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>
42 #include "transaction.h"
43 #include "btrfs_inode.h"
45 #include "print-tree.h"
47 #include "ordered-data.h"
50 #include "compression.h"
53 struct btrfs_iget_args
{
55 struct btrfs_root
*root
;
58 static const struct inode_operations btrfs_dir_inode_operations
;
59 static const struct inode_operations btrfs_symlink_inode_operations
;
60 static const struct inode_operations btrfs_dir_ro_inode_operations
;
61 static const struct inode_operations btrfs_special_inode_operations
;
62 static const struct inode_operations btrfs_file_inode_operations
;
63 static const struct address_space_operations btrfs_aops
;
64 static const struct address_space_operations btrfs_symlink_aops
;
65 static const struct file_operations btrfs_dir_file_operations
;
66 static struct extent_io_ops btrfs_extent_io_ops
;
68 static struct kmem_cache
*btrfs_inode_cachep
;
69 struct kmem_cache
*btrfs_trans_handle_cachep
;
70 struct kmem_cache
*btrfs_transaction_cachep
;
71 struct kmem_cache
*btrfs_path_cachep
;
74 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
75 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
76 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
77 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
78 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
79 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
80 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
81 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
84 static void btrfs_truncate(struct inode
*inode
);
85 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
86 static noinline
int cow_file_range(struct inode
*inode
,
87 struct page
*locked_page
,
88 u64 start
, u64 end
, int *page_started
,
89 unsigned long *nr_written
, int unlock
);
91 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
92 struct inode
*inode
, struct inode
*dir
)
96 err
= btrfs_init_acl(trans
, inode
, dir
);
98 err
= btrfs_xattr_security_init(trans
, inode
, dir
);
103 * this does all the hard work for inserting an inline extent into
104 * the btree. The caller should have done a btrfs_drop_extents so that
105 * no overlapping inline items exist in the btree
107 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
108 struct btrfs_root
*root
, struct inode
*inode
,
109 u64 start
, size_t size
, size_t compressed_size
,
110 struct page
**compressed_pages
)
112 struct btrfs_key key
;
113 struct btrfs_path
*path
;
114 struct extent_buffer
*leaf
;
115 struct page
*page
= NULL
;
118 struct btrfs_file_extent_item
*ei
;
121 size_t cur_size
= size
;
123 unsigned long offset
;
124 int use_compress
= 0;
126 if (compressed_size
&& compressed_pages
) {
128 cur_size
= compressed_size
;
131 path
= btrfs_alloc_path();
135 path
->leave_spinning
= 1;
136 btrfs_set_trans_block_group(trans
, inode
);
138 key
.objectid
= inode
->i_ino
;
140 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
141 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
143 inode_add_bytes(inode
, size
);
144 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
151 leaf
= path
->nodes
[0];
152 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
153 struct btrfs_file_extent_item
);
154 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
155 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
156 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
157 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
158 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
159 ptr
= btrfs_file_extent_inline_start(ei
);
164 while (compressed_size
> 0) {
165 cpage
= compressed_pages
[i
];
166 cur_size
= min_t(unsigned long, compressed_size
,
169 kaddr
= kmap_atomic(cpage
, KM_USER0
);
170 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
171 kunmap_atomic(kaddr
, KM_USER0
);
175 compressed_size
-= cur_size
;
177 btrfs_set_file_extent_compression(leaf
, ei
,
178 BTRFS_COMPRESS_ZLIB
);
180 page
= find_get_page(inode
->i_mapping
,
181 start
>> PAGE_CACHE_SHIFT
);
182 btrfs_set_file_extent_compression(leaf
, ei
, 0);
183 kaddr
= kmap_atomic(page
, KM_USER0
);
184 offset
= start
& (PAGE_CACHE_SIZE
- 1);
185 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
186 kunmap_atomic(kaddr
, KM_USER0
);
187 page_cache_release(page
);
189 btrfs_mark_buffer_dirty(leaf
);
190 btrfs_free_path(path
);
193 * we're an inline extent, so nobody can
194 * extend the file past i_size without locking
195 * a page we already have locked.
197 * We must do any isize and inode updates
198 * before we unlock the pages. Otherwise we
199 * could end up racing with unlink.
201 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
202 btrfs_update_inode(trans
, root
, inode
);
206 btrfs_free_path(path
);
212 * conditionally insert an inline extent into the file. This
213 * does the checks required to make sure the data is small enough
214 * to fit as an inline extent.
216 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
217 struct btrfs_root
*root
,
218 struct inode
*inode
, u64 start
, u64 end
,
219 size_t compressed_size
,
220 struct page
**compressed_pages
)
222 u64 isize
= i_size_read(inode
);
223 u64 actual_end
= min(end
+ 1, isize
);
224 u64 inline_len
= actual_end
- start
;
225 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
226 ~((u64
)root
->sectorsize
- 1);
228 u64 data_len
= inline_len
;
232 data_len
= compressed_size
;
235 actual_end
>= PAGE_CACHE_SIZE
||
236 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
238 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
240 data_len
> root
->fs_info
->max_inline
) {
244 ret
= btrfs_drop_extents(trans
, inode
, start
, aligned_end
,
248 if (isize
> actual_end
)
249 inline_len
= min_t(u64
, isize
, actual_end
);
250 ret
= insert_inline_extent(trans
, root
, inode
, start
,
251 inline_len
, compressed_size
,
254 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
258 struct async_extent
{
263 unsigned long nr_pages
;
264 struct list_head list
;
269 struct btrfs_root
*root
;
270 struct page
*locked_page
;
273 struct list_head extents
;
274 struct btrfs_work work
;
277 static noinline
int add_async_extent(struct async_cow
*cow
,
278 u64 start
, u64 ram_size
,
281 unsigned long nr_pages
)
283 struct async_extent
*async_extent
;
285 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
286 async_extent
->start
= start
;
287 async_extent
->ram_size
= ram_size
;
288 async_extent
->compressed_size
= compressed_size
;
289 async_extent
->pages
= pages
;
290 async_extent
->nr_pages
= nr_pages
;
291 list_add_tail(&async_extent
->list
, &cow
->extents
);
296 * we create compressed extents in two phases. The first
297 * phase compresses a range of pages that have already been
298 * locked (both pages and state bits are locked).
300 * This is done inside an ordered work queue, and the compression
301 * is spread across many cpus. The actual IO submission is step
302 * two, and the ordered work queue takes care of making sure that
303 * happens in the same order things were put onto the queue by
304 * writepages and friends.
306 * If this code finds it can't get good compression, it puts an
307 * entry onto the work queue to write the uncompressed bytes. This
308 * makes sure that both compressed inodes and uncompressed inodes
309 * are written in the same order that pdflush sent them down.
311 static noinline
int compress_file_range(struct inode
*inode
,
312 struct page
*locked_page
,
314 struct async_cow
*async_cow
,
317 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
318 struct btrfs_trans_handle
*trans
;
322 u64 blocksize
= root
->sectorsize
;
324 u64 isize
= i_size_read(inode
);
326 struct page
**pages
= NULL
;
327 unsigned long nr_pages
;
328 unsigned long nr_pages_ret
= 0;
329 unsigned long total_compressed
= 0;
330 unsigned long total_in
= 0;
331 unsigned long max_compressed
= 128 * 1024;
332 unsigned long max_uncompressed
= 128 * 1024;
338 actual_end
= min_t(u64
, isize
, end
+ 1);
341 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
342 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
345 * we don't want to send crud past the end of i_size through
346 * compression, that's just a waste of CPU time. So, if the
347 * end of the file is before the start of our current
348 * requested range of bytes, we bail out to the uncompressed
349 * cleanup code that can deal with all of this.
351 * It isn't really the fastest way to fix things, but this is a
352 * very uncommon corner.
354 if (actual_end
<= start
)
355 goto cleanup_and_bail_uncompressed
;
357 total_compressed
= actual_end
- start
;
359 /* we want to make sure that amount of ram required to uncompress
360 * an extent is reasonable, so we limit the total size in ram
361 * of a compressed extent to 128k. This is a crucial number
362 * because it also controls how easily we can spread reads across
363 * cpus for decompression.
365 * We also want to make sure the amount of IO required to do
366 * a random read is reasonably small, so we limit the size of
367 * a compressed extent to 128k.
369 total_compressed
= min(total_compressed
, max_uncompressed
);
370 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
371 num_bytes
= max(blocksize
, num_bytes
);
372 disk_num_bytes
= num_bytes
;
377 * we do compression for mount -o compress and when the
378 * inode has not been flagged as nocompress. This flag can
379 * change at any time if we discover bad compression ratios.
381 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
382 btrfs_test_opt(root
, COMPRESS
)) {
384 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
386 ret
= btrfs_zlib_compress_pages(inode
->i_mapping
, start
,
387 total_compressed
, pages
,
388 nr_pages
, &nr_pages_ret
,
394 unsigned long offset
= total_compressed
&
395 (PAGE_CACHE_SIZE
- 1);
396 struct page
*page
= pages
[nr_pages_ret
- 1];
399 /* zero the tail end of the last page, we might be
400 * sending it down to disk
403 kaddr
= kmap_atomic(page
, KM_USER0
);
404 memset(kaddr
+ offset
, 0,
405 PAGE_CACHE_SIZE
- offset
);
406 kunmap_atomic(kaddr
, KM_USER0
);
412 trans
= btrfs_join_transaction(root
, 1);
414 btrfs_set_trans_block_group(trans
, inode
);
416 /* lets try to make an inline extent */
417 if (ret
|| total_in
< (actual_end
- start
)) {
418 /* we didn't compress the entire range, try
419 * to make an uncompressed inline extent.
421 ret
= cow_file_range_inline(trans
, root
, inode
,
422 start
, end
, 0, NULL
);
424 /* try making a compressed inline extent */
425 ret
= cow_file_range_inline(trans
, root
, inode
,
427 total_compressed
, pages
);
431 * inline extent creation worked, we don't need
432 * to create any more async work items. Unlock
433 * and free up our temp pages.
435 extent_clear_unlock_delalloc(inode
,
436 &BTRFS_I(inode
)->io_tree
,
438 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
439 EXTENT_CLEAR_DELALLOC
|
440 EXTENT_CLEAR_ACCOUNTING
|
441 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
443 btrfs_end_transaction(trans
, root
);
446 btrfs_end_transaction(trans
, root
);
451 * we aren't doing an inline extent round the compressed size
452 * up to a block size boundary so the allocator does sane
455 total_compressed
= (total_compressed
+ blocksize
- 1) &
459 * one last check to make sure the compression is really a
460 * win, compare the page count read with the blocks on disk
462 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
463 ~(PAGE_CACHE_SIZE
- 1);
464 if (total_compressed
>= total_in
) {
467 disk_num_bytes
= total_compressed
;
468 num_bytes
= total_in
;
471 if (!will_compress
&& pages
) {
473 * the compression code ran but failed to make things smaller,
474 * free any pages it allocated and our page pointer array
476 for (i
= 0; i
< nr_pages_ret
; i
++) {
477 WARN_ON(pages
[i
]->mapping
);
478 page_cache_release(pages
[i
]);
482 total_compressed
= 0;
485 /* flag the file so we don't compress in the future */
486 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
491 /* the async work queues will take care of doing actual
492 * allocation on disk for these compressed pages,
493 * and will submit them to the elevator.
495 add_async_extent(async_cow
, start
, num_bytes
,
496 total_compressed
, pages
, nr_pages_ret
);
498 if (start
+ num_bytes
< end
&& start
+ num_bytes
< actual_end
) {
505 cleanup_and_bail_uncompressed
:
507 * No compression, but we still need to write the pages in
508 * the file we've been given so far. redirty the locked
509 * page if it corresponds to our extent and set things up
510 * for the async work queue to run cow_file_range to do
511 * the normal delalloc dance
513 if (page_offset(locked_page
) >= start
&&
514 page_offset(locked_page
) <= end
) {
515 __set_page_dirty_nobuffers(locked_page
);
516 /* unlocked later on in the async handlers */
518 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0);
526 for (i
= 0; i
< nr_pages_ret
; i
++) {
527 WARN_ON(pages
[i
]->mapping
);
528 page_cache_release(pages
[i
]);
536 * phase two of compressed writeback. This is the ordered portion
537 * of the code, which only gets called in the order the work was
538 * queued. We walk all the async extents created by compress_file_range
539 * and send them down to the disk.
541 static noinline
int submit_compressed_extents(struct inode
*inode
,
542 struct async_cow
*async_cow
)
544 struct async_extent
*async_extent
;
546 struct btrfs_trans_handle
*trans
;
547 struct btrfs_key ins
;
548 struct extent_map
*em
;
549 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
550 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
551 struct extent_io_tree
*io_tree
;
554 if (list_empty(&async_cow
->extents
))
558 while (!list_empty(&async_cow
->extents
)) {
559 async_extent
= list_entry(async_cow
->extents
.next
,
560 struct async_extent
, list
);
561 list_del(&async_extent
->list
);
563 io_tree
= &BTRFS_I(inode
)->io_tree
;
566 /* did the compression code fall back to uncompressed IO? */
567 if (!async_extent
->pages
) {
568 int page_started
= 0;
569 unsigned long nr_written
= 0;
571 lock_extent(io_tree
, async_extent
->start
,
572 async_extent
->start
+
573 async_extent
->ram_size
- 1, GFP_NOFS
);
575 /* allocate blocks */
576 ret
= cow_file_range(inode
, async_cow
->locked_page
,
578 async_extent
->start
+
579 async_extent
->ram_size
- 1,
580 &page_started
, &nr_written
, 0);
583 * if page_started, cow_file_range inserted an
584 * inline extent and took care of all the unlocking
585 * and IO for us. Otherwise, we need to submit
586 * all those pages down to the drive.
588 if (!page_started
&& !ret
)
589 extent_write_locked_range(io_tree
,
590 inode
, async_extent
->start
,
591 async_extent
->start
+
592 async_extent
->ram_size
- 1,
600 lock_extent(io_tree
, async_extent
->start
,
601 async_extent
->start
+ async_extent
->ram_size
- 1,
604 trans
= btrfs_join_transaction(root
, 1);
605 ret
= btrfs_reserve_extent(trans
, root
,
606 async_extent
->compressed_size
,
607 async_extent
->compressed_size
,
610 btrfs_end_transaction(trans
, root
);
614 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
615 WARN_ON(async_extent
->pages
[i
]->mapping
);
616 page_cache_release(async_extent
->pages
[i
]);
618 kfree(async_extent
->pages
);
619 async_extent
->nr_pages
= 0;
620 async_extent
->pages
= NULL
;
621 unlock_extent(io_tree
, async_extent
->start
,
622 async_extent
->start
+
623 async_extent
->ram_size
- 1, GFP_NOFS
);
628 * here we're doing allocation and writeback of the
631 btrfs_drop_extent_cache(inode
, async_extent
->start
,
632 async_extent
->start
+
633 async_extent
->ram_size
- 1, 0);
635 em
= alloc_extent_map(GFP_NOFS
);
636 em
->start
= async_extent
->start
;
637 em
->len
= async_extent
->ram_size
;
638 em
->orig_start
= em
->start
;
640 em
->block_start
= ins
.objectid
;
641 em
->block_len
= ins
.offset
;
642 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
643 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
644 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
647 write_lock(&em_tree
->lock
);
648 ret
= add_extent_mapping(em_tree
, em
);
649 write_unlock(&em_tree
->lock
);
650 if (ret
!= -EEXIST
) {
654 btrfs_drop_extent_cache(inode
, async_extent
->start
,
655 async_extent
->start
+
656 async_extent
->ram_size
- 1, 0);
659 ret
= btrfs_add_ordered_extent(inode
, async_extent
->start
,
661 async_extent
->ram_size
,
663 BTRFS_ORDERED_COMPRESSED
);
667 * clear dirty, set writeback and unlock the pages.
669 extent_clear_unlock_delalloc(inode
,
670 &BTRFS_I(inode
)->io_tree
,
672 async_extent
->start
+
673 async_extent
->ram_size
- 1,
674 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
675 EXTENT_CLEAR_UNLOCK
|
676 EXTENT_CLEAR_DELALLOC
|
677 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
679 ret
= btrfs_submit_compressed_write(inode
,
681 async_extent
->ram_size
,
683 ins
.offset
, async_extent
->pages
,
684 async_extent
->nr_pages
);
687 alloc_hint
= ins
.objectid
+ ins
.offset
;
696 * when extent_io.c finds a delayed allocation range in the file,
697 * the call backs end up in this code. The basic idea is to
698 * allocate extents on disk for the range, and create ordered data structs
699 * in ram to track those extents.
701 * locked_page is the page that writepage had locked already. We use
702 * it to make sure we don't do extra locks or unlocks.
704 * *page_started is set to one if we unlock locked_page and do everything
705 * required to start IO on it. It may be clean and already done with
708 static noinline
int cow_file_range(struct inode
*inode
,
709 struct page
*locked_page
,
710 u64 start
, u64 end
, int *page_started
,
711 unsigned long *nr_written
,
714 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
715 struct btrfs_trans_handle
*trans
;
718 unsigned long ram_size
;
721 u64 blocksize
= root
->sectorsize
;
723 u64 isize
= i_size_read(inode
);
724 struct btrfs_key ins
;
725 struct extent_map
*em
;
726 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
729 trans
= btrfs_join_transaction(root
, 1);
731 btrfs_set_trans_block_group(trans
, inode
);
733 actual_end
= min_t(u64
, isize
, end
+ 1);
735 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
736 num_bytes
= max(blocksize
, num_bytes
);
737 disk_num_bytes
= num_bytes
;
741 /* lets try to make an inline extent */
742 ret
= cow_file_range_inline(trans
, root
, inode
,
743 start
, end
, 0, NULL
);
745 extent_clear_unlock_delalloc(inode
,
746 &BTRFS_I(inode
)->io_tree
,
748 EXTENT_CLEAR_UNLOCK_PAGE
|
749 EXTENT_CLEAR_UNLOCK
|
750 EXTENT_CLEAR_DELALLOC
|
751 EXTENT_CLEAR_ACCOUNTING
|
753 EXTENT_SET_WRITEBACK
|
754 EXTENT_END_WRITEBACK
);
756 *nr_written
= *nr_written
+
757 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
764 BUG_ON(disk_num_bytes
>
765 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
768 read_lock(&BTRFS_I(inode
)->extent_tree
.lock
);
769 em
= search_extent_mapping(&BTRFS_I(inode
)->extent_tree
,
773 * if block start isn't an actual block number then find the
774 * first block in this inode and use that as a hint. If that
775 * block is also bogus then just don't worry about it.
777 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
779 em
= search_extent_mapping(em_tree
, 0, 0);
780 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
781 alloc_hint
= em
->block_start
;
785 alloc_hint
= em
->block_start
;
789 read_unlock(&BTRFS_I(inode
)->extent_tree
.lock
);
790 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
792 while (disk_num_bytes
> 0) {
795 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
796 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
797 root
->sectorsize
, 0, alloc_hint
,
801 em
= alloc_extent_map(GFP_NOFS
);
803 em
->orig_start
= em
->start
;
804 ram_size
= ins
.offset
;
805 em
->len
= ins
.offset
;
807 em
->block_start
= ins
.objectid
;
808 em
->block_len
= ins
.offset
;
809 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
810 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
813 write_lock(&em_tree
->lock
);
814 ret
= add_extent_mapping(em_tree
, em
);
815 write_unlock(&em_tree
->lock
);
816 if (ret
!= -EEXIST
) {
820 btrfs_drop_extent_cache(inode
, start
,
821 start
+ ram_size
- 1, 0);
824 cur_alloc_size
= ins
.offset
;
825 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
826 ram_size
, cur_alloc_size
, 0);
829 if (root
->root_key
.objectid
==
830 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
831 ret
= btrfs_reloc_clone_csums(inode
, start
,
836 if (disk_num_bytes
< cur_alloc_size
)
839 /* we're not doing compressed IO, don't unlock the first
840 * page (which the caller expects to stay locked), don't
841 * clear any dirty bits and don't set any writeback bits
843 * Do set the Private2 bit so we know this page was properly
844 * setup for writepage
846 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
847 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
850 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
851 start
, start
+ ram_size
- 1,
853 disk_num_bytes
-= cur_alloc_size
;
854 num_bytes
-= cur_alloc_size
;
855 alloc_hint
= ins
.objectid
+ ins
.offset
;
856 start
+= cur_alloc_size
;
860 btrfs_end_transaction(trans
, root
);
866 * work queue call back to started compression on a file and pages
868 static noinline
void async_cow_start(struct btrfs_work
*work
)
870 struct async_cow
*async_cow
;
872 async_cow
= container_of(work
, struct async_cow
, work
);
874 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
875 async_cow
->start
, async_cow
->end
, async_cow
,
878 async_cow
->inode
= NULL
;
882 * work queue call back to submit previously compressed pages
884 static noinline
void async_cow_submit(struct btrfs_work
*work
)
886 struct async_cow
*async_cow
;
887 struct btrfs_root
*root
;
888 unsigned long nr_pages
;
890 async_cow
= container_of(work
, struct async_cow
, work
);
892 root
= async_cow
->root
;
893 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
896 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
898 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
900 waitqueue_active(&root
->fs_info
->async_submit_wait
))
901 wake_up(&root
->fs_info
->async_submit_wait
);
903 if (async_cow
->inode
)
904 submit_compressed_extents(async_cow
->inode
, async_cow
);
907 static noinline
void async_cow_free(struct btrfs_work
*work
)
909 struct async_cow
*async_cow
;
910 async_cow
= container_of(work
, struct async_cow
, work
);
914 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
915 u64 start
, u64 end
, int *page_started
,
916 unsigned long *nr_written
)
918 struct async_cow
*async_cow
;
919 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
920 unsigned long nr_pages
;
922 int limit
= 10 * 1024 * 1042;
924 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
925 1, 0, NULL
, GFP_NOFS
);
926 while (start
< end
) {
927 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
928 async_cow
->inode
= inode
;
929 async_cow
->root
= root
;
930 async_cow
->locked_page
= locked_page
;
931 async_cow
->start
= start
;
933 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
936 cur_end
= min(end
, start
+ 512 * 1024 - 1);
938 async_cow
->end
= cur_end
;
939 INIT_LIST_HEAD(&async_cow
->extents
);
941 async_cow
->work
.func
= async_cow_start
;
942 async_cow
->work
.ordered_func
= async_cow_submit
;
943 async_cow
->work
.ordered_free
= async_cow_free
;
944 async_cow
->work
.flags
= 0;
946 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
948 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
950 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
953 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
954 wait_event(root
->fs_info
->async_submit_wait
,
955 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
959 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
960 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
961 wait_event(root
->fs_info
->async_submit_wait
,
962 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
966 *nr_written
+= nr_pages
;
973 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
974 u64 bytenr
, u64 num_bytes
)
977 struct btrfs_ordered_sum
*sums
;
980 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
981 bytenr
+ num_bytes
- 1, &list
);
982 if (ret
== 0 && list_empty(&list
))
985 while (!list_empty(&list
)) {
986 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
987 list_del(&sums
->list
);
994 * when nowcow writeback call back. This checks for snapshots or COW copies
995 * of the extents that exist in the file, and COWs the file as required.
997 * If no cow copies or snapshots exist, we write directly to the existing
1000 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1001 struct page
*locked_page
,
1002 u64 start
, u64 end
, int *page_started
, int force
,
1003 unsigned long *nr_written
)
1005 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1006 struct btrfs_trans_handle
*trans
;
1007 struct extent_buffer
*leaf
;
1008 struct btrfs_path
*path
;
1009 struct btrfs_file_extent_item
*fi
;
1010 struct btrfs_key found_key
;
1023 path
= btrfs_alloc_path();
1025 trans
= btrfs_join_transaction(root
, 1);
1028 cow_start
= (u64
)-1;
1031 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
1034 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1035 leaf
= path
->nodes
[0];
1036 btrfs_item_key_to_cpu(leaf
, &found_key
,
1037 path
->slots
[0] - 1);
1038 if (found_key
.objectid
== inode
->i_ino
&&
1039 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1044 leaf
= path
->nodes
[0];
1045 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1046 ret
= btrfs_next_leaf(root
, path
);
1051 leaf
= path
->nodes
[0];
1057 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1059 if (found_key
.objectid
> inode
->i_ino
||
1060 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1061 found_key
.offset
> end
)
1064 if (found_key
.offset
> cur_offset
) {
1065 extent_end
= found_key
.offset
;
1070 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1071 struct btrfs_file_extent_item
);
1072 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1074 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1075 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1076 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1077 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1078 extent_end
= found_key
.offset
+
1079 btrfs_file_extent_num_bytes(leaf
, fi
);
1080 if (extent_end
<= start
) {
1084 if (disk_bytenr
== 0)
1086 if (btrfs_file_extent_compression(leaf
, fi
) ||
1087 btrfs_file_extent_encryption(leaf
, fi
) ||
1088 btrfs_file_extent_other_encoding(leaf
, fi
))
1090 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1092 if (btrfs_extent_readonly(root
, disk_bytenr
))
1094 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1096 extent_offset
, disk_bytenr
))
1098 disk_bytenr
+= extent_offset
;
1099 disk_bytenr
+= cur_offset
- found_key
.offset
;
1100 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1102 * force cow if csum exists in the range.
1103 * this ensure that csum for a given extent are
1104 * either valid or do not exist.
1106 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1109 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1110 extent_end
= found_key
.offset
+
1111 btrfs_file_extent_inline_len(leaf
, fi
);
1112 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1117 if (extent_end
<= start
) {
1122 if (cow_start
== (u64
)-1)
1123 cow_start
= cur_offset
;
1124 cur_offset
= extent_end
;
1125 if (cur_offset
> end
)
1131 btrfs_release_path(root
, path
);
1132 if (cow_start
!= (u64
)-1) {
1133 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1134 found_key
.offset
- 1, page_started
,
1137 cow_start
= (u64
)-1;
1140 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1141 struct extent_map
*em
;
1142 struct extent_map_tree
*em_tree
;
1143 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1144 em
= alloc_extent_map(GFP_NOFS
);
1145 em
->start
= cur_offset
;
1146 em
->orig_start
= em
->start
;
1147 em
->len
= num_bytes
;
1148 em
->block_len
= num_bytes
;
1149 em
->block_start
= disk_bytenr
;
1150 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1151 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1153 write_lock(&em_tree
->lock
);
1154 ret
= add_extent_mapping(em_tree
, em
);
1155 write_unlock(&em_tree
->lock
);
1156 if (ret
!= -EEXIST
) {
1157 free_extent_map(em
);
1160 btrfs_drop_extent_cache(inode
, em
->start
,
1161 em
->start
+ em
->len
- 1, 0);
1163 type
= BTRFS_ORDERED_PREALLOC
;
1165 type
= BTRFS_ORDERED_NOCOW
;
1168 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1169 num_bytes
, num_bytes
, type
);
1172 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1173 cur_offset
, cur_offset
+ num_bytes
- 1,
1174 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1175 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1176 EXTENT_SET_PRIVATE2
);
1177 cur_offset
= extent_end
;
1178 if (cur_offset
> end
)
1181 btrfs_release_path(root
, path
);
1183 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1184 cow_start
= cur_offset
;
1185 if (cow_start
!= (u64
)-1) {
1186 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1187 page_started
, nr_written
, 1);
1191 ret
= btrfs_end_transaction(trans
, root
);
1193 btrfs_free_path(path
);
1198 * extent_io.c call back to do delayed allocation processing
1200 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1201 u64 start
, u64 end
, int *page_started
,
1202 unsigned long *nr_written
)
1205 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1207 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1208 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1209 page_started
, 1, nr_written
);
1210 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1211 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1212 page_started
, 0, nr_written
);
1213 else if (!btrfs_test_opt(root
, COMPRESS
))
1214 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1215 page_started
, nr_written
, 1);
1217 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1218 page_started
, nr_written
);
1222 static int btrfs_split_extent_hook(struct inode
*inode
,
1223 struct extent_state
*orig
, u64 split
)
1225 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1228 if (!(orig
->state
& EXTENT_DELALLOC
))
1231 size
= orig
->end
- orig
->start
+ 1;
1232 if (size
> root
->fs_info
->max_extent
) {
1236 new_size
= orig
->end
- split
+ 1;
1237 num_extents
= div64_u64(size
+ root
->fs_info
->max_extent
- 1,
1238 root
->fs_info
->max_extent
);
1241 * if we break a large extent up then leave oustanding_extents
1242 * be, since we've already accounted for the large extent.
1244 if (div64_u64(new_size
+ root
->fs_info
->max_extent
- 1,
1245 root
->fs_info
->max_extent
) < num_extents
)
1249 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
1250 BTRFS_I(inode
)->outstanding_extents
++;
1251 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
1257 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1258 * extents so we can keep track of new extents that are just merged onto old
1259 * extents, such as when we are doing sequential writes, so we can properly
1260 * account for the metadata space we'll need.
1262 static int btrfs_merge_extent_hook(struct inode
*inode
,
1263 struct extent_state
*new,
1264 struct extent_state
*other
)
1266 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1267 u64 new_size
, old_size
;
1270 /* not delalloc, ignore it */
1271 if (!(other
->state
& EXTENT_DELALLOC
))
1274 old_size
= other
->end
- other
->start
+ 1;
1275 if (new->start
< other
->start
)
1276 new_size
= other
->end
- new->start
+ 1;
1278 new_size
= new->end
- other
->start
+ 1;
1280 /* we're not bigger than the max, unreserve the space and go */
1281 if (new_size
<= root
->fs_info
->max_extent
) {
1282 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
1283 BTRFS_I(inode
)->outstanding_extents
--;
1284 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
1289 * If we grew by another max_extent, just return, we want to keep that
1292 num_extents
= div64_u64(old_size
+ root
->fs_info
->max_extent
- 1,
1293 root
->fs_info
->max_extent
);
1294 if (div64_u64(new_size
+ root
->fs_info
->max_extent
- 1,
1295 root
->fs_info
->max_extent
) > num_extents
)
1298 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
1299 BTRFS_I(inode
)->outstanding_extents
--;
1300 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
1306 * extent_io.c set_bit_hook, used to track delayed allocation
1307 * bytes in this file, and to maintain the list of inodes that
1308 * have pending delalloc work to be done.
1310 static int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1311 unsigned long old
, unsigned long bits
)
1315 * set_bit and clear bit hooks normally require _irqsave/restore
1316 * but in this case, we are only testeing for the DELALLOC
1317 * bit, which is only set or cleared with irqs on
1319 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1320 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1322 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
1323 BTRFS_I(inode
)->outstanding_extents
++;
1324 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
1325 btrfs_delalloc_reserve_space(root
, inode
, end
- start
+ 1);
1326 spin_lock(&root
->fs_info
->delalloc_lock
);
1327 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
1328 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
1329 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1330 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1331 &root
->fs_info
->delalloc_inodes
);
1333 spin_unlock(&root
->fs_info
->delalloc_lock
);
1339 * extent_io.c clear_bit_hook, see set_bit_hook for why
1341 static int btrfs_clear_bit_hook(struct inode
*inode
,
1342 struct extent_state
*state
, unsigned long bits
)
1345 * set_bit and clear bit hooks normally require _irqsave/restore
1346 * but in this case, we are only testeing for the DELALLOC
1347 * bit, which is only set or cleared with irqs on
1349 if ((state
->state
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1350 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1352 if (bits
& EXTENT_DO_ACCOUNTING
) {
1353 spin_lock(&BTRFS_I(inode
)->accounting_lock
);
1354 BTRFS_I(inode
)->outstanding_extents
--;
1355 spin_unlock(&BTRFS_I(inode
)->accounting_lock
);
1356 btrfs_unreserve_metadata_for_delalloc(root
, inode
, 1);
1359 spin_lock(&root
->fs_info
->delalloc_lock
);
1360 if (state
->end
- state
->start
+ 1 >
1361 root
->fs_info
->delalloc_bytes
) {
1362 printk(KERN_INFO
"btrfs warning: delalloc account "
1364 (unsigned long long)
1365 state
->end
- state
->start
+ 1,
1366 (unsigned long long)
1367 root
->fs_info
->delalloc_bytes
);
1368 btrfs_delalloc_free_space(root
, inode
, (u64
)-1);
1369 root
->fs_info
->delalloc_bytes
= 0;
1370 BTRFS_I(inode
)->delalloc_bytes
= 0;
1372 btrfs_delalloc_free_space(root
, inode
,
1375 root
->fs_info
->delalloc_bytes
-= state
->end
-
1377 BTRFS_I(inode
)->delalloc_bytes
-= state
->end
-
1380 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1381 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1382 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1384 spin_unlock(&root
->fs_info
->delalloc_lock
);
1390 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1391 * we don't create bios that span stripes or chunks
1393 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1394 size_t size
, struct bio
*bio
,
1395 unsigned long bio_flags
)
1397 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1398 struct btrfs_mapping_tree
*map_tree
;
1399 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1404 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1407 length
= bio
->bi_size
;
1408 map_tree
= &root
->fs_info
->mapping_tree
;
1409 map_length
= length
;
1410 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1411 &map_length
, NULL
, 0);
1413 if (map_length
< length
+ size
)
1419 * in order to insert checksums into the metadata in large chunks,
1420 * we wait until bio submission time. All the pages in the bio are
1421 * checksummed and sums are attached onto the ordered extent record.
1423 * At IO completion time the cums attached on the ordered extent record
1424 * are inserted into the btree
1426 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1427 struct bio
*bio
, int mirror_num
,
1428 unsigned long bio_flags
)
1430 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1433 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1439 * in order to insert checksums into the metadata in large chunks,
1440 * we wait until bio submission time. All the pages in the bio are
1441 * checksummed and sums are attached onto the ordered extent record.
1443 * At IO completion time the cums attached on the ordered extent record
1444 * are inserted into the btree
1446 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1447 int mirror_num
, unsigned long bio_flags
)
1449 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1450 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1454 * extent_io.c submission hook. This does the right thing for csum calculation
1455 * on write, or reading the csums from the tree before a read
1457 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1458 int mirror_num
, unsigned long bio_flags
)
1460 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1464 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1466 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1469 if (!(rw
& (1 << BIO_RW
))) {
1470 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1471 return btrfs_submit_compressed_read(inode
, bio
,
1472 mirror_num
, bio_flags
);
1473 } else if (!skip_sum
)
1474 btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1476 } else if (!skip_sum
) {
1477 /* csum items have already been cloned */
1478 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1480 /* we're doing a write, do the async checksumming */
1481 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1482 inode
, rw
, bio
, mirror_num
,
1483 bio_flags
, __btrfs_submit_bio_start
,
1484 __btrfs_submit_bio_done
);
1488 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1492 * given a list of ordered sums record them in the inode. This happens
1493 * at IO completion time based on sums calculated at bio submission time.
1495 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1496 struct inode
*inode
, u64 file_offset
,
1497 struct list_head
*list
)
1499 struct btrfs_ordered_sum
*sum
;
1501 btrfs_set_trans_block_group(trans
, inode
);
1503 list_for_each_entry(sum
, list
, list
) {
1504 btrfs_csum_file_blocks(trans
,
1505 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1510 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
1512 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1514 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1518 /* see btrfs_writepage_start_hook for details on why this is required */
1519 struct btrfs_writepage_fixup
{
1521 struct btrfs_work work
;
1524 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1526 struct btrfs_writepage_fixup
*fixup
;
1527 struct btrfs_ordered_extent
*ordered
;
1529 struct inode
*inode
;
1533 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1537 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1538 ClearPageChecked(page
);
1542 inode
= page
->mapping
->host
;
1543 page_start
= page_offset(page
);
1544 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1546 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1548 /* already ordered? We're done */
1549 if (PagePrivate2(page
))
1552 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1554 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
1555 page_end
, GFP_NOFS
);
1557 btrfs_start_ordered_extent(inode
, ordered
, 1);
1561 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
1562 ClearPageChecked(page
);
1564 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1567 page_cache_release(page
);
1571 * There are a few paths in the higher layers of the kernel that directly
1572 * set the page dirty bit without asking the filesystem if it is a
1573 * good idea. This causes problems because we want to make sure COW
1574 * properly happens and the data=ordered rules are followed.
1576 * In our case any range that doesn't have the ORDERED bit set
1577 * hasn't been properly setup for IO. We kick off an async process
1578 * to fix it up. The async helper will wait for ordered extents, set
1579 * the delalloc bit and make it safe to write the page.
1581 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1583 struct inode
*inode
= page
->mapping
->host
;
1584 struct btrfs_writepage_fixup
*fixup
;
1585 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1587 /* this page is properly in the ordered list */
1588 if (TestClearPagePrivate2(page
))
1591 if (PageChecked(page
))
1594 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1598 SetPageChecked(page
);
1599 page_cache_get(page
);
1600 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1602 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1606 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1607 struct inode
*inode
, u64 file_pos
,
1608 u64 disk_bytenr
, u64 disk_num_bytes
,
1609 u64 num_bytes
, u64 ram_bytes
,
1610 u8 compression
, u8 encryption
,
1611 u16 other_encoding
, int extent_type
)
1613 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1614 struct btrfs_file_extent_item
*fi
;
1615 struct btrfs_path
*path
;
1616 struct extent_buffer
*leaf
;
1617 struct btrfs_key ins
;
1621 path
= btrfs_alloc_path();
1624 path
->leave_spinning
= 1;
1627 * we may be replacing one extent in the tree with another.
1628 * The new extent is pinned in the extent map, and we don't want
1629 * to drop it from the cache until it is completely in the btree.
1631 * So, tell btrfs_drop_extents to leave this extent in the cache.
1632 * the caller is expected to unpin it and allow it to be merged
1635 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1639 ins
.objectid
= inode
->i_ino
;
1640 ins
.offset
= file_pos
;
1641 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1642 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1644 leaf
= path
->nodes
[0];
1645 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1646 struct btrfs_file_extent_item
);
1647 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1648 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1649 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1650 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1651 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1652 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1653 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1654 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1655 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1656 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1658 btrfs_unlock_up_safe(path
, 1);
1659 btrfs_set_lock_blocking(leaf
);
1661 btrfs_mark_buffer_dirty(leaf
);
1663 inode_add_bytes(inode
, num_bytes
);
1665 ins
.objectid
= disk_bytenr
;
1666 ins
.offset
= disk_num_bytes
;
1667 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1668 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1669 root
->root_key
.objectid
,
1670 inode
->i_ino
, file_pos
, &ins
);
1672 btrfs_free_path(path
);
1678 * helper function for btrfs_finish_ordered_io, this
1679 * just reads in some of the csum leaves to prime them into ram
1680 * before we start the transaction. It limits the amount of btree
1681 * reads required while inside the transaction.
1683 static noinline
void reada_csum(struct btrfs_root
*root
,
1684 struct btrfs_path
*path
,
1685 struct btrfs_ordered_extent
*ordered_extent
)
1687 struct btrfs_ordered_sum
*sum
;
1690 sum
= list_entry(ordered_extent
->list
.next
, struct btrfs_ordered_sum
,
1692 bytenr
= sum
->sums
[0].bytenr
;
1695 * we don't care about the results, the point of this search is
1696 * just to get the btree leaves into ram
1698 btrfs_lookup_csum(NULL
, root
->fs_info
->csum_root
, path
, bytenr
, 0);
1701 /* as ordered data IO finishes, this gets called so we can finish
1702 * an ordered extent if the range of bytes in the file it covers are
1705 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1707 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1708 struct btrfs_trans_handle
*trans
;
1709 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1710 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1711 struct btrfs_path
*path
;
1715 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1720 * before we join the transaction, try to do some of our IO.
1721 * This will limit the amount of IO that we have to do with
1722 * the transaction running. We're unlikely to need to do any
1723 * IO if the file extents are new, the disk_i_size checks
1724 * covers the most common case.
1726 if (start
< BTRFS_I(inode
)->disk_i_size
) {
1727 path
= btrfs_alloc_path();
1729 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
1732 ordered_extent
= btrfs_lookup_ordered_extent(inode
,
1734 if (!list_empty(&ordered_extent
->list
)) {
1735 btrfs_release_path(root
, path
);
1736 reada_csum(root
, path
, ordered_extent
);
1738 btrfs_free_path(path
);
1742 if (!ordered_extent
)
1743 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1744 BUG_ON(!ordered_extent
);
1745 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1746 BUG_ON(!list_empty(&ordered_extent
->list
));
1747 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1749 trans
= btrfs_join_transaction(root
, 1);
1750 ret
= btrfs_update_inode(trans
, root
, inode
);
1752 btrfs_end_transaction(trans
, root
);
1757 lock_extent(io_tree
, ordered_extent
->file_offset
,
1758 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1761 trans
= btrfs_join_transaction(root
, 1);
1763 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1765 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1767 ret
= btrfs_mark_extent_written(trans
, inode
,
1768 ordered_extent
->file_offset
,
1769 ordered_extent
->file_offset
+
1770 ordered_extent
->len
);
1773 ret
= insert_reserved_file_extent(trans
, inode
,
1774 ordered_extent
->file_offset
,
1775 ordered_extent
->start
,
1776 ordered_extent
->disk_len
,
1777 ordered_extent
->len
,
1778 ordered_extent
->len
,
1780 BTRFS_FILE_EXTENT_REG
);
1781 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1782 ordered_extent
->file_offset
,
1783 ordered_extent
->len
);
1786 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1787 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1789 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1790 &ordered_extent
->list
);
1792 /* this also removes the ordered extent from the tree */
1793 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1794 ret
= btrfs_update_inode(trans
, root
, inode
);
1796 btrfs_end_transaction(trans
, root
);
1799 btrfs_put_ordered_extent(ordered_extent
);
1800 /* once for the tree */
1801 btrfs_put_ordered_extent(ordered_extent
);
1806 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1807 struct extent_state
*state
, int uptodate
)
1809 ClearPagePrivate2(page
);
1810 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1814 * When IO fails, either with EIO or csum verification fails, we
1815 * try other mirrors that might have a good copy of the data. This
1816 * io_failure_record is used to record state as we go through all the
1817 * mirrors. If another mirror has good data, the page is set up to date
1818 * and things continue. If a good mirror can't be found, the original
1819 * bio end_io callback is called to indicate things have failed.
1821 struct io_failure_record
{
1826 unsigned long bio_flags
;
1830 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1831 struct page
*page
, u64 start
, u64 end
,
1832 struct extent_state
*state
)
1834 struct io_failure_record
*failrec
= NULL
;
1836 struct extent_map
*em
;
1837 struct inode
*inode
= page
->mapping
->host
;
1838 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1839 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1846 ret
= get_state_private(failure_tree
, start
, &private);
1848 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1851 failrec
->start
= start
;
1852 failrec
->len
= end
- start
+ 1;
1853 failrec
->last_mirror
= 0;
1854 failrec
->bio_flags
= 0;
1856 read_lock(&em_tree
->lock
);
1857 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1858 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1859 free_extent_map(em
);
1862 read_unlock(&em_tree
->lock
);
1864 if (!em
|| IS_ERR(em
)) {
1868 logical
= start
- em
->start
;
1869 logical
= em
->block_start
+ logical
;
1870 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1871 logical
= em
->block_start
;
1872 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1874 failrec
->logical
= logical
;
1875 free_extent_map(em
);
1876 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1877 EXTENT_DIRTY
, GFP_NOFS
);
1878 set_state_private(failure_tree
, start
,
1879 (u64
)(unsigned long)failrec
);
1881 failrec
= (struct io_failure_record
*)(unsigned long)private;
1883 num_copies
= btrfs_num_copies(
1884 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1885 failrec
->logical
, failrec
->len
);
1886 failrec
->last_mirror
++;
1888 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1889 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1892 if (state
&& state
->start
!= failrec
->start
)
1894 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1896 if (!state
|| failrec
->last_mirror
> num_copies
) {
1897 set_state_private(failure_tree
, failrec
->start
, 0);
1898 clear_extent_bits(failure_tree
, failrec
->start
,
1899 failrec
->start
+ failrec
->len
- 1,
1900 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1904 bio
= bio_alloc(GFP_NOFS
, 1);
1905 bio
->bi_private
= state
;
1906 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1907 bio
->bi_sector
= failrec
->logical
>> 9;
1908 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1911 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1912 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1917 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1918 failrec
->last_mirror
,
1919 failrec
->bio_flags
);
1924 * each time an IO finishes, we do a fast check in the IO failure tree
1925 * to see if we need to process or clean up an io_failure_record
1927 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1930 u64 private_failure
;
1931 struct io_failure_record
*failure
;
1935 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1936 (u64
)-1, 1, EXTENT_DIRTY
)) {
1937 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1938 start
, &private_failure
);
1940 failure
= (struct io_failure_record
*)(unsigned long)
1942 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1944 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1946 failure
->start
+ failure
->len
- 1,
1947 EXTENT_DIRTY
| EXTENT_LOCKED
,
1956 * when reads are done, we need to check csums to verify the data is correct
1957 * if there's a match, we allow the bio to finish. If not, we go through
1958 * the io_failure_record routines to find good copies
1960 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1961 struct extent_state
*state
)
1963 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1964 struct inode
*inode
= page
->mapping
->host
;
1965 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1967 u64
private = ~(u32
)0;
1969 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1972 if (PageChecked(page
)) {
1973 ClearPageChecked(page
);
1977 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1980 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1981 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1982 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1987 if (state
&& state
->start
== start
) {
1988 private = state
->private;
1991 ret
= get_state_private(io_tree
, start
, &private);
1993 kaddr
= kmap_atomic(page
, KM_USER0
);
1997 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1998 btrfs_csum_final(csum
, (char *)&csum
);
1999 if (csum
!= private)
2002 kunmap_atomic(kaddr
, KM_USER0
);
2004 /* if the io failure tree for this inode is non-empty,
2005 * check to see if we've recovered from a failed IO
2007 btrfs_clean_io_failures(inode
, start
);
2011 if (printk_ratelimit()) {
2012 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
2013 "private %llu\n", page
->mapping
->host
->i_ino
,
2014 (unsigned long long)start
, csum
,
2015 (unsigned long long)private);
2017 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2018 flush_dcache_page(page
);
2019 kunmap_atomic(kaddr
, KM_USER0
);
2025 struct delayed_iput
{
2026 struct list_head list
;
2027 struct inode
*inode
;
2030 void btrfs_add_delayed_iput(struct inode
*inode
)
2032 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2033 struct delayed_iput
*delayed
;
2035 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2038 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2039 delayed
->inode
= inode
;
2041 spin_lock(&fs_info
->delayed_iput_lock
);
2042 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2043 spin_unlock(&fs_info
->delayed_iput_lock
);
2046 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2049 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2050 struct delayed_iput
*delayed
;
2053 spin_lock(&fs_info
->delayed_iput_lock
);
2054 empty
= list_empty(&fs_info
->delayed_iputs
);
2055 spin_unlock(&fs_info
->delayed_iput_lock
);
2059 down_read(&root
->fs_info
->cleanup_work_sem
);
2060 spin_lock(&fs_info
->delayed_iput_lock
);
2061 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2062 spin_unlock(&fs_info
->delayed_iput_lock
);
2064 while (!list_empty(&list
)) {
2065 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2066 list_del(&delayed
->list
);
2067 iput(delayed
->inode
);
2070 up_read(&root
->fs_info
->cleanup_work_sem
);
2074 * This creates an orphan entry for the given inode in case something goes
2075 * wrong in the middle of an unlink/truncate.
2077 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2079 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2082 spin_lock(&root
->list_lock
);
2084 /* already on the orphan list, we're good */
2085 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2086 spin_unlock(&root
->list_lock
);
2090 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2092 spin_unlock(&root
->list_lock
);
2095 * insert an orphan item to track this unlinked/truncated file
2097 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
2103 * We have done the truncate/delete so we can go ahead and remove the orphan
2104 * item for this particular inode.
2106 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2108 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2111 spin_lock(&root
->list_lock
);
2113 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2114 spin_unlock(&root
->list_lock
);
2118 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2120 spin_unlock(&root
->list_lock
);
2124 spin_unlock(&root
->list_lock
);
2126 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
2132 * this cleans up any orphans that may be left on the list from the last use
2135 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
2137 struct btrfs_path
*path
;
2138 struct extent_buffer
*leaf
;
2139 struct btrfs_item
*item
;
2140 struct btrfs_key key
, found_key
;
2141 struct btrfs_trans_handle
*trans
;
2142 struct inode
*inode
;
2143 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2145 if (!xchg(&root
->clean_orphans
, 0))
2148 path
= btrfs_alloc_path();
2152 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2153 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2154 key
.offset
= (u64
)-1;
2157 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2159 printk(KERN_ERR
"Error searching slot for orphan: %d"
2165 * if ret == 0 means we found what we were searching for, which
2166 * is weird, but possible, so only screw with path if we didnt
2167 * find the key and see if we have stuff that matches
2170 if (path
->slots
[0] == 0)
2175 /* pull out the item */
2176 leaf
= path
->nodes
[0];
2177 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
2178 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2180 /* make sure the item matches what we want */
2181 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2183 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2186 /* release the path since we're done with it */
2187 btrfs_release_path(root
, path
);
2190 * this is where we are basically btrfs_lookup, without the
2191 * crossing root thing. we store the inode number in the
2192 * offset of the orphan item.
2194 found_key
.objectid
= found_key
.offset
;
2195 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2196 found_key
.offset
= 0;
2197 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
);
2202 * add this inode to the orphan list so btrfs_orphan_del does
2203 * the proper thing when we hit it
2205 spin_lock(&root
->list_lock
);
2206 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2207 spin_unlock(&root
->list_lock
);
2210 * if this is a bad inode, means we actually succeeded in
2211 * removing the inode, but not the orphan record, which means
2212 * we need to manually delete the orphan since iput will just
2213 * do a destroy_inode
2215 if (is_bad_inode(inode
)) {
2216 trans
= btrfs_start_transaction(root
, 1);
2217 btrfs_orphan_del(trans
, inode
);
2218 btrfs_end_transaction(trans
, root
);
2223 /* if we have links, this was a truncate, lets do that */
2224 if (inode
->i_nlink
) {
2226 btrfs_truncate(inode
);
2231 /* this will do delete_inode and everything for us */
2236 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2238 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2240 btrfs_free_path(path
);
2244 * very simple check to peek ahead in the leaf looking for xattrs. If we
2245 * don't find any xattrs, we know there can't be any acls.
2247 * slot is the slot the inode is in, objectid is the objectid of the inode
2249 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2250 int slot
, u64 objectid
)
2252 u32 nritems
= btrfs_header_nritems(leaf
);
2253 struct btrfs_key found_key
;
2257 while (slot
< nritems
) {
2258 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2260 /* we found a different objectid, there must not be acls */
2261 if (found_key
.objectid
!= objectid
)
2264 /* we found an xattr, assume we've got an acl */
2265 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2269 * we found a key greater than an xattr key, there can't
2270 * be any acls later on
2272 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2279 * it goes inode, inode backrefs, xattrs, extents,
2280 * so if there are a ton of hard links to an inode there can
2281 * be a lot of backrefs. Don't waste time searching too hard,
2282 * this is just an optimization
2287 /* we hit the end of the leaf before we found an xattr or
2288 * something larger than an xattr. We have to assume the inode
2295 * read an inode from the btree into the in-memory inode
2297 static void btrfs_read_locked_inode(struct inode
*inode
)
2299 struct btrfs_path
*path
;
2300 struct extent_buffer
*leaf
;
2301 struct btrfs_inode_item
*inode_item
;
2302 struct btrfs_timespec
*tspec
;
2303 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2304 struct btrfs_key location
;
2306 u64 alloc_group_block
;
2310 path
= btrfs_alloc_path();
2312 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2314 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2318 leaf
= path
->nodes
[0];
2319 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2320 struct btrfs_inode_item
);
2322 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2323 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2324 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2325 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2326 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2328 tspec
= btrfs_inode_atime(inode_item
);
2329 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2330 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2332 tspec
= btrfs_inode_mtime(inode_item
);
2333 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2334 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2336 tspec
= btrfs_inode_ctime(inode_item
);
2337 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2338 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2340 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2341 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2342 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2343 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2345 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2347 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2348 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2350 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2353 * try to precache a NULL acl entry for files that don't have
2354 * any xattrs or acls
2356 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0], inode
->i_ino
);
2358 cache_no_acl(inode
);
2360 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2361 alloc_group_block
, 0);
2362 btrfs_free_path(path
);
2365 switch (inode
->i_mode
& S_IFMT
) {
2367 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2368 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2369 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2370 inode
->i_fop
= &btrfs_file_operations
;
2371 inode
->i_op
= &btrfs_file_inode_operations
;
2374 inode
->i_fop
= &btrfs_dir_file_operations
;
2375 if (root
== root
->fs_info
->tree_root
)
2376 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2378 inode
->i_op
= &btrfs_dir_inode_operations
;
2381 inode
->i_op
= &btrfs_symlink_inode_operations
;
2382 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2383 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2386 inode
->i_op
= &btrfs_special_inode_operations
;
2387 init_special_inode(inode
, inode
->i_mode
, rdev
);
2391 btrfs_update_iflags(inode
);
2395 btrfs_free_path(path
);
2396 make_bad_inode(inode
);
2400 * given a leaf and an inode, copy the inode fields into the leaf
2402 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2403 struct extent_buffer
*leaf
,
2404 struct btrfs_inode_item
*item
,
2405 struct inode
*inode
)
2407 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2408 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2409 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2410 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2411 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2413 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2414 inode
->i_atime
.tv_sec
);
2415 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2416 inode
->i_atime
.tv_nsec
);
2418 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2419 inode
->i_mtime
.tv_sec
);
2420 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2421 inode
->i_mtime
.tv_nsec
);
2423 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2424 inode
->i_ctime
.tv_sec
);
2425 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2426 inode
->i_ctime
.tv_nsec
);
2428 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2429 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2430 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2431 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2432 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2433 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2434 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2438 * copy everything in the in-memory inode into the btree.
2440 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2441 struct btrfs_root
*root
, struct inode
*inode
)
2443 struct btrfs_inode_item
*inode_item
;
2444 struct btrfs_path
*path
;
2445 struct extent_buffer
*leaf
;
2448 path
= btrfs_alloc_path();
2450 path
->leave_spinning
= 1;
2451 ret
= btrfs_lookup_inode(trans
, root
, path
,
2452 &BTRFS_I(inode
)->location
, 1);
2459 btrfs_unlock_up_safe(path
, 1);
2460 leaf
= path
->nodes
[0];
2461 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2462 struct btrfs_inode_item
);
2464 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2465 btrfs_mark_buffer_dirty(leaf
);
2466 btrfs_set_inode_last_trans(trans
, inode
);
2469 btrfs_free_path(path
);
2475 * unlink helper that gets used here in inode.c and in the tree logging
2476 * recovery code. It remove a link in a directory with a given name, and
2477 * also drops the back refs in the inode to the directory
2479 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2480 struct btrfs_root
*root
,
2481 struct inode
*dir
, struct inode
*inode
,
2482 const char *name
, int name_len
)
2484 struct btrfs_path
*path
;
2486 struct extent_buffer
*leaf
;
2487 struct btrfs_dir_item
*di
;
2488 struct btrfs_key key
;
2491 path
= btrfs_alloc_path();
2497 path
->leave_spinning
= 1;
2498 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2499 name
, name_len
, -1);
2508 leaf
= path
->nodes
[0];
2509 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2510 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2513 btrfs_release_path(root
, path
);
2515 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2517 dir
->i_ino
, &index
);
2519 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2520 "inode %lu parent %lu\n", name_len
, name
,
2521 inode
->i_ino
, dir
->i_ino
);
2525 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2526 index
, name
, name_len
, -1);
2535 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2536 btrfs_release_path(root
, path
);
2538 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2540 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2542 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2546 btrfs_free_path(path
);
2550 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2551 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2552 btrfs_update_inode(trans
, root
, dir
);
2553 btrfs_drop_nlink(inode
);
2554 ret
= btrfs_update_inode(trans
, root
, inode
);
2559 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2561 struct btrfs_root
*root
;
2562 struct btrfs_trans_handle
*trans
;
2563 struct inode
*inode
= dentry
->d_inode
;
2565 unsigned long nr
= 0;
2567 root
= BTRFS_I(dir
)->root
;
2570 * 5 items for unlink inode
2573 ret
= btrfs_reserve_metadata_space(root
, 6);
2577 trans
= btrfs_start_transaction(root
, 1);
2578 if (IS_ERR(trans
)) {
2579 btrfs_unreserve_metadata_space(root
, 6);
2580 return PTR_ERR(trans
);
2583 btrfs_set_trans_block_group(trans
, dir
);
2585 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2587 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2588 dentry
->d_name
.name
, dentry
->d_name
.len
);
2590 if (inode
->i_nlink
== 0)
2591 ret
= btrfs_orphan_add(trans
, inode
);
2593 nr
= trans
->blocks_used
;
2595 btrfs_end_transaction_throttle(trans
, root
);
2596 btrfs_unreserve_metadata_space(root
, 6);
2597 btrfs_btree_balance_dirty(root
, nr
);
2601 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2602 struct btrfs_root
*root
,
2603 struct inode
*dir
, u64 objectid
,
2604 const char *name
, int name_len
)
2606 struct btrfs_path
*path
;
2607 struct extent_buffer
*leaf
;
2608 struct btrfs_dir_item
*di
;
2609 struct btrfs_key key
;
2613 path
= btrfs_alloc_path();
2617 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2618 name
, name_len
, -1);
2619 BUG_ON(!di
|| IS_ERR(di
));
2621 leaf
= path
->nodes
[0];
2622 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2623 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
2624 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2626 btrfs_release_path(root
, path
);
2628 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
2629 objectid
, root
->root_key
.objectid
,
2630 dir
->i_ino
, &index
, name
, name_len
);
2632 BUG_ON(ret
!= -ENOENT
);
2633 di
= btrfs_search_dir_index_item(root
, path
, dir
->i_ino
,
2635 BUG_ON(!di
|| IS_ERR(di
));
2637 leaf
= path
->nodes
[0];
2638 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2639 btrfs_release_path(root
, path
);
2643 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2644 index
, name
, name_len
, -1);
2645 BUG_ON(!di
|| IS_ERR(di
));
2647 leaf
= path
->nodes
[0];
2648 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2649 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
2650 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2652 btrfs_release_path(root
, path
);
2654 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2655 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2656 ret
= btrfs_update_inode(trans
, root
, dir
);
2658 dir
->i_sb
->s_dirt
= 1;
2660 btrfs_free_path(path
);
2664 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2666 struct inode
*inode
= dentry
->d_inode
;
2669 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2670 struct btrfs_trans_handle
*trans
;
2671 unsigned long nr
= 0;
2673 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2674 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
2677 ret
= btrfs_reserve_metadata_space(root
, 5);
2681 trans
= btrfs_start_transaction(root
, 1);
2682 if (IS_ERR(trans
)) {
2683 btrfs_unreserve_metadata_space(root
, 5);
2684 return PTR_ERR(trans
);
2687 btrfs_set_trans_block_group(trans
, dir
);
2689 if (unlikely(inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
2690 err
= btrfs_unlink_subvol(trans
, root
, dir
,
2691 BTRFS_I(inode
)->location
.objectid
,
2692 dentry
->d_name
.name
,
2693 dentry
->d_name
.len
);
2697 err
= btrfs_orphan_add(trans
, inode
);
2701 /* now the directory is empty */
2702 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2703 dentry
->d_name
.name
, dentry
->d_name
.len
);
2705 btrfs_i_size_write(inode
, 0);
2707 nr
= trans
->blocks_used
;
2708 ret
= btrfs_end_transaction_throttle(trans
, root
);
2709 btrfs_unreserve_metadata_space(root
, 5);
2710 btrfs_btree_balance_dirty(root
, nr
);
2719 * when truncating bytes in a file, it is possible to avoid reading
2720 * the leaves that contain only checksum items. This can be the
2721 * majority of the IO required to delete a large file, but it must
2722 * be done carefully.
2724 * The keys in the level just above the leaves are checked to make sure
2725 * the lowest key in a given leaf is a csum key, and starts at an offset
2726 * after the new size.
2728 * Then the key for the next leaf is checked to make sure it also has
2729 * a checksum item for the same file. If it does, we know our target leaf
2730 * contains only checksum items, and it can be safely freed without reading
2733 * This is just an optimization targeted at large files. It may do
2734 * nothing. It will return 0 unless things went badly.
2736 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
2737 struct btrfs_root
*root
,
2738 struct btrfs_path
*path
,
2739 struct inode
*inode
, u64 new_size
)
2741 struct btrfs_key key
;
2744 struct btrfs_key found_key
;
2745 struct btrfs_key other_key
;
2746 struct btrfs_leaf_ref
*ref
;
2750 path
->lowest_level
= 1;
2751 key
.objectid
= inode
->i_ino
;
2752 key
.type
= BTRFS_CSUM_ITEM_KEY
;
2753 key
.offset
= new_size
;
2755 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2759 if (path
->nodes
[1] == NULL
) {
2764 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
2765 nritems
= btrfs_header_nritems(path
->nodes
[1]);
2770 if (path
->slots
[1] >= nritems
)
2773 /* did we find a key greater than anything we want to delete? */
2774 if (found_key
.objectid
> inode
->i_ino
||
2775 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
2778 /* we check the next key in the node to make sure the leave contains
2779 * only checksum items. This comparison doesn't work if our
2780 * leaf is the last one in the node
2782 if (path
->slots
[1] + 1 >= nritems
) {
2784 /* search forward from the last key in the node, this
2785 * will bring us into the next node in the tree
2787 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
2789 /* unlikely, but we inc below, so check to be safe */
2790 if (found_key
.offset
== (u64
)-1)
2793 /* search_forward needs a path with locks held, do the
2794 * search again for the original key. It is possible
2795 * this will race with a balance and return a path that
2796 * we could modify, but this drop is just an optimization
2797 * and is allowed to miss some leaves.
2799 btrfs_release_path(root
, path
);
2802 /* setup a max key for search_forward */
2803 other_key
.offset
= (u64
)-1;
2804 other_key
.type
= key
.type
;
2805 other_key
.objectid
= key
.objectid
;
2807 path
->keep_locks
= 1;
2808 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
2810 path
->keep_locks
= 0;
2811 if (ret
|| found_key
.objectid
!= key
.objectid
||
2812 found_key
.type
!= key
.type
) {
2817 key
.offset
= found_key
.offset
;
2818 btrfs_release_path(root
, path
);
2823 /* we know there's one more slot after us in the tree,
2824 * read that key so we can verify it is also a checksum item
2826 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
2828 if (found_key
.objectid
< inode
->i_ino
)
2831 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
2835 * if the key for the next leaf isn't a csum key from this objectid,
2836 * we can't be sure there aren't good items inside this leaf.
2839 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
2842 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
2843 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
2845 * it is safe to delete this leaf, it contains only
2846 * csum items from this inode at an offset >= new_size
2848 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
2851 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
2852 ref
= btrfs_alloc_leaf_ref(root
, 0);
2854 ref
->root_gen
= root
->root_key
.offset
;
2855 ref
->bytenr
= leaf_start
;
2857 ref
->generation
= leaf_gen
;
2860 btrfs_sort_leaf_ref(ref
);
2862 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
2864 btrfs_free_leaf_ref(root
, ref
);
2870 btrfs_release_path(root
, path
);
2872 if (other_key
.objectid
== inode
->i_ino
&&
2873 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
2874 key
.offset
= other_key
.offset
;
2880 /* fixup any changes we've made to the path */
2881 path
->lowest_level
= 0;
2882 path
->keep_locks
= 0;
2883 btrfs_release_path(root
, path
);
2890 * this can truncate away extent items, csum items and directory items.
2891 * It starts at a high offset and removes keys until it can't find
2892 * any higher than new_size
2894 * csum items that cross the new i_size are truncated to the new size
2897 * min_type is the minimum key type to truncate down to. If set to 0, this
2898 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2900 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2901 struct btrfs_root
*root
,
2902 struct inode
*inode
,
2903 u64 new_size
, u32 min_type
)
2905 struct btrfs_path
*path
;
2906 struct extent_buffer
*leaf
;
2907 struct btrfs_file_extent_item
*fi
;
2908 struct btrfs_key key
;
2909 struct btrfs_key found_key
;
2910 u64 extent_start
= 0;
2911 u64 extent_num_bytes
= 0;
2912 u64 extent_offset
= 0;
2914 u64 mask
= root
->sectorsize
- 1;
2915 u32 found_type
= (u8
)-1;
2918 int pending_del_nr
= 0;
2919 int pending_del_slot
= 0;
2920 int extent_type
= -1;
2925 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
2928 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2930 path
= btrfs_alloc_path();
2934 key
.objectid
= inode
->i_ino
;
2935 key
.offset
= (u64
)-1;
2939 path
->leave_spinning
= 1;
2940 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2947 /* there are no items in the tree for us to truncate, we're
2950 if (path
->slots
[0] == 0)
2957 leaf
= path
->nodes
[0];
2958 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2959 found_type
= btrfs_key_type(&found_key
);
2962 if (found_key
.objectid
!= inode
->i_ino
)
2965 if (found_type
< min_type
)
2968 item_end
= found_key
.offset
;
2969 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2970 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2971 struct btrfs_file_extent_item
);
2972 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2973 encoding
= btrfs_file_extent_compression(leaf
, fi
);
2974 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
2975 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
2977 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2979 btrfs_file_extent_num_bytes(leaf
, fi
);
2980 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2981 item_end
+= btrfs_file_extent_inline_len(leaf
,
2986 if (found_type
> min_type
) {
2989 if (item_end
< new_size
)
2991 if (found_key
.offset
>= new_size
)
2997 /* FIXME, shrink the extent if the ref count is only 1 */
2998 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3001 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3003 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3004 if (!del_item
&& !encoding
) {
3005 u64 orig_num_bytes
=
3006 btrfs_file_extent_num_bytes(leaf
, fi
);
3007 extent_num_bytes
= new_size
-
3008 found_key
.offset
+ root
->sectorsize
- 1;
3009 extent_num_bytes
= extent_num_bytes
&
3010 ~((u64
)root
->sectorsize
- 1);
3011 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3013 num_dec
= (orig_num_bytes
-
3015 if (root
->ref_cows
&& extent_start
!= 0)
3016 inode_sub_bytes(inode
, num_dec
);
3017 btrfs_mark_buffer_dirty(leaf
);
3020 btrfs_file_extent_disk_num_bytes(leaf
,
3022 extent_offset
= found_key
.offset
-
3023 btrfs_file_extent_offset(leaf
, fi
);
3025 /* FIXME blocksize != 4096 */
3026 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3027 if (extent_start
!= 0) {
3030 inode_sub_bytes(inode
, num_dec
);
3033 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3035 * we can't truncate inline items that have had
3039 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3040 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3041 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3042 u32 size
= new_size
- found_key
.offset
;
3044 if (root
->ref_cows
) {
3045 inode_sub_bytes(inode
, item_end
+ 1 -
3049 btrfs_file_extent_calc_inline_size(size
);
3050 ret
= btrfs_truncate_item(trans
, root
, path
,
3053 } else if (root
->ref_cows
) {
3054 inode_sub_bytes(inode
, item_end
+ 1 -
3060 if (!pending_del_nr
) {
3061 /* no pending yet, add ourselves */
3062 pending_del_slot
= path
->slots
[0];
3064 } else if (pending_del_nr
&&
3065 path
->slots
[0] + 1 == pending_del_slot
) {
3066 /* hop on the pending chunk */
3068 pending_del_slot
= path
->slots
[0];
3075 if (found_extent
&& root
->ref_cows
) {
3076 btrfs_set_path_blocking(path
);
3077 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3078 extent_num_bytes
, 0,
3079 btrfs_header_owner(leaf
),
3080 inode
->i_ino
, extent_offset
);
3084 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3087 if (path
->slots
[0] == 0 ||
3088 path
->slots
[0] != pending_del_slot
) {
3089 if (root
->ref_cows
) {
3093 if (pending_del_nr
) {
3094 ret
= btrfs_del_items(trans
, root
, path
,
3100 btrfs_release_path(root
, path
);
3107 if (pending_del_nr
) {
3108 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3111 btrfs_free_path(path
);
3116 * taken from block_truncate_page, but does cow as it zeros out
3117 * any bytes left in the last page in the file.
3119 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3121 struct inode
*inode
= mapping
->host
;
3122 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3123 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3124 struct btrfs_ordered_extent
*ordered
;
3126 u32 blocksize
= root
->sectorsize
;
3127 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3128 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3134 if ((offset
& (blocksize
- 1)) == 0)
3136 ret
= btrfs_check_data_free_space(root
, inode
, PAGE_CACHE_SIZE
);
3140 ret
= btrfs_reserve_metadata_for_delalloc(root
, inode
, 1);
3146 page
= grab_cache_page(mapping
, index
);
3148 btrfs_free_reserved_data_space(root
, inode
, PAGE_CACHE_SIZE
);
3149 btrfs_unreserve_metadata_for_delalloc(root
, inode
, 1);
3153 page_start
= page_offset(page
);
3154 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3156 if (!PageUptodate(page
)) {
3157 ret
= btrfs_readpage(NULL
, page
);
3159 if (page
->mapping
!= mapping
) {
3161 page_cache_release(page
);
3164 if (!PageUptodate(page
)) {
3169 wait_on_page_writeback(page
);
3171 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3172 set_page_extent_mapped(page
);
3174 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3176 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3178 page_cache_release(page
);
3179 btrfs_start_ordered_extent(inode
, ordered
, 1);
3180 btrfs_put_ordered_extent(ordered
);
3184 clear_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3185 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3188 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
3190 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3195 if (offset
!= PAGE_CACHE_SIZE
) {
3197 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3198 flush_dcache_page(page
);
3201 ClearPageChecked(page
);
3202 set_page_dirty(page
);
3203 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3207 btrfs_free_reserved_data_space(root
, inode
, PAGE_CACHE_SIZE
);
3208 btrfs_unreserve_metadata_for_delalloc(root
, inode
, 1);
3210 page_cache_release(page
);
3215 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
3217 struct btrfs_trans_handle
*trans
;
3218 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3219 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3220 struct extent_map
*em
;
3221 u64 mask
= root
->sectorsize
- 1;
3222 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
3223 u64 block_end
= (size
+ mask
) & ~mask
;
3229 if (size
<= hole_start
)
3233 struct btrfs_ordered_extent
*ordered
;
3234 btrfs_wait_ordered_range(inode
, hole_start
,
3235 block_end
- hole_start
);
3236 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
3237 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3240 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
3241 btrfs_put_ordered_extent(ordered
);
3244 cur_offset
= hole_start
;
3246 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3247 block_end
- cur_offset
, 0);
3248 BUG_ON(IS_ERR(em
) || !em
);
3249 last_byte
= min(extent_map_end(em
), block_end
);
3250 last_byte
= (last_byte
+ mask
) & ~mask
;
3251 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3253 hole_size
= last_byte
- cur_offset
;
3255 err
= btrfs_reserve_metadata_space(root
, 2);
3259 trans
= btrfs_start_transaction(root
, 1);
3260 btrfs_set_trans_block_group(trans
, inode
);
3262 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3263 cur_offset
+ hole_size
,
3267 err
= btrfs_insert_file_extent(trans
, root
,
3268 inode
->i_ino
, cur_offset
, 0,
3269 0, hole_size
, 0, hole_size
,
3273 btrfs_drop_extent_cache(inode
, hole_start
,
3276 btrfs_end_transaction(trans
, root
);
3277 btrfs_unreserve_metadata_space(root
, 2);
3279 free_extent_map(em
);
3280 cur_offset
= last_byte
;
3281 if (cur_offset
>= block_end
)
3285 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
3289 static int btrfs_setattr_size(struct inode
*inode
, struct iattr
*attr
)
3291 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3292 struct btrfs_trans_handle
*trans
;
3296 if (attr
->ia_size
== inode
->i_size
)
3299 if (attr
->ia_size
> inode
->i_size
) {
3300 unsigned long limit
;
3301 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
3302 if (attr
->ia_size
> inode
->i_sb
->s_maxbytes
)
3304 if (limit
!= RLIM_INFINITY
&& attr
->ia_size
> limit
) {
3305 send_sig(SIGXFSZ
, current
, 0);
3310 ret
= btrfs_reserve_metadata_space(root
, 1);
3314 trans
= btrfs_start_transaction(root
, 1);
3315 btrfs_set_trans_block_group(trans
, inode
);
3317 ret
= btrfs_orphan_add(trans
, inode
);
3320 nr
= trans
->blocks_used
;
3321 btrfs_end_transaction(trans
, root
);
3322 btrfs_unreserve_metadata_space(root
, 1);
3323 btrfs_btree_balance_dirty(root
, nr
);
3325 if (attr
->ia_size
> inode
->i_size
) {
3326 ret
= btrfs_cont_expand(inode
, attr
->ia_size
);
3328 btrfs_truncate(inode
);
3332 i_size_write(inode
, attr
->ia_size
);
3333 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
3335 trans
= btrfs_start_transaction(root
, 1);
3336 btrfs_set_trans_block_group(trans
, inode
);
3338 ret
= btrfs_update_inode(trans
, root
, inode
);
3340 if (inode
->i_nlink
> 0) {
3341 ret
= btrfs_orphan_del(trans
, inode
);
3344 nr
= trans
->blocks_used
;
3345 btrfs_end_transaction(trans
, root
);
3346 btrfs_btree_balance_dirty(root
, nr
);
3351 * We're truncating a file that used to have good data down to
3352 * zero. Make sure it gets into the ordered flush list so that
3353 * any new writes get down to disk quickly.
3355 if (attr
->ia_size
== 0)
3356 BTRFS_I(inode
)->ordered_data_close
= 1;
3358 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3359 ret
= vmtruncate(inode
, attr
->ia_size
);
3365 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3367 struct inode
*inode
= dentry
->d_inode
;
3370 err
= inode_change_ok(inode
, attr
);
3374 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3375 err
= btrfs_setattr_size(inode
, attr
);
3379 attr
->ia_valid
&= ~ATTR_SIZE
;
3382 err
= inode_setattr(inode
, attr
);
3384 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
3385 err
= btrfs_acl_chmod(inode
);
3389 void btrfs_delete_inode(struct inode
*inode
)
3391 struct btrfs_trans_handle
*trans
;
3392 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3396 truncate_inode_pages(&inode
->i_data
, 0);
3397 if (is_bad_inode(inode
)) {
3398 btrfs_orphan_del(NULL
, inode
);
3401 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3403 if (root
->fs_info
->log_root_recovering
) {
3404 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3408 if (inode
->i_nlink
> 0) {
3409 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3413 btrfs_i_size_write(inode
, 0);
3416 trans
= btrfs_start_transaction(root
, 1);
3417 btrfs_set_trans_block_group(trans
, inode
);
3418 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3423 nr
= trans
->blocks_used
;
3424 btrfs_end_transaction(trans
, root
);
3426 btrfs_btree_balance_dirty(root
, nr
);
3430 ret
= btrfs_orphan_del(trans
, inode
);
3434 nr
= trans
->blocks_used
;
3435 btrfs_end_transaction(trans
, root
);
3436 btrfs_btree_balance_dirty(root
, nr
);
3443 * this returns the key found in the dir entry in the location pointer.
3444 * If no dir entries were found, location->objectid is 0.
3446 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3447 struct btrfs_key
*location
)
3449 const char *name
= dentry
->d_name
.name
;
3450 int namelen
= dentry
->d_name
.len
;
3451 struct btrfs_dir_item
*di
;
3452 struct btrfs_path
*path
;
3453 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3456 path
= btrfs_alloc_path();
3459 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
3464 if (!di
|| IS_ERR(di
))
3467 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3469 btrfs_free_path(path
);
3472 location
->objectid
= 0;
3477 * when we hit a tree root in a directory, the btrfs part of the inode
3478 * needs to be changed to reflect the root directory of the tree root. This
3479 * is kind of like crossing a mount point.
3481 static int fixup_tree_root_location(struct btrfs_root
*root
,
3483 struct dentry
*dentry
,
3484 struct btrfs_key
*location
,
3485 struct btrfs_root
**sub_root
)
3487 struct btrfs_path
*path
;
3488 struct btrfs_root
*new_root
;
3489 struct btrfs_root_ref
*ref
;
3490 struct extent_buffer
*leaf
;
3494 path
= btrfs_alloc_path();
3501 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3502 BTRFS_I(dir
)->root
->root_key
.objectid
,
3503 location
->objectid
);
3510 leaf
= path
->nodes
[0];
3511 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3512 if (btrfs_root_ref_dirid(leaf
, ref
) != dir
->i_ino
||
3513 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3516 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3517 (unsigned long)(ref
+ 1),
3518 dentry
->d_name
.len
);
3522 btrfs_release_path(root
->fs_info
->tree_root
, path
);
3524 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3525 if (IS_ERR(new_root
)) {
3526 err
= PTR_ERR(new_root
);
3530 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3535 *sub_root
= new_root
;
3536 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3537 location
->type
= BTRFS_INODE_ITEM_KEY
;
3538 location
->offset
= 0;
3541 btrfs_free_path(path
);
3545 static void inode_tree_add(struct inode
*inode
)
3547 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3548 struct btrfs_inode
*entry
;
3550 struct rb_node
*parent
;
3552 p
= &root
->inode_tree
.rb_node
;
3555 if (hlist_unhashed(&inode
->i_hash
))
3558 spin_lock(&root
->inode_lock
);
3561 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3563 if (inode
->i_ino
< entry
->vfs_inode
.i_ino
)
3564 p
= &parent
->rb_left
;
3565 else if (inode
->i_ino
> entry
->vfs_inode
.i_ino
)
3566 p
= &parent
->rb_right
;
3568 WARN_ON(!(entry
->vfs_inode
.i_state
&
3569 (I_WILL_FREE
| I_FREEING
| I_CLEAR
)));
3570 rb_erase(parent
, &root
->inode_tree
);
3571 RB_CLEAR_NODE(parent
);
3572 spin_unlock(&root
->inode_lock
);
3576 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3577 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3578 spin_unlock(&root
->inode_lock
);
3581 static void inode_tree_del(struct inode
*inode
)
3583 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3586 spin_lock(&root
->inode_lock
);
3587 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3588 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3589 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3590 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3592 spin_unlock(&root
->inode_lock
);
3594 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
3595 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3596 spin_lock(&root
->inode_lock
);
3597 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3598 spin_unlock(&root
->inode_lock
);
3600 btrfs_add_dead_root(root
);
3604 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3606 struct rb_node
*node
;
3607 struct rb_node
*prev
;
3608 struct btrfs_inode
*entry
;
3609 struct inode
*inode
;
3612 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3614 spin_lock(&root
->inode_lock
);
3616 node
= root
->inode_tree
.rb_node
;
3620 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3622 if (objectid
< entry
->vfs_inode
.i_ino
)
3623 node
= node
->rb_left
;
3624 else if (objectid
> entry
->vfs_inode
.i_ino
)
3625 node
= node
->rb_right
;
3631 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
3632 if (objectid
<= entry
->vfs_inode
.i_ino
) {
3636 prev
= rb_next(prev
);
3640 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3641 objectid
= entry
->vfs_inode
.i_ino
+ 1;
3642 inode
= igrab(&entry
->vfs_inode
);
3644 spin_unlock(&root
->inode_lock
);
3645 if (atomic_read(&inode
->i_count
) > 1)
3646 d_prune_aliases(inode
);
3648 * btrfs_drop_inode will remove it from
3649 * the inode cache when its usage count
3654 spin_lock(&root
->inode_lock
);
3658 if (cond_resched_lock(&root
->inode_lock
))
3661 node
= rb_next(node
);
3663 spin_unlock(&root
->inode_lock
);
3667 static noinline
void init_btrfs_i(struct inode
*inode
)
3669 struct btrfs_inode
*bi
= BTRFS_I(inode
);
3674 bi
->last_sub_trans
= 0;
3675 bi
->logged_trans
= 0;
3676 bi
->delalloc_bytes
= 0;
3677 bi
->reserved_bytes
= 0;
3678 bi
->disk_i_size
= 0;
3680 bi
->index_cnt
= (u64
)-1;
3681 bi
->last_unlink_trans
= 0;
3682 bi
->ordered_data_close
= 0;
3683 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
3684 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
3685 inode
->i_mapping
, GFP_NOFS
);
3686 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
3687 inode
->i_mapping
, GFP_NOFS
);
3688 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
3689 INIT_LIST_HEAD(&BTRFS_I(inode
)->ordered_operations
);
3690 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3691 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
3692 mutex_init(&BTRFS_I(inode
)->log_mutex
);
3695 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3697 struct btrfs_iget_args
*args
= p
;
3698 inode
->i_ino
= args
->ino
;
3699 init_btrfs_i(inode
);
3700 BTRFS_I(inode
)->root
= args
->root
;
3701 btrfs_set_inode_space_info(args
->root
, inode
);
3705 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3707 struct btrfs_iget_args
*args
= opaque
;
3708 return args
->ino
== inode
->i_ino
&&
3709 args
->root
== BTRFS_I(inode
)->root
;
3712 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
3714 struct btrfs_root
*root
)
3716 struct inode
*inode
;
3717 struct btrfs_iget_args args
;
3718 args
.ino
= objectid
;
3721 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3722 btrfs_init_locked_inode
,
3727 /* Get an inode object given its location and corresponding root.
3728 * Returns in *is_new if the inode was read from disk
3730 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3731 struct btrfs_root
*root
)
3733 struct inode
*inode
;
3735 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3737 return ERR_PTR(-ENOMEM
);
3739 if (inode
->i_state
& I_NEW
) {
3740 BTRFS_I(inode
)->root
= root
;
3741 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3742 btrfs_read_locked_inode(inode
);
3744 inode_tree_add(inode
);
3745 unlock_new_inode(inode
);
3751 static struct inode
*new_simple_dir(struct super_block
*s
,
3752 struct btrfs_key
*key
,
3753 struct btrfs_root
*root
)
3755 struct inode
*inode
= new_inode(s
);
3758 return ERR_PTR(-ENOMEM
);
3760 init_btrfs_i(inode
);
3762 BTRFS_I(inode
)->root
= root
;
3763 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
3764 BTRFS_I(inode
)->dummy_inode
= 1;
3766 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
3767 inode
->i_op
= &simple_dir_inode_operations
;
3768 inode
->i_fop
= &simple_dir_operations
;
3769 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
3770 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3775 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3777 struct inode
*inode
;
3778 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3779 struct btrfs_root
*sub_root
= root
;
3780 struct btrfs_key location
;
3784 dentry
->d_op
= &btrfs_dentry_operations
;
3786 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3787 return ERR_PTR(-ENAMETOOLONG
);
3789 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3792 return ERR_PTR(ret
);
3794 if (location
.objectid
== 0)
3797 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
3798 inode
= btrfs_iget(dir
->i_sb
, &location
, root
);
3802 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
3804 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
3805 ret
= fixup_tree_root_location(root
, dir
, dentry
,
3806 &location
, &sub_root
);
3809 inode
= ERR_PTR(ret
);
3811 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
3813 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
);
3815 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
3817 if (root
!= sub_root
) {
3818 down_read(&root
->fs_info
->cleanup_work_sem
);
3819 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
3820 btrfs_orphan_cleanup(sub_root
);
3821 up_read(&root
->fs_info
->cleanup_work_sem
);
3827 static int btrfs_dentry_delete(struct dentry
*dentry
)
3829 struct btrfs_root
*root
;
3831 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
3832 dentry
= dentry
->d_parent
;
3834 if (dentry
->d_inode
) {
3835 root
= BTRFS_I(dentry
->d_inode
)->root
;
3836 if (btrfs_root_refs(&root
->root_item
) == 0)
3842 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
3843 struct nameidata
*nd
)
3845 struct inode
*inode
;
3847 inode
= btrfs_lookup_dentry(dir
, dentry
);
3849 return ERR_CAST(inode
);
3851 return d_splice_alias(inode
, dentry
);
3854 static unsigned char btrfs_filetype_table
[] = {
3855 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
3858 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
3861 struct inode
*inode
= filp
->f_dentry
->d_inode
;
3862 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3863 struct btrfs_item
*item
;
3864 struct btrfs_dir_item
*di
;
3865 struct btrfs_key key
;
3866 struct btrfs_key found_key
;
3867 struct btrfs_path
*path
;
3870 struct extent_buffer
*leaf
;
3873 unsigned char d_type
;
3878 int key_type
= BTRFS_DIR_INDEX_KEY
;
3883 /* FIXME, use a real flag for deciding about the key type */
3884 if (root
->fs_info
->tree_root
== root
)
3885 key_type
= BTRFS_DIR_ITEM_KEY
;
3887 /* special case for "." */
3888 if (filp
->f_pos
== 0) {
3889 over
= filldir(dirent
, ".", 1,
3896 /* special case for .., just use the back ref */
3897 if (filp
->f_pos
== 1) {
3898 u64 pino
= parent_ino(filp
->f_path
.dentry
);
3899 over
= filldir(dirent
, "..", 2,
3905 path
= btrfs_alloc_path();
3908 btrfs_set_key_type(&key
, key_type
);
3909 key
.offset
= filp
->f_pos
;
3910 key
.objectid
= inode
->i_ino
;
3912 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3918 leaf
= path
->nodes
[0];
3919 nritems
= btrfs_header_nritems(leaf
);
3920 slot
= path
->slots
[0];
3921 if (advance
|| slot
>= nritems
) {
3922 if (slot
>= nritems
- 1) {
3923 ret
= btrfs_next_leaf(root
, path
);
3926 leaf
= path
->nodes
[0];
3927 nritems
= btrfs_header_nritems(leaf
);
3928 slot
= path
->slots
[0];
3936 item
= btrfs_item_nr(leaf
, slot
);
3937 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3939 if (found_key
.objectid
!= key
.objectid
)
3941 if (btrfs_key_type(&found_key
) != key_type
)
3943 if (found_key
.offset
< filp
->f_pos
)
3946 filp
->f_pos
= found_key
.offset
;
3948 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
3950 di_total
= btrfs_item_size(leaf
, item
);
3952 while (di_cur
< di_total
) {
3953 struct btrfs_key location
;
3955 name_len
= btrfs_dir_name_len(leaf
, di
);
3956 if (name_len
<= sizeof(tmp_name
)) {
3957 name_ptr
= tmp_name
;
3959 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
3965 read_extent_buffer(leaf
, name_ptr
,
3966 (unsigned long)(di
+ 1), name_len
);
3968 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
3969 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
3971 /* is this a reference to our own snapshot? If so
3974 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
3975 location
.objectid
== root
->root_key
.objectid
) {
3979 over
= filldir(dirent
, name_ptr
, name_len
,
3980 found_key
.offset
, location
.objectid
,
3984 if (name_ptr
!= tmp_name
)
3989 di_len
= btrfs_dir_name_len(leaf
, di
) +
3990 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
3992 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
3996 /* Reached end of directory/root. Bump pos past the last item. */
3997 if (key_type
== BTRFS_DIR_INDEX_KEY
)
3998 filp
->f_pos
= INT_LIMIT(off_t
);
4004 btrfs_free_path(path
);
4008 int btrfs_write_inode(struct inode
*inode
, int wait
)
4010 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4011 struct btrfs_trans_handle
*trans
;
4014 if (root
->fs_info
->btree_inode
== inode
)
4018 trans
= btrfs_join_transaction(root
, 1);
4019 btrfs_set_trans_block_group(trans
, inode
);
4020 ret
= btrfs_commit_transaction(trans
, root
);
4026 * This is somewhat expensive, updating the tree every time the
4027 * inode changes. But, it is most likely to find the inode in cache.
4028 * FIXME, needs more benchmarking...there are no reasons other than performance
4029 * to keep or drop this code.
4031 void btrfs_dirty_inode(struct inode
*inode
)
4033 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4034 struct btrfs_trans_handle
*trans
;
4036 trans
= btrfs_join_transaction(root
, 1);
4037 btrfs_set_trans_block_group(trans
, inode
);
4038 btrfs_update_inode(trans
, root
, inode
);
4039 btrfs_end_transaction(trans
, root
);
4043 * find the highest existing sequence number in a directory
4044 * and then set the in-memory index_cnt variable to reflect
4045 * free sequence numbers
4047 static int btrfs_set_inode_index_count(struct inode
*inode
)
4049 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4050 struct btrfs_key key
, found_key
;
4051 struct btrfs_path
*path
;
4052 struct extent_buffer
*leaf
;
4055 key
.objectid
= inode
->i_ino
;
4056 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4057 key
.offset
= (u64
)-1;
4059 path
= btrfs_alloc_path();
4063 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4066 /* FIXME: we should be able to handle this */
4072 * MAGIC NUMBER EXPLANATION:
4073 * since we search a directory based on f_pos we have to start at 2
4074 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4075 * else has to start at 2
4077 if (path
->slots
[0] == 0) {
4078 BTRFS_I(inode
)->index_cnt
= 2;
4084 leaf
= path
->nodes
[0];
4085 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4087 if (found_key
.objectid
!= inode
->i_ino
||
4088 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4089 BTRFS_I(inode
)->index_cnt
= 2;
4093 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4095 btrfs_free_path(path
);
4100 * helper to find a free sequence number in a given directory. This current
4101 * code is very simple, later versions will do smarter things in the btree
4103 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4107 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4108 ret
= btrfs_set_inode_index_count(dir
);
4113 *index
= BTRFS_I(dir
)->index_cnt
;
4114 BTRFS_I(dir
)->index_cnt
++;
4119 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4120 struct btrfs_root
*root
,
4122 const char *name
, int name_len
,
4123 u64 ref_objectid
, u64 objectid
,
4124 u64 alloc_hint
, int mode
, u64
*index
)
4126 struct inode
*inode
;
4127 struct btrfs_inode_item
*inode_item
;
4128 struct btrfs_key
*location
;
4129 struct btrfs_path
*path
;
4130 struct btrfs_inode_ref
*ref
;
4131 struct btrfs_key key
[2];
4137 path
= btrfs_alloc_path();
4140 inode
= new_inode(root
->fs_info
->sb
);
4142 return ERR_PTR(-ENOMEM
);
4145 ret
= btrfs_set_inode_index(dir
, index
);
4148 return ERR_PTR(ret
);
4152 * index_cnt is ignored for everything but a dir,
4153 * btrfs_get_inode_index_count has an explanation for the magic
4156 init_btrfs_i(inode
);
4157 BTRFS_I(inode
)->index_cnt
= 2;
4158 BTRFS_I(inode
)->root
= root
;
4159 BTRFS_I(inode
)->generation
= trans
->transid
;
4160 btrfs_set_inode_space_info(root
, inode
);
4166 BTRFS_I(inode
)->block_group
=
4167 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
4169 key
[0].objectid
= objectid
;
4170 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4173 key
[1].objectid
= objectid
;
4174 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4175 key
[1].offset
= ref_objectid
;
4177 sizes
[0] = sizeof(struct btrfs_inode_item
);
4178 sizes
[1] = name_len
+ sizeof(*ref
);
4180 path
->leave_spinning
= 1;
4181 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4185 inode
->i_uid
= current_fsuid();
4187 if (dir
&& (dir
->i_mode
& S_ISGID
)) {
4188 inode
->i_gid
= dir
->i_gid
;
4192 inode
->i_gid
= current_fsgid();
4194 inode
->i_mode
= mode
;
4195 inode
->i_ino
= objectid
;
4196 inode_set_bytes(inode
, 0);
4197 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4198 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4199 struct btrfs_inode_item
);
4200 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4202 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4203 struct btrfs_inode_ref
);
4204 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4205 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4206 ptr
= (unsigned long)(ref
+ 1);
4207 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4209 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4210 btrfs_free_path(path
);
4212 location
= &BTRFS_I(inode
)->location
;
4213 location
->objectid
= objectid
;
4214 location
->offset
= 0;
4215 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4217 btrfs_inherit_iflags(inode
, dir
);
4219 if ((mode
& S_IFREG
)) {
4220 if (btrfs_test_opt(root
, NODATASUM
))
4221 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4222 if (btrfs_test_opt(root
, NODATACOW
))
4223 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4226 insert_inode_hash(inode
);
4227 inode_tree_add(inode
);
4231 BTRFS_I(dir
)->index_cnt
--;
4232 btrfs_free_path(path
);
4234 return ERR_PTR(ret
);
4237 static inline u8
btrfs_inode_type(struct inode
*inode
)
4239 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4243 * utility function to add 'inode' into 'parent_inode' with
4244 * a give name and a given sequence number.
4245 * if 'add_backref' is true, also insert a backref from the
4246 * inode to the parent directory.
4248 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4249 struct inode
*parent_inode
, struct inode
*inode
,
4250 const char *name
, int name_len
, int add_backref
, u64 index
)
4253 struct btrfs_key key
;
4254 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4256 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4257 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4259 key
.objectid
= inode
->i_ino
;
4260 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4264 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4265 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4266 key
.objectid
, root
->root_key
.objectid
,
4267 parent_inode
->i_ino
,
4268 index
, name
, name_len
);
4269 } else if (add_backref
) {
4270 ret
= btrfs_insert_inode_ref(trans
, root
,
4271 name
, name_len
, inode
->i_ino
,
4272 parent_inode
->i_ino
, index
);
4276 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4277 parent_inode
->i_ino
, &key
,
4278 btrfs_inode_type(inode
), index
);
4281 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4283 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4284 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4289 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4290 struct dentry
*dentry
, struct inode
*inode
,
4291 int backref
, u64 index
)
4293 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
4294 inode
, dentry
->d_name
.name
,
4295 dentry
->d_name
.len
, backref
, index
);
4297 d_instantiate(dentry
, inode
);
4305 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4306 int mode
, dev_t rdev
)
4308 struct btrfs_trans_handle
*trans
;
4309 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4310 struct inode
*inode
= NULL
;
4314 unsigned long nr
= 0;
4317 if (!new_valid_dev(rdev
))
4321 * 2 for inode item and ref
4323 * 1 for xattr if selinux is on
4325 err
= btrfs_reserve_metadata_space(root
, 5);
4329 trans
= btrfs_start_transaction(root
, 1);
4332 btrfs_set_trans_block_group(trans
, dir
);
4334 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4340 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4342 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4343 BTRFS_I(dir
)->block_group
, mode
, &index
);
4344 err
= PTR_ERR(inode
);
4348 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4354 btrfs_set_trans_block_group(trans
, inode
);
4355 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4359 inode
->i_op
= &btrfs_special_inode_operations
;
4360 init_special_inode(inode
, inode
->i_mode
, rdev
);
4361 btrfs_update_inode(trans
, root
, inode
);
4363 btrfs_update_inode_block_group(trans
, inode
);
4364 btrfs_update_inode_block_group(trans
, dir
);
4366 nr
= trans
->blocks_used
;
4367 btrfs_end_transaction_throttle(trans
, root
);
4369 btrfs_unreserve_metadata_space(root
, 5);
4371 inode_dec_link_count(inode
);
4374 btrfs_btree_balance_dirty(root
, nr
);
4378 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4379 int mode
, struct nameidata
*nd
)
4381 struct btrfs_trans_handle
*trans
;
4382 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4383 struct inode
*inode
= NULL
;
4386 unsigned long nr
= 0;
4391 * 2 for inode item and ref
4393 * 1 for xattr if selinux is on
4395 err
= btrfs_reserve_metadata_space(root
, 5);
4399 trans
= btrfs_start_transaction(root
, 1);
4402 btrfs_set_trans_block_group(trans
, dir
);
4404 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4410 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4412 dentry
->d_parent
->d_inode
->i_ino
,
4413 objectid
, BTRFS_I(dir
)->block_group
, mode
,
4415 err
= PTR_ERR(inode
);
4419 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4425 btrfs_set_trans_block_group(trans
, inode
);
4426 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4430 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4431 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4432 inode
->i_fop
= &btrfs_file_operations
;
4433 inode
->i_op
= &btrfs_file_inode_operations
;
4434 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4436 btrfs_update_inode_block_group(trans
, inode
);
4437 btrfs_update_inode_block_group(trans
, dir
);
4439 nr
= trans
->blocks_used
;
4440 btrfs_end_transaction_throttle(trans
, root
);
4442 btrfs_unreserve_metadata_space(root
, 5);
4444 inode_dec_link_count(inode
);
4447 btrfs_btree_balance_dirty(root
, nr
);
4451 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4452 struct dentry
*dentry
)
4454 struct btrfs_trans_handle
*trans
;
4455 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4456 struct inode
*inode
= old_dentry
->d_inode
;
4458 unsigned long nr
= 0;
4462 if (inode
->i_nlink
== 0)
4465 /* do not allow sys_link's with other subvols of the same device */
4466 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4470 * 1 item for inode ref
4471 * 2 items for dir items
4473 err
= btrfs_reserve_metadata_space(root
, 3);
4477 btrfs_inc_nlink(inode
);
4479 err
= btrfs_set_inode_index(dir
, &index
);
4483 trans
= btrfs_start_transaction(root
, 1);
4485 btrfs_set_trans_block_group(trans
, dir
);
4486 atomic_inc(&inode
->i_count
);
4488 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
4493 btrfs_update_inode_block_group(trans
, dir
);
4494 err
= btrfs_update_inode(trans
, root
, inode
);
4496 btrfs_log_new_name(trans
, inode
, NULL
, dentry
->d_parent
);
4499 nr
= trans
->blocks_used
;
4500 btrfs_end_transaction_throttle(trans
, root
);
4502 btrfs_unreserve_metadata_space(root
, 3);
4504 inode_dec_link_count(inode
);
4507 btrfs_btree_balance_dirty(root
, nr
);
4511 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4513 struct inode
*inode
= NULL
;
4514 struct btrfs_trans_handle
*trans
;
4515 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4517 int drop_on_err
= 0;
4520 unsigned long nr
= 1;
4523 * 2 items for inode and ref
4524 * 2 items for dir items
4525 * 1 for xattr if selinux is on
4527 err
= btrfs_reserve_metadata_space(root
, 5);
4531 trans
= btrfs_start_transaction(root
, 1);
4536 btrfs_set_trans_block_group(trans
, dir
);
4538 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4544 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4546 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4547 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
4549 if (IS_ERR(inode
)) {
4550 err
= PTR_ERR(inode
);
4556 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4560 inode
->i_op
= &btrfs_dir_inode_operations
;
4561 inode
->i_fop
= &btrfs_dir_file_operations
;
4562 btrfs_set_trans_block_group(trans
, inode
);
4564 btrfs_i_size_write(inode
, 0);
4565 err
= btrfs_update_inode(trans
, root
, inode
);
4569 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
4570 inode
, dentry
->d_name
.name
,
4571 dentry
->d_name
.len
, 0, index
);
4575 d_instantiate(dentry
, inode
);
4577 btrfs_update_inode_block_group(trans
, inode
);
4578 btrfs_update_inode_block_group(trans
, dir
);
4581 nr
= trans
->blocks_used
;
4582 btrfs_end_transaction_throttle(trans
, root
);
4585 btrfs_unreserve_metadata_space(root
, 5);
4588 btrfs_btree_balance_dirty(root
, nr
);
4592 /* helper for btfs_get_extent. Given an existing extent in the tree,
4593 * and an extent that you want to insert, deal with overlap and insert
4594 * the new extent into the tree.
4596 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4597 struct extent_map
*existing
,
4598 struct extent_map
*em
,
4599 u64 map_start
, u64 map_len
)
4603 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4604 start_diff
= map_start
- em
->start
;
4605 em
->start
= map_start
;
4607 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4608 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4609 em
->block_start
+= start_diff
;
4610 em
->block_len
-= start_diff
;
4612 return add_extent_mapping(em_tree
, em
);
4615 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4616 struct inode
*inode
, struct page
*page
,
4617 size_t pg_offset
, u64 extent_offset
,
4618 struct btrfs_file_extent_item
*item
)
4621 struct extent_buffer
*leaf
= path
->nodes
[0];
4624 unsigned long inline_size
;
4627 WARN_ON(pg_offset
!= 0);
4628 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4629 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4630 btrfs_item_nr(leaf
, path
->slots
[0]));
4631 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4632 ptr
= btrfs_file_extent_inline_start(item
);
4634 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4636 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4637 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
4638 inline_size
, max_size
);
4640 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4641 unsigned long copy_size
= min_t(u64
,
4642 PAGE_CACHE_SIZE
- pg_offset
,
4643 max_size
- extent_offset
);
4644 memset(kaddr
+ pg_offset
, 0, copy_size
);
4645 kunmap_atomic(kaddr
, KM_USER0
);
4652 * a bit scary, this does extent mapping from logical file offset to the disk.
4653 * the ugly parts come from merging extents from the disk with the in-ram
4654 * representation. This gets more complex because of the data=ordered code,
4655 * where the in-ram extents might be locked pending data=ordered completion.
4657 * This also copies inline extents directly into the page.
4660 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
4661 size_t pg_offset
, u64 start
, u64 len
,
4667 u64 extent_start
= 0;
4669 u64 objectid
= inode
->i_ino
;
4671 struct btrfs_path
*path
= NULL
;
4672 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4673 struct btrfs_file_extent_item
*item
;
4674 struct extent_buffer
*leaf
;
4675 struct btrfs_key found_key
;
4676 struct extent_map
*em
= NULL
;
4677 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4678 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4679 struct btrfs_trans_handle
*trans
= NULL
;
4683 read_lock(&em_tree
->lock
);
4684 em
= lookup_extent_mapping(em_tree
, start
, len
);
4686 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4687 read_unlock(&em_tree
->lock
);
4690 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
4691 free_extent_map(em
);
4692 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
4693 free_extent_map(em
);
4697 em
= alloc_extent_map(GFP_NOFS
);
4702 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4703 em
->start
= EXTENT_MAP_HOLE
;
4704 em
->orig_start
= EXTENT_MAP_HOLE
;
4706 em
->block_len
= (u64
)-1;
4709 path
= btrfs_alloc_path();
4713 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
4714 objectid
, start
, trans
!= NULL
);
4721 if (path
->slots
[0] == 0)
4726 leaf
= path
->nodes
[0];
4727 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4728 struct btrfs_file_extent_item
);
4729 /* are we inside the extent that was found? */
4730 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4731 found_type
= btrfs_key_type(&found_key
);
4732 if (found_key
.objectid
!= objectid
||
4733 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4737 found_type
= btrfs_file_extent_type(leaf
, item
);
4738 extent_start
= found_key
.offset
;
4739 compressed
= btrfs_file_extent_compression(leaf
, item
);
4740 if (found_type
== BTRFS_FILE_EXTENT_REG
||
4741 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4742 extent_end
= extent_start
+
4743 btrfs_file_extent_num_bytes(leaf
, item
);
4744 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4746 size
= btrfs_file_extent_inline_len(leaf
, item
);
4747 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
4748 ~((u64
)root
->sectorsize
- 1);
4751 if (start
>= extent_end
) {
4753 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
4754 ret
= btrfs_next_leaf(root
, path
);
4761 leaf
= path
->nodes
[0];
4763 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4764 if (found_key
.objectid
!= objectid
||
4765 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4767 if (start
+ len
<= found_key
.offset
)
4770 em
->len
= found_key
.offset
- start
;
4774 if (found_type
== BTRFS_FILE_EXTENT_REG
||
4775 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4776 em
->start
= extent_start
;
4777 em
->len
= extent_end
- extent_start
;
4778 em
->orig_start
= extent_start
-
4779 btrfs_file_extent_offset(leaf
, item
);
4780 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
4782 em
->block_start
= EXTENT_MAP_HOLE
;
4786 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4787 em
->block_start
= bytenr
;
4788 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
4791 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
4792 em
->block_start
= bytenr
;
4793 em
->block_len
= em
->len
;
4794 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
4795 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
4798 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4802 size_t extent_offset
;
4805 em
->block_start
= EXTENT_MAP_INLINE
;
4806 if (!page
|| create
) {
4807 em
->start
= extent_start
;
4808 em
->len
= extent_end
- extent_start
;
4812 size
= btrfs_file_extent_inline_len(leaf
, item
);
4813 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
4814 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
4815 size
- extent_offset
);
4816 em
->start
= extent_start
+ extent_offset
;
4817 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
4818 ~((u64
)root
->sectorsize
- 1);
4819 em
->orig_start
= EXTENT_MAP_INLINE
;
4821 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4822 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
4823 if (create
== 0 && !PageUptodate(page
)) {
4824 if (btrfs_file_extent_compression(leaf
, item
) ==
4825 BTRFS_COMPRESS_ZLIB
) {
4826 ret
= uncompress_inline(path
, inode
, page
,
4828 extent_offset
, item
);
4832 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4834 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
4835 memset(map
+ pg_offset
+ copy_size
, 0,
4836 PAGE_CACHE_SIZE
- pg_offset
-
4841 flush_dcache_page(page
);
4842 } else if (create
&& PageUptodate(page
)) {
4845 free_extent_map(em
);
4847 btrfs_release_path(root
, path
);
4848 trans
= btrfs_join_transaction(root
, 1);
4852 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4855 btrfs_mark_buffer_dirty(leaf
);
4857 set_extent_uptodate(io_tree
, em
->start
,
4858 extent_map_end(em
) - 1, GFP_NOFS
);
4861 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
4868 em
->block_start
= EXTENT_MAP_HOLE
;
4869 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
4871 btrfs_release_path(root
, path
);
4872 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
4873 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
4874 "[%llu %llu]\n", (unsigned long long)em
->start
,
4875 (unsigned long long)em
->len
,
4876 (unsigned long long)start
,
4877 (unsigned long long)len
);
4883 write_lock(&em_tree
->lock
);
4884 ret
= add_extent_mapping(em_tree
, em
);
4885 /* it is possible that someone inserted the extent into the tree
4886 * while we had the lock dropped. It is also possible that
4887 * an overlapping map exists in the tree
4889 if (ret
== -EEXIST
) {
4890 struct extent_map
*existing
;
4894 existing
= lookup_extent_mapping(em_tree
, start
, len
);
4895 if (existing
&& (existing
->start
> start
||
4896 existing
->start
+ existing
->len
<= start
)) {
4897 free_extent_map(existing
);
4901 existing
= lookup_extent_mapping(em_tree
, em
->start
,
4904 err
= merge_extent_mapping(em_tree
, existing
,
4907 free_extent_map(existing
);
4909 free_extent_map(em
);
4914 free_extent_map(em
);
4918 free_extent_map(em
);
4923 write_unlock(&em_tree
->lock
);
4926 btrfs_free_path(path
);
4928 ret
= btrfs_end_transaction(trans
, root
);
4933 free_extent_map(em
);
4934 return ERR_PTR(err
);
4939 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
4940 const struct iovec
*iov
, loff_t offset
,
4941 unsigned long nr_segs
)
4946 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4947 __u64 start
, __u64 len
)
4949 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent
);
4952 int btrfs_readpage(struct file
*file
, struct page
*page
)
4954 struct extent_io_tree
*tree
;
4955 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4956 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
4959 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
4961 struct extent_io_tree
*tree
;
4964 if (current
->flags
& PF_MEMALLOC
) {
4965 redirty_page_for_writepage(wbc
, page
);
4969 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4970 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
4973 int btrfs_writepages(struct address_space
*mapping
,
4974 struct writeback_control
*wbc
)
4976 struct extent_io_tree
*tree
;
4978 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4979 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
4983 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
4984 struct list_head
*pages
, unsigned nr_pages
)
4986 struct extent_io_tree
*tree
;
4987 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4988 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
4991 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4993 struct extent_io_tree
*tree
;
4994 struct extent_map_tree
*map
;
4997 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4998 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
4999 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
5001 ClearPagePrivate(page
);
5002 set_page_private(page
, 0);
5003 page_cache_release(page
);
5008 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
5010 if (PageWriteback(page
) || PageDirty(page
))
5012 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
5015 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
5017 struct extent_io_tree
*tree
;
5018 struct btrfs_ordered_extent
*ordered
;
5019 u64 page_start
= page_offset(page
);
5020 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
5024 * we have the page locked, so new writeback can't start,
5025 * and the dirty bit won't be cleared while we are here.
5027 * Wait for IO on this page so that we can safely clear
5028 * the PagePrivate2 bit and do ordered accounting
5030 wait_on_page_writeback(page
);
5032 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
5034 btrfs_releasepage(page
, GFP_NOFS
);
5037 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
5038 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
5042 * IO on this page will never be started, so we need
5043 * to account for any ordered extents now
5045 clear_extent_bit(tree
, page_start
, page_end
,
5046 EXTENT_DIRTY
| EXTENT_DELALLOC
|
5047 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
5050 * whoever cleared the private bit is responsible
5051 * for the finish_ordered_io
5053 if (TestClearPagePrivate2(page
)) {
5054 btrfs_finish_ordered_io(page
->mapping
->host
,
5055 page_start
, page_end
);
5057 btrfs_put_ordered_extent(ordered
);
5058 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
5060 clear_extent_bit(tree
, page_start
, page_end
,
5061 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
5062 EXTENT_DO_ACCOUNTING
, 1, 1, NULL
, GFP_NOFS
);
5063 __btrfs_releasepage(page
, GFP_NOFS
);
5065 ClearPageChecked(page
);
5066 if (PagePrivate(page
)) {
5067 ClearPagePrivate(page
);
5068 set_page_private(page
, 0);
5069 page_cache_release(page
);
5074 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5075 * called from a page fault handler when a page is first dirtied. Hence we must
5076 * be careful to check for EOF conditions here. We set the page up correctly
5077 * for a written page which means we get ENOSPC checking when writing into
5078 * holes and correct delalloc and unwritten extent mapping on filesystems that
5079 * support these features.
5081 * We are not allowed to take the i_mutex here so we have to play games to
5082 * protect against truncate races as the page could now be beyond EOF. Because
5083 * vmtruncate() writes the inode size before removing pages, once we have the
5084 * page lock we can determine safely if the page is beyond EOF. If it is not
5085 * beyond EOF, then the page is guaranteed safe against truncation until we
5088 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5090 struct page
*page
= vmf
->page
;
5091 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
5092 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5093 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5094 struct btrfs_ordered_extent
*ordered
;
5096 unsigned long zero_start
;
5102 ret
= btrfs_check_data_free_space(root
, inode
, PAGE_CACHE_SIZE
);
5106 else /* -ENOSPC, -EIO, etc */
5107 ret
= VM_FAULT_SIGBUS
;
5111 ret
= btrfs_reserve_metadata_for_delalloc(root
, inode
, 1);
5113 btrfs_free_reserved_data_space(root
, inode
, PAGE_CACHE_SIZE
);
5114 ret
= VM_FAULT_SIGBUS
;
5118 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
5121 size
= i_size_read(inode
);
5122 page_start
= page_offset(page
);
5123 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
5125 if ((page
->mapping
!= inode
->i_mapping
) ||
5126 (page_start
>= size
)) {
5127 btrfs_free_reserved_data_space(root
, inode
, PAGE_CACHE_SIZE
);
5128 /* page got truncated out from underneath us */
5131 wait_on_page_writeback(page
);
5133 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5134 set_page_extent_mapped(page
);
5137 * we can't set the delalloc bits if there are pending ordered
5138 * extents. Drop our locks and wait for them to finish
5140 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
5142 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5144 btrfs_start_ordered_extent(inode
, ordered
, 1);
5145 btrfs_put_ordered_extent(ordered
);
5150 * XXX - page_mkwrite gets called every time the page is dirtied, even
5151 * if it was already dirty, so for space accounting reasons we need to
5152 * clear any delalloc bits for the range we are fixing to save. There
5153 * is probably a better way to do this, but for now keep consistent with
5154 * prepare_pages in the normal write path.
5156 clear_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
5157 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
5160 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
5162 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5163 ret
= VM_FAULT_SIGBUS
;
5164 btrfs_free_reserved_data_space(root
, inode
, PAGE_CACHE_SIZE
);
5169 /* page is wholly or partially inside EOF */
5170 if (page_start
+ PAGE_CACHE_SIZE
> size
)
5171 zero_start
= size
& ~PAGE_CACHE_MASK
;
5173 zero_start
= PAGE_CACHE_SIZE
;
5175 if (zero_start
!= PAGE_CACHE_SIZE
) {
5177 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
5178 flush_dcache_page(page
);
5181 ClearPageChecked(page
);
5182 set_page_dirty(page
);
5183 SetPageUptodate(page
);
5185 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
5186 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
5188 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
5191 btrfs_unreserve_metadata_for_delalloc(root
, inode
, 1);
5193 return VM_FAULT_LOCKED
;
5199 static void btrfs_truncate(struct inode
*inode
)
5201 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5203 struct btrfs_trans_handle
*trans
;
5205 u64 mask
= root
->sectorsize
- 1;
5207 if (!S_ISREG(inode
->i_mode
)) {
5212 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
5216 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
5217 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
5219 trans
= btrfs_start_transaction(root
, 1);
5220 btrfs_set_trans_block_group(trans
, inode
);
5223 * setattr is responsible for setting the ordered_data_close flag,
5224 * but that is only tested during the last file release. That
5225 * could happen well after the next commit, leaving a great big
5226 * window where new writes may get lost if someone chooses to write
5227 * to this file after truncating to zero
5229 * The inode doesn't have any dirty data here, and so if we commit
5230 * this is a noop. If someone immediately starts writing to the inode
5231 * it is very likely we'll catch some of their writes in this
5232 * transaction, and the commit will find this file on the ordered
5233 * data list with good things to send down.
5235 * This is a best effort solution, there is still a window where
5236 * using truncate to replace the contents of the file will
5237 * end up with a zero length file after a crash.
5239 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
5240 btrfs_add_ordered_operation(trans
, root
, inode
);
5243 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
5245 BTRFS_EXTENT_DATA_KEY
);
5249 ret
= btrfs_update_inode(trans
, root
, inode
);
5252 nr
= trans
->blocks_used
;
5253 btrfs_end_transaction(trans
, root
);
5254 btrfs_btree_balance_dirty(root
, nr
);
5256 trans
= btrfs_start_transaction(root
, 1);
5257 btrfs_set_trans_block_group(trans
, inode
);
5260 if (ret
== 0 && inode
->i_nlink
> 0) {
5261 ret
= btrfs_orphan_del(trans
, inode
);
5265 ret
= btrfs_update_inode(trans
, root
, inode
);
5268 nr
= trans
->blocks_used
;
5269 ret
= btrfs_end_transaction_throttle(trans
, root
);
5271 btrfs_btree_balance_dirty(root
, nr
);
5275 * create a new subvolume directory/inode (helper for the ioctl).
5277 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
5278 struct btrfs_root
*new_root
,
5279 u64 new_dirid
, u64 alloc_hint
)
5281 struct inode
*inode
;
5285 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
5286 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
5288 return PTR_ERR(inode
);
5289 inode
->i_op
= &btrfs_dir_inode_operations
;
5290 inode
->i_fop
= &btrfs_dir_file_operations
;
5293 btrfs_i_size_write(inode
, 0);
5295 err
= btrfs_update_inode(trans
, new_root
, inode
);
5302 /* helper function for file defrag and space balancing. This
5303 * forces readahead on a given range of bytes in an inode
5305 unsigned long btrfs_force_ra(struct address_space
*mapping
,
5306 struct file_ra_state
*ra
, struct file
*file
,
5307 pgoff_t offset
, pgoff_t last_index
)
5309 pgoff_t req_size
= last_index
- offset
+ 1;
5311 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
5312 return offset
+ req_size
;
5315 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
5317 struct btrfs_inode
*ei
;
5319 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
5323 ei
->last_sub_trans
= 0;
5324 ei
->logged_trans
= 0;
5325 ei
->outstanding_extents
= 0;
5326 ei
->reserved_extents
= 0;
5328 spin_lock_init(&ei
->accounting_lock
);
5329 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
5330 INIT_LIST_HEAD(&ei
->i_orphan
);
5331 INIT_LIST_HEAD(&ei
->ordered_operations
);
5332 return &ei
->vfs_inode
;
5335 void btrfs_destroy_inode(struct inode
*inode
)
5337 struct btrfs_ordered_extent
*ordered
;
5338 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5340 WARN_ON(!list_empty(&inode
->i_dentry
));
5341 WARN_ON(inode
->i_data
.nrpages
);
5344 * This can happen where we create an inode, but somebody else also
5345 * created the same inode and we need to destroy the one we already
5352 * Make sure we're properly removed from the ordered operation
5356 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
5357 spin_lock(&root
->fs_info
->ordered_extent_lock
);
5358 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
5359 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
5362 spin_lock(&root
->list_lock
);
5363 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
5364 printk(KERN_INFO
"BTRFS: inode %lu still on the orphan list\n",
5366 list_del_init(&BTRFS_I(inode
)->i_orphan
);
5368 spin_unlock(&root
->list_lock
);
5371 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
5375 printk(KERN_ERR
"btrfs found ordered "
5376 "extent %llu %llu on inode cleanup\n",
5377 (unsigned long long)ordered
->file_offset
,
5378 (unsigned long long)ordered
->len
);
5379 btrfs_remove_ordered_extent(inode
, ordered
);
5380 btrfs_put_ordered_extent(ordered
);
5381 btrfs_put_ordered_extent(ordered
);
5384 inode_tree_del(inode
);
5385 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
5387 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
5390 void btrfs_drop_inode(struct inode
*inode
)
5392 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5394 if (inode
->i_nlink
> 0 && btrfs_root_refs(&root
->root_item
) == 0)
5395 generic_delete_inode(inode
);
5397 generic_drop_inode(inode
);
5400 static void init_once(void *foo
)
5402 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
5404 inode_init_once(&ei
->vfs_inode
);
5407 void btrfs_destroy_cachep(void)
5409 if (btrfs_inode_cachep
)
5410 kmem_cache_destroy(btrfs_inode_cachep
);
5411 if (btrfs_trans_handle_cachep
)
5412 kmem_cache_destroy(btrfs_trans_handle_cachep
);
5413 if (btrfs_transaction_cachep
)
5414 kmem_cache_destroy(btrfs_transaction_cachep
);
5415 if (btrfs_path_cachep
)
5416 kmem_cache_destroy(btrfs_path_cachep
);
5419 int btrfs_init_cachep(void)
5421 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
5422 sizeof(struct btrfs_inode
), 0,
5423 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
5424 if (!btrfs_inode_cachep
)
5427 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
5428 sizeof(struct btrfs_trans_handle
), 0,
5429 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
5430 if (!btrfs_trans_handle_cachep
)
5433 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
5434 sizeof(struct btrfs_transaction
), 0,
5435 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
5436 if (!btrfs_transaction_cachep
)
5439 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
5440 sizeof(struct btrfs_path
), 0,
5441 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
5442 if (!btrfs_path_cachep
)
5447 btrfs_destroy_cachep();
5451 static int btrfs_getattr(struct vfsmount
*mnt
,
5452 struct dentry
*dentry
, struct kstat
*stat
)
5454 struct inode
*inode
= dentry
->d_inode
;
5455 generic_fillattr(inode
, stat
);
5456 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
5457 stat
->blksize
= PAGE_CACHE_SIZE
;
5458 stat
->blocks
= (inode_get_bytes(inode
) +
5459 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
5463 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
5464 struct inode
*new_dir
, struct dentry
*new_dentry
)
5466 struct btrfs_trans_handle
*trans
;
5467 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
5468 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
5469 struct inode
*new_inode
= new_dentry
->d_inode
;
5470 struct inode
*old_inode
= old_dentry
->d_inode
;
5471 struct timespec ctime
= CURRENT_TIME
;
5476 if (new_dir
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5479 /* we only allow rename subvolume link between subvolumes */
5480 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
5483 if (old_inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
5484 (new_inode
&& new_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
))
5487 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
5488 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
5492 * We want to reserve the absolute worst case amount of items. So if
5493 * both inodes are subvols and we need to unlink them then that would
5494 * require 4 item modifications, but if they are both normal inodes it
5495 * would require 5 item modifications, so we'll assume their normal
5496 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5497 * should cover the worst case number of items we'll modify.
5499 ret
= btrfs_reserve_metadata_space(root
, 11);
5504 * we're using rename to replace one file with another.
5505 * and the replacement file is large. Start IO on it now so
5506 * we don't add too much work to the end of the transaction
5508 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
5509 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
5510 filemap_flush(old_inode
->i_mapping
);
5512 /* close the racy window with snapshot create/destroy ioctl */
5513 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
5514 down_read(&root
->fs_info
->subvol_sem
);
5516 trans
= btrfs_start_transaction(root
, 1);
5517 btrfs_set_trans_block_group(trans
, new_dir
);
5520 btrfs_record_root_in_trans(trans
, dest
);
5522 ret
= btrfs_set_inode_index(new_dir
, &index
);
5526 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5527 /* force full log commit if subvolume involved. */
5528 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
5530 ret
= btrfs_insert_inode_ref(trans
, dest
,
5531 new_dentry
->d_name
.name
,
5532 new_dentry
->d_name
.len
,
5534 new_dir
->i_ino
, index
);
5538 * this is an ugly little race, but the rename is required
5539 * to make sure that if we crash, the inode is either at the
5540 * old name or the new one. pinning the log transaction lets
5541 * us make sure we don't allow a log commit to come in after
5542 * we unlink the name but before we add the new name back in.
5544 btrfs_pin_log_trans(root
);
5547 * make sure the inode gets flushed if it is replacing
5550 if (new_inode
&& new_inode
->i_size
&&
5551 old_inode
&& S_ISREG(old_inode
->i_mode
)) {
5552 btrfs_add_ordered_operation(trans
, root
, old_inode
);
5555 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
5556 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
5557 old_inode
->i_ctime
= ctime
;
5559 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
5560 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
5562 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5563 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
5564 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
5565 old_dentry
->d_name
.name
,
5566 old_dentry
->d_name
.len
);
5568 btrfs_inc_nlink(old_dentry
->d_inode
);
5569 ret
= btrfs_unlink_inode(trans
, root
, old_dir
,
5570 old_dentry
->d_inode
,
5571 old_dentry
->d_name
.name
,
5572 old_dentry
->d_name
.len
);
5577 new_inode
->i_ctime
= CURRENT_TIME
;
5578 if (unlikely(new_inode
->i_ino
==
5579 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
5580 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
5581 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
5583 new_dentry
->d_name
.name
,
5584 new_dentry
->d_name
.len
);
5585 BUG_ON(new_inode
->i_nlink
== 0);
5587 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
5588 new_dentry
->d_inode
,
5589 new_dentry
->d_name
.name
,
5590 new_dentry
->d_name
.len
);
5593 if (new_inode
->i_nlink
== 0) {
5594 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
5599 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
5600 new_dentry
->d_name
.name
,
5601 new_dentry
->d_name
.len
, 0, index
);
5604 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
5605 btrfs_log_new_name(trans
, old_inode
, old_dir
,
5606 new_dentry
->d_parent
);
5607 btrfs_end_log_trans(root
);
5610 btrfs_end_transaction_throttle(trans
, root
);
5612 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
5613 up_read(&root
->fs_info
->subvol_sem
);
5615 btrfs_unreserve_metadata_space(root
, 11);
5620 * some fairly slow code that needs optimization. This walks the list
5621 * of all the inodes with pending delalloc and forces them to disk.
5623 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
5625 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
5626 struct btrfs_inode
*binode
;
5627 struct inode
*inode
;
5629 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
5632 spin_lock(&root
->fs_info
->delalloc_lock
);
5633 while (!list_empty(head
)) {
5634 binode
= list_entry(head
->next
, struct btrfs_inode
,
5636 inode
= igrab(&binode
->vfs_inode
);
5638 list_del_init(&binode
->delalloc_inodes
);
5639 spin_unlock(&root
->fs_info
->delalloc_lock
);
5641 filemap_flush(inode
->i_mapping
);
5643 btrfs_add_delayed_iput(inode
);
5648 spin_lock(&root
->fs_info
->delalloc_lock
);
5650 spin_unlock(&root
->fs_info
->delalloc_lock
);
5652 /* the filemap_flush will queue IO into the worker threads, but
5653 * we have to make sure the IO is actually started and that
5654 * ordered extents get created before we return
5656 atomic_inc(&root
->fs_info
->async_submit_draining
);
5657 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
5658 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
5659 wait_event(root
->fs_info
->async_submit_wait
,
5660 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
5661 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
5663 atomic_dec(&root
->fs_info
->async_submit_draining
);
5667 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
5668 const char *symname
)
5670 struct btrfs_trans_handle
*trans
;
5671 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5672 struct btrfs_path
*path
;
5673 struct btrfs_key key
;
5674 struct inode
*inode
= NULL
;
5682 struct btrfs_file_extent_item
*ei
;
5683 struct extent_buffer
*leaf
;
5684 unsigned long nr
= 0;
5686 name_len
= strlen(symname
) + 1;
5687 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
5688 return -ENAMETOOLONG
;
5691 * 2 items for inode item and ref
5692 * 2 items for dir items
5693 * 1 item for xattr if selinux is on
5695 err
= btrfs_reserve_metadata_space(root
, 5);
5699 trans
= btrfs_start_transaction(root
, 1);
5702 btrfs_set_trans_block_group(trans
, dir
);
5704 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
5710 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5712 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
5713 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
5715 err
= PTR_ERR(inode
);
5719 err
= btrfs_init_inode_security(trans
, inode
, dir
);
5725 btrfs_set_trans_block_group(trans
, inode
);
5726 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
5730 inode
->i_mapping
->a_ops
= &btrfs_aops
;
5731 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
5732 inode
->i_fop
= &btrfs_file_operations
;
5733 inode
->i_op
= &btrfs_file_inode_operations
;
5734 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
5736 btrfs_update_inode_block_group(trans
, inode
);
5737 btrfs_update_inode_block_group(trans
, dir
);
5741 path
= btrfs_alloc_path();
5743 key
.objectid
= inode
->i_ino
;
5745 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
5746 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
5747 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
5753 leaf
= path
->nodes
[0];
5754 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
5755 struct btrfs_file_extent_item
);
5756 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
5757 btrfs_set_file_extent_type(leaf
, ei
,
5758 BTRFS_FILE_EXTENT_INLINE
);
5759 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
5760 btrfs_set_file_extent_compression(leaf
, ei
, 0);
5761 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
5762 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
5764 ptr
= btrfs_file_extent_inline_start(ei
);
5765 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
5766 btrfs_mark_buffer_dirty(leaf
);
5767 btrfs_free_path(path
);
5769 inode
->i_op
= &btrfs_symlink_inode_operations
;
5770 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
5771 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
5772 inode_set_bytes(inode
, name_len
);
5773 btrfs_i_size_write(inode
, name_len
- 1);
5774 err
= btrfs_update_inode(trans
, root
, inode
);
5779 nr
= trans
->blocks_used
;
5780 btrfs_end_transaction_throttle(trans
, root
);
5782 btrfs_unreserve_metadata_space(root
, 5);
5784 inode_dec_link_count(inode
);
5787 btrfs_btree_balance_dirty(root
, nr
);
5791 static int prealloc_file_range(struct inode
*inode
, u64 start
, u64 end
,
5792 u64 alloc_hint
, int mode
)
5794 struct btrfs_trans_handle
*trans
;
5795 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5796 struct btrfs_key ins
;
5798 u64 cur_offset
= start
;
5799 u64 num_bytes
= end
- start
;
5802 while (num_bytes
> 0) {
5803 alloc_size
= min(num_bytes
, root
->fs_info
->max_extent
);
5805 trans
= btrfs_start_transaction(root
, 1);
5807 ret
= btrfs_reserve_extent(trans
, root
, alloc_size
,
5808 root
->sectorsize
, 0, alloc_hint
,
5815 ret
= btrfs_reserve_metadata_space(root
, 3);
5817 btrfs_free_reserved_extent(root
, ins
.objectid
,
5822 ret
= insert_reserved_file_extent(trans
, inode
,
5823 cur_offset
, ins
.objectid
,
5824 ins
.offset
, ins
.offset
,
5825 ins
.offset
, 0, 0, 0,
5826 BTRFS_FILE_EXTENT_PREALLOC
);
5828 btrfs_drop_extent_cache(inode
, cur_offset
,
5829 cur_offset
+ ins
.offset
-1, 0);
5831 num_bytes
-= ins
.offset
;
5832 cur_offset
+= ins
.offset
;
5833 alloc_hint
= ins
.objectid
+ ins
.offset
;
5835 inode
->i_ctime
= CURRENT_TIME
;
5836 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
5837 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
5838 cur_offset
> inode
->i_size
) {
5839 i_size_write(inode
, cur_offset
);
5840 btrfs_ordered_update_i_size(inode
, cur_offset
, NULL
);
5843 ret
= btrfs_update_inode(trans
, root
, inode
);
5846 btrfs_end_transaction(trans
, root
);
5847 btrfs_unreserve_metadata_space(root
, 3);
5852 btrfs_end_transaction(trans
, root
);
5857 static long btrfs_fallocate(struct inode
*inode
, int mode
,
5858 loff_t offset
, loff_t len
)
5866 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
5867 struct extent_map
*em
;
5870 alloc_start
= offset
& ~mask
;
5871 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
5874 * wait for ordered IO before we have any locks. We'll loop again
5875 * below with the locks held.
5877 btrfs_wait_ordered_range(inode
, alloc_start
, alloc_end
- alloc_start
);
5879 mutex_lock(&inode
->i_mutex
);
5880 if (alloc_start
> inode
->i_size
) {
5881 ret
= btrfs_cont_expand(inode
, alloc_start
);
5886 ret
= btrfs_check_data_free_space(BTRFS_I(inode
)->root
, inode
,
5887 alloc_end
- alloc_start
);
5891 locked_end
= alloc_end
- 1;
5893 struct btrfs_ordered_extent
*ordered
;
5895 /* the extent lock is ordered inside the running
5898 lock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
5900 ordered
= btrfs_lookup_first_ordered_extent(inode
,
5903 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
5904 ordered
->file_offset
< alloc_end
) {
5905 btrfs_put_ordered_extent(ordered
);
5906 unlock_extent(&BTRFS_I(inode
)->io_tree
,
5907 alloc_start
, locked_end
, GFP_NOFS
);
5909 * we can't wait on the range with the transaction
5910 * running or with the extent lock held
5912 btrfs_wait_ordered_range(inode
, alloc_start
,
5913 alloc_end
- alloc_start
);
5916 btrfs_put_ordered_extent(ordered
);
5921 cur_offset
= alloc_start
;
5923 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
5924 alloc_end
- cur_offset
, 0);
5925 BUG_ON(IS_ERR(em
) || !em
);
5926 last_byte
= min(extent_map_end(em
), alloc_end
);
5927 last_byte
= (last_byte
+ mask
) & ~mask
;
5928 if (em
->block_start
== EXTENT_MAP_HOLE
||
5929 (cur_offset
>= inode
->i_size
&&
5930 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5931 ret
= prealloc_file_range(inode
,
5932 cur_offset
, last_byte
,
5935 free_extent_map(em
);
5939 if (em
->block_start
<= EXTENT_MAP_LAST_BYTE
)
5940 alloc_hint
= em
->block_start
;
5941 free_extent_map(em
);
5943 cur_offset
= last_byte
;
5944 if (cur_offset
>= alloc_end
) {
5949 unlock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
5952 btrfs_free_reserved_data_space(BTRFS_I(inode
)->root
, inode
,
5953 alloc_end
- alloc_start
);
5955 mutex_unlock(&inode
->i_mutex
);
5959 static int btrfs_set_page_dirty(struct page
*page
)
5961 return __set_page_dirty_nobuffers(page
);
5964 static int btrfs_permission(struct inode
*inode
, int mask
)
5966 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
5968 return generic_permission(inode
, mask
, btrfs_check_acl
);
5971 static const struct inode_operations btrfs_dir_inode_operations
= {
5972 .getattr
= btrfs_getattr
,
5973 .lookup
= btrfs_lookup
,
5974 .create
= btrfs_create
,
5975 .unlink
= btrfs_unlink
,
5977 .mkdir
= btrfs_mkdir
,
5978 .rmdir
= btrfs_rmdir
,
5979 .rename
= btrfs_rename
,
5980 .symlink
= btrfs_symlink
,
5981 .setattr
= btrfs_setattr
,
5982 .mknod
= btrfs_mknod
,
5983 .setxattr
= btrfs_setxattr
,
5984 .getxattr
= btrfs_getxattr
,
5985 .listxattr
= btrfs_listxattr
,
5986 .removexattr
= btrfs_removexattr
,
5987 .permission
= btrfs_permission
,
5989 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
5990 .lookup
= btrfs_lookup
,
5991 .permission
= btrfs_permission
,
5994 static const struct file_operations btrfs_dir_file_operations
= {
5995 .llseek
= generic_file_llseek
,
5996 .read
= generic_read_dir
,
5997 .readdir
= btrfs_real_readdir
,
5998 .unlocked_ioctl
= btrfs_ioctl
,
5999 #ifdef CONFIG_COMPAT
6000 .compat_ioctl
= btrfs_ioctl
,
6002 .release
= btrfs_release_file
,
6003 .fsync
= btrfs_sync_file
,
6006 static struct extent_io_ops btrfs_extent_io_ops
= {
6007 .fill_delalloc
= run_delalloc_range
,
6008 .submit_bio_hook
= btrfs_submit_bio_hook
,
6009 .merge_bio_hook
= btrfs_merge_bio_hook
,
6010 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
6011 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
6012 .writepage_start_hook
= btrfs_writepage_start_hook
,
6013 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
6014 .set_bit_hook
= btrfs_set_bit_hook
,
6015 .clear_bit_hook
= btrfs_clear_bit_hook
,
6016 .merge_extent_hook
= btrfs_merge_extent_hook
,
6017 .split_extent_hook
= btrfs_split_extent_hook
,
6021 * btrfs doesn't support the bmap operation because swapfiles
6022 * use bmap to make a mapping of extents in the file. They assume
6023 * these extents won't change over the life of the file and they
6024 * use the bmap result to do IO directly to the drive.
6026 * the btrfs bmap call would return logical addresses that aren't
6027 * suitable for IO and they also will change frequently as COW
6028 * operations happen. So, swapfile + btrfs == corruption.
6030 * For now we're avoiding this by dropping bmap.
6032 static const struct address_space_operations btrfs_aops
= {
6033 .readpage
= btrfs_readpage
,
6034 .writepage
= btrfs_writepage
,
6035 .writepages
= btrfs_writepages
,
6036 .readpages
= btrfs_readpages
,
6037 .sync_page
= block_sync_page
,
6038 .direct_IO
= btrfs_direct_IO
,
6039 .invalidatepage
= btrfs_invalidatepage
,
6040 .releasepage
= btrfs_releasepage
,
6041 .set_page_dirty
= btrfs_set_page_dirty
,
6042 .error_remove_page
= generic_error_remove_page
,
6045 static const struct address_space_operations btrfs_symlink_aops
= {
6046 .readpage
= btrfs_readpage
,
6047 .writepage
= btrfs_writepage
,
6048 .invalidatepage
= btrfs_invalidatepage
,
6049 .releasepage
= btrfs_releasepage
,
6052 static const struct inode_operations btrfs_file_inode_operations
= {
6053 .truncate
= btrfs_truncate
,
6054 .getattr
= btrfs_getattr
,
6055 .setattr
= btrfs_setattr
,
6056 .setxattr
= btrfs_setxattr
,
6057 .getxattr
= btrfs_getxattr
,
6058 .listxattr
= btrfs_listxattr
,
6059 .removexattr
= btrfs_removexattr
,
6060 .permission
= btrfs_permission
,
6061 .fallocate
= btrfs_fallocate
,
6062 .fiemap
= btrfs_fiemap
,
6064 static const struct inode_operations btrfs_special_inode_operations
= {
6065 .getattr
= btrfs_getattr
,
6066 .setattr
= btrfs_setattr
,
6067 .permission
= btrfs_permission
,
6068 .setxattr
= btrfs_setxattr
,
6069 .getxattr
= btrfs_getxattr
,
6070 .listxattr
= btrfs_listxattr
,
6071 .removexattr
= btrfs_removexattr
,
6073 static const struct inode_operations btrfs_symlink_inode_operations
= {
6074 .readlink
= generic_readlink
,
6075 .follow_link
= page_follow_link_light
,
6076 .put_link
= page_put_link
,
6077 .permission
= btrfs_permission
,
6078 .setxattr
= btrfs_setxattr
,
6079 .getxattr
= btrfs_getxattr
,
6080 .listxattr
= btrfs_listxattr
,
6081 .removexattr
= btrfs_removexattr
,
6084 const struct dentry_operations btrfs_dentry_operations
= {
6085 .d_delete
= btrfs_dentry_delete
,