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 static const struct inode_operations btrfs_dir_inode_operations
;
70 static const struct inode_operations btrfs_symlink_inode_operations
;
71 static const struct inode_operations btrfs_dir_ro_inode_operations
;
72 static const struct inode_operations btrfs_special_inode_operations
;
73 static const struct inode_operations btrfs_file_inode_operations
;
74 static const struct address_space_operations btrfs_aops
;
75 static const struct address_space_operations btrfs_symlink_aops
;
76 static const struct file_operations btrfs_dir_file_operations
;
77 static struct extent_io_ops btrfs_extent_io_ops
;
79 static struct kmem_cache
*btrfs_inode_cachep
;
80 static struct kmem_cache
*btrfs_delalloc_work_cachep
;
81 struct kmem_cache
*btrfs_trans_handle_cachep
;
82 struct kmem_cache
*btrfs_transaction_cachep
;
83 struct kmem_cache
*btrfs_path_cachep
;
84 struct kmem_cache
*btrfs_free_space_cachep
;
87 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
88 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
89 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
90 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
91 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
92 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
93 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
94 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
97 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
98 static int btrfs_truncate(struct inode
*inode
);
99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
100 static noinline
int cow_file_range(struct inode
*inode
,
101 struct page
*locked_page
,
102 u64 start
, u64 end
, int *page_started
,
103 unsigned long *nr_written
, int unlock
);
104 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
105 u64 len
, u64 orig_start
,
106 u64 block_start
, u64 block_len
,
107 u64 orig_block_len
, u64 ram_bytes
,
110 static int btrfs_dirty_inode(struct inode
*inode
);
112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113 void btrfs_test_inode_set_ops(struct inode
*inode
)
115 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
119 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
120 struct inode
*inode
, struct inode
*dir
,
121 const struct qstr
*qstr
)
125 err
= btrfs_init_acl(trans
, inode
, dir
);
127 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
136 static int insert_inline_extent(struct btrfs_trans_handle
*trans
,
137 struct btrfs_path
*path
, int extent_inserted
,
138 struct btrfs_root
*root
, struct inode
*inode
,
139 u64 start
, size_t size
, size_t compressed_size
,
141 struct page
**compressed_pages
)
143 struct extent_buffer
*leaf
;
144 struct page
*page
= NULL
;
147 struct btrfs_file_extent_item
*ei
;
150 size_t cur_size
= size
;
151 unsigned long offset
;
153 if (compressed_size
&& compressed_pages
)
154 cur_size
= compressed_size
;
156 inode_add_bytes(inode
, size
);
158 if (!extent_inserted
) {
159 struct btrfs_key key
;
162 key
.objectid
= btrfs_ino(inode
);
164 key
.type
= BTRFS_EXTENT_DATA_KEY
;
166 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
167 path
->leave_spinning
= 1;
168 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
175 leaf
= path
->nodes
[0];
176 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
177 struct btrfs_file_extent_item
);
178 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
179 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
180 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
181 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
182 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
183 ptr
= btrfs_file_extent_inline_start(ei
);
185 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
188 while (compressed_size
> 0) {
189 cpage
= compressed_pages
[i
];
190 cur_size
= min_t(unsigned long, compressed_size
,
193 kaddr
= kmap_atomic(cpage
);
194 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
195 kunmap_atomic(kaddr
);
199 compressed_size
-= cur_size
;
201 btrfs_set_file_extent_compression(leaf
, ei
,
204 page
= find_get_page(inode
->i_mapping
,
205 start
>> PAGE_CACHE_SHIFT
);
206 btrfs_set_file_extent_compression(leaf
, ei
, 0);
207 kaddr
= kmap_atomic(page
);
208 offset
= start
& (PAGE_CACHE_SIZE
- 1);
209 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
210 kunmap_atomic(kaddr
);
211 page_cache_release(page
);
213 btrfs_mark_buffer_dirty(leaf
);
214 btrfs_release_path(path
);
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
225 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
226 ret
= btrfs_update_inode(trans
, root
, inode
);
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
239 static noinline
int cow_file_range_inline(struct btrfs_root
*root
,
240 struct inode
*inode
, u64 start
,
241 u64 end
, size_t compressed_size
,
243 struct page
**compressed_pages
)
245 struct btrfs_trans_handle
*trans
;
246 u64 isize
= i_size_read(inode
);
247 u64 actual_end
= min(end
+ 1, isize
);
248 u64 inline_len
= actual_end
- start
;
249 u64 aligned_end
= ALIGN(end
, root
->sectorsize
);
250 u64 data_len
= inline_len
;
252 struct btrfs_path
*path
;
253 int extent_inserted
= 0;
254 u32 extent_item_size
;
257 data_len
= compressed_size
;
260 actual_end
> PAGE_CACHE_SIZE
||
261 data_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
263 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
265 data_len
> root
->fs_info
->max_inline
) {
269 path
= btrfs_alloc_path();
273 trans
= btrfs_join_transaction(root
);
275 btrfs_free_path(path
);
276 return PTR_ERR(trans
);
278 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
280 if (compressed_size
&& compressed_pages
)
281 extent_item_size
= btrfs_file_extent_calc_inline_size(
284 extent_item_size
= btrfs_file_extent_calc_inline_size(
287 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
288 start
, aligned_end
, NULL
,
289 1, 1, extent_item_size
, &extent_inserted
);
291 btrfs_abort_transaction(trans
, root
, ret
);
295 if (isize
> actual_end
)
296 inline_len
= min_t(u64
, isize
, actual_end
);
297 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
299 inline_len
, compressed_size
,
300 compress_type
, compressed_pages
);
301 if (ret
&& ret
!= -ENOSPC
) {
302 btrfs_abort_transaction(trans
, root
, ret
);
304 } else if (ret
== -ENOSPC
) {
309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
310 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
311 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
313 btrfs_free_path(path
);
314 btrfs_end_transaction(trans
, root
);
318 struct async_extent
{
323 unsigned long nr_pages
;
325 struct list_head list
;
330 struct btrfs_root
*root
;
331 struct page
*locked_page
;
334 struct list_head extents
;
335 struct btrfs_work work
;
338 static noinline
int add_async_extent(struct async_cow
*cow
,
339 u64 start
, u64 ram_size
,
342 unsigned long nr_pages
,
345 struct async_extent
*async_extent
;
347 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
348 BUG_ON(!async_extent
); /* -ENOMEM */
349 async_extent
->start
= start
;
350 async_extent
->ram_size
= ram_size
;
351 async_extent
->compressed_size
= compressed_size
;
352 async_extent
->pages
= pages
;
353 async_extent
->nr_pages
= nr_pages
;
354 async_extent
->compress_type
= compress_type
;
355 list_add_tail(&async_extent
->list
, &cow
->extents
);
359 static inline int inode_need_compress(struct inode
*inode
)
361 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
364 if (btrfs_test_opt(root
, FORCE_COMPRESS
))
366 /* bad compression ratios */
367 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
369 if (btrfs_test_opt(root
, COMPRESS
) ||
370 BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
||
371 BTRFS_I(inode
)->force_compress
)
377 * we create compressed extents in two phases. The first
378 * phase compresses a range of pages that have already been
379 * locked (both pages and state bits are locked).
381 * This is done inside an ordered work queue, and the compression
382 * is spread across many cpus. The actual IO submission is step
383 * two, and the ordered work queue takes care of making sure that
384 * happens in the same order things were put onto the queue by
385 * writepages and friends.
387 * If this code finds it can't get good compression, it puts an
388 * entry onto the work queue to write the uncompressed bytes. This
389 * makes sure that both compressed inodes and uncompressed inodes
390 * are written in the same order that the flusher thread sent them
393 static noinline
void compress_file_range(struct inode
*inode
,
394 struct page
*locked_page
,
396 struct async_cow
*async_cow
,
399 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
401 u64 blocksize
= root
->sectorsize
;
403 u64 isize
= i_size_read(inode
);
405 struct page
**pages
= NULL
;
406 unsigned long nr_pages
;
407 unsigned long nr_pages_ret
= 0;
408 unsigned long total_compressed
= 0;
409 unsigned long total_in
= 0;
410 unsigned long max_compressed
= 128 * 1024;
411 unsigned long max_uncompressed
= 128 * 1024;
414 int compress_type
= root
->fs_info
->compress_type
;
417 /* if this is a small write inside eof, kick off a defrag */
418 if ((end
- start
+ 1) < 16 * 1024 &&
419 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
420 btrfs_add_inode_defrag(NULL
, inode
);
422 actual_end
= min_t(u64
, isize
, end
+ 1);
425 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
426 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
429 * we don't want to send crud past the end of i_size through
430 * compression, that's just a waste of CPU time. So, if the
431 * end of the file is before the start of our current
432 * requested range of bytes, we bail out to the uncompressed
433 * cleanup code that can deal with all of this.
435 * It isn't really the fastest way to fix things, but this is a
436 * very uncommon corner.
438 if (actual_end
<= start
)
439 goto cleanup_and_bail_uncompressed
;
441 total_compressed
= actual_end
- start
;
444 * skip compression for a small file range(<=blocksize) that
445 * isn't an inline extent, since it dosen't save disk space at all.
447 if (total_compressed
<= blocksize
&&
448 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
449 goto cleanup_and_bail_uncompressed
;
451 /* we want to make sure that amount of ram required to uncompress
452 * an extent is reasonable, so we limit the total size in ram
453 * of a compressed extent to 128k. This is a crucial number
454 * because it also controls how easily we can spread reads across
455 * cpus for decompression.
457 * We also want to make sure the amount of IO required to do
458 * a random read is reasonably small, so we limit the size of
459 * a compressed extent to 128k.
461 total_compressed
= min(total_compressed
, max_uncompressed
);
462 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
463 num_bytes
= max(blocksize
, num_bytes
);
468 * we do compression for mount -o compress and when the
469 * inode has not been flagged as nocompress. This flag can
470 * change at any time if we discover bad compression ratios.
472 if (inode_need_compress(inode
)) {
474 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_NOFS
);
476 /* just bail out to the uncompressed code */
480 if (BTRFS_I(inode
)->force_compress
)
481 compress_type
= BTRFS_I(inode
)->force_compress
;
484 * we need to call clear_page_dirty_for_io on each
485 * page in the range. Otherwise applications with the file
486 * mmap'd can wander in and change the page contents while
487 * we are compressing them.
489 * If the compression fails for any reason, we set the pages
490 * dirty again later on.
492 extent_range_clear_dirty_for_io(inode
, start
, end
);
494 ret
= btrfs_compress_pages(compress_type
,
495 inode
->i_mapping
, start
,
496 total_compressed
, pages
,
497 nr_pages
, &nr_pages_ret
,
503 unsigned long offset
= total_compressed
&
504 (PAGE_CACHE_SIZE
- 1);
505 struct page
*page
= pages
[nr_pages_ret
- 1];
508 /* zero the tail end of the last page, we might be
509 * sending it down to disk
512 kaddr
= kmap_atomic(page
);
513 memset(kaddr
+ offset
, 0,
514 PAGE_CACHE_SIZE
- offset
);
515 kunmap_atomic(kaddr
);
522 /* lets try to make an inline extent */
523 if (ret
|| total_in
< (actual_end
- start
)) {
524 /* we didn't compress the entire range, try
525 * to make an uncompressed inline extent.
527 ret
= cow_file_range_inline(root
, inode
, start
, end
,
530 /* try making a compressed inline extent */
531 ret
= cow_file_range_inline(root
, inode
, start
, end
,
533 compress_type
, pages
);
536 unsigned long clear_flags
= EXTENT_DELALLOC
|
538 unsigned long page_error_op
;
540 clear_flags
|= (ret
< 0) ? EXTENT_DO_ACCOUNTING
: 0;
541 page_error_op
= ret
< 0 ? PAGE_SET_ERROR
: 0;
544 * inline extent creation worked or returned error,
545 * we don't need to create any more async work items.
546 * Unlock and free up our temp pages.
548 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
549 clear_flags
, PAGE_UNLOCK
|
560 * we aren't doing an inline extent round the compressed size
561 * up to a block size boundary so the allocator does sane
564 total_compressed
= ALIGN(total_compressed
, blocksize
);
567 * one last check to make sure the compression is really a
568 * win, compare the page count read with the blocks on disk
570 total_in
= ALIGN(total_in
, PAGE_CACHE_SIZE
);
571 if (total_compressed
>= total_in
) {
574 num_bytes
= total_in
;
577 if (!will_compress
&& pages
) {
579 * the compression code ran but failed to make things smaller,
580 * free any pages it allocated and our page pointer array
582 for (i
= 0; i
< nr_pages_ret
; i
++) {
583 WARN_ON(pages
[i
]->mapping
);
584 page_cache_release(pages
[i
]);
588 total_compressed
= 0;
591 /* flag the file so we don't compress in the future */
592 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
593 !(BTRFS_I(inode
)->force_compress
)) {
594 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
600 /* the async work queues will take care of doing actual
601 * allocation on disk for these compressed pages,
602 * and will submit them to the elevator.
604 add_async_extent(async_cow
, start
, num_bytes
,
605 total_compressed
, pages
, nr_pages_ret
,
608 if (start
+ num_bytes
< end
) {
615 cleanup_and_bail_uncompressed
:
617 * No compression, but we still need to write the pages in
618 * the file we've been given so far. redirty the locked
619 * page if it corresponds to our extent and set things up
620 * for the async work queue to run cow_file_range to do
621 * the normal delalloc dance
623 if (page_offset(locked_page
) >= start
&&
624 page_offset(locked_page
) <= end
) {
625 __set_page_dirty_nobuffers(locked_page
);
626 /* unlocked later on in the async handlers */
629 extent_range_redirty_for_io(inode
, start
, end
);
630 add_async_extent(async_cow
, start
, end
- start
+ 1,
631 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
638 for (i
= 0; i
< nr_pages_ret
; i
++) {
639 WARN_ON(pages
[i
]->mapping
);
640 page_cache_release(pages
[i
]);
645 static void free_async_extent_pages(struct async_extent
*async_extent
)
649 if (!async_extent
->pages
)
652 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
653 WARN_ON(async_extent
->pages
[i
]->mapping
);
654 page_cache_release(async_extent
->pages
[i
]);
656 kfree(async_extent
->pages
);
657 async_extent
->nr_pages
= 0;
658 async_extent
->pages
= NULL
;
662 * phase two of compressed writeback. This is the ordered portion
663 * of the code, which only gets called in the order the work was
664 * queued. We walk all the async extents created by compress_file_range
665 * and send them down to the disk.
667 static noinline
void submit_compressed_extents(struct inode
*inode
,
668 struct async_cow
*async_cow
)
670 struct async_extent
*async_extent
;
672 struct btrfs_key ins
;
673 struct extent_map
*em
;
674 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
675 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
676 struct extent_io_tree
*io_tree
;
680 while (!list_empty(&async_cow
->extents
)) {
681 async_extent
= list_entry(async_cow
->extents
.next
,
682 struct async_extent
, list
);
683 list_del(&async_extent
->list
);
685 io_tree
= &BTRFS_I(inode
)->io_tree
;
688 /* did the compression code fall back to uncompressed IO? */
689 if (!async_extent
->pages
) {
690 int page_started
= 0;
691 unsigned long nr_written
= 0;
693 lock_extent(io_tree
, async_extent
->start
,
694 async_extent
->start
+
695 async_extent
->ram_size
- 1);
697 /* allocate blocks */
698 ret
= cow_file_range(inode
, async_cow
->locked_page
,
700 async_extent
->start
+
701 async_extent
->ram_size
- 1,
702 &page_started
, &nr_written
, 0);
707 * if page_started, cow_file_range inserted an
708 * inline extent and took care of all the unlocking
709 * and IO for us. Otherwise, we need to submit
710 * all those pages down to the drive.
712 if (!page_started
&& !ret
)
713 extent_write_locked_range(io_tree
,
714 inode
, async_extent
->start
,
715 async_extent
->start
+
716 async_extent
->ram_size
- 1,
720 unlock_page(async_cow
->locked_page
);
726 lock_extent(io_tree
, async_extent
->start
,
727 async_extent
->start
+ async_extent
->ram_size
- 1);
729 ret
= btrfs_reserve_extent(root
,
730 async_extent
->compressed_size
,
731 async_extent
->compressed_size
,
732 0, alloc_hint
, &ins
, 1, 1);
734 free_async_extent_pages(async_extent
);
736 if (ret
== -ENOSPC
) {
737 unlock_extent(io_tree
, async_extent
->start
,
738 async_extent
->start
+
739 async_extent
->ram_size
- 1);
742 * we need to redirty the pages if we decide to
743 * fallback to uncompressed IO, otherwise we
744 * will not submit these pages down to lower
747 extent_range_redirty_for_io(inode
,
749 async_extent
->start
+
750 async_extent
->ram_size
- 1);
757 * here we're doing allocation and writeback of the
760 btrfs_drop_extent_cache(inode
, async_extent
->start
,
761 async_extent
->start
+
762 async_extent
->ram_size
- 1, 0);
764 em
= alloc_extent_map();
767 goto out_free_reserve
;
769 em
->start
= async_extent
->start
;
770 em
->len
= async_extent
->ram_size
;
771 em
->orig_start
= em
->start
;
772 em
->mod_start
= em
->start
;
773 em
->mod_len
= em
->len
;
775 em
->block_start
= ins
.objectid
;
776 em
->block_len
= ins
.offset
;
777 em
->orig_block_len
= ins
.offset
;
778 em
->ram_bytes
= async_extent
->ram_size
;
779 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
780 em
->compress_type
= async_extent
->compress_type
;
781 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
782 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
786 write_lock(&em_tree
->lock
);
787 ret
= add_extent_mapping(em_tree
, em
, 1);
788 write_unlock(&em_tree
->lock
);
789 if (ret
!= -EEXIST
) {
793 btrfs_drop_extent_cache(inode
, async_extent
->start
,
794 async_extent
->start
+
795 async_extent
->ram_size
- 1, 0);
799 goto out_free_reserve
;
801 ret
= btrfs_add_ordered_extent_compress(inode
,
804 async_extent
->ram_size
,
806 BTRFS_ORDERED_COMPRESSED
,
807 async_extent
->compress_type
);
809 btrfs_drop_extent_cache(inode
, async_extent
->start
,
810 async_extent
->start
+
811 async_extent
->ram_size
- 1, 0);
812 goto out_free_reserve
;
816 * clear dirty, set writeback and unlock the pages.
818 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
819 async_extent
->start
+
820 async_extent
->ram_size
- 1,
821 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
822 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
824 ret
= btrfs_submit_compressed_write(inode
,
826 async_extent
->ram_size
,
828 ins
.offset
, async_extent
->pages
,
829 async_extent
->nr_pages
);
831 struct extent_io_tree
*tree
= &BTRFS_I(inode
)->io_tree
;
832 struct page
*p
= async_extent
->pages
[0];
833 const u64 start
= async_extent
->start
;
834 const u64 end
= start
+ async_extent
->ram_size
- 1;
836 p
->mapping
= inode
->i_mapping
;
837 tree
->ops
->writepage_end_io_hook(p
, start
, end
,
840 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
, 0,
843 free_async_extent_pages(async_extent
);
845 alloc_hint
= ins
.objectid
+ ins
.offset
;
851 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
853 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
854 async_extent
->start
+
855 async_extent
->ram_size
- 1,
856 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
857 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
858 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
859 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
|
861 free_async_extent_pages(async_extent
);
866 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
869 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
870 struct extent_map
*em
;
873 read_lock(&em_tree
->lock
);
874 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
877 * if block start isn't an actual block number then find the
878 * first block in this inode and use that as a hint. If that
879 * block is also bogus then just don't worry about it.
881 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
883 em
= search_extent_mapping(em_tree
, 0, 0);
884 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
885 alloc_hint
= em
->block_start
;
889 alloc_hint
= em
->block_start
;
893 read_unlock(&em_tree
->lock
);
899 * when extent_io.c finds a delayed allocation range in the file,
900 * the call backs end up in this code. The basic idea is to
901 * allocate extents on disk for the range, and create ordered data structs
902 * in ram to track those extents.
904 * locked_page is the page that writepage had locked already. We use
905 * it to make sure we don't do extra locks or unlocks.
907 * *page_started is set to one if we unlock locked_page and do everything
908 * required to start IO on it. It may be clean and already done with
911 static noinline
int cow_file_range(struct inode
*inode
,
912 struct page
*locked_page
,
913 u64 start
, u64 end
, int *page_started
,
914 unsigned long *nr_written
,
917 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
920 unsigned long ram_size
;
923 u64 blocksize
= root
->sectorsize
;
924 struct btrfs_key ins
;
925 struct extent_map
*em
;
926 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
929 if (btrfs_is_free_space_inode(inode
)) {
935 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
936 num_bytes
= max(blocksize
, num_bytes
);
937 disk_num_bytes
= num_bytes
;
939 /* if this is a small write inside eof, kick off defrag */
940 if (num_bytes
< 64 * 1024 &&
941 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
942 btrfs_add_inode_defrag(NULL
, inode
);
945 /* lets try to make an inline extent */
946 ret
= cow_file_range_inline(root
, inode
, start
, end
, 0, 0,
949 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
950 EXTENT_LOCKED
| EXTENT_DELALLOC
|
951 EXTENT_DEFRAG
, PAGE_UNLOCK
|
952 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
955 *nr_written
= *nr_written
+
956 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
959 } else if (ret
< 0) {
964 BUG_ON(disk_num_bytes
>
965 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
967 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
968 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
970 while (disk_num_bytes
> 0) {
973 cur_alloc_size
= disk_num_bytes
;
974 ret
= btrfs_reserve_extent(root
, cur_alloc_size
,
975 root
->sectorsize
, 0, alloc_hint
,
980 em
= alloc_extent_map();
986 em
->orig_start
= em
->start
;
987 ram_size
= ins
.offset
;
988 em
->len
= ins
.offset
;
989 em
->mod_start
= em
->start
;
990 em
->mod_len
= em
->len
;
992 em
->block_start
= ins
.objectid
;
993 em
->block_len
= ins
.offset
;
994 em
->orig_block_len
= ins
.offset
;
995 em
->ram_bytes
= ram_size
;
996 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
997 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1001 write_lock(&em_tree
->lock
);
1002 ret
= add_extent_mapping(em_tree
, em
, 1);
1003 write_unlock(&em_tree
->lock
);
1004 if (ret
!= -EEXIST
) {
1005 free_extent_map(em
);
1008 btrfs_drop_extent_cache(inode
, start
,
1009 start
+ ram_size
- 1, 0);
1014 cur_alloc_size
= ins
.offset
;
1015 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
1016 ram_size
, cur_alloc_size
, 0);
1018 goto out_drop_extent_cache
;
1020 if (root
->root_key
.objectid
==
1021 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1022 ret
= btrfs_reloc_clone_csums(inode
, start
,
1025 goto out_drop_extent_cache
;
1028 if (disk_num_bytes
< cur_alloc_size
)
1031 /* we're not doing compressed IO, don't unlock the first
1032 * page (which the caller expects to stay locked), don't
1033 * clear any dirty bits and don't set any writeback bits
1035 * Do set the Private2 bit so we know this page was properly
1036 * setup for writepage
1038 op
= unlock
? PAGE_UNLOCK
: 0;
1039 op
|= PAGE_SET_PRIVATE2
;
1041 extent_clear_unlock_delalloc(inode
, start
,
1042 start
+ ram_size
- 1, locked_page
,
1043 EXTENT_LOCKED
| EXTENT_DELALLOC
,
1045 disk_num_bytes
-= cur_alloc_size
;
1046 num_bytes
-= cur_alloc_size
;
1047 alloc_hint
= ins
.objectid
+ ins
.offset
;
1048 start
+= cur_alloc_size
;
1053 out_drop_extent_cache
:
1054 btrfs_drop_extent_cache(inode
, start
, start
+ ram_size
- 1, 0);
1056 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
1058 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1059 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
1060 EXTENT_DELALLOC
| EXTENT_DEFRAG
,
1061 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
1062 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
);
1067 * work queue call back to started compression on a file and pages
1069 static noinline
void async_cow_start(struct btrfs_work
*work
)
1071 struct async_cow
*async_cow
;
1073 async_cow
= container_of(work
, struct async_cow
, work
);
1075 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1076 async_cow
->start
, async_cow
->end
, async_cow
,
1078 if (num_added
== 0) {
1079 btrfs_add_delayed_iput(async_cow
->inode
);
1080 async_cow
->inode
= NULL
;
1085 * work queue call back to submit previously compressed pages
1087 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1089 struct async_cow
*async_cow
;
1090 struct btrfs_root
*root
;
1091 unsigned long nr_pages
;
1093 async_cow
= container_of(work
, struct async_cow
, work
);
1095 root
= async_cow
->root
;
1096 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1100 * atomic_sub_return implies a barrier for waitqueue_active
1102 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1104 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1105 wake_up(&root
->fs_info
->async_submit_wait
);
1107 if (async_cow
->inode
)
1108 submit_compressed_extents(async_cow
->inode
, async_cow
);
1111 static noinline
void async_cow_free(struct btrfs_work
*work
)
1113 struct async_cow
*async_cow
;
1114 async_cow
= container_of(work
, struct async_cow
, work
);
1115 if (async_cow
->inode
)
1116 btrfs_add_delayed_iput(async_cow
->inode
);
1120 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1121 u64 start
, u64 end
, int *page_started
,
1122 unsigned long *nr_written
)
1124 struct async_cow
*async_cow
;
1125 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1126 unsigned long nr_pages
;
1128 int limit
= 10 * 1024 * 1024;
1130 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1131 1, 0, NULL
, GFP_NOFS
);
1132 while (start
< end
) {
1133 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1134 BUG_ON(!async_cow
); /* -ENOMEM */
1135 async_cow
->inode
= igrab(inode
);
1136 async_cow
->root
= root
;
1137 async_cow
->locked_page
= locked_page
;
1138 async_cow
->start
= start
;
1140 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1141 !btrfs_test_opt(root
, FORCE_COMPRESS
))
1144 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1146 async_cow
->end
= cur_end
;
1147 INIT_LIST_HEAD(&async_cow
->extents
);
1149 btrfs_init_work(&async_cow
->work
,
1150 btrfs_delalloc_helper
,
1151 async_cow_start
, async_cow_submit
,
1154 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1156 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1158 btrfs_queue_work(root
->fs_info
->delalloc_workers
,
1161 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1162 wait_event(root
->fs_info
->async_submit_wait
,
1163 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1167 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1168 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1169 wait_event(root
->fs_info
->async_submit_wait
,
1170 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1174 *nr_written
+= nr_pages
;
1175 start
= cur_end
+ 1;
1181 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1182 u64 bytenr
, u64 num_bytes
)
1185 struct btrfs_ordered_sum
*sums
;
1188 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1189 bytenr
+ num_bytes
- 1, &list
, 0);
1190 if (ret
== 0 && list_empty(&list
))
1193 while (!list_empty(&list
)) {
1194 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1195 list_del(&sums
->list
);
1202 * when nowcow writeback call back. This checks for snapshots or COW copies
1203 * of the extents that exist in the file, and COWs the file as required.
1205 * If no cow copies or snapshots exist, we write directly to the existing
1208 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1209 struct page
*locked_page
,
1210 u64 start
, u64 end
, int *page_started
, int force
,
1211 unsigned long *nr_written
)
1213 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1214 struct btrfs_trans_handle
*trans
;
1215 struct extent_buffer
*leaf
;
1216 struct btrfs_path
*path
;
1217 struct btrfs_file_extent_item
*fi
;
1218 struct btrfs_key found_key
;
1233 u64 ino
= btrfs_ino(inode
);
1235 path
= btrfs_alloc_path();
1237 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1238 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1239 EXTENT_DO_ACCOUNTING
|
1240 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1242 PAGE_SET_WRITEBACK
|
1243 PAGE_END_WRITEBACK
);
1247 nolock
= btrfs_is_free_space_inode(inode
);
1250 trans
= btrfs_join_transaction_nolock(root
);
1252 trans
= btrfs_join_transaction(root
);
1254 if (IS_ERR(trans
)) {
1255 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1256 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1257 EXTENT_DO_ACCOUNTING
|
1258 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1260 PAGE_SET_WRITEBACK
|
1261 PAGE_END_WRITEBACK
);
1262 btrfs_free_path(path
);
1263 return PTR_ERR(trans
);
1266 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1268 cow_start
= (u64
)-1;
1271 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1275 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1276 leaf
= path
->nodes
[0];
1277 btrfs_item_key_to_cpu(leaf
, &found_key
,
1278 path
->slots
[0] - 1);
1279 if (found_key
.objectid
== ino
&&
1280 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1285 leaf
= path
->nodes
[0];
1286 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1287 ret
= btrfs_next_leaf(root
, path
);
1292 leaf
= path
->nodes
[0];
1298 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1300 if (found_key
.objectid
> ino
||
1301 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1302 found_key
.offset
> end
)
1305 if (found_key
.offset
> cur_offset
) {
1306 extent_end
= found_key
.offset
;
1311 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1312 struct btrfs_file_extent_item
);
1313 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1315 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1316 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1317 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1318 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1319 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1320 extent_end
= found_key
.offset
+
1321 btrfs_file_extent_num_bytes(leaf
, fi
);
1323 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1324 if (extent_end
<= start
) {
1328 if (disk_bytenr
== 0)
1330 if (btrfs_file_extent_compression(leaf
, fi
) ||
1331 btrfs_file_extent_encryption(leaf
, fi
) ||
1332 btrfs_file_extent_other_encoding(leaf
, fi
))
1334 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1336 if (btrfs_extent_readonly(root
, disk_bytenr
))
1338 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1340 extent_offset
, disk_bytenr
))
1342 disk_bytenr
+= extent_offset
;
1343 disk_bytenr
+= cur_offset
- found_key
.offset
;
1344 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1346 * if there are pending snapshots for this root,
1347 * we fall into common COW way.
1350 err
= btrfs_start_write_no_snapshoting(root
);
1355 * force cow if csum exists in the range.
1356 * this ensure that csum for a given extent are
1357 * either valid or do not exist.
1359 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1362 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1363 extent_end
= found_key
.offset
+
1364 btrfs_file_extent_inline_len(leaf
,
1365 path
->slots
[0], fi
);
1366 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1371 if (extent_end
<= start
) {
1373 if (!nolock
&& nocow
)
1374 btrfs_end_write_no_snapshoting(root
);
1378 if (cow_start
== (u64
)-1)
1379 cow_start
= cur_offset
;
1380 cur_offset
= extent_end
;
1381 if (cur_offset
> end
)
1387 btrfs_release_path(path
);
1388 if (cow_start
!= (u64
)-1) {
1389 ret
= cow_file_range(inode
, locked_page
,
1390 cow_start
, found_key
.offset
- 1,
1391 page_started
, nr_written
, 1);
1393 if (!nolock
&& nocow
)
1394 btrfs_end_write_no_snapshoting(root
);
1397 cow_start
= (u64
)-1;
1400 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1401 struct extent_map
*em
;
1402 struct extent_map_tree
*em_tree
;
1403 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1404 em
= alloc_extent_map();
1405 BUG_ON(!em
); /* -ENOMEM */
1406 em
->start
= cur_offset
;
1407 em
->orig_start
= found_key
.offset
- extent_offset
;
1408 em
->len
= num_bytes
;
1409 em
->block_len
= num_bytes
;
1410 em
->block_start
= disk_bytenr
;
1411 em
->orig_block_len
= disk_num_bytes
;
1412 em
->ram_bytes
= ram_bytes
;
1413 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1414 em
->mod_start
= em
->start
;
1415 em
->mod_len
= em
->len
;
1416 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1417 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1418 em
->generation
= -1;
1420 write_lock(&em_tree
->lock
);
1421 ret
= add_extent_mapping(em_tree
, em
, 1);
1422 write_unlock(&em_tree
->lock
);
1423 if (ret
!= -EEXIST
) {
1424 free_extent_map(em
);
1427 btrfs_drop_extent_cache(inode
, em
->start
,
1428 em
->start
+ em
->len
- 1, 0);
1430 type
= BTRFS_ORDERED_PREALLOC
;
1432 type
= BTRFS_ORDERED_NOCOW
;
1435 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1436 num_bytes
, num_bytes
, type
);
1437 BUG_ON(ret
); /* -ENOMEM */
1439 if (root
->root_key
.objectid
==
1440 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1441 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1444 if (!nolock
&& nocow
)
1445 btrfs_end_write_no_snapshoting(root
);
1450 extent_clear_unlock_delalloc(inode
, cur_offset
,
1451 cur_offset
+ num_bytes
- 1,
1452 locked_page
, EXTENT_LOCKED
|
1453 EXTENT_DELALLOC
, PAGE_UNLOCK
|
1455 if (!nolock
&& nocow
)
1456 btrfs_end_write_no_snapshoting(root
);
1457 cur_offset
= extent_end
;
1458 if (cur_offset
> end
)
1461 btrfs_release_path(path
);
1463 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1464 cow_start
= cur_offset
;
1468 if (cow_start
!= (u64
)-1) {
1469 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1470 page_started
, nr_written
, 1);
1476 err
= btrfs_end_transaction(trans
, root
);
1480 if (ret
&& cur_offset
< end
)
1481 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1482 locked_page
, EXTENT_LOCKED
|
1483 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1484 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1486 PAGE_SET_WRITEBACK
|
1487 PAGE_END_WRITEBACK
);
1488 btrfs_free_path(path
);
1492 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1495 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1496 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1500 * @defrag_bytes is a hint value, no spinlock held here,
1501 * if is not zero, it means the file is defragging.
1502 * Force cow if given extent needs to be defragged.
1504 if (BTRFS_I(inode
)->defrag_bytes
&&
1505 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1506 EXTENT_DEFRAG
, 0, NULL
))
1513 * extent_io.c call back to do delayed allocation processing
1515 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1516 u64 start
, u64 end
, int *page_started
,
1517 unsigned long *nr_written
)
1520 int force_cow
= need_force_cow(inode
, start
, end
);
1522 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1523 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1524 page_started
, 1, nr_written
);
1525 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1526 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1527 page_started
, 0, nr_written
);
1528 } else if (!inode_need_compress(inode
)) {
1529 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1530 page_started
, nr_written
, 1);
1532 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1533 &BTRFS_I(inode
)->runtime_flags
);
1534 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1535 page_started
, nr_written
);
1540 static void btrfs_split_extent_hook(struct inode
*inode
,
1541 struct extent_state
*orig
, u64 split
)
1545 /* not delalloc, ignore it */
1546 if (!(orig
->state
& EXTENT_DELALLOC
))
1549 size
= orig
->end
- orig
->start
+ 1;
1550 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1555 * See the explanation in btrfs_merge_extent_hook, the same
1556 * applies here, just in reverse.
1558 new_size
= orig
->end
- split
+ 1;
1559 num_extents
= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1560 BTRFS_MAX_EXTENT_SIZE
);
1561 new_size
= split
- orig
->start
;
1562 num_extents
+= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1563 BTRFS_MAX_EXTENT_SIZE
);
1564 if (div64_u64(size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1565 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1569 spin_lock(&BTRFS_I(inode
)->lock
);
1570 BTRFS_I(inode
)->outstanding_extents
++;
1571 spin_unlock(&BTRFS_I(inode
)->lock
);
1575 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1576 * extents so we can keep track of new extents that are just merged onto old
1577 * extents, such as when we are doing sequential writes, so we can properly
1578 * account for the metadata space we'll need.
1580 static void btrfs_merge_extent_hook(struct inode
*inode
,
1581 struct extent_state
*new,
1582 struct extent_state
*other
)
1584 u64 new_size
, old_size
;
1587 /* not delalloc, ignore it */
1588 if (!(other
->state
& EXTENT_DELALLOC
))
1591 if (new->start
> other
->start
)
1592 new_size
= new->end
- other
->start
+ 1;
1594 new_size
= other
->end
- new->start
+ 1;
1596 /* we're not bigger than the max, unreserve the space and go */
1597 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1598 spin_lock(&BTRFS_I(inode
)->lock
);
1599 BTRFS_I(inode
)->outstanding_extents
--;
1600 spin_unlock(&BTRFS_I(inode
)->lock
);
1605 * We have to add up either side to figure out how many extents were
1606 * accounted for before we merged into one big extent. If the number of
1607 * extents we accounted for is <= the amount we need for the new range
1608 * then we can return, otherwise drop. Think of it like this
1612 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1613 * need 2 outstanding extents, on one side we have 1 and the other side
1614 * we have 1 so they are == and we can return. But in this case
1616 * [MAX_SIZE+4k][MAX_SIZE+4k]
1618 * Each range on their own accounts for 2 extents, but merged together
1619 * they are only 3 extents worth of accounting, so we need to drop in
1622 old_size
= other
->end
- other
->start
+ 1;
1623 num_extents
= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1624 BTRFS_MAX_EXTENT_SIZE
);
1625 old_size
= new->end
- new->start
+ 1;
1626 num_extents
+= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1627 BTRFS_MAX_EXTENT_SIZE
);
1629 if (div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1630 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1633 spin_lock(&BTRFS_I(inode
)->lock
);
1634 BTRFS_I(inode
)->outstanding_extents
--;
1635 spin_unlock(&BTRFS_I(inode
)->lock
);
1638 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1639 struct inode
*inode
)
1641 spin_lock(&root
->delalloc_lock
);
1642 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1643 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1644 &root
->delalloc_inodes
);
1645 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1646 &BTRFS_I(inode
)->runtime_flags
);
1647 root
->nr_delalloc_inodes
++;
1648 if (root
->nr_delalloc_inodes
== 1) {
1649 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1650 BUG_ON(!list_empty(&root
->delalloc_root
));
1651 list_add_tail(&root
->delalloc_root
,
1652 &root
->fs_info
->delalloc_roots
);
1653 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1656 spin_unlock(&root
->delalloc_lock
);
1659 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1660 struct inode
*inode
)
1662 spin_lock(&root
->delalloc_lock
);
1663 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1664 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1665 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1666 &BTRFS_I(inode
)->runtime_flags
);
1667 root
->nr_delalloc_inodes
--;
1668 if (!root
->nr_delalloc_inodes
) {
1669 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1670 BUG_ON(list_empty(&root
->delalloc_root
));
1671 list_del_init(&root
->delalloc_root
);
1672 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1675 spin_unlock(&root
->delalloc_lock
);
1679 * extent_io.c set_bit_hook, used to track delayed allocation
1680 * bytes in this file, and to maintain the list of inodes that
1681 * have pending delalloc work to be done.
1683 static void btrfs_set_bit_hook(struct inode
*inode
,
1684 struct extent_state
*state
, unsigned *bits
)
1687 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1690 * set_bit and clear bit hooks normally require _irqsave/restore
1691 * but in this case, we are only testing for the DELALLOC
1692 * bit, which is only set or cleared with irqs on
1694 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1695 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1696 u64 len
= state
->end
+ 1 - state
->start
;
1697 bool do_list
= !btrfs_is_free_space_inode(inode
);
1699 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1700 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1702 spin_lock(&BTRFS_I(inode
)->lock
);
1703 BTRFS_I(inode
)->outstanding_extents
++;
1704 spin_unlock(&BTRFS_I(inode
)->lock
);
1707 /* For sanity tests */
1708 if (btrfs_test_is_dummy_root(root
))
1711 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1712 root
->fs_info
->delalloc_batch
);
1713 spin_lock(&BTRFS_I(inode
)->lock
);
1714 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1715 if (*bits
& EXTENT_DEFRAG
)
1716 BTRFS_I(inode
)->defrag_bytes
+= len
;
1717 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1718 &BTRFS_I(inode
)->runtime_flags
))
1719 btrfs_add_delalloc_inodes(root
, inode
);
1720 spin_unlock(&BTRFS_I(inode
)->lock
);
1725 * extent_io.c clear_bit_hook, see set_bit_hook for why
1727 static void btrfs_clear_bit_hook(struct inode
*inode
,
1728 struct extent_state
*state
,
1731 u64 len
= state
->end
+ 1 - state
->start
;
1732 u64 num_extents
= div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
-1,
1733 BTRFS_MAX_EXTENT_SIZE
);
1735 spin_lock(&BTRFS_I(inode
)->lock
);
1736 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
))
1737 BTRFS_I(inode
)->defrag_bytes
-= len
;
1738 spin_unlock(&BTRFS_I(inode
)->lock
);
1741 * set_bit and clear bit hooks normally require _irqsave/restore
1742 * but in this case, we are only testing for the DELALLOC
1743 * bit, which is only set or cleared with irqs on
1745 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1746 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1747 bool do_list
= !btrfs_is_free_space_inode(inode
);
1749 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1750 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1751 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1752 spin_lock(&BTRFS_I(inode
)->lock
);
1753 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
1754 spin_unlock(&BTRFS_I(inode
)->lock
);
1758 * We don't reserve metadata space for space cache inodes so we
1759 * don't need to call dellalloc_release_metadata if there is an
1762 if (*bits
& EXTENT_DO_ACCOUNTING
&&
1763 root
!= root
->fs_info
->tree_root
)
1764 btrfs_delalloc_release_metadata(inode
, len
);
1766 /* For sanity tests. */
1767 if (btrfs_test_is_dummy_root(root
))
1770 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1771 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1772 btrfs_free_reserved_data_space(inode
, len
);
1774 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1775 root
->fs_info
->delalloc_batch
);
1776 spin_lock(&BTRFS_I(inode
)->lock
);
1777 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1778 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1779 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1780 &BTRFS_I(inode
)->runtime_flags
))
1781 btrfs_del_delalloc_inode(root
, inode
);
1782 spin_unlock(&BTRFS_I(inode
)->lock
);
1787 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1788 * we don't create bios that span stripes or chunks
1790 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1791 size_t size
, struct bio
*bio
,
1792 unsigned long bio_flags
)
1794 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1795 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1800 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1803 length
= bio
->bi_iter
.bi_size
;
1804 map_length
= length
;
1805 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1806 &map_length
, NULL
, 0);
1807 /* Will always return 0 with map_multi == NULL */
1809 if (map_length
< length
+ size
)
1815 * in order to insert checksums into the metadata in large chunks,
1816 * we wait until bio submission time. All the pages in the bio are
1817 * checksummed and sums are attached onto the ordered extent record.
1819 * At IO completion time the cums attached on the ordered extent record
1820 * are inserted into the btree
1822 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1823 struct bio
*bio
, int mirror_num
,
1824 unsigned long bio_flags
,
1827 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1830 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1831 BUG_ON(ret
); /* -ENOMEM */
1836 * in order to insert checksums into the metadata in large chunks,
1837 * we wait until bio submission time. All the pages in the bio are
1838 * checksummed and sums are attached onto the ordered extent record.
1840 * At IO completion time the cums attached on the ordered extent record
1841 * are inserted into the btree
1843 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1844 int mirror_num
, unsigned long bio_flags
,
1847 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1850 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1852 bio
->bi_error
= ret
;
1859 * extent_io.c submission hook. This does the right thing for csum calculation
1860 * on write, or reading the csums from the tree before a read
1862 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1863 int mirror_num
, unsigned long bio_flags
,
1866 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1867 enum btrfs_wq_endio_type metadata
= BTRFS_WQ_ENDIO_DATA
;
1870 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1872 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1874 if (btrfs_is_free_space_inode(inode
))
1875 metadata
= BTRFS_WQ_ENDIO_FREE_SPACE
;
1877 if (!(rw
& REQ_WRITE
)) {
1878 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1882 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1883 ret
= btrfs_submit_compressed_read(inode
, bio
,
1887 } else if (!skip_sum
) {
1888 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1893 } else if (async
&& !skip_sum
) {
1894 /* csum items have already been cloned */
1895 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1897 /* we're doing a write, do the async checksumming */
1898 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1899 inode
, rw
, bio
, mirror_num
,
1900 bio_flags
, bio_offset
,
1901 __btrfs_submit_bio_start
,
1902 __btrfs_submit_bio_done
);
1904 } else if (!skip_sum
) {
1905 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1911 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1915 bio
->bi_error
= ret
;
1922 * given a list of ordered sums record them in the inode. This happens
1923 * at IO completion time based on sums calculated at bio submission time.
1925 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1926 struct inode
*inode
, u64 file_offset
,
1927 struct list_head
*list
)
1929 struct btrfs_ordered_sum
*sum
;
1931 list_for_each_entry(sum
, list
, list
) {
1932 trans
->adding_csums
= 1;
1933 btrfs_csum_file_blocks(trans
,
1934 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1935 trans
->adding_csums
= 0;
1940 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1941 struct extent_state
**cached_state
)
1943 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1944 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1945 cached_state
, GFP_NOFS
);
1948 /* see btrfs_writepage_start_hook for details on why this is required */
1949 struct btrfs_writepage_fixup
{
1951 struct btrfs_work work
;
1954 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1956 struct btrfs_writepage_fixup
*fixup
;
1957 struct btrfs_ordered_extent
*ordered
;
1958 struct extent_state
*cached_state
= NULL
;
1960 struct inode
*inode
;
1965 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1969 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1970 ClearPageChecked(page
);
1974 inode
= page
->mapping
->host
;
1975 page_start
= page_offset(page
);
1976 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1978 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1981 /* already ordered? We're done */
1982 if (PagePrivate2(page
))
1985 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1987 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1988 page_end
, &cached_state
, GFP_NOFS
);
1990 btrfs_start_ordered_extent(inode
, ordered
, 1);
1991 btrfs_put_ordered_extent(ordered
);
1995 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1997 mapping_set_error(page
->mapping
, ret
);
1998 end_extent_writepage(page
, ret
, page_start
, page_end
);
1999 ClearPageChecked(page
);
2003 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
2004 ClearPageChecked(page
);
2005 set_page_dirty(page
);
2007 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2008 &cached_state
, GFP_NOFS
);
2011 page_cache_release(page
);
2016 * There are a few paths in the higher layers of the kernel that directly
2017 * set the page dirty bit without asking the filesystem if it is a
2018 * good idea. This causes problems because we want to make sure COW
2019 * properly happens and the data=ordered rules are followed.
2021 * In our case any range that doesn't have the ORDERED bit set
2022 * hasn't been properly setup for IO. We kick off an async process
2023 * to fix it up. The async helper will wait for ordered extents, set
2024 * the delalloc bit and make it safe to write the page.
2026 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2028 struct inode
*inode
= page
->mapping
->host
;
2029 struct btrfs_writepage_fixup
*fixup
;
2030 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2032 /* this page is properly in the ordered list */
2033 if (TestClearPagePrivate2(page
))
2036 if (PageChecked(page
))
2039 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2043 SetPageChecked(page
);
2044 page_cache_get(page
);
2045 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2046 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2048 btrfs_queue_work(root
->fs_info
->fixup_workers
, &fixup
->work
);
2052 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2053 struct inode
*inode
, u64 file_pos
,
2054 u64 disk_bytenr
, u64 disk_num_bytes
,
2055 u64 num_bytes
, u64 ram_bytes
,
2056 u8 compression
, u8 encryption
,
2057 u16 other_encoding
, int extent_type
)
2059 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2060 struct btrfs_file_extent_item
*fi
;
2061 struct btrfs_path
*path
;
2062 struct extent_buffer
*leaf
;
2063 struct btrfs_key ins
;
2064 int extent_inserted
= 0;
2067 path
= btrfs_alloc_path();
2072 * we may be replacing one extent in the tree with another.
2073 * The new extent is pinned in the extent map, and we don't want
2074 * to drop it from the cache until it is completely in the btree.
2076 * So, tell btrfs_drop_extents to leave this extent in the cache.
2077 * the caller is expected to unpin it and allow it to be merged
2080 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2081 file_pos
+ num_bytes
, NULL
, 0,
2082 1, sizeof(*fi
), &extent_inserted
);
2086 if (!extent_inserted
) {
2087 ins
.objectid
= btrfs_ino(inode
);
2088 ins
.offset
= file_pos
;
2089 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2091 path
->leave_spinning
= 1;
2092 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2097 leaf
= path
->nodes
[0];
2098 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2099 struct btrfs_file_extent_item
);
2100 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2101 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2102 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2103 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2104 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2105 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2106 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2107 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2108 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2109 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2111 btrfs_mark_buffer_dirty(leaf
);
2112 btrfs_release_path(path
);
2114 inode_add_bytes(inode
, num_bytes
);
2116 ins
.objectid
= disk_bytenr
;
2117 ins
.offset
= disk_num_bytes
;
2118 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2119 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2120 root
->root_key
.objectid
,
2121 btrfs_ino(inode
), file_pos
, &ins
);
2123 btrfs_free_path(path
);
2128 /* snapshot-aware defrag */
2129 struct sa_defrag_extent_backref
{
2130 struct rb_node node
;
2131 struct old_sa_defrag_extent
*old
;
2140 struct old_sa_defrag_extent
{
2141 struct list_head list
;
2142 struct new_sa_defrag_extent
*new;
2151 struct new_sa_defrag_extent
{
2152 struct rb_root root
;
2153 struct list_head head
;
2154 struct btrfs_path
*path
;
2155 struct inode
*inode
;
2163 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2164 struct sa_defrag_extent_backref
*b2
)
2166 if (b1
->root_id
< b2
->root_id
)
2168 else if (b1
->root_id
> b2
->root_id
)
2171 if (b1
->inum
< b2
->inum
)
2173 else if (b1
->inum
> b2
->inum
)
2176 if (b1
->file_pos
< b2
->file_pos
)
2178 else if (b1
->file_pos
> b2
->file_pos
)
2182 * [------------------------------] ===> (a range of space)
2183 * |<--->| |<---->| =============> (fs/file tree A)
2184 * |<---------------------------->| ===> (fs/file tree B)
2186 * A range of space can refer to two file extents in one tree while
2187 * refer to only one file extent in another tree.
2189 * So we may process a disk offset more than one time(two extents in A)
2190 * and locate at the same extent(one extent in B), then insert two same
2191 * backrefs(both refer to the extent in B).
2196 static void backref_insert(struct rb_root
*root
,
2197 struct sa_defrag_extent_backref
*backref
)
2199 struct rb_node
**p
= &root
->rb_node
;
2200 struct rb_node
*parent
= NULL
;
2201 struct sa_defrag_extent_backref
*entry
;
2206 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2208 ret
= backref_comp(backref
, entry
);
2212 p
= &(*p
)->rb_right
;
2215 rb_link_node(&backref
->node
, parent
, p
);
2216 rb_insert_color(&backref
->node
, root
);
2220 * Note the backref might has changed, and in this case we just return 0.
2222 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2225 struct btrfs_file_extent_item
*extent
;
2226 struct btrfs_fs_info
*fs_info
;
2227 struct old_sa_defrag_extent
*old
= ctx
;
2228 struct new_sa_defrag_extent
*new = old
->new;
2229 struct btrfs_path
*path
= new->path
;
2230 struct btrfs_key key
;
2231 struct btrfs_root
*root
;
2232 struct sa_defrag_extent_backref
*backref
;
2233 struct extent_buffer
*leaf
;
2234 struct inode
*inode
= new->inode
;
2240 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2241 inum
== btrfs_ino(inode
))
2244 key
.objectid
= root_id
;
2245 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2246 key
.offset
= (u64
)-1;
2248 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2249 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2251 if (PTR_ERR(root
) == -ENOENT
)
2254 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2255 inum
, offset
, root_id
);
2256 return PTR_ERR(root
);
2259 key
.objectid
= inum
;
2260 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2261 if (offset
> (u64
)-1 << 32)
2264 key
.offset
= offset
;
2266 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2267 if (WARN_ON(ret
< 0))
2274 leaf
= path
->nodes
[0];
2275 slot
= path
->slots
[0];
2277 if (slot
>= btrfs_header_nritems(leaf
)) {
2278 ret
= btrfs_next_leaf(root
, path
);
2281 } else if (ret
> 0) {
2290 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2292 if (key
.objectid
> inum
)
2295 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2298 extent
= btrfs_item_ptr(leaf
, slot
,
2299 struct btrfs_file_extent_item
);
2301 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2305 * 'offset' refers to the exact key.offset,
2306 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2307 * (key.offset - extent_offset).
2309 if (key
.offset
!= offset
)
2312 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2313 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2315 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2316 old
->len
|| extent_offset
+ num_bytes
<=
2317 old
->extent_offset
+ old
->offset
)
2322 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2328 backref
->root_id
= root_id
;
2329 backref
->inum
= inum
;
2330 backref
->file_pos
= offset
;
2331 backref
->num_bytes
= num_bytes
;
2332 backref
->extent_offset
= extent_offset
;
2333 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2335 backref_insert(&new->root
, backref
);
2338 btrfs_release_path(path
);
2343 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2344 struct new_sa_defrag_extent
*new)
2346 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2347 struct old_sa_defrag_extent
*old
, *tmp
;
2352 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2353 ret
= iterate_inodes_from_logical(old
->bytenr
+
2354 old
->extent_offset
, fs_info
,
2355 path
, record_one_backref
,
2357 if (ret
< 0 && ret
!= -ENOENT
)
2360 /* no backref to be processed for this extent */
2362 list_del(&old
->list
);
2367 if (list_empty(&new->head
))
2373 static int relink_is_mergable(struct extent_buffer
*leaf
,
2374 struct btrfs_file_extent_item
*fi
,
2375 struct new_sa_defrag_extent
*new)
2377 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2380 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2383 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2386 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2387 btrfs_file_extent_other_encoding(leaf
, fi
))
2394 * Note the backref might has changed, and in this case we just return 0.
2396 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2397 struct sa_defrag_extent_backref
*prev
,
2398 struct sa_defrag_extent_backref
*backref
)
2400 struct btrfs_file_extent_item
*extent
;
2401 struct btrfs_file_extent_item
*item
;
2402 struct btrfs_ordered_extent
*ordered
;
2403 struct btrfs_trans_handle
*trans
;
2404 struct btrfs_fs_info
*fs_info
;
2405 struct btrfs_root
*root
;
2406 struct btrfs_key key
;
2407 struct extent_buffer
*leaf
;
2408 struct old_sa_defrag_extent
*old
= backref
->old
;
2409 struct new_sa_defrag_extent
*new = old
->new;
2410 struct inode
*src_inode
= new->inode
;
2411 struct inode
*inode
;
2412 struct extent_state
*cached
= NULL
;
2421 if (prev
&& prev
->root_id
== backref
->root_id
&&
2422 prev
->inum
== backref
->inum
&&
2423 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2426 /* step 1: get root */
2427 key
.objectid
= backref
->root_id
;
2428 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2429 key
.offset
= (u64
)-1;
2431 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2432 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2434 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2436 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2437 if (PTR_ERR(root
) == -ENOENT
)
2439 return PTR_ERR(root
);
2442 if (btrfs_root_readonly(root
)) {
2443 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2447 /* step 2: get inode */
2448 key
.objectid
= backref
->inum
;
2449 key
.type
= BTRFS_INODE_ITEM_KEY
;
2452 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2453 if (IS_ERR(inode
)) {
2454 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2458 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2460 /* step 3: relink backref */
2461 lock_start
= backref
->file_pos
;
2462 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2463 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2466 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2468 btrfs_put_ordered_extent(ordered
);
2472 trans
= btrfs_join_transaction(root
);
2473 if (IS_ERR(trans
)) {
2474 ret
= PTR_ERR(trans
);
2478 key
.objectid
= backref
->inum
;
2479 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2480 key
.offset
= backref
->file_pos
;
2482 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2485 } else if (ret
> 0) {
2490 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2491 struct btrfs_file_extent_item
);
2493 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2494 backref
->generation
)
2497 btrfs_release_path(path
);
2499 start
= backref
->file_pos
;
2500 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2501 start
+= old
->extent_offset
+ old
->offset
-
2502 backref
->extent_offset
;
2504 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2505 old
->extent_offset
+ old
->offset
+ old
->len
);
2506 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2508 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2513 key
.objectid
= btrfs_ino(inode
);
2514 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2517 path
->leave_spinning
= 1;
2519 struct btrfs_file_extent_item
*fi
;
2521 struct btrfs_key found_key
;
2523 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2528 leaf
= path
->nodes
[0];
2529 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2531 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2532 struct btrfs_file_extent_item
);
2533 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2535 if (extent_len
+ found_key
.offset
== start
&&
2536 relink_is_mergable(leaf
, fi
, new)) {
2537 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2539 btrfs_mark_buffer_dirty(leaf
);
2540 inode_add_bytes(inode
, len
);
2546 btrfs_release_path(path
);
2551 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2554 btrfs_abort_transaction(trans
, root
, ret
);
2558 leaf
= path
->nodes
[0];
2559 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2560 struct btrfs_file_extent_item
);
2561 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2562 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2563 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2564 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2565 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2566 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2567 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2568 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2569 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2570 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2572 btrfs_mark_buffer_dirty(leaf
);
2573 inode_add_bytes(inode
, len
);
2574 btrfs_release_path(path
);
2576 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2578 backref
->root_id
, backref
->inum
,
2579 new->file_pos
, 0); /* start - extent_offset */
2581 btrfs_abort_transaction(trans
, root
, ret
);
2587 btrfs_release_path(path
);
2588 path
->leave_spinning
= 0;
2589 btrfs_end_transaction(trans
, root
);
2591 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2597 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2599 struct old_sa_defrag_extent
*old
, *tmp
;
2604 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2610 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2612 struct btrfs_path
*path
;
2613 struct sa_defrag_extent_backref
*backref
;
2614 struct sa_defrag_extent_backref
*prev
= NULL
;
2615 struct inode
*inode
;
2616 struct btrfs_root
*root
;
2617 struct rb_node
*node
;
2621 root
= BTRFS_I(inode
)->root
;
2623 path
= btrfs_alloc_path();
2627 if (!record_extent_backrefs(path
, new)) {
2628 btrfs_free_path(path
);
2631 btrfs_release_path(path
);
2634 node
= rb_first(&new->root
);
2637 rb_erase(node
, &new->root
);
2639 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2641 ret
= relink_extent_backref(path
, prev
, backref
);
2654 btrfs_free_path(path
);
2656 free_sa_defrag_extent(new);
2658 atomic_dec(&root
->fs_info
->defrag_running
);
2659 wake_up(&root
->fs_info
->transaction_wait
);
2662 static struct new_sa_defrag_extent
*
2663 record_old_file_extents(struct inode
*inode
,
2664 struct btrfs_ordered_extent
*ordered
)
2666 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2667 struct btrfs_path
*path
;
2668 struct btrfs_key key
;
2669 struct old_sa_defrag_extent
*old
;
2670 struct new_sa_defrag_extent
*new;
2673 new = kmalloc(sizeof(*new), GFP_NOFS
);
2678 new->file_pos
= ordered
->file_offset
;
2679 new->len
= ordered
->len
;
2680 new->bytenr
= ordered
->start
;
2681 new->disk_len
= ordered
->disk_len
;
2682 new->compress_type
= ordered
->compress_type
;
2683 new->root
= RB_ROOT
;
2684 INIT_LIST_HEAD(&new->head
);
2686 path
= btrfs_alloc_path();
2690 key
.objectid
= btrfs_ino(inode
);
2691 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2692 key
.offset
= new->file_pos
;
2694 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2697 if (ret
> 0 && path
->slots
[0] > 0)
2700 /* find out all the old extents for the file range */
2702 struct btrfs_file_extent_item
*extent
;
2703 struct extent_buffer
*l
;
2712 slot
= path
->slots
[0];
2714 if (slot
>= btrfs_header_nritems(l
)) {
2715 ret
= btrfs_next_leaf(root
, path
);
2723 btrfs_item_key_to_cpu(l
, &key
, slot
);
2725 if (key
.objectid
!= btrfs_ino(inode
))
2727 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2729 if (key
.offset
>= new->file_pos
+ new->len
)
2732 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2734 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2735 if (key
.offset
+ num_bytes
< new->file_pos
)
2738 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2742 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2744 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2748 offset
= max(new->file_pos
, key
.offset
);
2749 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2751 old
->bytenr
= disk_bytenr
;
2752 old
->extent_offset
= extent_offset
;
2753 old
->offset
= offset
- key
.offset
;
2754 old
->len
= end
- offset
;
2757 list_add_tail(&old
->list
, &new->head
);
2763 btrfs_free_path(path
);
2764 atomic_inc(&root
->fs_info
->defrag_running
);
2769 btrfs_free_path(path
);
2771 free_sa_defrag_extent(new);
2775 static void btrfs_release_delalloc_bytes(struct btrfs_root
*root
,
2778 struct btrfs_block_group_cache
*cache
;
2780 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2783 spin_lock(&cache
->lock
);
2784 cache
->delalloc_bytes
-= len
;
2785 spin_unlock(&cache
->lock
);
2787 btrfs_put_block_group(cache
);
2790 /* as ordered data IO finishes, this gets called so we can finish
2791 * an ordered extent if the range of bytes in the file it covers are
2794 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2796 struct inode
*inode
= ordered_extent
->inode
;
2797 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2798 struct btrfs_trans_handle
*trans
= NULL
;
2799 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2800 struct extent_state
*cached_state
= NULL
;
2801 struct new_sa_defrag_extent
*new = NULL
;
2802 int compress_type
= 0;
2804 u64 logical_len
= ordered_extent
->len
;
2806 bool truncated
= false;
2808 nolock
= btrfs_is_free_space_inode(inode
);
2810 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2815 btrfs_free_io_failure_record(inode
, ordered_extent
->file_offset
,
2816 ordered_extent
->file_offset
+
2817 ordered_extent
->len
- 1);
2819 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2821 logical_len
= ordered_extent
->truncated_len
;
2822 /* Truncated the entire extent, don't bother adding */
2827 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2828 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2829 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2831 trans
= btrfs_join_transaction_nolock(root
);
2833 trans
= btrfs_join_transaction(root
);
2834 if (IS_ERR(trans
)) {
2835 ret
= PTR_ERR(trans
);
2839 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2840 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2841 if (ret
) /* -ENOMEM or corruption */
2842 btrfs_abort_transaction(trans
, root
, ret
);
2846 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2847 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2850 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2851 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2852 EXTENT_DEFRAG
, 1, cached_state
);
2854 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2855 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2856 /* the inode is shared */
2857 new = record_old_file_extents(inode
, ordered_extent
);
2859 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2860 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2861 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2865 trans
= btrfs_join_transaction_nolock(root
);
2867 trans
= btrfs_join_transaction(root
);
2868 if (IS_ERR(trans
)) {
2869 ret
= PTR_ERR(trans
);
2874 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2876 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2877 compress_type
= ordered_extent
->compress_type
;
2878 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2879 BUG_ON(compress_type
);
2880 ret
= btrfs_mark_extent_written(trans
, inode
,
2881 ordered_extent
->file_offset
,
2882 ordered_extent
->file_offset
+
2885 BUG_ON(root
== root
->fs_info
->tree_root
);
2886 ret
= insert_reserved_file_extent(trans
, inode
,
2887 ordered_extent
->file_offset
,
2888 ordered_extent
->start
,
2889 ordered_extent
->disk_len
,
2890 logical_len
, logical_len
,
2891 compress_type
, 0, 0,
2892 BTRFS_FILE_EXTENT_REG
);
2894 btrfs_release_delalloc_bytes(root
,
2895 ordered_extent
->start
,
2896 ordered_extent
->disk_len
);
2898 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2899 ordered_extent
->file_offset
, ordered_extent
->len
,
2902 btrfs_abort_transaction(trans
, root
, ret
);
2906 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2907 &ordered_extent
->list
);
2909 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2910 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2911 if (ret
) { /* -ENOMEM or corruption */
2912 btrfs_abort_transaction(trans
, root
, ret
);
2917 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2918 ordered_extent
->file_offset
+
2919 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2921 if (root
!= root
->fs_info
->tree_root
)
2922 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2924 btrfs_end_transaction(trans
, root
);
2926 if (ret
|| truncated
) {
2930 start
= ordered_extent
->file_offset
+ logical_len
;
2932 start
= ordered_extent
->file_offset
;
2933 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
2934 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
2936 /* Drop the cache for the part of the extent we didn't write. */
2937 btrfs_drop_extent_cache(inode
, start
, end
, 0);
2940 * If the ordered extent had an IOERR or something else went
2941 * wrong we need to return the space for this ordered extent
2942 * back to the allocator. We only free the extent in the
2943 * truncated case if we didn't write out the extent at all.
2945 if ((ret
|| !logical_len
) &&
2946 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2947 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2948 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
2949 ordered_extent
->disk_len
, 1);
2954 * This needs to be done to make sure anybody waiting knows we are done
2955 * updating everything for this ordered extent.
2957 btrfs_remove_ordered_extent(inode
, ordered_extent
);
2959 /* for snapshot-aware defrag */
2962 free_sa_defrag_extent(new);
2963 atomic_dec(&root
->fs_info
->defrag_running
);
2965 relink_file_extents(new);
2970 btrfs_put_ordered_extent(ordered_extent
);
2971 /* once for the tree */
2972 btrfs_put_ordered_extent(ordered_extent
);
2977 static void finish_ordered_fn(struct btrfs_work
*work
)
2979 struct btrfs_ordered_extent
*ordered_extent
;
2980 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
2981 btrfs_finish_ordered_io(ordered_extent
);
2984 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2985 struct extent_state
*state
, int uptodate
)
2987 struct inode
*inode
= page
->mapping
->host
;
2988 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2989 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
2990 struct btrfs_workqueue
*wq
;
2991 btrfs_work_func_t func
;
2993 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
2995 ClearPagePrivate2(page
);
2996 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
2997 end
- start
+ 1, uptodate
))
3000 if (btrfs_is_free_space_inode(inode
)) {
3001 wq
= root
->fs_info
->endio_freespace_worker
;
3002 func
= btrfs_freespace_write_helper
;
3004 wq
= root
->fs_info
->endio_write_workers
;
3005 func
= btrfs_endio_write_helper
;
3008 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3010 btrfs_queue_work(wq
, &ordered_extent
->work
);
3015 static int __readpage_endio_check(struct inode
*inode
,
3016 struct btrfs_io_bio
*io_bio
,
3017 int icsum
, struct page
*page
,
3018 int pgoff
, u64 start
, size_t len
)
3024 csum_expected
= *(((u32
*)io_bio
->csum
) + icsum
);
3026 kaddr
= kmap_atomic(page
);
3027 csum
= btrfs_csum_data(kaddr
+ pgoff
, csum
, len
);
3028 btrfs_csum_final(csum
, (char *)&csum
);
3029 if (csum
!= csum_expected
)
3032 kunmap_atomic(kaddr
);
3035 btrfs_warn_rl(BTRFS_I(inode
)->root
->fs_info
,
3036 "csum failed ino %llu off %llu csum %u expected csum %u",
3037 btrfs_ino(inode
), start
, csum
, csum_expected
);
3038 memset(kaddr
+ pgoff
, 1, len
);
3039 flush_dcache_page(page
);
3040 kunmap_atomic(kaddr
);
3041 if (csum_expected
== 0)
3047 * when reads are done, we need to check csums to verify the data is correct
3048 * if there's a match, we allow the bio to finish. If not, the code in
3049 * extent_io.c will try to find good copies for us.
3051 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3052 u64 phy_offset
, struct page
*page
,
3053 u64 start
, u64 end
, int mirror
)
3055 size_t offset
= start
- page_offset(page
);
3056 struct inode
*inode
= page
->mapping
->host
;
3057 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3058 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3060 if (PageChecked(page
)) {
3061 ClearPageChecked(page
);
3065 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3068 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3069 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3070 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
3075 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3076 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3077 start
, (size_t)(end
- start
+ 1));
3080 struct delayed_iput
{
3081 struct list_head list
;
3082 struct inode
*inode
;
3085 /* JDM: If this is fs-wide, why can't we add a pointer to
3086 * btrfs_inode instead and avoid the allocation? */
3087 void btrfs_add_delayed_iput(struct inode
*inode
)
3089 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
3090 struct delayed_iput
*delayed
;
3092 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3095 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
3096 delayed
->inode
= inode
;
3098 spin_lock(&fs_info
->delayed_iput_lock
);
3099 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
3100 spin_unlock(&fs_info
->delayed_iput_lock
);
3103 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
3106 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3107 struct delayed_iput
*delayed
;
3110 spin_lock(&fs_info
->delayed_iput_lock
);
3111 empty
= list_empty(&fs_info
->delayed_iputs
);
3112 spin_unlock(&fs_info
->delayed_iput_lock
);
3116 down_read(&fs_info
->delayed_iput_sem
);
3118 spin_lock(&fs_info
->delayed_iput_lock
);
3119 list_splice_init(&fs_info
->delayed_iputs
, &list
);
3120 spin_unlock(&fs_info
->delayed_iput_lock
);
3122 while (!list_empty(&list
)) {
3123 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
3124 list_del(&delayed
->list
);
3125 iput(delayed
->inode
);
3129 up_read(&root
->fs_info
->delayed_iput_sem
);
3133 * This is called in transaction commit time. If there are no orphan
3134 * files in the subvolume, it removes orphan item and frees block_rsv
3137 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3138 struct btrfs_root
*root
)
3140 struct btrfs_block_rsv
*block_rsv
;
3143 if (atomic_read(&root
->orphan_inodes
) ||
3144 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3147 spin_lock(&root
->orphan_lock
);
3148 if (atomic_read(&root
->orphan_inodes
)) {
3149 spin_unlock(&root
->orphan_lock
);
3153 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3154 spin_unlock(&root
->orphan_lock
);
3158 block_rsv
= root
->orphan_block_rsv
;
3159 root
->orphan_block_rsv
= NULL
;
3160 spin_unlock(&root
->orphan_lock
);
3162 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3163 btrfs_root_refs(&root
->root_item
) > 0) {
3164 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
3165 root
->root_key
.objectid
);
3167 btrfs_abort_transaction(trans
, root
, ret
);
3169 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3174 WARN_ON(block_rsv
->size
> 0);
3175 btrfs_free_block_rsv(root
, block_rsv
);
3180 * This creates an orphan entry for the given inode in case something goes
3181 * wrong in the middle of an unlink/truncate.
3183 * NOTE: caller of this function should reserve 5 units of metadata for
3186 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
3188 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3189 struct btrfs_block_rsv
*block_rsv
= NULL
;
3194 if (!root
->orphan_block_rsv
) {
3195 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3200 spin_lock(&root
->orphan_lock
);
3201 if (!root
->orphan_block_rsv
) {
3202 root
->orphan_block_rsv
= block_rsv
;
3203 } else if (block_rsv
) {
3204 btrfs_free_block_rsv(root
, block_rsv
);
3208 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3209 &BTRFS_I(inode
)->runtime_flags
)) {
3212 * For proper ENOSPC handling, we should do orphan
3213 * cleanup when mounting. But this introduces backward
3214 * compatibility issue.
3216 if (!xchg(&root
->orphan_item_inserted
, 1))
3222 atomic_inc(&root
->orphan_inodes
);
3225 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3226 &BTRFS_I(inode
)->runtime_flags
))
3228 spin_unlock(&root
->orphan_lock
);
3230 /* grab metadata reservation from transaction handle */
3232 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3233 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
3236 /* insert an orphan item to track this unlinked/truncated file */
3238 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3240 atomic_dec(&root
->orphan_inodes
);
3242 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3243 &BTRFS_I(inode
)->runtime_flags
);
3244 btrfs_orphan_release_metadata(inode
);
3246 if (ret
!= -EEXIST
) {
3247 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3248 &BTRFS_I(inode
)->runtime_flags
);
3249 btrfs_abort_transaction(trans
, root
, ret
);
3256 /* insert an orphan item to track subvolume contains orphan files */
3258 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3259 root
->root_key
.objectid
);
3260 if (ret
&& ret
!= -EEXIST
) {
3261 btrfs_abort_transaction(trans
, root
, ret
);
3269 * We have done the truncate/delete so we can go ahead and remove the orphan
3270 * item for this particular inode.
3272 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3273 struct inode
*inode
)
3275 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3276 int delete_item
= 0;
3277 int release_rsv
= 0;
3280 spin_lock(&root
->orphan_lock
);
3281 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3282 &BTRFS_I(inode
)->runtime_flags
))
3285 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3286 &BTRFS_I(inode
)->runtime_flags
))
3288 spin_unlock(&root
->orphan_lock
);
3291 atomic_dec(&root
->orphan_inodes
);
3293 ret
= btrfs_del_orphan_item(trans
, root
,
3298 btrfs_orphan_release_metadata(inode
);
3304 * this cleans up any orphans that may be left on the list from the last use
3307 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3309 struct btrfs_path
*path
;
3310 struct extent_buffer
*leaf
;
3311 struct btrfs_key key
, found_key
;
3312 struct btrfs_trans_handle
*trans
;
3313 struct inode
*inode
;
3314 u64 last_objectid
= 0;
3315 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3317 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3320 path
= btrfs_alloc_path();
3327 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3328 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3329 key
.offset
= (u64
)-1;
3332 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3337 * if ret == 0 means we found what we were searching for, which
3338 * is weird, but possible, so only screw with path if we didn't
3339 * find the key and see if we have stuff that matches
3343 if (path
->slots
[0] == 0)
3348 /* pull out the item */
3349 leaf
= path
->nodes
[0];
3350 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3352 /* make sure the item matches what we want */
3353 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3355 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3358 /* release the path since we're done with it */
3359 btrfs_release_path(path
);
3362 * this is where we are basically btrfs_lookup, without the
3363 * crossing root thing. we store the inode number in the
3364 * offset of the orphan item.
3367 if (found_key
.offset
== last_objectid
) {
3368 btrfs_err(root
->fs_info
,
3369 "Error removing orphan entry, stopping orphan cleanup");
3374 last_objectid
= found_key
.offset
;
3376 found_key
.objectid
= found_key
.offset
;
3377 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3378 found_key
.offset
= 0;
3379 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3380 ret
= PTR_ERR_OR_ZERO(inode
);
3381 if (ret
&& ret
!= -ESTALE
)
3384 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3385 struct btrfs_root
*dead_root
;
3386 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3387 int is_dead_root
= 0;
3390 * this is an orphan in the tree root. Currently these
3391 * could come from 2 sources:
3392 * a) a snapshot deletion in progress
3393 * b) a free space cache inode
3394 * We need to distinguish those two, as the snapshot
3395 * orphan must not get deleted.
3396 * find_dead_roots already ran before us, so if this
3397 * is a snapshot deletion, we should find the root
3398 * in the dead_roots list
3400 spin_lock(&fs_info
->trans_lock
);
3401 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3403 if (dead_root
->root_key
.objectid
==
3404 found_key
.objectid
) {
3409 spin_unlock(&fs_info
->trans_lock
);
3411 /* prevent this orphan from being found again */
3412 key
.offset
= found_key
.objectid
- 1;
3417 * Inode is already gone but the orphan item is still there,
3418 * kill the orphan item.
3420 if (ret
== -ESTALE
) {
3421 trans
= btrfs_start_transaction(root
, 1);
3422 if (IS_ERR(trans
)) {
3423 ret
= PTR_ERR(trans
);
3426 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3427 found_key
.objectid
);
3428 ret
= btrfs_del_orphan_item(trans
, root
,
3429 found_key
.objectid
);
3430 btrfs_end_transaction(trans
, root
);
3437 * add this inode to the orphan list so btrfs_orphan_del does
3438 * the proper thing when we hit it
3440 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3441 &BTRFS_I(inode
)->runtime_flags
);
3442 atomic_inc(&root
->orphan_inodes
);
3444 /* if we have links, this was a truncate, lets do that */
3445 if (inode
->i_nlink
) {
3446 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3452 /* 1 for the orphan item deletion. */
3453 trans
= btrfs_start_transaction(root
, 1);
3454 if (IS_ERR(trans
)) {
3456 ret
= PTR_ERR(trans
);
3459 ret
= btrfs_orphan_add(trans
, inode
);
3460 btrfs_end_transaction(trans
, root
);
3466 ret
= btrfs_truncate(inode
);
3468 btrfs_orphan_del(NULL
, inode
);
3473 /* this will do delete_inode and everything for us */
3478 /* release the path since we're done with it */
3479 btrfs_release_path(path
);
3481 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3483 if (root
->orphan_block_rsv
)
3484 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3487 if (root
->orphan_block_rsv
||
3488 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3489 trans
= btrfs_join_transaction(root
);
3491 btrfs_end_transaction(trans
, root
);
3495 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3497 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3501 btrfs_err(root
->fs_info
,
3502 "could not do orphan cleanup %d", ret
);
3503 btrfs_free_path(path
);
3508 * very simple check to peek ahead in the leaf looking for xattrs. If we
3509 * don't find any xattrs, we know there can't be any acls.
3511 * slot is the slot the inode is in, objectid is the objectid of the inode
3513 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3514 int slot
, u64 objectid
,
3515 int *first_xattr_slot
)
3517 u32 nritems
= btrfs_header_nritems(leaf
);
3518 struct btrfs_key found_key
;
3519 static u64 xattr_access
= 0;
3520 static u64 xattr_default
= 0;
3523 if (!xattr_access
) {
3524 xattr_access
= btrfs_name_hash(POSIX_ACL_XATTR_ACCESS
,
3525 strlen(POSIX_ACL_XATTR_ACCESS
));
3526 xattr_default
= btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT
,
3527 strlen(POSIX_ACL_XATTR_DEFAULT
));
3531 *first_xattr_slot
= -1;
3532 while (slot
< nritems
) {
3533 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3535 /* we found a different objectid, there must not be acls */
3536 if (found_key
.objectid
!= objectid
)
3539 /* we found an xattr, assume we've got an acl */
3540 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3541 if (*first_xattr_slot
== -1)
3542 *first_xattr_slot
= slot
;
3543 if (found_key
.offset
== xattr_access
||
3544 found_key
.offset
== xattr_default
)
3549 * we found a key greater than an xattr key, there can't
3550 * be any acls later on
3552 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3559 * it goes inode, inode backrefs, xattrs, extents,
3560 * so if there are a ton of hard links to an inode there can
3561 * be a lot of backrefs. Don't waste time searching too hard,
3562 * this is just an optimization
3567 /* we hit the end of the leaf before we found an xattr or
3568 * something larger than an xattr. We have to assume the inode
3571 if (*first_xattr_slot
== -1)
3572 *first_xattr_slot
= slot
;
3577 * read an inode from the btree into the in-memory inode
3579 static void btrfs_read_locked_inode(struct inode
*inode
)
3581 struct btrfs_path
*path
;
3582 struct extent_buffer
*leaf
;
3583 struct btrfs_inode_item
*inode_item
;
3584 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3585 struct btrfs_key location
;
3590 bool filled
= false;
3591 int first_xattr_slot
;
3593 ret
= btrfs_fill_inode(inode
, &rdev
);
3597 path
= btrfs_alloc_path();
3601 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3603 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3607 leaf
= path
->nodes
[0];
3612 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3613 struct btrfs_inode_item
);
3614 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3615 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3616 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3617 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3618 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3620 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3621 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3623 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3624 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3626 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3627 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3629 BTRFS_I(inode
)->i_otime
.tv_sec
=
3630 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3631 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3632 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3634 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3635 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3636 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3638 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3639 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3641 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3643 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3644 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3648 * If we were modified in the current generation and evicted from memory
3649 * and then re-read we need to do a full sync since we don't have any
3650 * idea about which extents were modified before we were evicted from
3653 * This is required for both inode re-read from disk and delayed inode
3654 * in delayed_nodes_tree.
3656 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3657 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3658 &BTRFS_I(inode
)->runtime_flags
);
3661 * We don't persist the id of the transaction where an unlink operation
3662 * against the inode was last made. So here we assume the inode might
3663 * have been evicted, and therefore the exact value of last_unlink_trans
3664 * lost, and set it to last_trans to avoid metadata inconsistencies
3665 * between the inode and its parent if the inode is fsync'ed and the log
3666 * replayed. For example, in the scenario:
3669 * ln mydir/foo mydir/bar
3672 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3673 * xfs_io -c fsync mydir/foo
3675 * mount fs, triggers fsync log replay
3677 * We must make sure that when we fsync our inode foo we also log its
3678 * parent inode, otherwise after log replay the parent still has the
3679 * dentry with the "bar" name but our inode foo has a link count of 1
3680 * and doesn't have an inode ref with the name "bar" anymore.
3682 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3683 * but it guarantees correctness at the expense of ocassional full
3684 * transaction commits on fsync if our inode is a directory, or if our
3685 * inode is not a directory, logging its parent unnecessarily.
3687 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3690 if (inode
->i_nlink
!= 1 ||
3691 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3694 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3695 if (location
.objectid
!= btrfs_ino(inode
))
3698 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3699 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3700 struct btrfs_inode_ref
*ref
;
3702 ref
= (struct btrfs_inode_ref
*)ptr
;
3703 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3704 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3705 struct btrfs_inode_extref
*extref
;
3707 extref
= (struct btrfs_inode_extref
*)ptr
;
3708 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3713 * try to precache a NULL acl entry for files that don't have
3714 * any xattrs or acls
3716 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3717 btrfs_ino(inode
), &first_xattr_slot
);
3718 if (first_xattr_slot
!= -1) {
3719 path
->slots
[0] = first_xattr_slot
;
3720 ret
= btrfs_load_inode_props(inode
, path
);
3722 btrfs_err(root
->fs_info
,
3723 "error loading props for ino %llu (root %llu): %d",
3725 root
->root_key
.objectid
, ret
);
3727 btrfs_free_path(path
);
3730 cache_no_acl(inode
);
3732 switch (inode
->i_mode
& S_IFMT
) {
3734 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3735 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3736 inode
->i_fop
= &btrfs_file_operations
;
3737 inode
->i_op
= &btrfs_file_inode_operations
;
3740 inode
->i_fop
= &btrfs_dir_file_operations
;
3741 if (root
== root
->fs_info
->tree_root
)
3742 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3744 inode
->i_op
= &btrfs_dir_inode_operations
;
3747 inode
->i_op
= &btrfs_symlink_inode_operations
;
3748 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3751 inode
->i_op
= &btrfs_special_inode_operations
;
3752 init_special_inode(inode
, inode
->i_mode
, rdev
);
3756 btrfs_update_iflags(inode
);
3760 btrfs_free_path(path
);
3761 make_bad_inode(inode
);
3765 * given a leaf and an inode, copy the inode fields into the leaf
3767 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3768 struct extent_buffer
*leaf
,
3769 struct btrfs_inode_item
*item
,
3770 struct inode
*inode
)
3772 struct btrfs_map_token token
;
3774 btrfs_init_map_token(&token
);
3776 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3777 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3778 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3780 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3781 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3783 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3784 inode
->i_atime
.tv_sec
, &token
);
3785 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3786 inode
->i_atime
.tv_nsec
, &token
);
3788 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3789 inode
->i_mtime
.tv_sec
, &token
);
3790 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3791 inode
->i_mtime
.tv_nsec
, &token
);
3793 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3794 inode
->i_ctime
.tv_sec
, &token
);
3795 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3796 inode
->i_ctime
.tv_nsec
, &token
);
3798 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3799 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3800 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3801 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3803 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3805 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3807 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3808 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3809 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3810 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3811 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3815 * copy everything in the in-memory inode into the btree.
3817 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3818 struct btrfs_root
*root
, struct inode
*inode
)
3820 struct btrfs_inode_item
*inode_item
;
3821 struct btrfs_path
*path
;
3822 struct extent_buffer
*leaf
;
3825 path
= btrfs_alloc_path();
3829 path
->leave_spinning
= 1;
3830 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3838 leaf
= path
->nodes
[0];
3839 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3840 struct btrfs_inode_item
);
3842 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3843 btrfs_mark_buffer_dirty(leaf
);
3844 btrfs_set_inode_last_trans(trans
, inode
);
3847 btrfs_free_path(path
);
3852 * copy everything in the in-memory inode into the btree.
3854 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3855 struct btrfs_root
*root
, struct inode
*inode
)
3860 * If the inode is a free space inode, we can deadlock during commit
3861 * if we put it into the delayed code.
3863 * The data relocation inode should also be directly updated
3866 if (!btrfs_is_free_space_inode(inode
)
3867 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3868 && !root
->fs_info
->log_root_recovering
) {
3869 btrfs_update_root_times(trans
, root
);
3871 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3873 btrfs_set_inode_last_trans(trans
, inode
);
3877 return btrfs_update_inode_item(trans
, root
, inode
);
3880 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3881 struct btrfs_root
*root
,
3882 struct inode
*inode
)
3886 ret
= btrfs_update_inode(trans
, root
, inode
);
3888 return btrfs_update_inode_item(trans
, root
, inode
);
3893 * unlink helper that gets used here in inode.c and in the tree logging
3894 * recovery code. It remove a link in a directory with a given name, and
3895 * also drops the back refs in the inode to the directory
3897 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3898 struct btrfs_root
*root
,
3899 struct inode
*dir
, struct inode
*inode
,
3900 const char *name
, int name_len
)
3902 struct btrfs_path
*path
;
3904 struct extent_buffer
*leaf
;
3905 struct btrfs_dir_item
*di
;
3906 struct btrfs_key key
;
3908 u64 ino
= btrfs_ino(inode
);
3909 u64 dir_ino
= btrfs_ino(dir
);
3911 path
= btrfs_alloc_path();
3917 path
->leave_spinning
= 1;
3918 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3919 name
, name_len
, -1);
3928 leaf
= path
->nodes
[0];
3929 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3930 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3933 btrfs_release_path(path
);
3936 * If we don't have dir index, we have to get it by looking up
3937 * the inode ref, since we get the inode ref, remove it directly,
3938 * it is unnecessary to do delayed deletion.
3940 * But if we have dir index, needn't search inode ref to get it.
3941 * Since the inode ref is close to the inode item, it is better
3942 * that we delay to delete it, and just do this deletion when
3943 * we update the inode item.
3945 if (BTRFS_I(inode
)->dir_index
) {
3946 ret
= btrfs_delayed_delete_inode_ref(inode
);
3948 index
= BTRFS_I(inode
)->dir_index
;
3953 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3956 btrfs_info(root
->fs_info
,
3957 "failed to delete reference to %.*s, inode %llu parent %llu",
3958 name_len
, name
, ino
, dir_ino
);
3959 btrfs_abort_transaction(trans
, root
, ret
);
3963 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3965 btrfs_abort_transaction(trans
, root
, ret
);
3969 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
3971 if (ret
!= 0 && ret
!= -ENOENT
) {
3972 btrfs_abort_transaction(trans
, root
, ret
);
3976 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
3981 btrfs_abort_transaction(trans
, root
, ret
);
3983 btrfs_free_path(path
);
3987 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3988 inode_inc_iversion(inode
);
3989 inode_inc_iversion(dir
);
3990 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3991 ret
= btrfs_update_inode(trans
, root
, dir
);
3996 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3997 struct btrfs_root
*root
,
3998 struct inode
*dir
, struct inode
*inode
,
3999 const char *name
, int name_len
)
4002 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4005 ret
= btrfs_update_inode(trans
, root
, inode
);
4011 * helper to start transaction for unlink and rmdir.
4013 * unlink and rmdir are special in btrfs, they do not always free space, so
4014 * if we cannot make our reservations the normal way try and see if there is
4015 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4016 * allow the unlink to occur.
4018 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4020 struct btrfs_trans_handle
*trans
;
4021 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4025 * 1 for the possible orphan item
4026 * 1 for the dir item
4027 * 1 for the dir index
4028 * 1 for the inode ref
4031 trans
= btrfs_start_transaction(root
, 5);
4032 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
4035 if (PTR_ERR(trans
) == -ENOSPC
) {
4036 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
4038 trans
= btrfs_start_transaction(root
, 0);
4041 ret
= btrfs_cond_migrate_bytes(root
->fs_info
,
4042 &root
->fs_info
->trans_block_rsv
,
4045 btrfs_end_transaction(trans
, root
);
4046 return ERR_PTR(ret
);
4048 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
4049 trans
->bytes_reserved
= num_bytes
;
4054 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4056 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4057 struct btrfs_trans_handle
*trans
;
4058 struct inode
*inode
= d_inode(dentry
);
4061 trans
= __unlink_start_trans(dir
);
4063 return PTR_ERR(trans
);
4065 btrfs_record_unlink_dir(trans
, dir
, d_inode(dentry
), 0);
4067 ret
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4068 dentry
->d_name
.name
, dentry
->d_name
.len
);
4072 if (inode
->i_nlink
== 0) {
4073 ret
= btrfs_orphan_add(trans
, inode
);
4079 btrfs_end_transaction(trans
, root
);
4080 btrfs_btree_balance_dirty(root
);
4084 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4085 struct btrfs_root
*root
,
4086 struct inode
*dir
, u64 objectid
,
4087 const char *name
, int name_len
)
4089 struct btrfs_path
*path
;
4090 struct extent_buffer
*leaf
;
4091 struct btrfs_dir_item
*di
;
4092 struct btrfs_key key
;
4095 u64 dir_ino
= btrfs_ino(dir
);
4097 path
= btrfs_alloc_path();
4101 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4102 name
, name_len
, -1);
4103 if (IS_ERR_OR_NULL(di
)) {
4111 leaf
= path
->nodes
[0];
4112 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4113 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4114 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4116 btrfs_abort_transaction(trans
, root
, ret
);
4119 btrfs_release_path(path
);
4121 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4122 objectid
, root
->root_key
.objectid
,
4123 dir_ino
, &index
, name
, name_len
);
4125 if (ret
!= -ENOENT
) {
4126 btrfs_abort_transaction(trans
, root
, ret
);
4129 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4131 if (IS_ERR_OR_NULL(di
)) {
4136 btrfs_abort_transaction(trans
, root
, ret
);
4140 leaf
= path
->nodes
[0];
4141 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4142 btrfs_release_path(path
);
4145 btrfs_release_path(path
);
4147 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4149 btrfs_abort_transaction(trans
, root
, ret
);
4153 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4154 inode_inc_iversion(dir
);
4155 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4156 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4158 btrfs_abort_transaction(trans
, root
, ret
);
4160 btrfs_free_path(path
);
4164 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4166 struct inode
*inode
= d_inode(dentry
);
4168 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4169 struct btrfs_trans_handle
*trans
;
4171 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4173 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
4176 trans
= __unlink_start_trans(dir
);
4178 return PTR_ERR(trans
);
4180 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4181 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4182 BTRFS_I(inode
)->location
.objectid
,
4183 dentry
->d_name
.name
,
4184 dentry
->d_name
.len
);
4188 err
= btrfs_orphan_add(trans
, inode
);
4192 /* now the directory is empty */
4193 err
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4194 dentry
->d_name
.name
, dentry
->d_name
.len
);
4196 btrfs_i_size_write(inode
, 0);
4198 btrfs_end_transaction(trans
, root
);
4199 btrfs_btree_balance_dirty(root
);
4204 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4205 struct btrfs_root
*root
,
4210 bytes_deleted
= btrfs_csum_bytes_to_leaves(root
, bytes_deleted
);
4211 ret
= btrfs_block_rsv_add(root
, &root
->fs_info
->trans_block_rsv
,
4212 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4214 trans
->bytes_reserved
+= bytes_deleted
;
4220 * this can truncate away extent items, csum items and directory items.
4221 * It starts at a high offset and removes keys until it can't find
4222 * any higher than new_size
4224 * csum items that cross the new i_size are truncated to the new size
4227 * min_type is the minimum key type to truncate down to. If set to 0, this
4228 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4230 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4231 struct btrfs_root
*root
,
4232 struct inode
*inode
,
4233 u64 new_size
, u32 min_type
)
4235 struct btrfs_path
*path
;
4236 struct extent_buffer
*leaf
;
4237 struct btrfs_file_extent_item
*fi
;
4238 struct btrfs_key key
;
4239 struct btrfs_key found_key
;
4240 u64 extent_start
= 0;
4241 u64 extent_num_bytes
= 0;
4242 u64 extent_offset
= 0;
4244 u64 last_size
= new_size
;
4245 u32 found_type
= (u8
)-1;
4248 int pending_del_nr
= 0;
4249 int pending_del_slot
= 0;
4250 int extent_type
= -1;
4253 u64 ino
= btrfs_ino(inode
);
4254 u64 bytes_deleted
= 0;
4256 bool should_throttle
= 0;
4257 bool should_end
= 0;
4259 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4262 * for non-free space inodes and ref cows, we want to back off from
4265 if (!btrfs_is_free_space_inode(inode
) &&
4266 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4269 path
= btrfs_alloc_path();
4275 * We want to drop from the next block forward in case this new size is
4276 * not block aligned since we will be keeping the last block of the
4277 * extent just the way it is.
4279 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4280 root
== root
->fs_info
->tree_root
)
4281 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
4282 root
->sectorsize
), (u64
)-1, 0);
4285 * This function is also used to drop the items in the log tree before
4286 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4287 * it is used to drop the loged items. So we shouldn't kill the delayed
4290 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4291 btrfs_kill_delayed_inode_items(inode
);
4294 key
.offset
= (u64
)-1;
4299 * with a 16K leaf size and 128MB extents, you can actually queue
4300 * up a huge file in a single leaf. Most of the time that
4301 * bytes_deleted is > 0, it will be huge by the time we get here
4303 if (be_nice
&& bytes_deleted
> 32 * 1024 * 1024) {
4304 if (btrfs_should_end_transaction(trans
, root
)) {
4311 path
->leave_spinning
= 1;
4312 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4319 /* there are no items in the tree for us to truncate, we're
4322 if (path
->slots
[0] == 0)
4329 leaf
= path
->nodes
[0];
4330 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4331 found_type
= found_key
.type
;
4333 if (found_key
.objectid
!= ino
)
4336 if (found_type
< min_type
)
4339 item_end
= found_key
.offset
;
4340 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4341 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4342 struct btrfs_file_extent_item
);
4343 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4344 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4346 btrfs_file_extent_num_bytes(leaf
, fi
);
4347 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4348 item_end
+= btrfs_file_extent_inline_len(leaf
,
4349 path
->slots
[0], fi
);
4353 if (found_type
> min_type
) {
4356 if (item_end
< new_size
)
4358 if (found_key
.offset
>= new_size
)
4364 /* FIXME, shrink the extent if the ref count is only 1 */
4365 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4369 last_size
= found_key
.offset
;
4371 last_size
= new_size
;
4373 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4375 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4377 u64 orig_num_bytes
=
4378 btrfs_file_extent_num_bytes(leaf
, fi
);
4379 extent_num_bytes
= ALIGN(new_size
-
4382 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4384 num_dec
= (orig_num_bytes
-
4386 if (test_bit(BTRFS_ROOT_REF_COWS
,
4389 inode_sub_bytes(inode
, num_dec
);
4390 btrfs_mark_buffer_dirty(leaf
);
4393 btrfs_file_extent_disk_num_bytes(leaf
,
4395 extent_offset
= found_key
.offset
-
4396 btrfs_file_extent_offset(leaf
, fi
);
4398 /* FIXME blocksize != 4096 */
4399 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4400 if (extent_start
!= 0) {
4402 if (test_bit(BTRFS_ROOT_REF_COWS
,
4404 inode_sub_bytes(inode
, num_dec
);
4407 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4409 * we can't truncate inline items that have had
4413 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
4414 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4415 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4416 u32 size
= new_size
- found_key
.offset
;
4418 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4419 inode_sub_bytes(inode
, item_end
+ 1 -
4423 * update the ram bytes to properly reflect
4424 * the new size of our item
4426 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4428 btrfs_file_extent_calc_inline_size(size
);
4429 btrfs_truncate_item(root
, path
, size
, 1);
4430 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4432 inode_sub_bytes(inode
, item_end
+ 1 -
4438 if (!pending_del_nr
) {
4439 /* no pending yet, add ourselves */
4440 pending_del_slot
= path
->slots
[0];
4442 } else if (pending_del_nr
&&
4443 path
->slots
[0] + 1 == pending_del_slot
) {
4444 /* hop on the pending chunk */
4446 pending_del_slot
= path
->slots
[0];
4453 should_throttle
= 0;
4456 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4457 root
== root
->fs_info
->tree_root
)) {
4458 btrfs_set_path_blocking(path
);
4459 bytes_deleted
+= extent_num_bytes
;
4460 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4461 extent_num_bytes
, 0,
4462 btrfs_header_owner(leaf
),
4463 ino
, extent_offset
, 0);
4465 if (btrfs_should_throttle_delayed_refs(trans
, root
))
4466 btrfs_async_run_delayed_refs(root
,
4467 trans
->delayed_ref_updates
* 2, 0);
4469 if (truncate_space_check(trans
, root
,
4470 extent_num_bytes
)) {
4473 if (btrfs_should_throttle_delayed_refs(trans
,
4475 should_throttle
= 1;
4480 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4483 if (path
->slots
[0] == 0 ||
4484 path
->slots
[0] != pending_del_slot
||
4485 should_throttle
|| should_end
) {
4486 if (pending_del_nr
) {
4487 ret
= btrfs_del_items(trans
, root
, path
,
4491 btrfs_abort_transaction(trans
,
4497 btrfs_release_path(path
);
4498 if (should_throttle
) {
4499 unsigned long updates
= trans
->delayed_ref_updates
;
4501 trans
->delayed_ref_updates
= 0;
4502 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4508 * if we failed to refill our space rsv, bail out
4509 * and let the transaction restart
4521 if (pending_del_nr
) {
4522 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4525 btrfs_abort_transaction(trans
, root
, ret
);
4528 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4529 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4531 btrfs_free_path(path
);
4533 if (be_nice
&& bytes_deleted
> 32 * 1024 * 1024) {
4534 unsigned long updates
= trans
->delayed_ref_updates
;
4536 trans
->delayed_ref_updates
= 0;
4537 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4546 * btrfs_truncate_page - read, zero a chunk and write a page
4547 * @inode - inode that we're zeroing
4548 * @from - the offset to start zeroing
4549 * @len - the length to zero, 0 to zero the entire range respective to the
4551 * @front - zero up to the offset instead of from the offset on
4553 * This will find the page for the "from" offset and cow the page and zero the
4554 * part we want to zero. This is used with truncate and hole punching.
4556 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
4559 struct address_space
*mapping
= inode
->i_mapping
;
4560 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4561 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4562 struct btrfs_ordered_extent
*ordered
;
4563 struct extent_state
*cached_state
= NULL
;
4565 u32 blocksize
= root
->sectorsize
;
4566 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4567 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4569 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4574 if ((offset
& (blocksize
- 1)) == 0 &&
4575 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4577 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
4582 page
= find_or_create_page(mapping
, index
, mask
);
4584 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4589 page_start
= page_offset(page
);
4590 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4592 if (!PageUptodate(page
)) {
4593 ret
= btrfs_readpage(NULL
, page
);
4595 if (page
->mapping
!= mapping
) {
4597 page_cache_release(page
);
4600 if (!PageUptodate(page
)) {
4605 wait_on_page_writeback(page
);
4607 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
4608 set_page_extent_mapped(page
);
4610 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4612 unlock_extent_cached(io_tree
, page_start
, page_end
,
4613 &cached_state
, GFP_NOFS
);
4615 page_cache_release(page
);
4616 btrfs_start_ordered_extent(inode
, ordered
, 1);
4617 btrfs_put_ordered_extent(ordered
);
4621 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
4622 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4623 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4624 0, 0, &cached_state
, GFP_NOFS
);
4626 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
4629 unlock_extent_cached(io_tree
, page_start
, page_end
,
4630 &cached_state
, GFP_NOFS
);
4634 if (offset
!= PAGE_CACHE_SIZE
) {
4636 len
= PAGE_CACHE_SIZE
- offset
;
4639 memset(kaddr
, 0, offset
);
4641 memset(kaddr
+ offset
, 0, len
);
4642 flush_dcache_page(page
);
4645 ClearPageChecked(page
);
4646 set_page_dirty(page
);
4647 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
4652 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4654 page_cache_release(page
);
4659 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4660 u64 offset
, u64 len
)
4662 struct btrfs_trans_handle
*trans
;
4666 * Still need to make sure the inode looks like it's been updated so
4667 * that any holes get logged if we fsync.
4669 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
)) {
4670 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
4671 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4672 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4677 * 1 - for the one we're dropping
4678 * 1 - for the one we're adding
4679 * 1 - for updating the inode.
4681 trans
= btrfs_start_transaction(root
, 3);
4683 return PTR_ERR(trans
);
4685 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4687 btrfs_abort_transaction(trans
, root
, ret
);
4688 btrfs_end_transaction(trans
, root
);
4692 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
4693 0, 0, len
, 0, len
, 0, 0, 0);
4695 btrfs_abort_transaction(trans
, root
, ret
);
4697 btrfs_update_inode(trans
, root
, inode
);
4698 btrfs_end_transaction(trans
, root
);
4703 * This function puts in dummy file extents for the area we're creating a hole
4704 * for. So if we are truncating this file to a larger size we need to insert
4705 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4706 * the range between oldsize and size
4708 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4710 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4711 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4712 struct extent_map
*em
= NULL
;
4713 struct extent_state
*cached_state
= NULL
;
4714 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4715 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4716 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4723 * If our size started in the middle of a page we need to zero out the
4724 * rest of the page before we expand the i_size, otherwise we could
4725 * expose stale data.
4727 err
= btrfs_truncate_page(inode
, oldsize
, 0, 0);
4731 if (size
<= hole_start
)
4735 struct btrfs_ordered_extent
*ordered
;
4737 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
4739 ordered
= btrfs_lookup_ordered_range(inode
, hole_start
,
4740 block_end
- hole_start
);
4743 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4744 &cached_state
, GFP_NOFS
);
4745 btrfs_start_ordered_extent(inode
, ordered
, 1);
4746 btrfs_put_ordered_extent(ordered
);
4749 cur_offset
= hole_start
;
4751 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4752 block_end
- cur_offset
, 0);
4758 last_byte
= min(extent_map_end(em
), block_end
);
4759 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4760 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4761 struct extent_map
*hole_em
;
4762 hole_size
= last_byte
- cur_offset
;
4764 err
= maybe_insert_hole(root
, inode
, cur_offset
,
4768 btrfs_drop_extent_cache(inode
, cur_offset
,
4769 cur_offset
+ hole_size
- 1, 0);
4770 hole_em
= alloc_extent_map();
4772 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4773 &BTRFS_I(inode
)->runtime_flags
);
4776 hole_em
->start
= cur_offset
;
4777 hole_em
->len
= hole_size
;
4778 hole_em
->orig_start
= cur_offset
;
4780 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4781 hole_em
->block_len
= 0;
4782 hole_em
->orig_block_len
= 0;
4783 hole_em
->ram_bytes
= hole_size
;
4784 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4785 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4786 hole_em
->generation
= root
->fs_info
->generation
;
4789 write_lock(&em_tree
->lock
);
4790 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4791 write_unlock(&em_tree
->lock
);
4794 btrfs_drop_extent_cache(inode
, cur_offset
,
4798 free_extent_map(hole_em
);
4801 free_extent_map(em
);
4803 cur_offset
= last_byte
;
4804 if (cur_offset
>= block_end
)
4807 free_extent_map(em
);
4808 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4813 static int wait_snapshoting_atomic_t(atomic_t
*a
)
4819 static void wait_for_snapshot_creation(struct btrfs_root
*root
)
4824 ret
= btrfs_start_write_no_snapshoting(root
);
4827 wait_on_atomic_t(&root
->will_be_snapshoted
,
4828 wait_snapshoting_atomic_t
,
4829 TASK_UNINTERRUPTIBLE
);
4833 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4835 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4836 struct btrfs_trans_handle
*trans
;
4837 loff_t oldsize
= i_size_read(inode
);
4838 loff_t newsize
= attr
->ia_size
;
4839 int mask
= attr
->ia_valid
;
4843 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4844 * special case where we need to update the times despite not having
4845 * these flags set. For all other operations the VFS set these flags
4846 * explicitly if it wants a timestamp update.
4848 if (newsize
!= oldsize
) {
4849 inode_inc_iversion(inode
);
4850 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
4851 inode
->i_ctime
= inode
->i_mtime
=
4852 current_fs_time(inode
->i_sb
);
4855 if (newsize
> oldsize
) {
4856 truncate_pagecache(inode
, newsize
);
4858 * Don't do an expanding truncate while snapshoting is ongoing.
4859 * This is to ensure the snapshot captures a fully consistent
4860 * state of this file - if the snapshot captures this expanding
4861 * truncation, it must capture all writes that happened before
4864 wait_for_snapshot_creation(root
);
4865 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4867 btrfs_end_write_no_snapshoting(root
);
4871 trans
= btrfs_start_transaction(root
, 1);
4872 if (IS_ERR(trans
)) {
4873 btrfs_end_write_no_snapshoting(root
);
4874 return PTR_ERR(trans
);
4877 i_size_write(inode
, newsize
);
4878 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4879 ret
= btrfs_update_inode(trans
, root
, inode
);
4880 btrfs_end_write_no_snapshoting(root
);
4881 btrfs_end_transaction(trans
, root
);
4885 * We're truncating a file that used to have good data down to
4886 * zero. Make sure it gets into the ordered flush list so that
4887 * any new writes get down to disk quickly.
4890 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4891 &BTRFS_I(inode
)->runtime_flags
);
4894 * 1 for the orphan item we're going to add
4895 * 1 for the orphan item deletion.
4897 trans
= btrfs_start_transaction(root
, 2);
4899 return PTR_ERR(trans
);
4902 * We need to do this in case we fail at _any_ point during the
4903 * actual truncate. Once we do the truncate_setsize we could
4904 * invalidate pages which forces any outstanding ordered io to
4905 * be instantly completed which will give us extents that need
4906 * to be truncated. If we fail to get an orphan inode down we
4907 * could have left over extents that were never meant to live,
4908 * so we need to garuntee from this point on that everything
4909 * will be consistent.
4911 ret
= btrfs_orphan_add(trans
, inode
);
4912 btrfs_end_transaction(trans
, root
);
4916 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4917 truncate_setsize(inode
, newsize
);
4919 /* Disable nonlocked read DIO to avoid the end less truncate */
4920 btrfs_inode_block_unlocked_dio(inode
);
4921 inode_dio_wait(inode
);
4922 btrfs_inode_resume_unlocked_dio(inode
);
4924 ret
= btrfs_truncate(inode
);
4925 if (ret
&& inode
->i_nlink
) {
4929 * failed to truncate, disk_i_size is only adjusted down
4930 * as we remove extents, so it should represent the true
4931 * size of the inode, so reset the in memory size and
4932 * delete our orphan entry.
4934 trans
= btrfs_join_transaction(root
);
4935 if (IS_ERR(trans
)) {
4936 btrfs_orphan_del(NULL
, inode
);
4939 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
4940 err
= btrfs_orphan_del(trans
, inode
);
4942 btrfs_abort_transaction(trans
, root
, err
);
4943 btrfs_end_transaction(trans
, root
);
4950 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4952 struct inode
*inode
= d_inode(dentry
);
4953 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4956 if (btrfs_root_readonly(root
))
4959 err
= inode_change_ok(inode
, attr
);
4963 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
4964 err
= btrfs_setsize(inode
, attr
);
4969 if (attr
->ia_valid
) {
4970 setattr_copy(inode
, attr
);
4971 inode_inc_iversion(inode
);
4972 err
= btrfs_dirty_inode(inode
);
4974 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
4975 err
= posix_acl_chmod(inode
, inode
->i_mode
);
4982 * While truncating the inode pages during eviction, we get the VFS calling
4983 * btrfs_invalidatepage() against each page of the inode. This is slow because
4984 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4985 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4986 * extent_state structures over and over, wasting lots of time.
4988 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4989 * those expensive operations on a per page basis and do only the ordered io
4990 * finishing, while we release here the extent_map and extent_state structures,
4991 * without the excessive merging and splitting.
4993 static void evict_inode_truncate_pages(struct inode
*inode
)
4995 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4996 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
4997 struct rb_node
*node
;
4999 ASSERT(inode
->i_state
& I_FREEING
);
5000 truncate_inode_pages_final(&inode
->i_data
);
5002 write_lock(&map_tree
->lock
);
5003 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5004 struct extent_map
*em
;
5006 node
= rb_first(&map_tree
->map
);
5007 em
= rb_entry(node
, struct extent_map
, rb_node
);
5008 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5009 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5010 remove_extent_mapping(map_tree
, em
);
5011 free_extent_map(em
);
5012 if (need_resched()) {
5013 write_unlock(&map_tree
->lock
);
5015 write_lock(&map_tree
->lock
);
5018 write_unlock(&map_tree
->lock
);
5021 * Keep looping until we have no more ranges in the io tree.
5022 * We can have ongoing bios started by readpages (called from readahead)
5023 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5024 * still in progress (unlocked the pages in the bio but did not yet
5025 * unlocked the ranges in the io tree). Therefore this means some
5026 * ranges can still be locked and eviction started because before
5027 * submitting those bios, which are executed by a separate task (work
5028 * queue kthread), inode references (inode->i_count) were not taken
5029 * (which would be dropped in the end io callback of each bio).
5030 * Therefore here we effectively end up waiting for those bios and
5031 * anyone else holding locked ranges without having bumped the inode's
5032 * reference count - if we don't do it, when they access the inode's
5033 * io_tree to unlock a range it may be too late, leading to an
5034 * use-after-free issue.
5036 spin_lock(&io_tree
->lock
);
5037 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5038 struct extent_state
*state
;
5039 struct extent_state
*cached_state
= NULL
;
5043 node
= rb_first(&io_tree
->state
);
5044 state
= rb_entry(node
, struct extent_state
, rb_node
);
5045 start
= state
->start
;
5047 spin_unlock(&io_tree
->lock
);
5049 lock_extent_bits(io_tree
, start
, end
, 0, &cached_state
);
5050 clear_extent_bit(io_tree
, start
, end
,
5051 EXTENT_LOCKED
| EXTENT_DIRTY
|
5052 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5053 EXTENT_DEFRAG
, 1, 1,
5054 &cached_state
, GFP_NOFS
);
5057 spin_lock(&io_tree
->lock
);
5059 spin_unlock(&io_tree
->lock
);
5062 void btrfs_evict_inode(struct inode
*inode
)
5064 struct btrfs_trans_handle
*trans
;
5065 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5066 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5067 int steal_from_global
= 0;
5068 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
5071 trace_btrfs_inode_evict(inode
);
5073 evict_inode_truncate_pages(inode
);
5075 if (inode
->i_nlink
&&
5076 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5077 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5078 btrfs_is_free_space_inode(inode
)))
5081 if (is_bad_inode(inode
)) {
5082 btrfs_orphan_del(NULL
, inode
);
5085 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5086 if (!special_file(inode
->i_mode
))
5087 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5089 btrfs_free_io_failure_record(inode
, 0, (u64
)-1);
5091 if (root
->fs_info
->log_root_recovering
) {
5092 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5093 &BTRFS_I(inode
)->runtime_flags
));
5097 if (inode
->i_nlink
> 0) {
5098 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5099 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5103 ret
= btrfs_commit_inode_delayed_inode(inode
);
5105 btrfs_orphan_del(NULL
, inode
);
5109 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
5111 btrfs_orphan_del(NULL
, inode
);
5114 rsv
->size
= min_size
;
5116 global_rsv
= &root
->fs_info
->global_block_rsv
;
5118 btrfs_i_size_write(inode
, 0);
5121 * This is a bit simpler than btrfs_truncate since we've already
5122 * reserved our space for our orphan item in the unlink, so we just
5123 * need to reserve some slack space in case we add bytes and update
5124 * inode item when doing the truncate.
5127 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5128 BTRFS_RESERVE_FLUSH_LIMIT
);
5131 * Try and steal from the global reserve since we will
5132 * likely not use this space anyway, we want to try as
5133 * hard as possible to get this to work.
5136 steal_from_global
++;
5138 steal_from_global
= 0;
5142 * steal_from_global == 0: we reserved stuff, hooray!
5143 * steal_from_global == 1: we didn't reserve stuff, boo!
5144 * steal_from_global == 2: we've committed, still not a lot of
5145 * room but maybe we'll have room in the global reserve this
5147 * steal_from_global == 3: abandon all hope!
5149 if (steal_from_global
> 2) {
5150 btrfs_warn(root
->fs_info
,
5151 "Could not get space for a delete, will truncate on mount %d",
5153 btrfs_orphan_del(NULL
, inode
);
5154 btrfs_free_block_rsv(root
, rsv
);
5158 trans
= btrfs_join_transaction(root
);
5159 if (IS_ERR(trans
)) {
5160 btrfs_orphan_del(NULL
, inode
);
5161 btrfs_free_block_rsv(root
, rsv
);
5166 * We can't just steal from the global reserve, we need tomake
5167 * sure there is room to do it, if not we need to commit and try
5170 if (steal_from_global
) {
5171 if (!btrfs_check_space_for_delayed_refs(trans
, root
))
5172 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5179 * Couldn't steal from the global reserve, we have too much
5180 * pending stuff built up, commit the transaction and try it
5184 ret
= btrfs_commit_transaction(trans
, root
);
5186 btrfs_orphan_del(NULL
, inode
);
5187 btrfs_free_block_rsv(root
, rsv
);
5192 steal_from_global
= 0;
5195 trans
->block_rsv
= rsv
;
5197 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5198 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5201 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5202 btrfs_end_transaction(trans
, root
);
5204 btrfs_btree_balance_dirty(root
);
5207 btrfs_free_block_rsv(root
, rsv
);
5210 * Errors here aren't a big deal, it just means we leave orphan items
5211 * in the tree. They will be cleaned up on the next mount.
5214 trans
->block_rsv
= root
->orphan_block_rsv
;
5215 btrfs_orphan_del(trans
, inode
);
5217 btrfs_orphan_del(NULL
, inode
);
5220 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5221 if (!(root
== root
->fs_info
->tree_root
||
5222 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5223 btrfs_return_ino(root
, btrfs_ino(inode
));
5225 btrfs_end_transaction(trans
, root
);
5226 btrfs_btree_balance_dirty(root
);
5228 btrfs_remove_delayed_node(inode
);
5234 * this returns the key found in the dir entry in the location pointer.
5235 * If no dir entries were found, location->objectid is 0.
5237 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5238 struct btrfs_key
*location
)
5240 const char *name
= dentry
->d_name
.name
;
5241 int namelen
= dentry
->d_name
.len
;
5242 struct btrfs_dir_item
*di
;
5243 struct btrfs_path
*path
;
5244 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5247 path
= btrfs_alloc_path();
5251 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
5256 if (IS_ERR_OR_NULL(di
))
5259 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5261 btrfs_free_path(path
);
5264 location
->objectid
= 0;
5269 * when we hit a tree root in a directory, the btrfs part of the inode
5270 * needs to be changed to reflect the root directory of the tree root. This
5271 * is kind of like crossing a mount point.
5273 static int fixup_tree_root_location(struct btrfs_root
*root
,
5275 struct dentry
*dentry
,
5276 struct btrfs_key
*location
,
5277 struct btrfs_root
**sub_root
)
5279 struct btrfs_path
*path
;
5280 struct btrfs_root
*new_root
;
5281 struct btrfs_root_ref
*ref
;
5282 struct extent_buffer
*leaf
;
5283 struct btrfs_key key
;
5287 path
= btrfs_alloc_path();
5294 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5295 key
.type
= BTRFS_ROOT_REF_KEY
;
5296 key
.offset
= location
->objectid
;
5298 ret
= btrfs_search_slot(NULL
, root
->fs_info
->tree_root
, &key
, path
,
5306 leaf
= path
->nodes
[0];
5307 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5308 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
5309 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5312 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5313 (unsigned long)(ref
+ 1),
5314 dentry
->d_name
.len
);
5318 btrfs_release_path(path
);
5320 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
5321 if (IS_ERR(new_root
)) {
5322 err
= PTR_ERR(new_root
);
5326 *sub_root
= new_root
;
5327 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5328 location
->type
= BTRFS_INODE_ITEM_KEY
;
5329 location
->offset
= 0;
5332 btrfs_free_path(path
);
5336 static void inode_tree_add(struct inode
*inode
)
5338 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5339 struct btrfs_inode
*entry
;
5341 struct rb_node
*parent
;
5342 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5343 u64 ino
= btrfs_ino(inode
);
5345 if (inode_unhashed(inode
))
5348 spin_lock(&root
->inode_lock
);
5349 p
= &root
->inode_tree
.rb_node
;
5352 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5354 if (ino
< btrfs_ino(&entry
->vfs_inode
))
5355 p
= &parent
->rb_left
;
5356 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
5357 p
= &parent
->rb_right
;
5359 WARN_ON(!(entry
->vfs_inode
.i_state
&
5360 (I_WILL_FREE
| I_FREEING
)));
5361 rb_replace_node(parent
, new, &root
->inode_tree
);
5362 RB_CLEAR_NODE(parent
);
5363 spin_unlock(&root
->inode_lock
);
5367 rb_link_node(new, parent
, p
);
5368 rb_insert_color(new, &root
->inode_tree
);
5369 spin_unlock(&root
->inode_lock
);
5372 static void inode_tree_del(struct inode
*inode
)
5374 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5377 spin_lock(&root
->inode_lock
);
5378 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5379 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5380 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5381 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5383 spin_unlock(&root
->inode_lock
);
5385 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5386 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
5387 spin_lock(&root
->inode_lock
);
5388 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5389 spin_unlock(&root
->inode_lock
);
5391 btrfs_add_dead_root(root
);
5395 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5397 struct rb_node
*node
;
5398 struct rb_node
*prev
;
5399 struct btrfs_inode
*entry
;
5400 struct inode
*inode
;
5403 if (!test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
5404 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5406 spin_lock(&root
->inode_lock
);
5408 node
= root
->inode_tree
.rb_node
;
5412 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5414 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
5415 node
= node
->rb_left
;
5416 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
5417 node
= node
->rb_right
;
5423 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5424 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
5428 prev
= rb_next(prev
);
5432 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5433 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
5434 inode
= igrab(&entry
->vfs_inode
);
5436 spin_unlock(&root
->inode_lock
);
5437 if (atomic_read(&inode
->i_count
) > 1)
5438 d_prune_aliases(inode
);
5440 * btrfs_drop_inode will have it removed from
5441 * the inode cache when its usage count
5446 spin_lock(&root
->inode_lock
);
5450 if (cond_resched_lock(&root
->inode_lock
))
5453 node
= rb_next(node
);
5455 spin_unlock(&root
->inode_lock
);
5458 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5460 struct btrfs_iget_args
*args
= p
;
5461 inode
->i_ino
= args
->location
->objectid
;
5462 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5463 sizeof(*args
->location
));
5464 BTRFS_I(inode
)->root
= args
->root
;
5468 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5470 struct btrfs_iget_args
*args
= opaque
;
5471 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5472 args
->root
== BTRFS_I(inode
)->root
;
5475 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5476 struct btrfs_key
*location
,
5477 struct btrfs_root
*root
)
5479 struct inode
*inode
;
5480 struct btrfs_iget_args args
;
5481 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5483 args
.location
= location
;
5486 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5487 btrfs_init_locked_inode
,
5492 /* Get an inode object given its location and corresponding root.
5493 * Returns in *is_new if the inode was read from disk
5495 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5496 struct btrfs_root
*root
, int *new)
5498 struct inode
*inode
;
5500 inode
= btrfs_iget_locked(s
, location
, root
);
5502 return ERR_PTR(-ENOMEM
);
5504 if (inode
->i_state
& I_NEW
) {
5505 btrfs_read_locked_inode(inode
);
5506 if (!is_bad_inode(inode
)) {
5507 inode_tree_add(inode
);
5508 unlock_new_inode(inode
);
5512 unlock_new_inode(inode
);
5514 inode
= ERR_PTR(-ESTALE
);
5521 static struct inode
*new_simple_dir(struct super_block
*s
,
5522 struct btrfs_key
*key
,
5523 struct btrfs_root
*root
)
5525 struct inode
*inode
= new_inode(s
);
5528 return ERR_PTR(-ENOMEM
);
5530 BTRFS_I(inode
)->root
= root
;
5531 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5532 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5534 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5535 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5536 inode
->i_fop
= &simple_dir_operations
;
5537 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5538 inode
->i_mtime
= CURRENT_TIME
;
5539 inode
->i_atime
= inode
->i_mtime
;
5540 inode
->i_ctime
= inode
->i_mtime
;
5541 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5546 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5548 struct inode
*inode
;
5549 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5550 struct btrfs_root
*sub_root
= root
;
5551 struct btrfs_key location
;
5555 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5556 return ERR_PTR(-ENAMETOOLONG
);
5558 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5560 return ERR_PTR(ret
);
5562 if (location
.objectid
== 0)
5563 return ERR_PTR(-ENOENT
);
5565 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5566 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5570 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5572 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5573 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5574 &location
, &sub_root
);
5577 inode
= ERR_PTR(ret
);
5579 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5581 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5583 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5585 if (!IS_ERR(inode
) && root
!= sub_root
) {
5586 down_read(&root
->fs_info
->cleanup_work_sem
);
5587 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5588 ret
= btrfs_orphan_cleanup(sub_root
);
5589 up_read(&root
->fs_info
->cleanup_work_sem
);
5592 inode
= ERR_PTR(ret
);
5599 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5601 struct btrfs_root
*root
;
5602 struct inode
*inode
= d_inode(dentry
);
5604 if (!inode
&& !IS_ROOT(dentry
))
5605 inode
= d_inode(dentry
->d_parent
);
5608 root
= BTRFS_I(inode
)->root
;
5609 if (btrfs_root_refs(&root
->root_item
) == 0)
5612 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5618 static void btrfs_dentry_release(struct dentry
*dentry
)
5620 kfree(dentry
->d_fsdata
);
5623 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5626 struct inode
*inode
;
5628 inode
= btrfs_lookup_dentry(dir
, dentry
);
5629 if (IS_ERR(inode
)) {
5630 if (PTR_ERR(inode
) == -ENOENT
)
5633 return ERR_CAST(inode
);
5636 return d_splice_alias(inode
, dentry
);
5639 unsigned char btrfs_filetype_table
[] = {
5640 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5643 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5645 struct inode
*inode
= file_inode(file
);
5646 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5647 struct btrfs_item
*item
;
5648 struct btrfs_dir_item
*di
;
5649 struct btrfs_key key
;
5650 struct btrfs_key found_key
;
5651 struct btrfs_path
*path
;
5652 struct list_head ins_list
;
5653 struct list_head del_list
;
5655 struct extent_buffer
*leaf
;
5657 unsigned char d_type
;
5662 int key_type
= BTRFS_DIR_INDEX_KEY
;
5666 int is_curr
= 0; /* ctx->pos points to the current index? */
5668 /* FIXME, use a real flag for deciding about the key type */
5669 if (root
->fs_info
->tree_root
== root
)
5670 key_type
= BTRFS_DIR_ITEM_KEY
;
5672 if (!dir_emit_dots(file
, ctx
))
5675 path
= btrfs_alloc_path();
5681 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5682 INIT_LIST_HEAD(&ins_list
);
5683 INIT_LIST_HEAD(&del_list
);
5684 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5687 key
.type
= key_type
;
5688 key
.offset
= ctx
->pos
;
5689 key
.objectid
= btrfs_ino(inode
);
5691 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5696 leaf
= path
->nodes
[0];
5697 slot
= path
->slots
[0];
5698 if (slot
>= btrfs_header_nritems(leaf
)) {
5699 ret
= btrfs_next_leaf(root
, path
);
5707 item
= btrfs_item_nr(slot
);
5708 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5710 if (found_key
.objectid
!= key
.objectid
)
5712 if (found_key
.type
!= key_type
)
5714 if (found_key
.offset
< ctx
->pos
)
5716 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5717 btrfs_should_delete_dir_index(&del_list
,
5721 ctx
->pos
= found_key
.offset
;
5724 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5726 di_total
= btrfs_item_size(leaf
, item
);
5728 while (di_cur
< di_total
) {
5729 struct btrfs_key location
;
5731 if (verify_dir_item(root
, leaf
, di
))
5734 name_len
= btrfs_dir_name_len(leaf
, di
);
5735 if (name_len
<= sizeof(tmp_name
)) {
5736 name_ptr
= tmp_name
;
5738 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5744 read_extent_buffer(leaf
, name_ptr
,
5745 (unsigned long)(di
+ 1), name_len
);
5747 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5748 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5751 /* is this a reference to our own snapshot? If so
5754 * In contrast to old kernels, we insert the snapshot's
5755 * dir item and dir index after it has been created, so
5756 * we won't find a reference to our own snapshot. We
5757 * still keep the following code for backward
5760 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5761 location
.objectid
== root
->root_key
.objectid
) {
5765 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5766 location
.objectid
, d_type
);
5769 if (name_ptr
!= tmp_name
)
5774 di_len
= btrfs_dir_name_len(leaf
, di
) +
5775 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5777 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5783 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5786 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
5791 /* Reached end of directory/root. Bump pos past the last item. */
5795 * Stop new entries from being returned after we return the last
5798 * New directory entries are assigned a strictly increasing
5799 * offset. This means that new entries created during readdir
5800 * are *guaranteed* to be seen in the future by that readdir.
5801 * This has broken buggy programs which operate on names as
5802 * they're returned by readdir. Until we re-use freed offsets
5803 * we have this hack to stop new entries from being returned
5804 * under the assumption that they'll never reach this huge
5807 * This is being careful not to overflow 32bit loff_t unless the
5808 * last entry requires it because doing so has broken 32bit apps
5811 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5812 if (ctx
->pos
>= INT_MAX
)
5813 ctx
->pos
= LLONG_MAX
;
5820 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5821 btrfs_put_delayed_items(&ins_list
, &del_list
);
5822 btrfs_free_path(path
);
5826 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5828 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5829 struct btrfs_trans_handle
*trans
;
5831 bool nolock
= false;
5833 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5836 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5839 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5841 trans
= btrfs_join_transaction_nolock(root
);
5843 trans
= btrfs_join_transaction(root
);
5845 return PTR_ERR(trans
);
5846 ret
= btrfs_commit_transaction(trans
, root
);
5852 * This is somewhat expensive, updating the tree every time the
5853 * inode changes. But, it is most likely to find the inode in cache.
5854 * FIXME, needs more benchmarking...there are no reasons other than performance
5855 * to keep or drop this code.
5857 static int btrfs_dirty_inode(struct inode
*inode
)
5859 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5860 struct btrfs_trans_handle
*trans
;
5863 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5866 trans
= btrfs_join_transaction(root
);
5868 return PTR_ERR(trans
);
5870 ret
= btrfs_update_inode(trans
, root
, inode
);
5871 if (ret
&& ret
== -ENOSPC
) {
5872 /* whoops, lets try again with the full transaction */
5873 btrfs_end_transaction(trans
, root
);
5874 trans
= btrfs_start_transaction(root
, 1);
5876 return PTR_ERR(trans
);
5878 ret
= btrfs_update_inode(trans
, root
, inode
);
5880 btrfs_end_transaction(trans
, root
);
5881 if (BTRFS_I(inode
)->delayed_node
)
5882 btrfs_balance_delayed_items(root
);
5888 * This is a copy of file_update_time. We need this so we can return error on
5889 * ENOSPC for updating the inode in the case of file write and mmap writes.
5891 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
5894 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5896 if (btrfs_root_readonly(root
))
5899 if (flags
& S_VERSION
)
5900 inode_inc_iversion(inode
);
5901 if (flags
& S_CTIME
)
5902 inode
->i_ctime
= *now
;
5903 if (flags
& S_MTIME
)
5904 inode
->i_mtime
= *now
;
5905 if (flags
& S_ATIME
)
5906 inode
->i_atime
= *now
;
5907 return btrfs_dirty_inode(inode
);
5911 * find the highest existing sequence number in a directory
5912 * and then set the in-memory index_cnt variable to reflect
5913 * free sequence numbers
5915 static int btrfs_set_inode_index_count(struct inode
*inode
)
5917 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5918 struct btrfs_key key
, found_key
;
5919 struct btrfs_path
*path
;
5920 struct extent_buffer
*leaf
;
5923 key
.objectid
= btrfs_ino(inode
);
5924 key
.type
= BTRFS_DIR_INDEX_KEY
;
5925 key
.offset
= (u64
)-1;
5927 path
= btrfs_alloc_path();
5931 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5934 /* FIXME: we should be able to handle this */
5940 * MAGIC NUMBER EXPLANATION:
5941 * since we search a directory based on f_pos we have to start at 2
5942 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5943 * else has to start at 2
5945 if (path
->slots
[0] == 0) {
5946 BTRFS_I(inode
)->index_cnt
= 2;
5952 leaf
= path
->nodes
[0];
5953 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5955 if (found_key
.objectid
!= btrfs_ino(inode
) ||
5956 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
5957 BTRFS_I(inode
)->index_cnt
= 2;
5961 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
5963 btrfs_free_path(path
);
5968 * helper to find a free sequence number in a given directory. This current
5969 * code is very simple, later versions will do smarter things in the btree
5971 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
5975 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
5976 ret
= btrfs_inode_delayed_dir_index_count(dir
);
5978 ret
= btrfs_set_inode_index_count(dir
);
5984 *index
= BTRFS_I(dir
)->index_cnt
;
5985 BTRFS_I(dir
)->index_cnt
++;
5990 static int btrfs_insert_inode_locked(struct inode
*inode
)
5992 struct btrfs_iget_args args
;
5993 args
.location
= &BTRFS_I(inode
)->location
;
5994 args
.root
= BTRFS_I(inode
)->root
;
5996 return insert_inode_locked4(inode
,
5997 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
5998 btrfs_find_actor
, &args
);
6001 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6002 struct btrfs_root
*root
,
6004 const char *name
, int name_len
,
6005 u64 ref_objectid
, u64 objectid
,
6006 umode_t mode
, u64
*index
)
6008 struct inode
*inode
;
6009 struct btrfs_inode_item
*inode_item
;
6010 struct btrfs_key
*location
;
6011 struct btrfs_path
*path
;
6012 struct btrfs_inode_ref
*ref
;
6013 struct btrfs_key key
[2];
6015 int nitems
= name
? 2 : 1;
6019 path
= btrfs_alloc_path();
6021 return ERR_PTR(-ENOMEM
);
6023 inode
= new_inode(root
->fs_info
->sb
);
6025 btrfs_free_path(path
);
6026 return ERR_PTR(-ENOMEM
);
6030 * O_TMPFILE, set link count to 0, so that after this point,
6031 * we fill in an inode item with the correct link count.
6034 set_nlink(inode
, 0);
6037 * we have to initialize this early, so we can reclaim the inode
6038 * number if we fail afterwards in this function.
6040 inode
->i_ino
= objectid
;
6043 trace_btrfs_inode_request(dir
);
6045 ret
= btrfs_set_inode_index(dir
, index
);
6047 btrfs_free_path(path
);
6049 return ERR_PTR(ret
);
6055 * index_cnt is ignored for everything but a dir,
6056 * btrfs_get_inode_index_count has an explanation for the magic
6059 BTRFS_I(inode
)->index_cnt
= 2;
6060 BTRFS_I(inode
)->dir_index
= *index
;
6061 BTRFS_I(inode
)->root
= root
;
6062 BTRFS_I(inode
)->generation
= trans
->transid
;
6063 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6066 * We could have gotten an inode number from somebody who was fsynced
6067 * and then removed in this same transaction, so let's just set full
6068 * sync since it will be a full sync anyway and this will blow away the
6069 * old info in the log.
6071 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6073 key
[0].objectid
= objectid
;
6074 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6077 sizes
[0] = sizeof(struct btrfs_inode_item
);
6081 * Start new inodes with an inode_ref. This is slightly more
6082 * efficient for small numbers of hard links since they will
6083 * be packed into one item. Extended refs will kick in if we
6084 * add more hard links than can fit in the ref item.
6086 key
[1].objectid
= objectid
;
6087 key
[1].type
= BTRFS_INODE_REF_KEY
;
6088 key
[1].offset
= ref_objectid
;
6090 sizes
[1] = name_len
+ sizeof(*ref
);
6093 location
= &BTRFS_I(inode
)->location
;
6094 location
->objectid
= objectid
;
6095 location
->offset
= 0;
6096 location
->type
= BTRFS_INODE_ITEM_KEY
;
6098 ret
= btrfs_insert_inode_locked(inode
);
6102 path
->leave_spinning
= 1;
6103 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6107 inode_init_owner(inode
, dir
, mode
);
6108 inode_set_bytes(inode
, 0);
6110 inode
->i_mtime
= CURRENT_TIME
;
6111 inode
->i_atime
= inode
->i_mtime
;
6112 inode
->i_ctime
= inode
->i_mtime
;
6113 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6115 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6116 struct btrfs_inode_item
);
6117 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
6118 sizeof(*inode_item
));
6119 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6122 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6123 struct btrfs_inode_ref
);
6124 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6125 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6126 ptr
= (unsigned long)(ref
+ 1);
6127 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6130 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6131 btrfs_free_path(path
);
6133 btrfs_inherit_iflags(inode
, dir
);
6135 if (S_ISREG(mode
)) {
6136 if (btrfs_test_opt(root
, NODATASUM
))
6137 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6138 if (btrfs_test_opt(root
, NODATACOW
))
6139 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6140 BTRFS_INODE_NODATASUM
;
6143 inode_tree_add(inode
);
6145 trace_btrfs_inode_new(inode
);
6146 btrfs_set_inode_last_trans(trans
, inode
);
6148 btrfs_update_root_times(trans
, root
);
6150 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6152 btrfs_err(root
->fs_info
,
6153 "error inheriting props for ino %llu (root %llu): %d",
6154 btrfs_ino(inode
), root
->root_key
.objectid
, ret
);
6159 unlock_new_inode(inode
);
6162 BTRFS_I(dir
)->index_cnt
--;
6163 btrfs_free_path(path
);
6165 return ERR_PTR(ret
);
6168 static inline u8
btrfs_inode_type(struct inode
*inode
)
6170 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6174 * utility function to add 'inode' into 'parent_inode' with
6175 * a give name and a given sequence number.
6176 * if 'add_backref' is true, also insert a backref from the
6177 * inode to the parent directory.
6179 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6180 struct inode
*parent_inode
, struct inode
*inode
,
6181 const char *name
, int name_len
, int add_backref
, u64 index
)
6184 struct btrfs_key key
;
6185 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
6186 u64 ino
= btrfs_ino(inode
);
6187 u64 parent_ino
= btrfs_ino(parent_inode
);
6189 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6190 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
6193 key
.type
= BTRFS_INODE_ITEM_KEY
;
6197 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6198 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
6199 key
.objectid
, root
->root_key
.objectid
,
6200 parent_ino
, index
, name
, name_len
);
6201 } else if (add_backref
) {
6202 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6206 /* Nothing to clean up yet */
6210 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6212 btrfs_inode_type(inode
), index
);
6213 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6216 btrfs_abort_transaction(trans
, root
, ret
);
6220 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
6222 inode_inc_iversion(parent_inode
);
6223 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
6224 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
6226 btrfs_abort_transaction(trans
, root
, ret
);
6230 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6233 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
6234 key
.objectid
, root
->root_key
.objectid
,
6235 parent_ino
, &local_index
, name
, name_len
);
6237 } else if (add_backref
) {
6241 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6242 ino
, parent_ino
, &local_index
);
6247 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6248 struct inode
*dir
, struct dentry
*dentry
,
6249 struct inode
*inode
, int backref
, u64 index
)
6251 int err
= btrfs_add_link(trans
, dir
, inode
,
6252 dentry
->d_name
.name
, dentry
->d_name
.len
,
6259 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6260 umode_t mode
, dev_t rdev
)
6262 struct btrfs_trans_handle
*trans
;
6263 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6264 struct inode
*inode
= NULL
;
6270 if (!new_valid_dev(rdev
))
6274 * 2 for inode item and ref
6276 * 1 for xattr if selinux is on
6278 trans
= btrfs_start_transaction(root
, 5);
6280 return PTR_ERR(trans
);
6282 err
= btrfs_find_free_ino(root
, &objectid
);
6286 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6287 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6289 if (IS_ERR(inode
)) {
6290 err
= PTR_ERR(inode
);
6295 * If the active LSM wants to access the inode during
6296 * d_instantiate it needs these. Smack checks to see
6297 * if the filesystem supports xattrs by looking at the
6300 inode
->i_op
= &btrfs_special_inode_operations
;
6301 init_special_inode(inode
, inode
->i_mode
, rdev
);
6303 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6305 goto out_unlock_inode
;
6307 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6309 goto out_unlock_inode
;
6311 btrfs_update_inode(trans
, root
, inode
);
6312 unlock_new_inode(inode
);
6313 d_instantiate(dentry
, inode
);
6317 btrfs_end_transaction(trans
, root
);
6318 btrfs_balance_delayed_items(root
);
6319 btrfs_btree_balance_dirty(root
);
6321 inode_dec_link_count(inode
);
6328 unlock_new_inode(inode
);
6333 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6334 umode_t mode
, bool excl
)
6336 struct btrfs_trans_handle
*trans
;
6337 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6338 struct inode
*inode
= NULL
;
6339 int drop_inode_on_err
= 0;
6345 * 2 for inode item and ref
6347 * 1 for xattr if selinux is on
6349 trans
= btrfs_start_transaction(root
, 5);
6351 return PTR_ERR(trans
);
6353 err
= btrfs_find_free_ino(root
, &objectid
);
6357 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6358 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6360 if (IS_ERR(inode
)) {
6361 err
= PTR_ERR(inode
);
6364 drop_inode_on_err
= 1;
6366 * If the active LSM wants to access the inode during
6367 * d_instantiate it needs these. Smack checks to see
6368 * if the filesystem supports xattrs by looking at the
6371 inode
->i_fop
= &btrfs_file_operations
;
6372 inode
->i_op
= &btrfs_file_inode_operations
;
6373 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6375 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6377 goto out_unlock_inode
;
6379 err
= btrfs_update_inode(trans
, root
, inode
);
6381 goto out_unlock_inode
;
6383 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6385 goto out_unlock_inode
;
6387 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6388 unlock_new_inode(inode
);
6389 d_instantiate(dentry
, inode
);
6392 btrfs_end_transaction(trans
, root
);
6393 if (err
&& drop_inode_on_err
) {
6394 inode_dec_link_count(inode
);
6397 btrfs_balance_delayed_items(root
);
6398 btrfs_btree_balance_dirty(root
);
6402 unlock_new_inode(inode
);
6407 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6408 struct dentry
*dentry
)
6410 struct btrfs_trans_handle
*trans
;
6411 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6412 struct inode
*inode
= d_inode(old_dentry
);
6417 /* do not allow sys_link's with other subvols of the same device */
6418 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6421 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6424 err
= btrfs_set_inode_index(dir
, &index
);
6429 * 2 items for inode and inode ref
6430 * 2 items for dir items
6431 * 1 item for parent inode
6433 trans
= btrfs_start_transaction(root
, 5);
6434 if (IS_ERR(trans
)) {
6435 err
= PTR_ERR(trans
);
6439 /* There are several dir indexes for this inode, clear the cache. */
6440 BTRFS_I(inode
)->dir_index
= 0ULL;
6442 inode_inc_iversion(inode
);
6443 inode
->i_ctime
= CURRENT_TIME
;
6445 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6447 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
6452 struct dentry
*parent
= dentry
->d_parent
;
6453 err
= btrfs_update_inode(trans
, root
, inode
);
6456 if (inode
->i_nlink
== 1) {
6458 * If new hard link count is 1, it's a file created
6459 * with open(2) O_TMPFILE flag.
6461 err
= btrfs_orphan_del(trans
, inode
);
6465 d_instantiate(dentry
, inode
);
6466 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
6469 btrfs_end_transaction(trans
, root
);
6470 btrfs_balance_delayed_items(root
);
6473 inode_dec_link_count(inode
);
6476 btrfs_btree_balance_dirty(root
);
6480 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6482 struct inode
*inode
= NULL
;
6483 struct btrfs_trans_handle
*trans
;
6484 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6486 int drop_on_err
= 0;
6491 * 2 items for inode and ref
6492 * 2 items for dir items
6493 * 1 for xattr if selinux is on
6495 trans
= btrfs_start_transaction(root
, 5);
6497 return PTR_ERR(trans
);
6499 err
= btrfs_find_free_ino(root
, &objectid
);
6503 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6504 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6505 S_IFDIR
| mode
, &index
);
6506 if (IS_ERR(inode
)) {
6507 err
= PTR_ERR(inode
);
6512 /* these must be set before we unlock the inode */
6513 inode
->i_op
= &btrfs_dir_inode_operations
;
6514 inode
->i_fop
= &btrfs_dir_file_operations
;
6516 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6518 goto out_fail_inode
;
6520 btrfs_i_size_write(inode
, 0);
6521 err
= btrfs_update_inode(trans
, root
, inode
);
6523 goto out_fail_inode
;
6525 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
6526 dentry
->d_name
.len
, 0, index
);
6528 goto out_fail_inode
;
6530 d_instantiate(dentry
, inode
);
6532 * mkdir is special. We're unlocking after we call d_instantiate
6533 * to avoid a race with nfsd calling d_instantiate.
6535 unlock_new_inode(inode
);
6539 btrfs_end_transaction(trans
, root
);
6541 inode_dec_link_count(inode
);
6544 btrfs_balance_delayed_items(root
);
6545 btrfs_btree_balance_dirty(root
);
6549 unlock_new_inode(inode
);
6553 /* Find next extent map of a given extent map, caller needs to ensure locks */
6554 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6556 struct rb_node
*next
;
6558 next
= rb_next(&em
->rb_node
);
6561 return container_of(next
, struct extent_map
, rb_node
);
6564 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6566 struct rb_node
*prev
;
6568 prev
= rb_prev(&em
->rb_node
);
6571 return container_of(prev
, struct extent_map
, rb_node
);
6574 /* helper for btfs_get_extent. Given an existing extent in the tree,
6575 * the existing extent is the nearest extent to map_start,
6576 * and an extent that you want to insert, deal with overlap and insert
6577 * the best fitted new extent into the tree.
6579 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6580 struct extent_map
*existing
,
6581 struct extent_map
*em
,
6584 struct extent_map
*prev
;
6585 struct extent_map
*next
;
6590 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6592 if (existing
->start
> map_start
) {
6594 prev
= prev_extent_map(next
);
6597 next
= next_extent_map(prev
);
6600 start
= prev
? extent_map_end(prev
) : em
->start
;
6601 start
= max_t(u64
, start
, em
->start
);
6602 end
= next
? next
->start
: extent_map_end(em
);
6603 end
= min_t(u64
, end
, extent_map_end(em
));
6604 start_diff
= start
- em
->start
;
6606 em
->len
= end
- start
;
6607 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6608 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6609 em
->block_start
+= start_diff
;
6610 em
->block_len
-= start_diff
;
6612 return add_extent_mapping(em_tree
, em
, 0);
6615 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6616 struct inode
*inode
, struct page
*page
,
6617 size_t pg_offset
, u64 extent_offset
,
6618 struct btrfs_file_extent_item
*item
)
6621 struct extent_buffer
*leaf
= path
->nodes
[0];
6624 unsigned long inline_size
;
6628 WARN_ON(pg_offset
!= 0);
6629 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6630 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6631 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6632 btrfs_item_nr(path
->slots
[0]));
6633 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6636 ptr
= btrfs_file_extent_inline_start(item
);
6638 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6640 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
6641 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6642 extent_offset
, inline_size
, max_size
);
6648 * a bit scary, this does extent mapping from logical file offset to the disk.
6649 * the ugly parts come from merging extents from the disk with the in-ram
6650 * representation. This gets more complex because of the data=ordered code,
6651 * where the in-ram extents might be locked pending data=ordered completion.
6653 * This also copies inline extents directly into the page.
6656 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6657 size_t pg_offset
, u64 start
, u64 len
,
6662 u64 extent_start
= 0;
6664 u64 objectid
= btrfs_ino(inode
);
6666 struct btrfs_path
*path
= NULL
;
6667 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6668 struct btrfs_file_extent_item
*item
;
6669 struct extent_buffer
*leaf
;
6670 struct btrfs_key found_key
;
6671 struct extent_map
*em
= NULL
;
6672 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6673 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6674 struct btrfs_trans_handle
*trans
= NULL
;
6675 const bool new_inline
= !page
|| create
;
6678 read_lock(&em_tree
->lock
);
6679 em
= lookup_extent_mapping(em_tree
, start
, len
);
6681 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6682 read_unlock(&em_tree
->lock
);
6685 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6686 free_extent_map(em
);
6687 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6688 free_extent_map(em
);
6692 em
= alloc_extent_map();
6697 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6698 em
->start
= EXTENT_MAP_HOLE
;
6699 em
->orig_start
= EXTENT_MAP_HOLE
;
6701 em
->block_len
= (u64
)-1;
6704 path
= btrfs_alloc_path();
6710 * Chances are we'll be called again, so go ahead and do
6716 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6717 objectid
, start
, trans
!= NULL
);
6724 if (path
->slots
[0] == 0)
6729 leaf
= path
->nodes
[0];
6730 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6731 struct btrfs_file_extent_item
);
6732 /* are we inside the extent that was found? */
6733 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6734 found_type
= found_key
.type
;
6735 if (found_key
.objectid
!= objectid
||
6736 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6738 * If we backup past the first extent we want to move forward
6739 * and see if there is an extent in front of us, otherwise we'll
6740 * say there is a hole for our whole search range which can
6747 found_type
= btrfs_file_extent_type(leaf
, item
);
6748 extent_start
= found_key
.offset
;
6749 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6750 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6751 extent_end
= extent_start
+
6752 btrfs_file_extent_num_bytes(leaf
, item
);
6753 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6755 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6756 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6759 if (start
>= extent_end
) {
6761 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6762 ret
= btrfs_next_leaf(root
, path
);
6769 leaf
= path
->nodes
[0];
6771 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6772 if (found_key
.objectid
!= objectid
||
6773 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6775 if (start
+ len
<= found_key
.offset
)
6777 if (start
> found_key
.offset
)
6780 em
->orig_start
= start
;
6781 em
->len
= found_key
.offset
- start
;
6785 btrfs_extent_item_to_extent_map(inode
, path
, item
, new_inline
, em
);
6787 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6788 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6790 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6794 size_t extent_offset
;
6800 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6801 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6802 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6803 size
- extent_offset
);
6804 em
->start
= extent_start
+ extent_offset
;
6805 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6806 em
->orig_block_len
= em
->len
;
6807 em
->orig_start
= em
->start
;
6808 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6809 if (create
== 0 && !PageUptodate(page
)) {
6810 if (btrfs_file_extent_compression(leaf
, item
) !=
6811 BTRFS_COMPRESS_NONE
) {
6812 ret
= uncompress_inline(path
, inode
, page
,
6814 extent_offset
, item
);
6821 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6823 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6824 memset(map
+ pg_offset
+ copy_size
, 0,
6825 PAGE_CACHE_SIZE
- pg_offset
-
6830 flush_dcache_page(page
);
6831 } else if (create
&& PageUptodate(page
)) {
6835 free_extent_map(em
);
6838 btrfs_release_path(path
);
6839 trans
= btrfs_join_transaction(root
);
6842 return ERR_CAST(trans
);
6846 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6849 btrfs_mark_buffer_dirty(leaf
);
6851 set_extent_uptodate(io_tree
, em
->start
,
6852 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6857 em
->orig_start
= start
;
6860 em
->block_start
= EXTENT_MAP_HOLE
;
6861 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6863 btrfs_release_path(path
);
6864 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6865 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6866 em
->start
, em
->len
, start
, len
);
6872 write_lock(&em_tree
->lock
);
6873 ret
= add_extent_mapping(em_tree
, em
, 0);
6874 /* it is possible that someone inserted the extent into the tree
6875 * while we had the lock dropped. It is also possible that
6876 * an overlapping map exists in the tree
6878 if (ret
== -EEXIST
) {
6879 struct extent_map
*existing
;
6883 existing
= search_extent_mapping(em_tree
, start
, len
);
6885 * existing will always be non-NULL, since there must be
6886 * extent causing the -EEXIST.
6888 if (start
>= extent_map_end(existing
) ||
6889 start
<= existing
->start
) {
6891 * The existing extent map is the one nearest to
6892 * the [start, start + len) range which overlaps
6894 err
= merge_extent_mapping(em_tree
, existing
,
6896 free_extent_map(existing
);
6898 free_extent_map(em
);
6902 free_extent_map(em
);
6907 write_unlock(&em_tree
->lock
);
6910 trace_btrfs_get_extent(root
, em
);
6912 btrfs_free_path(path
);
6914 ret
= btrfs_end_transaction(trans
, root
);
6919 free_extent_map(em
);
6920 return ERR_PTR(err
);
6922 BUG_ON(!em
); /* Error is always set */
6926 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
6927 size_t pg_offset
, u64 start
, u64 len
,
6930 struct extent_map
*em
;
6931 struct extent_map
*hole_em
= NULL
;
6932 u64 range_start
= start
;
6938 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
6945 * - a pre-alloc extent,
6946 * there might actually be delalloc bytes behind it.
6948 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
6949 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6955 /* check to see if we've wrapped (len == -1 or similar) */
6964 /* ok, we didn't find anything, lets look for delalloc */
6965 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
6966 end
, len
, EXTENT_DELALLOC
, 1);
6967 found_end
= range_start
+ found
;
6968 if (found_end
< range_start
)
6969 found_end
= (u64
)-1;
6972 * we didn't find anything useful, return
6973 * the original results from get_extent()
6975 if (range_start
> end
|| found_end
<= start
) {
6981 /* adjust the range_start to make sure it doesn't
6982 * go backwards from the start they passed in
6984 range_start
= max(start
, range_start
);
6985 found
= found_end
- range_start
;
6988 u64 hole_start
= start
;
6991 em
= alloc_extent_map();
6997 * when btrfs_get_extent can't find anything it
6998 * returns one huge hole
7000 * make sure what it found really fits our range, and
7001 * adjust to make sure it is based on the start from
7005 u64 calc_end
= extent_map_end(hole_em
);
7007 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7008 free_extent_map(hole_em
);
7011 hole_start
= max(hole_em
->start
, start
);
7012 hole_len
= calc_end
- hole_start
;
7016 if (hole_em
&& range_start
> hole_start
) {
7017 /* our hole starts before our delalloc, so we
7018 * have to return just the parts of the hole
7019 * that go until the delalloc starts
7021 em
->len
= min(hole_len
,
7022 range_start
- hole_start
);
7023 em
->start
= hole_start
;
7024 em
->orig_start
= hole_start
;
7026 * don't adjust block start at all,
7027 * it is fixed at EXTENT_MAP_HOLE
7029 em
->block_start
= hole_em
->block_start
;
7030 em
->block_len
= hole_len
;
7031 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7032 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7034 em
->start
= range_start
;
7036 em
->orig_start
= range_start
;
7037 em
->block_start
= EXTENT_MAP_DELALLOC
;
7038 em
->block_len
= found
;
7040 } else if (hole_em
) {
7045 free_extent_map(hole_em
);
7047 free_extent_map(em
);
7048 return ERR_PTR(err
);
7053 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7056 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7057 struct extent_map
*em
;
7058 struct btrfs_key ins
;
7062 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7063 ret
= btrfs_reserve_extent(root
, len
, root
->sectorsize
, 0,
7064 alloc_hint
, &ins
, 1, 1);
7066 return ERR_PTR(ret
);
7068 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
7069 ins
.offset
, ins
.offset
, ins
.offset
, 0);
7071 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7075 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
7076 ins
.offset
, ins
.offset
, 0);
7078 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7079 free_extent_map(em
);
7080 return ERR_PTR(ret
);
7087 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7088 * block must be cow'd
7090 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7091 u64
*orig_start
, u64
*orig_block_len
,
7094 struct btrfs_trans_handle
*trans
;
7095 struct btrfs_path
*path
;
7097 struct extent_buffer
*leaf
;
7098 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7099 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7100 struct btrfs_file_extent_item
*fi
;
7101 struct btrfs_key key
;
7108 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7110 path
= btrfs_alloc_path();
7114 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
7119 slot
= path
->slots
[0];
7122 /* can't find the item, must cow */
7129 leaf
= path
->nodes
[0];
7130 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7131 if (key
.objectid
!= btrfs_ino(inode
) ||
7132 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7133 /* not our file or wrong item type, must cow */
7137 if (key
.offset
> offset
) {
7138 /* Wrong offset, must cow */
7142 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7143 found_type
= btrfs_file_extent_type(leaf
, fi
);
7144 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7145 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7146 /* not a regular extent, must cow */
7150 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7153 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7154 if (extent_end
<= offset
)
7157 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7158 if (disk_bytenr
== 0)
7161 if (btrfs_file_extent_compression(leaf
, fi
) ||
7162 btrfs_file_extent_encryption(leaf
, fi
) ||
7163 btrfs_file_extent_other_encoding(leaf
, fi
))
7166 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7169 *orig_start
= key
.offset
- backref_offset
;
7170 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7171 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7174 if (btrfs_extent_readonly(root
, disk_bytenr
))
7177 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7178 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7181 range_end
= round_up(offset
+ num_bytes
, root
->sectorsize
) - 1;
7182 ret
= test_range_bit(io_tree
, offset
, range_end
,
7183 EXTENT_DELALLOC
, 0, NULL
);
7190 btrfs_release_path(path
);
7193 * look for other files referencing this extent, if we
7194 * find any we must cow
7196 trans
= btrfs_join_transaction(root
);
7197 if (IS_ERR(trans
)) {
7202 ret
= btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
7203 key
.offset
- backref_offset
, disk_bytenr
);
7204 btrfs_end_transaction(trans
, root
);
7211 * adjust disk_bytenr and num_bytes to cover just the bytes
7212 * in this extent we are about to write. If there
7213 * are any csums in that range we have to cow in order
7214 * to keep the csums correct
7216 disk_bytenr
+= backref_offset
;
7217 disk_bytenr
+= offset
- key
.offset
;
7218 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
7221 * all of the above have passed, it is safe to overwrite this extent
7227 btrfs_free_path(path
);
7231 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7233 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7235 void **pagep
= NULL
;
7236 struct page
*page
= NULL
;
7240 start_idx
= start
>> PAGE_CACHE_SHIFT
;
7243 * end is the last byte in the last page. end == start is legal
7245 end_idx
= end
>> PAGE_CACHE_SHIFT
;
7249 /* Most of the code in this while loop is lifted from
7250 * find_get_page. It's been modified to begin searching from a
7251 * page and return just the first page found in that range. If the
7252 * found idx is less than or equal to the end idx then we know that
7253 * a page exists. If no pages are found or if those pages are
7254 * outside of the range then we're fine (yay!) */
7255 while (page
== NULL
&&
7256 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7257 page
= radix_tree_deref_slot(pagep
);
7258 if (unlikely(!page
))
7261 if (radix_tree_exception(page
)) {
7262 if (radix_tree_deref_retry(page
)) {
7267 * Otherwise, shmem/tmpfs must be storing a swap entry
7268 * here as an exceptional entry: so return it without
7269 * attempting to raise page count.
7272 break; /* TODO: Is this relevant for this use case? */
7275 if (!page_cache_get_speculative(page
)) {
7281 * Has the page moved?
7282 * This is part of the lockless pagecache protocol. See
7283 * include/linux/pagemap.h for details.
7285 if (unlikely(page
!= *pagep
)) {
7286 page_cache_release(page
);
7292 if (page
->index
<= end_idx
)
7294 page_cache_release(page
);
7301 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7302 struct extent_state
**cached_state
, int writing
)
7304 struct btrfs_ordered_extent
*ordered
;
7308 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7311 * We're concerned with the entire range that we're going to be
7312 * doing DIO to, so we need to make sure theres no ordered
7313 * extents in this range.
7315 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
7316 lockend
- lockstart
+ 1);
7319 * We need to make sure there are no buffered pages in this
7320 * range either, we could have raced between the invalidate in
7321 * generic_file_direct_write and locking the extent. The
7322 * invalidate needs to happen so that reads after a write do not
7327 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7330 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7331 cached_state
, GFP_NOFS
);
7334 btrfs_start_ordered_extent(inode
, ordered
, 1);
7335 btrfs_put_ordered_extent(ordered
);
7337 /* Screw you mmap */
7338 ret
= btrfs_fdatawrite_range(inode
, lockstart
, lockend
);
7341 ret
= filemap_fdatawait_range(inode
->i_mapping
,
7348 * If we found a page that couldn't be invalidated just
7349 * fall back to buffered.
7351 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
7352 lockstart
>> PAGE_CACHE_SHIFT
,
7353 lockend
>> PAGE_CACHE_SHIFT
);
7364 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
7365 u64 len
, u64 orig_start
,
7366 u64 block_start
, u64 block_len
,
7367 u64 orig_block_len
, u64 ram_bytes
,
7370 struct extent_map_tree
*em_tree
;
7371 struct extent_map
*em
;
7372 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7375 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7376 em
= alloc_extent_map();
7378 return ERR_PTR(-ENOMEM
);
7381 em
->orig_start
= orig_start
;
7382 em
->mod_start
= start
;
7385 em
->block_len
= block_len
;
7386 em
->block_start
= block_start
;
7387 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7388 em
->orig_block_len
= orig_block_len
;
7389 em
->ram_bytes
= ram_bytes
;
7390 em
->generation
= -1;
7391 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7392 if (type
== BTRFS_ORDERED_PREALLOC
)
7393 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7396 btrfs_drop_extent_cache(inode
, em
->start
,
7397 em
->start
+ em
->len
- 1, 0);
7398 write_lock(&em_tree
->lock
);
7399 ret
= add_extent_mapping(em_tree
, em
, 1);
7400 write_unlock(&em_tree
->lock
);
7401 } while (ret
== -EEXIST
);
7404 free_extent_map(em
);
7405 return ERR_PTR(ret
);
7411 struct btrfs_dio_data
{
7412 u64 outstanding_extents
;
7416 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7417 struct buffer_head
*bh_result
, int create
)
7419 struct extent_map
*em
;
7420 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7421 struct extent_state
*cached_state
= NULL
;
7422 struct btrfs_dio_data
*dio_data
= NULL
;
7423 u64 start
= iblock
<< inode
->i_blkbits
;
7424 u64 lockstart
, lockend
;
7425 u64 len
= bh_result
->b_size
;
7426 int unlock_bits
= EXTENT_LOCKED
;
7430 unlock_bits
|= EXTENT_DIRTY
;
7432 len
= min_t(u64
, len
, root
->sectorsize
);
7435 lockend
= start
+ len
- 1;
7437 if (current
->journal_info
) {
7439 * Need to pull our outstanding extents and set journal_info to NULL so
7440 * that anything that needs to check if there's a transction doesn't get
7443 dio_data
= current
->journal_info
;
7444 current
->journal_info
= NULL
;
7448 * If this errors out it's because we couldn't invalidate pagecache for
7449 * this range and we need to fallback to buffered.
7451 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
, create
))
7454 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7461 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7462 * io. INLINE is special, and we could probably kludge it in here, but
7463 * it's still buffered so for safety lets just fall back to the generic
7466 * For COMPRESSED we _have_ to read the entire extent in so we can
7467 * decompress it, so there will be buffering required no matter what we
7468 * do, so go ahead and fallback to buffered.
7470 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7471 * to buffered IO. Don't blame me, this is the price we pay for using
7474 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7475 em
->block_start
== EXTENT_MAP_INLINE
) {
7476 free_extent_map(em
);
7481 /* Just a good old fashioned hole, return */
7482 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7483 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7484 free_extent_map(em
);
7489 * We don't allocate a new extent in the following cases
7491 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7493 * 2) The extent is marked as PREALLOC. We're good to go here and can
7494 * just use the extent.
7498 len
= min(len
, em
->len
- (start
- em
->start
));
7499 lockstart
= start
+ len
;
7503 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7504 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7505 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7507 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7509 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7510 type
= BTRFS_ORDERED_PREALLOC
;
7512 type
= BTRFS_ORDERED_NOCOW
;
7513 len
= min(len
, em
->len
- (start
- em
->start
));
7514 block_start
= em
->block_start
+ (start
- em
->start
);
7516 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7517 &orig_block_len
, &ram_bytes
) == 1) {
7518 if (type
== BTRFS_ORDERED_PREALLOC
) {
7519 free_extent_map(em
);
7520 em
= create_pinned_em(inode
, start
, len
,
7531 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
7532 block_start
, len
, len
, type
);
7534 free_extent_map(em
);
7542 * this will cow the extent, reset the len in case we changed
7545 len
= bh_result
->b_size
;
7546 free_extent_map(em
);
7547 em
= btrfs_new_extent_direct(inode
, start
, len
);
7552 len
= min(len
, em
->len
- (start
- em
->start
));
7554 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7556 bh_result
->b_size
= len
;
7557 bh_result
->b_bdev
= em
->bdev
;
7558 set_buffer_mapped(bh_result
);
7560 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7561 set_buffer_new(bh_result
);
7564 * Need to update the i_size under the extent lock so buffered
7565 * readers will get the updated i_size when we unlock.
7567 if (start
+ len
> i_size_read(inode
))
7568 i_size_write(inode
, start
+ len
);
7571 * If we have an outstanding_extents count still set then we're
7572 * within our reservation, otherwise we need to adjust our inode
7573 * counter appropriately.
7575 if (dio_data
->outstanding_extents
) {
7576 (dio_data
->outstanding_extents
)--;
7578 spin_lock(&BTRFS_I(inode
)->lock
);
7579 BTRFS_I(inode
)->outstanding_extents
++;
7580 spin_unlock(&BTRFS_I(inode
)->lock
);
7583 btrfs_free_reserved_data_space(inode
, len
);
7584 WARN_ON(dio_data
->reserve
< len
);
7585 dio_data
->reserve
-= len
;
7586 current
->journal_info
= dio_data
;
7590 * In the case of write we need to clear and unlock the entire range,
7591 * in the case of read we need to unlock only the end area that we
7592 * aren't using if there is any left over space.
7594 if (lockstart
< lockend
) {
7595 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7596 lockend
, unlock_bits
, 1, 0,
7597 &cached_state
, GFP_NOFS
);
7599 free_extent_state(cached_state
);
7602 free_extent_map(em
);
7607 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7608 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
7610 current
->journal_info
= dio_data
;
7614 static inline int submit_dio_repair_bio(struct inode
*inode
, struct bio
*bio
,
7615 int rw
, int mirror_num
)
7617 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7620 BUG_ON(rw
& REQ_WRITE
);
7624 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
7625 BTRFS_WQ_ENDIO_DIO_REPAIR
);
7629 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
7635 static int btrfs_check_dio_repairable(struct inode
*inode
,
7636 struct bio
*failed_bio
,
7637 struct io_failure_record
*failrec
,
7642 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
7643 failrec
->logical
, failrec
->len
);
7644 if (num_copies
== 1) {
7646 * we only have a single copy of the data, so don't bother with
7647 * all the retry and error correction code that follows. no
7648 * matter what the error is, it is very likely to persist.
7650 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7651 num_copies
, failrec
->this_mirror
, failed_mirror
);
7655 failrec
->failed_mirror
= failed_mirror
;
7656 failrec
->this_mirror
++;
7657 if (failrec
->this_mirror
== failed_mirror
)
7658 failrec
->this_mirror
++;
7660 if (failrec
->this_mirror
> num_copies
) {
7661 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7662 num_copies
, failrec
->this_mirror
, failed_mirror
);
7669 static int dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7670 struct page
*page
, u64 start
, u64 end
,
7671 int failed_mirror
, bio_end_io_t
*repair_endio
,
7674 struct io_failure_record
*failrec
;
7680 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
7682 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7686 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7689 free_io_failure(inode
, failrec
);
7693 if (failed_bio
->bi_vcnt
> 1)
7694 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
7696 read_mode
= READ_SYNC
;
7698 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7699 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7700 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7701 0, isector
, repair_endio
, repair_arg
);
7703 free_io_failure(inode
, failrec
);
7707 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7708 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7709 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7711 ret
= submit_dio_repair_bio(inode
, bio
, read_mode
,
7712 failrec
->this_mirror
);
7714 free_io_failure(inode
, failrec
);
7721 struct btrfs_retry_complete
{
7722 struct completion done
;
7723 struct inode
*inode
;
7728 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7730 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7731 struct bio_vec
*bvec
;
7738 bio_for_each_segment_all(bvec
, bio
, i
)
7739 clean_io_failure(done
->inode
, done
->start
, bvec
->bv_page
, 0);
7741 complete(&done
->done
);
7745 static int __btrfs_correct_data_nocsum(struct inode
*inode
,
7746 struct btrfs_io_bio
*io_bio
)
7748 struct bio_vec
*bvec
;
7749 struct btrfs_retry_complete done
;
7754 start
= io_bio
->logical
;
7757 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7761 init_completion(&done
.done
);
7763 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7764 start
+ bvec
->bv_len
- 1,
7766 btrfs_retry_endio_nocsum
, &done
);
7770 wait_for_completion(&done
.done
);
7772 if (!done
.uptodate
) {
7773 /* We might have another mirror, so try again */
7777 start
+= bvec
->bv_len
;
7783 static void btrfs_retry_endio(struct bio
*bio
)
7785 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7786 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7787 struct bio_vec
*bvec
;
7796 bio_for_each_segment_all(bvec
, bio
, i
) {
7797 ret
= __readpage_endio_check(done
->inode
, io_bio
, i
,
7799 done
->start
, bvec
->bv_len
);
7801 clean_io_failure(done
->inode
, done
->start
,
7807 done
->uptodate
= uptodate
;
7809 complete(&done
->done
);
7813 static int __btrfs_subio_endio_read(struct inode
*inode
,
7814 struct btrfs_io_bio
*io_bio
, int err
)
7816 struct bio_vec
*bvec
;
7817 struct btrfs_retry_complete done
;
7824 start
= io_bio
->logical
;
7827 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7828 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
7829 0, start
, bvec
->bv_len
);
7835 init_completion(&done
.done
);
7837 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7838 start
+ bvec
->bv_len
- 1,
7840 btrfs_retry_endio
, &done
);
7846 wait_for_completion(&done
.done
);
7848 if (!done
.uptodate
) {
7849 /* We might have another mirror, so try again */
7853 offset
+= bvec
->bv_len
;
7854 start
+= bvec
->bv_len
;
7860 static int btrfs_subio_endio_read(struct inode
*inode
,
7861 struct btrfs_io_bio
*io_bio
, int err
)
7863 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
7867 return __btrfs_correct_data_nocsum(inode
, io_bio
);
7871 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
7875 static void btrfs_endio_direct_read(struct bio
*bio
)
7877 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7878 struct inode
*inode
= dip
->inode
;
7879 struct bio
*dio_bio
;
7880 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7881 int err
= bio
->bi_error
;
7883 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
7884 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
7886 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
7887 dip
->logical_offset
+ dip
->bytes
- 1);
7888 dio_bio
= dip
->dio_bio
;
7892 dio_end_io(dio_bio
, bio
->bi_error
);
7895 io_bio
->end_io(io_bio
, err
);
7899 static void btrfs_endio_direct_write(struct bio
*bio
)
7901 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7902 struct inode
*inode
= dip
->inode
;
7903 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7904 struct btrfs_ordered_extent
*ordered
= NULL
;
7905 u64 ordered_offset
= dip
->logical_offset
;
7906 u64 ordered_bytes
= dip
->bytes
;
7907 struct bio
*dio_bio
;
7911 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
7918 btrfs_init_work(&ordered
->work
, btrfs_endio_write_helper
,
7919 finish_ordered_fn
, NULL
, NULL
);
7920 btrfs_queue_work(root
->fs_info
->endio_write_workers
,
7924 * our bio might span multiple ordered extents. If we haven't
7925 * completed the accounting for the whole dio, go back and try again
7927 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
7928 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
7933 dio_bio
= dip
->dio_bio
;
7937 dio_end_io(dio_bio
, bio
->bi_error
);
7941 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
7942 struct bio
*bio
, int mirror_num
,
7943 unsigned long bio_flags
, u64 offset
)
7946 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7947 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
7948 BUG_ON(ret
); /* -ENOMEM */
7952 static void btrfs_end_dio_bio(struct bio
*bio
)
7954 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7955 int err
= bio
->bi_error
;
7958 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
7959 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7960 btrfs_ino(dip
->inode
), bio
->bi_rw
,
7961 (unsigned long long)bio
->bi_iter
.bi_sector
,
7962 bio
->bi_iter
.bi_size
, err
);
7964 if (dip
->subio_endio
)
7965 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
7971 * before atomic variable goto zero, we must make sure
7972 * dip->errors is perceived to be set.
7974 smp_mb__before_atomic();
7977 /* if there are more bios still pending for this dio, just exit */
7978 if (!atomic_dec_and_test(&dip
->pending_bios
))
7982 bio_io_error(dip
->orig_bio
);
7984 dip
->dio_bio
->bi_error
= 0;
7985 bio_endio(dip
->orig_bio
);
7991 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
7992 u64 first_sector
, gfp_t gfp_flags
)
7995 bio
= btrfs_bio_alloc(bdev
, first_sector
, BIO_MAX_PAGES
, gfp_flags
);
7997 bio_associate_current(bio
);
8001 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root
*root
,
8002 struct inode
*inode
,
8003 struct btrfs_dio_private
*dip
,
8007 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8008 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8012 * We load all the csum data we need when we submit
8013 * the first bio to reduce the csum tree search and
8016 if (dip
->logical_offset
== file_offset
) {
8017 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, dip
->orig_bio
,
8023 if (bio
== dip
->orig_bio
)
8026 file_offset
-= dip
->logical_offset
;
8027 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8028 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8033 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
8034 int rw
, u64 file_offset
, int skip_sum
,
8037 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8038 int write
= rw
& REQ_WRITE
;
8039 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8043 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8048 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
8049 BTRFS_WQ_ENDIO_DATA
);
8057 if (write
&& async_submit
) {
8058 ret
= btrfs_wq_submit_bio(root
->fs_info
,
8059 inode
, rw
, bio
, 0, 0,
8061 __btrfs_submit_bio_start_direct_io
,
8062 __btrfs_submit_bio_done
);
8066 * If we aren't doing async submit, calculate the csum of the
8069 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
8073 ret
= btrfs_lookup_and_bind_dio_csum(root
, inode
, dip
, bio
,
8079 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
8085 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
8088 struct inode
*inode
= dip
->inode
;
8089 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8091 struct bio
*orig_bio
= dip
->orig_bio
;
8092 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
8093 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8094 u64 file_offset
= dip
->logical_offset
;
8099 int async_submit
= 0;
8101 map_length
= orig_bio
->bi_iter
.bi_size
;
8102 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
8103 &map_length
, NULL
, 0);
8107 if (map_length
>= orig_bio
->bi_iter
.bi_size
) {
8109 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8113 /* async crcs make it difficult to collect full stripe writes. */
8114 if (btrfs_get_alloc_profile(root
, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8119 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
8123 bio
->bi_private
= dip
;
8124 bio
->bi_end_io
= btrfs_end_dio_bio
;
8125 btrfs_io_bio(bio
)->logical
= file_offset
;
8126 atomic_inc(&dip
->pending_bios
);
8128 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
8129 if (map_length
< submit_len
+ bvec
->bv_len
||
8130 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
8131 bvec
->bv_offset
) < bvec
->bv_len
) {
8133 * inc the count before we submit the bio so
8134 * we know the end IO handler won't happen before
8135 * we inc the count. Otherwise, the dip might get freed
8136 * before we're done setting it up
8138 atomic_inc(&dip
->pending_bios
);
8139 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
8140 file_offset
, skip_sum
,
8144 atomic_dec(&dip
->pending_bios
);
8148 start_sector
+= submit_len
>> 9;
8149 file_offset
+= submit_len
;
8154 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
8155 start_sector
, GFP_NOFS
);
8158 bio
->bi_private
= dip
;
8159 bio
->bi_end_io
= btrfs_end_dio_bio
;
8160 btrfs_io_bio(bio
)->logical
= file_offset
;
8162 map_length
= orig_bio
->bi_iter
.bi_size
;
8163 ret
= btrfs_map_block(root
->fs_info
, rw
,
8165 &map_length
, NULL
, 0);
8171 submit_len
+= bvec
->bv_len
;
8178 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
8187 * before atomic variable goto zero, we must
8188 * make sure dip->errors is perceived to be set.
8190 smp_mb__before_atomic();
8191 if (atomic_dec_and_test(&dip
->pending_bios
))
8192 bio_io_error(dip
->orig_bio
);
8194 /* bio_end_io() will handle error, so we needn't return it */
8198 static void btrfs_submit_direct(int rw
, struct bio
*dio_bio
,
8199 struct inode
*inode
, loff_t file_offset
)
8201 struct btrfs_dio_private
*dip
= NULL
;
8202 struct bio
*io_bio
= NULL
;
8203 struct btrfs_io_bio
*btrfs_bio
;
8205 int write
= rw
& REQ_WRITE
;
8208 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8210 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
8216 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8222 dip
->private = dio_bio
->bi_private
;
8224 dip
->logical_offset
= file_offset
;
8225 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8226 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8227 io_bio
->bi_private
= dip
;
8228 dip
->orig_bio
= io_bio
;
8229 dip
->dio_bio
= dio_bio
;
8230 atomic_set(&dip
->pending_bios
, 0);
8231 btrfs_bio
= btrfs_io_bio(io_bio
);
8232 btrfs_bio
->logical
= file_offset
;
8235 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
8237 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
8238 dip
->subio_endio
= btrfs_subio_endio_read
;
8241 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
8245 if (btrfs_bio
->end_io
)
8246 btrfs_bio
->end_io(btrfs_bio
, ret
);
8250 * If we arrived here it means either we failed to submit the dip
8251 * or we either failed to clone the dio_bio or failed to allocate the
8252 * dip. If we cloned the dio_bio and allocated the dip, we can just
8253 * call bio_endio against our io_bio so that we get proper resource
8254 * cleanup if we fail to submit the dip, otherwise, we must do the
8255 * same as btrfs_endio_direct_[write|read] because we can't call these
8256 * callbacks - they require an allocated dip and a clone of dio_bio.
8258 if (io_bio
&& dip
) {
8259 io_bio
->bi_error
= -EIO
;
8262 * The end io callbacks free our dip, do the final put on io_bio
8263 * and all the cleanup and final put for dio_bio (through
8270 struct btrfs_ordered_extent
*ordered
;
8272 ordered
= btrfs_lookup_ordered_extent(inode
,
8274 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
8276 * Decrements our ref on the ordered extent and removes
8277 * the ordered extent from the inode's ordered tree,
8278 * doing all the proper resource cleanup such as for the
8279 * reserved space and waking up any waiters for this
8280 * ordered extent (through btrfs_remove_ordered_extent).
8282 btrfs_finish_ordered_io(ordered
);
8284 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8285 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8287 dio_bio
->bi_error
= -EIO
;
8289 * Releases and cleans up our dio_bio, no need to bio_put()
8290 * nor bio_endio()/bio_io_error() against dio_bio.
8292 dio_end_io(dio_bio
, ret
);
8299 static ssize_t
check_direct_IO(struct btrfs_root
*root
, struct kiocb
*iocb
,
8300 const struct iov_iter
*iter
, loff_t offset
)
8304 unsigned blocksize_mask
= root
->sectorsize
- 1;
8305 ssize_t retval
= -EINVAL
;
8307 if (offset
& blocksize_mask
)
8310 if (iov_iter_alignment(iter
) & blocksize_mask
)
8313 /* If this is a write we don't need to check anymore */
8314 if (iov_iter_rw(iter
) == WRITE
)
8317 * Check to make sure we don't have duplicate iov_base's in this
8318 * iovec, if so return EINVAL, otherwise we'll get csum errors
8319 * when reading back.
8321 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8322 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8323 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8332 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
8335 struct file
*file
= iocb
->ki_filp
;
8336 struct inode
*inode
= file
->f_mapping
->host
;
8337 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8338 struct btrfs_dio_data dio_data
= { 0 };
8342 bool relock
= false;
8345 if (check_direct_IO(BTRFS_I(inode
)->root
, iocb
, iter
, offset
))
8348 inode_dio_begin(inode
);
8349 smp_mb__after_atomic();
8352 * The generic stuff only does filemap_write_and_wait_range, which
8353 * isn't enough if we've written compressed pages to this area, so
8354 * we need to flush the dirty pages again to make absolutely sure
8355 * that any outstanding dirty pages are on disk.
8357 count
= iov_iter_count(iter
);
8358 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8359 &BTRFS_I(inode
)->runtime_flags
))
8360 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8361 offset
+ count
- 1);
8363 if (iov_iter_rw(iter
) == WRITE
) {
8365 * If the write DIO is beyond the EOF, we need update
8366 * the isize, but it is protected by i_mutex. So we can
8367 * not unlock the i_mutex at this case.
8369 if (offset
+ count
<= inode
->i_size
) {
8370 mutex_unlock(&inode
->i_mutex
);
8373 ret
= btrfs_delalloc_reserve_space(inode
, count
);
8376 dio_data
.outstanding_extents
= div64_u64(count
+
8377 BTRFS_MAX_EXTENT_SIZE
- 1,
8378 BTRFS_MAX_EXTENT_SIZE
);
8381 * We need to know how many extents we reserved so that we can
8382 * do the accounting properly if we go over the number we
8383 * originally calculated. Abuse current->journal_info for this.
8385 dio_data
.reserve
= round_up(count
, root
->sectorsize
);
8386 current
->journal_info
= &dio_data
;
8387 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8388 &BTRFS_I(inode
)->runtime_flags
)) {
8389 inode_dio_end(inode
);
8390 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8394 ret
= __blockdev_direct_IO(iocb
, inode
,
8395 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
8396 iter
, offset
, btrfs_get_blocks_direct
, NULL
,
8397 btrfs_submit_direct
, flags
);
8398 if (iov_iter_rw(iter
) == WRITE
) {
8399 current
->journal_info
= NULL
;
8400 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8401 if (dio_data
.reserve
)
8402 btrfs_delalloc_release_space(inode
,
8404 } else if (ret
>= 0 && (size_t)ret
< count
)
8405 btrfs_delalloc_release_space(inode
,
8406 count
- (size_t)ret
);
8410 inode_dio_end(inode
);
8412 mutex_lock(&inode
->i_mutex
);
8417 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8419 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8420 __u64 start
, __u64 len
)
8424 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8428 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
8431 int btrfs_readpage(struct file
*file
, struct page
*page
)
8433 struct extent_io_tree
*tree
;
8434 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8435 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8438 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8440 struct extent_io_tree
*tree
;
8443 if (current
->flags
& PF_MEMALLOC
) {
8444 redirty_page_for_writepage(wbc
, page
);
8448 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8449 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
8452 static int btrfs_writepages(struct address_space
*mapping
,
8453 struct writeback_control
*wbc
)
8455 struct extent_io_tree
*tree
;
8457 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8458 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
8462 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8463 struct list_head
*pages
, unsigned nr_pages
)
8465 struct extent_io_tree
*tree
;
8466 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8467 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
8470 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8472 struct extent_io_tree
*tree
;
8473 struct extent_map_tree
*map
;
8476 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8477 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
8478 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
8480 ClearPagePrivate(page
);
8481 set_page_private(page
, 0);
8482 page_cache_release(page
);
8487 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8489 if (PageWriteback(page
) || PageDirty(page
))
8491 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
8494 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8495 unsigned int length
)
8497 struct inode
*inode
= page
->mapping
->host
;
8498 struct extent_io_tree
*tree
;
8499 struct btrfs_ordered_extent
*ordered
;
8500 struct extent_state
*cached_state
= NULL
;
8501 u64 page_start
= page_offset(page
);
8502 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8503 int inode_evicting
= inode
->i_state
& I_FREEING
;
8506 * we have the page locked, so new writeback can't start,
8507 * and the dirty bit won't be cleared while we are here.
8509 * Wait for IO on this page so that we can safely clear
8510 * the PagePrivate2 bit and do ordered accounting
8512 wait_on_page_writeback(page
);
8514 tree
= &BTRFS_I(inode
)->io_tree
;
8516 btrfs_releasepage(page
, GFP_NOFS
);
8520 if (!inode_evicting
)
8521 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
8522 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8525 * IO on this page will never be started, so we need
8526 * to account for any ordered extents now
8528 if (!inode_evicting
)
8529 clear_extent_bit(tree
, page_start
, page_end
,
8530 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8531 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8532 EXTENT_DEFRAG
, 1, 0, &cached_state
,
8535 * whoever cleared the private bit is responsible
8536 * for the finish_ordered_io
8538 if (TestClearPagePrivate2(page
)) {
8539 struct btrfs_ordered_inode_tree
*tree
;
8542 tree
= &BTRFS_I(inode
)->ordered_tree
;
8544 spin_lock_irq(&tree
->lock
);
8545 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8546 new_len
= page_start
- ordered
->file_offset
;
8547 if (new_len
< ordered
->truncated_len
)
8548 ordered
->truncated_len
= new_len
;
8549 spin_unlock_irq(&tree
->lock
);
8551 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8553 PAGE_CACHE_SIZE
, 1))
8554 btrfs_finish_ordered_io(ordered
);
8556 btrfs_put_ordered_extent(ordered
);
8557 if (!inode_evicting
) {
8558 cached_state
= NULL
;
8559 lock_extent_bits(tree
, page_start
, page_end
, 0,
8564 if (!inode_evicting
) {
8565 clear_extent_bit(tree
, page_start
, page_end
,
8566 EXTENT_LOCKED
| EXTENT_DIRTY
|
8567 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
8568 EXTENT_DEFRAG
, 1, 1,
8569 &cached_state
, GFP_NOFS
);
8571 __btrfs_releasepage(page
, GFP_NOFS
);
8574 ClearPageChecked(page
);
8575 if (PagePrivate(page
)) {
8576 ClearPagePrivate(page
);
8577 set_page_private(page
, 0);
8578 page_cache_release(page
);
8583 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8584 * called from a page fault handler when a page is first dirtied. Hence we must
8585 * be careful to check for EOF conditions here. We set the page up correctly
8586 * for a written page which means we get ENOSPC checking when writing into
8587 * holes and correct delalloc and unwritten extent mapping on filesystems that
8588 * support these features.
8590 * We are not allowed to take the i_mutex here so we have to play games to
8591 * protect against truncate races as the page could now be beyond EOF. Because
8592 * vmtruncate() writes the inode size before removing pages, once we have the
8593 * page lock we can determine safely if the page is beyond EOF. If it is not
8594 * beyond EOF, then the page is guaranteed safe against truncation until we
8597 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
8599 struct page
*page
= vmf
->page
;
8600 struct inode
*inode
= file_inode(vma
->vm_file
);
8601 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8602 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8603 struct btrfs_ordered_extent
*ordered
;
8604 struct extent_state
*cached_state
= NULL
;
8606 unsigned long zero_start
;
8613 sb_start_pagefault(inode
->i_sb
);
8614 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
8616 ret
= file_update_time(vma
->vm_file
);
8622 else /* -ENOSPC, -EIO, etc */
8623 ret
= VM_FAULT_SIGBUS
;
8629 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8632 size
= i_size_read(inode
);
8633 page_start
= page_offset(page
);
8634 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8636 if ((page
->mapping
!= inode
->i_mapping
) ||
8637 (page_start
>= size
)) {
8638 /* page got truncated out from underneath us */
8641 wait_on_page_writeback(page
);
8643 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
8644 set_page_extent_mapped(page
);
8647 * we can't set the delalloc bits if there are pending ordered
8648 * extents. Drop our locks and wait for them to finish
8650 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8652 unlock_extent_cached(io_tree
, page_start
, page_end
,
8653 &cached_state
, GFP_NOFS
);
8655 btrfs_start_ordered_extent(inode
, ordered
, 1);
8656 btrfs_put_ordered_extent(ordered
);
8661 * XXX - page_mkwrite gets called every time the page is dirtied, even
8662 * if it was already dirty, so for space accounting reasons we need to
8663 * clear any delalloc bits for the range we are fixing to save. There
8664 * is probably a better way to do this, but for now keep consistent with
8665 * prepare_pages in the normal write path.
8667 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
8668 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8669 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
8670 0, 0, &cached_state
, GFP_NOFS
);
8672 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
8675 unlock_extent_cached(io_tree
, page_start
, page_end
,
8676 &cached_state
, GFP_NOFS
);
8677 ret
= VM_FAULT_SIGBUS
;
8682 /* page is wholly or partially inside EOF */
8683 if (page_start
+ PAGE_CACHE_SIZE
> size
)
8684 zero_start
= size
& ~PAGE_CACHE_MASK
;
8686 zero_start
= PAGE_CACHE_SIZE
;
8688 if (zero_start
!= PAGE_CACHE_SIZE
) {
8690 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
8691 flush_dcache_page(page
);
8694 ClearPageChecked(page
);
8695 set_page_dirty(page
);
8696 SetPageUptodate(page
);
8698 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
8699 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
8700 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
8702 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
8706 sb_end_pagefault(inode
->i_sb
);
8707 return VM_FAULT_LOCKED
;
8711 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
8713 sb_end_pagefault(inode
->i_sb
);
8717 static int btrfs_truncate(struct inode
*inode
)
8719 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8720 struct btrfs_block_rsv
*rsv
;
8723 struct btrfs_trans_handle
*trans
;
8724 u64 mask
= root
->sectorsize
- 1;
8725 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
8727 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
8733 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8734 * 3 things going on here
8736 * 1) We need to reserve space for our orphan item and the space to
8737 * delete our orphan item. Lord knows we don't want to have a dangling
8738 * orphan item because we didn't reserve space to remove it.
8740 * 2) We need to reserve space to update our inode.
8742 * 3) We need to have something to cache all the space that is going to
8743 * be free'd up by the truncate operation, but also have some slack
8744 * space reserved in case it uses space during the truncate (thank you
8745 * very much snapshotting).
8747 * And we need these to all be seperate. The fact is we can use alot of
8748 * space doing the truncate, and we have no earthly idea how much space
8749 * we will use, so we need the truncate reservation to be seperate so it
8750 * doesn't end up using space reserved for updating the inode or
8751 * removing the orphan item. We also need to be able to stop the
8752 * transaction and start a new one, which means we need to be able to
8753 * update the inode several times, and we have no idea of knowing how
8754 * many times that will be, so we can't just reserve 1 item for the
8755 * entirety of the opration, so that has to be done seperately as well.
8756 * Then there is the orphan item, which does indeed need to be held on
8757 * to for the whole operation, and we need nobody to touch this reserved
8758 * space except the orphan code.
8760 * So that leaves us with
8762 * 1) root->orphan_block_rsv - for the orphan deletion.
8763 * 2) rsv - for the truncate reservation, which we will steal from the
8764 * transaction reservation.
8765 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8766 * updating the inode.
8768 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
8771 rsv
->size
= min_size
;
8775 * 1 for the truncate slack space
8776 * 1 for updating the inode.
8778 trans
= btrfs_start_transaction(root
, 2);
8779 if (IS_ERR(trans
)) {
8780 err
= PTR_ERR(trans
);
8784 /* Migrate the slack space for the truncate to our reserve */
8785 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
8790 * So if we truncate and then write and fsync we normally would just
8791 * write the extents that changed, which is a problem if we need to
8792 * first truncate that entire inode. So set this flag so we write out
8793 * all of the extents in the inode to the sync log so we're completely
8796 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
8797 trans
->block_rsv
= rsv
;
8800 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
8802 BTRFS_EXTENT_DATA_KEY
);
8803 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
8808 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
8809 ret
= btrfs_update_inode(trans
, root
, inode
);
8815 btrfs_end_transaction(trans
, root
);
8816 btrfs_btree_balance_dirty(root
);
8818 trans
= btrfs_start_transaction(root
, 2);
8819 if (IS_ERR(trans
)) {
8820 ret
= err
= PTR_ERR(trans
);
8825 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
8827 BUG_ON(ret
); /* shouldn't happen */
8828 trans
->block_rsv
= rsv
;
8831 if (ret
== 0 && inode
->i_nlink
> 0) {
8832 trans
->block_rsv
= root
->orphan_block_rsv
;
8833 ret
= btrfs_orphan_del(trans
, inode
);
8839 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
8840 ret
= btrfs_update_inode(trans
, root
, inode
);
8844 ret
= btrfs_end_transaction(trans
, root
);
8845 btrfs_btree_balance_dirty(root
);
8849 btrfs_free_block_rsv(root
, rsv
);
8858 * create a new subvolume directory/inode (helper for the ioctl).
8860 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
8861 struct btrfs_root
*new_root
,
8862 struct btrfs_root
*parent_root
,
8865 struct inode
*inode
;
8869 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
8870 new_dirid
, new_dirid
,
8871 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
8874 return PTR_ERR(inode
);
8875 inode
->i_op
= &btrfs_dir_inode_operations
;
8876 inode
->i_fop
= &btrfs_dir_file_operations
;
8878 set_nlink(inode
, 1);
8879 btrfs_i_size_write(inode
, 0);
8880 unlock_new_inode(inode
);
8882 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
8884 btrfs_err(new_root
->fs_info
,
8885 "error inheriting subvolume %llu properties: %d",
8886 new_root
->root_key
.objectid
, err
);
8888 err
= btrfs_update_inode(trans
, new_root
, inode
);
8894 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
8896 struct btrfs_inode
*ei
;
8897 struct inode
*inode
;
8899 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
8906 ei
->last_sub_trans
= 0;
8907 ei
->logged_trans
= 0;
8908 ei
->delalloc_bytes
= 0;
8909 ei
->defrag_bytes
= 0;
8910 ei
->disk_i_size
= 0;
8913 ei
->index_cnt
= (u64
)-1;
8915 ei
->last_unlink_trans
= 0;
8916 ei
->last_log_commit
= 0;
8918 spin_lock_init(&ei
->lock
);
8919 ei
->outstanding_extents
= 0;
8920 ei
->reserved_extents
= 0;
8922 ei
->runtime_flags
= 0;
8923 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
8925 ei
->delayed_node
= NULL
;
8927 ei
->i_otime
.tv_sec
= 0;
8928 ei
->i_otime
.tv_nsec
= 0;
8930 inode
= &ei
->vfs_inode
;
8931 extent_map_tree_init(&ei
->extent_tree
);
8932 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
8933 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
8934 ei
->io_tree
.track_uptodate
= 1;
8935 ei
->io_failure_tree
.track_uptodate
= 1;
8936 atomic_set(&ei
->sync_writers
, 0);
8937 mutex_init(&ei
->log_mutex
);
8938 mutex_init(&ei
->delalloc_mutex
);
8939 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
8940 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
8941 RB_CLEAR_NODE(&ei
->rb_node
);
8946 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8947 void btrfs_test_destroy_inode(struct inode
*inode
)
8949 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
8950 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
8954 static void btrfs_i_callback(struct rcu_head
*head
)
8956 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
8957 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
8960 void btrfs_destroy_inode(struct inode
*inode
)
8962 struct btrfs_ordered_extent
*ordered
;
8963 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8965 WARN_ON(!hlist_empty(&inode
->i_dentry
));
8966 WARN_ON(inode
->i_data
.nrpages
);
8967 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
8968 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
8969 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
8970 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
8971 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
8974 * This can happen where we create an inode, but somebody else also
8975 * created the same inode and we need to destroy the one we already
8981 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
8982 &BTRFS_I(inode
)->runtime_flags
)) {
8983 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
8985 atomic_dec(&root
->orphan_inodes
);
8989 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
8993 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
8994 ordered
->file_offset
, ordered
->len
);
8995 btrfs_remove_ordered_extent(inode
, ordered
);
8996 btrfs_put_ordered_extent(ordered
);
8997 btrfs_put_ordered_extent(ordered
);
9000 inode_tree_del(inode
);
9001 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9003 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9006 int btrfs_drop_inode(struct inode
*inode
)
9008 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9013 /* the snap/subvol tree is on deleting */
9014 if (btrfs_root_refs(&root
->root_item
) == 0)
9017 return generic_drop_inode(inode
);
9020 static void init_once(void *foo
)
9022 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9024 inode_init_once(&ei
->vfs_inode
);
9027 void btrfs_destroy_cachep(void)
9030 * Make sure all delayed rcu free inodes are flushed before we
9034 if (btrfs_inode_cachep
)
9035 kmem_cache_destroy(btrfs_inode_cachep
);
9036 if (btrfs_trans_handle_cachep
)
9037 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9038 if (btrfs_transaction_cachep
)
9039 kmem_cache_destroy(btrfs_transaction_cachep
);
9040 if (btrfs_path_cachep
)
9041 kmem_cache_destroy(btrfs_path_cachep
);
9042 if (btrfs_free_space_cachep
)
9043 kmem_cache_destroy(btrfs_free_space_cachep
);
9044 if (btrfs_delalloc_work_cachep
)
9045 kmem_cache_destroy(btrfs_delalloc_work_cachep
);
9048 int btrfs_init_cachep(void)
9050 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9051 sizeof(struct btrfs_inode
), 0,
9052 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
9053 if (!btrfs_inode_cachep
)
9056 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9057 sizeof(struct btrfs_trans_handle
), 0,
9058 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9059 if (!btrfs_trans_handle_cachep
)
9062 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
9063 sizeof(struct btrfs_transaction
), 0,
9064 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9065 if (!btrfs_transaction_cachep
)
9068 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9069 sizeof(struct btrfs_path
), 0,
9070 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9071 if (!btrfs_path_cachep
)
9074 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9075 sizeof(struct btrfs_free_space
), 0,
9076 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9077 if (!btrfs_free_space_cachep
)
9080 btrfs_delalloc_work_cachep
= kmem_cache_create("btrfs_delalloc_work",
9081 sizeof(struct btrfs_delalloc_work
), 0,
9082 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
9084 if (!btrfs_delalloc_work_cachep
)
9089 btrfs_destroy_cachep();
9093 static int btrfs_getattr(struct vfsmount
*mnt
,
9094 struct dentry
*dentry
, struct kstat
*stat
)
9097 struct inode
*inode
= d_inode(dentry
);
9098 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9100 generic_fillattr(inode
, stat
);
9101 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9102 stat
->blksize
= PAGE_CACHE_SIZE
;
9104 spin_lock(&BTRFS_I(inode
)->lock
);
9105 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
9106 spin_unlock(&BTRFS_I(inode
)->lock
);
9107 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9108 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9112 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9113 struct inode
*new_dir
, struct dentry
*new_dentry
)
9115 struct btrfs_trans_handle
*trans
;
9116 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9117 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9118 struct inode
*new_inode
= d_inode(new_dentry
);
9119 struct inode
*old_inode
= d_inode(old_dentry
);
9120 struct timespec ctime
= CURRENT_TIME
;
9124 u64 old_ino
= btrfs_ino(old_inode
);
9126 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9129 /* we only allow rename subvolume link between subvolumes */
9130 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9133 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9134 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
9137 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9138 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9142 /* check for collisions, even if the name isn't there */
9143 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9144 new_dentry
->d_name
.name
,
9145 new_dentry
->d_name
.len
);
9148 if (ret
== -EEXIST
) {
9150 * eexist without a new_inode */
9151 if (WARN_ON(!new_inode
)) {
9155 /* maybe -EOVERFLOW */
9162 * we're using rename to replace one file with another. Start IO on it
9163 * now so we don't add too much work to the end of the transaction
9165 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9166 filemap_flush(old_inode
->i_mapping
);
9168 /* close the racy window with snapshot create/destroy ioctl */
9169 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9170 down_read(&root
->fs_info
->subvol_sem
);
9172 * We want to reserve the absolute worst case amount of items. So if
9173 * both inodes are subvols and we need to unlink them then that would
9174 * require 4 item modifications, but if they are both normal inodes it
9175 * would require 5 item modifications, so we'll assume their normal
9176 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9177 * should cover the worst case number of items we'll modify.
9179 trans
= btrfs_start_transaction(root
, 11);
9180 if (IS_ERR(trans
)) {
9181 ret
= PTR_ERR(trans
);
9186 btrfs_record_root_in_trans(trans
, dest
);
9188 ret
= btrfs_set_inode_index(new_dir
, &index
);
9192 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9193 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9194 /* force full log commit if subvolume involved. */
9195 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9197 ret
= btrfs_insert_inode_ref(trans
, dest
,
9198 new_dentry
->d_name
.name
,
9199 new_dentry
->d_name
.len
,
9201 btrfs_ino(new_dir
), index
);
9205 * this is an ugly little race, but the rename is required
9206 * to make sure that if we crash, the inode is either at the
9207 * old name or the new one. pinning the log transaction lets
9208 * us make sure we don't allow a log commit to come in after
9209 * we unlink the name but before we add the new name back in.
9211 btrfs_pin_log_trans(root
);
9214 inode_inc_iversion(old_dir
);
9215 inode_inc_iversion(new_dir
);
9216 inode_inc_iversion(old_inode
);
9217 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9218 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9219 old_inode
->i_ctime
= ctime
;
9221 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9222 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
9224 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9225 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9226 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
9227 old_dentry
->d_name
.name
,
9228 old_dentry
->d_name
.len
);
9230 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
9231 d_inode(old_dentry
),
9232 old_dentry
->d_name
.name
,
9233 old_dentry
->d_name
.len
);
9235 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9238 btrfs_abort_transaction(trans
, root
, ret
);
9243 inode_inc_iversion(new_inode
);
9244 new_inode
->i_ctime
= CURRENT_TIME
;
9245 if (unlikely(btrfs_ino(new_inode
) ==
9246 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9247 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9248 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9250 new_dentry
->d_name
.name
,
9251 new_dentry
->d_name
.len
);
9252 BUG_ON(new_inode
->i_nlink
== 0);
9254 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
9255 d_inode(new_dentry
),
9256 new_dentry
->d_name
.name
,
9257 new_dentry
->d_name
.len
);
9259 if (!ret
&& new_inode
->i_nlink
== 0)
9260 ret
= btrfs_orphan_add(trans
, d_inode(new_dentry
));
9262 btrfs_abort_transaction(trans
, root
, ret
);
9267 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
9268 new_dentry
->d_name
.name
,
9269 new_dentry
->d_name
.len
, 0, index
);
9271 btrfs_abort_transaction(trans
, root
, ret
);
9275 if (old_inode
->i_nlink
== 1)
9276 BTRFS_I(old_inode
)->dir_index
= index
;
9278 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
9279 struct dentry
*parent
= new_dentry
->d_parent
;
9280 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
9281 btrfs_end_log_trans(root
);
9284 btrfs_end_transaction(trans
, root
);
9286 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9287 up_read(&root
->fs_info
->subvol_sem
);
9292 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
9293 struct inode
*new_dir
, struct dentry
*new_dentry
,
9296 if (flags
& ~RENAME_NOREPLACE
)
9299 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
9302 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9304 struct btrfs_delalloc_work
*delalloc_work
;
9305 struct inode
*inode
;
9307 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9309 inode
= delalloc_work
->inode
;
9310 if (delalloc_work
->wait
) {
9311 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
9313 filemap_flush(inode
->i_mapping
);
9314 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9315 &BTRFS_I(inode
)->runtime_flags
))
9316 filemap_flush(inode
->i_mapping
);
9319 if (delalloc_work
->delay_iput
)
9320 btrfs_add_delayed_iput(inode
);
9323 complete(&delalloc_work
->completion
);
9326 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
9327 int wait
, int delay_iput
)
9329 struct btrfs_delalloc_work
*work
;
9331 work
= kmem_cache_zalloc(btrfs_delalloc_work_cachep
, GFP_NOFS
);
9335 init_completion(&work
->completion
);
9336 INIT_LIST_HEAD(&work
->list
);
9337 work
->inode
= inode
;
9339 work
->delay_iput
= delay_iput
;
9340 WARN_ON_ONCE(!inode
);
9341 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
9342 btrfs_run_delalloc_work
, NULL
, NULL
);
9347 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
9349 wait_for_completion(&work
->completion
);
9350 kmem_cache_free(btrfs_delalloc_work_cachep
, work
);
9354 * some fairly slow code that needs optimization. This walks the list
9355 * of all the inodes with pending delalloc and forces them to disk.
9357 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
9360 struct btrfs_inode
*binode
;
9361 struct inode
*inode
;
9362 struct btrfs_delalloc_work
*work
, *next
;
9363 struct list_head works
;
9364 struct list_head splice
;
9367 INIT_LIST_HEAD(&works
);
9368 INIT_LIST_HEAD(&splice
);
9370 mutex_lock(&root
->delalloc_mutex
);
9371 spin_lock(&root
->delalloc_lock
);
9372 list_splice_init(&root
->delalloc_inodes
, &splice
);
9373 while (!list_empty(&splice
)) {
9374 binode
= list_entry(splice
.next
, struct btrfs_inode
,
9377 list_move_tail(&binode
->delalloc_inodes
,
9378 &root
->delalloc_inodes
);
9379 inode
= igrab(&binode
->vfs_inode
);
9381 cond_resched_lock(&root
->delalloc_lock
);
9384 spin_unlock(&root
->delalloc_lock
);
9386 work
= btrfs_alloc_delalloc_work(inode
, 0, delay_iput
);
9389 btrfs_add_delayed_iput(inode
);
9395 list_add_tail(&work
->list
, &works
);
9396 btrfs_queue_work(root
->fs_info
->flush_workers
,
9399 if (nr
!= -1 && ret
>= nr
)
9402 spin_lock(&root
->delalloc_lock
);
9404 spin_unlock(&root
->delalloc_lock
);
9407 list_for_each_entry_safe(work
, next
, &works
, list
) {
9408 list_del_init(&work
->list
);
9409 btrfs_wait_and_free_delalloc_work(work
);
9412 if (!list_empty_careful(&splice
)) {
9413 spin_lock(&root
->delalloc_lock
);
9414 list_splice_tail(&splice
, &root
->delalloc_inodes
);
9415 spin_unlock(&root
->delalloc_lock
);
9417 mutex_unlock(&root
->delalloc_mutex
);
9421 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
9425 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
9428 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
9432 * the filemap_flush will queue IO into the worker threads, but
9433 * we have to make sure the IO is actually started and that
9434 * ordered extents get created before we return
9436 atomic_inc(&root
->fs_info
->async_submit_draining
);
9437 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
9438 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
9439 wait_event(root
->fs_info
->async_submit_wait
,
9440 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
9441 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
9443 atomic_dec(&root
->fs_info
->async_submit_draining
);
9447 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
9450 struct btrfs_root
*root
;
9451 struct list_head splice
;
9454 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
9457 INIT_LIST_HEAD(&splice
);
9459 mutex_lock(&fs_info
->delalloc_root_mutex
);
9460 spin_lock(&fs_info
->delalloc_root_lock
);
9461 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
9462 while (!list_empty(&splice
) && nr
) {
9463 root
= list_first_entry(&splice
, struct btrfs_root
,
9465 root
= btrfs_grab_fs_root(root
);
9467 list_move_tail(&root
->delalloc_root
,
9468 &fs_info
->delalloc_roots
);
9469 spin_unlock(&fs_info
->delalloc_root_lock
);
9471 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
9472 btrfs_put_fs_root(root
);
9480 spin_lock(&fs_info
->delalloc_root_lock
);
9482 spin_unlock(&fs_info
->delalloc_root_lock
);
9485 atomic_inc(&fs_info
->async_submit_draining
);
9486 while (atomic_read(&fs_info
->nr_async_submits
) ||
9487 atomic_read(&fs_info
->async_delalloc_pages
)) {
9488 wait_event(fs_info
->async_submit_wait
,
9489 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
9490 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
9492 atomic_dec(&fs_info
->async_submit_draining
);
9494 if (!list_empty_careful(&splice
)) {
9495 spin_lock(&fs_info
->delalloc_root_lock
);
9496 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
9497 spin_unlock(&fs_info
->delalloc_root_lock
);
9499 mutex_unlock(&fs_info
->delalloc_root_mutex
);
9503 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
9504 const char *symname
)
9506 struct btrfs_trans_handle
*trans
;
9507 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9508 struct btrfs_path
*path
;
9509 struct btrfs_key key
;
9510 struct inode
*inode
= NULL
;
9518 struct btrfs_file_extent_item
*ei
;
9519 struct extent_buffer
*leaf
;
9521 name_len
= strlen(symname
);
9522 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
9523 return -ENAMETOOLONG
;
9526 * 2 items for inode item and ref
9527 * 2 items for dir items
9528 * 1 item for xattr if selinux is on
9530 trans
= btrfs_start_transaction(root
, 5);
9532 return PTR_ERR(trans
);
9534 err
= btrfs_find_free_ino(root
, &objectid
);
9538 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
9539 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
9540 S_IFLNK
|S_IRWXUGO
, &index
);
9541 if (IS_ERR(inode
)) {
9542 err
= PTR_ERR(inode
);
9547 * If the active LSM wants to access the inode during
9548 * d_instantiate it needs these. Smack checks to see
9549 * if the filesystem supports xattrs by looking at the
9552 inode
->i_fop
= &btrfs_file_operations
;
9553 inode
->i_op
= &btrfs_file_inode_operations
;
9554 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9555 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9557 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
9559 goto out_unlock_inode
;
9561 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
9563 goto out_unlock_inode
;
9565 path
= btrfs_alloc_path();
9568 goto out_unlock_inode
;
9570 key
.objectid
= btrfs_ino(inode
);
9572 key
.type
= BTRFS_EXTENT_DATA_KEY
;
9573 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
9574 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
9577 btrfs_free_path(path
);
9578 goto out_unlock_inode
;
9580 leaf
= path
->nodes
[0];
9581 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
9582 struct btrfs_file_extent_item
);
9583 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
9584 btrfs_set_file_extent_type(leaf
, ei
,
9585 BTRFS_FILE_EXTENT_INLINE
);
9586 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
9587 btrfs_set_file_extent_compression(leaf
, ei
, 0);
9588 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
9589 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
9591 ptr
= btrfs_file_extent_inline_start(ei
);
9592 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
9593 btrfs_mark_buffer_dirty(leaf
);
9594 btrfs_free_path(path
);
9596 inode
->i_op
= &btrfs_symlink_inode_operations
;
9597 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
9598 inode_set_bytes(inode
, name_len
);
9599 btrfs_i_size_write(inode
, name_len
);
9600 err
= btrfs_update_inode(trans
, root
, inode
);
9603 goto out_unlock_inode
;
9606 unlock_new_inode(inode
);
9607 d_instantiate(dentry
, inode
);
9610 btrfs_end_transaction(trans
, root
);
9612 inode_dec_link_count(inode
);
9615 btrfs_btree_balance_dirty(root
);
9620 unlock_new_inode(inode
);
9624 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9625 u64 start
, u64 num_bytes
, u64 min_size
,
9626 loff_t actual_len
, u64
*alloc_hint
,
9627 struct btrfs_trans_handle
*trans
)
9629 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
9630 struct extent_map
*em
;
9631 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9632 struct btrfs_key ins
;
9633 u64 cur_offset
= start
;
9637 bool own_trans
= true;
9641 while (num_bytes
> 0) {
9643 trans
= btrfs_start_transaction(root
, 3);
9644 if (IS_ERR(trans
)) {
9645 ret
= PTR_ERR(trans
);
9650 cur_bytes
= min(num_bytes
, 256ULL * 1024 * 1024);
9651 cur_bytes
= max(cur_bytes
, min_size
);
9652 ret
= btrfs_reserve_extent(root
, cur_bytes
, min_size
, 0,
9653 *alloc_hint
, &ins
, 1, 0);
9656 btrfs_end_transaction(trans
, root
);
9660 ret
= insert_reserved_file_extent(trans
, inode
,
9661 cur_offset
, ins
.objectid
,
9662 ins
.offset
, ins
.offset
,
9663 ins
.offset
, 0, 0, 0,
9664 BTRFS_FILE_EXTENT_PREALLOC
);
9666 btrfs_free_reserved_extent(root
, ins
.objectid
,
9668 btrfs_abort_transaction(trans
, root
, ret
);
9670 btrfs_end_transaction(trans
, root
);
9674 btrfs_drop_extent_cache(inode
, cur_offset
,
9675 cur_offset
+ ins
.offset
-1, 0);
9677 em
= alloc_extent_map();
9679 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
9680 &BTRFS_I(inode
)->runtime_flags
);
9684 em
->start
= cur_offset
;
9685 em
->orig_start
= cur_offset
;
9686 em
->len
= ins
.offset
;
9687 em
->block_start
= ins
.objectid
;
9688 em
->block_len
= ins
.offset
;
9689 em
->orig_block_len
= ins
.offset
;
9690 em
->ram_bytes
= ins
.offset
;
9691 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
9692 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
9693 em
->generation
= trans
->transid
;
9696 write_lock(&em_tree
->lock
);
9697 ret
= add_extent_mapping(em_tree
, em
, 1);
9698 write_unlock(&em_tree
->lock
);
9701 btrfs_drop_extent_cache(inode
, cur_offset
,
9702 cur_offset
+ ins
.offset
- 1,
9705 free_extent_map(em
);
9707 num_bytes
-= ins
.offset
;
9708 cur_offset
+= ins
.offset
;
9709 *alloc_hint
= ins
.objectid
+ ins
.offset
;
9711 inode_inc_iversion(inode
);
9712 inode
->i_ctime
= CURRENT_TIME
;
9713 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
9714 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
9715 (actual_len
> inode
->i_size
) &&
9716 (cur_offset
> inode
->i_size
)) {
9717 if (cur_offset
> actual_len
)
9718 i_size
= actual_len
;
9720 i_size
= cur_offset
;
9721 i_size_write(inode
, i_size
);
9722 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
9725 ret
= btrfs_update_inode(trans
, root
, inode
);
9728 btrfs_abort_transaction(trans
, root
, ret
);
9730 btrfs_end_transaction(trans
, root
);
9735 btrfs_end_transaction(trans
, root
);
9740 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9741 u64 start
, u64 num_bytes
, u64 min_size
,
9742 loff_t actual_len
, u64
*alloc_hint
)
9744 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9745 min_size
, actual_len
, alloc_hint
,
9749 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
9750 struct btrfs_trans_handle
*trans
, int mode
,
9751 u64 start
, u64 num_bytes
, u64 min_size
,
9752 loff_t actual_len
, u64
*alloc_hint
)
9754 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9755 min_size
, actual_len
, alloc_hint
, trans
);
9758 static int btrfs_set_page_dirty(struct page
*page
)
9760 return __set_page_dirty_nobuffers(page
);
9763 static int btrfs_permission(struct inode
*inode
, int mask
)
9765 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9766 umode_t mode
= inode
->i_mode
;
9768 if (mask
& MAY_WRITE
&&
9769 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
9770 if (btrfs_root_readonly(root
))
9772 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
9775 return generic_permission(inode
, mask
);
9778 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
9780 struct btrfs_trans_handle
*trans
;
9781 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9782 struct inode
*inode
= NULL
;
9788 * 5 units required for adding orphan entry
9790 trans
= btrfs_start_transaction(root
, 5);
9792 return PTR_ERR(trans
);
9794 ret
= btrfs_find_free_ino(root
, &objectid
);
9798 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
9799 btrfs_ino(dir
), objectid
, mode
, &index
);
9800 if (IS_ERR(inode
)) {
9801 ret
= PTR_ERR(inode
);
9806 inode
->i_fop
= &btrfs_file_operations
;
9807 inode
->i_op
= &btrfs_file_inode_operations
;
9809 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9810 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9812 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
9816 ret
= btrfs_update_inode(trans
, root
, inode
);
9819 ret
= btrfs_orphan_add(trans
, inode
);
9824 * We set number of links to 0 in btrfs_new_inode(), and here we set
9825 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9828 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9830 set_nlink(inode
, 1);
9831 unlock_new_inode(inode
);
9832 d_tmpfile(dentry
, inode
);
9833 mark_inode_dirty(inode
);
9836 btrfs_end_transaction(trans
, root
);
9839 btrfs_balance_delayed_items(root
);
9840 btrfs_btree_balance_dirty(root
);
9844 unlock_new_inode(inode
);
9849 /* Inspired by filemap_check_errors() */
9850 int btrfs_inode_check_errors(struct inode
*inode
)
9854 if (test_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
) &&
9855 test_and_clear_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
))
9857 if (test_bit(AS_EIO
, &inode
->i_mapping
->flags
) &&
9858 test_and_clear_bit(AS_EIO
, &inode
->i_mapping
->flags
))
9864 static const struct inode_operations btrfs_dir_inode_operations
= {
9865 .getattr
= btrfs_getattr
,
9866 .lookup
= btrfs_lookup
,
9867 .create
= btrfs_create
,
9868 .unlink
= btrfs_unlink
,
9870 .mkdir
= btrfs_mkdir
,
9871 .rmdir
= btrfs_rmdir
,
9872 .rename2
= btrfs_rename2
,
9873 .symlink
= btrfs_symlink
,
9874 .setattr
= btrfs_setattr
,
9875 .mknod
= btrfs_mknod
,
9876 .setxattr
= btrfs_setxattr
,
9877 .getxattr
= btrfs_getxattr
,
9878 .listxattr
= btrfs_listxattr
,
9879 .removexattr
= btrfs_removexattr
,
9880 .permission
= btrfs_permission
,
9881 .get_acl
= btrfs_get_acl
,
9882 .set_acl
= btrfs_set_acl
,
9883 .update_time
= btrfs_update_time
,
9884 .tmpfile
= btrfs_tmpfile
,
9886 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
9887 .lookup
= btrfs_lookup
,
9888 .permission
= btrfs_permission
,
9889 .get_acl
= btrfs_get_acl
,
9890 .set_acl
= btrfs_set_acl
,
9891 .update_time
= btrfs_update_time
,
9894 static const struct file_operations btrfs_dir_file_operations
= {
9895 .llseek
= generic_file_llseek
,
9896 .read
= generic_read_dir
,
9897 .iterate
= btrfs_real_readdir
,
9898 .unlocked_ioctl
= btrfs_ioctl
,
9899 #ifdef CONFIG_COMPAT
9900 .compat_ioctl
= btrfs_ioctl
,
9902 .release
= btrfs_release_file
,
9903 .fsync
= btrfs_sync_file
,
9906 static struct extent_io_ops btrfs_extent_io_ops
= {
9907 .fill_delalloc
= run_delalloc_range
,
9908 .submit_bio_hook
= btrfs_submit_bio_hook
,
9909 .merge_bio_hook
= btrfs_merge_bio_hook
,
9910 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
9911 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
9912 .writepage_start_hook
= btrfs_writepage_start_hook
,
9913 .set_bit_hook
= btrfs_set_bit_hook
,
9914 .clear_bit_hook
= btrfs_clear_bit_hook
,
9915 .merge_extent_hook
= btrfs_merge_extent_hook
,
9916 .split_extent_hook
= btrfs_split_extent_hook
,
9920 * btrfs doesn't support the bmap operation because swapfiles
9921 * use bmap to make a mapping of extents in the file. They assume
9922 * these extents won't change over the life of the file and they
9923 * use the bmap result to do IO directly to the drive.
9925 * the btrfs bmap call would return logical addresses that aren't
9926 * suitable for IO and they also will change frequently as COW
9927 * operations happen. So, swapfile + btrfs == corruption.
9929 * For now we're avoiding this by dropping bmap.
9931 static const struct address_space_operations btrfs_aops
= {
9932 .readpage
= btrfs_readpage
,
9933 .writepage
= btrfs_writepage
,
9934 .writepages
= btrfs_writepages
,
9935 .readpages
= btrfs_readpages
,
9936 .direct_IO
= btrfs_direct_IO
,
9937 .invalidatepage
= btrfs_invalidatepage
,
9938 .releasepage
= btrfs_releasepage
,
9939 .set_page_dirty
= btrfs_set_page_dirty
,
9940 .error_remove_page
= generic_error_remove_page
,
9943 static const struct address_space_operations btrfs_symlink_aops
= {
9944 .readpage
= btrfs_readpage
,
9945 .writepage
= btrfs_writepage
,
9946 .invalidatepage
= btrfs_invalidatepage
,
9947 .releasepage
= btrfs_releasepage
,
9950 static const struct inode_operations btrfs_file_inode_operations
= {
9951 .getattr
= btrfs_getattr
,
9952 .setattr
= btrfs_setattr
,
9953 .setxattr
= btrfs_setxattr
,
9954 .getxattr
= btrfs_getxattr
,
9955 .listxattr
= btrfs_listxattr
,
9956 .removexattr
= btrfs_removexattr
,
9957 .permission
= btrfs_permission
,
9958 .fiemap
= btrfs_fiemap
,
9959 .get_acl
= btrfs_get_acl
,
9960 .set_acl
= btrfs_set_acl
,
9961 .update_time
= btrfs_update_time
,
9963 static const struct inode_operations btrfs_special_inode_operations
= {
9964 .getattr
= btrfs_getattr
,
9965 .setattr
= btrfs_setattr
,
9966 .permission
= btrfs_permission
,
9967 .setxattr
= btrfs_setxattr
,
9968 .getxattr
= btrfs_getxattr
,
9969 .listxattr
= btrfs_listxattr
,
9970 .removexattr
= btrfs_removexattr
,
9971 .get_acl
= btrfs_get_acl
,
9972 .set_acl
= btrfs_set_acl
,
9973 .update_time
= btrfs_update_time
,
9975 static const struct inode_operations btrfs_symlink_inode_operations
= {
9976 .readlink
= generic_readlink
,
9977 .follow_link
= page_follow_link_light
,
9978 .put_link
= page_put_link
,
9979 .getattr
= btrfs_getattr
,
9980 .setattr
= btrfs_setattr
,
9981 .permission
= btrfs_permission
,
9982 .setxattr
= btrfs_setxattr
,
9983 .getxattr
= btrfs_getxattr
,
9984 .listxattr
= btrfs_listxattr
,
9985 .removexattr
= btrfs_removexattr
,
9986 .update_time
= btrfs_update_time
,
9989 const struct dentry_operations btrfs_dentry_operations
= {
9990 .d_delete
= btrfs_dentry_delete
,
9991 .d_release
= btrfs_dentry_release
,