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
) >>
1099 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1101 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1102 wake_up(&root
->fs_info
->async_submit_wait
);
1104 if (async_cow
->inode
)
1105 submit_compressed_extents(async_cow
->inode
, async_cow
);
1108 static noinline
void async_cow_free(struct btrfs_work
*work
)
1110 struct async_cow
*async_cow
;
1111 async_cow
= container_of(work
, struct async_cow
, work
);
1112 if (async_cow
->inode
)
1113 btrfs_add_delayed_iput(async_cow
->inode
);
1117 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1118 u64 start
, u64 end
, int *page_started
,
1119 unsigned long *nr_written
)
1121 struct async_cow
*async_cow
;
1122 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1123 unsigned long nr_pages
;
1125 int limit
= 10 * 1024 * 1024;
1127 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1128 1, 0, NULL
, GFP_NOFS
);
1129 while (start
< end
) {
1130 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1131 BUG_ON(!async_cow
); /* -ENOMEM */
1132 async_cow
->inode
= igrab(inode
);
1133 async_cow
->root
= root
;
1134 async_cow
->locked_page
= locked_page
;
1135 async_cow
->start
= start
;
1137 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
&&
1138 !btrfs_test_opt(root
, FORCE_COMPRESS
))
1141 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1143 async_cow
->end
= cur_end
;
1144 INIT_LIST_HEAD(&async_cow
->extents
);
1146 btrfs_init_work(&async_cow
->work
,
1147 btrfs_delalloc_helper
,
1148 async_cow_start
, async_cow_submit
,
1151 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1153 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1155 btrfs_queue_work(root
->fs_info
->delalloc_workers
,
1158 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1159 wait_event(root
->fs_info
->async_submit_wait
,
1160 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1164 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1165 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1166 wait_event(root
->fs_info
->async_submit_wait
,
1167 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1171 *nr_written
+= nr_pages
;
1172 start
= cur_end
+ 1;
1178 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1179 u64 bytenr
, u64 num_bytes
)
1182 struct btrfs_ordered_sum
*sums
;
1185 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1186 bytenr
+ num_bytes
- 1, &list
, 0);
1187 if (ret
== 0 && list_empty(&list
))
1190 while (!list_empty(&list
)) {
1191 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1192 list_del(&sums
->list
);
1199 * when nowcow writeback call back. This checks for snapshots or COW copies
1200 * of the extents that exist in the file, and COWs the file as required.
1202 * If no cow copies or snapshots exist, we write directly to the existing
1205 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1206 struct page
*locked_page
,
1207 u64 start
, u64 end
, int *page_started
, int force
,
1208 unsigned long *nr_written
)
1210 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1211 struct btrfs_trans_handle
*trans
;
1212 struct extent_buffer
*leaf
;
1213 struct btrfs_path
*path
;
1214 struct btrfs_file_extent_item
*fi
;
1215 struct btrfs_key found_key
;
1230 u64 ino
= btrfs_ino(inode
);
1232 path
= btrfs_alloc_path();
1234 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1235 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1236 EXTENT_DO_ACCOUNTING
|
1237 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1239 PAGE_SET_WRITEBACK
|
1240 PAGE_END_WRITEBACK
);
1244 nolock
= btrfs_is_free_space_inode(inode
);
1247 trans
= btrfs_join_transaction_nolock(root
);
1249 trans
= btrfs_join_transaction(root
);
1251 if (IS_ERR(trans
)) {
1252 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1253 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1254 EXTENT_DO_ACCOUNTING
|
1255 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1257 PAGE_SET_WRITEBACK
|
1258 PAGE_END_WRITEBACK
);
1259 btrfs_free_path(path
);
1260 return PTR_ERR(trans
);
1263 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1265 cow_start
= (u64
)-1;
1268 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1272 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1273 leaf
= path
->nodes
[0];
1274 btrfs_item_key_to_cpu(leaf
, &found_key
,
1275 path
->slots
[0] - 1);
1276 if (found_key
.objectid
== ino
&&
1277 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1282 leaf
= path
->nodes
[0];
1283 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1284 ret
= btrfs_next_leaf(root
, path
);
1289 leaf
= path
->nodes
[0];
1295 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1297 if (found_key
.objectid
> ino
||
1298 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1299 found_key
.offset
> end
)
1302 if (found_key
.offset
> cur_offset
) {
1303 extent_end
= found_key
.offset
;
1308 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1309 struct btrfs_file_extent_item
);
1310 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1312 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1313 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1314 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1315 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1316 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1317 extent_end
= found_key
.offset
+
1318 btrfs_file_extent_num_bytes(leaf
, fi
);
1320 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1321 if (extent_end
<= start
) {
1325 if (disk_bytenr
== 0)
1327 if (btrfs_file_extent_compression(leaf
, fi
) ||
1328 btrfs_file_extent_encryption(leaf
, fi
) ||
1329 btrfs_file_extent_other_encoding(leaf
, fi
))
1331 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1333 if (btrfs_extent_readonly(root
, disk_bytenr
))
1335 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1337 extent_offset
, disk_bytenr
))
1339 disk_bytenr
+= extent_offset
;
1340 disk_bytenr
+= cur_offset
- found_key
.offset
;
1341 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1343 * if there are pending snapshots for this root,
1344 * we fall into common COW way.
1347 err
= btrfs_start_write_no_snapshoting(root
);
1352 * force cow if csum exists in the range.
1353 * this ensure that csum for a given extent are
1354 * either valid or do not exist.
1356 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1359 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1360 extent_end
= found_key
.offset
+
1361 btrfs_file_extent_inline_len(leaf
,
1362 path
->slots
[0], fi
);
1363 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1368 if (extent_end
<= start
) {
1370 if (!nolock
&& nocow
)
1371 btrfs_end_write_no_snapshoting(root
);
1375 if (cow_start
== (u64
)-1)
1376 cow_start
= cur_offset
;
1377 cur_offset
= extent_end
;
1378 if (cur_offset
> end
)
1384 btrfs_release_path(path
);
1385 if (cow_start
!= (u64
)-1) {
1386 ret
= cow_file_range(inode
, locked_page
,
1387 cow_start
, found_key
.offset
- 1,
1388 page_started
, nr_written
, 1);
1390 if (!nolock
&& nocow
)
1391 btrfs_end_write_no_snapshoting(root
);
1394 cow_start
= (u64
)-1;
1397 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1398 struct extent_map
*em
;
1399 struct extent_map_tree
*em_tree
;
1400 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1401 em
= alloc_extent_map();
1402 BUG_ON(!em
); /* -ENOMEM */
1403 em
->start
= cur_offset
;
1404 em
->orig_start
= found_key
.offset
- extent_offset
;
1405 em
->len
= num_bytes
;
1406 em
->block_len
= num_bytes
;
1407 em
->block_start
= disk_bytenr
;
1408 em
->orig_block_len
= disk_num_bytes
;
1409 em
->ram_bytes
= ram_bytes
;
1410 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1411 em
->mod_start
= em
->start
;
1412 em
->mod_len
= em
->len
;
1413 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1414 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1415 em
->generation
= -1;
1417 write_lock(&em_tree
->lock
);
1418 ret
= add_extent_mapping(em_tree
, em
, 1);
1419 write_unlock(&em_tree
->lock
);
1420 if (ret
!= -EEXIST
) {
1421 free_extent_map(em
);
1424 btrfs_drop_extent_cache(inode
, em
->start
,
1425 em
->start
+ em
->len
- 1, 0);
1427 type
= BTRFS_ORDERED_PREALLOC
;
1429 type
= BTRFS_ORDERED_NOCOW
;
1432 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1433 num_bytes
, num_bytes
, type
);
1434 BUG_ON(ret
); /* -ENOMEM */
1436 if (root
->root_key
.objectid
==
1437 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1438 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1441 if (!nolock
&& nocow
)
1442 btrfs_end_write_no_snapshoting(root
);
1447 extent_clear_unlock_delalloc(inode
, cur_offset
,
1448 cur_offset
+ num_bytes
- 1,
1449 locked_page
, EXTENT_LOCKED
|
1450 EXTENT_DELALLOC
, PAGE_UNLOCK
|
1452 if (!nolock
&& nocow
)
1453 btrfs_end_write_no_snapshoting(root
);
1454 cur_offset
= extent_end
;
1455 if (cur_offset
> end
)
1458 btrfs_release_path(path
);
1460 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1461 cow_start
= cur_offset
;
1465 if (cow_start
!= (u64
)-1) {
1466 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1467 page_started
, nr_written
, 1);
1473 err
= btrfs_end_transaction(trans
, root
);
1477 if (ret
&& cur_offset
< end
)
1478 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1479 locked_page
, EXTENT_LOCKED
|
1480 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1481 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1483 PAGE_SET_WRITEBACK
|
1484 PAGE_END_WRITEBACK
);
1485 btrfs_free_path(path
);
1489 static inline int need_force_cow(struct inode
*inode
, u64 start
, u64 end
)
1492 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
1493 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
))
1497 * @defrag_bytes is a hint value, no spinlock held here,
1498 * if is not zero, it means the file is defragging.
1499 * Force cow if given extent needs to be defragged.
1501 if (BTRFS_I(inode
)->defrag_bytes
&&
1502 test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1503 EXTENT_DEFRAG
, 0, NULL
))
1510 * extent_io.c call back to do delayed allocation processing
1512 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1513 u64 start
, u64 end
, int *page_started
,
1514 unsigned long *nr_written
)
1517 int force_cow
= need_force_cow(inode
, start
, end
);
1519 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
&& !force_cow
) {
1520 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1521 page_started
, 1, nr_written
);
1522 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
&& !force_cow
) {
1523 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1524 page_started
, 0, nr_written
);
1525 } else if (!inode_need_compress(inode
)) {
1526 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1527 page_started
, nr_written
, 1);
1529 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1530 &BTRFS_I(inode
)->runtime_flags
);
1531 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1532 page_started
, nr_written
);
1537 static void btrfs_split_extent_hook(struct inode
*inode
,
1538 struct extent_state
*orig
, u64 split
)
1542 /* not delalloc, ignore it */
1543 if (!(orig
->state
& EXTENT_DELALLOC
))
1546 size
= orig
->end
- orig
->start
+ 1;
1547 if (size
> BTRFS_MAX_EXTENT_SIZE
) {
1552 * See the explanation in btrfs_merge_extent_hook, the same
1553 * applies here, just in reverse.
1555 new_size
= orig
->end
- split
+ 1;
1556 num_extents
= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1557 BTRFS_MAX_EXTENT_SIZE
);
1558 new_size
= split
- orig
->start
;
1559 num_extents
+= div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1560 BTRFS_MAX_EXTENT_SIZE
);
1561 if (div64_u64(size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1562 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1566 spin_lock(&BTRFS_I(inode
)->lock
);
1567 BTRFS_I(inode
)->outstanding_extents
++;
1568 spin_unlock(&BTRFS_I(inode
)->lock
);
1572 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1573 * extents so we can keep track of new extents that are just merged onto old
1574 * extents, such as when we are doing sequential writes, so we can properly
1575 * account for the metadata space we'll need.
1577 static void btrfs_merge_extent_hook(struct inode
*inode
,
1578 struct extent_state
*new,
1579 struct extent_state
*other
)
1581 u64 new_size
, old_size
;
1584 /* not delalloc, ignore it */
1585 if (!(other
->state
& EXTENT_DELALLOC
))
1588 if (new->start
> other
->start
)
1589 new_size
= new->end
- other
->start
+ 1;
1591 new_size
= other
->end
- new->start
+ 1;
1593 /* we're not bigger than the max, unreserve the space and go */
1594 if (new_size
<= BTRFS_MAX_EXTENT_SIZE
) {
1595 spin_lock(&BTRFS_I(inode
)->lock
);
1596 BTRFS_I(inode
)->outstanding_extents
--;
1597 spin_unlock(&BTRFS_I(inode
)->lock
);
1602 * We have to add up either side to figure out how many extents were
1603 * accounted for before we merged into one big extent. If the number of
1604 * extents we accounted for is <= the amount we need for the new range
1605 * then we can return, otherwise drop. Think of it like this
1609 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1610 * need 2 outstanding extents, on one side we have 1 and the other side
1611 * we have 1 so they are == and we can return. But in this case
1613 * [MAX_SIZE+4k][MAX_SIZE+4k]
1615 * Each range on their own accounts for 2 extents, but merged together
1616 * they are only 3 extents worth of accounting, so we need to drop in
1619 old_size
= other
->end
- other
->start
+ 1;
1620 num_extents
= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1621 BTRFS_MAX_EXTENT_SIZE
);
1622 old_size
= new->end
- new->start
+ 1;
1623 num_extents
+= div64_u64(old_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1624 BTRFS_MAX_EXTENT_SIZE
);
1626 if (div64_u64(new_size
+ BTRFS_MAX_EXTENT_SIZE
- 1,
1627 BTRFS_MAX_EXTENT_SIZE
) >= num_extents
)
1630 spin_lock(&BTRFS_I(inode
)->lock
);
1631 BTRFS_I(inode
)->outstanding_extents
--;
1632 spin_unlock(&BTRFS_I(inode
)->lock
);
1635 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1636 struct inode
*inode
)
1638 spin_lock(&root
->delalloc_lock
);
1639 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1640 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1641 &root
->delalloc_inodes
);
1642 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1643 &BTRFS_I(inode
)->runtime_flags
);
1644 root
->nr_delalloc_inodes
++;
1645 if (root
->nr_delalloc_inodes
== 1) {
1646 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1647 BUG_ON(!list_empty(&root
->delalloc_root
));
1648 list_add_tail(&root
->delalloc_root
,
1649 &root
->fs_info
->delalloc_roots
);
1650 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1653 spin_unlock(&root
->delalloc_lock
);
1656 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1657 struct inode
*inode
)
1659 spin_lock(&root
->delalloc_lock
);
1660 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1661 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1662 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1663 &BTRFS_I(inode
)->runtime_flags
);
1664 root
->nr_delalloc_inodes
--;
1665 if (!root
->nr_delalloc_inodes
) {
1666 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1667 BUG_ON(list_empty(&root
->delalloc_root
));
1668 list_del_init(&root
->delalloc_root
);
1669 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1672 spin_unlock(&root
->delalloc_lock
);
1676 * extent_io.c set_bit_hook, used to track delayed allocation
1677 * bytes in this file, and to maintain the list of inodes that
1678 * have pending delalloc work to be done.
1680 static void btrfs_set_bit_hook(struct inode
*inode
,
1681 struct extent_state
*state
, unsigned *bits
)
1684 if ((*bits
& EXTENT_DEFRAG
) && !(*bits
& EXTENT_DELALLOC
))
1687 * set_bit and clear bit hooks normally require _irqsave/restore
1688 * but in this case, we are only testing for the DELALLOC
1689 * bit, which is only set or cleared with irqs on
1691 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1692 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1693 u64 len
= state
->end
+ 1 - state
->start
;
1694 bool do_list
= !btrfs_is_free_space_inode(inode
);
1696 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1697 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1699 spin_lock(&BTRFS_I(inode
)->lock
);
1700 BTRFS_I(inode
)->outstanding_extents
++;
1701 spin_unlock(&BTRFS_I(inode
)->lock
);
1704 /* For sanity tests */
1705 if (btrfs_test_is_dummy_root(root
))
1708 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1709 root
->fs_info
->delalloc_batch
);
1710 spin_lock(&BTRFS_I(inode
)->lock
);
1711 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1712 if (*bits
& EXTENT_DEFRAG
)
1713 BTRFS_I(inode
)->defrag_bytes
+= len
;
1714 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1715 &BTRFS_I(inode
)->runtime_flags
))
1716 btrfs_add_delalloc_inodes(root
, inode
);
1717 spin_unlock(&BTRFS_I(inode
)->lock
);
1722 * extent_io.c clear_bit_hook, see set_bit_hook for why
1724 static void btrfs_clear_bit_hook(struct inode
*inode
,
1725 struct extent_state
*state
,
1728 u64 len
= state
->end
+ 1 - state
->start
;
1729 u64 num_extents
= div64_u64(len
+ BTRFS_MAX_EXTENT_SIZE
-1,
1730 BTRFS_MAX_EXTENT_SIZE
);
1732 spin_lock(&BTRFS_I(inode
)->lock
);
1733 if ((state
->state
& EXTENT_DEFRAG
) && (*bits
& EXTENT_DEFRAG
))
1734 BTRFS_I(inode
)->defrag_bytes
-= len
;
1735 spin_unlock(&BTRFS_I(inode
)->lock
);
1738 * set_bit and clear bit hooks normally require _irqsave/restore
1739 * but in this case, we are only testing for the DELALLOC
1740 * bit, which is only set or cleared with irqs on
1742 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1743 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1744 bool do_list
= !btrfs_is_free_space_inode(inode
);
1746 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1747 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1748 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1749 spin_lock(&BTRFS_I(inode
)->lock
);
1750 BTRFS_I(inode
)->outstanding_extents
-= num_extents
;
1751 spin_unlock(&BTRFS_I(inode
)->lock
);
1755 * We don't reserve metadata space for space cache inodes so we
1756 * don't need to call dellalloc_release_metadata if there is an
1759 if (*bits
& EXTENT_DO_ACCOUNTING
&&
1760 root
!= root
->fs_info
->tree_root
)
1761 btrfs_delalloc_release_metadata(inode
, len
);
1763 /* For sanity tests. */
1764 if (btrfs_test_is_dummy_root(root
))
1767 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1768 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1769 btrfs_free_reserved_data_space(inode
, len
);
1771 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1772 root
->fs_info
->delalloc_batch
);
1773 spin_lock(&BTRFS_I(inode
)->lock
);
1774 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1775 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1776 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1777 &BTRFS_I(inode
)->runtime_flags
))
1778 btrfs_del_delalloc_inode(root
, inode
);
1779 spin_unlock(&BTRFS_I(inode
)->lock
);
1784 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1785 * we don't create bios that span stripes or chunks
1787 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1788 size_t size
, struct bio
*bio
,
1789 unsigned long bio_flags
)
1791 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1792 u64 logical
= (u64
)bio
->bi_iter
.bi_sector
<< 9;
1797 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1800 length
= bio
->bi_iter
.bi_size
;
1801 map_length
= length
;
1802 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1803 &map_length
, NULL
, 0);
1804 /* Will always return 0 with map_multi == NULL */
1806 if (map_length
< length
+ size
)
1812 * in order to insert checksums into the metadata in large chunks,
1813 * we wait until bio submission time. All the pages in the bio are
1814 * checksummed and sums are attached onto the ordered extent record.
1816 * At IO completion time the cums attached on the ordered extent record
1817 * are inserted into the btree
1819 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1820 struct bio
*bio
, int mirror_num
,
1821 unsigned long bio_flags
,
1824 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1827 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1828 BUG_ON(ret
); /* -ENOMEM */
1833 * in order to insert checksums into the metadata in large chunks,
1834 * we wait until bio submission time. All the pages in the bio are
1835 * checksummed and sums are attached onto the ordered extent record.
1837 * At IO completion time the cums attached on the ordered extent record
1838 * are inserted into the btree
1840 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1841 int mirror_num
, unsigned long bio_flags
,
1844 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1847 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1849 bio
->bi_error
= ret
;
1856 * extent_io.c submission hook. This does the right thing for csum calculation
1857 * on write, or reading the csums from the tree before a read
1859 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1860 int mirror_num
, unsigned long bio_flags
,
1863 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1867 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1869 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1871 if (btrfs_is_free_space_inode(inode
))
1874 if (!(rw
& REQ_WRITE
)) {
1875 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1879 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1880 ret
= btrfs_submit_compressed_read(inode
, bio
,
1884 } else if (!skip_sum
) {
1885 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1890 } else if (async
&& !skip_sum
) {
1891 /* csum items have already been cloned */
1892 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1894 /* we're doing a write, do the async checksumming */
1895 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1896 inode
, rw
, bio
, mirror_num
,
1897 bio_flags
, bio_offset
,
1898 __btrfs_submit_bio_start
,
1899 __btrfs_submit_bio_done
);
1901 } else if (!skip_sum
) {
1902 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1908 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1912 bio
->bi_error
= ret
;
1919 * given a list of ordered sums record them in the inode. This happens
1920 * at IO completion time based on sums calculated at bio submission time.
1922 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1923 struct inode
*inode
, u64 file_offset
,
1924 struct list_head
*list
)
1926 struct btrfs_ordered_sum
*sum
;
1928 list_for_each_entry(sum
, list
, list
) {
1929 trans
->adding_csums
= 1;
1930 btrfs_csum_file_blocks(trans
,
1931 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1932 trans
->adding_csums
= 0;
1937 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1938 struct extent_state
**cached_state
)
1940 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1941 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1942 cached_state
, GFP_NOFS
);
1945 /* see btrfs_writepage_start_hook for details on why this is required */
1946 struct btrfs_writepage_fixup
{
1948 struct btrfs_work work
;
1951 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1953 struct btrfs_writepage_fixup
*fixup
;
1954 struct btrfs_ordered_extent
*ordered
;
1955 struct extent_state
*cached_state
= NULL
;
1957 struct inode
*inode
;
1962 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1966 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1967 ClearPageChecked(page
);
1971 inode
= page
->mapping
->host
;
1972 page_start
= page_offset(page
);
1973 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1975 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1978 /* already ordered? We're done */
1979 if (PagePrivate2(page
))
1982 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1984 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1985 page_end
, &cached_state
, GFP_NOFS
);
1987 btrfs_start_ordered_extent(inode
, ordered
, 1);
1988 btrfs_put_ordered_extent(ordered
);
1992 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1994 mapping_set_error(page
->mapping
, ret
);
1995 end_extent_writepage(page
, ret
, page_start
, page_end
);
1996 ClearPageChecked(page
);
2000 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
2001 ClearPageChecked(page
);
2002 set_page_dirty(page
);
2004 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
2005 &cached_state
, GFP_NOFS
);
2008 page_cache_release(page
);
2013 * There are a few paths in the higher layers of the kernel that directly
2014 * set the page dirty bit without asking the filesystem if it is a
2015 * good idea. This causes problems because we want to make sure COW
2016 * properly happens and the data=ordered rules are followed.
2018 * In our case any range that doesn't have the ORDERED bit set
2019 * hasn't been properly setup for IO. We kick off an async process
2020 * to fix it up. The async helper will wait for ordered extents, set
2021 * the delalloc bit and make it safe to write the page.
2023 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
2025 struct inode
*inode
= page
->mapping
->host
;
2026 struct btrfs_writepage_fixup
*fixup
;
2027 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2029 /* this page is properly in the ordered list */
2030 if (TestClearPagePrivate2(page
))
2033 if (PageChecked(page
))
2036 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
2040 SetPageChecked(page
);
2041 page_cache_get(page
);
2042 btrfs_init_work(&fixup
->work
, btrfs_fixup_helper
,
2043 btrfs_writepage_fixup_worker
, NULL
, NULL
);
2045 btrfs_queue_work(root
->fs_info
->fixup_workers
, &fixup
->work
);
2049 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
2050 struct inode
*inode
, u64 file_pos
,
2051 u64 disk_bytenr
, u64 disk_num_bytes
,
2052 u64 num_bytes
, u64 ram_bytes
,
2053 u8 compression
, u8 encryption
,
2054 u16 other_encoding
, int extent_type
)
2056 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2057 struct btrfs_file_extent_item
*fi
;
2058 struct btrfs_path
*path
;
2059 struct extent_buffer
*leaf
;
2060 struct btrfs_key ins
;
2061 int extent_inserted
= 0;
2064 path
= btrfs_alloc_path();
2069 * we may be replacing one extent in the tree with another.
2070 * The new extent is pinned in the extent map, and we don't want
2071 * to drop it from the cache until it is completely in the btree.
2073 * So, tell btrfs_drop_extents to leave this extent in the cache.
2074 * the caller is expected to unpin it and allow it to be merged
2077 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
2078 file_pos
+ num_bytes
, NULL
, 0,
2079 1, sizeof(*fi
), &extent_inserted
);
2083 if (!extent_inserted
) {
2084 ins
.objectid
= btrfs_ino(inode
);
2085 ins
.offset
= file_pos
;
2086 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
2088 path
->leave_spinning
= 1;
2089 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
2094 leaf
= path
->nodes
[0];
2095 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2096 struct btrfs_file_extent_item
);
2097 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
2098 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
2099 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
2100 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
2101 btrfs_set_file_extent_offset(leaf
, fi
, 0);
2102 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
2103 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
2104 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
2105 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
2106 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
2108 btrfs_mark_buffer_dirty(leaf
);
2109 btrfs_release_path(path
);
2111 inode_add_bytes(inode
, num_bytes
);
2113 ins
.objectid
= disk_bytenr
;
2114 ins
.offset
= disk_num_bytes
;
2115 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2116 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
2117 root
->root_key
.objectid
,
2118 btrfs_ino(inode
), file_pos
, &ins
);
2120 btrfs_free_path(path
);
2125 /* snapshot-aware defrag */
2126 struct sa_defrag_extent_backref
{
2127 struct rb_node node
;
2128 struct old_sa_defrag_extent
*old
;
2137 struct old_sa_defrag_extent
{
2138 struct list_head list
;
2139 struct new_sa_defrag_extent
*new;
2148 struct new_sa_defrag_extent
{
2149 struct rb_root root
;
2150 struct list_head head
;
2151 struct btrfs_path
*path
;
2152 struct inode
*inode
;
2160 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
2161 struct sa_defrag_extent_backref
*b2
)
2163 if (b1
->root_id
< b2
->root_id
)
2165 else if (b1
->root_id
> b2
->root_id
)
2168 if (b1
->inum
< b2
->inum
)
2170 else if (b1
->inum
> b2
->inum
)
2173 if (b1
->file_pos
< b2
->file_pos
)
2175 else if (b1
->file_pos
> b2
->file_pos
)
2179 * [------------------------------] ===> (a range of space)
2180 * |<--->| |<---->| =============> (fs/file tree A)
2181 * |<---------------------------->| ===> (fs/file tree B)
2183 * A range of space can refer to two file extents in one tree while
2184 * refer to only one file extent in another tree.
2186 * So we may process a disk offset more than one time(two extents in A)
2187 * and locate at the same extent(one extent in B), then insert two same
2188 * backrefs(both refer to the extent in B).
2193 static void backref_insert(struct rb_root
*root
,
2194 struct sa_defrag_extent_backref
*backref
)
2196 struct rb_node
**p
= &root
->rb_node
;
2197 struct rb_node
*parent
= NULL
;
2198 struct sa_defrag_extent_backref
*entry
;
2203 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2205 ret
= backref_comp(backref
, entry
);
2209 p
= &(*p
)->rb_right
;
2212 rb_link_node(&backref
->node
, parent
, p
);
2213 rb_insert_color(&backref
->node
, root
);
2217 * Note the backref might has changed, and in this case we just return 0.
2219 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2222 struct btrfs_file_extent_item
*extent
;
2223 struct btrfs_fs_info
*fs_info
;
2224 struct old_sa_defrag_extent
*old
= ctx
;
2225 struct new_sa_defrag_extent
*new = old
->new;
2226 struct btrfs_path
*path
= new->path
;
2227 struct btrfs_key key
;
2228 struct btrfs_root
*root
;
2229 struct sa_defrag_extent_backref
*backref
;
2230 struct extent_buffer
*leaf
;
2231 struct inode
*inode
= new->inode
;
2237 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2238 inum
== btrfs_ino(inode
))
2241 key
.objectid
= root_id
;
2242 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2243 key
.offset
= (u64
)-1;
2245 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2246 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2248 if (PTR_ERR(root
) == -ENOENT
)
2251 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2252 inum
, offset
, root_id
);
2253 return PTR_ERR(root
);
2256 key
.objectid
= inum
;
2257 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2258 if (offset
> (u64
)-1 << 32)
2261 key
.offset
= offset
;
2263 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2264 if (WARN_ON(ret
< 0))
2271 leaf
= path
->nodes
[0];
2272 slot
= path
->slots
[0];
2274 if (slot
>= btrfs_header_nritems(leaf
)) {
2275 ret
= btrfs_next_leaf(root
, path
);
2278 } else if (ret
> 0) {
2287 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2289 if (key
.objectid
> inum
)
2292 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2295 extent
= btrfs_item_ptr(leaf
, slot
,
2296 struct btrfs_file_extent_item
);
2298 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2302 * 'offset' refers to the exact key.offset,
2303 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2304 * (key.offset - extent_offset).
2306 if (key
.offset
!= offset
)
2309 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2310 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2312 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2313 old
->len
|| extent_offset
+ num_bytes
<=
2314 old
->extent_offset
+ old
->offset
)
2319 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2325 backref
->root_id
= root_id
;
2326 backref
->inum
= inum
;
2327 backref
->file_pos
= offset
;
2328 backref
->num_bytes
= num_bytes
;
2329 backref
->extent_offset
= extent_offset
;
2330 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2332 backref_insert(&new->root
, backref
);
2335 btrfs_release_path(path
);
2340 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2341 struct new_sa_defrag_extent
*new)
2343 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2344 struct old_sa_defrag_extent
*old
, *tmp
;
2349 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2350 ret
= iterate_inodes_from_logical(old
->bytenr
+
2351 old
->extent_offset
, fs_info
,
2352 path
, record_one_backref
,
2354 if (ret
< 0 && ret
!= -ENOENT
)
2357 /* no backref to be processed for this extent */
2359 list_del(&old
->list
);
2364 if (list_empty(&new->head
))
2370 static int relink_is_mergable(struct extent_buffer
*leaf
,
2371 struct btrfs_file_extent_item
*fi
,
2372 struct new_sa_defrag_extent
*new)
2374 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2377 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2380 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2383 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2384 btrfs_file_extent_other_encoding(leaf
, fi
))
2391 * Note the backref might has changed, and in this case we just return 0.
2393 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2394 struct sa_defrag_extent_backref
*prev
,
2395 struct sa_defrag_extent_backref
*backref
)
2397 struct btrfs_file_extent_item
*extent
;
2398 struct btrfs_file_extent_item
*item
;
2399 struct btrfs_ordered_extent
*ordered
;
2400 struct btrfs_trans_handle
*trans
;
2401 struct btrfs_fs_info
*fs_info
;
2402 struct btrfs_root
*root
;
2403 struct btrfs_key key
;
2404 struct extent_buffer
*leaf
;
2405 struct old_sa_defrag_extent
*old
= backref
->old
;
2406 struct new_sa_defrag_extent
*new = old
->new;
2407 struct inode
*src_inode
= new->inode
;
2408 struct inode
*inode
;
2409 struct extent_state
*cached
= NULL
;
2418 if (prev
&& prev
->root_id
== backref
->root_id
&&
2419 prev
->inum
== backref
->inum
&&
2420 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2423 /* step 1: get root */
2424 key
.objectid
= backref
->root_id
;
2425 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2426 key
.offset
= (u64
)-1;
2428 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2429 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2431 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2433 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2434 if (PTR_ERR(root
) == -ENOENT
)
2436 return PTR_ERR(root
);
2439 if (btrfs_root_readonly(root
)) {
2440 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2444 /* step 2: get inode */
2445 key
.objectid
= backref
->inum
;
2446 key
.type
= BTRFS_INODE_ITEM_KEY
;
2449 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2450 if (IS_ERR(inode
)) {
2451 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2455 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2457 /* step 3: relink backref */
2458 lock_start
= backref
->file_pos
;
2459 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2460 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2463 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2465 btrfs_put_ordered_extent(ordered
);
2469 trans
= btrfs_join_transaction(root
);
2470 if (IS_ERR(trans
)) {
2471 ret
= PTR_ERR(trans
);
2475 key
.objectid
= backref
->inum
;
2476 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2477 key
.offset
= backref
->file_pos
;
2479 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2482 } else if (ret
> 0) {
2487 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2488 struct btrfs_file_extent_item
);
2490 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2491 backref
->generation
)
2494 btrfs_release_path(path
);
2496 start
= backref
->file_pos
;
2497 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2498 start
+= old
->extent_offset
+ old
->offset
-
2499 backref
->extent_offset
;
2501 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2502 old
->extent_offset
+ old
->offset
+ old
->len
);
2503 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2505 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2510 key
.objectid
= btrfs_ino(inode
);
2511 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2514 path
->leave_spinning
= 1;
2516 struct btrfs_file_extent_item
*fi
;
2518 struct btrfs_key found_key
;
2520 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2525 leaf
= path
->nodes
[0];
2526 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2528 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2529 struct btrfs_file_extent_item
);
2530 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2532 if (extent_len
+ found_key
.offset
== start
&&
2533 relink_is_mergable(leaf
, fi
, new)) {
2534 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2536 btrfs_mark_buffer_dirty(leaf
);
2537 inode_add_bytes(inode
, len
);
2543 btrfs_release_path(path
);
2548 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2551 btrfs_abort_transaction(trans
, root
, ret
);
2555 leaf
= path
->nodes
[0];
2556 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2557 struct btrfs_file_extent_item
);
2558 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2559 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2560 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2561 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2562 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2563 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2564 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2565 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2566 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2567 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2569 btrfs_mark_buffer_dirty(leaf
);
2570 inode_add_bytes(inode
, len
);
2571 btrfs_release_path(path
);
2573 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2575 backref
->root_id
, backref
->inum
,
2576 new->file_pos
, 0); /* start - extent_offset */
2578 btrfs_abort_transaction(trans
, root
, ret
);
2584 btrfs_release_path(path
);
2585 path
->leave_spinning
= 0;
2586 btrfs_end_transaction(trans
, root
);
2588 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2594 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2596 struct old_sa_defrag_extent
*old
, *tmp
;
2601 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2602 list_del(&old
->list
);
2608 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2610 struct btrfs_path
*path
;
2611 struct sa_defrag_extent_backref
*backref
;
2612 struct sa_defrag_extent_backref
*prev
= NULL
;
2613 struct inode
*inode
;
2614 struct btrfs_root
*root
;
2615 struct rb_node
*node
;
2619 root
= BTRFS_I(inode
)->root
;
2621 path
= btrfs_alloc_path();
2625 if (!record_extent_backrefs(path
, new)) {
2626 btrfs_free_path(path
);
2629 btrfs_release_path(path
);
2632 node
= rb_first(&new->root
);
2635 rb_erase(node
, &new->root
);
2637 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2639 ret
= relink_extent_backref(path
, prev
, backref
);
2652 btrfs_free_path(path
);
2654 free_sa_defrag_extent(new);
2656 atomic_dec(&root
->fs_info
->defrag_running
);
2657 wake_up(&root
->fs_info
->transaction_wait
);
2660 static struct new_sa_defrag_extent
*
2661 record_old_file_extents(struct inode
*inode
,
2662 struct btrfs_ordered_extent
*ordered
)
2664 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2665 struct btrfs_path
*path
;
2666 struct btrfs_key key
;
2667 struct old_sa_defrag_extent
*old
;
2668 struct new_sa_defrag_extent
*new;
2671 new = kmalloc(sizeof(*new), GFP_NOFS
);
2676 new->file_pos
= ordered
->file_offset
;
2677 new->len
= ordered
->len
;
2678 new->bytenr
= ordered
->start
;
2679 new->disk_len
= ordered
->disk_len
;
2680 new->compress_type
= ordered
->compress_type
;
2681 new->root
= RB_ROOT
;
2682 INIT_LIST_HEAD(&new->head
);
2684 path
= btrfs_alloc_path();
2688 key
.objectid
= btrfs_ino(inode
);
2689 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2690 key
.offset
= new->file_pos
;
2692 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2695 if (ret
> 0 && path
->slots
[0] > 0)
2698 /* find out all the old extents for the file range */
2700 struct btrfs_file_extent_item
*extent
;
2701 struct extent_buffer
*l
;
2710 slot
= path
->slots
[0];
2712 if (slot
>= btrfs_header_nritems(l
)) {
2713 ret
= btrfs_next_leaf(root
, path
);
2721 btrfs_item_key_to_cpu(l
, &key
, slot
);
2723 if (key
.objectid
!= btrfs_ino(inode
))
2725 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2727 if (key
.offset
>= new->file_pos
+ new->len
)
2730 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2732 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2733 if (key
.offset
+ num_bytes
< new->file_pos
)
2736 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2740 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2742 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2746 offset
= max(new->file_pos
, key
.offset
);
2747 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2749 old
->bytenr
= disk_bytenr
;
2750 old
->extent_offset
= extent_offset
;
2751 old
->offset
= offset
- key
.offset
;
2752 old
->len
= end
- offset
;
2755 list_add_tail(&old
->list
, &new->head
);
2761 btrfs_free_path(path
);
2762 atomic_inc(&root
->fs_info
->defrag_running
);
2767 btrfs_free_path(path
);
2769 free_sa_defrag_extent(new);
2773 static void btrfs_release_delalloc_bytes(struct btrfs_root
*root
,
2776 struct btrfs_block_group_cache
*cache
;
2778 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
2781 spin_lock(&cache
->lock
);
2782 cache
->delalloc_bytes
-= len
;
2783 spin_unlock(&cache
->lock
);
2785 btrfs_put_block_group(cache
);
2788 /* as ordered data IO finishes, this gets called so we can finish
2789 * an ordered extent if the range of bytes in the file it covers are
2792 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2794 struct inode
*inode
= ordered_extent
->inode
;
2795 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2796 struct btrfs_trans_handle
*trans
= NULL
;
2797 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2798 struct extent_state
*cached_state
= NULL
;
2799 struct new_sa_defrag_extent
*new = NULL
;
2800 int compress_type
= 0;
2802 u64 logical_len
= ordered_extent
->len
;
2804 bool truncated
= false;
2806 nolock
= btrfs_is_free_space_inode(inode
);
2808 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2813 btrfs_free_io_failure_record(inode
, ordered_extent
->file_offset
,
2814 ordered_extent
->file_offset
+
2815 ordered_extent
->len
- 1);
2817 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2819 logical_len
= ordered_extent
->truncated_len
;
2820 /* Truncated the entire extent, don't bother adding */
2825 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2826 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2827 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2829 trans
= btrfs_join_transaction_nolock(root
);
2831 trans
= btrfs_join_transaction(root
);
2832 if (IS_ERR(trans
)) {
2833 ret
= PTR_ERR(trans
);
2837 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2838 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2839 if (ret
) /* -ENOMEM or corruption */
2840 btrfs_abort_transaction(trans
, root
, ret
);
2844 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2845 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2848 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2849 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2850 EXTENT_DEFRAG
, 1, cached_state
);
2852 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2853 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2854 /* the inode is shared */
2855 new = record_old_file_extents(inode
, ordered_extent
);
2857 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2858 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2859 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2863 trans
= btrfs_join_transaction_nolock(root
);
2865 trans
= btrfs_join_transaction(root
);
2866 if (IS_ERR(trans
)) {
2867 ret
= PTR_ERR(trans
);
2872 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2874 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2875 compress_type
= ordered_extent
->compress_type
;
2876 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2877 BUG_ON(compress_type
);
2878 ret
= btrfs_mark_extent_written(trans
, inode
,
2879 ordered_extent
->file_offset
,
2880 ordered_extent
->file_offset
+
2883 BUG_ON(root
== root
->fs_info
->tree_root
);
2884 ret
= insert_reserved_file_extent(trans
, inode
,
2885 ordered_extent
->file_offset
,
2886 ordered_extent
->start
,
2887 ordered_extent
->disk_len
,
2888 logical_len
, logical_len
,
2889 compress_type
, 0, 0,
2890 BTRFS_FILE_EXTENT_REG
);
2892 btrfs_release_delalloc_bytes(root
,
2893 ordered_extent
->start
,
2894 ordered_extent
->disk_len
);
2896 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2897 ordered_extent
->file_offset
, ordered_extent
->len
,
2900 btrfs_abort_transaction(trans
, root
, ret
);
2904 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2905 &ordered_extent
->list
);
2907 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2908 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2909 if (ret
) { /* -ENOMEM or corruption */
2910 btrfs_abort_transaction(trans
, root
, ret
);
2915 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2916 ordered_extent
->file_offset
+
2917 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2919 if (root
!= root
->fs_info
->tree_root
)
2920 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2922 btrfs_end_transaction(trans
, root
);
2924 if (ret
|| truncated
) {
2928 start
= ordered_extent
->file_offset
+ logical_len
;
2930 start
= ordered_extent
->file_offset
;
2931 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
2932 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
2934 /* Drop the cache for the part of the extent we didn't write. */
2935 btrfs_drop_extent_cache(inode
, start
, end
, 0);
2938 * If the ordered extent had an IOERR or something else went
2939 * wrong we need to return the space for this ordered extent
2940 * back to the allocator. We only free the extent in the
2941 * truncated case if we didn't write out the extent at all.
2943 if ((ret
|| !logical_len
) &&
2944 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2945 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2946 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
2947 ordered_extent
->disk_len
, 1);
2952 * This needs to be done to make sure anybody waiting knows we are done
2953 * updating everything for this ordered extent.
2955 btrfs_remove_ordered_extent(inode
, ordered_extent
);
2957 /* for snapshot-aware defrag */
2960 free_sa_defrag_extent(new);
2961 atomic_dec(&root
->fs_info
->defrag_running
);
2963 relink_file_extents(new);
2968 btrfs_put_ordered_extent(ordered_extent
);
2969 /* once for the tree */
2970 btrfs_put_ordered_extent(ordered_extent
);
2975 static void finish_ordered_fn(struct btrfs_work
*work
)
2977 struct btrfs_ordered_extent
*ordered_extent
;
2978 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
2979 btrfs_finish_ordered_io(ordered_extent
);
2982 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2983 struct extent_state
*state
, int uptodate
)
2985 struct inode
*inode
= page
->mapping
->host
;
2986 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2987 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
2988 struct btrfs_workqueue
*wq
;
2989 btrfs_work_func_t func
;
2991 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
2993 ClearPagePrivate2(page
);
2994 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
2995 end
- start
+ 1, uptodate
))
2998 if (btrfs_is_free_space_inode(inode
)) {
2999 wq
= root
->fs_info
->endio_freespace_worker
;
3000 func
= btrfs_freespace_write_helper
;
3002 wq
= root
->fs_info
->endio_write_workers
;
3003 func
= btrfs_endio_write_helper
;
3006 btrfs_init_work(&ordered_extent
->work
, func
, finish_ordered_fn
, NULL
,
3008 btrfs_queue_work(wq
, &ordered_extent
->work
);
3013 static int __readpage_endio_check(struct inode
*inode
,
3014 struct btrfs_io_bio
*io_bio
,
3015 int icsum
, struct page
*page
,
3016 int pgoff
, u64 start
, size_t len
)
3021 static DEFINE_RATELIMIT_STATE(_rs
, DEFAULT_RATELIMIT_INTERVAL
,
3022 DEFAULT_RATELIMIT_BURST
);
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 if (__ratelimit(&_rs
))
3036 btrfs_warn(BTRFS_I(inode
)->root
->fs_info
,
3037 "csum failed ino %llu off %llu csum %u expected csum %u",
3038 btrfs_ino(inode
), start
, csum
, csum_expected
);
3039 memset(kaddr
+ pgoff
, 1, len
);
3040 flush_dcache_page(page
);
3041 kunmap_atomic(kaddr
);
3042 if (csum_expected
== 0)
3048 * when reads are done, we need to check csums to verify the data is correct
3049 * if there's a match, we allow the bio to finish. If not, the code in
3050 * extent_io.c will try to find good copies for us.
3052 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
3053 u64 phy_offset
, struct page
*page
,
3054 u64 start
, u64 end
, int mirror
)
3056 size_t offset
= start
- page_offset(page
);
3057 struct inode
*inode
= page
->mapping
->host
;
3058 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3059 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3061 if (PageChecked(page
)) {
3062 ClearPageChecked(page
);
3066 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
3069 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
3070 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
3071 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
3076 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
3077 return __readpage_endio_check(inode
, io_bio
, phy_offset
, page
, offset
,
3078 start
, (size_t)(end
- start
+ 1));
3081 struct delayed_iput
{
3082 struct list_head list
;
3083 struct inode
*inode
;
3086 /* JDM: If this is fs-wide, why can't we add a pointer to
3087 * btrfs_inode instead and avoid the allocation? */
3088 void btrfs_add_delayed_iput(struct inode
*inode
)
3090 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
3091 struct delayed_iput
*delayed
;
3093 if (atomic_add_unless(&inode
->i_count
, -1, 1))
3096 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
3097 delayed
->inode
= inode
;
3099 spin_lock(&fs_info
->delayed_iput_lock
);
3100 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
3101 spin_unlock(&fs_info
->delayed_iput_lock
);
3104 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
3107 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3108 struct delayed_iput
*delayed
;
3111 spin_lock(&fs_info
->delayed_iput_lock
);
3112 empty
= list_empty(&fs_info
->delayed_iputs
);
3113 spin_unlock(&fs_info
->delayed_iput_lock
);
3117 down_read(&fs_info
->delayed_iput_sem
);
3119 spin_lock(&fs_info
->delayed_iput_lock
);
3120 list_splice_init(&fs_info
->delayed_iputs
, &list
);
3121 spin_unlock(&fs_info
->delayed_iput_lock
);
3123 while (!list_empty(&list
)) {
3124 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
3125 list_del(&delayed
->list
);
3126 iput(delayed
->inode
);
3130 up_read(&root
->fs_info
->delayed_iput_sem
);
3134 * This is called in transaction commit time. If there are no orphan
3135 * files in the subvolume, it removes orphan item and frees block_rsv
3138 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
3139 struct btrfs_root
*root
)
3141 struct btrfs_block_rsv
*block_rsv
;
3144 if (atomic_read(&root
->orphan_inodes
) ||
3145 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
3148 spin_lock(&root
->orphan_lock
);
3149 if (atomic_read(&root
->orphan_inodes
)) {
3150 spin_unlock(&root
->orphan_lock
);
3154 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
3155 spin_unlock(&root
->orphan_lock
);
3159 block_rsv
= root
->orphan_block_rsv
;
3160 root
->orphan_block_rsv
= NULL
;
3161 spin_unlock(&root
->orphan_lock
);
3163 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
) &&
3164 btrfs_root_refs(&root
->root_item
) > 0) {
3165 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
3166 root
->root_key
.objectid
);
3168 btrfs_abort_transaction(trans
, root
, ret
);
3170 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
,
3175 WARN_ON(block_rsv
->size
> 0);
3176 btrfs_free_block_rsv(root
, block_rsv
);
3181 * This creates an orphan entry for the given inode in case something goes
3182 * wrong in the middle of an unlink/truncate.
3184 * NOTE: caller of this function should reserve 5 units of metadata for
3187 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
3189 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3190 struct btrfs_block_rsv
*block_rsv
= NULL
;
3195 if (!root
->orphan_block_rsv
) {
3196 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3201 spin_lock(&root
->orphan_lock
);
3202 if (!root
->orphan_block_rsv
) {
3203 root
->orphan_block_rsv
= block_rsv
;
3204 } else if (block_rsv
) {
3205 btrfs_free_block_rsv(root
, block_rsv
);
3209 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3210 &BTRFS_I(inode
)->runtime_flags
)) {
3213 * For proper ENOSPC handling, we should do orphan
3214 * cleanup when mounting. But this introduces backward
3215 * compatibility issue.
3217 if (!xchg(&root
->orphan_item_inserted
, 1))
3223 atomic_inc(&root
->orphan_inodes
);
3226 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3227 &BTRFS_I(inode
)->runtime_flags
))
3229 spin_unlock(&root
->orphan_lock
);
3231 /* grab metadata reservation from transaction handle */
3233 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3234 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
3237 /* insert an orphan item to track this unlinked/truncated file */
3239 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3241 atomic_dec(&root
->orphan_inodes
);
3243 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3244 &BTRFS_I(inode
)->runtime_flags
);
3245 btrfs_orphan_release_metadata(inode
);
3247 if (ret
!= -EEXIST
) {
3248 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3249 &BTRFS_I(inode
)->runtime_flags
);
3250 btrfs_abort_transaction(trans
, root
, ret
);
3257 /* insert an orphan item to track subvolume contains orphan files */
3259 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3260 root
->root_key
.objectid
);
3261 if (ret
&& ret
!= -EEXIST
) {
3262 btrfs_abort_transaction(trans
, root
, ret
);
3270 * We have done the truncate/delete so we can go ahead and remove the orphan
3271 * item for this particular inode.
3273 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3274 struct inode
*inode
)
3276 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3277 int delete_item
= 0;
3278 int release_rsv
= 0;
3281 spin_lock(&root
->orphan_lock
);
3282 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3283 &BTRFS_I(inode
)->runtime_flags
))
3286 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3287 &BTRFS_I(inode
)->runtime_flags
))
3289 spin_unlock(&root
->orphan_lock
);
3292 atomic_dec(&root
->orphan_inodes
);
3294 ret
= btrfs_del_orphan_item(trans
, root
,
3299 btrfs_orphan_release_metadata(inode
);
3305 * this cleans up any orphans that may be left on the list from the last use
3308 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3310 struct btrfs_path
*path
;
3311 struct extent_buffer
*leaf
;
3312 struct btrfs_key key
, found_key
;
3313 struct btrfs_trans_handle
*trans
;
3314 struct inode
*inode
;
3315 u64 last_objectid
= 0;
3316 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3318 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3321 path
= btrfs_alloc_path();
3328 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3329 key
.type
= BTRFS_ORPHAN_ITEM_KEY
;
3330 key
.offset
= (u64
)-1;
3333 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3338 * if ret == 0 means we found what we were searching for, which
3339 * is weird, but possible, so only screw with path if we didn't
3340 * find the key and see if we have stuff that matches
3344 if (path
->slots
[0] == 0)
3349 /* pull out the item */
3350 leaf
= path
->nodes
[0];
3351 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3353 /* make sure the item matches what we want */
3354 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3356 if (found_key
.type
!= BTRFS_ORPHAN_ITEM_KEY
)
3359 /* release the path since we're done with it */
3360 btrfs_release_path(path
);
3363 * this is where we are basically btrfs_lookup, without the
3364 * crossing root thing. we store the inode number in the
3365 * offset of the orphan item.
3368 if (found_key
.offset
== last_objectid
) {
3369 btrfs_err(root
->fs_info
,
3370 "Error removing orphan entry, stopping orphan cleanup");
3375 last_objectid
= found_key
.offset
;
3377 found_key
.objectid
= found_key
.offset
;
3378 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3379 found_key
.offset
= 0;
3380 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3381 ret
= PTR_ERR_OR_ZERO(inode
);
3382 if (ret
&& ret
!= -ESTALE
)
3385 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3386 struct btrfs_root
*dead_root
;
3387 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3388 int is_dead_root
= 0;
3391 * this is an orphan in the tree root. Currently these
3392 * could come from 2 sources:
3393 * a) a snapshot deletion in progress
3394 * b) a free space cache inode
3395 * We need to distinguish those two, as the snapshot
3396 * orphan must not get deleted.
3397 * find_dead_roots already ran before us, so if this
3398 * is a snapshot deletion, we should find the root
3399 * in the dead_roots list
3401 spin_lock(&fs_info
->trans_lock
);
3402 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3404 if (dead_root
->root_key
.objectid
==
3405 found_key
.objectid
) {
3410 spin_unlock(&fs_info
->trans_lock
);
3412 /* prevent this orphan from being found again */
3413 key
.offset
= found_key
.objectid
- 1;
3418 * Inode is already gone but the orphan item is still there,
3419 * kill the orphan item.
3421 if (ret
== -ESTALE
) {
3422 trans
= btrfs_start_transaction(root
, 1);
3423 if (IS_ERR(trans
)) {
3424 ret
= PTR_ERR(trans
);
3427 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3428 found_key
.objectid
);
3429 ret
= btrfs_del_orphan_item(trans
, root
,
3430 found_key
.objectid
);
3431 btrfs_end_transaction(trans
, root
);
3438 * add this inode to the orphan list so btrfs_orphan_del does
3439 * the proper thing when we hit it
3441 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3442 &BTRFS_I(inode
)->runtime_flags
);
3443 atomic_inc(&root
->orphan_inodes
);
3445 /* if we have links, this was a truncate, lets do that */
3446 if (inode
->i_nlink
) {
3447 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3453 /* 1 for the orphan item deletion. */
3454 trans
= btrfs_start_transaction(root
, 1);
3455 if (IS_ERR(trans
)) {
3457 ret
= PTR_ERR(trans
);
3460 ret
= btrfs_orphan_add(trans
, inode
);
3461 btrfs_end_transaction(trans
, root
);
3467 ret
= btrfs_truncate(inode
);
3469 btrfs_orphan_del(NULL
, inode
);
3474 /* this will do delete_inode and everything for us */
3479 /* release the path since we're done with it */
3480 btrfs_release_path(path
);
3482 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3484 if (root
->orphan_block_rsv
)
3485 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3488 if (root
->orphan_block_rsv
||
3489 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED
, &root
->state
)) {
3490 trans
= btrfs_join_transaction(root
);
3492 btrfs_end_transaction(trans
, root
);
3496 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3498 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3502 btrfs_err(root
->fs_info
,
3503 "could not do orphan cleanup %d", ret
);
3504 btrfs_free_path(path
);
3509 * very simple check to peek ahead in the leaf looking for xattrs. If we
3510 * don't find any xattrs, we know there can't be any acls.
3512 * slot is the slot the inode is in, objectid is the objectid of the inode
3514 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3515 int slot
, u64 objectid
,
3516 int *first_xattr_slot
)
3518 u32 nritems
= btrfs_header_nritems(leaf
);
3519 struct btrfs_key found_key
;
3520 static u64 xattr_access
= 0;
3521 static u64 xattr_default
= 0;
3524 if (!xattr_access
) {
3525 xattr_access
= btrfs_name_hash(POSIX_ACL_XATTR_ACCESS
,
3526 strlen(POSIX_ACL_XATTR_ACCESS
));
3527 xattr_default
= btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT
,
3528 strlen(POSIX_ACL_XATTR_DEFAULT
));
3532 *first_xattr_slot
= -1;
3533 while (slot
< nritems
) {
3534 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3536 /* we found a different objectid, there must not be acls */
3537 if (found_key
.objectid
!= objectid
)
3540 /* we found an xattr, assume we've got an acl */
3541 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3542 if (*first_xattr_slot
== -1)
3543 *first_xattr_slot
= slot
;
3544 if (found_key
.offset
== xattr_access
||
3545 found_key
.offset
== xattr_default
)
3550 * we found a key greater than an xattr key, there can't
3551 * be any acls later on
3553 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3560 * it goes inode, inode backrefs, xattrs, extents,
3561 * so if there are a ton of hard links to an inode there can
3562 * be a lot of backrefs. Don't waste time searching too hard,
3563 * this is just an optimization
3568 /* we hit the end of the leaf before we found an xattr or
3569 * something larger than an xattr. We have to assume the inode
3572 if (*first_xattr_slot
== -1)
3573 *first_xattr_slot
= slot
;
3578 * read an inode from the btree into the in-memory inode
3580 static void btrfs_read_locked_inode(struct inode
*inode
)
3582 struct btrfs_path
*path
;
3583 struct extent_buffer
*leaf
;
3584 struct btrfs_inode_item
*inode_item
;
3585 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3586 struct btrfs_key location
;
3591 bool filled
= false;
3592 int first_xattr_slot
;
3594 ret
= btrfs_fill_inode(inode
, &rdev
);
3598 path
= btrfs_alloc_path();
3602 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3604 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3608 leaf
= path
->nodes
[0];
3613 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3614 struct btrfs_inode_item
);
3615 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3616 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3617 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3618 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3619 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3621 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->atime
);
3622 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->atime
);
3624 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->mtime
);
3625 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->mtime
);
3627 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, &inode_item
->ctime
);
3628 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, &inode_item
->ctime
);
3630 BTRFS_I(inode
)->i_otime
.tv_sec
=
3631 btrfs_timespec_sec(leaf
, &inode_item
->otime
);
3632 BTRFS_I(inode
)->i_otime
.tv_nsec
=
3633 btrfs_timespec_nsec(leaf
, &inode_item
->otime
);
3635 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3636 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3637 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3639 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3640 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3642 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3644 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3645 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3649 * If we were modified in the current generation and evicted from memory
3650 * and then re-read we need to do a full sync since we don't have any
3651 * idea about which extents were modified before we were evicted from
3654 * This is required for both inode re-read from disk and delayed inode
3655 * in delayed_nodes_tree.
3657 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3658 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3659 &BTRFS_I(inode
)->runtime_flags
);
3662 * We don't persist the id of the transaction where an unlink operation
3663 * against the inode was last made. So here we assume the inode might
3664 * have been evicted, and therefore the exact value of last_unlink_trans
3665 * lost, and set it to last_trans to avoid metadata inconsistencies
3666 * between the inode and its parent if the inode is fsync'ed and the log
3667 * replayed. For example, in the scenario:
3670 * ln mydir/foo mydir/bar
3673 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3674 * xfs_io -c fsync mydir/foo
3676 * mount fs, triggers fsync log replay
3678 * We must make sure that when we fsync our inode foo we also log its
3679 * parent inode, otherwise after log replay the parent still has the
3680 * dentry with the "bar" name but our inode foo has a link count of 1
3681 * and doesn't have an inode ref with the name "bar" anymore.
3683 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3684 * but it guarantees correctness at the expense of ocassional full
3685 * transaction commits on fsync if our inode is a directory, or if our
3686 * inode is not a directory, logging its parent unnecessarily.
3688 BTRFS_I(inode
)->last_unlink_trans
= BTRFS_I(inode
)->last_trans
;
3691 if (inode
->i_nlink
!= 1 ||
3692 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3695 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3696 if (location
.objectid
!= btrfs_ino(inode
))
3699 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3700 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3701 struct btrfs_inode_ref
*ref
;
3703 ref
= (struct btrfs_inode_ref
*)ptr
;
3704 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3705 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3706 struct btrfs_inode_extref
*extref
;
3708 extref
= (struct btrfs_inode_extref
*)ptr
;
3709 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3714 * try to precache a NULL acl entry for files that don't have
3715 * any xattrs or acls
3717 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3718 btrfs_ino(inode
), &first_xattr_slot
);
3719 if (first_xattr_slot
!= -1) {
3720 path
->slots
[0] = first_xattr_slot
;
3721 ret
= btrfs_load_inode_props(inode
, path
);
3723 btrfs_err(root
->fs_info
,
3724 "error loading props for ino %llu (root %llu): %d",
3726 root
->root_key
.objectid
, ret
);
3728 btrfs_free_path(path
);
3731 cache_no_acl(inode
);
3733 switch (inode
->i_mode
& S_IFMT
) {
3735 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3736 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3737 inode
->i_fop
= &btrfs_file_operations
;
3738 inode
->i_op
= &btrfs_file_inode_operations
;
3741 inode
->i_fop
= &btrfs_dir_file_operations
;
3742 if (root
== root
->fs_info
->tree_root
)
3743 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3745 inode
->i_op
= &btrfs_dir_inode_operations
;
3748 inode
->i_op
= &btrfs_symlink_inode_operations
;
3749 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3752 inode
->i_op
= &btrfs_special_inode_operations
;
3753 init_special_inode(inode
, inode
->i_mode
, rdev
);
3757 btrfs_update_iflags(inode
);
3761 btrfs_free_path(path
);
3762 make_bad_inode(inode
);
3766 * given a leaf and an inode, copy the inode fields into the leaf
3768 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3769 struct extent_buffer
*leaf
,
3770 struct btrfs_inode_item
*item
,
3771 struct inode
*inode
)
3773 struct btrfs_map_token token
;
3775 btrfs_init_map_token(&token
);
3777 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3778 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3779 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3781 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3782 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3784 btrfs_set_token_timespec_sec(leaf
, &item
->atime
,
3785 inode
->i_atime
.tv_sec
, &token
);
3786 btrfs_set_token_timespec_nsec(leaf
, &item
->atime
,
3787 inode
->i_atime
.tv_nsec
, &token
);
3789 btrfs_set_token_timespec_sec(leaf
, &item
->mtime
,
3790 inode
->i_mtime
.tv_sec
, &token
);
3791 btrfs_set_token_timespec_nsec(leaf
, &item
->mtime
,
3792 inode
->i_mtime
.tv_nsec
, &token
);
3794 btrfs_set_token_timespec_sec(leaf
, &item
->ctime
,
3795 inode
->i_ctime
.tv_sec
, &token
);
3796 btrfs_set_token_timespec_nsec(leaf
, &item
->ctime
,
3797 inode
->i_ctime
.tv_nsec
, &token
);
3799 btrfs_set_token_timespec_sec(leaf
, &item
->otime
,
3800 BTRFS_I(inode
)->i_otime
.tv_sec
, &token
);
3801 btrfs_set_token_timespec_nsec(leaf
, &item
->otime
,
3802 BTRFS_I(inode
)->i_otime
.tv_nsec
, &token
);
3804 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3806 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3808 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3809 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3810 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3811 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3812 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3816 * copy everything in the in-memory inode into the btree.
3818 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3819 struct btrfs_root
*root
, struct inode
*inode
)
3821 struct btrfs_inode_item
*inode_item
;
3822 struct btrfs_path
*path
;
3823 struct extent_buffer
*leaf
;
3826 path
= btrfs_alloc_path();
3830 path
->leave_spinning
= 1;
3831 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3839 leaf
= path
->nodes
[0];
3840 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3841 struct btrfs_inode_item
);
3843 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3844 btrfs_mark_buffer_dirty(leaf
);
3845 btrfs_set_inode_last_trans(trans
, inode
);
3848 btrfs_free_path(path
);
3853 * copy everything in the in-memory inode into the btree.
3855 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3856 struct btrfs_root
*root
, struct inode
*inode
)
3861 * If the inode is a free space inode, we can deadlock during commit
3862 * if we put it into the delayed code.
3864 * The data relocation inode should also be directly updated
3867 if (!btrfs_is_free_space_inode(inode
)
3868 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
3869 && !root
->fs_info
->log_root_recovering
) {
3870 btrfs_update_root_times(trans
, root
);
3872 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3874 btrfs_set_inode_last_trans(trans
, inode
);
3878 return btrfs_update_inode_item(trans
, root
, inode
);
3881 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3882 struct btrfs_root
*root
,
3883 struct inode
*inode
)
3887 ret
= btrfs_update_inode(trans
, root
, inode
);
3889 return btrfs_update_inode_item(trans
, root
, inode
);
3894 * unlink helper that gets used here in inode.c and in the tree logging
3895 * recovery code. It remove a link in a directory with a given name, and
3896 * also drops the back refs in the inode to the directory
3898 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3899 struct btrfs_root
*root
,
3900 struct inode
*dir
, struct inode
*inode
,
3901 const char *name
, int name_len
)
3903 struct btrfs_path
*path
;
3905 struct extent_buffer
*leaf
;
3906 struct btrfs_dir_item
*di
;
3907 struct btrfs_key key
;
3909 u64 ino
= btrfs_ino(inode
);
3910 u64 dir_ino
= btrfs_ino(dir
);
3912 path
= btrfs_alloc_path();
3918 path
->leave_spinning
= 1;
3919 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3920 name
, name_len
, -1);
3929 leaf
= path
->nodes
[0];
3930 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3931 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3934 btrfs_release_path(path
);
3937 * If we don't have dir index, we have to get it by looking up
3938 * the inode ref, since we get the inode ref, remove it directly,
3939 * it is unnecessary to do delayed deletion.
3941 * But if we have dir index, needn't search inode ref to get it.
3942 * Since the inode ref is close to the inode item, it is better
3943 * that we delay to delete it, and just do this deletion when
3944 * we update the inode item.
3946 if (BTRFS_I(inode
)->dir_index
) {
3947 ret
= btrfs_delayed_delete_inode_ref(inode
);
3949 index
= BTRFS_I(inode
)->dir_index
;
3954 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3957 btrfs_info(root
->fs_info
,
3958 "failed to delete reference to %.*s, inode %llu parent %llu",
3959 name_len
, name
, ino
, dir_ino
);
3960 btrfs_abort_transaction(trans
, root
, ret
);
3964 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3966 btrfs_abort_transaction(trans
, root
, ret
);
3970 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
3972 if (ret
!= 0 && ret
!= -ENOENT
) {
3973 btrfs_abort_transaction(trans
, root
, ret
);
3977 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
3982 btrfs_abort_transaction(trans
, root
, ret
);
3984 btrfs_free_path(path
);
3988 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3989 inode_inc_iversion(inode
);
3990 inode_inc_iversion(dir
);
3991 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3992 ret
= btrfs_update_inode(trans
, root
, dir
);
3997 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3998 struct btrfs_root
*root
,
3999 struct inode
*dir
, struct inode
*inode
,
4000 const char *name
, int name_len
)
4003 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
4006 ret
= btrfs_update_inode(trans
, root
, inode
);
4012 * helper to start transaction for unlink and rmdir.
4014 * unlink and rmdir are special in btrfs, they do not always free space, so
4015 * if we cannot make our reservations the normal way try and see if there is
4016 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4017 * allow the unlink to occur.
4019 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
4021 struct btrfs_trans_handle
*trans
;
4022 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4026 * 1 for the possible orphan item
4027 * 1 for the dir item
4028 * 1 for the dir index
4029 * 1 for the inode ref
4032 trans
= btrfs_start_transaction(root
, 5);
4033 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
4036 if (PTR_ERR(trans
) == -ENOSPC
) {
4037 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
4039 trans
= btrfs_start_transaction(root
, 0);
4042 ret
= btrfs_cond_migrate_bytes(root
->fs_info
,
4043 &root
->fs_info
->trans_block_rsv
,
4046 btrfs_end_transaction(trans
, root
);
4047 return ERR_PTR(ret
);
4049 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
4050 trans
->bytes_reserved
= num_bytes
;
4055 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
4057 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4058 struct btrfs_trans_handle
*trans
;
4059 struct inode
*inode
= d_inode(dentry
);
4062 trans
= __unlink_start_trans(dir
);
4064 return PTR_ERR(trans
);
4066 btrfs_record_unlink_dir(trans
, dir
, d_inode(dentry
), 0);
4068 ret
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4069 dentry
->d_name
.name
, dentry
->d_name
.len
);
4073 if (inode
->i_nlink
== 0) {
4074 ret
= btrfs_orphan_add(trans
, inode
);
4080 btrfs_end_transaction(trans
, root
);
4081 btrfs_btree_balance_dirty(root
);
4085 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
4086 struct btrfs_root
*root
,
4087 struct inode
*dir
, u64 objectid
,
4088 const char *name
, int name_len
)
4090 struct btrfs_path
*path
;
4091 struct extent_buffer
*leaf
;
4092 struct btrfs_dir_item
*di
;
4093 struct btrfs_key key
;
4096 u64 dir_ino
= btrfs_ino(dir
);
4098 path
= btrfs_alloc_path();
4102 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
4103 name
, name_len
, -1);
4104 if (IS_ERR_OR_NULL(di
)) {
4112 leaf
= path
->nodes
[0];
4113 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
4114 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
4115 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
4117 btrfs_abort_transaction(trans
, root
, ret
);
4120 btrfs_release_path(path
);
4122 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4123 objectid
, root
->root_key
.objectid
,
4124 dir_ino
, &index
, name
, name_len
);
4126 if (ret
!= -ENOENT
) {
4127 btrfs_abort_transaction(trans
, root
, ret
);
4130 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
4132 if (IS_ERR_OR_NULL(di
)) {
4137 btrfs_abort_transaction(trans
, root
, ret
);
4141 leaf
= path
->nodes
[0];
4142 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4143 btrfs_release_path(path
);
4146 btrfs_release_path(path
);
4148 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
4150 btrfs_abort_transaction(trans
, root
, ret
);
4154 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
4155 inode_inc_iversion(dir
);
4156 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
4157 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
4159 btrfs_abort_transaction(trans
, root
, ret
);
4161 btrfs_free_path(path
);
4165 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
4167 struct inode
*inode
= d_inode(dentry
);
4169 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4170 struct btrfs_trans_handle
*trans
;
4172 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
4174 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
4177 trans
= __unlink_start_trans(dir
);
4179 return PTR_ERR(trans
);
4181 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4182 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4183 BTRFS_I(inode
)->location
.objectid
,
4184 dentry
->d_name
.name
,
4185 dentry
->d_name
.len
);
4189 err
= btrfs_orphan_add(trans
, inode
);
4193 /* now the directory is empty */
4194 err
= btrfs_unlink_inode(trans
, root
, dir
, d_inode(dentry
),
4195 dentry
->d_name
.name
, dentry
->d_name
.len
);
4197 btrfs_i_size_write(inode
, 0);
4199 btrfs_end_transaction(trans
, root
);
4200 btrfs_btree_balance_dirty(root
);
4205 static int truncate_space_check(struct btrfs_trans_handle
*trans
,
4206 struct btrfs_root
*root
,
4211 bytes_deleted
= btrfs_csum_bytes_to_leaves(root
, bytes_deleted
);
4212 ret
= btrfs_block_rsv_add(root
, &root
->fs_info
->trans_block_rsv
,
4213 bytes_deleted
, BTRFS_RESERVE_NO_FLUSH
);
4215 trans
->bytes_reserved
+= bytes_deleted
;
4221 * this can truncate away extent items, csum items and directory items.
4222 * It starts at a high offset and removes keys until it can't find
4223 * any higher than new_size
4225 * csum items that cross the new i_size are truncated to the new size
4228 * min_type is the minimum key type to truncate down to. If set to 0, this
4229 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4231 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4232 struct btrfs_root
*root
,
4233 struct inode
*inode
,
4234 u64 new_size
, u32 min_type
)
4236 struct btrfs_path
*path
;
4237 struct extent_buffer
*leaf
;
4238 struct btrfs_file_extent_item
*fi
;
4239 struct btrfs_key key
;
4240 struct btrfs_key found_key
;
4241 u64 extent_start
= 0;
4242 u64 extent_num_bytes
= 0;
4243 u64 extent_offset
= 0;
4245 u64 last_size
= new_size
;
4246 u32 found_type
= (u8
)-1;
4249 int pending_del_nr
= 0;
4250 int pending_del_slot
= 0;
4251 int extent_type
= -1;
4254 u64 ino
= btrfs_ino(inode
);
4255 u64 bytes_deleted
= 0;
4257 bool should_throttle
= 0;
4258 bool should_end
= 0;
4260 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4263 * for non-free space inodes and ref cows, we want to back off from
4266 if (!btrfs_is_free_space_inode(inode
) &&
4267 test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4270 path
= btrfs_alloc_path();
4276 * We want to drop from the next block forward in case this new size is
4277 * not block aligned since we will be keeping the last block of the
4278 * extent just the way it is.
4280 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4281 root
== root
->fs_info
->tree_root
)
4282 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
4283 root
->sectorsize
), (u64
)-1, 0);
4286 * This function is also used to drop the items in the log tree before
4287 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4288 * it is used to drop the loged items. So we shouldn't kill the delayed
4291 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4292 btrfs_kill_delayed_inode_items(inode
);
4295 key
.offset
= (u64
)-1;
4300 * with a 16K leaf size and 128MB extents, you can actually queue
4301 * up a huge file in a single leaf. Most of the time that
4302 * bytes_deleted is > 0, it will be huge by the time we get here
4304 if (be_nice
&& bytes_deleted
> 32 * 1024 * 1024) {
4305 if (btrfs_should_end_transaction(trans
, root
)) {
4312 path
->leave_spinning
= 1;
4313 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4320 /* there are no items in the tree for us to truncate, we're
4323 if (path
->slots
[0] == 0)
4330 leaf
= path
->nodes
[0];
4331 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4332 found_type
= found_key
.type
;
4334 if (found_key
.objectid
!= ino
)
4337 if (found_type
< min_type
)
4340 item_end
= found_key
.offset
;
4341 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4342 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4343 struct btrfs_file_extent_item
);
4344 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4345 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4347 btrfs_file_extent_num_bytes(leaf
, fi
);
4348 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4349 item_end
+= btrfs_file_extent_inline_len(leaf
,
4350 path
->slots
[0], fi
);
4354 if (found_type
> min_type
) {
4357 if (item_end
< new_size
)
4359 if (found_key
.offset
>= new_size
)
4365 /* FIXME, shrink the extent if the ref count is only 1 */
4366 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4370 last_size
= found_key
.offset
;
4372 last_size
= new_size
;
4374 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4376 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4378 u64 orig_num_bytes
=
4379 btrfs_file_extent_num_bytes(leaf
, fi
);
4380 extent_num_bytes
= ALIGN(new_size
-
4383 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4385 num_dec
= (orig_num_bytes
-
4387 if (test_bit(BTRFS_ROOT_REF_COWS
,
4390 inode_sub_bytes(inode
, num_dec
);
4391 btrfs_mark_buffer_dirty(leaf
);
4394 btrfs_file_extent_disk_num_bytes(leaf
,
4396 extent_offset
= found_key
.offset
-
4397 btrfs_file_extent_offset(leaf
, fi
);
4399 /* FIXME blocksize != 4096 */
4400 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4401 if (extent_start
!= 0) {
4403 if (test_bit(BTRFS_ROOT_REF_COWS
,
4405 inode_sub_bytes(inode
, num_dec
);
4408 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4410 * we can't truncate inline items that have had
4414 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
4415 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4416 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4417 u32 size
= new_size
- found_key
.offset
;
4419 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
))
4420 inode_sub_bytes(inode
, item_end
+ 1 -
4424 * update the ram bytes to properly reflect
4425 * the new size of our item
4427 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4429 btrfs_file_extent_calc_inline_size(size
);
4430 btrfs_truncate_item(root
, path
, size
, 1);
4431 } else if (test_bit(BTRFS_ROOT_REF_COWS
,
4433 inode_sub_bytes(inode
, item_end
+ 1 -
4439 if (!pending_del_nr
) {
4440 /* no pending yet, add ourselves */
4441 pending_del_slot
= path
->slots
[0];
4443 } else if (pending_del_nr
&&
4444 path
->slots
[0] + 1 == pending_del_slot
) {
4445 /* hop on the pending chunk */
4447 pending_del_slot
= path
->slots
[0];
4454 should_throttle
= 0;
4457 (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) ||
4458 root
== root
->fs_info
->tree_root
)) {
4459 btrfs_set_path_blocking(path
);
4460 bytes_deleted
+= extent_num_bytes
;
4461 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4462 extent_num_bytes
, 0,
4463 btrfs_header_owner(leaf
),
4464 ino
, extent_offset
, 0);
4466 if (btrfs_should_throttle_delayed_refs(trans
, root
))
4467 btrfs_async_run_delayed_refs(root
,
4468 trans
->delayed_ref_updates
* 2, 0);
4470 if (truncate_space_check(trans
, root
,
4471 extent_num_bytes
)) {
4474 if (btrfs_should_throttle_delayed_refs(trans
,
4476 should_throttle
= 1;
4481 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4484 if (path
->slots
[0] == 0 ||
4485 path
->slots
[0] != pending_del_slot
||
4486 should_throttle
|| should_end
) {
4487 if (pending_del_nr
) {
4488 ret
= btrfs_del_items(trans
, root
, path
,
4492 btrfs_abort_transaction(trans
,
4498 btrfs_release_path(path
);
4499 if (should_throttle
) {
4500 unsigned long updates
= trans
->delayed_ref_updates
;
4502 trans
->delayed_ref_updates
= 0;
4503 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4509 * if we failed to refill our space rsv, bail out
4510 * and let the transaction restart
4522 if (pending_del_nr
) {
4523 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4526 btrfs_abort_transaction(trans
, root
, ret
);
4529 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
)
4530 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4532 btrfs_free_path(path
);
4534 if (be_nice
&& bytes_deleted
> 32 * 1024 * 1024) {
4535 unsigned long updates
= trans
->delayed_ref_updates
;
4537 trans
->delayed_ref_updates
= 0;
4538 ret
= btrfs_run_delayed_refs(trans
, root
, updates
* 2);
4547 * btrfs_truncate_page - read, zero a chunk and write a page
4548 * @inode - inode that we're zeroing
4549 * @from - the offset to start zeroing
4550 * @len - the length to zero, 0 to zero the entire range respective to the
4552 * @front - zero up to the offset instead of from the offset on
4554 * This will find the page for the "from" offset and cow the page and zero the
4555 * part we want to zero. This is used with truncate and hole punching.
4557 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
4560 struct address_space
*mapping
= inode
->i_mapping
;
4561 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4562 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4563 struct btrfs_ordered_extent
*ordered
;
4564 struct extent_state
*cached_state
= NULL
;
4566 u32 blocksize
= root
->sectorsize
;
4567 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4568 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4570 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4575 if ((offset
& (blocksize
- 1)) == 0 &&
4576 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4578 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
4583 page
= find_or_create_page(mapping
, index
, mask
);
4585 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4590 page_start
= page_offset(page
);
4591 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4593 if (!PageUptodate(page
)) {
4594 ret
= btrfs_readpage(NULL
, page
);
4596 if (page
->mapping
!= mapping
) {
4598 page_cache_release(page
);
4601 if (!PageUptodate(page
)) {
4606 wait_on_page_writeback(page
);
4608 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
4609 set_page_extent_mapped(page
);
4611 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4613 unlock_extent_cached(io_tree
, page_start
, page_end
,
4614 &cached_state
, GFP_NOFS
);
4616 page_cache_release(page
);
4617 btrfs_start_ordered_extent(inode
, ordered
, 1);
4618 btrfs_put_ordered_extent(ordered
);
4622 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
4623 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4624 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4625 0, 0, &cached_state
, GFP_NOFS
);
4627 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
4630 unlock_extent_cached(io_tree
, page_start
, page_end
,
4631 &cached_state
, GFP_NOFS
);
4635 if (offset
!= PAGE_CACHE_SIZE
) {
4637 len
= PAGE_CACHE_SIZE
- offset
;
4640 memset(kaddr
, 0, offset
);
4642 memset(kaddr
+ offset
, 0, len
);
4643 flush_dcache_page(page
);
4646 ClearPageChecked(page
);
4647 set_page_dirty(page
);
4648 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
4653 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4655 page_cache_release(page
);
4660 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4661 u64 offset
, u64 len
)
4663 struct btrfs_trans_handle
*trans
;
4667 * Still need to make sure the inode looks like it's been updated so
4668 * that any holes get logged if we fsync.
4670 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
)) {
4671 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
4672 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4673 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4678 * 1 - for the one we're dropping
4679 * 1 - for the one we're adding
4680 * 1 - for updating the inode.
4682 trans
= btrfs_start_transaction(root
, 3);
4684 return PTR_ERR(trans
);
4686 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4688 btrfs_abort_transaction(trans
, root
, ret
);
4689 btrfs_end_transaction(trans
, root
);
4693 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
4694 0, 0, len
, 0, len
, 0, 0, 0);
4696 btrfs_abort_transaction(trans
, root
, ret
);
4698 btrfs_update_inode(trans
, root
, inode
);
4699 btrfs_end_transaction(trans
, root
);
4704 * This function puts in dummy file extents for the area we're creating a hole
4705 * for. So if we are truncating this file to a larger size we need to insert
4706 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4707 * the range between oldsize and size
4709 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4711 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4712 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4713 struct extent_map
*em
= NULL
;
4714 struct extent_state
*cached_state
= NULL
;
4715 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4716 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4717 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4724 * If our size started in the middle of a page we need to zero out the
4725 * rest of the page before we expand the i_size, otherwise we could
4726 * expose stale data.
4728 err
= btrfs_truncate_page(inode
, oldsize
, 0, 0);
4732 if (size
<= hole_start
)
4736 struct btrfs_ordered_extent
*ordered
;
4738 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
4740 ordered
= btrfs_lookup_ordered_range(inode
, hole_start
,
4741 block_end
- hole_start
);
4744 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4745 &cached_state
, GFP_NOFS
);
4746 btrfs_start_ordered_extent(inode
, ordered
, 1);
4747 btrfs_put_ordered_extent(ordered
);
4750 cur_offset
= hole_start
;
4752 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4753 block_end
- cur_offset
, 0);
4759 last_byte
= min(extent_map_end(em
), block_end
);
4760 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4761 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4762 struct extent_map
*hole_em
;
4763 hole_size
= last_byte
- cur_offset
;
4765 err
= maybe_insert_hole(root
, inode
, cur_offset
,
4769 btrfs_drop_extent_cache(inode
, cur_offset
,
4770 cur_offset
+ hole_size
- 1, 0);
4771 hole_em
= alloc_extent_map();
4773 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4774 &BTRFS_I(inode
)->runtime_flags
);
4777 hole_em
->start
= cur_offset
;
4778 hole_em
->len
= hole_size
;
4779 hole_em
->orig_start
= cur_offset
;
4781 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4782 hole_em
->block_len
= 0;
4783 hole_em
->orig_block_len
= 0;
4784 hole_em
->ram_bytes
= hole_size
;
4785 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4786 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4787 hole_em
->generation
= root
->fs_info
->generation
;
4790 write_lock(&em_tree
->lock
);
4791 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4792 write_unlock(&em_tree
->lock
);
4795 btrfs_drop_extent_cache(inode
, cur_offset
,
4799 free_extent_map(hole_em
);
4802 free_extent_map(em
);
4804 cur_offset
= last_byte
;
4805 if (cur_offset
>= block_end
)
4808 free_extent_map(em
);
4809 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4814 static int wait_snapshoting_atomic_t(atomic_t
*a
)
4820 static void wait_for_snapshot_creation(struct btrfs_root
*root
)
4825 ret
= btrfs_start_write_no_snapshoting(root
);
4828 wait_on_atomic_t(&root
->will_be_snapshoted
,
4829 wait_snapshoting_atomic_t
,
4830 TASK_UNINTERRUPTIBLE
);
4834 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4836 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4837 struct btrfs_trans_handle
*trans
;
4838 loff_t oldsize
= i_size_read(inode
);
4839 loff_t newsize
= attr
->ia_size
;
4840 int mask
= attr
->ia_valid
;
4844 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4845 * special case where we need to update the times despite not having
4846 * these flags set. For all other operations the VFS set these flags
4847 * explicitly if it wants a timestamp update.
4849 if (newsize
!= oldsize
) {
4850 inode_inc_iversion(inode
);
4851 if (!(mask
& (ATTR_CTIME
| ATTR_MTIME
)))
4852 inode
->i_ctime
= inode
->i_mtime
=
4853 current_fs_time(inode
->i_sb
);
4856 if (newsize
> oldsize
) {
4857 truncate_pagecache(inode
, newsize
);
4859 * Don't do an expanding truncate while snapshoting is ongoing.
4860 * This is to ensure the snapshot captures a fully consistent
4861 * state of this file - if the snapshot captures this expanding
4862 * truncation, it must capture all writes that happened before
4865 wait_for_snapshot_creation(root
);
4866 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4868 btrfs_end_write_no_snapshoting(root
);
4872 trans
= btrfs_start_transaction(root
, 1);
4873 if (IS_ERR(trans
)) {
4874 btrfs_end_write_no_snapshoting(root
);
4875 return PTR_ERR(trans
);
4878 i_size_write(inode
, newsize
);
4879 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4880 ret
= btrfs_update_inode(trans
, root
, inode
);
4881 btrfs_end_write_no_snapshoting(root
);
4882 btrfs_end_transaction(trans
, root
);
4886 * We're truncating a file that used to have good data down to
4887 * zero. Make sure it gets into the ordered flush list so that
4888 * any new writes get down to disk quickly.
4891 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4892 &BTRFS_I(inode
)->runtime_flags
);
4895 * 1 for the orphan item we're going to add
4896 * 1 for the orphan item deletion.
4898 trans
= btrfs_start_transaction(root
, 2);
4900 return PTR_ERR(trans
);
4903 * We need to do this in case we fail at _any_ point during the
4904 * actual truncate. Once we do the truncate_setsize we could
4905 * invalidate pages which forces any outstanding ordered io to
4906 * be instantly completed which will give us extents that need
4907 * to be truncated. If we fail to get an orphan inode down we
4908 * could have left over extents that were never meant to live,
4909 * so we need to garuntee from this point on that everything
4910 * will be consistent.
4912 ret
= btrfs_orphan_add(trans
, inode
);
4913 btrfs_end_transaction(trans
, root
);
4917 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4918 truncate_setsize(inode
, newsize
);
4920 /* Disable nonlocked read DIO to avoid the end less truncate */
4921 btrfs_inode_block_unlocked_dio(inode
);
4922 inode_dio_wait(inode
);
4923 btrfs_inode_resume_unlocked_dio(inode
);
4925 ret
= btrfs_truncate(inode
);
4926 if (ret
&& inode
->i_nlink
) {
4930 * failed to truncate, disk_i_size is only adjusted down
4931 * as we remove extents, so it should represent the true
4932 * size of the inode, so reset the in memory size and
4933 * delete our orphan entry.
4935 trans
= btrfs_join_transaction(root
);
4936 if (IS_ERR(trans
)) {
4937 btrfs_orphan_del(NULL
, inode
);
4940 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
4941 err
= btrfs_orphan_del(trans
, inode
);
4943 btrfs_abort_transaction(trans
, root
, err
);
4944 btrfs_end_transaction(trans
, root
);
4951 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4953 struct inode
*inode
= d_inode(dentry
);
4954 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4957 if (btrfs_root_readonly(root
))
4960 err
= inode_change_ok(inode
, attr
);
4964 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
4965 err
= btrfs_setsize(inode
, attr
);
4970 if (attr
->ia_valid
) {
4971 setattr_copy(inode
, attr
);
4972 inode_inc_iversion(inode
);
4973 err
= btrfs_dirty_inode(inode
);
4975 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
4976 err
= posix_acl_chmod(inode
, inode
->i_mode
);
4983 * While truncating the inode pages during eviction, we get the VFS calling
4984 * btrfs_invalidatepage() against each page of the inode. This is slow because
4985 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4986 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4987 * extent_state structures over and over, wasting lots of time.
4989 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4990 * those expensive operations on a per page basis and do only the ordered io
4991 * finishing, while we release here the extent_map and extent_state structures,
4992 * without the excessive merging and splitting.
4994 static void evict_inode_truncate_pages(struct inode
*inode
)
4996 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4997 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
4998 struct rb_node
*node
;
5000 ASSERT(inode
->i_state
& I_FREEING
);
5001 truncate_inode_pages_final(&inode
->i_data
);
5003 write_lock(&map_tree
->lock
);
5004 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
5005 struct extent_map
*em
;
5007 node
= rb_first(&map_tree
->map
);
5008 em
= rb_entry(node
, struct extent_map
, rb_node
);
5009 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5010 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
5011 remove_extent_mapping(map_tree
, em
);
5012 free_extent_map(em
);
5013 if (need_resched()) {
5014 write_unlock(&map_tree
->lock
);
5016 write_lock(&map_tree
->lock
);
5019 write_unlock(&map_tree
->lock
);
5022 * Keep looping until we have no more ranges in the io tree.
5023 * We can have ongoing bios started by readpages (called from readahead)
5024 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5025 * still in progress (unlocked the pages in the bio but did not yet
5026 * unlocked the ranges in the io tree). Therefore this means some
5027 * ranges can still be locked and eviction started because before
5028 * submitting those bios, which are executed by a separate task (work
5029 * queue kthread), inode references (inode->i_count) were not taken
5030 * (which would be dropped in the end io callback of each bio).
5031 * Therefore here we effectively end up waiting for those bios and
5032 * anyone else holding locked ranges without having bumped the inode's
5033 * reference count - if we don't do it, when they access the inode's
5034 * io_tree to unlock a range it may be too late, leading to an
5035 * use-after-free issue.
5037 spin_lock(&io_tree
->lock
);
5038 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
5039 struct extent_state
*state
;
5040 struct extent_state
*cached_state
= NULL
;
5044 node
= rb_first(&io_tree
->state
);
5045 state
= rb_entry(node
, struct extent_state
, rb_node
);
5046 start
= state
->start
;
5048 spin_unlock(&io_tree
->lock
);
5050 lock_extent_bits(io_tree
, start
, end
, 0, &cached_state
);
5051 clear_extent_bit(io_tree
, start
, end
,
5052 EXTENT_LOCKED
| EXTENT_DIRTY
|
5053 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
5054 EXTENT_DEFRAG
, 1, 1,
5055 &cached_state
, GFP_NOFS
);
5058 spin_lock(&io_tree
->lock
);
5060 spin_unlock(&io_tree
->lock
);
5063 void btrfs_evict_inode(struct inode
*inode
)
5065 struct btrfs_trans_handle
*trans
;
5066 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5067 struct btrfs_block_rsv
*rsv
, *global_rsv
;
5068 int steal_from_global
= 0;
5069 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
5072 trace_btrfs_inode_evict(inode
);
5074 evict_inode_truncate_pages(inode
);
5076 if (inode
->i_nlink
&&
5077 ((btrfs_root_refs(&root
->root_item
) != 0 &&
5078 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
5079 btrfs_is_free_space_inode(inode
)))
5082 if (is_bad_inode(inode
)) {
5083 btrfs_orphan_del(NULL
, inode
);
5086 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5087 if (!special_file(inode
->i_mode
))
5088 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
5090 btrfs_free_io_failure_record(inode
, 0, (u64
)-1);
5092 if (root
->fs_info
->log_root_recovering
) {
5093 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
5094 &BTRFS_I(inode
)->runtime_flags
));
5098 if (inode
->i_nlink
> 0) {
5099 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
5100 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
5104 ret
= btrfs_commit_inode_delayed_inode(inode
);
5106 btrfs_orphan_del(NULL
, inode
);
5110 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
5112 btrfs_orphan_del(NULL
, inode
);
5115 rsv
->size
= min_size
;
5117 global_rsv
= &root
->fs_info
->global_block_rsv
;
5119 btrfs_i_size_write(inode
, 0);
5122 * This is a bit simpler than btrfs_truncate since we've already
5123 * reserved our space for our orphan item in the unlink, so we just
5124 * need to reserve some slack space in case we add bytes and update
5125 * inode item when doing the truncate.
5128 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
5129 BTRFS_RESERVE_FLUSH_LIMIT
);
5132 * Try and steal from the global reserve since we will
5133 * likely not use this space anyway, we want to try as
5134 * hard as possible to get this to work.
5137 steal_from_global
++;
5139 steal_from_global
= 0;
5143 * steal_from_global == 0: we reserved stuff, hooray!
5144 * steal_from_global == 1: we didn't reserve stuff, boo!
5145 * steal_from_global == 2: we've committed, still not a lot of
5146 * room but maybe we'll have room in the global reserve this
5148 * steal_from_global == 3: abandon all hope!
5150 if (steal_from_global
> 2) {
5151 btrfs_warn(root
->fs_info
,
5152 "Could not get space for a delete, will truncate on mount %d",
5154 btrfs_orphan_del(NULL
, inode
);
5155 btrfs_free_block_rsv(root
, rsv
);
5159 trans
= btrfs_join_transaction(root
);
5160 if (IS_ERR(trans
)) {
5161 btrfs_orphan_del(NULL
, inode
);
5162 btrfs_free_block_rsv(root
, rsv
);
5167 * We can't just steal from the global reserve, we need tomake
5168 * sure there is room to do it, if not we need to commit and try
5171 if (steal_from_global
) {
5172 if (!btrfs_check_space_for_delayed_refs(trans
, root
))
5173 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
,
5180 * Couldn't steal from the global reserve, we have too much
5181 * pending stuff built up, commit the transaction and try it
5185 ret
= btrfs_commit_transaction(trans
, root
);
5187 btrfs_orphan_del(NULL
, inode
);
5188 btrfs_free_block_rsv(root
, rsv
);
5193 steal_from_global
= 0;
5196 trans
->block_rsv
= rsv
;
5198 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
5199 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
)
5202 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5203 btrfs_end_transaction(trans
, root
);
5205 btrfs_btree_balance_dirty(root
);
5208 btrfs_free_block_rsv(root
, rsv
);
5211 * Errors here aren't a big deal, it just means we leave orphan items
5212 * in the tree. They will be cleaned up on the next mount.
5215 trans
->block_rsv
= root
->orphan_block_rsv
;
5216 btrfs_orphan_del(trans
, inode
);
5218 btrfs_orphan_del(NULL
, inode
);
5221 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
5222 if (!(root
== root
->fs_info
->tree_root
||
5223 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
5224 btrfs_return_ino(root
, btrfs_ino(inode
));
5226 btrfs_end_transaction(trans
, root
);
5227 btrfs_btree_balance_dirty(root
);
5229 btrfs_remove_delayed_node(inode
);
5235 * this returns the key found in the dir entry in the location pointer.
5236 * If no dir entries were found, location->objectid is 0.
5238 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
5239 struct btrfs_key
*location
)
5241 const char *name
= dentry
->d_name
.name
;
5242 int namelen
= dentry
->d_name
.len
;
5243 struct btrfs_dir_item
*di
;
5244 struct btrfs_path
*path
;
5245 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5248 path
= btrfs_alloc_path();
5252 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
5257 if (IS_ERR_OR_NULL(di
))
5260 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
5262 btrfs_free_path(path
);
5265 location
->objectid
= 0;
5270 * when we hit a tree root in a directory, the btrfs part of the inode
5271 * needs to be changed to reflect the root directory of the tree root. This
5272 * is kind of like crossing a mount point.
5274 static int fixup_tree_root_location(struct btrfs_root
*root
,
5276 struct dentry
*dentry
,
5277 struct btrfs_key
*location
,
5278 struct btrfs_root
**sub_root
)
5280 struct btrfs_path
*path
;
5281 struct btrfs_root
*new_root
;
5282 struct btrfs_root_ref
*ref
;
5283 struct extent_buffer
*leaf
;
5284 struct btrfs_key key
;
5288 path
= btrfs_alloc_path();
5295 key
.objectid
= BTRFS_I(dir
)->root
->root_key
.objectid
;
5296 key
.type
= BTRFS_ROOT_REF_KEY
;
5297 key
.offset
= location
->objectid
;
5299 ret
= btrfs_search_slot(NULL
, root
->fs_info
->tree_root
, &key
, path
,
5307 leaf
= path
->nodes
[0];
5308 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
5309 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
5310 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
5313 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
5314 (unsigned long)(ref
+ 1),
5315 dentry
->d_name
.len
);
5319 btrfs_release_path(path
);
5321 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
5322 if (IS_ERR(new_root
)) {
5323 err
= PTR_ERR(new_root
);
5327 *sub_root
= new_root
;
5328 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
5329 location
->type
= BTRFS_INODE_ITEM_KEY
;
5330 location
->offset
= 0;
5333 btrfs_free_path(path
);
5337 static void inode_tree_add(struct inode
*inode
)
5339 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5340 struct btrfs_inode
*entry
;
5342 struct rb_node
*parent
;
5343 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
5344 u64 ino
= btrfs_ino(inode
);
5346 if (inode_unhashed(inode
))
5349 spin_lock(&root
->inode_lock
);
5350 p
= &root
->inode_tree
.rb_node
;
5353 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
5355 if (ino
< btrfs_ino(&entry
->vfs_inode
))
5356 p
= &parent
->rb_left
;
5357 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
5358 p
= &parent
->rb_right
;
5360 WARN_ON(!(entry
->vfs_inode
.i_state
&
5361 (I_WILL_FREE
| I_FREEING
)));
5362 rb_replace_node(parent
, new, &root
->inode_tree
);
5363 RB_CLEAR_NODE(parent
);
5364 spin_unlock(&root
->inode_lock
);
5368 rb_link_node(new, parent
, p
);
5369 rb_insert_color(new, &root
->inode_tree
);
5370 spin_unlock(&root
->inode_lock
);
5373 static void inode_tree_del(struct inode
*inode
)
5375 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5378 spin_lock(&root
->inode_lock
);
5379 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
5380 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
5381 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
5382 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5384 spin_unlock(&root
->inode_lock
);
5386 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
5387 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
5388 spin_lock(&root
->inode_lock
);
5389 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
5390 spin_unlock(&root
->inode_lock
);
5392 btrfs_add_dead_root(root
);
5396 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
5398 struct rb_node
*node
;
5399 struct rb_node
*prev
;
5400 struct btrfs_inode
*entry
;
5401 struct inode
*inode
;
5404 if (!test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
5405 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
5407 spin_lock(&root
->inode_lock
);
5409 node
= root
->inode_tree
.rb_node
;
5413 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5415 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
5416 node
= node
->rb_left
;
5417 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
5418 node
= node
->rb_right
;
5424 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
5425 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
5429 prev
= rb_next(prev
);
5433 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
5434 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
5435 inode
= igrab(&entry
->vfs_inode
);
5437 spin_unlock(&root
->inode_lock
);
5438 if (atomic_read(&inode
->i_count
) > 1)
5439 d_prune_aliases(inode
);
5441 * btrfs_drop_inode will have it removed from
5442 * the inode cache when its usage count
5447 spin_lock(&root
->inode_lock
);
5451 if (cond_resched_lock(&root
->inode_lock
))
5454 node
= rb_next(node
);
5456 spin_unlock(&root
->inode_lock
);
5459 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
5461 struct btrfs_iget_args
*args
= p
;
5462 inode
->i_ino
= args
->location
->objectid
;
5463 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
5464 sizeof(*args
->location
));
5465 BTRFS_I(inode
)->root
= args
->root
;
5469 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
5471 struct btrfs_iget_args
*args
= opaque
;
5472 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
5473 args
->root
== BTRFS_I(inode
)->root
;
5476 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5477 struct btrfs_key
*location
,
5478 struct btrfs_root
*root
)
5480 struct inode
*inode
;
5481 struct btrfs_iget_args args
;
5482 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5484 args
.location
= location
;
5487 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5488 btrfs_init_locked_inode
,
5493 /* Get an inode object given its location and corresponding root.
5494 * Returns in *is_new if the inode was read from disk
5496 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5497 struct btrfs_root
*root
, int *new)
5499 struct inode
*inode
;
5501 inode
= btrfs_iget_locked(s
, location
, root
);
5503 return ERR_PTR(-ENOMEM
);
5505 if (inode
->i_state
& I_NEW
) {
5506 btrfs_read_locked_inode(inode
);
5507 if (!is_bad_inode(inode
)) {
5508 inode_tree_add(inode
);
5509 unlock_new_inode(inode
);
5513 unlock_new_inode(inode
);
5515 inode
= ERR_PTR(-ESTALE
);
5522 static struct inode
*new_simple_dir(struct super_block
*s
,
5523 struct btrfs_key
*key
,
5524 struct btrfs_root
*root
)
5526 struct inode
*inode
= new_inode(s
);
5529 return ERR_PTR(-ENOMEM
);
5531 BTRFS_I(inode
)->root
= root
;
5532 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5533 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5535 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5536 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5537 inode
->i_fop
= &simple_dir_operations
;
5538 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5539 inode
->i_mtime
= CURRENT_TIME
;
5540 inode
->i_atime
= inode
->i_mtime
;
5541 inode
->i_ctime
= inode
->i_mtime
;
5542 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
5547 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5549 struct inode
*inode
;
5550 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5551 struct btrfs_root
*sub_root
= root
;
5552 struct btrfs_key location
;
5556 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5557 return ERR_PTR(-ENAMETOOLONG
);
5559 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5561 return ERR_PTR(ret
);
5563 if (location
.objectid
== 0)
5564 return ERR_PTR(-ENOENT
);
5566 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5567 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5571 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5573 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5574 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5575 &location
, &sub_root
);
5578 inode
= ERR_PTR(ret
);
5580 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5582 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5584 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5586 if (!IS_ERR(inode
) && root
!= sub_root
) {
5587 down_read(&root
->fs_info
->cleanup_work_sem
);
5588 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5589 ret
= btrfs_orphan_cleanup(sub_root
);
5590 up_read(&root
->fs_info
->cleanup_work_sem
);
5593 inode
= ERR_PTR(ret
);
5600 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5602 struct btrfs_root
*root
;
5603 struct inode
*inode
= d_inode(dentry
);
5605 if (!inode
&& !IS_ROOT(dentry
))
5606 inode
= d_inode(dentry
->d_parent
);
5609 root
= BTRFS_I(inode
)->root
;
5610 if (btrfs_root_refs(&root
->root_item
) == 0)
5613 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5619 static void btrfs_dentry_release(struct dentry
*dentry
)
5621 kfree(dentry
->d_fsdata
);
5624 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5627 struct inode
*inode
;
5629 inode
= btrfs_lookup_dentry(dir
, dentry
);
5630 if (IS_ERR(inode
)) {
5631 if (PTR_ERR(inode
) == -ENOENT
)
5634 return ERR_CAST(inode
);
5637 return d_splice_alias(inode
, dentry
);
5640 unsigned char btrfs_filetype_table
[] = {
5641 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5644 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5646 struct inode
*inode
= file_inode(file
);
5647 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5648 struct btrfs_item
*item
;
5649 struct btrfs_dir_item
*di
;
5650 struct btrfs_key key
;
5651 struct btrfs_key found_key
;
5652 struct btrfs_path
*path
;
5653 struct list_head ins_list
;
5654 struct list_head del_list
;
5656 struct extent_buffer
*leaf
;
5658 unsigned char d_type
;
5663 int key_type
= BTRFS_DIR_INDEX_KEY
;
5667 int is_curr
= 0; /* ctx->pos points to the current index? */
5669 /* FIXME, use a real flag for deciding about the key type */
5670 if (root
->fs_info
->tree_root
== root
)
5671 key_type
= BTRFS_DIR_ITEM_KEY
;
5673 if (!dir_emit_dots(file
, ctx
))
5676 path
= btrfs_alloc_path();
5682 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5683 INIT_LIST_HEAD(&ins_list
);
5684 INIT_LIST_HEAD(&del_list
);
5685 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5688 key
.type
= key_type
;
5689 key
.offset
= ctx
->pos
;
5690 key
.objectid
= btrfs_ino(inode
);
5692 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5697 leaf
= path
->nodes
[0];
5698 slot
= path
->slots
[0];
5699 if (slot
>= btrfs_header_nritems(leaf
)) {
5700 ret
= btrfs_next_leaf(root
, path
);
5708 item
= btrfs_item_nr(slot
);
5709 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5711 if (found_key
.objectid
!= key
.objectid
)
5713 if (found_key
.type
!= key_type
)
5715 if (found_key
.offset
< ctx
->pos
)
5717 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5718 btrfs_should_delete_dir_index(&del_list
,
5722 ctx
->pos
= found_key
.offset
;
5725 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5727 di_total
= btrfs_item_size(leaf
, item
);
5729 while (di_cur
< di_total
) {
5730 struct btrfs_key location
;
5732 if (verify_dir_item(root
, leaf
, di
))
5735 name_len
= btrfs_dir_name_len(leaf
, di
);
5736 if (name_len
<= sizeof(tmp_name
)) {
5737 name_ptr
= tmp_name
;
5739 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5745 read_extent_buffer(leaf
, name_ptr
,
5746 (unsigned long)(di
+ 1), name_len
);
5748 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5749 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5752 /* is this a reference to our own snapshot? If so
5755 * In contrast to old kernels, we insert the snapshot's
5756 * dir item and dir index after it has been created, so
5757 * we won't find a reference to our own snapshot. We
5758 * still keep the following code for backward
5761 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5762 location
.objectid
== root
->root_key
.objectid
) {
5766 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5767 location
.objectid
, d_type
);
5770 if (name_ptr
!= tmp_name
)
5775 di_len
= btrfs_dir_name_len(leaf
, di
) +
5776 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5778 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5784 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5787 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
5792 /* Reached end of directory/root. Bump pos past the last item. */
5796 * Stop new entries from being returned after we return the last
5799 * New directory entries are assigned a strictly increasing
5800 * offset. This means that new entries created during readdir
5801 * are *guaranteed* to be seen in the future by that readdir.
5802 * This has broken buggy programs which operate on names as
5803 * they're returned by readdir. Until we re-use freed offsets
5804 * we have this hack to stop new entries from being returned
5805 * under the assumption that they'll never reach this huge
5808 * This is being careful not to overflow 32bit loff_t unless the
5809 * last entry requires it because doing so has broken 32bit apps
5812 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5813 if (ctx
->pos
>= INT_MAX
)
5814 ctx
->pos
= LLONG_MAX
;
5821 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5822 btrfs_put_delayed_items(&ins_list
, &del_list
);
5823 btrfs_free_path(path
);
5827 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5829 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5830 struct btrfs_trans_handle
*trans
;
5832 bool nolock
= false;
5834 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5837 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5840 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5842 trans
= btrfs_join_transaction_nolock(root
);
5844 trans
= btrfs_join_transaction(root
);
5846 return PTR_ERR(trans
);
5847 ret
= btrfs_commit_transaction(trans
, root
);
5853 * This is somewhat expensive, updating the tree every time the
5854 * inode changes. But, it is most likely to find the inode in cache.
5855 * FIXME, needs more benchmarking...there are no reasons other than performance
5856 * to keep or drop this code.
5858 static int btrfs_dirty_inode(struct inode
*inode
)
5860 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5861 struct btrfs_trans_handle
*trans
;
5864 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5867 trans
= btrfs_join_transaction(root
);
5869 return PTR_ERR(trans
);
5871 ret
= btrfs_update_inode(trans
, root
, inode
);
5872 if (ret
&& ret
== -ENOSPC
) {
5873 /* whoops, lets try again with the full transaction */
5874 btrfs_end_transaction(trans
, root
);
5875 trans
= btrfs_start_transaction(root
, 1);
5877 return PTR_ERR(trans
);
5879 ret
= btrfs_update_inode(trans
, root
, inode
);
5881 btrfs_end_transaction(trans
, root
);
5882 if (BTRFS_I(inode
)->delayed_node
)
5883 btrfs_balance_delayed_items(root
);
5889 * This is a copy of file_update_time. We need this so we can return error on
5890 * ENOSPC for updating the inode in the case of file write and mmap writes.
5892 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
5895 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5897 if (btrfs_root_readonly(root
))
5900 if (flags
& S_VERSION
)
5901 inode_inc_iversion(inode
);
5902 if (flags
& S_CTIME
)
5903 inode
->i_ctime
= *now
;
5904 if (flags
& S_MTIME
)
5905 inode
->i_mtime
= *now
;
5906 if (flags
& S_ATIME
)
5907 inode
->i_atime
= *now
;
5908 return btrfs_dirty_inode(inode
);
5912 * find the highest existing sequence number in a directory
5913 * and then set the in-memory index_cnt variable to reflect
5914 * free sequence numbers
5916 static int btrfs_set_inode_index_count(struct inode
*inode
)
5918 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5919 struct btrfs_key key
, found_key
;
5920 struct btrfs_path
*path
;
5921 struct extent_buffer
*leaf
;
5924 key
.objectid
= btrfs_ino(inode
);
5925 key
.type
= BTRFS_DIR_INDEX_KEY
;
5926 key
.offset
= (u64
)-1;
5928 path
= btrfs_alloc_path();
5932 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5935 /* FIXME: we should be able to handle this */
5941 * MAGIC NUMBER EXPLANATION:
5942 * since we search a directory based on f_pos we have to start at 2
5943 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5944 * else has to start at 2
5946 if (path
->slots
[0] == 0) {
5947 BTRFS_I(inode
)->index_cnt
= 2;
5953 leaf
= path
->nodes
[0];
5954 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5956 if (found_key
.objectid
!= btrfs_ino(inode
) ||
5957 found_key
.type
!= BTRFS_DIR_INDEX_KEY
) {
5958 BTRFS_I(inode
)->index_cnt
= 2;
5962 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
5964 btrfs_free_path(path
);
5969 * helper to find a free sequence number in a given directory. This current
5970 * code is very simple, later versions will do smarter things in the btree
5972 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
5976 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
5977 ret
= btrfs_inode_delayed_dir_index_count(dir
);
5979 ret
= btrfs_set_inode_index_count(dir
);
5985 *index
= BTRFS_I(dir
)->index_cnt
;
5986 BTRFS_I(dir
)->index_cnt
++;
5991 static int btrfs_insert_inode_locked(struct inode
*inode
)
5993 struct btrfs_iget_args args
;
5994 args
.location
= &BTRFS_I(inode
)->location
;
5995 args
.root
= BTRFS_I(inode
)->root
;
5997 return insert_inode_locked4(inode
,
5998 btrfs_inode_hash(inode
->i_ino
, BTRFS_I(inode
)->root
),
5999 btrfs_find_actor
, &args
);
6002 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
6003 struct btrfs_root
*root
,
6005 const char *name
, int name_len
,
6006 u64 ref_objectid
, u64 objectid
,
6007 umode_t mode
, u64
*index
)
6009 struct inode
*inode
;
6010 struct btrfs_inode_item
*inode_item
;
6011 struct btrfs_key
*location
;
6012 struct btrfs_path
*path
;
6013 struct btrfs_inode_ref
*ref
;
6014 struct btrfs_key key
[2];
6016 int nitems
= name
? 2 : 1;
6020 path
= btrfs_alloc_path();
6022 return ERR_PTR(-ENOMEM
);
6024 inode
= new_inode(root
->fs_info
->sb
);
6026 btrfs_free_path(path
);
6027 return ERR_PTR(-ENOMEM
);
6031 * O_TMPFILE, set link count to 0, so that after this point,
6032 * we fill in an inode item with the correct link count.
6035 set_nlink(inode
, 0);
6038 * we have to initialize this early, so we can reclaim the inode
6039 * number if we fail afterwards in this function.
6041 inode
->i_ino
= objectid
;
6044 trace_btrfs_inode_request(dir
);
6046 ret
= btrfs_set_inode_index(dir
, index
);
6048 btrfs_free_path(path
);
6050 return ERR_PTR(ret
);
6056 * index_cnt is ignored for everything but a dir,
6057 * btrfs_get_inode_index_count has an explanation for the magic
6060 BTRFS_I(inode
)->index_cnt
= 2;
6061 BTRFS_I(inode
)->dir_index
= *index
;
6062 BTRFS_I(inode
)->root
= root
;
6063 BTRFS_I(inode
)->generation
= trans
->transid
;
6064 inode
->i_generation
= BTRFS_I(inode
)->generation
;
6067 * We could have gotten an inode number from somebody who was fsynced
6068 * and then removed in this same transaction, so let's just set full
6069 * sync since it will be a full sync anyway and this will blow away the
6070 * old info in the log.
6072 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6074 key
[0].objectid
= objectid
;
6075 key
[0].type
= BTRFS_INODE_ITEM_KEY
;
6078 sizes
[0] = sizeof(struct btrfs_inode_item
);
6082 * Start new inodes with an inode_ref. This is slightly more
6083 * efficient for small numbers of hard links since they will
6084 * be packed into one item. Extended refs will kick in if we
6085 * add more hard links than can fit in the ref item.
6087 key
[1].objectid
= objectid
;
6088 key
[1].type
= BTRFS_INODE_REF_KEY
;
6089 key
[1].offset
= ref_objectid
;
6091 sizes
[1] = name_len
+ sizeof(*ref
);
6094 location
= &BTRFS_I(inode
)->location
;
6095 location
->objectid
= objectid
;
6096 location
->offset
= 0;
6097 location
->type
= BTRFS_INODE_ITEM_KEY
;
6099 ret
= btrfs_insert_inode_locked(inode
);
6103 path
->leave_spinning
= 1;
6104 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, nitems
);
6108 inode_init_owner(inode
, dir
, mode
);
6109 inode_set_bytes(inode
, 0);
6111 inode
->i_mtime
= CURRENT_TIME
;
6112 inode
->i_atime
= inode
->i_mtime
;
6113 inode
->i_ctime
= inode
->i_mtime
;
6114 BTRFS_I(inode
)->i_otime
= inode
->i_mtime
;
6116 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
6117 struct btrfs_inode_item
);
6118 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
6119 sizeof(*inode_item
));
6120 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
6123 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
6124 struct btrfs_inode_ref
);
6125 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
6126 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
6127 ptr
= (unsigned long)(ref
+ 1);
6128 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
6131 btrfs_mark_buffer_dirty(path
->nodes
[0]);
6132 btrfs_free_path(path
);
6134 btrfs_inherit_iflags(inode
, dir
);
6136 if (S_ISREG(mode
)) {
6137 if (btrfs_test_opt(root
, NODATASUM
))
6138 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
6139 if (btrfs_test_opt(root
, NODATACOW
))
6140 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
6141 BTRFS_INODE_NODATASUM
;
6144 inode_tree_add(inode
);
6146 trace_btrfs_inode_new(inode
);
6147 btrfs_set_inode_last_trans(trans
, inode
);
6149 btrfs_update_root_times(trans
, root
);
6151 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
6153 btrfs_err(root
->fs_info
,
6154 "error inheriting props for ino %llu (root %llu): %d",
6155 btrfs_ino(inode
), root
->root_key
.objectid
, ret
);
6160 unlock_new_inode(inode
);
6163 BTRFS_I(dir
)->index_cnt
--;
6164 btrfs_free_path(path
);
6166 return ERR_PTR(ret
);
6169 static inline u8
btrfs_inode_type(struct inode
*inode
)
6171 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
6175 * utility function to add 'inode' into 'parent_inode' with
6176 * a give name and a given sequence number.
6177 * if 'add_backref' is true, also insert a backref from the
6178 * inode to the parent directory.
6180 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
6181 struct inode
*parent_inode
, struct inode
*inode
,
6182 const char *name
, int name_len
, int add_backref
, u64 index
)
6185 struct btrfs_key key
;
6186 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
6187 u64 ino
= btrfs_ino(inode
);
6188 u64 parent_ino
= btrfs_ino(parent_inode
);
6190 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6191 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
6194 key
.type
= BTRFS_INODE_ITEM_KEY
;
6198 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6199 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
6200 key
.objectid
, root
->root_key
.objectid
,
6201 parent_ino
, index
, name
, name_len
);
6202 } else if (add_backref
) {
6203 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
6207 /* Nothing to clean up yet */
6211 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
6213 btrfs_inode_type(inode
), index
);
6214 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
6217 btrfs_abort_transaction(trans
, root
, ret
);
6221 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
6223 inode_inc_iversion(parent_inode
);
6224 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
6225 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
6227 btrfs_abort_transaction(trans
, root
, ret
);
6231 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6234 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
6235 key
.objectid
, root
->root_key
.objectid
,
6236 parent_ino
, &local_index
, name
, name_len
);
6238 } else if (add_backref
) {
6242 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
6243 ino
, parent_ino
, &local_index
);
6248 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
6249 struct inode
*dir
, struct dentry
*dentry
,
6250 struct inode
*inode
, int backref
, u64 index
)
6252 int err
= btrfs_add_link(trans
, dir
, inode
,
6253 dentry
->d_name
.name
, dentry
->d_name
.len
,
6260 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
6261 umode_t mode
, dev_t rdev
)
6263 struct btrfs_trans_handle
*trans
;
6264 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6265 struct inode
*inode
= NULL
;
6271 if (!new_valid_dev(rdev
))
6275 * 2 for inode item and ref
6277 * 1 for xattr if selinux is on
6279 trans
= btrfs_start_transaction(root
, 5);
6281 return PTR_ERR(trans
);
6283 err
= btrfs_find_free_ino(root
, &objectid
);
6287 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6288 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6290 if (IS_ERR(inode
)) {
6291 err
= PTR_ERR(inode
);
6296 * If the active LSM wants to access the inode during
6297 * d_instantiate it needs these. Smack checks to see
6298 * if the filesystem supports xattrs by looking at the
6301 inode
->i_op
= &btrfs_special_inode_operations
;
6302 init_special_inode(inode
, inode
->i_mode
, rdev
);
6304 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6306 goto out_unlock_inode
;
6308 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6310 goto out_unlock_inode
;
6312 btrfs_update_inode(trans
, root
, inode
);
6313 unlock_new_inode(inode
);
6314 d_instantiate(dentry
, inode
);
6318 btrfs_end_transaction(trans
, root
);
6319 btrfs_balance_delayed_items(root
);
6320 btrfs_btree_balance_dirty(root
);
6322 inode_dec_link_count(inode
);
6329 unlock_new_inode(inode
);
6334 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
6335 umode_t mode
, bool excl
)
6337 struct btrfs_trans_handle
*trans
;
6338 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6339 struct inode
*inode
= NULL
;
6340 int drop_inode_on_err
= 0;
6346 * 2 for inode item and ref
6348 * 1 for xattr if selinux is on
6350 trans
= btrfs_start_transaction(root
, 5);
6352 return PTR_ERR(trans
);
6354 err
= btrfs_find_free_ino(root
, &objectid
);
6358 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6359 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6361 if (IS_ERR(inode
)) {
6362 err
= PTR_ERR(inode
);
6365 drop_inode_on_err
= 1;
6367 * If the active LSM wants to access the inode during
6368 * d_instantiate it needs these. Smack checks to see
6369 * if the filesystem supports xattrs by looking at the
6372 inode
->i_fop
= &btrfs_file_operations
;
6373 inode
->i_op
= &btrfs_file_inode_operations
;
6374 inode
->i_mapping
->a_ops
= &btrfs_aops
;
6376 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6378 goto out_unlock_inode
;
6380 err
= btrfs_update_inode(trans
, root
, inode
);
6382 goto out_unlock_inode
;
6384 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
6386 goto out_unlock_inode
;
6388 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
6389 unlock_new_inode(inode
);
6390 d_instantiate(dentry
, inode
);
6393 btrfs_end_transaction(trans
, root
);
6394 if (err
&& drop_inode_on_err
) {
6395 inode_dec_link_count(inode
);
6398 btrfs_balance_delayed_items(root
);
6399 btrfs_btree_balance_dirty(root
);
6403 unlock_new_inode(inode
);
6408 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
6409 struct dentry
*dentry
)
6411 struct btrfs_trans_handle
*trans
;
6412 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6413 struct inode
*inode
= d_inode(old_dentry
);
6418 /* do not allow sys_link's with other subvols of the same device */
6419 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
6422 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
6425 err
= btrfs_set_inode_index(dir
, &index
);
6430 * 2 items for inode and inode ref
6431 * 2 items for dir items
6432 * 1 item for parent inode
6434 trans
= btrfs_start_transaction(root
, 5);
6435 if (IS_ERR(trans
)) {
6436 err
= PTR_ERR(trans
);
6440 /* There are several dir indexes for this inode, clear the cache. */
6441 BTRFS_I(inode
)->dir_index
= 0ULL;
6443 inode_inc_iversion(inode
);
6444 inode
->i_ctime
= CURRENT_TIME
;
6446 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
6448 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
6453 struct dentry
*parent
= dentry
->d_parent
;
6454 err
= btrfs_update_inode(trans
, root
, inode
);
6457 if (inode
->i_nlink
== 1) {
6459 * If new hard link count is 1, it's a file created
6460 * with open(2) O_TMPFILE flag.
6462 err
= btrfs_orphan_del(trans
, inode
);
6466 d_instantiate(dentry
, inode
);
6467 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
6470 btrfs_end_transaction(trans
, root
);
6471 btrfs_balance_delayed_items(root
);
6474 inode_dec_link_count(inode
);
6477 btrfs_btree_balance_dirty(root
);
6481 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
6483 struct inode
*inode
= NULL
;
6484 struct btrfs_trans_handle
*trans
;
6485 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
6487 int drop_on_err
= 0;
6492 * 2 items for inode and ref
6493 * 2 items for dir items
6494 * 1 for xattr if selinux is on
6496 trans
= btrfs_start_transaction(root
, 5);
6498 return PTR_ERR(trans
);
6500 err
= btrfs_find_free_ino(root
, &objectid
);
6504 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
6505 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
6506 S_IFDIR
| mode
, &index
);
6507 if (IS_ERR(inode
)) {
6508 err
= PTR_ERR(inode
);
6513 /* these must be set before we unlock the inode */
6514 inode
->i_op
= &btrfs_dir_inode_operations
;
6515 inode
->i_fop
= &btrfs_dir_file_operations
;
6517 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
6519 goto out_fail_inode
;
6521 btrfs_i_size_write(inode
, 0);
6522 err
= btrfs_update_inode(trans
, root
, inode
);
6524 goto out_fail_inode
;
6526 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
6527 dentry
->d_name
.len
, 0, index
);
6529 goto out_fail_inode
;
6531 d_instantiate(dentry
, inode
);
6533 * mkdir is special. We're unlocking after we call d_instantiate
6534 * to avoid a race with nfsd calling d_instantiate.
6536 unlock_new_inode(inode
);
6540 btrfs_end_transaction(trans
, root
);
6542 inode_dec_link_count(inode
);
6545 btrfs_balance_delayed_items(root
);
6546 btrfs_btree_balance_dirty(root
);
6550 unlock_new_inode(inode
);
6554 /* Find next extent map of a given extent map, caller needs to ensure locks */
6555 static struct extent_map
*next_extent_map(struct extent_map
*em
)
6557 struct rb_node
*next
;
6559 next
= rb_next(&em
->rb_node
);
6562 return container_of(next
, struct extent_map
, rb_node
);
6565 static struct extent_map
*prev_extent_map(struct extent_map
*em
)
6567 struct rb_node
*prev
;
6569 prev
= rb_prev(&em
->rb_node
);
6572 return container_of(prev
, struct extent_map
, rb_node
);
6575 /* helper for btfs_get_extent. Given an existing extent in the tree,
6576 * the existing extent is the nearest extent to map_start,
6577 * and an extent that you want to insert, deal with overlap and insert
6578 * the best fitted new extent into the tree.
6580 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6581 struct extent_map
*existing
,
6582 struct extent_map
*em
,
6585 struct extent_map
*prev
;
6586 struct extent_map
*next
;
6591 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6593 if (existing
->start
> map_start
) {
6595 prev
= prev_extent_map(next
);
6598 next
= next_extent_map(prev
);
6601 start
= prev
? extent_map_end(prev
) : em
->start
;
6602 start
= max_t(u64
, start
, em
->start
);
6603 end
= next
? next
->start
: extent_map_end(em
);
6604 end
= min_t(u64
, end
, extent_map_end(em
));
6605 start_diff
= start
- em
->start
;
6607 em
->len
= end
- start
;
6608 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6609 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6610 em
->block_start
+= start_diff
;
6611 em
->block_len
-= start_diff
;
6613 return add_extent_mapping(em_tree
, em
, 0);
6616 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6617 struct inode
*inode
, struct page
*page
,
6618 size_t pg_offset
, u64 extent_offset
,
6619 struct btrfs_file_extent_item
*item
)
6622 struct extent_buffer
*leaf
= path
->nodes
[0];
6625 unsigned long inline_size
;
6629 WARN_ON(pg_offset
!= 0);
6630 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6631 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6632 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6633 btrfs_item_nr(path
->slots
[0]));
6634 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6637 ptr
= btrfs_file_extent_inline_start(item
);
6639 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6641 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
6642 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6643 extent_offset
, inline_size
, max_size
);
6649 * a bit scary, this does extent mapping from logical file offset to the disk.
6650 * the ugly parts come from merging extents from the disk with the in-ram
6651 * representation. This gets more complex because of the data=ordered code,
6652 * where the in-ram extents might be locked pending data=ordered completion.
6654 * This also copies inline extents directly into the page.
6657 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6658 size_t pg_offset
, u64 start
, u64 len
,
6663 u64 extent_start
= 0;
6665 u64 objectid
= btrfs_ino(inode
);
6667 struct btrfs_path
*path
= NULL
;
6668 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6669 struct btrfs_file_extent_item
*item
;
6670 struct extent_buffer
*leaf
;
6671 struct btrfs_key found_key
;
6672 struct extent_map
*em
= NULL
;
6673 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6674 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6675 struct btrfs_trans_handle
*trans
= NULL
;
6676 const bool new_inline
= !page
|| create
;
6679 read_lock(&em_tree
->lock
);
6680 em
= lookup_extent_mapping(em_tree
, start
, len
);
6682 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6683 read_unlock(&em_tree
->lock
);
6686 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6687 free_extent_map(em
);
6688 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6689 free_extent_map(em
);
6693 em
= alloc_extent_map();
6698 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6699 em
->start
= EXTENT_MAP_HOLE
;
6700 em
->orig_start
= EXTENT_MAP_HOLE
;
6702 em
->block_len
= (u64
)-1;
6705 path
= btrfs_alloc_path();
6711 * Chances are we'll be called again, so go ahead and do
6717 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6718 objectid
, start
, trans
!= NULL
);
6725 if (path
->slots
[0] == 0)
6730 leaf
= path
->nodes
[0];
6731 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6732 struct btrfs_file_extent_item
);
6733 /* are we inside the extent that was found? */
6734 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6735 found_type
= found_key
.type
;
6736 if (found_key
.objectid
!= objectid
||
6737 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6739 * If we backup past the first extent we want to move forward
6740 * and see if there is an extent in front of us, otherwise we'll
6741 * say there is a hole for our whole search range which can
6748 found_type
= btrfs_file_extent_type(leaf
, item
);
6749 extent_start
= found_key
.offset
;
6750 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6751 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6752 extent_end
= extent_start
+
6753 btrfs_file_extent_num_bytes(leaf
, item
);
6754 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6756 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6757 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6760 if (start
>= extent_end
) {
6762 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6763 ret
= btrfs_next_leaf(root
, path
);
6770 leaf
= path
->nodes
[0];
6772 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6773 if (found_key
.objectid
!= objectid
||
6774 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6776 if (start
+ len
<= found_key
.offset
)
6778 if (start
> found_key
.offset
)
6781 em
->orig_start
= start
;
6782 em
->len
= found_key
.offset
- start
;
6786 btrfs_extent_item_to_extent_map(inode
, path
, item
, new_inline
, em
);
6788 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6789 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6791 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6795 size_t extent_offset
;
6801 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6802 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6803 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6804 size
- extent_offset
);
6805 em
->start
= extent_start
+ extent_offset
;
6806 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6807 em
->orig_block_len
= em
->len
;
6808 em
->orig_start
= em
->start
;
6809 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6810 if (create
== 0 && !PageUptodate(page
)) {
6811 if (btrfs_file_extent_compression(leaf
, item
) !=
6812 BTRFS_COMPRESS_NONE
) {
6813 ret
= uncompress_inline(path
, inode
, page
,
6815 extent_offset
, item
);
6822 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6824 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6825 memset(map
+ pg_offset
+ copy_size
, 0,
6826 PAGE_CACHE_SIZE
- pg_offset
-
6831 flush_dcache_page(page
);
6832 } else if (create
&& PageUptodate(page
)) {
6836 free_extent_map(em
);
6839 btrfs_release_path(path
);
6840 trans
= btrfs_join_transaction(root
);
6843 return ERR_CAST(trans
);
6847 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6850 btrfs_mark_buffer_dirty(leaf
);
6852 set_extent_uptodate(io_tree
, em
->start
,
6853 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6858 em
->orig_start
= start
;
6861 em
->block_start
= EXTENT_MAP_HOLE
;
6862 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6864 btrfs_release_path(path
);
6865 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6866 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6867 em
->start
, em
->len
, start
, len
);
6873 write_lock(&em_tree
->lock
);
6874 ret
= add_extent_mapping(em_tree
, em
, 0);
6875 /* it is possible that someone inserted the extent into the tree
6876 * while we had the lock dropped. It is also possible that
6877 * an overlapping map exists in the tree
6879 if (ret
== -EEXIST
) {
6880 struct extent_map
*existing
;
6884 existing
= search_extent_mapping(em_tree
, start
, len
);
6886 * existing will always be non-NULL, since there must be
6887 * extent causing the -EEXIST.
6889 if (start
>= extent_map_end(existing
) ||
6890 start
<= existing
->start
) {
6892 * The existing extent map is the one nearest to
6893 * the [start, start + len) range which overlaps
6895 err
= merge_extent_mapping(em_tree
, existing
,
6897 free_extent_map(existing
);
6899 free_extent_map(em
);
6903 free_extent_map(em
);
6908 write_unlock(&em_tree
->lock
);
6911 trace_btrfs_get_extent(root
, em
);
6913 btrfs_free_path(path
);
6915 ret
= btrfs_end_transaction(trans
, root
);
6920 free_extent_map(em
);
6921 return ERR_PTR(err
);
6923 BUG_ON(!em
); /* Error is always set */
6927 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
6928 size_t pg_offset
, u64 start
, u64 len
,
6931 struct extent_map
*em
;
6932 struct extent_map
*hole_em
= NULL
;
6933 u64 range_start
= start
;
6939 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
6946 * - a pre-alloc extent,
6947 * there might actually be delalloc bytes behind it.
6949 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
6950 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6956 /* check to see if we've wrapped (len == -1 or similar) */
6965 /* ok, we didn't find anything, lets look for delalloc */
6966 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
6967 end
, len
, EXTENT_DELALLOC
, 1);
6968 found_end
= range_start
+ found
;
6969 if (found_end
< range_start
)
6970 found_end
= (u64
)-1;
6973 * we didn't find anything useful, return
6974 * the original results from get_extent()
6976 if (range_start
> end
|| found_end
<= start
) {
6982 /* adjust the range_start to make sure it doesn't
6983 * go backwards from the start they passed in
6985 range_start
= max(start
, range_start
);
6986 found
= found_end
- range_start
;
6989 u64 hole_start
= start
;
6992 em
= alloc_extent_map();
6998 * when btrfs_get_extent can't find anything it
6999 * returns one huge hole
7001 * make sure what it found really fits our range, and
7002 * adjust to make sure it is based on the start from
7006 u64 calc_end
= extent_map_end(hole_em
);
7008 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
7009 free_extent_map(hole_em
);
7012 hole_start
= max(hole_em
->start
, start
);
7013 hole_len
= calc_end
- hole_start
;
7017 if (hole_em
&& range_start
> hole_start
) {
7018 /* our hole starts before our delalloc, so we
7019 * have to return just the parts of the hole
7020 * that go until the delalloc starts
7022 em
->len
= min(hole_len
,
7023 range_start
- hole_start
);
7024 em
->start
= hole_start
;
7025 em
->orig_start
= hole_start
;
7027 * don't adjust block start at all,
7028 * it is fixed at EXTENT_MAP_HOLE
7030 em
->block_start
= hole_em
->block_start
;
7031 em
->block_len
= hole_len
;
7032 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
7033 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7035 em
->start
= range_start
;
7037 em
->orig_start
= range_start
;
7038 em
->block_start
= EXTENT_MAP_DELALLOC
;
7039 em
->block_len
= found
;
7041 } else if (hole_em
) {
7046 free_extent_map(hole_em
);
7048 free_extent_map(em
);
7049 return ERR_PTR(err
);
7054 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
7057 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7058 struct extent_map
*em
;
7059 struct btrfs_key ins
;
7063 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
7064 ret
= btrfs_reserve_extent(root
, len
, root
->sectorsize
, 0,
7065 alloc_hint
, &ins
, 1, 1);
7067 return ERR_PTR(ret
);
7069 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
7070 ins
.offset
, ins
.offset
, ins
.offset
, 0);
7072 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7076 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
7077 ins
.offset
, ins
.offset
, 0);
7079 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
, 1);
7080 free_extent_map(em
);
7081 return ERR_PTR(ret
);
7088 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7089 * block must be cow'd
7091 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
7092 u64
*orig_start
, u64
*orig_block_len
,
7095 struct btrfs_trans_handle
*trans
;
7096 struct btrfs_path
*path
;
7098 struct extent_buffer
*leaf
;
7099 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7100 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7101 struct btrfs_file_extent_item
*fi
;
7102 struct btrfs_key key
;
7109 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
7111 path
= btrfs_alloc_path();
7115 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
7120 slot
= path
->slots
[0];
7123 /* can't find the item, must cow */
7130 leaf
= path
->nodes
[0];
7131 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
7132 if (key
.objectid
!= btrfs_ino(inode
) ||
7133 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
7134 /* not our file or wrong item type, must cow */
7138 if (key
.offset
> offset
) {
7139 /* Wrong offset, must cow */
7143 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
7144 found_type
= btrfs_file_extent_type(leaf
, fi
);
7145 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
7146 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
7147 /* not a regular extent, must cow */
7151 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
7154 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
7155 if (extent_end
<= offset
)
7158 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
7159 if (disk_bytenr
== 0)
7162 if (btrfs_file_extent_compression(leaf
, fi
) ||
7163 btrfs_file_extent_encryption(leaf
, fi
) ||
7164 btrfs_file_extent_other_encoding(leaf
, fi
))
7167 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
7170 *orig_start
= key
.offset
- backref_offset
;
7171 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
7172 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
7175 if (btrfs_extent_readonly(root
, disk_bytenr
))
7178 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
7179 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
7182 range_end
= round_up(offset
+ num_bytes
, root
->sectorsize
) - 1;
7183 ret
= test_range_bit(io_tree
, offset
, range_end
,
7184 EXTENT_DELALLOC
, 0, NULL
);
7191 btrfs_release_path(path
);
7194 * look for other files referencing this extent, if we
7195 * find any we must cow
7197 trans
= btrfs_join_transaction(root
);
7198 if (IS_ERR(trans
)) {
7203 ret
= btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
7204 key
.offset
- backref_offset
, disk_bytenr
);
7205 btrfs_end_transaction(trans
, root
);
7212 * adjust disk_bytenr and num_bytes to cover just the bytes
7213 * in this extent we are about to write. If there
7214 * are any csums in that range we have to cow in order
7215 * to keep the csums correct
7217 disk_bytenr
+= backref_offset
;
7218 disk_bytenr
+= offset
- key
.offset
;
7219 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
7222 * all of the above have passed, it is safe to overwrite this extent
7228 btrfs_free_path(path
);
7232 bool btrfs_page_exists_in_range(struct inode
*inode
, loff_t start
, loff_t end
)
7234 struct radix_tree_root
*root
= &inode
->i_mapping
->page_tree
;
7236 void **pagep
= NULL
;
7237 struct page
*page
= NULL
;
7241 start_idx
= start
>> PAGE_CACHE_SHIFT
;
7244 * end is the last byte in the last page. end == start is legal
7246 end_idx
= end
>> PAGE_CACHE_SHIFT
;
7250 /* Most of the code in this while loop is lifted from
7251 * find_get_page. It's been modified to begin searching from a
7252 * page and return just the first page found in that range. If the
7253 * found idx is less than or equal to the end idx then we know that
7254 * a page exists. If no pages are found or if those pages are
7255 * outside of the range then we're fine (yay!) */
7256 while (page
== NULL
&&
7257 radix_tree_gang_lookup_slot(root
, &pagep
, NULL
, start_idx
, 1)) {
7258 page
= radix_tree_deref_slot(pagep
);
7259 if (unlikely(!page
))
7262 if (radix_tree_exception(page
)) {
7263 if (radix_tree_deref_retry(page
)) {
7268 * Otherwise, shmem/tmpfs must be storing a swap entry
7269 * here as an exceptional entry: so return it without
7270 * attempting to raise page count.
7273 break; /* TODO: Is this relevant for this use case? */
7276 if (!page_cache_get_speculative(page
)) {
7282 * Has the page moved?
7283 * This is part of the lockless pagecache protocol. See
7284 * include/linux/pagemap.h for details.
7286 if (unlikely(page
!= *pagep
)) {
7287 page_cache_release(page
);
7293 if (page
->index
<= end_idx
)
7295 page_cache_release(page
);
7302 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
7303 struct extent_state
**cached_state
, int writing
)
7305 struct btrfs_ordered_extent
*ordered
;
7309 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7312 * We're concerned with the entire range that we're going to be
7313 * doing DIO to, so we need to make sure theres no ordered
7314 * extents in this range.
7316 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
7317 lockend
- lockstart
+ 1);
7320 * We need to make sure there are no buffered pages in this
7321 * range either, we could have raced between the invalidate in
7322 * generic_file_direct_write and locking the extent. The
7323 * invalidate needs to happen so that reads after a write do not
7328 !btrfs_page_exists_in_range(inode
, lockstart
, lockend
)))
7331 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7332 cached_state
, GFP_NOFS
);
7335 btrfs_start_ordered_extent(inode
, ordered
, 1);
7336 btrfs_put_ordered_extent(ordered
);
7338 /* Screw you mmap */
7339 ret
= btrfs_fdatawrite_range(inode
, lockstart
, lockend
);
7342 ret
= filemap_fdatawait_range(inode
->i_mapping
,
7349 * If we found a page that couldn't be invalidated just
7350 * fall back to buffered.
7352 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
7353 lockstart
>> PAGE_CACHE_SHIFT
,
7354 lockend
>> PAGE_CACHE_SHIFT
);
7365 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
7366 u64 len
, u64 orig_start
,
7367 u64 block_start
, u64 block_len
,
7368 u64 orig_block_len
, u64 ram_bytes
,
7371 struct extent_map_tree
*em_tree
;
7372 struct extent_map
*em
;
7373 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7376 em_tree
= &BTRFS_I(inode
)->extent_tree
;
7377 em
= alloc_extent_map();
7379 return ERR_PTR(-ENOMEM
);
7382 em
->orig_start
= orig_start
;
7383 em
->mod_start
= start
;
7386 em
->block_len
= block_len
;
7387 em
->block_start
= block_start
;
7388 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7389 em
->orig_block_len
= orig_block_len
;
7390 em
->ram_bytes
= ram_bytes
;
7391 em
->generation
= -1;
7392 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
7393 if (type
== BTRFS_ORDERED_PREALLOC
)
7394 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
7397 btrfs_drop_extent_cache(inode
, em
->start
,
7398 em
->start
+ em
->len
- 1, 0);
7399 write_lock(&em_tree
->lock
);
7400 ret
= add_extent_mapping(em_tree
, em
, 1);
7401 write_unlock(&em_tree
->lock
);
7402 } while (ret
== -EEXIST
);
7405 free_extent_map(em
);
7406 return ERR_PTR(ret
);
7412 struct btrfs_dio_data
{
7413 u64 outstanding_extents
;
7417 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
7418 struct buffer_head
*bh_result
, int create
)
7420 struct extent_map
*em
;
7421 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7422 struct extent_state
*cached_state
= NULL
;
7423 struct btrfs_dio_data
*dio_data
= NULL
;
7424 u64 start
= iblock
<< inode
->i_blkbits
;
7425 u64 lockstart
, lockend
;
7426 u64 len
= bh_result
->b_size
;
7427 int unlock_bits
= EXTENT_LOCKED
;
7431 unlock_bits
|= EXTENT_DIRTY
;
7433 len
= min_t(u64
, len
, root
->sectorsize
);
7436 lockend
= start
+ len
- 1;
7438 if (current
->journal_info
) {
7440 * Need to pull our outstanding extents and set journal_info to NULL so
7441 * that anything that needs to check if there's a transction doesn't get
7444 dio_data
= current
->journal_info
;
7445 current
->journal_info
= NULL
;
7449 * If this errors out it's because we couldn't invalidate pagecache for
7450 * this range and we need to fallback to buffered.
7452 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
, create
))
7455 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
7462 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7463 * io. INLINE is special, and we could probably kludge it in here, but
7464 * it's still buffered so for safety lets just fall back to the generic
7467 * For COMPRESSED we _have_ to read the entire extent in so we can
7468 * decompress it, so there will be buffering required no matter what we
7469 * do, so go ahead and fallback to buffered.
7471 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7472 * to buffered IO. Don't blame me, this is the price we pay for using
7475 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
7476 em
->block_start
== EXTENT_MAP_INLINE
) {
7477 free_extent_map(em
);
7482 /* Just a good old fashioned hole, return */
7483 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
7484 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
7485 free_extent_map(em
);
7490 * We don't allocate a new extent in the following cases
7492 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7494 * 2) The extent is marked as PREALLOC. We're good to go here and can
7495 * just use the extent.
7499 len
= min(len
, em
->len
- (start
- em
->start
));
7500 lockstart
= start
+ len
;
7504 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
7505 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
7506 em
->block_start
!= EXTENT_MAP_HOLE
)) {
7508 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
7510 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7511 type
= BTRFS_ORDERED_PREALLOC
;
7513 type
= BTRFS_ORDERED_NOCOW
;
7514 len
= min(len
, em
->len
- (start
- em
->start
));
7515 block_start
= em
->block_start
+ (start
- em
->start
);
7517 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
7518 &orig_block_len
, &ram_bytes
) == 1) {
7519 if (type
== BTRFS_ORDERED_PREALLOC
) {
7520 free_extent_map(em
);
7521 em
= create_pinned_em(inode
, start
, len
,
7532 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
7533 block_start
, len
, len
, type
);
7535 free_extent_map(em
);
7543 * this will cow the extent, reset the len in case we changed
7546 len
= bh_result
->b_size
;
7547 free_extent_map(em
);
7548 em
= btrfs_new_extent_direct(inode
, start
, len
);
7553 len
= min(len
, em
->len
- (start
- em
->start
));
7555 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
7557 bh_result
->b_size
= len
;
7558 bh_result
->b_bdev
= em
->bdev
;
7559 set_buffer_mapped(bh_result
);
7561 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
7562 set_buffer_new(bh_result
);
7565 * Need to update the i_size under the extent lock so buffered
7566 * readers will get the updated i_size when we unlock.
7568 if (start
+ len
> i_size_read(inode
))
7569 i_size_write(inode
, start
+ len
);
7572 * If we have an outstanding_extents count still set then we're
7573 * within our reservation, otherwise we need to adjust our inode
7574 * counter appropriately.
7576 if (dio_data
->outstanding_extents
) {
7577 (dio_data
->outstanding_extents
)--;
7579 spin_lock(&BTRFS_I(inode
)->lock
);
7580 BTRFS_I(inode
)->outstanding_extents
++;
7581 spin_unlock(&BTRFS_I(inode
)->lock
);
7584 btrfs_free_reserved_data_space(inode
, len
);
7585 WARN_ON(dio_data
->reserve
< len
);
7586 dio_data
->reserve
-= len
;
7587 current
->journal_info
= dio_data
;
7591 * In the case of write we need to clear and unlock the entire range,
7592 * in the case of read we need to unlock only the end area that we
7593 * aren't using if there is any left over space.
7595 if (lockstart
< lockend
) {
7596 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
7597 lockend
, unlock_bits
, 1, 0,
7598 &cached_state
, GFP_NOFS
);
7600 free_extent_state(cached_state
);
7603 free_extent_map(em
);
7608 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
7609 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
7611 current
->journal_info
= dio_data
;
7615 static inline int submit_dio_repair_bio(struct inode
*inode
, struct bio
*bio
,
7616 int rw
, int mirror_num
)
7618 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7621 BUG_ON(rw
& REQ_WRITE
);
7625 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
7626 BTRFS_WQ_ENDIO_DIO_REPAIR
);
7630 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
7636 static int btrfs_check_dio_repairable(struct inode
*inode
,
7637 struct bio
*failed_bio
,
7638 struct io_failure_record
*failrec
,
7643 num_copies
= btrfs_num_copies(BTRFS_I(inode
)->root
->fs_info
,
7644 failrec
->logical
, failrec
->len
);
7645 if (num_copies
== 1) {
7647 * we only have a single copy of the data, so don't bother with
7648 * all the retry and error correction code that follows. no
7649 * matter what the error is, it is very likely to persist.
7651 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7652 num_copies
, failrec
->this_mirror
, failed_mirror
);
7656 failrec
->failed_mirror
= failed_mirror
;
7657 failrec
->this_mirror
++;
7658 if (failrec
->this_mirror
== failed_mirror
)
7659 failrec
->this_mirror
++;
7661 if (failrec
->this_mirror
> num_copies
) {
7662 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7663 num_copies
, failrec
->this_mirror
, failed_mirror
);
7670 static int dio_read_error(struct inode
*inode
, struct bio
*failed_bio
,
7671 struct page
*page
, u64 start
, u64 end
,
7672 int failed_mirror
, bio_end_io_t
*repair_endio
,
7675 struct io_failure_record
*failrec
;
7681 BUG_ON(failed_bio
->bi_rw
& REQ_WRITE
);
7683 ret
= btrfs_get_io_failure_record(inode
, start
, end
, &failrec
);
7687 ret
= btrfs_check_dio_repairable(inode
, failed_bio
, failrec
,
7690 free_io_failure(inode
, failrec
);
7694 if (failed_bio
->bi_vcnt
> 1)
7695 read_mode
= READ_SYNC
| REQ_FAILFAST_DEV
;
7697 read_mode
= READ_SYNC
;
7699 isector
= start
- btrfs_io_bio(failed_bio
)->logical
;
7700 isector
>>= inode
->i_sb
->s_blocksize_bits
;
7701 bio
= btrfs_create_repair_bio(inode
, failed_bio
, failrec
, page
,
7702 0, isector
, repair_endio
, repair_arg
);
7704 free_io_failure(inode
, failrec
);
7708 btrfs_debug(BTRFS_I(inode
)->root
->fs_info
,
7709 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7710 read_mode
, failrec
->this_mirror
, failrec
->in_validation
);
7712 ret
= submit_dio_repair_bio(inode
, bio
, read_mode
,
7713 failrec
->this_mirror
);
7715 free_io_failure(inode
, failrec
);
7722 struct btrfs_retry_complete
{
7723 struct completion done
;
7724 struct inode
*inode
;
7729 static void btrfs_retry_endio_nocsum(struct bio
*bio
)
7731 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7732 struct bio_vec
*bvec
;
7739 bio_for_each_segment_all(bvec
, bio
, i
)
7740 clean_io_failure(done
->inode
, done
->start
, bvec
->bv_page
, 0);
7742 complete(&done
->done
);
7746 static int __btrfs_correct_data_nocsum(struct inode
*inode
,
7747 struct btrfs_io_bio
*io_bio
)
7749 struct bio_vec
*bvec
;
7750 struct btrfs_retry_complete done
;
7755 start
= io_bio
->logical
;
7758 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7762 init_completion(&done
.done
);
7764 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7765 start
+ bvec
->bv_len
- 1,
7767 btrfs_retry_endio_nocsum
, &done
);
7771 wait_for_completion(&done
.done
);
7773 if (!done
.uptodate
) {
7774 /* We might have another mirror, so try again */
7778 start
+= bvec
->bv_len
;
7784 static void btrfs_retry_endio(struct bio
*bio
)
7786 struct btrfs_retry_complete
*done
= bio
->bi_private
;
7787 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7788 struct bio_vec
*bvec
;
7797 bio_for_each_segment_all(bvec
, bio
, i
) {
7798 ret
= __readpage_endio_check(done
->inode
, io_bio
, i
,
7800 done
->start
, bvec
->bv_len
);
7802 clean_io_failure(done
->inode
, done
->start
,
7808 done
->uptodate
= uptodate
;
7810 complete(&done
->done
);
7814 static int __btrfs_subio_endio_read(struct inode
*inode
,
7815 struct btrfs_io_bio
*io_bio
, int err
)
7817 struct bio_vec
*bvec
;
7818 struct btrfs_retry_complete done
;
7825 start
= io_bio
->logical
;
7828 bio_for_each_segment_all(bvec
, &io_bio
->bio
, i
) {
7829 ret
= __readpage_endio_check(inode
, io_bio
, i
, bvec
->bv_page
,
7830 0, start
, bvec
->bv_len
);
7836 init_completion(&done
.done
);
7838 ret
= dio_read_error(inode
, &io_bio
->bio
, bvec
->bv_page
, start
,
7839 start
+ bvec
->bv_len
- 1,
7841 btrfs_retry_endio
, &done
);
7847 wait_for_completion(&done
.done
);
7849 if (!done
.uptodate
) {
7850 /* We might have another mirror, so try again */
7854 offset
+= bvec
->bv_len
;
7855 start
+= bvec
->bv_len
;
7861 static int btrfs_subio_endio_read(struct inode
*inode
,
7862 struct btrfs_io_bio
*io_bio
, int err
)
7864 bool skip_csum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
7868 return __btrfs_correct_data_nocsum(inode
, io_bio
);
7872 return __btrfs_subio_endio_read(inode
, io_bio
, err
);
7876 static void btrfs_endio_direct_read(struct bio
*bio
)
7878 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7879 struct inode
*inode
= dip
->inode
;
7880 struct bio
*dio_bio
;
7881 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
7882 int err
= bio
->bi_error
;
7884 if (dip
->flags
& BTRFS_DIO_ORIG_BIO_SUBMITTED
)
7885 err
= btrfs_subio_endio_read(inode
, io_bio
, err
);
7887 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
7888 dip
->logical_offset
+ dip
->bytes
- 1);
7889 dio_bio
= dip
->dio_bio
;
7893 dio_end_io(dio_bio
, bio
->bi_error
);
7896 io_bio
->end_io(io_bio
, err
);
7900 static void btrfs_endio_direct_write(struct bio
*bio
)
7902 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7903 struct inode
*inode
= dip
->inode
;
7904 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7905 struct btrfs_ordered_extent
*ordered
= NULL
;
7906 u64 ordered_offset
= dip
->logical_offset
;
7907 u64 ordered_bytes
= dip
->bytes
;
7908 struct bio
*dio_bio
;
7912 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
7919 btrfs_init_work(&ordered
->work
, btrfs_endio_write_helper
,
7920 finish_ordered_fn
, NULL
, NULL
);
7921 btrfs_queue_work(root
->fs_info
->endio_write_workers
,
7925 * our bio might span multiple ordered extents. If we haven't
7926 * completed the accounting for the whole dio, go back and try again
7928 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
7929 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
7934 dio_bio
= dip
->dio_bio
;
7938 dio_end_io(dio_bio
, bio
->bi_error
);
7942 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
7943 struct bio
*bio
, int mirror_num
,
7944 unsigned long bio_flags
, u64 offset
)
7947 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7948 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
7949 BUG_ON(ret
); /* -ENOMEM */
7953 static void btrfs_end_dio_bio(struct bio
*bio
)
7955 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7956 int err
= bio
->bi_error
;
7959 btrfs_warn(BTRFS_I(dip
->inode
)->root
->fs_info
,
7960 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7961 btrfs_ino(dip
->inode
), bio
->bi_rw
,
7962 (unsigned long long)bio
->bi_iter
.bi_sector
,
7963 bio
->bi_iter
.bi_size
, err
);
7965 if (dip
->subio_endio
)
7966 err
= dip
->subio_endio(dip
->inode
, btrfs_io_bio(bio
), err
);
7972 * before atomic variable goto zero, we must make sure
7973 * dip->errors is perceived to be set.
7975 smp_mb__before_atomic();
7978 /* if there are more bios still pending for this dio, just exit */
7979 if (!atomic_dec_and_test(&dip
->pending_bios
))
7983 bio_io_error(dip
->orig_bio
);
7985 dip
->dio_bio
->bi_error
= 0;
7986 bio_endio(dip
->orig_bio
);
7992 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
7993 u64 first_sector
, gfp_t gfp_flags
)
7996 bio
= btrfs_bio_alloc(bdev
, first_sector
, BIO_MAX_PAGES
, gfp_flags
);
7998 bio_associate_current(bio
);
8002 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root
*root
,
8003 struct inode
*inode
,
8004 struct btrfs_dio_private
*dip
,
8008 struct btrfs_io_bio
*io_bio
= btrfs_io_bio(bio
);
8009 struct btrfs_io_bio
*orig_io_bio
= btrfs_io_bio(dip
->orig_bio
);
8013 * We load all the csum data we need when we submit
8014 * the first bio to reduce the csum tree search and
8017 if (dip
->logical_offset
== file_offset
) {
8018 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, dip
->orig_bio
,
8024 if (bio
== dip
->orig_bio
)
8027 file_offset
-= dip
->logical_offset
;
8028 file_offset
>>= inode
->i_sb
->s_blocksize_bits
;
8029 io_bio
->csum
= (u8
*)(((u32
*)orig_io_bio
->csum
) + file_offset
);
8034 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
8035 int rw
, u64 file_offset
, int skip_sum
,
8038 struct btrfs_dio_private
*dip
= bio
->bi_private
;
8039 int write
= rw
& REQ_WRITE
;
8040 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8044 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
8049 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
,
8050 BTRFS_WQ_ENDIO_DATA
);
8058 if (write
&& async_submit
) {
8059 ret
= btrfs_wq_submit_bio(root
->fs_info
,
8060 inode
, rw
, bio
, 0, 0,
8062 __btrfs_submit_bio_start_direct_io
,
8063 __btrfs_submit_bio_done
);
8067 * If we aren't doing async submit, calculate the csum of the
8070 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
8074 ret
= btrfs_lookup_and_bind_dio_csum(root
, inode
, dip
, bio
,
8080 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
8086 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
8089 struct inode
*inode
= dip
->inode
;
8090 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8092 struct bio
*orig_bio
= dip
->orig_bio
;
8093 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
8094 u64 start_sector
= orig_bio
->bi_iter
.bi_sector
;
8095 u64 file_offset
= dip
->logical_offset
;
8100 int async_submit
= 0;
8102 map_length
= orig_bio
->bi_iter
.bi_size
;
8103 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
8104 &map_length
, NULL
, 0);
8108 if (map_length
>= orig_bio
->bi_iter
.bi_size
) {
8110 dip
->flags
|= BTRFS_DIO_ORIG_BIO_SUBMITTED
;
8114 /* async crcs make it difficult to collect full stripe writes. */
8115 if (btrfs_get_alloc_profile(root
, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK
)
8120 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
8124 bio
->bi_private
= dip
;
8125 bio
->bi_end_io
= btrfs_end_dio_bio
;
8126 btrfs_io_bio(bio
)->logical
= file_offset
;
8127 atomic_inc(&dip
->pending_bios
);
8129 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
8130 if (map_length
< submit_len
+ bvec
->bv_len
||
8131 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
8132 bvec
->bv_offset
) < bvec
->bv_len
) {
8134 * inc the count before we submit the bio so
8135 * we know the end IO handler won't happen before
8136 * we inc the count. Otherwise, the dip might get freed
8137 * before we're done setting it up
8139 atomic_inc(&dip
->pending_bios
);
8140 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
8141 file_offset
, skip_sum
,
8145 atomic_dec(&dip
->pending_bios
);
8149 start_sector
+= submit_len
>> 9;
8150 file_offset
+= submit_len
;
8155 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
8156 start_sector
, GFP_NOFS
);
8159 bio
->bi_private
= dip
;
8160 bio
->bi_end_io
= btrfs_end_dio_bio
;
8161 btrfs_io_bio(bio
)->logical
= file_offset
;
8163 map_length
= orig_bio
->bi_iter
.bi_size
;
8164 ret
= btrfs_map_block(root
->fs_info
, rw
,
8166 &map_length
, NULL
, 0);
8172 submit_len
+= bvec
->bv_len
;
8179 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
8188 * before atomic variable goto zero, we must
8189 * make sure dip->errors is perceived to be set.
8191 smp_mb__before_atomic();
8192 if (atomic_dec_and_test(&dip
->pending_bios
))
8193 bio_io_error(dip
->orig_bio
);
8195 /* bio_end_io() will handle error, so we needn't return it */
8199 static void btrfs_submit_direct(int rw
, struct bio
*dio_bio
,
8200 struct inode
*inode
, loff_t file_offset
)
8202 struct btrfs_dio_private
*dip
= NULL
;
8203 struct bio
*io_bio
= NULL
;
8204 struct btrfs_io_bio
*btrfs_bio
;
8206 int write
= rw
& REQ_WRITE
;
8209 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
8211 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
8217 dip
= kzalloc(sizeof(*dip
), GFP_NOFS
);
8223 dip
->private = dio_bio
->bi_private
;
8225 dip
->logical_offset
= file_offset
;
8226 dip
->bytes
= dio_bio
->bi_iter
.bi_size
;
8227 dip
->disk_bytenr
= (u64
)dio_bio
->bi_iter
.bi_sector
<< 9;
8228 io_bio
->bi_private
= dip
;
8229 dip
->orig_bio
= io_bio
;
8230 dip
->dio_bio
= dio_bio
;
8231 atomic_set(&dip
->pending_bios
, 0);
8232 btrfs_bio
= btrfs_io_bio(io_bio
);
8233 btrfs_bio
->logical
= file_offset
;
8236 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
8238 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
8239 dip
->subio_endio
= btrfs_subio_endio_read
;
8242 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
8246 if (btrfs_bio
->end_io
)
8247 btrfs_bio
->end_io(btrfs_bio
, ret
);
8251 * If we arrived here it means either we failed to submit the dip
8252 * or we either failed to clone the dio_bio or failed to allocate the
8253 * dip. If we cloned the dio_bio and allocated the dip, we can just
8254 * call bio_endio against our io_bio so that we get proper resource
8255 * cleanup if we fail to submit the dip, otherwise, we must do the
8256 * same as btrfs_endio_direct_[write|read] because we can't call these
8257 * callbacks - they require an allocated dip and a clone of dio_bio.
8259 if (io_bio
&& dip
) {
8260 io_bio
->bi_error
= -EIO
;
8263 * The end io callbacks free our dip, do the final put on io_bio
8264 * and all the cleanup and final put for dio_bio (through
8271 struct btrfs_ordered_extent
*ordered
;
8273 ordered
= btrfs_lookup_ordered_extent(inode
,
8275 set_bit(BTRFS_ORDERED_IOERR
, &ordered
->flags
);
8277 * Decrements our ref on the ordered extent and removes
8278 * the ordered extent from the inode's ordered tree,
8279 * doing all the proper resource cleanup such as for the
8280 * reserved space and waking up any waiters for this
8281 * ordered extent (through btrfs_remove_ordered_extent).
8283 btrfs_finish_ordered_io(ordered
);
8285 unlock_extent(&BTRFS_I(inode
)->io_tree
, file_offset
,
8286 file_offset
+ dio_bio
->bi_iter
.bi_size
- 1);
8288 dio_bio
->bi_error
= -EIO
;
8290 * Releases and cleans up our dio_bio, no need to bio_put()
8291 * nor bio_endio()/bio_io_error() against dio_bio.
8293 dio_end_io(dio_bio
, ret
);
8300 static ssize_t
check_direct_IO(struct btrfs_root
*root
, struct kiocb
*iocb
,
8301 const struct iov_iter
*iter
, loff_t offset
)
8305 unsigned blocksize_mask
= root
->sectorsize
- 1;
8306 ssize_t retval
= -EINVAL
;
8308 if (offset
& blocksize_mask
)
8311 if (iov_iter_alignment(iter
) & blocksize_mask
)
8314 /* If this is a write we don't need to check anymore */
8315 if (iov_iter_rw(iter
) == WRITE
)
8318 * Check to make sure we don't have duplicate iov_base's in this
8319 * iovec, if so return EINVAL, otherwise we'll get csum errors
8320 * when reading back.
8322 for (seg
= 0; seg
< iter
->nr_segs
; seg
++) {
8323 for (i
= seg
+ 1; i
< iter
->nr_segs
; i
++) {
8324 if (iter
->iov
[seg
].iov_base
== iter
->iov
[i
].iov_base
)
8333 static ssize_t
btrfs_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
,
8336 struct file
*file
= iocb
->ki_filp
;
8337 struct inode
*inode
= file
->f_mapping
->host
;
8338 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8339 struct btrfs_dio_data dio_data
= { 0 };
8343 bool relock
= false;
8346 if (check_direct_IO(BTRFS_I(inode
)->root
, iocb
, iter
, offset
))
8349 inode_dio_begin(inode
);
8350 smp_mb__after_atomic();
8353 * The generic stuff only does filemap_write_and_wait_range, which
8354 * isn't enough if we've written compressed pages to this area, so
8355 * we need to flush the dirty pages again to make absolutely sure
8356 * that any outstanding dirty pages are on disk.
8358 count
= iov_iter_count(iter
);
8359 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8360 &BTRFS_I(inode
)->runtime_flags
))
8361 filemap_fdatawrite_range(inode
->i_mapping
, offset
,
8362 offset
+ count
- 1);
8364 if (iov_iter_rw(iter
) == WRITE
) {
8366 * If the write DIO is beyond the EOF, we need update
8367 * the isize, but it is protected by i_mutex. So we can
8368 * not unlock the i_mutex at this case.
8370 if (offset
+ count
<= inode
->i_size
) {
8371 mutex_unlock(&inode
->i_mutex
);
8374 ret
= btrfs_delalloc_reserve_space(inode
, count
);
8377 dio_data
.outstanding_extents
= div64_u64(count
+
8378 BTRFS_MAX_EXTENT_SIZE
- 1,
8379 BTRFS_MAX_EXTENT_SIZE
);
8382 * We need to know how many extents we reserved so that we can
8383 * do the accounting properly if we go over the number we
8384 * originally calculated. Abuse current->journal_info for this.
8386 dio_data
.reserve
= round_up(count
, root
->sectorsize
);
8387 current
->journal_info
= &dio_data
;
8388 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
8389 &BTRFS_I(inode
)->runtime_flags
)) {
8390 inode_dio_end(inode
);
8391 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
8395 ret
= __blockdev_direct_IO(iocb
, inode
,
8396 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
8397 iter
, offset
, btrfs_get_blocks_direct
, NULL
,
8398 btrfs_submit_direct
, flags
);
8399 if (iov_iter_rw(iter
) == WRITE
) {
8400 current
->journal_info
= NULL
;
8401 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
8402 if (dio_data
.reserve
)
8403 btrfs_delalloc_release_space(inode
,
8405 } else if (ret
>= 0 && (size_t)ret
< count
)
8406 btrfs_delalloc_release_space(inode
,
8407 count
- (size_t)ret
);
8411 inode_dio_end(inode
);
8413 mutex_lock(&inode
->i_mutex
);
8418 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8420 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
8421 __u64 start
, __u64 len
)
8425 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
8429 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
8432 int btrfs_readpage(struct file
*file
, struct page
*page
)
8434 struct extent_io_tree
*tree
;
8435 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8436 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
8439 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
8441 struct extent_io_tree
*tree
;
8444 if (current
->flags
& PF_MEMALLOC
) {
8445 redirty_page_for_writepage(wbc
, page
);
8449 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8450 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
8453 static int btrfs_writepages(struct address_space
*mapping
,
8454 struct writeback_control
*wbc
)
8456 struct extent_io_tree
*tree
;
8458 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8459 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
8463 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
8464 struct list_head
*pages
, unsigned nr_pages
)
8466 struct extent_io_tree
*tree
;
8467 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
8468 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
8471 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8473 struct extent_io_tree
*tree
;
8474 struct extent_map_tree
*map
;
8477 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
8478 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
8479 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
8481 ClearPagePrivate(page
);
8482 set_page_private(page
, 0);
8483 page_cache_release(page
);
8488 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
8490 if (PageWriteback(page
) || PageDirty(page
))
8492 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
8495 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
8496 unsigned int length
)
8498 struct inode
*inode
= page
->mapping
->host
;
8499 struct extent_io_tree
*tree
;
8500 struct btrfs_ordered_extent
*ordered
;
8501 struct extent_state
*cached_state
= NULL
;
8502 u64 page_start
= page_offset(page
);
8503 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8504 int inode_evicting
= inode
->i_state
& I_FREEING
;
8507 * we have the page locked, so new writeback can't start,
8508 * and the dirty bit won't be cleared while we are here.
8510 * Wait for IO on this page so that we can safely clear
8511 * the PagePrivate2 bit and do ordered accounting
8513 wait_on_page_writeback(page
);
8515 tree
= &BTRFS_I(inode
)->io_tree
;
8517 btrfs_releasepage(page
, GFP_NOFS
);
8521 if (!inode_evicting
)
8522 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
8523 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8526 * IO on this page will never be started, so we need
8527 * to account for any ordered extents now
8529 if (!inode_evicting
)
8530 clear_extent_bit(tree
, page_start
, page_end
,
8531 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8532 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
8533 EXTENT_DEFRAG
, 1, 0, &cached_state
,
8536 * whoever cleared the private bit is responsible
8537 * for the finish_ordered_io
8539 if (TestClearPagePrivate2(page
)) {
8540 struct btrfs_ordered_inode_tree
*tree
;
8543 tree
= &BTRFS_I(inode
)->ordered_tree
;
8545 spin_lock_irq(&tree
->lock
);
8546 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
8547 new_len
= page_start
- ordered
->file_offset
;
8548 if (new_len
< ordered
->truncated_len
)
8549 ordered
->truncated_len
= new_len
;
8550 spin_unlock_irq(&tree
->lock
);
8552 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
8554 PAGE_CACHE_SIZE
, 1))
8555 btrfs_finish_ordered_io(ordered
);
8557 btrfs_put_ordered_extent(ordered
);
8558 if (!inode_evicting
) {
8559 cached_state
= NULL
;
8560 lock_extent_bits(tree
, page_start
, page_end
, 0,
8565 if (!inode_evicting
) {
8566 clear_extent_bit(tree
, page_start
, page_end
,
8567 EXTENT_LOCKED
| EXTENT_DIRTY
|
8568 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
8569 EXTENT_DEFRAG
, 1, 1,
8570 &cached_state
, GFP_NOFS
);
8572 __btrfs_releasepage(page
, GFP_NOFS
);
8575 ClearPageChecked(page
);
8576 if (PagePrivate(page
)) {
8577 ClearPagePrivate(page
);
8578 set_page_private(page
, 0);
8579 page_cache_release(page
);
8584 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8585 * called from a page fault handler when a page is first dirtied. Hence we must
8586 * be careful to check for EOF conditions here. We set the page up correctly
8587 * for a written page which means we get ENOSPC checking when writing into
8588 * holes and correct delalloc and unwritten extent mapping on filesystems that
8589 * support these features.
8591 * We are not allowed to take the i_mutex here so we have to play games to
8592 * protect against truncate races as the page could now be beyond EOF. Because
8593 * vmtruncate() writes the inode size before removing pages, once we have the
8594 * page lock we can determine safely if the page is beyond EOF. If it is not
8595 * beyond EOF, then the page is guaranteed safe against truncation until we
8598 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
8600 struct page
*page
= vmf
->page
;
8601 struct inode
*inode
= file_inode(vma
->vm_file
);
8602 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8603 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
8604 struct btrfs_ordered_extent
*ordered
;
8605 struct extent_state
*cached_state
= NULL
;
8607 unsigned long zero_start
;
8614 sb_start_pagefault(inode
->i_sb
);
8615 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
8617 ret
= file_update_time(vma
->vm_file
);
8623 else /* -ENOSPC, -EIO, etc */
8624 ret
= VM_FAULT_SIGBUS
;
8630 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
8633 size
= i_size_read(inode
);
8634 page_start
= page_offset(page
);
8635 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
8637 if ((page
->mapping
!= inode
->i_mapping
) ||
8638 (page_start
>= size
)) {
8639 /* page got truncated out from underneath us */
8642 wait_on_page_writeback(page
);
8644 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
8645 set_page_extent_mapped(page
);
8648 * we can't set the delalloc bits if there are pending ordered
8649 * extents. Drop our locks and wait for them to finish
8651 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
8653 unlock_extent_cached(io_tree
, page_start
, page_end
,
8654 &cached_state
, GFP_NOFS
);
8656 btrfs_start_ordered_extent(inode
, ordered
, 1);
8657 btrfs_put_ordered_extent(ordered
);
8662 * XXX - page_mkwrite gets called every time the page is dirtied, even
8663 * if it was already dirty, so for space accounting reasons we need to
8664 * clear any delalloc bits for the range we are fixing to save. There
8665 * is probably a better way to do this, but for now keep consistent with
8666 * prepare_pages in the normal write path.
8668 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
8669 EXTENT_DIRTY
| EXTENT_DELALLOC
|
8670 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
8671 0, 0, &cached_state
, GFP_NOFS
);
8673 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
8676 unlock_extent_cached(io_tree
, page_start
, page_end
,
8677 &cached_state
, GFP_NOFS
);
8678 ret
= VM_FAULT_SIGBUS
;
8683 /* page is wholly or partially inside EOF */
8684 if (page_start
+ PAGE_CACHE_SIZE
> size
)
8685 zero_start
= size
& ~PAGE_CACHE_MASK
;
8687 zero_start
= PAGE_CACHE_SIZE
;
8689 if (zero_start
!= PAGE_CACHE_SIZE
) {
8691 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
8692 flush_dcache_page(page
);
8695 ClearPageChecked(page
);
8696 set_page_dirty(page
);
8697 SetPageUptodate(page
);
8699 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
8700 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
8701 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
8703 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
8707 sb_end_pagefault(inode
->i_sb
);
8708 return VM_FAULT_LOCKED
;
8712 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
8714 sb_end_pagefault(inode
->i_sb
);
8718 static int btrfs_truncate(struct inode
*inode
)
8720 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8721 struct btrfs_block_rsv
*rsv
;
8724 struct btrfs_trans_handle
*trans
;
8725 u64 mask
= root
->sectorsize
- 1;
8726 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
8728 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
8734 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8735 * 3 things going on here
8737 * 1) We need to reserve space for our orphan item and the space to
8738 * delete our orphan item. Lord knows we don't want to have a dangling
8739 * orphan item because we didn't reserve space to remove it.
8741 * 2) We need to reserve space to update our inode.
8743 * 3) We need to have something to cache all the space that is going to
8744 * be free'd up by the truncate operation, but also have some slack
8745 * space reserved in case it uses space during the truncate (thank you
8746 * very much snapshotting).
8748 * And we need these to all be seperate. The fact is we can use alot of
8749 * space doing the truncate, and we have no earthly idea how much space
8750 * we will use, so we need the truncate reservation to be seperate so it
8751 * doesn't end up using space reserved for updating the inode or
8752 * removing the orphan item. We also need to be able to stop the
8753 * transaction and start a new one, which means we need to be able to
8754 * update the inode several times, and we have no idea of knowing how
8755 * many times that will be, so we can't just reserve 1 item for the
8756 * entirety of the opration, so that has to be done seperately as well.
8757 * Then there is the orphan item, which does indeed need to be held on
8758 * to for the whole operation, and we need nobody to touch this reserved
8759 * space except the orphan code.
8761 * So that leaves us with
8763 * 1) root->orphan_block_rsv - for the orphan deletion.
8764 * 2) rsv - for the truncate reservation, which we will steal from the
8765 * transaction reservation.
8766 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8767 * updating the inode.
8769 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
8772 rsv
->size
= min_size
;
8776 * 1 for the truncate slack space
8777 * 1 for updating the inode.
8779 trans
= btrfs_start_transaction(root
, 2);
8780 if (IS_ERR(trans
)) {
8781 err
= PTR_ERR(trans
);
8785 /* Migrate the slack space for the truncate to our reserve */
8786 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
8791 * So if we truncate and then write and fsync we normally would just
8792 * write the extents that changed, which is a problem if we need to
8793 * first truncate that entire inode. So set this flag so we write out
8794 * all of the extents in the inode to the sync log so we're completely
8797 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
8798 trans
->block_rsv
= rsv
;
8801 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
8803 BTRFS_EXTENT_DATA_KEY
);
8804 if (ret
!= -ENOSPC
&& ret
!= -EAGAIN
) {
8809 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
8810 ret
= btrfs_update_inode(trans
, root
, inode
);
8816 btrfs_end_transaction(trans
, root
);
8817 btrfs_btree_balance_dirty(root
);
8819 trans
= btrfs_start_transaction(root
, 2);
8820 if (IS_ERR(trans
)) {
8821 ret
= err
= PTR_ERR(trans
);
8826 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
8828 BUG_ON(ret
); /* shouldn't happen */
8829 trans
->block_rsv
= rsv
;
8832 if (ret
== 0 && inode
->i_nlink
> 0) {
8833 trans
->block_rsv
= root
->orphan_block_rsv
;
8834 ret
= btrfs_orphan_del(trans
, inode
);
8840 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
8841 ret
= btrfs_update_inode(trans
, root
, inode
);
8845 ret
= btrfs_end_transaction(trans
, root
);
8846 btrfs_btree_balance_dirty(root
);
8850 btrfs_free_block_rsv(root
, rsv
);
8859 * create a new subvolume directory/inode (helper for the ioctl).
8861 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
8862 struct btrfs_root
*new_root
,
8863 struct btrfs_root
*parent_root
,
8866 struct inode
*inode
;
8870 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
8871 new_dirid
, new_dirid
,
8872 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
8875 return PTR_ERR(inode
);
8876 inode
->i_op
= &btrfs_dir_inode_operations
;
8877 inode
->i_fop
= &btrfs_dir_file_operations
;
8879 set_nlink(inode
, 1);
8880 btrfs_i_size_write(inode
, 0);
8881 unlock_new_inode(inode
);
8883 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
8885 btrfs_err(new_root
->fs_info
,
8886 "error inheriting subvolume %llu properties: %d",
8887 new_root
->root_key
.objectid
, err
);
8889 err
= btrfs_update_inode(trans
, new_root
, inode
);
8895 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
8897 struct btrfs_inode
*ei
;
8898 struct inode
*inode
;
8900 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
8907 ei
->last_sub_trans
= 0;
8908 ei
->logged_trans
= 0;
8909 ei
->delalloc_bytes
= 0;
8910 ei
->defrag_bytes
= 0;
8911 ei
->disk_i_size
= 0;
8914 ei
->index_cnt
= (u64
)-1;
8916 ei
->last_unlink_trans
= 0;
8917 ei
->last_log_commit
= 0;
8919 spin_lock_init(&ei
->lock
);
8920 ei
->outstanding_extents
= 0;
8921 ei
->reserved_extents
= 0;
8923 ei
->runtime_flags
= 0;
8924 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
8926 ei
->delayed_node
= NULL
;
8928 ei
->i_otime
.tv_sec
= 0;
8929 ei
->i_otime
.tv_nsec
= 0;
8931 inode
= &ei
->vfs_inode
;
8932 extent_map_tree_init(&ei
->extent_tree
);
8933 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
8934 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
8935 ei
->io_tree
.track_uptodate
= 1;
8936 ei
->io_failure_tree
.track_uptodate
= 1;
8937 atomic_set(&ei
->sync_writers
, 0);
8938 mutex_init(&ei
->log_mutex
);
8939 mutex_init(&ei
->delalloc_mutex
);
8940 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
8941 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
8942 RB_CLEAR_NODE(&ei
->rb_node
);
8947 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8948 void btrfs_test_destroy_inode(struct inode
*inode
)
8950 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
8951 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
8955 static void btrfs_i_callback(struct rcu_head
*head
)
8957 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
8958 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
8961 void btrfs_destroy_inode(struct inode
*inode
)
8963 struct btrfs_ordered_extent
*ordered
;
8964 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8966 WARN_ON(!hlist_empty(&inode
->i_dentry
));
8967 WARN_ON(inode
->i_data
.nrpages
);
8968 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
8969 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
8970 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
8971 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
8972 WARN_ON(BTRFS_I(inode
)->defrag_bytes
);
8975 * This can happen where we create an inode, but somebody else also
8976 * created the same inode and we need to destroy the one we already
8982 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
8983 &BTRFS_I(inode
)->runtime_flags
)) {
8984 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
8986 atomic_dec(&root
->orphan_inodes
);
8990 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
8994 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
8995 ordered
->file_offset
, ordered
->len
);
8996 btrfs_remove_ordered_extent(inode
, ordered
);
8997 btrfs_put_ordered_extent(ordered
);
8998 btrfs_put_ordered_extent(ordered
);
9001 inode_tree_del(inode
);
9002 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
9004 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
9007 int btrfs_drop_inode(struct inode
*inode
)
9009 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9014 /* the snap/subvol tree is on deleting */
9015 if (btrfs_root_refs(&root
->root_item
) == 0)
9018 return generic_drop_inode(inode
);
9021 static void init_once(void *foo
)
9023 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
9025 inode_init_once(&ei
->vfs_inode
);
9028 void btrfs_destroy_cachep(void)
9031 * Make sure all delayed rcu free inodes are flushed before we
9035 if (btrfs_inode_cachep
)
9036 kmem_cache_destroy(btrfs_inode_cachep
);
9037 if (btrfs_trans_handle_cachep
)
9038 kmem_cache_destroy(btrfs_trans_handle_cachep
);
9039 if (btrfs_transaction_cachep
)
9040 kmem_cache_destroy(btrfs_transaction_cachep
);
9041 if (btrfs_path_cachep
)
9042 kmem_cache_destroy(btrfs_path_cachep
);
9043 if (btrfs_free_space_cachep
)
9044 kmem_cache_destroy(btrfs_free_space_cachep
);
9045 if (btrfs_delalloc_work_cachep
)
9046 kmem_cache_destroy(btrfs_delalloc_work_cachep
);
9049 int btrfs_init_cachep(void)
9051 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
9052 sizeof(struct btrfs_inode
), 0,
9053 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
9054 if (!btrfs_inode_cachep
)
9057 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
9058 sizeof(struct btrfs_trans_handle
), 0,
9059 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9060 if (!btrfs_trans_handle_cachep
)
9063 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
9064 sizeof(struct btrfs_transaction
), 0,
9065 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9066 if (!btrfs_transaction_cachep
)
9069 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
9070 sizeof(struct btrfs_path
), 0,
9071 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9072 if (!btrfs_path_cachep
)
9075 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
9076 sizeof(struct btrfs_free_space
), 0,
9077 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
9078 if (!btrfs_free_space_cachep
)
9081 btrfs_delalloc_work_cachep
= kmem_cache_create("btrfs_delalloc_work",
9082 sizeof(struct btrfs_delalloc_work
), 0,
9083 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
9085 if (!btrfs_delalloc_work_cachep
)
9090 btrfs_destroy_cachep();
9094 static int btrfs_getattr(struct vfsmount
*mnt
,
9095 struct dentry
*dentry
, struct kstat
*stat
)
9098 struct inode
*inode
= d_inode(dentry
);
9099 u32 blocksize
= inode
->i_sb
->s_blocksize
;
9101 generic_fillattr(inode
, stat
);
9102 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
9103 stat
->blksize
= PAGE_CACHE_SIZE
;
9105 spin_lock(&BTRFS_I(inode
)->lock
);
9106 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
9107 spin_unlock(&BTRFS_I(inode
)->lock
);
9108 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
9109 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
9113 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
9114 struct inode
*new_dir
, struct dentry
*new_dentry
)
9116 struct btrfs_trans_handle
*trans
;
9117 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
9118 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
9119 struct inode
*new_inode
= d_inode(new_dentry
);
9120 struct inode
*old_inode
= d_inode(old_dentry
);
9121 struct timespec ctime
= CURRENT_TIME
;
9125 u64 old_ino
= btrfs_ino(old_inode
);
9127 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
9130 /* we only allow rename subvolume link between subvolumes */
9131 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
9134 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
9135 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
9138 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
9139 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
9143 /* check for collisions, even if the name isn't there */
9144 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
9145 new_dentry
->d_name
.name
,
9146 new_dentry
->d_name
.len
);
9149 if (ret
== -EEXIST
) {
9151 * eexist without a new_inode */
9152 if (WARN_ON(!new_inode
)) {
9156 /* maybe -EOVERFLOW */
9163 * we're using rename to replace one file with another. Start IO on it
9164 * now so we don't add too much work to the end of the transaction
9166 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
)
9167 filemap_flush(old_inode
->i_mapping
);
9169 /* close the racy window with snapshot create/destroy ioctl */
9170 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9171 down_read(&root
->fs_info
->subvol_sem
);
9173 * We want to reserve the absolute worst case amount of items. So if
9174 * both inodes are subvols and we need to unlink them then that would
9175 * require 4 item modifications, but if they are both normal inodes it
9176 * would require 5 item modifications, so we'll assume their normal
9177 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9178 * should cover the worst case number of items we'll modify.
9180 trans
= btrfs_start_transaction(root
, 11);
9181 if (IS_ERR(trans
)) {
9182 ret
= PTR_ERR(trans
);
9187 btrfs_record_root_in_trans(trans
, dest
);
9189 ret
= btrfs_set_inode_index(new_dir
, &index
);
9193 BTRFS_I(old_inode
)->dir_index
= 0ULL;
9194 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9195 /* force full log commit if subvolume involved. */
9196 btrfs_set_log_full_commit(root
->fs_info
, trans
);
9198 ret
= btrfs_insert_inode_ref(trans
, dest
,
9199 new_dentry
->d_name
.name
,
9200 new_dentry
->d_name
.len
,
9202 btrfs_ino(new_dir
), index
);
9206 * this is an ugly little race, but the rename is required
9207 * to make sure that if we crash, the inode is either at the
9208 * old name or the new one. pinning the log transaction lets
9209 * us make sure we don't allow a log commit to come in after
9210 * we unlink the name but before we add the new name back in.
9212 btrfs_pin_log_trans(root
);
9215 inode_inc_iversion(old_dir
);
9216 inode_inc_iversion(new_dir
);
9217 inode_inc_iversion(old_inode
);
9218 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
9219 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
9220 old_inode
->i_ctime
= ctime
;
9222 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
9223 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
9225 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
9226 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
9227 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
9228 old_dentry
->d_name
.name
,
9229 old_dentry
->d_name
.len
);
9231 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
9232 d_inode(old_dentry
),
9233 old_dentry
->d_name
.name
,
9234 old_dentry
->d_name
.len
);
9236 ret
= btrfs_update_inode(trans
, root
, old_inode
);
9239 btrfs_abort_transaction(trans
, root
, ret
);
9244 inode_inc_iversion(new_inode
);
9245 new_inode
->i_ctime
= CURRENT_TIME
;
9246 if (unlikely(btrfs_ino(new_inode
) ==
9247 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
9248 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
9249 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
9251 new_dentry
->d_name
.name
,
9252 new_dentry
->d_name
.len
);
9253 BUG_ON(new_inode
->i_nlink
== 0);
9255 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
9256 d_inode(new_dentry
),
9257 new_dentry
->d_name
.name
,
9258 new_dentry
->d_name
.len
);
9260 if (!ret
&& new_inode
->i_nlink
== 0)
9261 ret
= btrfs_orphan_add(trans
, d_inode(new_dentry
));
9263 btrfs_abort_transaction(trans
, root
, ret
);
9268 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
9269 new_dentry
->d_name
.name
,
9270 new_dentry
->d_name
.len
, 0, index
);
9272 btrfs_abort_transaction(trans
, root
, ret
);
9276 if (old_inode
->i_nlink
== 1)
9277 BTRFS_I(old_inode
)->dir_index
= index
;
9279 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
9280 struct dentry
*parent
= new_dentry
->d_parent
;
9281 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
9282 btrfs_end_log_trans(root
);
9285 btrfs_end_transaction(trans
, root
);
9287 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
9288 up_read(&root
->fs_info
->subvol_sem
);
9293 static int btrfs_rename2(struct inode
*old_dir
, struct dentry
*old_dentry
,
9294 struct inode
*new_dir
, struct dentry
*new_dentry
,
9297 if (flags
& ~RENAME_NOREPLACE
)
9300 return btrfs_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
9303 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
9305 struct btrfs_delalloc_work
*delalloc_work
;
9306 struct inode
*inode
;
9308 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
9310 inode
= delalloc_work
->inode
;
9311 if (delalloc_work
->wait
) {
9312 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
9314 filemap_flush(inode
->i_mapping
);
9315 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
9316 &BTRFS_I(inode
)->runtime_flags
))
9317 filemap_flush(inode
->i_mapping
);
9320 if (delalloc_work
->delay_iput
)
9321 btrfs_add_delayed_iput(inode
);
9324 complete(&delalloc_work
->completion
);
9327 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
9328 int wait
, int delay_iput
)
9330 struct btrfs_delalloc_work
*work
;
9332 work
= kmem_cache_zalloc(btrfs_delalloc_work_cachep
, GFP_NOFS
);
9336 init_completion(&work
->completion
);
9337 INIT_LIST_HEAD(&work
->list
);
9338 work
->inode
= inode
;
9340 work
->delay_iput
= delay_iput
;
9341 WARN_ON_ONCE(!inode
);
9342 btrfs_init_work(&work
->work
, btrfs_flush_delalloc_helper
,
9343 btrfs_run_delalloc_work
, NULL
, NULL
);
9348 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
9350 wait_for_completion(&work
->completion
);
9351 kmem_cache_free(btrfs_delalloc_work_cachep
, work
);
9355 * some fairly slow code that needs optimization. This walks the list
9356 * of all the inodes with pending delalloc and forces them to disk.
9358 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
,
9361 struct btrfs_inode
*binode
;
9362 struct inode
*inode
;
9363 struct btrfs_delalloc_work
*work
, *next
;
9364 struct list_head works
;
9365 struct list_head splice
;
9368 INIT_LIST_HEAD(&works
);
9369 INIT_LIST_HEAD(&splice
);
9371 mutex_lock(&root
->delalloc_mutex
);
9372 spin_lock(&root
->delalloc_lock
);
9373 list_splice_init(&root
->delalloc_inodes
, &splice
);
9374 while (!list_empty(&splice
)) {
9375 binode
= list_entry(splice
.next
, struct btrfs_inode
,
9378 list_move_tail(&binode
->delalloc_inodes
,
9379 &root
->delalloc_inodes
);
9380 inode
= igrab(&binode
->vfs_inode
);
9382 cond_resched_lock(&root
->delalloc_lock
);
9385 spin_unlock(&root
->delalloc_lock
);
9387 work
= btrfs_alloc_delalloc_work(inode
, 0, delay_iput
);
9390 btrfs_add_delayed_iput(inode
);
9396 list_add_tail(&work
->list
, &works
);
9397 btrfs_queue_work(root
->fs_info
->flush_workers
,
9400 if (nr
!= -1 && ret
>= nr
)
9403 spin_lock(&root
->delalloc_lock
);
9405 spin_unlock(&root
->delalloc_lock
);
9408 list_for_each_entry_safe(work
, next
, &works
, list
) {
9409 list_del_init(&work
->list
);
9410 btrfs_wait_and_free_delalloc_work(work
);
9413 if (!list_empty_careful(&splice
)) {
9414 spin_lock(&root
->delalloc_lock
);
9415 list_splice_tail(&splice
, &root
->delalloc_inodes
);
9416 spin_unlock(&root
->delalloc_lock
);
9418 mutex_unlock(&root
->delalloc_mutex
);
9422 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
9426 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
9429 ret
= __start_delalloc_inodes(root
, delay_iput
, -1);
9433 * the filemap_flush will queue IO into the worker threads, but
9434 * we have to make sure the IO is actually started and that
9435 * ordered extents get created before we return
9437 atomic_inc(&root
->fs_info
->async_submit_draining
);
9438 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
9439 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
9440 wait_event(root
->fs_info
->async_submit_wait
,
9441 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
9442 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
9444 atomic_dec(&root
->fs_info
->async_submit_draining
);
9448 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
,
9451 struct btrfs_root
*root
;
9452 struct list_head splice
;
9455 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
9458 INIT_LIST_HEAD(&splice
);
9460 mutex_lock(&fs_info
->delalloc_root_mutex
);
9461 spin_lock(&fs_info
->delalloc_root_lock
);
9462 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
9463 while (!list_empty(&splice
) && nr
) {
9464 root
= list_first_entry(&splice
, struct btrfs_root
,
9466 root
= btrfs_grab_fs_root(root
);
9468 list_move_tail(&root
->delalloc_root
,
9469 &fs_info
->delalloc_roots
);
9470 spin_unlock(&fs_info
->delalloc_root_lock
);
9472 ret
= __start_delalloc_inodes(root
, delay_iput
, nr
);
9473 btrfs_put_fs_root(root
);
9481 spin_lock(&fs_info
->delalloc_root_lock
);
9483 spin_unlock(&fs_info
->delalloc_root_lock
);
9486 atomic_inc(&fs_info
->async_submit_draining
);
9487 while (atomic_read(&fs_info
->nr_async_submits
) ||
9488 atomic_read(&fs_info
->async_delalloc_pages
)) {
9489 wait_event(fs_info
->async_submit_wait
,
9490 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
9491 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
9493 atomic_dec(&fs_info
->async_submit_draining
);
9495 if (!list_empty_careful(&splice
)) {
9496 spin_lock(&fs_info
->delalloc_root_lock
);
9497 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
9498 spin_unlock(&fs_info
->delalloc_root_lock
);
9500 mutex_unlock(&fs_info
->delalloc_root_mutex
);
9504 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
9505 const char *symname
)
9507 struct btrfs_trans_handle
*trans
;
9508 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9509 struct btrfs_path
*path
;
9510 struct btrfs_key key
;
9511 struct inode
*inode
= NULL
;
9519 struct btrfs_file_extent_item
*ei
;
9520 struct extent_buffer
*leaf
;
9522 name_len
= strlen(symname
);
9523 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
9524 return -ENAMETOOLONG
;
9527 * 2 items for inode item and ref
9528 * 2 items for dir items
9529 * 1 item for xattr if selinux is on
9531 trans
= btrfs_start_transaction(root
, 5);
9533 return PTR_ERR(trans
);
9535 err
= btrfs_find_free_ino(root
, &objectid
);
9539 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
9540 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
9541 S_IFLNK
|S_IRWXUGO
, &index
);
9542 if (IS_ERR(inode
)) {
9543 err
= PTR_ERR(inode
);
9548 * If the active LSM wants to access the inode during
9549 * d_instantiate it needs these. Smack checks to see
9550 * if the filesystem supports xattrs by looking at the
9553 inode
->i_fop
= &btrfs_file_operations
;
9554 inode
->i_op
= &btrfs_file_inode_operations
;
9555 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9556 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9558 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
9560 goto out_unlock_inode
;
9562 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
9564 goto out_unlock_inode
;
9566 path
= btrfs_alloc_path();
9569 goto out_unlock_inode
;
9571 key
.objectid
= btrfs_ino(inode
);
9573 key
.type
= BTRFS_EXTENT_DATA_KEY
;
9574 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
9575 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
9578 btrfs_free_path(path
);
9579 goto out_unlock_inode
;
9581 leaf
= path
->nodes
[0];
9582 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
9583 struct btrfs_file_extent_item
);
9584 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
9585 btrfs_set_file_extent_type(leaf
, ei
,
9586 BTRFS_FILE_EXTENT_INLINE
);
9587 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
9588 btrfs_set_file_extent_compression(leaf
, ei
, 0);
9589 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
9590 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
9592 ptr
= btrfs_file_extent_inline_start(ei
);
9593 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
9594 btrfs_mark_buffer_dirty(leaf
);
9595 btrfs_free_path(path
);
9597 inode
->i_op
= &btrfs_symlink_inode_operations
;
9598 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
9599 inode_set_bytes(inode
, name_len
);
9600 btrfs_i_size_write(inode
, name_len
);
9601 err
= btrfs_update_inode(trans
, root
, inode
);
9604 goto out_unlock_inode
;
9607 unlock_new_inode(inode
);
9608 d_instantiate(dentry
, inode
);
9611 btrfs_end_transaction(trans
, root
);
9613 inode_dec_link_count(inode
);
9616 btrfs_btree_balance_dirty(root
);
9621 unlock_new_inode(inode
);
9625 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9626 u64 start
, u64 num_bytes
, u64 min_size
,
9627 loff_t actual_len
, u64
*alloc_hint
,
9628 struct btrfs_trans_handle
*trans
)
9630 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
9631 struct extent_map
*em
;
9632 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9633 struct btrfs_key ins
;
9634 u64 cur_offset
= start
;
9638 bool own_trans
= true;
9642 while (num_bytes
> 0) {
9644 trans
= btrfs_start_transaction(root
, 3);
9645 if (IS_ERR(trans
)) {
9646 ret
= PTR_ERR(trans
);
9651 cur_bytes
= min(num_bytes
, 256ULL * 1024 * 1024);
9652 cur_bytes
= max(cur_bytes
, min_size
);
9653 ret
= btrfs_reserve_extent(root
, cur_bytes
, min_size
, 0,
9654 *alloc_hint
, &ins
, 1, 0);
9657 btrfs_end_transaction(trans
, root
);
9661 ret
= insert_reserved_file_extent(trans
, inode
,
9662 cur_offset
, ins
.objectid
,
9663 ins
.offset
, ins
.offset
,
9664 ins
.offset
, 0, 0, 0,
9665 BTRFS_FILE_EXTENT_PREALLOC
);
9667 btrfs_free_reserved_extent(root
, ins
.objectid
,
9669 btrfs_abort_transaction(trans
, root
, ret
);
9671 btrfs_end_transaction(trans
, root
);
9675 btrfs_drop_extent_cache(inode
, cur_offset
,
9676 cur_offset
+ ins
.offset
-1, 0);
9678 em
= alloc_extent_map();
9680 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
9681 &BTRFS_I(inode
)->runtime_flags
);
9685 em
->start
= cur_offset
;
9686 em
->orig_start
= cur_offset
;
9687 em
->len
= ins
.offset
;
9688 em
->block_start
= ins
.objectid
;
9689 em
->block_len
= ins
.offset
;
9690 em
->orig_block_len
= ins
.offset
;
9691 em
->ram_bytes
= ins
.offset
;
9692 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
9693 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
9694 em
->generation
= trans
->transid
;
9697 write_lock(&em_tree
->lock
);
9698 ret
= add_extent_mapping(em_tree
, em
, 1);
9699 write_unlock(&em_tree
->lock
);
9702 btrfs_drop_extent_cache(inode
, cur_offset
,
9703 cur_offset
+ ins
.offset
- 1,
9706 free_extent_map(em
);
9708 num_bytes
-= ins
.offset
;
9709 cur_offset
+= ins
.offset
;
9710 *alloc_hint
= ins
.objectid
+ ins
.offset
;
9712 inode_inc_iversion(inode
);
9713 inode
->i_ctime
= CURRENT_TIME
;
9714 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
9715 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
9716 (actual_len
> inode
->i_size
) &&
9717 (cur_offset
> inode
->i_size
)) {
9718 if (cur_offset
> actual_len
)
9719 i_size
= actual_len
;
9721 i_size
= cur_offset
;
9722 i_size_write(inode
, i_size
);
9723 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
9726 ret
= btrfs_update_inode(trans
, root
, inode
);
9729 btrfs_abort_transaction(trans
, root
, ret
);
9731 btrfs_end_transaction(trans
, root
);
9736 btrfs_end_transaction(trans
, root
);
9741 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
9742 u64 start
, u64 num_bytes
, u64 min_size
,
9743 loff_t actual_len
, u64
*alloc_hint
)
9745 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9746 min_size
, actual_len
, alloc_hint
,
9750 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
9751 struct btrfs_trans_handle
*trans
, int mode
,
9752 u64 start
, u64 num_bytes
, u64 min_size
,
9753 loff_t actual_len
, u64
*alloc_hint
)
9755 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
9756 min_size
, actual_len
, alloc_hint
, trans
);
9759 static int btrfs_set_page_dirty(struct page
*page
)
9761 return __set_page_dirty_nobuffers(page
);
9764 static int btrfs_permission(struct inode
*inode
, int mask
)
9766 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
9767 umode_t mode
= inode
->i_mode
;
9769 if (mask
& MAY_WRITE
&&
9770 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
9771 if (btrfs_root_readonly(root
))
9773 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
9776 return generic_permission(inode
, mask
);
9779 static int btrfs_tmpfile(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
9781 struct btrfs_trans_handle
*trans
;
9782 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
9783 struct inode
*inode
= NULL
;
9789 * 5 units required for adding orphan entry
9791 trans
= btrfs_start_transaction(root
, 5);
9793 return PTR_ERR(trans
);
9795 ret
= btrfs_find_free_ino(root
, &objectid
);
9799 inode
= btrfs_new_inode(trans
, root
, dir
, NULL
, 0,
9800 btrfs_ino(dir
), objectid
, mode
, &index
);
9801 if (IS_ERR(inode
)) {
9802 ret
= PTR_ERR(inode
);
9807 inode
->i_fop
= &btrfs_file_operations
;
9808 inode
->i_op
= &btrfs_file_inode_operations
;
9810 inode
->i_mapping
->a_ops
= &btrfs_aops
;
9811 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
9813 ret
= btrfs_init_inode_security(trans
, inode
, dir
, NULL
);
9817 ret
= btrfs_update_inode(trans
, root
, inode
);
9820 ret
= btrfs_orphan_add(trans
, inode
);
9825 * We set number of links to 0 in btrfs_new_inode(), and here we set
9826 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9829 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9831 set_nlink(inode
, 1);
9832 unlock_new_inode(inode
);
9833 d_tmpfile(dentry
, inode
);
9834 mark_inode_dirty(inode
);
9837 btrfs_end_transaction(trans
, root
);
9840 btrfs_balance_delayed_items(root
);
9841 btrfs_btree_balance_dirty(root
);
9845 unlock_new_inode(inode
);
9850 /* Inspired by filemap_check_errors() */
9851 int btrfs_inode_check_errors(struct inode
*inode
)
9855 if (test_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
) &&
9856 test_and_clear_bit(AS_ENOSPC
, &inode
->i_mapping
->flags
))
9858 if (test_bit(AS_EIO
, &inode
->i_mapping
->flags
) &&
9859 test_and_clear_bit(AS_EIO
, &inode
->i_mapping
->flags
))
9865 static const struct inode_operations btrfs_dir_inode_operations
= {
9866 .getattr
= btrfs_getattr
,
9867 .lookup
= btrfs_lookup
,
9868 .create
= btrfs_create
,
9869 .unlink
= btrfs_unlink
,
9871 .mkdir
= btrfs_mkdir
,
9872 .rmdir
= btrfs_rmdir
,
9873 .rename2
= btrfs_rename2
,
9874 .symlink
= btrfs_symlink
,
9875 .setattr
= btrfs_setattr
,
9876 .mknod
= btrfs_mknod
,
9877 .setxattr
= btrfs_setxattr
,
9878 .getxattr
= btrfs_getxattr
,
9879 .listxattr
= btrfs_listxattr
,
9880 .removexattr
= btrfs_removexattr
,
9881 .permission
= btrfs_permission
,
9882 .get_acl
= btrfs_get_acl
,
9883 .set_acl
= btrfs_set_acl
,
9884 .update_time
= btrfs_update_time
,
9885 .tmpfile
= btrfs_tmpfile
,
9887 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
9888 .lookup
= btrfs_lookup
,
9889 .permission
= btrfs_permission
,
9890 .get_acl
= btrfs_get_acl
,
9891 .set_acl
= btrfs_set_acl
,
9892 .update_time
= btrfs_update_time
,
9895 static const struct file_operations btrfs_dir_file_operations
= {
9896 .llseek
= generic_file_llseek
,
9897 .read
= generic_read_dir
,
9898 .iterate
= btrfs_real_readdir
,
9899 .unlocked_ioctl
= btrfs_ioctl
,
9900 #ifdef CONFIG_COMPAT
9901 .compat_ioctl
= btrfs_ioctl
,
9903 .release
= btrfs_release_file
,
9904 .fsync
= btrfs_sync_file
,
9907 static struct extent_io_ops btrfs_extent_io_ops
= {
9908 .fill_delalloc
= run_delalloc_range
,
9909 .submit_bio_hook
= btrfs_submit_bio_hook
,
9910 .merge_bio_hook
= btrfs_merge_bio_hook
,
9911 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
9912 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
9913 .writepage_start_hook
= btrfs_writepage_start_hook
,
9914 .set_bit_hook
= btrfs_set_bit_hook
,
9915 .clear_bit_hook
= btrfs_clear_bit_hook
,
9916 .merge_extent_hook
= btrfs_merge_extent_hook
,
9917 .split_extent_hook
= btrfs_split_extent_hook
,
9921 * btrfs doesn't support the bmap operation because swapfiles
9922 * use bmap to make a mapping of extents in the file. They assume
9923 * these extents won't change over the life of the file and they
9924 * use the bmap result to do IO directly to the drive.
9926 * the btrfs bmap call would return logical addresses that aren't
9927 * suitable for IO and they also will change frequently as COW
9928 * operations happen. So, swapfile + btrfs == corruption.
9930 * For now we're avoiding this by dropping bmap.
9932 static const struct address_space_operations btrfs_aops
= {
9933 .readpage
= btrfs_readpage
,
9934 .writepage
= btrfs_writepage
,
9935 .writepages
= btrfs_writepages
,
9936 .readpages
= btrfs_readpages
,
9937 .direct_IO
= btrfs_direct_IO
,
9938 .invalidatepage
= btrfs_invalidatepage
,
9939 .releasepage
= btrfs_releasepage
,
9940 .set_page_dirty
= btrfs_set_page_dirty
,
9941 .error_remove_page
= generic_error_remove_page
,
9944 static const struct address_space_operations btrfs_symlink_aops
= {
9945 .readpage
= btrfs_readpage
,
9946 .writepage
= btrfs_writepage
,
9947 .invalidatepage
= btrfs_invalidatepage
,
9948 .releasepage
= btrfs_releasepage
,
9951 static const struct inode_operations btrfs_file_inode_operations
= {
9952 .getattr
= btrfs_getattr
,
9953 .setattr
= btrfs_setattr
,
9954 .setxattr
= btrfs_setxattr
,
9955 .getxattr
= btrfs_getxattr
,
9956 .listxattr
= btrfs_listxattr
,
9957 .removexattr
= btrfs_removexattr
,
9958 .permission
= btrfs_permission
,
9959 .fiemap
= btrfs_fiemap
,
9960 .get_acl
= btrfs_get_acl
,
9961 .set_acl
= btrfs_set_acl
,
9962 .update_time
= btrfs_update_time
,
9964 static const struct inode_operations btrfs_special_inode_operations
= {
9965 .getattr
= btrfs_getattr
,
9966 .setattr
= btrfs_setattr
,
9967 .permission
= btrfs_permission
,
9968 .setxattr
= btrfs_setxattr
,
9969 .getxattr
= btrfs_getxattr
,
9970 .listxattr
= btrfs_listxattr
,
9971 .removexattr
= btrfs_removexattr
,
9972 .get_acl
= btrfs_get_acl
,
9973 .set_acl
= btrfs_set_acl
,
9974 .update_time
= btrfs_update_time
,
9976 static const struct inode_operations btrfs_symlink_inode_operations
= {
9977 .readlink
= generic_readlink
,
9978 .follow_link
= page_follow_link_light
,
9979 .put_link
= page_put_link
,
9980 .getattr
= btrfs_getattr
,
9981 .setattr
= btrfs_setattr
,
9982 .permission
= btrfs_permission
,
9983 .setxattr
= btrfs_setxattr
,
9984 .getxattr
= btrfs_getxattr
,
9985 .listxattr
= btrfs_listxattr
,
9986 .removexattr
= btrfs_removexattr
,
9987 .update_time
= btrfs_update_time
,
9990 const struct dentry_operations btrfs_dentry_operations
= {
9991 .d_delete
= btrfs_dentry_delete
,
9992 .d_release
= btrfs_dentry_release
,