660 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6661 em
->block_start
+= start_diff
;
6662 em
->block_len
-= start_diff
;
6664 return add_extent_mapping(em_tree
, em
, 0);
6667 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6669 size_t pg_offset
, u64 extent_offset
,
6670 struct btrfs_file_extent_item
*item
)
6673 struct extent_buffer
*leaf
= path
->nodes
[0];
6676 unsigned long inline_size
;
6680 WARN_ON(pg_offset
!= 0);
6681 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6682 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6683 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6684 btrfs_item_nr(path
->slots
[0]));
6685 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6688 ptr
= btrfs_file_extent_inline_start(item
);
6690 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6692 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
6693 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6694 extent_offset
, inline_size
, max_size
);
6700 * a bit scary, this does extent mapping from logical file offset to the disk.
6701 * the ugly parts come from merging extents from the disk with the in-ram
6702 * representation. This gets more complex because of the data=ordered code,
6703 * where the in-ram extents might be locked pending data=ordered completion.
6705 * This also copies inline extents directly into the page.
6708 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6709 size_t pg_offset
, u64 start
, u64 len
,
6714 u64 extent_start
= 0;
6716 u64 objectid
= btrfs_ino(inode
);
6718 struct btrfs_path
*path
= NULL
;
6719 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6720 struct btrfs_file_extent_item
*item
;
6721 struct extent_buffer
*leaf
;
6722 struct btrfs_key found_key
;
6723 struct extent_map
*em
= NULL
;
6724 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6725 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6726 struct btrfs_trans_handle
*trans
= NULL
;
6727 const bool new_inline
= !page
|| create
;
6730 read_lock(&em_tree
->lock
);
6731 em
= lookup_extent_mapping(em_tree
, start
, len
);
6733 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6734 read_unlock(&em_tree
->lock
);
6737 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6738 free_extent_map(em
);
6739 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6740 free_extent_map(em
);
6744 em
= alloc_extent_map();
6749 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6750 em
->start
= EXTENT_MAP_HOLE
;
6751 em
->orig_start
= EXTENT_MAP_HOLE
;
6753 em
->block_len
= (u64
)-1;
6756 path
= btrfs_alloc_path();
6762 * Chances are we'll be called again, so go ahead and do
6765 path
->reada
= READA_FORWARD
;
6768 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6769 objectid
, start
, trans
!= NULL
);
6776 if (path
->slots
[0] == 0)
6781 leaf
= path
->nodes
[0];
6782 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6783 struct btrfs_file_extent_item
);
6784 /* are we inside the extent that was found? */
6785 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6786 found_type
= found_key
.type
;
6787 if (found_key
.objectid
!= objectid
||
6788 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6790 * If we backup past the first extent we want to move forward
6791 * and see if there is an extent in front of us, otherwise we'll
6792 * say there is a hole for our whole search range which can
6799 found_type
= btrfs_file_extent_type(leaf
, item
);
6800 extent_start
= found_key
.offset
;
6801 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6802 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6803 extent_end
= extent_start
+
6804 btrfs_file_extent_num_bytes(leaf
, item
);
6805 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6807 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6808 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6811 if (start
>= extent_end
) {
6813 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6814 ret
= btrfs_next_leaf(root
, path
);
6821 leaf
= path
->nodes
[0];
6823 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6824 if (found_key
.objectid
!= objectid
||
6825 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6827 if (start
+ len
<= found_key
.offset
)
6829 if (start
> found_key
.offset
)
6832 em
->orig_start
= start
;
6833 em
->len
= found_key
.offset
- start
;
6837 btrfs_extent_item_to_extent_map(inode
, path
, item
, new_inline
, em
);
6839 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6840 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6842 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6846 size_t extent_offset
;
6852 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6853 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6854 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6855 size
- extent_offset
);
6856 em
->start
= extent_start
+ extent_offset
;
6857 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6858 em
->orig_block_len
= em
->len
;
6859 em
->orig_start
= em
->start
;
6860 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6861 if (create
== 0 && !PageUptodate(page
)) {
6862 if (btrfs_file_extent_compression(leaf
, item
) !=
6863 BTRFS_COMPRESS_NONE
) {
6864 ret
= uncompress_inline(path
, page
, pg_offset
,
6865 extent_offset
, item
);
6872 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6874 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6875 memset(map
+ pg_offset
+ copy_size
, 0,
6876 PAGE_CACHE_SIZE
- pg_offset
-
6881 flush_dcache_page(page
);
6882 } else if (create
&& PageUptodate(page
)) {
6886 free_extent_map(em
);
6889 btrfs_release_path(path
);
6890 trans
= btrfs_join_transaction(root
);
6893 return ERR_CAST(trans
);
6897 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6900 btrfs_mark_buffer_dirty(leaf
);
6902 set_extent_uptodate(io_tree
, em
->start
,
6903 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6908 em
->orig_start
= start
;
6911 em
->block_start
= EXTENT_MAP_HOLE
;
6912 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6914 btrfs_release_path(path
);
6915 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6916 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6917 em
->start
, em
->len
, start
, len
);
6923 write_lock(&em_tree
->lock
);
6924 ret
= add_extent_mapping(em_tree
, em
, 0);
6925 /* it is possible that someone inserted the extent into the tree
6926 * while we had the lock dropped. It is also possible that
6927 * an overlapping map exists in the tree
6929 if (ret
== -EEXIST
) {
6930 struct extent_map
*existing
;
6934 existing
= search_extent_mapping(em_tree
, start
, len
);
6936 * existing will always be non-NULL, since there must be
6937 * extent causing the -EEXIST.
6939 if (start
>= extent_map_end(existing
) ||
6940 start
<= existing
->start
) {
6942 * The existing extent map is the one nearest to
6943 * the [start, start + len) range which overlaps
6945 err
= merge_extent_mapping(em_tree
, existing
,
6947 free_extent_map(existing
);
6949 free_extent_map(em
);
6953 free_extent_map(em
);
6958 write_unlock(&em_tree
->lock
);
6961 trace_btrfs_get_extent(root
, em
);
6963 btrfs_free_path(path
);
6965 ret
= btrfs_end_transaction(trans
, root
);
6970 free_extent_map(em
);
6971 return ERR_PTR(err
);
6973 BUG_ON(!em
); /* Error is always set */
6977 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
6978 size_t pg_offset
, u64 start
, u64 len
,
6981 struct extent_map
*em
;
6982 struct extent_map
*hole_em
= NULL
;
6983 u64 range_start
= start
;
6989 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
6996 * - a pre-alloc extent,
6997 * there might actually be delalloc bytes behind it.
6999 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
7000 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7006 /* check to see if we've wrapped (len == -1 or similar) */
7015 /* ok, we didn't find anything, lets look for delalloc */
7016 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
7017 end
, len
, EXTENT_DELALLOC
, 1);
7018 found_end
= range_start
+ found
;
7019 if (found_end
< range_start
)
7020 found_end
= (u64
)-1;
7023 * we didn't find anything useful, return
7024 * the original results from get_extent()
7026 if (range_start
> end
|| found_end
<= start
) {
7032 /* adjust the range_start to make sure it doesn't
7033 * go backwards from the start they passed in
7035 range_start
= max(start
, range_start
);
7036 found
= found_end
- range_start
;
7039 u64 hole_start
= start
;
7042 em
= alloc_extent_map();
7048 * when btrfs_get_extent can't find anything it
7049 * returns one huge hole
7051 * make sure what it found really fits our range, and
7052 * adjust to make sure it is based on the start from
7056 u64 calc_end
= extent_map_end(hole_em
);
7058 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7059 free_extent_map(hole_em
);
7062 hole_start
= max(hole_em
->start
, start
);
7063 hole_len
= calc_end
- hole_start
;
7067 if (hole_em
&& range_start
> hole_start
) {
7068 /* our hole starts before our delalloc, so we
7069 * have to return just the parts of the hole
7070 * that go until the delalloc starts
7072 em
->len
= min(hole_len
,
7073 range_start
- hole_start
);
7074 em
->start
= hole_start
;
7075 em
->orig_start
= hole_start
;
7077 * don't adjust block start at all,
7078 * it is fixed at EXTENT_MAP_HOLE
7080 em
->block_start
= hole_em
->block_start
;
7081 em
->block_len
= hole_len
;
7082 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7083 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7085 em
->start
= range_start
;
7087 em
->orig_start
= range_start
;
7088 em
->block_start
= EXTENT_MAP_DELALLOC
;
7089 em
->block_len
= found
;
7091 } else if (hole_em
) {
7096 free_extent_map(hole_em
);
7098 free_extent_map(em
);
7099 return ERR_PTR(err
);
7104 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7107 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7108 struct extent_map
*em
;
7109 struct btrfs_key ins
;
7113 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7114 ret
= btrfs_reserve_extent(root
, len
, root
->sectorsize
, 0,
7115 alloc_hint
, &ins
, 1, 1);
7117 return ERR_PTR(ret
);
7120 * Create the ordered extent before the extent map. This is to avoid
7121 * races with the fast fsync path that would lead to it logging file
7122 * extent items that point to disk extents that were not yet written to.
7123 * The fast fsync path collects ordered extents into a local list and
7124 * then collects all the new extent maps, so we must create the ordered
7125 * extent first and make sure the fast fsync path collects any new
7126 * ordered extents after collecting new extent maps as well.
7127 * The fsync path simply can not rely on inode_dio_wait() because it
7128 * causes deadlock with AIO.
7130 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
7131 ins
.offset
, ins
.offset
, 0);
7133 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7134 return ERR_PTR(ret
);
7137 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
7138 ins
.offset
, ins
.offset
, ins
.offset
, 0);
7140 struct btrfs_ordered_extent
*oe
;
7142 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7143 oe
= btrfs_lookup_ordered_extent(inode
, start
);
7147 set_bit(BTRFS_ORDERED_IOERR
, &oe
->flags
);
7148 set_bit(BTRFS_ORDERED_IO_DONE
, &oe
->flags
);
7149 btrfs_remove_ordered_extent(inode
, oe
);
7150 /* Once for our lookup and once for the ordered extents tree. */
7151 btrfs_put_ordered_extent(oe
);
7152 btrfs_put_ordered_extent(oe
);
7158 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7159 * block must be cow'd
7161 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7162 u64
*orig_start
, u64
*orig_block_len
,
7165 struct btrfs_trans_handle
*trans
;
7166 struct btrfs_path
*path
;
7168 struct extent_buffer
*leaf
;
7169 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7170 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7171 struct btrfs_file_extent_item
*fi
;
7172 struct btrfs_key key
;
7179 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7181 path
= btrfs_alloc_path();
7185 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
7190 slot
= path
->slots
[0];
7193 /* can't find the item, must cow */
7200 leaf
= path
->nodes
[0];
7201 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7202 if (key
.objectid
!= btrfs_ino(inode
) ||
7203 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7204 /* not our file or wrong item type, must cow */
7208 if (key
.offset
> offset
) {
7209 /* Wrong offset, must cow */
7213 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7214 found_type
= btrfs_file_extent_type(leaf
, fi
);
7215 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7216 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7217 /* not a regular extent, must cow */
7221 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7224 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7225 if (extent_end
<= offset
)
7228 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7229 if (disk_bytenr
== 0)
7232 if (btrfs_file_extent_compression(leaf
, fi
) ||
7233 btrfs_file_extent_encryption(leaf
, fi
) ||
7234 btrfs_file_extent_other_encoding(leaf
, fi
))
7237 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7240 *orig_start
= key
.offset
- backref_offset
;
7241 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7242 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7245 if (btrfs_extent_readonly(root
, disk_bytenr
))
7248 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7249 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7252 range_end
= round_up(offset
+ num_bytes
, root
->sectorsize
) - 1;
7253 ret
= test_range_bit(io_tree
, offset
, range_end
,
7254 EXTENT_DELALLOC
, 0, NULL
);
7261 btrfs_release_path(path
);
7264 * look for other files referencing this extent, if we
7265 * find any we must cow
7267 trans
= btrfs_join_transaction(root
);
7268 if (IS_ERR(trans
)) {
7273 ret
= btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
7274 key
.offset
- backref_offset
, disk_bytenr
);
7275 btrfs_end_transaction(trans
, root
);
7282 * adjust disk_bytenr and num_bytes to cover just the bytes
7283 * in this extent we are about to write. If there
7284 * are any csums in that range we have to cow in order
7285 * to keep the csums correct
7287 disk_bytenr
+= backref_offset
;
7288 disk_bytenr
+= offset
- key
.offset
;
7289 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
7292 * all of the above have passed, it is safe to overwrite this extent
7298 btrfs_free_path(path
);
7302 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7304 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7306 void **pagep
= NULL
;
7307 struct page
*page
= NULL
;
7311 start_idx
= start
>> PAGE_CACHE_SHIFT
;
7314 * end is the last byte in the last page. end == start is legal
7316 end_idx
= end
>> PAGE_CACHE_SHIFT
;
7320 /* Most of the code in this while loop is lifted from
7321 * find_get_page. It's been modified to begin searching from a
7322 * page and return just the first page found in that range. If the
7323 * found idx is less than or equal to the end idx then we know that
7324 * a page exists. If no pages are found or if those pages are
7325 * outside of the range then we're fine (yay!) */
7326 while (page
== NULL
&&
7327 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7328 page
= radix_tree_deref_slot(pagep
);
7329 if (unlikely(!page
))
7332 if (radix_tree_exception(page
)) {
7333 if (radix_tree_deref_retry(page
)) {
7338 * Otherwise, shmem/tmpfs must be storing a swap entry
7339 * here as an exceptional entry: so return it without
7340 * attempting to raise page count.
7343 break; /* TODO: Is this relevant for this use case? */
7346 if (!page_cache_get_speculative(page
)) {
7352 * Has the page moved?
7353 * This is part of the lockless pagecache protocol. See
7354 * include/linux/pagemap.h for details.
7356 if (unlikely(page
!= *pagep
)) {
7357 page_cache_release(page
);
7363 if (page
->index
<= end_idx
)
7365 page_cache_release(page
);
7372 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7373 struct extent_state
**cached_state
, int writing
)
7375 struct btrfs_ordered_extent
*ordered
;
7379 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7382 * We're concerned with the entire range that we're going to be
7383 * doing DIO to, so we need to make sure theres no ordered
7384 * extents in this range.
7386 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
7387 lockend
- lockstart
+ 1);
7390 * We need to make sure there are no buffered pages in this
7391 * range either, we could have raced between the invalidate in
7392 * generic_file_direct_write and locking the extent. The
7393 * invalidate needs to happen so that reads after a write do not
7398 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7401 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7402 cached_state
, GFP_NOFS
);
7405 btrfs_start_ordered_extent(inode
, ordered
, 1);
7406 btrfs_put_ordered_extent(ordered
);
7409 * We could trigger writeback for this range (and wait
7410 * for it to complete) and then invalidate the pages for
7411 * this range (through invalidate_inode_pages2_range()),
7412 * but that can lead us to a deadlock with a concurrent
7413 * call to readpages() (a buffered read or a defrag call
7414 * triggered a readahead) on a page lock due to an
7415 * ordered dio extent we created before but did not have
7416 * yet a corresponding bio submitted (whence it can not
7417 * complete), which makes readpages() wait for that
7418 * ordered extent to complete while holding a lock on
7431 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
7432 u64 len
, u64 orig_start
,
7433 u64 block_start
, u64 block_len
,
7434 u64 orig_block_len
, u64 ram_bytes
,
7437 struct extent_map_tree
*em_tree
;
7438 struct extent_map
*em
;
7439 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7442 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7443 em
= alloc_extent_map();
7445 return ERR_PTR(-ENOMEM
);
7448 em
->orig_start
= orig_start
;
7449 em
->mod_start
= start
;
7452 em
->block_len
= block_len
;
7453 em
->block_start
= block_start
;
7454 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7455 em
->orig_block_len
= orig_block_len
;
7456 em
->ram_bytes
= ram_bytes
;
7457 em
->generation
= -1;
7458 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7459 if (type
== BTRFS_ORDERED_PREALLOC
)
7460 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7463 btrfs_drop_extent_cache(inode
, em
->start
,
7464 em
->start
+ em
->len
- 1, 0);
7465 write_lock(&em_tree
->lock
);
7466 ret
= add_extent_mapping(em_tree
, em
, 1);
7467 write_unlock(&em_tree
->lock
);
7468 } while (ret
== -EEXIST
);
7471 free_extent_map(em
);
7472 return ERR_PTR(ret
);
7478 static void adjust_dio_outstanding_extents(struct inode
*inode
,
7479 struct btrfs_dio_data
*dio_data
,
7482 unsigned num_extents
;
7484 num_extents
= (unsigned) div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
- 1,
7485 BTRFS_MAX_EXTENT_SIZE
);
7487 * If we have an outstanding_extents count still set then we're
7488 * within our reservation, otherwise we need to adjust our inode
7489 * counter appropriately.
7491 if (dio_data
->outstanding_extents
) {
7492 dio_data
->outstanding_extents
-= num_extents
;
7494 spin_lock(&BTRFS_I(inode
)->lock
);
7495 BTRFS_I(inode
)->outstanding_extents
+= num_extents
;
7496 spin_unlock(&BTRFS_I(inode
)->lock
);
7500 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7501 struct buffer_head
*bh_result
, int create
)
7503 struct extent_map
*em
;
7504 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7505 struct extent_state
*cached_state
= NULL
;
7506 struct btrfs_dio_data
*dio_data
= NULL
;
7507 u64 start
= iblock
<< inode
->i_blkbits
;
7508 u64 lockstart
, lockend
;
7509 u64 len
= bh_result
->b_size
;
7510 int unlock_bits
= EXTENT_LOCKED
;
7514 unlock_bits
|= EXTENT_DIRTY
;
7516 len
= min_t(u64
, len
, root
->sectorsize
);
7519 lockend
= start
+ len
- 1;
7521 if (current
->journal_info
) {
7523 * Need to pull our outstanding extents and set journal_info to NULL so
7524 * that anything that needs to check if there's a transction doesn't get
7527 dio_data
= current
->journal_info
;
7528 current
->journal_info
= NULL
;
7532 * If this errors out it's because we couldn't invalidate pagecache for
7533 * this range and we need to fallback to buffered.
7535 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
,
7541 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7548 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7549 * io. INLINE is special, and we could probably kludge it in here, but
7550 * it's still buffered so for safety lets just fall back to the generic
7553 * For COMPRESSED we _have_ to read the entire extent in so we can
7554 * decompress it, so there will be buffering required no matter what we
7555 * do, so go ahead and fallback to buffered.
7557 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7558 * to buffered IO. Don't blame me, this is the price we pay for using
7561 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7562 em
->block_start
== EXTENT_MAP_INLINE
) {
7563 free_extent_map(em
);
7568 /* Just a good old fashioned hole, return */
7569 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7570 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7571 free_extent_map(em
);
7576 * We don't allocate a new extent in the following cases
7578 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7580 * 2) The extent is marked as PREALLOC. We're good to go here and can
7581 * just use the extent.
7585 len
= min(len
, em
->len
- (start
- em
->start
));
7586 lockstart
= start
+ len
;
7590 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7591 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7592 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7594 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7596 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7597 type
= BTRFS_ORDERED_PREALLOC
;
7599 type
= BTRFS_ORDERED_NOCOW
;
7600 len
= min(len
, em
->len
- (start
- em
->start
));
7601 block_start
= em
->block_start
+ (start
- em
->start
);
7603 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7604 &orig_block_len
, &ram_bytes
) == 1) {
7605 if (type
== BTRFS_ORDERED_PREALLOC
) {
7606 free_extent_map(em
);
7607 em
= create_pinned_em(inode
, start
, len
,
7618 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
7619 block_start
, len
, len
, type
);
7621 free_extent_map(em
);
7629 * this will cow the extent, reset the len in case we changed
7632 len
= bh_result
->b_size
;
7633 free_extent_map(em
);
7634 em
= btrfs_new_extent_direct(inode
, start
, len
);
7639 len
= min(len
, em
->len
- (start
- em
->start
));
7641 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7643 bh_result
->b_size
= len
;
7644 bh_result
->b_bdev
= em
->bdev
;
7645 set_buffer_mapped(bh_result
);
7647 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7648 set_buffer_new(bh_result
);
7651 * Need to update the i_size under the extent lock so buffered
7652 * readers will get the updated i_size when we unlock.
7654 if (start
+ len
> i_size_read(inode
))
7655 i_size_write(inode
, start
+ len
);
7657 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7658 btrfs_free_reserved_data_space(inode
, start
, len
);
7659 WARN_ON(dio_data
->reserve
< len
);
7660 dio_data
->reserve
-= len
;
7661 dio_data
->unsubmitted_oe_range_end
= start
+ len
;
7662 current
->journal_info
= dio_data
;
7666 * In the case of write we need to clear and unlock the entire range,
7667 * in the case of read we need to unlock only the end area that we
7668 * aren't using if there is any left over space.
7670 if (lockstart
< lockend
) {
7671 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7672 lockend
, unlock_bits
, 1, 0,
7673 &cached_state
, GFP_NOFS
);
7675 free_extent_state(cached_state
);
7678 free_extent_map(em
);
7683 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7684 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
7687 current
->journal_info
= dio_data
;
7689 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7690 * write less data then expected, so that we don't underflow our inode's
7691 * outstanding extents counter.
7693 if (create
&& dio_data
)
7694 adjust_dio_outstanding_extents(inode
, dio_data
, len
);
7699 static inline int submit_dio_repair_bio(struct inode
*inode
, struct bio
*bio
,
7700 int rw
, int mirror_num
)
7702 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7705 BUG_ON(rw
& REQ_WRITE
);
7709 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
7710 BTRFS_WQ_ENDIO_DIO_REPAIR
);
7714 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
7720 static int btrfs_check_dio_repairable(struct inode
*inode
,
7721 struct bio
*failed_bio
,
7722 struct io_failure_record
*failrec
,
7727 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
7728 failrec
->logical
, failrec
->len
);
7729 if (num_copies
== 1) {
7731 * we only have a single copy of the data, so don't bother with
7732 * all the retry and error correction code that follows. no
7733 * matter what the error is, it is very likely to persist.
7735 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7736 num_copies
, failrec
->this_mirror
, failed_mirror
);
7740 failrec
->failed_mirror
= failed_mirror
;
7741 failrec
->this_mirror
++;
7742 if (failrec
->this_mirror
== failed_mirror
)
7743 failrec
->this_mirror
++;
7745 if (failrec
->this_mirror
> num_copies
) {
7746 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7747 num_copies
, failrec
->this_mirror
, failed_mirror
);
7754 static int dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7755 struct page
*page
, u64 start
, u64 end
,
7756 int failed_mirror
, bio_end_io_t
*repair_endio
,
7759 struct io_failure_record
*failrec
;
7765 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
7767 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7771 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7774 free_io_failure(inode
, failrec
);
7778 if (failed_bio
->bi_vcnt
> 1)
7779 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
7781 read_mode
= READ_SYNC
;
7783 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7784 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7785 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7786 0, isector
, repair_endio
, repair_arg
);
7788 free_io_failure(inode
, failrec
);
7792 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7793 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7794 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7796 ret
= submit_dio_repair_bio(inode
, bio
, read_mode
,
7797 failrec
->this_mirror
);
7799 free_io_failure(inode
, failrec
);
7806 struct btrfs_retry_complete
{
7807 struct completion done
;
7808 struct inode
*inode
;
7813 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7815 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7816 struct bio_vec
*bvec
;
7823 bio_for_each_segment_all(bvec
, bio
, i
)
7824 clean_io_failure(done
->inode
, done
->start
, bvec
->bv_page
, 0);
7826 complete(&done
->done
);
7830 static int __btrfs_correct_data_nocsum(struct inode
*inode
,
7831 struct btrfs_io_bio
*io_bio
)
7833 struct bio_vec
*bvec
;
7834 struct btrfs_retry_complete done
;
7839 start
= io_bio
->logical
;
7842 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7846 init_completion(&done
.done
);
7848 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7849 start
+ bvec
->bv_len
- 1,
7851 btrfs_retry_endio_nocsum
, &done
);
7855 wait_for_completion(&done
.done
);
7857 if (!done
.uptodate
) {
7858 /* We might have another mirror, so try again */
7862 start
+= bvec
->bv_len
;
7868 static void btrfs_retry_endio(struct bio
*bio
)
7870 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7871 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7872 struct bio_vec
*bvec
;
7881 bio_for_each_segment_all(bvec
, bio
, i
) {
7882 ret
= __readpage_endio_check(done
->inode
, io_bio
, i
,
7884 done
->start
, bvec
->bv_len
);
7886 clean_io_failure(done
->inode
, done
->start
,
7892 done
->uptodate
= uptodate
;
7894 complete(&done
->done
);
7898 static int __btrfs_subio_endio_read(struct inode
*inode
,
7899 struct btrfs_io_bio
*io_bio
, int err
)
7901 struct bio_vec
*bvec
;
7902 struct btrfs_retry_complete done
;
7909 start
= io_bio
->logical
;
7912 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7913 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
7914 0, start
, bvec
->bv_len
);
7920 init_completion(&done
.done
);
7922 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7923 start
+ bvec
->bv_len
- 1,
7925 btrfs_retry_endio
, &done
);
7931 wait_for_completion(&done
.done
);
7933 if (!done
.uptodate
) {
7934 /* We might have another mirror, so try again */
7938 offset
+= bvec
->bv_len
;
7939 start
+= bvec
->bv_len
;
7945 static int btrfs_subio_endio_read(struct inode
*inode
,
7946 struct btrfs_io_bio
*io_bio
, int err
)
7948 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
7952 return __btrfs_correct_data_nocsum(inode
, io_bio
);
7956 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
7960 static void btrfs_endio_direct_read(struct bio
*bio
)
7962 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7963 struct inode
*inode
= dip
->inode
;
7964 struct bio
*dio_bio
;
7965 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7966 int err
= bio
->bi_error
;
7968 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
7969 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
7971 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
7972 dip
->logical_offset
+ dip
->bytes
- 1);
7973 dio_bio
= dip
->dio_bio
;
7977 dio_end_io(dio_bio
, bio
->bi_error
);
7980 io_bio
->end_io(io_bio
, err
);
7984 static void btrfs_endio_direct_write_update_ordered(struct inode
*inode
,
7989 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7990 struct btrfs_ordered_extent
*ordered
= NULL
;
7991 u64 ordered_offset
= offset
;
7992 u64 ordered_bytes
= bytes
;
7996 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
8003 btrfs_init_work(&ordered
->work
, btrfs_endio_write_helper
,
8004 finish_ordered_fn
, NULL
, NULL
);
8005 btrfs_queue_work(root
->fs_info
->endio_write_workers
,
8009 * our bio might span multiple ordered extents. If we haven't
8010 * completed the accounting for the whole dio, go back and try again
8012 if (ordered_offset
< offset
+ bytes
) {
8013 ordered_bytes
= offset
+ bytes
- ordered_offset
;
8019 static void btrfs_endio_direct_write(struct bio
*bio
)
8021 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8022 struct bio
*dio_bio
= dip
->dio_bio
;
8024 btrfs_endio_direct_write_update_ordered(dip
->inode
,
8025 dip
->logical_offset
,
8031 dio_end_io(dio_bio
, bio
->bi_error
);
8035 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
8036 struct bio
*bio
, int mirror_num
,
8037 unsigned long bio_flags
, u64 offset
)
8040 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8041 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
8042 BUG_ON(ret
); /* -ENOMEM */
8046 static void btrfs_end_dio_bio(struct bio
*bio
)
8048 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8049 int err
= bio
->bi_error
;
8052 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
8053 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8054 btrfs_ino(dip
->inode
), bio
->bi_rw
,
8055 (unsigned long long)bio
->bi_iter
.bi_sector
,
8056 bio
->bi_iter
.bi_size
, err
);
8058 if (dip
->subio_endio
)
8059 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
8065 * before atomic variable goto zero, we must make sure
8066 * dip->errors is perceived to be set.
8068 smp_mb__before_atomic();
8071 /* if there are more bios still pending for this dio, just exit */
8072 if (!atomic_dec_and_test(&dip
->pending_bios
))
8076 bio_io_error(dip
->orig_bio
);
8078 dip
->dio_bio
->bi_error
= 0;
8079 bio_endio(dip
->orig_bio
);
8085 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
8086 u64 first_sector
, gfp_t gfp_flags
)
8089 bio
= btrfs_bio_alloc(bdev
, first_sector
, BIO_MAX_PAGES
, gfp_flags
);
8091 bio_associate_current(bio
);
8095 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root
*root
,
8096 struct inode
*inode
,
8097 struct btrfs_dio_private
*dip
,
8101 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8102 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8106 * We load all the csum data we need when we submit
8107 * the first bio to reduce the csum tree search and
8110 if (dip
->logical_offset
== file_offset
) {
8111 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, dip
->orig_bio
,
8117 if (bio
== dip
->orig_bio
)
8120 file_offset
-= dip
->logical_offset
;
8121 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8122 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8127 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
8128 int rw
, u64 file_offset
, int skip_sum
,
8131 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8132 int write
= rw
& REQ_WRITE
;
8133 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8137 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8142 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
8143 BTRFS_WQ_ENDIO_DATA
);
8151 if (write
&& async_submit
) {
8152 ret
= btrfs_wq_submit_bio(root
->fs_info
,
8153 inode
, rw
, bio
, 0, 0,
8155 __btrfs_submit_bio_start_direct_io
,
8156 __btrfs_submit_bio_done
);
8160 * If we aren't doing async submit, calculate the csum of the
8163 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
8167 ret
= btrfs_lookup_and_bind_dio_csum(root
, inode
, dip
, bio
,
8173 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
8179 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
8182 struct inode
*inode
= dip
->inode
;
8183 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8185 struct bio
*orig_bio
= dip
->orig_bio
;
8186 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
8187 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8188 u64 file_offset
= dip
->logical_offset
;
8193 int async_submit
= 0;
8195 map_length
= orig_bio
->bi_iter
.bi_size
;
8196 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
8197 &map_length
, NULL
, 0);
8201 if (map_length
>= orig_bio
->bi_iter
.bi_size
) {
8203 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8207 /* async crcs make it difficult to collect full stripe writes. */
8208 if (btrfs_get_alloc_profile(root
, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8213 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
8217 bio
->bi_private
= dip
;
8218 bio
->bi_end_io
= btrfs_end_dio_bio
;
8219 btrfs_io_bio(bio
)->logical
= file_offset
;
8220 atomic_inc(&dip
->pending_bios
);
8222 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
8223 if (map_length
< submit_len
+ bvec
->bv_len
||
8224 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
8225 bvec
->bv_offset
) < bvec
->bv_len
) {
8227 * inc the count before we submit the bio so
8228 * we know the end IO handler won't happen before
8229 * we inc the count. Otherwise, the dip might get freed
8230 * before we're done setting it up
8232 atomic_inc(&dip
->pending_bios
);
8233 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
8234 file_offset
, skip_sum
,
8238 atomic_dec(&dip
->pending_bios
);
8242 start_sector
+= submit_len
>> 9;
8243 file_offset
+= submit_len
;
8248 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
8249 start_sector
, GFP_NOFS
);
8252 bio
->bi_private
= dip
;
8253 bio
->bi_end_io
= btrfs_end_dio_bio
;
8254 btrfs_io_bio(bio
)->logical
= file_offset
;
8256 map_length
= orig_bio
->bi_iter
.bi_size
;
8257 ret
= btrfs_map_block(root
->fs_info
, rw
,
8259 &map_length
, NULL
, 0);
8265 submit_len
+= bvec
->bv_len
;
8272 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
8281 * before atomic variable goto zero, we must
8282 * make sure dip->errors is perceived to be set.
8284 smp_mb__before_atomic();
8285 if (atomic_dec_and_test(&dip
->pending_bios
))
8286 bio_io_error(dip
->orig_bio
);
8288 /* bio_end_io() will handle error, so we needn't return it */
8292 static void btrfs_submit_direct(int rw
, struct bio
*dio_bio
,
8293 struct inode
*inode
, loff_t file_offset
)
8295 struct btrfs_dio_private
*dip
= NULL
;
8296 struct bio
*io_bio
= NULL
;
8297 struct btrfs_io_bio
*btrfs_bio
;
8299 int write
= rw
& REQ_WRITE
;
8302 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8304 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
8310 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8316 dip
->private = dio_bio
->bi_private
;
8318 dip
->logical_offset
= file_offset
;
8319 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8320 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8321 io_bio
->bi_private
= dip
;
8322 dip
->orig_bio
= io_bio
;
8323 dip
->dio_bio
= dio_bio
;
8324 atomic_set(&dip
->pending_bios
, 0);
8325 btrfs_bio
= btrfs_io_bio(io_bio
);
8326 btrfs_bio
->logical
= file_offset
;
8329 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
8331 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
8332 dip
->subio_endio
= btrfs_subio_endio_read
;
8336 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8337 * even if we fail to submit a bio, because in such case we do the
8338 * corresponding error handling below and it must not be done a second
8339 * time by btrfs_direct_IO().
8342 struct btrfs_dio_data
*dio_data
= current
->journal_info
;
8344 dio_data
->unsubmitted_oe_range_end
= dip
->logical_offset
+
8346 dio_data
->unsubmitted_oe_range_start
=
8347 dio_data
->unsubmitted_oe_range_end
;
8350 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
8354 if (btrfs_bio
->end_io
)
8355 btrfs_bio
->end_io(btrfs_bio
, ret
);
8359 * If we arrived here it means either we failed to submit the dip
8360 * or we either failed to clone the dio_bio or failed to allocate the
8361 * dip. If we cloned the dio_bio and allocated the dip, we can just
8362 * call bio_endio against our io_bio so that we get proper resource
8363 * cleanup if we fail to submit the dip, otherwise, we must do the
8364 * same as btrfs_endio_direct_[write|read] because we can't call these
8365 * callbacks - they require an allocated dip and a clone of dio_bio.
8367 if (io_bio
&& dip
) {
8368 io_bio
->bi_error
= -EIO
;
8371 * The end io callbacks free our dip, do the final put on io_bio
8372 * and all the cleanup and final put for dio_bio (through
8379 btrfs_endio_direct_write_update_ordered(inode
,
8381 dio_bio
->bi_iter
.bi_size
,
8384 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8385 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8387 dio_bio
->bi_error
= -EIO
;
8389 * Releases and cleans up our dio_bio, no need to bio_put()
8390 * nor bio_endio()/bio_io_error() against dio_bio.
8392 dio_end_io(dio_bio
, ret
);
8399 static ssize_t
check_direct_IO(struct btrfs_root
*root
, struct kiocb
*iocb
,
8400 const struct iov_iter
*iter
, loff_t offset
)
8404 unsigned blocksize_mask
= root
->sectorsize
- 1;
8405 ssize_t retval
= -EINVAL
;
8407 if (offset
& blocksize_mask
)
8410 if (iov_iter_alignment(iter
) & blocksize_mask
)
8413 /* If this is a write we don't need to check anymore */
8414 if (iov_iter_rw(iter
) == WRITE
)
8417 * Check to make sure we don't have duplicate iov_base's in this
8418 * iovec, if so return EINVAL, otherwise we'll get csum errors
8419 * when reading back.
8421 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8422 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8423 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8432 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
8435 struct file
*file
= iocb
->ki_filp
;
8436 struct inode
*inode
= file
->f_mapping
->host
;
8437 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8438 struct btrfs_dio_data dio_data
= { 0 };
8442 bool relock
= false;
8445 if (check_direct_IO(BTRFS_I(inode
)->root
, iocb
, iter
, offset
))
8448 inode_dio_begin(inode
);
8449 smp_mb__after_atomic();
8452 * The generic stuff only does filemap_write_and_wait_range, which
8453 * isn't enough if we've written compressed pages to this area, so
8454 * we need to flush the dirty pages again to make absolutely sure
8455 * that any outstanding dirty pages are on disk.
8457 count
= iov_iter_count(iter
);
8458 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8459 &BTRFS_I(inode
)->runtime_flags
))
8460 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8461 offset
+ count
- 1);
8463 if (iov_iter_rw(iter
) == WRITE
) {
8465 * If the write DIO is beyond the EOF, we need update
8466 * the isize, but it is protected by i_mutex. So we can
8467 * not unlock the i_mutex at this case.
8469 if (offset
+ count
<= inode
->i_size
) {
8470 inode_unlock(inode
);
8473 ret
= btrfs_delalloc_reserve_space(inode
, offset
, count
);
8476 dio_data
.outstanding_extents
= div64_u64(count
+
8477 BTRFS_MAX_EXTENT_SIZE
- 1,
8478 BTRFS_MAX_EXTENT_SIZE
);
8481 * We need to know how many extents we reserved so that we can
8482 * do the accounting properly if we go over the number we
8483 * originally calculated. Abuse current->journal_info for this.
8485 dio_data
.reserve
= round_up(count
, root
->sectorsize
);
8486 dio_data
.unsubmitted_oe_range_start
= (u64
)offset
;
8487 dio_data
.unsubmitted_oe_range_end
= (u64
)offset
;
8488 current
->journal_info
= &dio_data
;
8489 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8490 &BTRFS_I(inode
)->runtime_flags
)) {
8491 inode_dio_end(inode
);
8492 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8496 ret
= __blockdev_direct_IO(iocb
, inode
,
8497 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
8498 iter
, offset
, btrfs_get_blocks_direct
, NULL
,
8499 btrfs_submit_direct
, flags
);
8500 if (iov_iter_rw(iter
) == WRITE
) {
8501 current
->journal_info
= NULL
;
8502 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8503 if (dio_data
.reserve
)
8504 btrfs_delalloc_release_space(inode
, offset
,
8507 * On error we might have left some ordered extents
8508 * without submitting corresponding bios for them, so
8509 * cleanup them up to avoid other tasks getting them
8510 * and waiting for them to complete forever.
8512 if (dio_data
.unsubmitted_oe_range_start
<
8513 dio_data
.unsubmitted_oe_range_end
)
8514 btrfs_endio_direct_write_update_ordered(inode
,
8515 dio_data
.unsubmitted_oe_range_start
,
8516 dio_data
.unsubmitted_oe_range_end
-
8517 dio_data
.unsubmitted_oe_range_start
,
8519 } else if (ret
>= 0 && (size_t)ret
< count
)
8520 btrfs_delalloc_release_space(inode
, offset
,
8521 count
- (size_t)ret
);
8525 inode_dio_end(inode
);
8532 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8534 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8535 __u64 start
, __u64 len
)
8539 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8543 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
8546 int btrfs_readpage(struct file
*file
, struct page
*page
)
8548 struct extent_io_tree
*tree
;
8549 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8550 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8553 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8555 struct extent_io_tree
*tree
;
8556 struct inode
*inode
= page
->mapping
->host
;
8559 if (current
->flags
& PF_MEMALLOC
) {
8560 redirty_page_for_writepage(wbc
, page
);
8566 * If we are under memory pressure we will call this directly from the
8567 * VM, we need to make sure we have the inode referenced for the ordered
8568 * extent. If not just return like we didn't do anything.
8570 if (!igrab(inode
)) {
8571 redirty_page_for_writepage(wbc
, page
);
8572 return AOP_WRITEPAGE_ACTIVATE
;
8574 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8575 ret
= extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
8576 btrfs_add_delayed_iput(inode
);
8580 static int btrfs_writepages(struct address_space
*mapping
,
8581 struct writeback_control
*wbc
)
8583 struct extent_io_tree
*tree
;
8585 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8586 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
8590 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8591 struct list_head
*pages
, unsigned nr_pages
)
8593 struct extent_io_tree
*tree
;
8594 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8595 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
8598 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8600 struct extent_io_tree
*tree
;
8601 struct extent_map_tree
*map
;
8604 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8605 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
8606 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
8608 ClearPagePrivate(page
);
8609 set_page_private(page
, 0);
8610 page_cache_release(page
);
8615 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8617 if (PageWriteback(page
) || PageDirty(page
))
8619 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
8622 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8623 unsigned int length
)
8625 struct inode
*inode
= page
->mapping
->host
;
8626 struct extent_io_tree
*tree
;
8627 struct btrfs_ordered_extent
*ordered
;
8628 struct extent_state
*cached_state
= NULL
;
8629 u64 page_start
= page_offset(page
);
8630 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8631 int inode_evicting
= inode
->i_state
& I_FREEING
;
8634 * we have the page locked, so new writeback can't start,
8635 * and the dirty bit won't be cleared while we are here.
8637 * Wait for IO on this page so that we can safely clear
8638 * the PagePrivate2 bit and do ordered accounting
8640 wait_on_page_writeback(page
);
8642 tree
= &BTRFS_I(inode
)->io_tree
;
8644 btrfs_releasepage(page
, GFP_NOFS
);
8648 if (!inode_evicting
)
8649 lock_extent_bits(tree
, page_start
, page_end
, &cached_state
);
8650 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8653 * IO on this page will never be started, so we need
8654 * to account for any ordered extents now
8656 if (!inode_evicting
)
8657 clear_extent_bit(tree
, page_start
, page_end
,
8658 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8659 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8660 EXTENT_DEFRAG
, 1, 0, &cached_state
,
8663 * whoever cleared the private bit is responsible
8664 * for the finish_ordered_io
8666 if (TestClearPagePrivate2(page
)) {
8667 struct btrfs_ordered_inode_tree
*tree
;
8670 tree
= &BTRFS_I(inode
)->ordered_tree
;
8672 spin_lock_irq(&tree
->lock
);
8673 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8674 new_len
= page_start
- ordered
->file_offset
;
8675 if (new_len
< ordered
->truncated_len
)
8676 ordered
->truncated_len
= new_len
;
8677 spin_unlock_irq(&tree
->lock
);
8679 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8681 PAGE_CACHE_SIZE
, 1))
8682 btrfs_finish_ordered_io(ordered
);
8684 btrfs_put_ordered_extent(ordered
);
8685 if (!inode_evicting
) {
8686 cached_state
= NULL
;
8687 lock_extent_bits(tree
, page_start
, page_end
,
8693 * Qgroup reserved space handler
8694 * Page here will be either
8695 * 1) Already written to disk
8696 * In this case, its reserved space is released from data rsv map
8697 * and will be freed by delayed_ref handler finally.
8698 * So even we call qgroup_free_data(), it won't decrease reserved
8700 * 2) Not written to disk
8701 * This means the reserved space should be freed here.
8703 btrfs_qgroup_free_data(inode
, page_start
, PAGE_CACHE_SIZE
);
8704 if (!inode_evicting
) {
8705 clear_extent_bit(tree
, page_start
, page_end
,
8706 EXTENT_LOCKED
| EXTENT_DIRTY
|
8707 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
8708 EXTENT_DEFRAG
, 1, 1,
8709 &cached_state
, GFP_NOFS
);
8711 __btrfs_releasepage(page
, GFP_NOFS
);
8714 ClearPageChecked(page
);
8715 if (PagePrivate(page
)) {
8716 ClearPagePrivate(page
);
8717 set_page_private(page
, 0);
8718 page_cache_release(page
);
8723 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8724 * called from a page fault handler when a page is first dirtied. Hence we must
8725 * be careful to check for EOF conditions here. We set the page up correctly
8726 * for a written page which means we get ENOSPC checking when writing into
8727 * holes and correct delalloc and unwritten extent mapping on filesystems that
8728 * support these features.
8730 * We are not allowed to take the i_mutex here so we have to play games to
8731 * protect against truncate races as the page could now be beyond EOF. Because
8732 * vmtruncate() writes the inode size before removing pages, once we have the
8733 * page lock we can determine safely if the page is beyond EOF. If it is not
8734 * beyond EOF, then the page is guaranteed safe against truncation until we
8737 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
8739 struct page
*page
= vmf
->page
;
8740 struct inode
*inode
= file_inode(vma
->vm_file
);
8741 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8742 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8743 struct btrfs_ordered_extent
*ordered
;
8744 struct extent_state
*cached_state
= NULL
;
8746 unsigned long zero_start
;
8753 sb_start_pagefault(inode
->i_sb
);
8754 page_start
= page_offset(page
);
8755 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8757 ret
= btrfs_delalloc_reserve_space(inode
, page_start
,
8760 ret
= file_update_time(vma
->vm_file
);
8766 else /* -ENOSPC, -EIO, etc */
8767 ret
= VM_FAULT_SIGBUS
;
8773 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8776 size
= i_size_read(inode
);
8778 if ((page
->mapping
!= inode
->i_mapping
) ||
8779 (page_start
>= size
)) {
8780 /* page got truncated out from underneath us */
8783 wait_on_page_writeback(page
);
8785 lock_extent_bits(io_tree
, page_start
, page_end
, &cached_state
);
8786 set_page_extent_mapped(page
);
8789 * we can't set the delalloc bits if there are pending ordered
8790 * extents. Drop our locks and wait for them to finish
8792 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8794 unlock_extent_cached(io_tree
, page_start
, page_end
,
8795 &cached_state
, GFP_NOFS
);
8797 btrfs_start_ordered_extent(inode
, ordered
, 1);
8798 btrfs_put_ordered_extent(ordered
);
8803 * XXX - page_mkwrite gets called every time the page is dirtied, even
8804 * if it was already dirty, so for space accounting reasons we need to
8805 * clear any delalloc bits for the range we are fixing to save. There
8806 * is probably a better way to do this, but for now keep consistent with
8807 * prepare_pages in the normal write path.
8809 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
8810 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8811 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
8812 0, 0, &cached_state
, GFP_NOFS
);
8814 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
8817 unlock_extent_cached(io_tree
, page_start
, page_end
,
8818 &cached_state
, GFP_NOFS
);
8819 ret
= VM_FAULT_SIGBUS
;
8824 /* page is wholly or partially inside EOF */
8825 if (page_start
+ PAGE_CACHE_SIZE
> size
)
8826 zero_start
= size
& ~PAGE_CACHE_MASK
;
8828 zero_start
= PAGE_CACHE_SIZE
;
8830 if (zero_start
!= PAGE_CACHE_SIZE
) {
8832 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
8833 flush_dcache_page(page
);
8836 ClearPageChecked(page
);
8837 set_page_dirty(page
);
8838 SetPageUptodate(page
);
8840 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
8841 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
8842 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
8844 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
8848 sb_end_pagefault(inode
->i_sb
);
8849 return VM_FAULT_LOCKED
;
8853 btrfs_delalloc_release_space(inode
, page_start
, PAGE_CACHE_SIZE
);
8855 sb_end_pagefault(inode
->i_sb
);
8859 static int btrfs_truncate(struct inode
*inode
)
8861 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8862 struct btrfs_block_rsv
*rsv
;
8865 struct btrfs_trans_handle
*trans
;
8866 u64 mask
= root
->sectorsize
- 1;
8867 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
8869 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
8875 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8876 * 3 things going on here
8878 * 1) We need to reserve space for our orphan item and the space to
8879 * delete our orphan item. Lord knows we don't want to have a dangling
8880 * orphan item because we didn't reserve space to remove it.
8882 * 2) We need to reserve space to update our inode.
8884 * 3) We need to have something to cache all the space that is going to
8885 * be free'd up by the truncate operation, but also have some slack
8886 * space reserved in case it uses space during the truncate (thank you
8887 * very much snapshotting).
8889 * And we need these to all be seperate. The fact is we can use alot of
8890 * space doing the truncate, and we have no earthly idea how much space
8891 * we will use, so we need the truncate reservation to be seperate so it
8892 * doesn't end up using space reserved for updating the inode or
8893 * removing the orphan item. We also need to be able to stop the
8894 * transaction and start a new one, which means we need to be able to
8895 * update the inode several times, and we have no idea of knowing how
8896 * many times that will be, so we can't just reserve 1 item for the
8897 * entirety of the opration, so that has to be done seperately as well.
8898 * Then there is the orphan item, which does indeed need to be held on
8899 * to for the whole operation, and we need nobody to touch this reserved
8900 * space except the orphan code.
8902 * So that leaves us with
8904 * 1) root->orphan_block_rsv - for the orphan deletion.
8905 * 2) rsv - for the truncate reservation, which we will steal from the
8906 * transaction reservation.
8907 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8908 * updating the inode.
8910 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
8913 rsv
->size
= min_size
;
8917 * 1 for the truncate slack space
8918 * 1 for updating the inode.
8920 trans
= btrfs_start_transaction(root
, 2);
8921 if (IS_ERR(trans
)) {
8922 err
= PTR_ERR(trans
);
8926 /* Migrate the slack space for the truncate to our reserve */
8927 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
8932 * So if we truncate and then write and fsync we normally would just
8933 * write the extents that changed, which is a problem if we need to
8934 * first truncate that entire inode. So set this flag so we write out
8935 * all of the extents in the inode to the sync log so we're completely
8938 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
8939 trans
->block_rsv
= rsv
;
8942 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
8944 BTRFS_EXTENT_DATA_KEY
);
8945 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
8950 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
8951 ret
= btrfs_update_inode(trans
, root
, inode
);
8957 btrfs_end_transaction(trans
, root
);
8958 btrfs_btree_balance_dirty(root
);
8960 trans
= btrfs_start_transaction(root
, 2);
8961 if (IS_ERR(trans
)) {
8962 ret
= err
= PTR_ERR(trans
);
8967 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
8969 BUG_ON(ret
); /* shouldn't happen */
8970 trans
->block_rsv
= rsv
;
8973 if (ret
== 0 && inode
->i_nlink
> 0) {
8974 trans
->block_rsv
= root
->orphan_block_rsv
;
8975 ret
= btrfs_orphan_del(trans
, inode
);
8981 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
8982 ret
= btrfs_update_inode(trans
, root
, inode
);
8986 ret
= btrfs_end_transaction(trans
, root
);
8987 btrfs_btree_balance_dirty(root
);
8991 btrfs_free_block_rsv(root
, rsv
);
9000 * create a new subvolume directory/inode (helper for the ioctl).
9002 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
9003 struct btrfs_root
*new_root
,
9004 struct btrfs_root
*parent_root
,
9007 struct inode
*inode
;
9011 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
9012 new_dirid
, new_dirid
,
9013 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
9016 return PTR_ERR(inode
);
9017 inode
->i_op
= &btrfs_dir_inode_operations
;
9018 inode
->i_fop
= &btrfs_dir_file_operations
;
9020 set_nlink(inode
, 1);
9021 btrfs_i_size_write(inode
, 0);
9022 unlock_new_inode(inode
);
9024 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
9026 btrfs_err(new_root
->fs_info
,
9027 "error inheriting subvolume %llu properties: %d",
9028 new_root
->root_key
.objectid
, err
);
9030 err
= btrfs_update_inode(trans
, new_root
, inode
);
9036 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
9038 struct btrfs_inode
*ei
;
9039 struct inode
*inode
;
9041 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
9048 ei
->last_sub_trans
= 0;
9049 ei
->logged_trans
= 0;
9050 ei
->delalloc_bytes
= 0;
9051 ei
->defrag_bytes
= 0;
9052 ei
->disk_i_size
= 0;
9055 ei
->index_cnt
= (u64
)-1;
9057 ei
->last_unlink_trans
= 0;
9058 ei
->last_log_commit
= 0;
9059 ei
->delayed_iput_count
= 0;
9061 spin_lock_init(&ei
->lock
);
9062 ei
->outstanding_extents
= 0;
9063 ei
->reserved_extents
= 0;
9065 ei
->runtime_flags
= 0;
9066 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
9068 ei
->delayed_node
= NULL
;
9070 ei
->i_otime
.tv_sec
= 0;
9071 ei
->i_otime
.tv_nsec
= 0;
9073 inode
= &ei
->vfs_inode
;
9074 extent_map_tree_init(&ei
->extent_tree
);
9075 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
9076 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
9077 ei
->io_tree
.track_uptodate
= 1;
9078 ei
->io_failure_tree
.track_uptodate
= 1;
9079 atomic_set(&ei
->sync_writers
, 0);
9080 mutex_init(&ei
->log_mutex
);
9081 mutex_init(&ei
->delalloc_mutex
);
9082 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
9083 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
9084 INIT_LIST_HEAD(&ei
->delayed_iput
);
9085 RB_CLEAR_NODE(&ei
->rb_node
);
9090 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9091 void btrfs_test_destroy_inode(struct inode
*inode
)
9093 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9094 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9098 static void btrfs_i_callback(struct rcu_head
*head
)
9100 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
9101 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
9104 void btrfs_destroy_inode(struct inode
*inode
)
9106 struct btrfs_ordered_extent
*ordered
;
9107 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9109 WARN_ON(!hlist_empty(&inode
->i_dentry
));
9110 WARN_ON(inode
->i_data
.nrpages
);
9111 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
9112 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
9113 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
9114 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
9115 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
9118 * This can happen where we create an inode, but somebody else also
9119 * created the same inode and we need to destroy the one we already
9125 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
9126 &BTRFS_I(inode
)->runtime_flags
)) {
9127 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
9129 atomic_dec(&root
->orphan_inodes
);
9133 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
9137 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
9138 ordered
->file_offset
, ordered
->len
);
9139 btrfs_remove_ordered_extent(inode
, ordered
);
9140 btrfs_put_ordered_extent(ordered
);
9141 btrfs_put_ordered_extent(ordered
);
9144 btrfs_qgroup_check_reserved_leak(inode
);
9145 inode_tree_del(inode
);
9146 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9148 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9151 int btrfs_drop_inode(struct inode
*inode
)
9153 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9158 /* the snap/subvol tree is on deleting */
9159 if (btrfs_root_refs(&root
->root_item
) == 0)
9162 return generic_drop_inode(inode
);
9165 static void init_once(void *foo
)
9167 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9169 inode_init_once(&ei
->vfs_inode
);
9172 void btrfs_destroy_cachep(void)
9175 * Make sure all delayed rcu free inodes are flushed before we
9179 if (btrfs_inode_cachep
)
9180 kmem_cache_destroy(btrfs_inode_cachep
);
9181 if (btrfs_trans_handle_cachep
)
9182 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9183 if (btrfs_transaction_cachep
)
9184 kmem_cache_destroy(btrfs_transaction_cachep
);
9185 if (btrfs_path_cachep
)
9186 kmem_cache_destroy(btrfs_path_cachep
);
9187 if (btrfs_free_space_cachep
)
9188 kmem_cache_destroy(btrfs_free_space_cachep
);
9191 int btrfs_init_cachep(void)
9193 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9194 sizeof(struct btrfs_inode
), 0,
9195 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
| SLAB_ACCOUNT
,
9197 if (!btrfs_inode_cachep
)
9200 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9201 sizeof(struct btrfs_trans_handle
), 0,
9202 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9203 if (!btrfs_trans_handle_cachep
)
9206 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
9207 sizeof(struct btrfs_transaction
), 0,
9208 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9209 if (!btrfs_transaction_cachep
)
9212 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9213 sizeof(struct btrfs_path
), 0,
9214 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9215 if (!btrfs_path_cachep
)
9218 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9219 sizeof(struct btrfs_free_space
), 0,
9220 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9221 if (!btrfs_free_space_cachep
)
9226 btrfs_destroy_cachep();
9230 static int btrfs_getattr(struct vfsmount
*mnt
,
9231 struct dentry
*dentry
, struct kstat
*stat
)
9234 struct inode
*inode
= d_inode(dentry
);
9235 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9237 generic_fillattr(inode
, stat
);
9238 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9239 stat
->blksize
= PAGE_CACHE_SIZE
;
9241 spin_lock(&BTRFS_I(inode
)->lock
);
9242 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
9243 spin_unlock(&BTRFS_I(inode
)->lock
);
9244 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9245 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9249 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9250 struct inode
*new_dir
, struct dentry
*new_dentry
)
9252 struct btrfs_trans_handle
*trans
;
9253 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9254 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9255 struct inode
*new_inode
= d_inode(new_dentry
);
9256 struct inode
*old_inode
= d_inode(old_dentry
);
9257 struct timespec ctime
= CURRENT_TIME
;
9261 u64 old_ino
= btrfs_ino(old_inode
);
9263 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9266 /* we only allow rename subvolume link between subvolumes */
9267 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9270 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9271 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
9274 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9275 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9279 /* check for collisions, even if the name isn't there */
9280 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9281 new_dentry
->d_name
.name
,
9282 new_dentry
->d_name
.len
);
9285 if (ret
== -EEXIST
) {
9287 * eexist without a new_inode */
9288 if (WARN_ON(!new_inode
)) {
9292 /* maybe -EOVERFLOW */
9299 * we're using rename to replace one file with another. Start IO on it
9300 * now so we don't add too much work to the end of the transaction
9302 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9303 filemap_flush(old_inode
->i_mapping
);
9305 /* close the racy window with snapshot create/destroy ioctl */
9306 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9307 down_read(&root
->fs_info
->subvol_sem
);
9309 * We want to reserve the absolute worst case amount of items. So if
9310 * both inodes are subvols and we need to unlink them then that would
9311 * require 4 item modifications, but if they are both normal inodes it
9312 * would require 5 item modifications, so we'll assume their normal
9313 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9314 * should cover the worst case number of items we'll modify.
9316 trans
= btrfs_start_transaction(root
, 11);
9317 if (IS_ERR(trans
)) {
9318 ret
= PTR_ERR(trans
);
9323 btrfs_record_root_in_trans(trans
, dest
);
9325 ret
= btrfs_set_inode_index(new_dir
, &index
);
9329 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9330 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9331 /* force full log commit if subvolume involved. */
9332 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9334 ret
= btrfs_insert_inode_ref(trans
, dest
,
9335 new_dentry
->d_name
.name
,
9336 new_dentry
->d_name
.len
,
9338 btrfs_ino(new_dir
), index
);
9342 * this is an ugly little race, but the rename is required
9343 * to make sure that if we crash, the inode is either at the
9344 * old name or the new one. pinning the log transaction lets
9345 * us make sure we don't allow a log commit to come in after
9346 * we unlink the name but before we add the new name back in.
9348 btrfs_pin_log_trans(root
);
9351 inode_inc_iversion(old_dir
);
9352 inode_inc_iversion(new_dir
);
9353 inode_inc_iversion(old_inode
);
9354 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9355 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9356 old_inode
->i_ctime
= ctime
;
9358 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9359 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
9361 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9362 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9363 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
9364 old_dentry
->d_name
.name
,
9365 old_dentry
->d_name
.len
);
9367 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
9368 d_inode(old_dentry
),
9369 old_dentry
->d_name
.name
,
9370 old_dentry
->d_name
.len
);
9372 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9375 btrfs_abort_transaction(trans
, root
, ret
);
9380 inode_inc_iversion(new_inode
);
9381 new_inode
->i_ctime
= CURRENT_TIME
;
9382 if (unlikely(btrfs_ino(new_inode
) ==
9383 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9384 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9385 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9387 new_dentry
->d_name
.name
,
9388 new_dentry
->d_name
.len
);
9389 BUG_ON(new_inode
->i_nlink
== 0);
9391 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
9392 d_inode(new_dentry
),
9393 new_dentry
->d_name
.name
,
9394 new_dentry
->d_name
.len
);
9396 if (!ret
&& new_inode
->i_nlink
== 0)
9397 ret
= btrfs_orphan_add(trans
, d_inode(new_dentry
));
9399 btrfs_abort_transaction(trans
, root
, ret
);
9404 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
9405 new_dentry
->d_name
.name
,
9406 new_dentry
->d_name
.len
, 0, index
);
9408 btrfs_abort_transaction(trans
, root
, ret
);
9412 if (old_inode
->i_nlink
== 1)
9413 BTRFS_I(old_inode
)->dir_index
= index
;
9415 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
9416 struct dentry
*parent
= new_dentry
->d_parent
;
9417 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
9418 btrfs_end_log_trans(root
);
9421 btrfs_end_transaction(trans
, root
);
9423 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9424 up_read(&root
->fs_info
->subvol_sem
);
9429 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
9430 struct inode
*new_dir
, struct dentry
*new_dentry
,
9433 if (flags
& ~RENAME_NOREPLACE
)
9436 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
9439 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9441 struct btrfs_delalloc_work
*delalloc_work
;
9442 struct inode
*inode
;
9444 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9446 inode
= delalloc_work
->inode
;
9447 filemap_flush(inode
->i_mapping
);
9448 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9449 &BTRFS_I(inode
)->runtime_flags
))
9450 filemap_flush(inode
->i_mapping
);
9452 if (delalloc_work
->delay_iput
)
9453 btrfs_add_delayed_iput(inode
);
9456 complete(&delalloc_work
->completion
);
9459 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
9462 struct btrfs_delalloc_work
*work
;
9464 work
= kmalloc(sizeof(*work
), GFP_NOFS
);
9468 init_completion(&work
->completion
);
9469 INIT_LIST_HEAD(&work
->list
);
9470 work
->inode
= inode
;
9471 work
->delay_iput
= delay_iput
;
9472 WARN_ON_ONCE(!inode
);
9473 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
9474 btrfs_run_delalloc_work
, NULL
, NULL
);
9479 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
9481 wait_for_completion(&work
->completion
);
9486 * some fairly slow code that needs optimization. This walks the list
9487 * of all the inodes with pending delalloc and forces them to disk.
9489 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
9492 struct btrfs_inode
*binode
;
9493 struct inode
*inode
;
9494 struct btrfs_delalloc_work
*work
, *next
;
9495 struct list_head works
;
9496 struct list_head splice
;
9499 INIT_LIST_HEAD(&works
);
9500 INIT_LIST_HEAD(&splice
);
9502 mutex_lock(&root
->delalloc_mutex
);
9503 spin_lock(&root
->delalloc_lock
);
9504 list_splice_init(&root
->delalloc_inodes
, &splice
);
9505 while (!list_empty(&splice
)) {
9506 binode
= list_entry(splice
.next
, struct btrfs_inode
,
9509 list_move_tail(&binode
->delalloc_inodes
,
9510 &root
->delalloc_inodes
);
9511 inode
= igrab(&binode
->vfs_inode
);
9513 cond_resched_lock(&root
->delalloc_lock
);
9516 spin_unlock(&root
->delalloc_lock
);
9518 work
= btrfs_alloc_delalloc_work(inode
, delay_iput
);
9521 btrfs_add_delayed_iput(inode
);
9527 list_add_tail(&work
->list
, &works
);
9528 btrfs_queue_work(root
->fs_info
->flush_workers
,
9531 if (nr
!= -1 && ret
>= nr
)
9534 spin_lock(&root
->delalloc_lock
);
9536 spin_unlock(&root
->delalloc_lock
);
9539 list_for_each_entry_safe(work
, next
, &works
, list
) {
9540 list_del_init(&work
->list
);
9541 btrfs_wait_and_free_delalloc_work(work
);
9544 if (!list_empty_careful(&splice
)) {
9545 spin_lock(&root
->delalloc_lock
);
9546 list_splice_tail(&splice
, &root
->delalloc_inodes
);
9547 spin_unlock(&root
->delalloc_lock
);
9549 mutex_unlock(&root
->delalloc_mutex
);
9553 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
9557 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
9560 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
9564 * the filemap_flush will queue IO into the worker threads, but
9565 * we have to make sure the IO is actually started and that
9566 * ordered extents get created before we return
9568 atomic_inc(&root
->fs_info
->async_submit_draining
);
9569 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
9570 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
9571 wait_event(root
->fs_info
->async_submit_wait
,
9572 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
9573 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
9575 atomic_dec(&root
->fs_info
->async_submit_draining
);
9579 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
9582 struct btrfs_root
*root
;
9583 struct list_head splice
;
9586 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
9589 INIT_LIST_HEAD(&splice
);
9591 mutex_lock(&fs_info
->delalloc_root_mutex
);
9592 spin_lock(&fs_info
->delalloc_root_lock
);
9593 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
9594 while (!list_empty(&splice
) && nr
) {
9595 root
= list_first_entry(&splice
, struct btrfs_root
,
9597 root
= btrfs_grab_fs_root(root
);
9599 list_move_tail(&root
->delalloc_root
,
9600 &fs_info
->delalloc_roots
);
9601 spin_unlock(&fs_info
->delalloc_root_lock
);
9603 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
9604 btrfs_put_fs_root(root
);
9612 spin_lock(&fs_info
->delalloc_root_lock
);
9614 spin_unlock(&fs_info
->delalloc_root_lock
);
9617 atomic_inc(&fs_info
->async_submit_draining
);
9618 while (atomic_read(&fs_info
->nr_async_submits
) ||
9619 atomic_read(&fs_info
->async_delalloc_pages
)) {
9620 wait_event(fs_info
->async_submit_wait
,
9621 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
9622 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
9624 atomic_dec(&fs_info
->async_submit_draining
);
9626 if (!list_empty_careful(&splice
)) {
9627 spin_lock(&fs_info
->delalloc_root_lock
);
9628 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
9629 spin_unlock(&fs_info
->delalloc_root_lock
);
9631 mutex_unlock(&fs_info
->delalloc_root_mutex
);
9635 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
9636 const char *symname
)
9638 struct btrfs_trans_handle
*trans
;
9639 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9640 struct btrfs_path
*path
;
9641 struct btrfs_key key
;
9642 struct inode
*inode
= NULL
;
9650 struct btrfs_file_extent_item
*ei
;
9651 struct extent_buffer
*leaf
;
9653 name_len
= strlen(symname
);
9654 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
9655 return -ENAMETOOLONG
;
9658 * 2 items for inode item and ref
9659 * 2 items for dir items
9660 * 1 item for updating parent inode item
9661 * 1 item for the inline extent item
9662 * 1 item for xattr if selinux is on
9664 trans
= btrfs_start_transaction(root
, 7);
9666 return PTR_ERR(trans
);
9668 err
= btrfs_find_free_ino(root
, &objectid
);
9672 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
9673 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
9674 S_IFLNK
|S_IRWXUGO
, &index
);
9675 if (IS_ERR(inode
)) {
9676 err
= PTR_ERR(inode
);
9681 * If the active LSM wants to access the inode during
9682 * d_instantiate it needs these. Smack checks to see
9683 * if the filesystem supports xattrs by looking at the
9686 inode
->i_fop
= &btrfs_file_operations
;
9687 inode
->i_op
= &btrfs_file_inode_operations
;
9688 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9689 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9691 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
9693 goto out_unlock_inode
;
9695 path
= btrfs_alloc_path();
9698 goto out_unlock_inode
;
9700 key
.objectid
= btrfs_ino(inode
);
9702 key
.type
= BTRFS_EXTENT_DATA_KEY
;
9703 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
9704 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
9707 btrfs_free_path(path
);
9708 goto out_unlock_inode
;
9710 leaf
= path
->nodes
[0];
9711 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
9712 struct btrfs_file_extent_item
);
9713 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
9714 btrfs_set_file_extent_type(leaf
, ei
,
9715 BTRFS_FILE_EXTENT_INLINE
);
9716 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
9717 btrfs_set_file_extent_compression(leaf
, ei
, 0);
9718 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
9719 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
9721 ptr
= btrfs_file_extent_inline_start(ei
);
9722 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
9723 btrfs_mark_buffer_dirty(leaf
);
9724 btrfs_free_path(path
);
9726 inode
->i_op
= &btrfs_symlink_inode_operations
;
9727 inode_nohighmem(inode
);
9728 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
9729 inode_set_bytes(inode
, name_len
);
9730 btrfs_i_size_write(inode
, name_len
);
9731 err
= btrfs_update_inode(trans
, root
, inode
);
9733 * Last step, add directory indexes for our symlink inode. This is the
9734 * last step to avoid extra cleanup of these indexes if an error happens
9738 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
9741 goto out_unlock_inode
;
9744 unlock_new_inode(inode
);
9745 d_instantiate(dentry
, inode
);
9748 btrfs_end_transaction(trans
, root
);
9750 inode_dec_link_count(inode
);
9753 btrfs_btree_balance_dirty(root
);
9758 unlock_new_inode(inode
);
9762 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9763 u64 start
, u64 num_bytes
, u64 min_size
,
9764 loff_t actual_len
, u64
*alloc_hint
,
9765 struct btrfs_trans_handle
*trans
)
9767 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
9768 struct extent_map
*em
;
9769 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9770 struct btrfs_key ins
;
9771 u64 cur_offset
= start
;
9774 u64 last_alloc
= (u64
)-1;
9776 bool own_trans
= true;
9780 while (num_bytes
> 0) {
9782 trans
= btrfs_start_transaction(root
, 3);
9783 if (IS_ERR(trans
)) {
9784 ret
= PTR_ERR(trans
);
9789 cur_bytes
= min_t(u64
, num_bytes
, SZ_256M
);
9790 cur_bytes
= max(cur_bytes
, min_size
);
9792 * If we are severely fragmented we could end up with really
9793 * small allocations, so if the allocator is returning small
9794 * chunks lets make its job easier by only searching for those
9797 cur_bytes
= min(cur_bytes
, last_alloc
);
9798 ret
= btrfs_reserve_extent(root
, cur_bytes
, min_size
, 0,
9799 *alloc_hint
, &ins
, 1, 0);
9802 btrfs_end_transaction(trans
, root
);
9806 last_alloc
= ins
.offset
;
9807 ret
= insert_reserved_file_extent(trans
, inode
,
9808 cur_offset
, ins
.objectid
,
9809 ins
.offset
, ins
.offset
,
9810 ins
.offset
, 0, 0, 0,
9811 BTRFS_FILE_EXTENT_PREALLOC
);
9813 btrfs_free_reserved_extent(root
, ins
.objectid
,
9815 btrfs_abort_transaction(trans
, root
, ret
);
9817 btrfs_end_transaction(trans
, root
);
9821 btrfs_drop_extent_cache(inode
, cur_offset
,
9822 cur_offset
+ ins
.offset
-1, 0);
9824 em
= alloc_extent_map();
9826 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
9827 &BTRFS_I(inode
)->runtime_flags
);
9831 em
->start
= cur_offset
;
9832 em
->orig_start
= cur_offset
;
9833 em
->len
= ins
.offset
;
9834 em
->block_start
= ins
.objectid
;
9835 em
->block_len
= ins
.offset
;
9836 em
->orig_block_len
= ins
.offset
;
9837 em
->ram_bytes
= ins
.offset
;
9838 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
9839 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
9840 em
->generation
= trans
->transid
;
9843 write_lock(&em_tree
->lock
);
9844 ret
= add_extent_mapping(em_tree
, em
, 1);
9845 write_unlock(&em_tree
->lock
);
9848 btrfs_drop_extent_cache(inode
, cur_offset
,
9849 cur_offset
+ ins
.offset
- 1,
9852 free_extent_map(em
);
9854 num_bytes
-= ins
.offset
;
9855 cur_offset
+= ins
.offset
;
9856 *alloc_hint
= ins
.objectid
+ ins
.offset
;
9858 inode_inc_iversion(inode
);
9859 inode
->i_ctime
= CURRENT_TIME
;
9860 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
9861 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
9862 (actual_len
> inode
->i_size
) &&
9863 (cur_offset
> inode
->i_size
)) {
9864 if (cur_offset
> actual_len
)
9865 i_size
= actual_len
;
9867 i_size
= cur_offset
;
9868 i_size_write(inode
, i_size
);
9869 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
9872 ret
= btrfs_update_inode(trans
, root
, inode
);
9875 btrfs_abort_transaction(trans
, root
, ret
);
9877 btrfs_end_transaction(trans
, root
);
9882 btrfs_end_transaction(trans
, root
);
9887 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9888 u64 start
, u64 num_bytes
, u64 min_size
,
9889 loff_t actual_len
, u64
*alloc_hint
)
9891 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9892 min_size
, actual_len
, alloc_hint
,
9896 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
9897 struct btrfs_trans_handle
*trans
, int mode
,
9898 u64 start
, u64 num_bytes
, u64 min_size
,
9899 loff_t actual_len
, u64
*alloc_hint
)
9901 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9902 min_size
, actual_len
, alloc_hint
, trans
);
9905 static int btrfs_set_page_dirty(struct page
*page
)
9907 return __set_page_dirty_nobuffers(page
);
9910 static int btrfs_permission(struct inode
*inode
, int mask
)
9912 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9913 umode_t mode
= inode
->i_mode
;
9915 if (mask
& MAY_WRITE
&&
9916 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
9917 if (btrfs_root_readonly(root
))
9919 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
9922 return generic_permission(inode
, mask
);
9925 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
9927 struct btrfs_trans_handle
*trans
;
9928 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9929 struct inode
*inode
= NULL
;
9935 * 5 units required for adding orphan entry
9937 trans
= btrfs_start_transaction(root
, 5);
9939 return PTR_ERR(trans
);
9941 ret
= btrfs_find_free_ino(root
, &objectid
);
9945 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
9946 btrfs_ino(dir
), objectid
, mode
, &index
);
9947 if (IS_ERR(inode
)) {
9948 ret
= PTR_ERR(inode
);
9953 inode
->i_fop
= &btrfs_file_operations
;
9954 inode
->i_op
= &btrfs_file_inode_operations
;
9956 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9957 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9959 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
9963 ret
= btrfs_update_inode(trans
, root
, inode
);
9966 ret
= btrfs_orphan_add(trans
, inode
);
9971 * We set number of links to 0 in btrfs_new_inode(), and here we set
9972 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9975 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9977 set_nlink(inode
, 1);
9978 unlock_new_inode(inode
);
9979 d_tmpfile(dentry
, inode
);
9980 mark_inode_dirty(inode
);
9983 btrfs_end_transaction(trans
, root
);
9986 btrfs_balance_delayed_items(root
);
9987 btrfs_btree_balance_dirty(root
);
9991 unlock_new_inode(inode
);
9996 /* Inspired by filemap_check_errors() */
9997 int btrfs_inode_check_errors(struct inode
*inode
)
10001 if (test_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
) &&
10002 test_and_clear_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
))
10004 if (test_bit(AS_EIO
, &inode
->i_mapping
->flags
) &&
10005 test_and_clear_bit(AS_EIO
, &inode
->i_mapping
->flags
))
10011 static const struct inode_operations btrfs_dir_inode_operations
= {
10012 .getattr
= btrfs_getattr
,
10013 .lookup
= btrfs_lookup
,
10014 .create
= btrfs_create
,
10015 .unlink
= btrfs_unlink
,
10016 .link
= btrfs_link
,
10017 .mkdir
= btrfs_mkdir
,
10018 .rmdir
= btrfs_rmdir
,
10019 .rename2
= btrfs_rename2
,
10020 .symlink
= btrfs_symlink
,
10021 .setattr
= btrfs_setattr
,
10022 .mknod
= btrfs_mknod
,
10023 .setxattr
= btrfs_setxattr
,
10024 .getxattr
= generic_getxattr
,
10025 .listxattr
= btrfs_listxattr
,
10026 .removexattr
= btrfs_removexattr
,
10027 .permission
= btrfs_permission
,
10028 .get_acl
= btrfs_get_acl
,
10029 .set_acl
= btrfs_set_acl
,
10030 .update_time
= btrfs_update_time
,
10031 .tmpfile
= btrfs_tmpfile
,
10033 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
10034 .lookup
= btrfs_lookup
,
10035 .permission
= btrfs_permission
,
10036 .get_acl
= btrfs_get_acl
,
10037 .set_acl
= btrfs_set_acl
,
10038 .update_time
= btrfs_update_time
,
10041 static const struct file_operations btrfs_dir_file_operations
= {
10042 .llseek
= generic_file_llseek
,
10043 .read
= generic_read_dir
,
10044 .iterate
= btrfs_real_readdir
,
10045 .unlocked_ioctl
= btrfs_ioctl
,
10046 #ifdef CONFIG_COMPAT
10047 .compat_ioctl
= btrfs_ioctl
,
10049 .release
= btrfs_release_file
,
10050 .fsync
= btrfs_sync_file
,
10053 static const struct extent_io_ops btrfs_extent_io_ops
= {
10054 .fill_delalloc
= run_delalloc_range
,
10055 .submit_bio_hook
= btrfs_submit_bio_hook
,
10056 .merge_bio_hook
= btrfs_merge_bio_hook
,
10057 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
10058 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
10059 .writepage_start_hook
= btrfs_writepage_start_hook
,
10060 .set_bit_hook
= btrfs_set_bit_hook
,
10061 .clear_bit_hook
= btrfs_clear_bit_hook
,
10062 .merge_extent_hook
= btrfs_merge_extent_hook
,
10063 .split_extent_hook
= btrfs_split_extent_hook
,
10067 * btrfs doesn't support the bmap operation because swapfiles
10068 * use bmap to make a mapping of extents in the file. They assume
10069 * these extents won't change over the life of the file and they
10070 * use the bmap result to do IO directly to the drive.
10072 * the btrfs bmap call would return logical addresses that aren't
10073 * suitable for IO and they also will change frequently as COW
10074 * operations happen. So, swapfile + btrfs == corruption.
10076 * For now we're avoiding this by dropping bmap.
10078 static const struct address_space_operations btrfs_aops
= {
10079 .readpage
= btrfs_readpage
,
10080 .writepage
= btrfs_writepage
,
10081 .writepages
= btrfs_writepages
,
10082 .readpages
= btrfs_readpages
,
10083 .direct_IO
= btrfs_direct_IO
,
10084 .invalidatepage
= btrfs_invalidatepage
,
10085 .releasepage
= btrfs_releasepage
,
10086 .set_page_dirty
= btrfs_set_page_dirty
,
10087 .error_remove_page
= generic_error_remove_page
,
10090 static const struct address_space_operations btrfs_symlink_aops
= {
10091 .readpage
= btrfs_readpage
,
10092 .writepage
= btrfs_writepage
,
10093 .invalidatepage
= btrfs_invalidatepage
,
10094 .releasepage
= btrfs_releasepage
,
10097 static const struct inode_operations btrfs_file_inode_operations
= {
10098 .getattr
= btrfs_getattr
,
10099 .setattr
= btrfs_setattr
,
10100 .setxattr
= btrfs_setxattr
,
10101 .getxattr
= generic_getxattr
,
10102 .listxattr
= btrfs_listxattr
,
10103 .removexattr
= btrfs_removexattr
,
10104 .permission
= btrfs_permission
,
10105 .fiemap
= btrfs_fiemap
,
10106 .get_acl
= btrfs_get_acl
,
10107 .set_acl
= btrfs_set_acl
,
10108 .update_time
= btrfs_update_time
,
10110 static const struct inode_operations btrfs_special_inode_operations
= {
10111 .getattr
= btrfs_getattr
,
10112 .setattr
= btrfs_setattr
,
10113 .permission
= btrfs_permission
,
10114 .setxattr
= btrfs_setxattr
,
10115 .getxattr
= generic_getxattr
,
10116 .listxattr
= btrfs_listxattr
,
10117 .removexattr
= btrfs_removexattr
,
10118 .get_acl
= btrfs_get_acl
,
10119 .set_acl
= btrfs_set_acl
,
10120 .update_time
= btrfs_update_time
,
10122 static const struct inode_operations btrfs_symlink_inode_operations
= {
10123 .readlink
= generic_readlink
,
10124 .get_link
= page_get_link
,
10125 .getattr
= btrfs_getattr
,
10126 .setattr
= btrfs_setattr
,
10127 .permission
= btrfs_permission
,
10128 .setxattr
= btrfs_setxattr
,
10129 .getxattr
= generic_getxattr
,
10130 .listxattr
= btrfs_listxattr
,
10131 .removexattr
= btrfs_removexattr
,
10132 .update_time
= btrfs_update_time
,
10135 const struct dentry_operations btrfs_dentry_operations
= {
10136 .d_delete
= btrfs_dentry_delete
,
10137 .d_release
= btrfs_dentry_release
,