2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args
{
60 struct btrfs_root
*root
;
63 static const struct inode_operations btrfs_dir_inode_operations
;
64 static const struct inode_operations btrfs_symlink_inode_operations
;
65 static const struct inode_operations btrfs_dir_ro_inode_operations
;
66 static const struct inode_operations btrfs_special_inode_operations
;
67 static const struct inode_operations btrfs_file_inode_operations
;
68 static const struct address_space_operations btrfs_aops
;
69 static const struct address_space_operations btrfs_symlink_aops
;
70 static const struct file_operations btrfs_dir_file_operations
;
71 static struct extent_io_ops btrfs_extent_io_ops
;
73 static struct kmem_cache
*btrfs_inode_cachep
;
74 struct kmem_cache
*btrfs_trans_handle_cachep
;
75 struct kmem_cache
*btrfs_transaction_cachep
;
76 struct kmem_cache
*btrfs_path_cachep
;
77 struct kmem_cache
*btrfs_free_space_cachep
;
80 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
81 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
82 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
83 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
84 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
85 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
86 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
87 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
90 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
);
91 static int btrfs_truncate(struct inode
*inode
);
92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
93 static noinline
int cow_file_range(struct inode
*inode
,
94 struct page
*locked_page
,
95 u64 start
, u64 end
, int *page_started
,
96 unsigned long *nr_written
, int unlock
);
97 static noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
98 struct btrfs_root
*root
, struct inode
*inode
);
100 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
101 struct inode
*inode
, struct inode
*dir
,
102 const struct qstr
*qstr
)
106 err
= btrfs_init_acl(trans
, inode
, dir
);
108 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
117 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
118 struct btrfs_root
*root
, struct inode
*inode
,
119 u64 start
, size_t size
, size_t compressed_size
,
121 struct page
**compressed_pages
)
123 struct btrfs_key key
;
124 struct btrfs_path
*path
;
125 struct extent_buffer
*leaf
;
126 struct page
*page
= NULL
;
129 struct btrfs_file_extent_item
*ei
;
132 size_t cur_size
= size
;
134 unsigned long offset
;
136 if (compressed_size
&& compressed_pages
)
137 cur_size
= compressed_size
;
139 path
= btrfs_alloc_path();
143 path
->leave_spinning
= 1;
145 key
.objectid
= btrfs_ino(inode
);
147 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
148 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
150 inode_add_bytes(inode
, size
);
151 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
157 leaf
= path
->nodes
[0];
158 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
159 struct btrfs_file_extent_item
);
160 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
161 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
162 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
163 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
164 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
165 ptr
= btrfs_file_extent_inline_start(ei
);
167 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
170 while (compressed_size
> 0) {
171 cpage
= compressed_pages
[i
];
172 cur_size
= min_t(unsigned long, compressed_size
,
175 kaddr
= kmap_atomic(cpage
);
176 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
177 kunmap_atomic(kaddr
);
181 compressed_size
-= cur_size
;
183 btrfs_set_file_extent_compression(leaf
, ei
,
186 page
= find_get_page(inode
->i_mapping
,
187 start
>> PAGE_CACHE_SHIFT
);
188 btrfs_set_file_extent_compression(leaf
, ei
, 0);
189 kaddr
= kmap_atomic(page
);
190 offset
= start
& (PAGE_CACHE_SIZE
- 1);
191 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
192 kunmap_atomic(kaddr
);
193 page_cache_release(page
);
195 btrfs_mark_buffer_dirty(leaf
);
196 btrfs_free_path(path
);
199 * we're an inline extent, so nobody can
200 * extend the file past i_size without locking
201 * a page we already have locked.
203 * We must do any isize and inode updates
204 * before we unlock the pages. Otherwise we
205 * could end up racing with unlink.
207 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
208 ret
= btrfs_update_inode(trans
, root
, inode
);
212 btrfs_free_path(path
);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
223 struct btrfs_root
*root
,
224 struct inode
*inode
, u64 start
, u64 end
,
225 size_t compressed_size
, int compress_type
,
226 struct page
**compressed_pages
)
228 u64 isize
= i_size_read(inode
);
229 u64 actual_end
= min(end
+ 1, isize
);
230 u64 inline_len
= actual_end
- start
;
231 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
232 ~((u64
)root
->sectorsize
- 1);
234 u64 data_len
= inline_len
;
238 data_len
= compressed_size
;
241 actual_end
>= PAGE_CACHE_SIZE
||
242 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
244 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
246 data_len
> root
->fs_info
->max_inline
) {
250 ret
= btrfs_drop_extents(trans
, inode
, start
, aligned_end
,
255 if (isize
> actual_end
)
256 inline_len
= min_t(u64
, isize
, actual_end
);
257 ret
= insert_inline_extent(trans
, root
, inode
, start
,
258 inline_len
, compressed_size
,
259 compress_type
, compressed_pages
);
260 if (ret
&& ret
!= -ENOSPC
) {
261 btrfs_abort_transaction(trans
, root
, ret
);
263 } else if (ret
== -ENOSPC
) {
267 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
268 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
272 struct async_extent
{
277 unsigned long nr_pages
;
279 struct list_head list
;
284 struct btrfs_root
*root
;
285 struct page
*locked_page
;
288 struct list_head extents
;
289 struct btrfs_work work
;
292 static noinline
int add_async_extent(struct async_cow
*cow
,
293 u64 start
, u64 ram_size
,
296 unsigned long nr_pages
,
299 struct async_extent
*async_extent
;
301 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
302 BUG_ON(!async_extent
); /* -ENOMEM */
303 async_extent
->start
= start
;
304 async_extent
->ram_size
= ram_size
;
305 async_extent
->compressed_size
= compressed_size
;
306 async_extent
->pages
= pages
;
307 async_extent
->nr_pages
= nr_pages
;
308 async_extent
->compress_type
= compress_type
;
309 list_add_tail(&async_extent
->list
, &cow
->extents
);
314 * we create compressed extents in two phases. The first
315 * phase compresses a range of pages that have already been
316 * locked (both pages and state bits are locked).
318 * This is done inside an ordered work queue, and the compression
319 * is spread across many cpus. The actual IO submission is step
320 * two, and the ordered work queue takes care of making sure that
321 * happens in the same order things were put onto the queue by
322 * writepages and friends.
324 * If this code finds it can't get good compression, it puts an
325 * entry onto the work queue to write the uncompressed bytes. This
326 * makes sure that both compressed inodes and uncompressed inodes
327 * are written in the same order that pdflush sent them down.
329 static noinline
int compress_file_range(struct inode
*inode
,
330 struct page
*locked_page
,
332 struct async_cow
*async_cow
,
335 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
336 struct btrfs_trans_handle
*trans
;
338 u64 blocksize
= root
->sectorsize
;
340 u64 isize
= i_size_read(inode
);
342 struct page
**pages
= NULL
;
343 unsigned long nr_pages
;
344 unsigned long nr_pages_ret
= 0;
345 unsigned long total_compressed
= 0;
346 unsigned long total_in
= 0;
347 unsigned long max_compressed
= 128 * 1024;
348 unsigned long max_uncompressed
= 128 * 1024;
351 int compress_type
= root
->fs_info
->compress_type
;
353 /* if this is a small write inside eof, kick off a defrag */
354 if ((end
- start
+ 1) < 16 * 1024 &&
355 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
356 btrfs_add_inode_defrag(NULL
, inode
);
358 actual_end
= min_t(u64
, isize
, end
+ 1);
361 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
362 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
365 * we don't want to send crud past the end of i_size through
366 * compression, that's just a waste of CPU time. So, if the
367 * end of the file is before the start of our current
368 * requested range of bytes, we bail out to the uncompressed
369 * cleanup code that can deal with all of this.
371 * It isn't really the fastest way to fix things, but this is a
372 * very uncommon corner.
374 if (actual_end
<= start
)
375 goto cleanup_and_bail_uncompressed
;
377 total_compressed
= actual_end
- start
;
379 /* we want to make sure that amount of ram required to uncompress
380 * an extent is reasonable, so we limit the total size in ram
381 * of a compressed extent to 128k. This is a crucial number
382 * because it also controls how easily we can spread reads across
383 * cpus for decompression.
385 * We also want to make sure the amount of IO required to do
386 * a random read is reasonably small, so we limit the size of
387 * a compressed extent to 128k.
389 total_compressed
= min(total_compressed
, max_uncompressed
);
390 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
391 num_bytes
= max(blocksize
, num_bytes
);
396 * we do compression for mount -o compress and when the
397 * inode has not been flagged as nocompress. This flag can
398 * change at any time if we discover bad compression ratios.
400 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
401 (btrfs_test_opt(root
, COMPRESS
) ||
402 (BTRFS_I(inode
)->force_compress
) ||
403 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
405 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
407 /* just bail out to the uncompressed code */
411 if (BTRFS_I(inode
)->force_compress
)
412 compress_type
= BTRFS_I(inode
)->force_compress
;
414 ret
= btrfs_compress_pages(compress_type
,
415 inode
->i_mapping
, start
,
416 total_compressed
, pages
,
417 nr_pages
, &nr_pages_ret
,
423 unsigned long offset
= total_compressed
&
424 (PAGE_CACHE_SIZE
- 1);
425 struct page
*page
= pages
[nr_pages_ret
- 1];
428 /* zero the tail end of the last page, we might be
429 * sending it down to disk
432 kaddr
= kmap_atomic(page
);
433 memset(kaddr
+ offset
, 0,
434 PAGE_CACHE_SIZE
- offset
);
435 kunmap_atomic(kaddr
);
442 trans
= btrfs_join_transaction(root
);
444 ret
= PTR_ERR(trans
);
446 goto cleanup_and_out
;
448 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
450 /* lets try to make an inline extent */
451 if (ret
|| total_in
< (actual_end
- start
)) {
452 /* we didn't compress the entire range, try
453 * to make an uncompressed inline extent.
455 ret
= cow_file_range_inline(trans
, root
, inode
,
456 start
, end
, 0, 0, NULL
);
458 /* try making a compressed inline extent */
459 ret
= cow_file_range_inline(trans
, root
, inode
,
462 compress_type
, pages
);
466 * inline extent creation worked or returned error,
467 * we don't need to create any more async work items.
468 * Unlock and free up our temp pages.
470 extent_clear_unlock_delalloc(inode
,
471 &BTRFS_I(inode
)->io_tree
,
473 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
474 EXTENT_CLEAR_DELALLOC
|
475 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
477 btrfs_end_transaction(trans
, root
);
480 btrfs_end_transaction(trans
, root
);
485 * we aren't doing an inline extent round the compressed size
486 * up to a block size boundary so the allocator does sane
489 total_compressed
= (total_compressed
+ blocksize
- 1) &
493 * one last check to make sure the compression is really a
494 * win, compare the page count read with the blocks on disk
496 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
497 ~(PAGE_CACHE_SIZE
- 1);
498 if (total_compressed
>= total_in
) {
501 num_bytes
= total_in
;
504 if (!will_compress
&& pages
) {
506 * the compression code ran but failed to make things smaller,
507 * free any pages it allocated and our page pointer array
509 for (i
= 0; i
< nr_pages_ret
; i
++) {
510 WARN_ON(pages
[i
]->mapping
);
511 page_cache_release(pages
[i
]);
515 total_compressed
= 0;
518 /* flag the file so we don't compress in the future */
519 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
520 !(BTRFS_I(inode
)->force_compress
)) {
521 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
527 /* the async work queues will take care of doing actual
528 * allocation on disk for these compressed pages,
529 * and will submit them to the elevator.
531 add_async_extent(async_cow
, start
, num_bytes
,
532 total_compressed
, pages
, nr_pages_ret
,
535 if (start
+ num_bytes
< end
) {
542 cleanup_and_bail_uncompressed
:
544 * No compression, but we still need to write the pages in
545 * the file we've been given so far. redirty the locked
546 * page if it corresponds to our extent and set things up
547 * for the async work queue to run cow_file_range to do
548 * the normal delalloc dance
550 if (page_offset(locked_page
) >= start
&&
551 page_offset(locked_page
) <= end
) {
552 __set_page_dirty_nobuffers(locked_page
);
553 /* unlocked later on in the async handlers */
555 add_async_extent(async_cow
, start
, end
- start
+ 1,
556 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
564 for (i
= 0; i
< nr_pages_ret
; i
++) {
565 WARN_ON(pages
[i
]->mapping
);
566 page_cache_release(pages
[i
]);
573 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
575 EXTENT_CLEAR_UNLOCK_PAGE
|
577 EXTENT_CLEAR_DELALLOC
|
578 EXTENT_SET_WRITEBACK
|
579 EXTENT_END_WRITEBACK
);
580 if (!trans
|| IS_ERR(trans
))
581 btrfs_error(root
->fs_info
, ret
, "Failed to join transaction");
583 btrfs_abort_transaction(trans
, root
, ret
);
588 * phase two of compressed writeback. This is the ordered portion
589 * of the code, which only gets called in the order the work was
590 * queued. We walk all the async extents created by compress_file_range
591 * and send them down to the disk.
593 static noinline
int submit_compressed_extents(struct inode
*inode
,
594 struct async_cow
*async_cow
)
596 struct async_extent
*async_extent
;
598 struct btrfs_trans_handle
*trans
;
599 struct btrfs_key ins
;
600 struct extent_map
*em
;
601 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
602 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
603 struct extent_io_tree
*io_tree
;
606 if (list_empty(&async_cow
->extents
))
610 while (!list_empty(&async_cow
->extents
)) {
611 async_extent
= list_entry(async_cow
->extents
.next
,
612 struct async_extent
, list
);
613 list_del(&async_extent
->list
);
615 io_tree
= &BTRFS_I(inode
)->io_tree
;
618 /* did the compression code fall back to uncompressed IO? */
619 if (!async_extent
->pages
) {
620 int page_started
= 0;
621 unsigned long nr_written
= 0;
623 lock_extent(io_tree
, async_extent
->start
,
624 async_extent
->start
+
625 async_extent
->ram_size
- 1);
627 /* allocate blocks */
628 ret
= cow_file_range(inode
, async_cow
->locked_page
,
630 async_extent
->start
+
631 async_extent
->ram_size
- 1,
632 &page_started
, &nr_written
, 0);
637 * if page_started, cow_file_range inserted an
638 * inline extent and took care of all the unlocking
639 * and IO for us. Otherwise, we need to submit
640 * all those pages down to the drive.
642 if (!page_started
&& !ret
)
643 extent_write_locked_range(io_tree
,
644 inode
, async_extent
->start
,
645 async_extent
->start
+
646 async_extent
->ram_size
- 1,
654 lock_extent(io_tree
, async_extent
->start
,
655 async_extent
->start
+ async_extent
->ram_size
- 1);
657 trans
= btrfs_join_transaction(root
);
659 ret
= PTR_ERR(trans
);
661 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
662 ret
= btrfs_reserve_extent(trans
, root
,
663 async_extent
->compressed_size
,
664 async_extent
->compressed_size
,
665 0, alloc_hint
, &ins
, 1);
667 btrfs_abort_transaction(trans
, root
, ret
);
668 btrfs_end_transaction(trans
, root
);
673 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
674 WARN_ON(async_extent
->pages
[i
]->mapping
);
675 page_cache_release(async_extent
->pages
[i
]);
677 kfree(async_extent
->pages
);
678 async_extent
->nr_pages
= 0;
679 async_extent
->pages
= NULL
;
680 unlock_extent(io_tree
, async_extent
->start
,
681 async_extent
->start
+
682 async_extent
->ram_size
- 1);
685 goto out_free
; /* JDM: Requeue? */
689 * here we're doing allocation and writeback of the
692 btrfs_drop_extent_cache(inode
, async_extent
->start
,
693 async_extent
->start
+
694 async_extent
->ram_size
- 1, 0);
696 em
= alloc_extent_map();
697 BUG_ON(!em
); /* -ENOMEM */
698 em
->start
= async_extent
->start
;
699 em
->len
= async_extent
->ram_size
;
700 em
->orig_start
= em
->start
;
702 em
->block_start
= ins
.objectid
;
703 em
->block_len
= ins
.offset
;
704 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
705 em
->compress_type
= async_extent
->compress_type
;
706 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
707 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
710 write_lock(&em_tree
->lock
);
711 ret
= add_extent_mapping(em_tree
, em
);
712 write_unlock(&em_tree
->lock
);
713 if (ret
!= -EEXIST
) {
717 btrfs_drop_extent_cache(inode
, async_extent
->start
,
718 async_extent
->start
+
719 async_extent
->ram_size
- 1, 0);
722 ret
= btrfs_add_ordered_extent_compress(inode
,
725 async_extent
->ram_size
,
727 BTRFS_ORDERED_COMPRESSED
,
728 async_extent
->compress_type
);
729 BUG_ON(ret
); /* -ENOMEM */
732 * clear dirty, set writeback and unlock the pages.
734 extent_clear_unlock_delalloc(inode
,
735 &BTRFS_I(inode
)->io_tree
,
737 async_extent
->start
+
738 async_extent
->ram_size
- 1,
739 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
740 EXTENT_CLEAR_UNLOCK
|
741 EXTENT_CLEAR_DELALLOC
|
742 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
744 ret
= btrfs_submit_compressed_write(inode
,
746 async_extent
->ram_size
,
748 ins
.offset
, async_extent
->pages
,
749 async_extent
->nr_pages
);
751 BUG_ON(ret
); /* -ENOMEM */
752 alloc_hint
= ins
.objectid
+ ins
.offset
;
764 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
767 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
768 struct extent_map
*em
;
771 read_lock(&em_tree
->lock
);
772 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
775 * if block start isn't an actual block number then find the
776 * first block in this inode and use that as a hint. If that
777 * block is also bogus then just don't worry about it.
779 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
781 em
= search_extent_mapping(em_tree
, 0, 0);
782 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
783 alloc_hint
= em
->block_start
;
787 alloc_hint
= em
->block_start
;
791 read_unlock(&em_tree
->lock
);
797 * when extent_io.c finds a delayed allocation range in the file,
798 * the call backs end up in this code. The basic idea is to
799 * allocate extents on disk for the range, and create ordered data structs
800 * in ram to track those extents.
802 * locked_page is the page that writepage had locked already. We use
803 * it to make sure we don't do extra locks or unlocks.
805 * *page_started is set to one if we unlock locked_page and do everything
806 * required to start IO on it. It may be clean and already done with
809 static noinline
int cow_file_range(struct inode
*inode
,
810 struct page
*locked_page
,
811 u64 start
, u64 end
, int *page_started
,
812 unsigned long *nr_written
,
815 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
816 struct btrfs_trans_handle
*trans
;
819 unsigned long ram_size
;
822 u64 blocksize
= root
->sectorsize
;
823 struct btrfs_key ins
;
824 struct extent_map
*em
;
825 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
828 BUG_ON(btrfs_is_free_space_inode(root
, inode
));
829 trans
= btrfs_join_transaction(root
);
831 extent_clear_unlock_delalloc(inode
,
832 &BTRFS_I(inode
)->io_tree
,
833 start
, end
, locked_page
,
834 EXTENT_CLEAR_UNLOCK_PAGE
|
835 EXTENT_CLEAR_UNLOCK
|
836 EXTENT_CLEAR_DELALLOC
|
838 EXTENT_SET_WRITEBACK
|
839 EXTENT_END_WRITEBACK
);
840 return PTR_ERR(trans
);
842 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
844 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
845 num_bytes
= max(blocksize
, num_bytes
);
846 disk_num_bytes
= num_bytes
;
849 /* if this is a small write inside eof, kick off defrag */
850 if (num_bytes
< 64 * 1024 &&
851 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
852 btrfs_add_inode_defrag(trans
, inode
);
855 /* lets try to make an inline extent */
856 ret
= cow_file_range_inline(trans
, root
, inode
,
857 start
, end
, 0, 0, NULL
);
859 extent_clear_unlock_delalloc(inode
,
860 &BTRFS_I(inode
)->io_tree
,
862 EXTENT_CLEAR_UNLOCK_PAGE
|
863 EXTENT_CLEAR_UNLOCK
|
864 EXTENT_CLEAR_DELALLOC
|
866 EXTENT_SET_WRITEBACK
|
867 EXTENT_END_WRITEBACK
);
869 *nr_written
= *nr_written
+
870 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
873 } else if (ret
< 0) {
874 btrfs_abort_transaction(trans
, root
, ret
);
879 BUG_ON(disk_num_bytes
>
880 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
882 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
883 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
885 while (disk_num_bytes
> 0) {
888 cur_alloc_size
= disk_num_bytes
;
889 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
890 root
->sectorsize
, 0, alloc_hint
,
893 btrfs_abort_transaction(trans
, root
, ret
);
897 em
= alloc_extent_map();
898 BUG_ON(!em
); /* -ENOMEM */
900 em
->orig_start
= em
->start
;
901 ram_size
= ins
.offset
;
902 em
->len
= ins
.offset
;
904 em
->block_start
= ins
.objectid
;
905 em
->block_len
= ins
.offset
;
906 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
907 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
910 write_lock(&em_tree
->lock
);
911 ret
= add_extent_mapping(em_tree
, em
);
912 write_unlock(&em_tree
->lock
);
913 if (ret
!= -EEXIST
) {
917 btrfs_drop_extent_cache(inode
, start
,
918 start
+ ram_size
- 1, 0);
921 cur_alloc_size
= ins
.offset
;
922 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
923 ram_size
, cur_alloc_size
, 0);
924 BUG_ON(ret
); /* -ENOMEM */
926 if (root
->root_key
.objectid
==
927 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
928 ret
= btrfs_reloc_clone_csums(inode
, start
,
931 btrfs_abort_transaction(trans
, root
, ret
);
936 if (disk_num_bytes
< cur_alloc_size
)
939 /* we're not doing compressed IO, don't unlock the first
940 * page (which the caller expects to stay locked), don't
941 * clear any dirty bits and don't set any writeback bits
943 * Do set the Private2 bit so we know this page was properly
944 * setup for writepage
946 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
947 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
950 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
951 start
, start
+ ram_size
- 1,
953 disk_num_bytes
-= cur_alloc_size
;
954 num_bytes
-= cur_alloc_size
;
955 alloc_hint
= ins
.objectid
+ ins
.offset
;
956 start
+= cur_alloc_size
;
960 btrfs_end_transaction(trans
, root
);
964 extent_clear_unlock_delalloc(inode
,
965 &BTRFS_I(inode
)->io_tree
,
966 start
, end
, locked_page
,
967 EXTENT_CLEAR_UNLOCK_PAGE
|
968 EXTENT_CLEAR_UNLOCK
|
969 EXTENT_CLEAR_DELALLOC
|
971 EXTENT_SET_WRITEBACK
|
972 EXTENT_END_WRITEBACK
);
978 * work queue call back to started compression on a file and pages
980 static noinline
void async_cow_start(struct btrfs_work
*work
)
982 struct async_cow
*async_cow
;
984 async_cow
= container_of(work
, struct async_cow
, work
);
986 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
987 async_cow
->start
, async_cow
->end
, async_cow
,
989 if (num_added
== 0) {
990 btrfs_add_delayed_iput(async_cow
->inode
);
991 async_cow
->inode
= NULL
;
996 * work queue call back to submit previously compressed pages
998 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1000 struct async_cow
*async_cow
;
1001 struct btrfs_root
*root
;
1002 unsigned long nr_pages
;
1004 async_cow
= container_of(work
, struct async_cow
, work
);
1006 root
= async_cow
->root
;
1007 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1010 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1012 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1014 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1015 wake_up(&root
->fs_info
->async_submit_wait
);
1017 if (async_cow
->inode
)
1018 submit_compressed_extents(async_cow
->inode
, async_cow
);
1021 static noinline
void async_cow_free(struct btrfs_work
*work
)
1023 struct async_cow
*async_cow
;
1024 async_cow
= container_of(work
, struct async_cow
, work
);
1025 if (async_cow
->inode
)
1026 btrfs_add_delayed_iput(async_cow
->inode
);
1030 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1031 u64 start
, u64 end
, int *page_started
,
1032 unsigned long *nr_written
)
1034 struct async_cow
*async_cow
;
1035 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1036 unsigned long nr_pages
;
1038 int limit
= 10 * 1024 * 1042;
1040 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1041 1, 0, NULL
, GFP_NOFS
);
1042 while (start
< end
) {
1043 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1044 BUG_ON(!async_cow
); /* -ENOMEM */
1045 async_cow
->inode
= igrab(inode
);
1046 async_cow
->root
= root
;
1047 async_cow
->locked_page
= locked_page
;
1048 async_cow
->start
= start
;
1050 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
1053 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1055 async_cow
->end
= cur_end
;
1056 INIT_LIST_HEAD(&async_cow
->extents
);
1058 async_cow
->work
.func
= async_cow_start
;
1059 async_cow
->work
.ordered_func
= async_cow_submit
;
1060 async_cow
->work
.ordered_free
= async_cow_free
;
1061 async_cow
->work
.flags
= 0;
1063 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1065 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1067 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
1070 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1071 wait_event(root
->fs_info
->async_submit_wait
,
1072 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1076 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1077 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1078 wait_event(root
->fs_info
->async_submit_wait
,
1079 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1083 *nr_written
+= nr_pages
;
1084 start
= cur_end
+ 1;
1090 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1091 u64 bytenr
, u64 num_bytes
)
1094 struct btrfs_ordered_sum
*sums
;
1097 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1098 bytenr
+ num_bytes
- 1, &list
, 0);
1099 if (ret
== 0 && list_empty(&list
))
1102 while (!list_empty(&list
)) {
1103 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1104 list_del(&sums
->list
);
1111 * when nowcow writeback call back. This checks for snapshots or COW copies
1112 * of the extents that exist in the file, and COWs the file as required.
1114 * If no cow copies or snapshots exist, we write directly to the existing
1117 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1118 struct page
*locked_page
,
1119 u64 start
, u64 end
, int *page_started
, int force
,
1120 unsigned long *nr_written
)
1122 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1123 struct btrfs_trans_handle
*trans
;
1124 struct extent_buffer
*leaf
;
1125 struct btrfs_path
*path
;
1126 struct btrfs_file_extent_item
*fi
;
1127 struct btrfs_key found_key
;
1140 u64 ino
= btrfs_ino(inode
);
1142 path
= btrfs_alloc_path();
1144 extent_clear_unlock_delalloc(inode
,
1145 &BTRFS_I(inode
)->io_tree
,
1146 start
, end
, locked_page
,
1147 EXTENT_CLEAR_UNLOCK_PAGE
|
1148 EXTENT_CLEAR_UNLOCK
|
1149 EXTENT_CLEAR_DELALLOC
|
1150 EXTENT_CLEAR_DIRTY
|
1151 EXTENT_SET_WRITEBACK
|
1152 EXTENT_END_WRITEBACK
);
1156 nolock
= btrfs_is_free_space_inode(root
, inode
);
1159 trans
= btrfs_join_transaction_nolock(root
);
1161 trans
= btrfs_join_transaction(root
);
1163 if (IS_ERR(trans
)) {
1164 extent_clear_unlock_delalloc(inode
,
1165 &BTRFS_I(inode
)->io_tree
,
1166 start
, end
, locked_page
,
1167 EXTENT_CLEAR_UNLOCK_PAGE
|
1168 EXTENT_CLEAR_UNLOCK
|
1169 EXTENT_CLEAR_DELALLOC
|
1170 EXTENT_CLEAR_DIRTY
|
1171 EXTENT_SET_WRITEBACK
|
1172 EXTENT_END_WRITEBACK
);
1173 btrfs_free_path(path
);
1174 return PTR_ERR(trans
);
1177 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1179 cow_start
= (u64
)-1;
1182 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1185 btrfs_abort_transaction(trans
, root
, ret
);
1188 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1189 leaf
= path
->nodes
[0];
1190 btrfs_item_key_to_cpu(leaf
, &found_key
,
1191 path
->slots
[0] - 1);
1192 if (found_key
.objectid
== ino
&&
1193 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1198 leaf
= path
->nodes
[0];
1199 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1200 ret
= btrfs_next_leaf(root
, path
);
1202 btrfs_abort_transaction(trans
, root
, ret
);
1207 leaf
= path
->nodes
[0];
1213 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1215 if (found_key
.objectid
> ino
||
1216 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1217 found_key
.offset
> end
)
1220 if (found_key
.offset
> cur_offset
) {
1221 extent_end
= found_key
.offset
;
1226 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1227 struct btrfs_file_extent_item
);
1228 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1230 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1231 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1232 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1233 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1234 extent_end
= found_key
.offset
+
1235 btrfs_file_extent_num_bytes(leaf
, fi
);
1236 if (extent_end
<= start
) {
1240 if (disk_bytenr
== 0)
1242 if (btrfs_file_extent_compression(leaf
, fi
) ||
1243 btrfs_file_extent_encryption(leaf
, fi
) ||
1244 btrfs_file_extent_other_encoding(leaf
, fi
))
1246 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1248 if (btrfs_extent_readonly(root
, disk_bytenr
))
1250 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1252 extent_offset
, disk_bytenr
))
1254 disk_bytenr
+= extent_offset
;
1255 disk_bytenr
+= cur_offset
- found_key
.offset
;
1256 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1258 * force cow if csum exists in the range.
1259 * this ensure that csum for a given extent are
1260 * either valid or do not exist.
1262 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1265 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1266 extent_end
= found_key
.offset
+
1267 btrfs_file_extent_inline_len(leaf
, fi
);
1268 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1273 if (extent_end
<= start
) {
1278 if (cow_start
== (u64
)-1)
1279 cow_start
= cur_offset
;
1280 cur_offset
= extent_end
;
1281 if (cur_offset
> end
)
1287 btrfs_release_path(path
);
1288 if (cow_start
!= (u64
)-1) {
1289 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1290 found_key
.offset
- 1, page_started
,
1293 btrfs_abort_transaction(trans
, root
, ret
);
1296 cow_start
= (u64
)-1;
1299 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1300 struct extent_map
*em
;
1301 struct extent_map_tree
*em_tree
;
1302 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1303 em
= alloc_extent_map();
1304 BUG_ON(!em
); /* -ENOMEM */
1305 em
->start
= cur_offset
;
1306 em
->orig_start
= em
->start
;
1307 em
->len
= num_bytes
;
1308 em
->block_len
= num_bytes
;
1309 em
->block_start
= disk_bytenr
;
1310 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1311 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1313 write_lock(&em_tree
->lock
);
1314 ret
= add_extent_mapping(em_tree
, em
);
1315 write_unlock(&em_tree
->lock
);
1316 if (ret
!= -EEXIST
) {
1317 free_extent_map(em
);
1320 btrfs_drop_extent_cache(inode
, em
->start
,
1321 em
->start
+ em
->len
- 1, 0);
1323 type
= BTRFS_ORDERED_PREALLOC
;
1325 type
= BTRFS_ORDERED_NOCOW
;
1328 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1329 num_bytes
, num_bytes
, type
);
1330 BUG_ON(ret
); /* -ENOMEM */
1332 if (root
->root_key
.objectid
==
1333 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1334 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1337 btrfs_abort_transaction(trans
, root
, ret
);
1342 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1343 cur_offset
, cur_offset
+ num_bytes
- 1,
1344 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1345 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1346 EXTENT_SET_PRIVATE2
);
1347 cur_offset
= extent_end
;
1348 if (cur_offset
> end
)
1351 btrfs_release_path(path
);
1353 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1354 cow_start
= cur_offset
;
1358 if (cow_start
!= (u64
)-1) {
1359 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1360 page_started
, nr_written
, 1);
1362 btrfs_abort_transaction(trans
, root
, ret
);
1369 err
= btrfs_end_transaction_nolock(trans
, root
);
1371 err
= btrfs_end_transaction(trans
, root
);
1376 if (ret
&& cur_offset
< end
)
1377 extent_clear_unlock_delalloc(inode
,
1378 &BTRFS_I(inode
)->io_tree
,
1379 cur_offset
, end
, locked_page
,
1380 EXTENT_CLEAR_UNLOCK_PAGE
|
1381 EXTENT_CLEAR_UNLOCK
|
1382 EXTENT_CLEAR_DELALLOC
|
1383 EXTENT_CLEAR_DIRTY
|
1384 EXTENT_SET_WRITEBACK
|
1385 EXTENT_END_WRITEBACK
);
1387 btrfs_free_path(path
);
1392 * extent_io.c call back to do delayed allocation processing
1394 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1395 u64 start
, u64 end
, int *page_started
,
1396 unsigned long *nr_written
)
1399 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1401 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) {
1402 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1403 page_started
, 1, nr_written
);
1404 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
) {
1405 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1406 page_started
, 0, nr_written
);
1407 } else if (!btrfs_test_opt(root
, COMPRESS
) &&
1408 !(BTRFS_I(inode
)->force_compress
) &&
1409 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
)) {
1410 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1411 page_started
, nr_written
, 1);
1413 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1414 &BTRFS_I(inode
)->runtime_flags
);
1415 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1416 page_started
, nr_written
);
1421 static void btrfs_split_extent_hook(struct inode
*inode
,
1422 struct extent_state
*orig
, u64 split
)
1424 /* not delalloc, ignore it */
1425 if (!(orig
->state
& EXTENT_DELALLOC
))
1428 spin_lock(&BTRFS_I(inode
)->lock
);
1429 BTRFS_I(inode
)->outstanding_extents
++;
1430 spin_unlock(&BTRFS_I(inode
)->lock
);
1434 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1435 * extents so we can keep track of new extents that are just merged onto old
1436 * extents, such as when we are doing sequential writes, so we can properly
1437 * account for the metadata space we'll need.
1439 static void btrfs_merge_extent_hook(struct inode
*inode
,
1440 struct extent_state
*new,
1441 struct extent_state
*other
)
1443 /* not delalloc, ignore it */
1444 if (!(other
->state
& EXTENT_DELALLOC
))
1447 spin_lock(&BTRFS_I(inode
)->lock
);
1448 BTRFS_I(inode
)->outstanding_extents
--;
1449 spin_unlock(&BTRFS_I(inode
)->lock
);
1453 * extent_io.c set_bit_hook, used to track delayed allocation
1454 * bytes in this file, and to maintain the list of inodes that
1455 * have pending delalloc work to be done.
1457 static void btrfs_set_bit_hook(struct inode
*inode
,
1458 struct extent_state
*state
, int *bits
)
1462 * set_bit and clear bit hooks normally require _irqsave/restore
1463 * but in this case, we are only testing for the DELALLOC
1464 * bit, which is only set or cleared with irqs on
1466 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1467 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1468 u64 len
= state
->end
+ 1 - state
->start
;
1469 bool do_list
= !btrfs_is_free_space_inode(root
, inode
);
1471 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1472 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1474 spin_lock(&BTRFS_I(inode
)->lock
);
1475 BTRFS_I(inode
)->outstanding_extents
++;
1476 spin_unlock(&BTRFS_I(inode
)->lock
);
1479 spin_lock(&root
->fs_info
->delalloc_lock
);
1480 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1481 root
->fs_info
->delalloc_bytes
+= len
;
1482 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1483 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1484 &root
->fs_info
->delalloc_inodes
);
1486 spin_unlock(&root
->fs_info
->delalloc_lock
);
1491 * extent_io.c clear_bit_hook, see set_bit_hook for why
1493 static void btrfs_clear_bit_hook(struct inode
*inode
,
1494 struct extent_state
*state
, int *bits
)
1497 * set_bit and clear bit hooks normally require _irqsave/restore
1498 * but in this case, we are only testing for the DELALLOC
1499 * bit, which is only set or cleared with irqs on
1501 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1502 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1503 u64 len
= state
->end
+ 1 - state
->start
;
1504 bool do_list
= !btrfs_is_free_space_inode(root
, inode
);
1506 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1507 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1508 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1509 spin_lock(&BTRFS_I(inode
)->lock
);
1510 BTRFS_I(inode
)->outstanding_extents
--;
1511 spin_unlock(&BTRFS_I(inode
)->lock
);
1514 if (*bits
& EXTENT_DO_ACCOUNTING
)
1515 btrfs_delalloc_release_metadata(inode
, len
);
1517 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1519 btrfs_free_reserved_data_space(inode
, len
);
1521 spin_lock(&root
->fs_info
->delalloc_lock
);
1522 root
->fs_info
->delalloc_bytes
-= len
;
1523 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1525 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1526 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1527 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1529 spin_unlock(&root
->fs_info
->delalloc_lock
);
1534 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1535 * we don't create bios that span stripes or chunks
1537 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1538 size_t size
, struct bio
*bio
,
1539 unsigned long bio_flags
)
1541 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1542 struct btrfs_mapping_tree
*map_tree
;
1543 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1548 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1551 length
= bio
->bi_size
;
1552 map_tree
= &root
->fs_info
->mapping_tree
;
1553 map_length
= length
;
1554 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1555 &map_length
, NULL
, 0);
1556 /* Will always return 0 or 1 with map_multi == NULL */
1558 if (map_length
< length
+ size
)
1564 * in order to insert checksums into the metadata in large chunks,
1565 * we wait until bio submission time. All the pages in the bio are
1566 * checksummed and sums are attached onto the ordered extent record.
1568 * At IO completion time the cums attached on the ordered extent record
1569 * are inserted into the btree
1571 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1572 struct bio
*bio
, int mirror_num
,
1573 unsigned long bio_flags
,
1576 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1579 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1580 BUG_ON(ret
); /* -ENOMEM */
1585 * in order to insert checksums into the metadata in large chunks,
1586 * we wait until bio submission time. All the pages in the bio are
1587 * checksummed and sums are attached onto the ordered extent record.
1589 * At IO completion time the cums attached on the ordered extent record
1590 * are inserted into the btree
1592 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1593 int mirror_num
, unsigned long bio_flags
,
1596 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1597 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1601 * extent_io.c submission hook. This does the right thing for csum calculation
1602 * on write, or reading the csums from the tree before a read
1604 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1605 int mirror_num
, unsigned long bio_flags
,
1608 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1613 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1615 if (btrfs_is_free_space_inode(root
, inode
))
1618 if (!(rw
& REQ_WRITE
)) {
1619 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1623 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1624 return btrfs_submit_compressed_read(inode
, bio
,
1625 mirror_num
, bio_flags
);
1626 } else if (!skip_sum
) {
1627 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1632 } else if (!skip_sum
) {
1633 /* csum items have already been cloned */
1634 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1636 /* we're doing a write, do the async checksumming */
1637 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1638 inode
, rw
, bio
, mirror_num
,
1639 bio_flags
, bio_offset
,
1640 __btrfs_submit_bio_start
,
1641 __btrfs_submit_bio_done
);
1645 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1649 * given a list of ordered sums record them in the inode. This happens
1650 * at IO completion time based on sums calculated at bio submission time.
1652 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1653 struct inode
*inode
, u64 file_offset
,
1654 struct list_head
*list
)
1656 struct btrfs_ordered_sum
*sum
;
1658 list_for_each_entry(sum
, list
, list
) {
1659 btrfs_csum_file_blocks(trans
,
1660 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1665 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1666 struct extent_state
**cached_state
)
1668 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1670 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1671 cached_state
, GFP_NOFS
);
1674 /* see btrfs_writepage_start_hook for details on why this is required */
1675 struct btrfs_writepage_fixup
{
1677 struct btrfs_work work
;
1680 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1682 struct btrfs_writepage_fixup
*fixup
;
1683 struct btrfs_ordered_extent
*ordered
;
1684 struct extent_state
*cached_state
= NULL
;
1686 struct inode
*inode
;
1691 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1695 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1696 ClearPageChecked(page
);
1700 inode
= page
->mapping
->host
;
1701 page_start
= page_offset(page
);
1702 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1704 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1707 /* already ordered? We're done */
1708 if (PagePrivate2(page
))
1711 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1713 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1714 page_end
, &cached_state
, GFP_NOFS
);
1716 btrfs_start_ordered_extent(inode
, ordered
, 1);
1717 btrfs_put_ordered_extent(ordered
);
1721 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1723 mapping_set_error(page
->mapping
, ret
);
1724 end_extent_writepage(page
, ret
, page_start
, page_end
);
1725 ClearPageChecked(page
);
1729 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1730 ClearPageChecked(page
);
1731 set_page_dirty(page
);
1733 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1734 &cached_state
, GFP_NOFS
);
1737 page_cache_release(page
);
1742 * There are a few paths in the higher layers of the kernel that directly
1743 * set the page dirty bit without asking the filesystem if it is a
1744 * good idea. This causes problems because we want to make sure COW
1745 * properly happens and the data=ordered rules are followed.
1747 * In our case any range that doesn't have the ORDERED bit set
1748 * hasn't been properly setup for IO. We kick off an async process
1749 * to fix it up. The async helper will wait for ordered extents, set
1750 * the delalloc bit and make it safe to write the page.
1752 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1754 struct inode
*inode
= page
->mapping
->host
;
1755 struct btrfs_writepage_fixup
*fixup
;
1756 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1758 /* this page is properly in the ordered list */
1759 if (TestClearPagePrivate2(page
))
1762 if (PageChecked(page
))
1765 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1769 SetPageChecked(page
);
1770 page_cache_get(page
);
1771 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1773 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1777 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1778 struct inode
*inode
, u64 file_pos
,
1779 u64 disk_bytenr
, u64 disk_num_bytes
,
1780 u64 num_bytes
, u64 ram_bytes
,
1781 u8 compression
, u8 encryption
,
1782 u16 other_encoding
, int extent_type
)
1784 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1785 struct btrfs_file_extent_item
*fi
;
1786 struct btrfs_path
*path
;
1787 struct extent_buffer
*leaf
;
1788 struct btrfs_key ins
;
1792 path
= btrfs_alloc_path();
1796 path
->leave_spinning
= 1;
1799 * we may be replacing one extent in the tree with another.
1800 * The new extent is pinned in the extent map, and we don't want
1801 * to drop it from the cache until it is completely in the btree.
1803 * So, tell btrfs_drop_extents to leave this extent in the cache.
1804 * the caller is expected to unpin it and allow it to be merged
1807 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1812 ins
.objectid
= btrfs_ino(inode
);
1813 ins
.offset
= file_pos
;
1814 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1815 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1818 leaf
= path
->nodes
[0];
1819 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1820 struct btrfs_file_extent_item
);
1821 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1822 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1823 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1824 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1825 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1826 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1827 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1828 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1829 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1830 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1832 btrfs_unlock_up_safe(path
, 1);
1833 btrfs_set_lock_blocking(leaf
);
1835 btrfs_mark_buffer_dirty(leaf
);
1837 inode_add_bytes(inode
, num_bytes
);
1839 ins
.objectid
= disk_bytenr
;
1840 ins
.offset
= disk_num_bytes
;
1841 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1842 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1843 root
->root_key
.objectid
,
1844 btrfs_ino(inode
), file_pos
, &ins
);
1846 btrfs_free_path(path
);
1852 * helper function for btrfs_finish_ordered_io, this
1853 * just reads in some of the csum leaves to prime them into ram
1854 * before we start the transaction. It limits the amount of btree
1855 * reads required while inside the transaction.
1857 /* as ordered data IO finishes, this gets called so we can finish
1858 * an ordered extent if the range of bytes in the file it covers are
1861 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
1863 struct inode
*inode
= ordered_extent
->inode
;
1864 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1865 struct btrfs_trans_handle
*trans
= NULL
;
1866 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1867 struct extent_state
*cached_state
= NULL
;
1868 int compress_type
= 0;
1872 nolock
= btrfs_is_free_space_inode(root
, inode
);
1874 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
1879 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1880 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
1881 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1884 trans
= btrfs_join_transaction_nolock(root
);
1886 trans
= btrfs_join_transaction(root
);
1888 return PTR_ERR(trans
);
1889 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1890 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
1891 if (ret
) /* -ENOMEM or corruption */
1892 btrfs_abort_transaction(trans
, root
, ret
);
1897 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1898 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1902 trans
= btrfs_join_transaction_nolock(root
);
1904 trans
= btrfs_join_transaction(root
);
1905 if (IS_ERR(trans
)) {
1906 ret
= PTR_ERR(trans
);
1910 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1912 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1913 compress_type
= ordered_extent
->compress_type
;
1914 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1915 BUG_ON(compress_type
);
1916 ret
= btrfs_mark_extent_written(trans
, inode
,
1917 ordered_extent
->file_offset
,
1918 ordered_extent
->file_offset
+
1919 ordered_extent
->len
);
1921 BUG_ON(root
== root
->fs_info
->tree_root
);
1922 ret
= insert_reserved_file_extent(trans
, inode
,
1923 ordered_extent
->file_offset
,
1924 ordered_extent
->start
,
1925 ordered_extent
->disk_len
,
1926 ordered_extent
->len
,
1927 ordered_extent
->len
,
1928 compress_type
, 0, 0,
1929 BTRFS_FILE_EXTENT_REG
);
1930 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1931 ordered_extent
->file_offset
,
1932 ordered_extent
->len
);
1936 btrfs_abort_transaction(trans
, root
, ret
);
1940 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1941 &ordered_extent
->list
);
1943 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1944 if (!ret
|| !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1945 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
1946 if (ret
) { /* -ENOMEM or corruption */
1947 btrfs_abort_transaction(trans
, root
, ret
);
1953 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1954 ordered_extent
->file_offset
+
1955 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1957 if (root
!= root
->fs_info
->tree_root
)
1958 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1961 btrfs_end_transaction_nolock(trans
, root
);
1963 btrfs_end_transaction(trans
, root
);
1967 clear_extent_uptodate(io_tree
, ordered_extent
->file_offset
,
1968 ordered_extent
->file_offset
+
1969 ordered_extent
->len
- 1, NULL
, GFP_NOFS
);
1972 * This needs to be dont to make sure anybody waiting knows we are done
1973 * upating everything for this ordered extent.
1975 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1978 btrfs_put_ordered_extent(ordered_extent
);
1979 /* once for the tree */
1980 btrfs_put_ordered_extent(ordered_extent
);
1985 static void finish_ordered_fn(struct btrfs_work
*work
)
1987 struct btrfs_ordered_extent
*ordered_extent
;
1988 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
1989 btrfs_finish_ordered_io(ordered_extent
);
1992 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1993 struct extent_state
*state
, int uptodate
)
1995 struct inode
*inode
= page
->mapping
->host
;
1996 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1997 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1998 struct btrfs_workers
*workers
;
2000 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
2002 ClearPagePrivate2(page
);
2003 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
2004 end
- start
+ 1, uptodate
))
2007 ordered_extent
->work
.func
= finish_ordered_fn
;
2008 ordered_extent
->work
.flags
= 0;
2010 if (btrfs_is_free_space_inode(root
, inode
))
2011 workers
= &root
->fs_info
->endio_freespace_worker
;
2013 workers
= &root
->fs_info
->endio_write_workers
;
2014 btrfs_queue_worker(workers
, &ordered_extent
->work
);
2020 * when reads are done, we need to check csums to verify the data is correct
2021 * if there's a match, we allow the bio to finish. If not, the code in
2022 * extent_io.c will try to find good copies for us.
2024 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2025 struct extent_state
*state
, int mirror
)
2027 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2028 struct inode
*inode
= page
->mapping
->host
;
2029 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2031 u64
private = ~(u32
)0;
2033 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2036 if (PageChecked(page
)) {
2037 ClearPageChecked(page
);
2041 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
2044 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
2045 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
2046 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
2051 if (state
&& state
->start
== start
) {
2052 private = state
->private;
2055 ret
= get_state_private(io_tree
, start
, &private);
2057 kaddr
= kmap_atomic(page
);
2061 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
2062 btrfs_csum_final(csum
, (char *)&csum
);
2063 if (csum
!= private)
2066 kunmap_atomic(kaddr
);
2071 printk_ratelimited(KERN_INFO
"btrfs csum failed ino %llu off %llu csum %u "
2073 (unsigned long long)btrfs_ino(page
->mapping
->host
),
2074 (unsigned long long)start
, csum
,
2075 (unsigned long long)private);
2076 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2077 flush_dcache_page(page
);
2078 kunmap_atomic(kaddr
);
2084 struct delayed_iput
{
2085 struct list_head list
;
2086 struct inode
*inode
;
2089 /* JDM: If this is fs-wide, why can't we add a pointer to
2090 * btrfs_inode instead and avoid the allocation? */
2091 void btrfs_add_delayed_iput(struct inode
*inode
)
2093 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2094 struct delayed_iput
*delayed
;
2096 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2099 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2100 delayed
->inode
= inode
;
2102 spin_lock(&fs_info
->delayed_iput_lock
);
2103 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2104 spin_unlock(&fs_info
->delayed_iput_lock
);
2107 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2110 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2111 struct delayed_iput
*delayed
;
2114 spin_lock(&fs_info
->delayed_iput_lock
);
2115 empty
= list_empty(&fs_info
->delayed_iputs
);
2116 spin_unlock(&fs_info
->delayed_iput_lock
);
2120 down_read(&root
->fs_info
->cleanup_work_sem
);
2121 spin_lock(&fs_info
->delayed_iput_lock
);
2122 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2123 spin_unlock(&fs_info
->delayed_iput_lock
);
2125 while (!list_empty(&list
)) {
2126 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2127 list_del(&delayed
->list
);
2128 iput(delayed
->inode
);
2131 up_read(&root
->fs_info
->cleanup_work_sem
);
2134 enum btrfs_orphan_cleanup_state
{
2135 ORPHAN_CLEANUP_STARTED
= 1,
2136 ORPHAN_CLEANUP_DONE
= 2,
2140 * This is called in transaction commit time. If there are no orphan
2141 * files in the subvolume, it removes orphan item and frees block_rsv
2144 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2145 struct btrfs_root
*root
)
2147 struct btrfs_block_rsv
*block_rsv
;
2150 if (atomic_read(&root
->orphan_inodes
) ||
2151 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2154 spin_lock(&root
->orphan_lock
);
2155 if (atomic_read(&root
->orphan_inodes
)) {
2156 spin_unlock(&root
->orphan_lock
);
2160 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
2161 spin_unlock(&root
->orphan_lock
);
2165 block_rsv
= root
->orphan_block_rsv
;
2166 root
->orphan_block_rsv
= NULL
;
2167 spin_unlock(&root
->orphan_lock
);
2169 if (root
->orphan_item_inserted
&&
2170 btrfs_root_refs(&root
->root_item
) > 0) {
2171 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2172 root
->root_key
.objectid
);
2174 root
->orphan_item_inserted
= 0;
2178 WARN_ON(block_rsv
->size
> 0);
2179 btrfs_free_block_rsv(root
, block_rsv
);
2184 * This creates an orphan entry for the given inode in case something goes
2185 * wrong in the middle of an unlink/truncate.
2187 * NOTE: caller of this function should reserve 5 units of metadata for
2190 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2192 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2193 struct btrfs_block_rsv
*block_rsv
= NULL
;
2198 if (!root
->orphan_block_rsv
) {
2199 block_rsv
= btrfs_alloc_block_rsv(root
);
2204 spin_lock(&root
->orphan_lock
);
2205 if (!root
->orphan_block_rsv
) {
2206 root
->orphan_block_rsv
= block_rsv
;
2207 } else if (block_rsv
) {
2208 btrfs_free_block_rsv(root
, block_rsv
);
2212 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2213 &BTRFS_I(inode
)->runtime_flags
)) {
2216 * For proper ENOSPC handling, we should do orphan
2217 * cleanup when mounting. But this introduces backward
2218 * compatibility issue.
2220 if (!xchg(&root
->orphan_item_inserted
, 1))
2226 atomic_dec(&root
->orphan_inodes
);
2229 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
2230 &BTRFS_I(inode
)->runtime_flags
))
2232 spin_unlock(&root
->orphan_lock
);
2234 /* grab metadata reservation from transaction handle */
2236 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2237 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
2240 /* insert an orphan item to track this unlinked/truncated file */
2242 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
2243 if (ret
&& ret
!= -EEXIST
) {
2244 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2245 &BTRFS_I(inode
)->runtime_flags
);
2246 btrfs_abort_transaction(trans
, root
, ret
);
2252 /* insert an orphan item to track subvolume contains orphan files */
2254 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2255 root
->root_key
.objectid
);
2256 if (ret
&& ret
!= -EEXIST
) {
2257 btrfs_abort_transaction(trans
, root
, ret
);
2265 * We have done the truncate/delete so we can go ahead and remove the orphan
2266 * item for this particular inode.
2268 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2270 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2271 int delete_item
= 0;
2272 int release_rsv
= 0;
2275 spin_lock(&root
->orphan_lock
);
2276 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2277 &BTRFS_I(inode
)->runtime_flags
))
2280 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
2281 &BTRFS_I(inode
)->runtime_flags
))
2283 spin_unlock(&root
->orphan_lock
);
2285 if (trans
&& delete_item
) {
2286 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
2287 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
2291 btrfs_orphan_release_metadata(inode
);
2292 atomic_dec(&root
->orphan_inodes
);
2299 * this cleans up any orphans that may be left on the list from the last use
2302 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2304 struct btrfs_path
*path
;
2305 struct extent_buffer
*leaf
;
2306 struct btrfs_key key
, found_key
;
2307 struct btrfs_trans_handle
*trans
;
2308 struct inode
*inode
;
2309 u64 last_objectid
= 0;
2310 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2312 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2315 path
= btrfs_alloc_path();
2322 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2323 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2324 key
.offset
= (u64
)-1;
2327 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2332 * if ret == 0 means we found what we were searching for, which
2333 * is weird, but possible, so only screw with path if we didn't
2334 * find the key and see if we have stuff that matches
2338 if (path
->slots
[0] == 0)
2343 /* pull out the item */
2344 leaf
= path
->nodes
[0];
2345 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2347 /* make sure the item matches what we want */
2348 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2350 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2353 /* release the path since we're done with it */
2354 btrfs_release_path(path
);
2357 * this is where we are basically btrfs_lookup, without the
2358 * crossing root thing. we store the inode number in the
2359 * offset of the orphan item.
2362 if (found_key
.offset
== last_objectid
) {
2363 printk(KERN_ERR
"btrfs: Error removing orphan entry, "
2364 "stopping orphan cleanup\n");
2369 last_objectid
= found_key
.offset
;
2371 found_key
.objectid
= found_key
.offset
;
2372 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2373 found_key
.offset
= 0;
2374 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2375 ret
= PTR_RET(inode
);
2376 if (ret
&& ret
!= -ESTALE
)
2379 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
2380 struct btrfs_root
*dead_root
;
2381 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2382 int is_dead_root
= 0;
2385 * this is an orphan in the tree root. Currently these
2386 * could come from 2 sources:
2387 * a) a snapshot deletion in progress
2388 * b) a free space cache inode
2389 * We need to distinguish those two, as the snapshot
2390 * orphan must not get deleted.
2391 * find_dead_roots already ran before us, so if this
2392 * is a snapshot deletion, we should find the root
2393 * in the dead_roots list
2395 spin_lock(&fs_info
->trans_lock
);
2396 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
2398 if (dead_root
->root_key
.objectid
==
2399 found_key
.objectid
) {
2404 spin_unlock(&fs_info
->trans_lock
);
2406 /* prevent this orphan from being found again */
2407 key
.offset
= found_key
.objectid
- 1;
2412 * Inode is already gone but the orphan item is still there,
2413 * kill the orphan item.
2415 if (ret
== -ESTALE
) {
2416 trans
= btrfs_start_transaction(root
, 1);
2417 if (IS_ERR(trans
)) {
2418 ret
= PTR_ERR(trans
);
2421 printk(KERN_ERR
"auto deleting %Lu\n",
2422 found_key
.objectid
);
2423 ret
= btrfs_del_orphan_item(trans
, root
,
2424 found_key
.objectid
);
2425 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
2426 btrfs_end_transaction(trans
, root
);
2431 * add this inode to the orphan list so btrfs_orphan_del does
2432 * the proper thing when we hit it
2434 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2435 &BTRFS_I(inode
)->runtime_flags
);
2437 /* if we have links, this was a truncate, lets do that */
2438 if (inode
->i_nlink
) {
2439 if (!S_ISREG(inode
->i_mode
)) {
2445 ret
= btrfs_truncate(inode
);
2450 /* this will do delete_inode and everything for us */
2455 /* release the path since we're done with it */
2456 btrfs_release_path(path
);
2458 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2460 if (root
->orphan_block_rsv
)
2461 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2464 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2465 trans
= btrfs_join_transaction(root
);
2467 btrfs_end_transaction(trans
, root
);
2471 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2473 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2477 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2478 btrfs_free_path(path
);
2483 * very simple check to peek ahead in the leaf looking for xattrs. If we
2484 * don't find any xattrs, we know there can't be any acls.
2486 * slot is the slot the inode is in, objectid is the objectid of the inode
2488 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2489 int slot
, u64 objectid
)
2491 u32 nritems
= btrfs_header_nritems(leaf
);
2492 struct btrfs_key found_key
;
2496 while (slot
< nritems
) {
2497 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2499 /* we found a different objectid, there must not be acls */
2500 if (found_key
.objectid
!= objectid
)
2503 /* we found an xattr, assume we've got an acl */
2504 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2508 * we found a key greater than an xattr key, there can't
2509 * be any acls later on
2511 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2518 * it goes inode, inode backrefs, xattrs, extents,
2519 * so if there are a ton of hard links to an inode there can
2520 * be a lot of backrefs. Don't waste time searching too hard,
2521 * this is just an optimization
2526 /* we hit the end of the leaf before we found an xattr or
2527 * something larger than an xattr. We have to assume the inode
2534 * read an inode from the btree into the in-memory inode
2536 static void btrfs_read_locked_inode(struct inode
*inode
)
2538 struct btrfs_path
*path
;
2539 struct extent_buffer
*leaf
;
2540 struct btrfs_inode_item
*inode_item
;
2541 struct btrfs_timespec
*tspec
;
2542 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2543 struct btrfs_key location
;
2547 bool filled
= false;
2549 ret
= btrfs_fill_inode(inode
, &rdev
);
2553 path
= btrfs_alloc_path();
2557 path
->leave_spinning
= 1;
2558 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2560 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2564 leaf
= path
->nodes
[0];
2569 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2570 struct btrfs_inode_item
);
2571 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2572 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
2573 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2574 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2575 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2577 tspec
= btrfs_inode_atime(inode_item
);
2578 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2579 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2581 tspec
= btrfs_inode_mtime(inode_item
);
2582 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2583 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2585 tspec
= btrfs_inode_ctime(inode_item
);
2586 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2587 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2589 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2590 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2591 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
2592 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2594 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2596 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2597 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2600 * try to precache a NULL acl entry for files that don't have
2601 * any xattrs or acls
2603 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
2606 cache_no_acl(inode
);
2608 btrfs_free_path(path
);
2610 switch (inode
->i_mode
& S_IFMT
) {
2612 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2613 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2614 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2615 inode
->i_fop
= &btrfs_file_operations
;
2616 inode
->i_op
= &btrfs_file_inode_operations
;
2619 inode
->i_fop
= &btrfs_dir_file_operations
;
2620 if (root
== root
->fs_info
->tree_root
)
2621 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2623 inode
->i_op
= &btrfs_dir_inode_operations
;
2626 inode
->i_op
= &btrfs_symlink_inode_operations
;
2627 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2628 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2631 inode
->i_op
= &btrfs_special_inode_operations
;
2632 init_special_inode(inode
, inode
->i_mode
, rdev
);
2636 btrfs_update_iflags(inode
);
2640 btrfs_free_path(path
);
2641 make_bad_inode(inode
);
2645 * given a leaf and an inode, copy the inode fields into the leaf
2647 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2648 struct extent_buffer
*leaf
,
2649 struct btrfs_inode_item
*item
,
2650 struct inode
*inode
)
2652 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2653 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2654 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2655 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2656 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2658 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2659 inode
->i_atime
.tv_sec
);
2660 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2661 inode
->i_atime
.tv_nsec
);
2663 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2664 inode
->i_mtime
.tv_sec
);
2665 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2666 inode
->i_mtime
.tv_nsec
);
2668 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2669 inode
->i_ctime
.tv_sec
);
2670 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2671 inode
->i_ctime
.tv_nsec
);
2673 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2674 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2675 btrfs_set_inode_sequence(leaf
, item
, inode
->i_version
);
2676 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2677 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2678 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2679 btrfs_set_inode_block_group(leaf
, item
, 0);
2683 * copy everything in the in-memory inode into the btree.
2685 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
2686 struct btrfs_root
*root
, struct inode
*inode
)
2688 struct btrfs_inode_item
*inode_item
;
2689 struct btrfs_path
*path
;
2690 struct extent_buffer
*leaf
;
2693 path
= btrfs_alloc_path();
2697 path
->leave_spinning
= 1;
2698 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
2706 btrfs_unlock_up_safe(path
, 1);
2707 leaf
= path
->nodes
[0];
2708 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2709 struct btrfs_inode_item
);
2711 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2712 btrfs_mark_buffer_dirty(leaf
);
2713 btrfs_set_inode_last_trans(trans
, inode
);
2716 btrfs_free_path(path
);
2721 * copy everything in the in-memory inode into the btree.
2723 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2724 struct btrfs_root
*root
, struct inode
*inode
)
2729 * If the inode is a free space inode, we can deadlock during commit
2730 * if we put it into the delayed code.
2732 * The data relocation inode should also be directly updated
2735 if (!btrfs_is_free_space_inode(root
, inode
)
2736 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
2737 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
2739 btrfs_set_inode_last_trans(trans
, inode
);
2743 return btrfs_update_inode_item(trans
, root
, inode
);
2746 static noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
2747 struct btrfs_root
*root
, struct inode
*inode
)
2751 ret
= btrfs_update_inode(trans
, root
, inode
);
2753 return btrfs_update_inode_item(trans
, root
, inode
);
2758 * unlink helper that gets used here in inode.c and in the tree logging
2759 * recovery code. It remove a link in a directory with a given name, and
2760 * also drops the back refs in the inode to the directory
2762 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2763 struct btrfs_root
*root
,
2764 struct inode
*dir
, struct inode
*inode
,
2765 const char *name
, int name_len
)
2767 struct btrfs_path
*path
;
2769 struct extent_buffer
*leaf
;
2770 struct btrfs_dir_item
*di
;
2771 struct btrfs_key key
;
2773 u64 ino
= btrfs_ino(inode
);
2774 u64 dir_ino
= btrfs_ino(dir
);
2776 path
= btrfs_alloc_path();
2782 path
->leave_spinning
= 1;
2783 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2784 name
, name_len
, -1);
2793 leaf
= path
->nodes
[0];
2794 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2795 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2798 btrfs_release_path(path
);
2800 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
2803 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2804 "inode %llu parent %llu\n", name_len
, name
,
2805 (unsigned long long)ino
, (unsigned long long)dir_ino
);
2806 btrfs_abort_transaction(trans
, root
, ret
);
2810 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2812 btrfs_abort_transaction(trans
, root
, ret
);
2816 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2818 if (ret
!= 0 && ret
!= -ENOENT
) {
2819 btrfs_abort_transaction(trans
, root
, ret
);
2823 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2828 btrfs_free_path(path
);
2832 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2833 inode_inc_iversion(inode
);
2834 inode_inc_iversion(dir
);
2835 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2836 btrfs_update_inode(trans
, root
, dir
);
2841 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2842 struct btrfs_root
*root
,
2843 struct inode
*dir
, struct inode
*inode
,
2844 const char *name
, int name_len
)
2847 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2849 btrfs_drop_nlink(inode
);
2850 ret
= btrfs_update_inode(trans
, root
, inode
);
2856 /* helper to check if there is any shared block in the path */
2857 static int check_path_shared(struct btrfs_root
*root
,
2858 struct btrfs_path
*path
)
2860 struct extent_buffer
*eb
;
2864 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2867 if (!path
->nodes
[level
])
2869 eb
= path
->nodes
[level
];
2870 if (!btrfs_block_can_be_shared(root
, eb
))
2872 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2881 * helper to start transaction for unlink and rmdir.
2883 * unlink and rmdir are special in btrfs, they do not always free space.
2884 * so in enospc case, we should make sure they will free space before
2885 * allowing them to use the global metadata reservation.
2887 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2888 struct dentry
*dentry
)
2890 struct btrfs_trans_handle
*trans
;
2891 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2892 struct btrfs_path
*path
;
2893 struct btrfs_inode_ref
*ref
;
2894 struct btrfs_dir_item
*di
;
2895 struct inode
*inode
= dentry
->d_inode
;
2900 u64 ino
= btrfs_ino(inode
);
2901 u64 dir_ino
= btrfs_ino(dir
);
2904 * 1 for the possible orphan item
2905 * 1 for the dir item
2906 * 1 for the dir index
2907 * 1 for the inode ref
2908 * 1 for the inode ref in the tree log
2909 * 2 for the dir entries in the log
2912 trans
= btrfs_start_transaction(root
, 8);
2913 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2916 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2917 return ERR_PTR(-ENOSPC
);
2919 /* check if there is someone else holds reference */
2920 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2921 return ERR_PTR(-ENOSPC
);
2923 if (atomic_read(&inode
->i_count
) > 2)
2924 return ERR_PTR(-ENOSPC
);
2926 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2927 return ERR_PTR(-ENOSPC
);
2929 path
= btrfs_alloc_path();
2931 root
->fs_info
->enospc_unlink
= 0;
2932 return ERR_PTR(-ENOMEM
);
2935 /* 1 for the orphan item */
2936 trans
= btrfs_start_transaction(root
, 1);
2937 if (IS_ERR(trans
)) {
2938 btrfs_free_path(path
);
2939 root
->fs_info
->enospc_unlink
= 0;
2943 path
->skip_locking
= 1;
2944 path
->search_commit_root
= 1;
2946 ret
= btrfs_lookup_inode(trans
, root
, path
,
2947 &BTRFS_I(dir
)->location
, 0);
2953 if (check_path_shared(root
, path
))
2958 btrfs_release_path(path
);
2960 ret
= btrfs_lookup_inode(trans
, root
, path
,
2961 &BTRFS_I(inode
)->location
, 0);
2967 if (check_path_shared(root
, path
))
2972 btrfs_release_path(path
);
2974 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2975 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2981 BUG_ON(ret
== 0); /* Corruption */
2982 if (check_path_shared(root
, path
))
2984 btrfs_release_path(path
);
2992 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2993 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2999 if (check_path_shared(root
, path
))
3005 btrfs_release_path(path
);
3007 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
3008 dentry
->d_name
.name
, dentry
->d_name
.len
,
3014 BUG_ON(!ref
); /* Logic error */
3015 if (check_path_shared(root
, path
))
3017 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
3018 btrfs_release_path(path
);
3021 * This is a commit root search, if we can lookup inode item and other
3022 * relative items in the commit root, it means the transaction of
3023 * dir/file creation has been committed, and the dir index item that we
3024 * delay to insert has also been inserted into the commit root. So
3025 * we needn't worry about the delayed insertion of the dir index item
3028 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
3029 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
3034 BUG_ON(ret
== -ENOENT
);
3035 if (check_path_shared(root
, path
))
3040 btrfs_free_path(path
);
3041 /* Migrate the orphan reservation over */
3043 err
= btrfs_block_rsv_migrate(trans
->block_rsv
,
3044 &root
->fs_info
->global_block_rsv
,
3045 trans
->bytes_reserved
);
3048 btrfs_end_transaction(trans
, root
);
3049 root
->fs_info
->enospc_unlink
= 0;
3050 return ERR_PTR(err
);
3053 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
3057 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
3058 struct btrfs_root
*root
)
3060 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
3061 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3062 trans
->bytes_reserved
);
3063 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3064 BUG_ON(!root
->fs_info
->enospc_unlink
);
3065 root
->fs_info
->enospc_unlink
= 0;
3067 btrfs_end_transaction(trans
, root
);
3070 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3072 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3073 struct btrfs_trans_handle
*trans
;
3074 struct inode
*inode
= dentry
->d_inode
;
3076 unsigned long nr
= 0;
3078 trans
= __unlink_start_trans(dir
, dentry
);
3080 return PTR_ERR(trans
);
3082 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
3084 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3085 dentry
->d_name
.name
, dentry
->d_name
.len
);
3089 if (inode
->i_nlink
== 0) {
3090 ret
= btrfs_orphan_add(trans
, inode
);
3096 nr
= trans
->blocks_used
;
3097 __unlink_end_trans(trans
, root
);
3098 btrfs_btree_balance_dirty(root
, nr
);
3102 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
3103 struct btrfs_root
*root
,
3104 struct inode
*dir
, u64 objectid
,
3105 const char *name
, int name_len
)
3107 struct btrfs_path
*path
;
3108 struct extent_buffer
*leaf
;
3109 struct btrfs_dir_item
*di
;
3110 struct btrfs_key key
;
3113 u64 dir_ino
= btrfs_ino(dir
);
3115 path
= btrfs_alloc_path();
3119 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3120 name
, name_len
, -1);
3121 if (IS_ERR_OR_NULL(di
)) {
3129 leaf
= path
->nodes
[0];
3130 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3131 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3132 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3134 btrfs_abort_transaction(trans
, root
, ret
);
3137 btrfs_release_path(path
);
3139 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3140 objectid
, root
->root_key
.objectid
,
3141 dir_ino
, &index
, name
, name_len
);
3143 if (ret
!= -ENOENT
) {
3144 btrfs_abort_transaction(trans
, root
, ret
);
3147 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
3149 if (IS_ERR_OR_NULL(di
)) {
3154 btrfs_abort_transaction(trans
, root
, ret
);
3158 leaf
= path
->nodes
[0];
3159 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3160 btrfs_release_path(path
);
3163 btrfs_release_path(path
);
3165 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3167 btrfs_abort_transaction(trans
, root
, ret
);
3171 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3172 inode_inc_iversion(dir
);
3173 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3174 ret
= btrfs_update_inode(trans
, root
, dir
);
3176 btrfs_abort_transaction(trans
, root
, ret
);
3178 btrfs_free_path(path
);
3182 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3184 struct inode
*inode
= dentry
->d_inode
;
3186 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3187 struct btrfs_trans_handle
*trans
;
3188 unsigned long nr
= 0;
3190 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3191 btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3194 trans
= __unlink_start_trans(dir
, dentry
);
3196 return PTR_ERR(trans
);
3198 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3199 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3200 BTRFS_I(inode
)->location
.objectid
,
3201 dentry
->d_name
.name
,
3202 dentry
->d_name
.len
);
3206 err
= btrfs_orphan_add(trans
, inode
);
3210 /* now the directory is empty */
3211 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3212 dentry
->d_name
.name
, dentry
->d_name
.len
);
3214 btrfs_i_size_write(inode
, 0);
3216 nr
= trans
->blocks_used
;
3217 __unlink_end_trans(trans
, root
);
3218 btrfs_btree_balance_dirty(root
, nr
);
3224 * this can truncate away extent items, csum items and directory items.
3225 * It starts at a high offset and removes keys until it can't find
3226 * any higher than new_size
3228 * csum items that cross the new i_size are truncated to the new size
3231 * min_type is the minimum key type to truncate down to. If set to 0, this
3232 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3234 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3235 struct btrfs_root
*root
,
3236 struct inode
*inode
,
3237 u64 new_size
, u32 min_type
)
3239 struct btrfs_path
*path
;
3240 struct extent_buffer
*leaf
;
3241 struct btrfs_file_extent_item
*fi
;
3242 struct btrfs_key key
;
3243 struct btrfs_key found_key
;
3244 u64 extent_start
= 0;
3245 u64 extent_num_bytes
= 0;
3246 u64 extent_offset
= 0;
3248 u64 mask
= root
->sectorsize
- 1;
3249 u32 found_type
= (u8
)-1;
3252 int pending_del_nr
= 0;
3253 int pending_del_slot
= 0;
3254 int extent_type
= -1;
3257 u64 ino
= btrfs_ino(inode
);
3259 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3261 path
= btrfs_alloc_path();
3266 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3267 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3270 * This function is also used to drop the items in the log tree before
3271 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3272 * it is used to drop the loged items. So we shouldn't kill the delayed
3275 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3276 btrfs_kill_delayed_inode_items(inode
);
3279 key
.offset
= (u64
)-1;
3283 path
->leave_spinning
= 1;
3284 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3291 /* there are no items in the tree for us to truncate, we're
3294 if (path
->slots
[0] == 0)
3301 leaf
= path
->nodes
[0];
3302 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3303 found_type
= btrfs_key_type(&found_key
);
3305 if (found_key
.objectid
!= ino
)
3308 if (found_type
< min_type
)
3311 item_end
= found_key
.offset
;
3312 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3313 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3314 struct btrfs_file_extent_item
);
3315 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3316 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3318 btrfs_file_extent_num_bytes(leaf
, fi
);
3319 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3320 item_end
+= btrfs_file_extent_inline_len(leaf
,
3325 if (found_type
> min_type
) {
3328 if (item_end
< new_size
)
3330 if (found_key
.offset
>= new_size
)
3336 /* FIXME, shrink the extent if the ref count is only 1 */
3337 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3340 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3342 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3344 u64 orig_num_bytes
=
3345 btrfs_file_extent_num_bytes(leaf
, fi
);
3346 extent_num_bytes
= new_size
-
3347 found_key
.offset
+ root
->sectorsize
- 1;
3348 extent_num_bytes
= extent_num_bytes
&
3349 ~((u64
)root
->sectorsize
- 1);
3350 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3352 num_dec
= (orig_num_bytes
-
3354 if (root
->ref_cows
&& extent_start
!= 0)
3355 inode_sub_bytes(inode
, num_dec
);
3356 btrfs_mark_buffer_dirty(leaf
);
3359 btrfs_file_extent_disk_num_bytes(leaf
,
3361 extent_offset
= found_key
.offset
-
3362 btrfs_file_extent_offset(leaf
, fi
);
3364 /* FIXME blocksize != 4096 */
3365 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3366 if (extent_start
!= 0) {
3369 inode_sub_bytes(inode
, num_dec
);
3372 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3374 * we can't truncate inline items that have had
3378 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3379 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3380 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3381 u32 size
= new_size
- found_key
.offset
;
3383 if (root
->ref_cows
) {
3384 inode_sub_bytes(inode
, item_end
+ 1 -
3388 btrfs_file_extent_calc_inline_size(size
);
3389 btrfs_truncate_item(trans
, root
, path
,
3391 } else if (root
->ref_cows
) {
3392 inode_sub_bytes(inode
, item_end
+ 1 -
3398 if (!pending_del_nr
) {
3399 /* no pending yet, add ourselves */
3400 pending_del_slot
= path
->slots
[0];
3402 } else if (pending_del_nr
&&
3403 path
->slots
[0] + 1 == pending_del_slot
) {
3404 /* hop on the pending chunk */
3406 pending_del_slot
= path
->slots
[0];
3413 if (found_extent
&& (root
->ref_cows
||
3414 root
== root
->fs_info
->tree_root
)) {
3415 btrfs_set_path_blocking(path
);
3416 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3417 extent_num_bytes
, 0,
3418 btrfs_header_owner(leaf
),
3419 ino
, extent_offset
, 0);
3423 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3426 if (path
->slots
[0] == 0 ||
3427 path
->slots
[0] != pending_del_slot
) {
3428 if (root
->ref_cows
&&
3429 BTRFS_I(inode
)->location
.objectid
!=
3430 BTRFS_FREE_INO_OBJECTID
) {
3434 if (pending_del_nr
) {
3435 ret
= btrfs_del_items(trans
, root
, path
,
3439 btrfs_abort_transaction(trans
,
3445 btrfs_release_path(path
);
3452 if (pending_del_nr
) {
3453 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3456 btrfs_abort_transaction(trans
, root
, ret
);
3459 btrfs_free_path(path
);
3464 * taken from block_truncate_page, but does cow as it zeros out
3465 * any bytes left in the last page in the file.
3467 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3469 struct inode
*inode
= mapping
->host
;
3470 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3471 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3472 struct btrfs_ordered_extent
*ordered
;
3473 struct extent_state
*cached_state
= NULL
;
3475 u32 blocksize
= root
->sectorsize
;
3476 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3477 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3479 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
3484 if ((offset
& (blocksize
- 1)) == 0)
3486 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3492 page
= find_or_create_page(mapping
, index
, mask
);
3494 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3498 page_start
= page_offset(page
);
3499 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3501 if (!PageUptodate(page
)) {
3502 ret
= btrfs_readpage(NULL
, page
);
3504 if (page
->mapping
!= mapping
) {
3506 page_cache_release(page
);
3509 if (!PageUptodate(page
)) {
3514 wait_on_page_writeback(page
);
3516 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
3517 set_page_extent_mapped(page
);
3519 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3521 unlock_extent_cached(io_tree
, page_start
, page_end
,
3522 &cached_state
, GFP_NOFS
);
3524 page_cache_release(page
);
3525 btrfs_start_ordered_extent(inode
, ordered
, 1);
3526 btrfs_put_ordered_extent(ordered
);
3530 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3531 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3532 0, 0, &cached_state
, GFP_NOFS
);
3534 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3537 unlock_extent_cached(io_tree
, page_start
, page_end
,
3538 &cached_state
, GFP_NOFS
);
3543 if (offset
!= PAGE_CACHE_SIZE
) {
3545 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3546 flush_dcache_page(page
);
3549 ClearPageChecked(page
);
3550 set_page_dirty(page
);
3551 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3556 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3558 page_cache_release(page
);
3564 * This function puts in dummy file extents for the area we're creating a hole
3565 * for. So if we are truncating this file to a larger size we need to insert
3566 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3567 * the range between oldsize and size
3569 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3571 struct btrfs_trans_handle
*trans
;
3572 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3573 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3574 struct extent_map
*em
= NULL
;
3575 struct extent_state
*cached_state
= NULL
;
3576 u64 mask
= root
->sectorsize
- 1;
3577 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3578 u64 block_end
= (size
+ mask
) & ~mask
;
3584 if (size
<= hole_start
)
3588 struct btrfs_ordered_extent
*ordered
;
3589 btrfs_wait_ordered_range(inode
, hole_start
,
3590 block_end
- hole_start
);
3591 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3593 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3596 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3597 &cached_state
, GFP_NOFS
);
3598 btrfs_put_ordered_extent(ordered
);
3601 cur_offset
= hole_start
;
3603 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3604 block_end
- cur_offset
, 0);
3609 last_byte
= min(extent_map_end(em
), block_end
);
3610 last_byte
= (last_byte
+ mask
) & ~mask
;
3611 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3613 hole_size
= last_byte
- cur_offset
;
3615 trans
= btrfs_start_transaction(root
, 3);
3616 if (IS_ERR(trans
)) {
3617 err
= PTR_ERR(trans
);
3621 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3622 cur_offset
+ hole_size
,
3625 btrfs_abort_transaction(trans
, root
, err
);
3626 btrfs_end_transaction(trans
, root
);
3630 err
= btrfs_insert_file_extent(trans
, root
,
3631 btrfs_ino(inode
), cur_offset
, 0,
3632 0, hole_size
, 0, hole_size
,
3635 btrfs_abort_transaction(trans
, root
, err
);
3636 btrfs_end_transaction(trans
, root
);
3640 btrfs_drop_extent_cache(inode
, hole_start
,
3643 btrfs_update_inode(trans
, root
, inode
);
3644 btrfs_end_transaction(trans
, root
);
3646 free_extent_map(em
);
3648 cur_offset
= last_byte
;
3649 if (cur_offset
>= block_end
)
3653 free_extent_map(em
);
3654 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3659 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3661 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3662 struct btrfs_trans_handle
*trans
;
3663 loff_t oldsize
= i_size_read(inode
);
3666 if (newsize
== oldsize
)
3669 if (newsize
> oldsize
) {
3670 truncate_pagecache(inode
, oldsize
, newsize
);
3671 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3675 trans
= btrfs_start_transaction(root
, 1);
3677 return PTR_ERR(trans
);
3679 i_size_write(inode
, newsize
);
3680 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3681 ret
= btrfs_update_inode(trans
, root
, inode
);
3682 btrfs_end_transaction(trans
, root
);
3686 * We're truncating a file that used to have good data down to
3687 * zero. Make sure it gets into the ordered flush list so that
3688 * any new writes get down to disk quickly.
3691 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
3692 &BTRFS_I(inode
)->runtime_flags
);
3694 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3695 truncate_setsize(inode
, newsize
);
3696 ret
= btrfs_truncate(inode
);
3702 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3704 struct inode
*inode
= dentry
->d_inode
;
3705 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3708 if (btrfs_root_readonly(root
))
3711 err
= inode_change_ok(inode
, attr
);
3715 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3716 err
= btrfs_setsize(inode
, attr
->ia_size
);
3721 if (attr
->ia_valid
) {
3722 setattr_copy(inode
, attr
);
3723 inode_inc_iversion(inode
);
3724 err
= btrfs_dirty_inode(inode
);
3726 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
3727 err
= btrfs_acl_chmod(inode
);
3733 void btrfs_evict_inode(struct inode
*inode
)
3735 struct btrfs_trans_handle
*trans
;
3736 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3737 struct btrfs_block_rsv
*rsv
, *global_rsv
;
3738 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
3742 trace_btrfs_inode_evict(inode
);
3744 truncate_inode_pages(&inode
->i_data
, 0);
3745 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3746 btrfs_is_free_space_inode(root
, inode
)))
3749 if (is_bad_inode(inode
)) {
3750 btrfs_orphan_del(NULL
, inode
);
3753 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3754 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3756 if (root
->fs_info
->log_root_recovering
) {
3757 BUG_ON(!test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3758 &BTRFS_I(inode
)->runtime_flags
));
3762 if (inode
->i_nlink
> 0) {
3763 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3767 rsv
= btrfs_alloc_block_rsv(root
);
3769 btrfs_orphan_del(NULL
, inode
);
3772 rsv
->size
= min_size
;
3773 global_rsv
= &root
->fs_info
->global_block_rsv
;
3775 btrfs_i_size_write(inode
, 0);
3778 * This is a bit simpler than btrfs_truncate since
3780 * 1) We've already reserved our space for our orphan item in the
3782 * 2) We're going to delete the inode item, so we don't need to update
3785 * So we just need to reserve some slack space in case we add bytes when
3786 * doing the truncate.
3789 ret
= btrfs_block_rsv_refill_noflush(root
, rsv
, min_size
);
3792 * Try and steal from the global reserve since we will
3793 * likely not use this space anyway, we want to try as
3794 * hard as possible to get this to work.
3797 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, min_size
);
3800 printk(KERN_WARNING
"Could not get space for a "
3801 "delete, will truncate on mount %d\n", ret
);
3802 btrfs_orphan_del(NULL
, inode
);
3803 btrfs_free_block_rsv(root
, rsv
);
3807 trans
= btrfs_start_transaction(root
, 0);
3808 if (IS_ERR(trans
)) {
3809 btrfs_orphan_del(NULL
, inode
);
3810 btrfs_free_block_rsv(root
, rsv
);
3814 trans
->block_rsv
= rsv
;
3816 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3820 nr
= trans
->blocks_used
;
3821 btrfs_end_transaction(trans
, root
);
3823 btrfs_btree_balance_dirty(root
, nr
);
3826 btrfs_free_block_rsv(root
, rsv
);
3829 trans
->block_rsv
= root
->orphan_block_rsv
;
3830 ret
= btrfs_orphan_del(trans
, inode
);
3834 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3835 if (!(root
== root
->fs_info
->tree_root
||
3836 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
3837 btrfs_return_ino(root
, btrfs_ino(inode
));
3839 nr
= trans
->blocks_used
;
3840 btrfs_end_transaction(trans
, root
);
3841 btrfs_btree_balance_dirty(root
, nr
);
3848 * this returns the key found in the dir entry in the location pointer.
3849 * If no dir entries were found, location->objectid is 0.
3851 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3852 struct btrfs_key
*location
)
3854 const char *name
= dentry
->d_name
.name
;
3855 int namelen
= dentry
->d_name
.len
;
3856 struct btrfs_dir_item
*di
;
3857 struct btrfs_path
*path
;
3858 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3861 path
= btrfs_alloc_path();
3865 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
3870 if (IS_ERR_OR_NULL(di
))
3873 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3875 btrfs_free_path(path
);
3878 location
->objectid
= 0;
3883 * when we hit a tree root in a directory, the btrfs part of the inode
3884 * needs to be changed to reflect the root directory of the tree root. This
3885 * is kind of like crossing a mount point.
3887 static int fixup_tree_root_location(struct btrfs_root
*root
,
3889 struct dentry
*dentry
,
3890 struct btrfs_key
*location
,
3891 struct btrfs_root
**sub_root
)
3893 struct btrfs_path
*path
;
3894 struct btrfs_root
*new_root
;
3895 struct btrfs_root_ref
*ref
;
3896 struct extent_buffer
*leaf
;
3900 path
= btrfs_alloc_path();
3907 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3908 BTRFS_I(dir
)->root
->root_key
.objectid
,
3909 location
->objectid
);
3916 leaf
= path
->nodes
[0];
3917 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3918 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
3919 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3922 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3923 (unsigned long)(ref
+ 1),
3924 dentry
->d_name
.len
);
3928 btrfs_release_path(path
);
3930 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3931 if (IS_ERR(new_root
)) {
3932 err
= PTR_ERR(new_root
);
3936 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3941 *sub_root
= new_root
;
3942 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3943 location
->type
= BTRFS_INODE_ITEM_KEY
;
3944 location
->offset
= 0;
3947 btrfs_free_path(path
);
3951 static void inode_tree_add(struct inode
*inode
)
3953 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3954 struct btrfs_inode
*entry
;
3956 struct rb_node
*parent
;
3957 u64 ino
= btrfs_ino(inode
);
3959 p
= &root
->inode_tree
.rb_node
;
3962 if (inode_unhashed(inode
))
3965 spin_lock(&root
->inode_lock
);
3968 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3970 if (ino
< btrfs_ino(&entry
->vfs_inode
))
3971 p
= &parent
->rb_left
;
3972 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
3973 p
= &parent
->rb_right
;
3975 WARN_ON(!(entry
->vfs_inode
.i_state
&
3976 (I_WILL_FREE
| I_FREEING
)));
3977 rb_erase(parent
, &root
->inode_tree
);
3978 RB_CLEAR_NODE(parent
);
3979 spin_unlock(&root
->inode_lock
);
3983 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3984 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3985 spin_unlock(&root
->inode_lock
);
3988 static void inode_tree_del(struct inode
*inode
)
3990 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3993 spin_lock(&root
->inode_lock
);
3994 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3995 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3996 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3997 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3999 spin_unlock(&root
->inode_lock
);
4002 * Free space cache has inodes in the tree root, but the tree root has a
4003 * root_refs of 0, so this could end up dropping the tree root as a
4004 * snapshot, so we need the extra !root->fs_info->tree_root check to
4005 * make sure we don't drop it.
4007 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
4008 root
!= root
->fs_info
->tree_root
) {
4009 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
4010 spin_lock(&root
->inode_lock
);
4011 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4012 spin_unlock(&root
->inode_lock
);
4014 btrfs_add_dead_root(root
);
4018 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
4020 struct rb_node
*node
;
4021 struct rb_node
*prev
;
4022 struct btrfs_inode
*entry
;
4023 struct inode
*inode
;
4026 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4028 spin_lock(&root
->inode_lock
);
4030 node
= root
->inode_tree
.rb_node
;
4034 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4036 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
4037 node
= node
->rb_left
;
4038 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
4039 node
= node
->rb_right
;
4045 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4046 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
4050 prev
= rb_next(prev
);
4054 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4055 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
4056 inode
= igrab(&entry
->vfs_inode
);
4058 spin_unlock(&root
->inode_lock
);
4059 if (atomic_read(&inode
->i_count
) > 1)
4060 d_prune_aliases(inode
);
4062 * btrfs_drop_inode will have it removed from
4063 * the inode cache when its usage count
4068 spin_lock(&root
->inode_lock
);
4072 if (cond_resched_lock(&root
->inode_lock
))
4075 node
= rb_next(node
);
4077 spin_unlock(&root
->inode_lock
);
4080 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4082 struct btrfs_iget_args
*args
= p
;
4083 inode
->i_ino
= args
->ino
;
4084 BTRFS_I(inode
)->root
= args
->root
;
4085 btrfs_set_inode_space_info(args
->root
, inode
);
4089 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4091 struct btrfs_iget_args
*args
= opaque
;
4092 return args
->ino
== btrfs_ino(inode
) &&
4093 args
->root
== BTRFS_I(inode
)->root
;
4096 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4098 struct btrfs_root
*root
)
4100 struct inode
*inode
;
4101 struct btrfs_iget_args args
;
4102 args
.ino
= objectid
;
4105 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4106 btrfs_init_locked_inode
,
4111 /* Get an inode object given its location and corresponding root.
4112 * Returns in *is_new if the inode was read from disk
4114 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4115 struct btrfs_root
*root
, int *new)
4117 struct inode
*inode
;
4119 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4121 return ERR_PTR(-ENOMEM
);
4123 if (inode
->i_state
& I_NEW
) {
4124 BTRFS_I(inode
)->root
= root
;
4125 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4126 btrfs_read_locked_inode(inode
);
4127 if (!is_bad_inode(inode
)) {
4128 inode_tree_add(inode
);
4129 unlock_new_inode(inode
);
4133 unlock_new_inode(inode
);
4135 inode
= ERR_PTR(-ESTALE
);
4142 static struct inode
*new_simple_dir(struct super_block
*s
,
4143 struct btrfs_key
*key
,
4144 struct btrfs_root
*root
)
4146 struct inode
*inode
= new_inode(s
);
4149 return ERR_PTR(-ENOMEM
);
4151 BTRFS_I(inode
)->root
= root
;
4152 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4153 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
4155 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4156 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
4157 inode
->i_fop
= &simple_dir_operations
;
4158 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4159 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4164 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4166 struct inode
*inode
;
4167 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4168 struct btrfs_root
*sub_root
= root
;
4169 struct btrfs_key location
;
4173 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4174 return ERR_PTR(-ENAMETOOLONG
);
4176 if (unlikely(d_need_lookup(dentry
))) {
4177 memcpy(&location
, dentry
->d_fsdata
, sizeof(struct btrfs_key
));
4178 kfree(dentry
->d_fsdata
);
4179 dentry
->d_fsdata
= NULL
;
4180 /* This thing is hashed, drop it for now */
4183 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4187 return ERR_PTR(ret
);
4189 if (location
.objectid
== 0)
4192 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4193 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4197 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4199 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4200 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4201 &location
, &sub_root
);
4204 inode
= ERR_PTR(ret
);
4206 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4208 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4210 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4212 if (!IS_ERR(inode
) && root
!= sub_root
) {
4213 down_read(&root
->fs_info
->cleanup_work_sem
);
4214 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4215 ret
= btrfs_orphan_cleanup(sub_root
);
4216 up_read(&root
->fs_info
->cleanup_work_sem
);
4218 inode
= ERR_PTR(ret
);
4224 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4226 struct btrfs_root
*root
;
4227 struct inode
*inode
= dentry
->d_inode
;
4229 if (!inode
&& !IS_ROOT(dentry
))
4230 inode
= dentry
->d_parent
->d_inode
;
4233 root
= BTRFS_I(inode
)->root
;
4234 if (btrfs_root_refs(&root
->root_item
) == 0)
4237 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
4243 static void btrfs_dentry_release(struct dentry
*dentry
)
4245 if (dentry
->d_fsdata
)
4246 kfree(dentry
->d_fsdata
);
4249 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4250 struct nameidata
*nd
)
4254 ret
= d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
4255 if (unlikely(d_need_lookup(dentry
))) {
4256 spin_lock(&dentry
->d_lock
);
4257 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
4258 spin_unlock(&dentry
->d_lock
);
4263 unsigned char btrfs_filetype_table
[] = {
4264 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4267 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4270 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4271 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4272 struct btrfs_item
*item
;
4273 struct btrfs_dir_item
*di
;
4274 struct btrfs_key key
;
4275 struct btrfs_key found_key
;
4276 struct btrfs_path
*path
;
4277 struct list_head ins_list
;
4278 struct list_head del_list
;
4280 struct extent_buffer
*leaf
;
4282 unsigned char d_type
;
4287 int key_type
= BTRFS_DIR_INDEX_KEY
;
4291 int is_curr
= 0; /* filp->f_pos points to the current index? */
4293 /* FIXME, use a real flag for deciding about the key type */
4294 if (root
->fs_info
->tree_root
== root
)
4295 key_type
= BTRFS_DIR_ITEM_KEY
;
4297 /* special case for "." */
4298 if (filp
->f_pos
== 0) {
4299 over
= filldir(dirent
, ".", 1,
4300 filp
->f_pos
, btrfs_ino(inode
), DT_DIR
);
4305 /* special case for .., just use the back ref */
4306 if (filp
->f_pos
== 1) {
4307 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4308 over
= filldir(dirent
, "..", 2,
4309 filp
->f_pos
, pino
, DT_DIR
);
4314 path
= btrfs_alloc_path();
4320 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4321 INIT_LIST_HEAD(&ins_list
);
4322 INIT_LIST_HEAD(&del_list
);
4323 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
4326 btrfs_set_key_type(&key
, key_type
);
4327 key
.offset
= filp
->f_pos
;
4328 key
.objectid
= btrfs_ino(inode
);
4330 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4335 leaf
= path
->nodes
[0];
4336 slot
= path
->slots
[0];
4337 if (slot
>= btrfs_header_nritems(leaf
)) {
4338 ret
= btrfs_next_leaf(root
, path
);
4346 item
= btrfs_item_nr(leaf
, slot
);
4347 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4349 if (found_key
.objectid
!= key
.objectid
)
4351 if (btrfs_key_type(&found_key
) != key_type
)
4353 if (found_key
.offset
< filp
->f_pos
)
4355 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
4356 btrfs_should_delete_dir_index(&del_list
,
4360 filp
->f_pos
= found_key
.offset
;
4363 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4365 di_total
= btrfs_item_size(leaf
, item
);
4367 while (di_cur
< di_total
) {
4368 struct btrfs_key location
;
4370 if (verify_dir_item(root
, leaf
, di
))
4373 name_len
= btrfs_dir_name_len(leaf
, di
);
4374 if (name_len
<= sizeof(tmp_name
)) {
4375 name_ptr
= tmp_name
;
4377 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4383 read_extent_buffer(leaf
, name_ptr
,
4384 (unsigned long)(di
+ 1), name_len
);
4386 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4387 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4390 /* is this a reference to our own snapshot? If so
4393 * In contrast to old kernels, we insert the snapshot's
4394 * dir item and dir index after it has been created, so
4395 * we won't find a reference to our own snapshot. We
4396 * still keep the following code for backward
4399 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4400 location
.objectid
== root
->root_key
.objectid
) {
4404 over
= filldir(dirent
, name_ptr
, name_len
,
4405 found_key
.offset
, location
.objectid
,
4409 if (name_ptr
!= tmp_name
)
4414 di_len
= btrfs_dir_name_len(leaf
, di
) +
4415 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4417 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4423 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4426 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
4432 /* Reached end of directory/root. Bump pos past the last item. */
4433 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4435 * 32-bit glibc will use getdents64, but then strtol -
4436 * so the last number we can serve is this.
4438 filp
->f_pos
= 0x7fffffff;
4444 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4445 btrfs_put_delayed_items(&ins_list
, &del_list
);
4446 btrfs_free_path(path
);
4450 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4452 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4453 struct btrfs_trans_handle
*trans
;
4455 bool nolock
= false;
4457 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
4460 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(root
, inode
))
4463 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4465 trans
= btrfs_join_transaction_nolock(root
);
4467 trans
= btrfs_join_transaction(root
);
4469 return PTR_ERR(trans
);
4471 ret
= btrfs_end_transaction_nolock(trans
, root
);
4473 ret
= btrfs_commit_transaction(trans
, root
);
4479 * This is somewhat expensive, updating the tree every time the
4480 * inode changes. But, it is most likely to find the inode in cache.
4481 * FIXME, needs more benchmarking...there are no reasons other than performance
4482 * to keep or drop this code.
4484 int btrfs_dirty_inode(struct inode
*inode
)
4486 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4487 struct btrfs_trans_handle
*trans
;
4490 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
4493 trans
= btrfs_join_transaction(root
);
4495 return PTR_ERR(trans
);
4497 ret
= btrfs_update_inode(trans
, root
, inode
);
4498 if (ret
&& ret
== -ENOSPC
) {
4499 /* whoops, lets try again with the full transaction */
4500 btrfs_end_transaction(trans
, root
);
4501 trans
= btrfs_start_transaction(root
, 1);
4503 return PTR_ERR(trans
);
4505 ret
= btrfs_update_inode(trans
, root
, inode
);
4507 btrfs_end_transaction(trans
, root
);
4508 if (BTRFS_I(inode
)->delayed_node
)
4509 btrfs_balance_delayed_items(root
);
4515 * This is a copy of file_update_time. We need this so we can return error on
4516 * ENOSPC for updating the inode in the case of file write and mmap writes.
4518 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
4521 if (flags
& S_VERSION
)
4522 inode_inc_iversion(inode
);
4523 if (flags
& S_CTIME
)
4524 inode
->i_ctime
= *now
;
4525 if (flags
& S_MTIME
)
4526 inode
->i_mtime
= *now
;
4527 if (flags
& S_ATIME
)
4528 inode
->i_atime
= *now
;
4529 return btrfs_dirty_inode(inode
);
4533 * find the highest existing sequence number in a directory
4534 * and then set the in-memory index_cnt variable to reflect
4535 * free sequence numbers
4537 static int btrfs_set_inode_index_count(struct inode
*inode
)
4539 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4540 struct btrfs_key key
, found_key
;
4541 struct btrfs_path
*path
;
4542 struct extent_buffer
*leaf
;
4545 key
.objectid
= btrfs_ino(inode
);
4546 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4547 key
.offset
= (u64
)-1;
4549 path
= btrfs_alloc_path();
4553 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4556 /* FIXME: we should be able to handle this */
4562 * MAGIC NUMBER EXPLANATION:
4563 * since we search a directory based on f_pos we have to start at 2
4564 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4565 * else has to start at 2
4567 if (path
->slots
[0] == 0) {
4568 BTRFS_I(inode
)->index_cnt
= 2;
4574 leaf
= path
->nodes
[0];
4575 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4577 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4578 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4579 BTRFS_I(inode
)->index_cnt
= 2;
4583 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4585 btrfs_free_path(path
);
4590 * helper to find a free sequence number in a given directory. This current
4591 * code is very simple, later versions will do smarter things in the btree
4593 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4597 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4598 ret
= btrfs_inode_delayed_dir_index_count(dir
);
4600 ret
= btrfs_set_inode_index_count(dir
);
4606 *index
= BTRFS_I(dir
)->index_cnt
;
4607 BTRFS_I(dir
)->index_cnt
++;
4612 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4613 struct btrfs_root
*root
,
4615 const char *name
, int name_len
,
4616 u64 ref_objectid
, u64 objectid
,
4617 umode_t mode
, u64
*index
)
4619 struct inode
*inode
;
4620 struct btrfs_inode_item
*inode_item
;
4621 struct btrfs_key
*location
;
4622 struct btrfs_path
*path
;
4623 struct btrfs_inode_ref
*ref
;
4624 struct btrfs_key key
[2];
4630 path
= btrfs_alloc_path();
4632 return ERR_PTR(-ENOMEM
);
4634 inode
= new_inode(root
->fs_info
->sb
);
4636 btrfs_free_path(path
);
4637 return ERR_PTR(-ENOMEM
);
4641 * we have to initialize this early, so we can reclaim the inode
4642 * number if we fail afterwards in this function.
4644 inode
->i_ino
= objectid
;
4647 trace_btrfs_inode_request(dir
);
4649 ret
= btrfs_set_inode_index(dir
, index
);
4651 btrfs_free_path(path
);
4653 return ERR_PTR(ret
);
4657 * index_cnt is ignored for everything but a dir,
4658 * btrfs_get_inode_index_count has an explanation for the magic
4661 BTRFS_I(inode
)->index_cnt
= 2;
4662 BTRFS_I(inode
)->root
= root
;
4663 BTRFS_I(inode
)->generation
= trans
->transid
;
4664 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4665 btrfs_set_inode_space_info(root
, inode
);
4672 key
[0].objectid
= objectid
;
4673 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4676 key
[1].objectid
= objectid
;
4677 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4678 key
[1].offset
= ref_objectid
;
4680 sizes
[0] = sizeof(struct btrfs_inode_item
);
4681 sizes
[1] = name_len
+ sizeof(*ref
);
4683 path
->leave_spinning
= 1;
4684 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4688 inode_init_owner(inode
, dir
, mode
);
4689 inode_set_bytes(inode
, 0);
4690 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4691 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4692 struct btrfs_inode_item
);
4693 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4695 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4696 struct btrfs_inode_ref
);
4697 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4698 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4699 ptr
= (unsigned long)(ref
+ 1);
4700 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4702 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4703 btrfs_free_path(path
);
4705 location
= &BTRFS_I(inode
)->location
;
4706 location
->objectid
= objectid
;
4707 location
->offset
= 0;
4708 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4710 btrfs_inherit_iflags(inode
, dir
);
4712 if (S_ISREG(mode
)) {
4713 if (btrfs_test_opt(root
, NODATASUM
))
4714 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4715 if (btrfs_test_opt(root
, NODATACOW
) ||
4716 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4717 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4720 insert_inode_hash(inode
);
4721 inode_tree_add(inode
);
4723 trace_btrfs_inode_new(inode
);
4724 btrfs_set_inode_last_trans(trans
, inode
);
4729 BTRFS_I(dir
)->index_cnt
--;
4730 btrfs_free_path(path
);
4732 return ERR_PTR(ret
);
4735 static inline u8
btrfs_inode_type(struct inode
*inode
)
4737 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4741 * utility function to add 'inode' into 'parent_inode' with
4742 * a give name and a given sequence number.
4743 * if 'add_backref' is true, also insert a backref from the
4744 * inode to the parent directory.
4746 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4747 struct inode
*parent_inode
, struct inode
*inode
,
4748 const char *name
, int name_len
, int add_backref
, u64 index
)
4751 struct btrfs_key key
;
4752 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4753 u64 ino
= btrfs_ino(inode
);
4754 u64 parent_ino
= btrfs_ino(parent_inode
);
4756 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4757 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4760 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4764 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4765 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4766 key
.objectid
, root
->root_key
.objectid
,
4767 parent_ino
, index
, name
, name_len
);
4768 } else if (add_backref
) {
4769 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
4773 /* Nothing to clean up yet */
4777 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4779 btrfs_inode_type(inode
), index
);
4783 btrfs_abort_transaction(trans
, root
, ret
);
4787 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4789 inode_inc_iversion(parent_inode
);
4790 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4791 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4793 btrfs_abort_transaction(trans
, root
, ret
);
4797 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4800 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4801 key
.objectid
, root
->root_key
.objectid
,
4802 parent_ino
, &local_index
, name
, name_len
);
4804 } else if (add_backref
) {
4808 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
4809 ino
, parent_ino
, &local_index
);
4814 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4815 struct inode
*dir
, struct dentry
*dentry
,
4816 struct inode
*inode
, int backref
, u64 index
)
4818 int err
= btrfs_add_link(trans
, dir
, inode
,
4819 dentry
->d_name
.name
, dentry
->d_name
.len
,
4826 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4827 umode_t mode
, dev_t rdev
)
4829 struct btrfs_trans_handle
*trans
;
4830 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4831 struct inode
*inode
= NULL
;
4835 unsigned long nr
= 0;
4838 if (!new_valid_dev(rdev
))
4842 * 2 for inode item and ref
4844 * 1 for xattr if selinux is on
4846 trans
= btrfs_start_transaction(root
, 5);
4848 return PTR_ERR(trans
);
4850 err
= btrfs_find_free_ino(root
, &objectid
);
4854 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4855 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4857 if (IS_ERR(inode
)) {
4858 err
= PTR_ERR(inode
);
4862 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4869 * If the active LSM wants to access the inode during
4870 * d_instantiate it needs these. Smack checks to see
4871 * if the filesystem supports xattrs by looking at the
4875 inode
->i_op
= &btrfs_special_inode_operations
;
4876 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4880 init_special_inode(inode
, inode
->i_mode
, rdev
);
4881 btrfs_update_inode(trans
, root
, inode
);
4882 d_instantiate(dentry
, inode
);
4885 nr
= trans
->blocks_used
;
4886 btrfs_end_transaction(trans
, root
);
4887 btrfs_btree_balance_dirty(root
, nr
);
4889 inode_dec_link_count(inode
);
4895 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4896 umode_t mode
, struct nameidata
*nd
)
4898 struct btrfs_trans_handle
*trans
;
4899 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4900 struct inode
*inode
= NULL
;
4903 unsigned long nr
= 0;
4908 * 2 for inode item and ref
4910 * 1 for xattr if selinux is on
4912 trans
= btrfs_start_transaction(root
, 5);
4914 return PTR_ERR(trans
);
4916 err
= btrfs_find_free_ino(root
, &objectid
);
4920 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4921 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4923 if (IS_ERR(inode
)) {
4924 err
= PTR_ERR(inode
);
4928 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4935 * If the active LSM wants to access the inode during
4936 * d_instantiate it needs these. Smack checks to see
4937 * if the filesystem supports xattrs by looking at the
4940 inode
->i_fop
= &btrfs_file_operations
;
4941 inode
->i_op
= &btrfs_file_inode_operations
;
4943 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4947 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4948 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4949 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4950 d_instantiate(dentry
, inode
);
4953 nr
= trans
->blocks_used
;
4954 btrfs_end_transaction(trans
, root
);
4956 inode_dec_link_count(inode
);
4959 btrfs_btree_balance_dirty(root
, nr
);
4963 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4964 struct dentry
*dentry
)
4966 struct btrfs_trans_handle
*trans
;
4967 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4968 struct inode
*inode
= old_dentry
->d_inode
;
4970 unsigned long nr
= 0;
4974 /* do not allow sys_link's with other subvols of the same device */
4975 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4978 if (inode
->i_nlink
== ~0U)
4981 err
= btrfs_set_inode_index(dir
, &index
);
4986 * 2 items for inode and inode ref
4987 * 2 items for dir items
4988 * 1 item for parent inode
4990 trans
= btrfs_start_transaction(root
, 5);
4991 if (IS_ERR(trans
)) {
4992 err
= PTR_ERR(trans
);
4996 btrfs_inc_nlink(inode
);
4997 inode_inc_iversion(inode
);
4998 inode
->i_ctime
= CURRENT_TIME
;
5001 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
5006 struct dentry
*parent
= dentry
->d_parent
;
5007 err
= btrfs_update_inode(trans
, root
, inode
);
5010 d_instantiate(dentry
, inode
);
5011 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
5014 nr
= trans
->blocks_used
;
5015 btrfs_end_transaction(trans
, root
);
5018 inode_dec_link_count(inode
);
5021 btrfs_btree_balance_dirty(root
, nr
);
5025 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
5027 struct inode
*inode
= NULL
;
5028 struct btrfs_trans_handle
*trans
;
5029 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5031 int drop_on_err
= 0;
5034 unsigned long nr
= 1;
5037 * 2 items for inode and ref
5038 * 2 items for dir items
5039 * 1 for xattr if selinux is on
5041 trans
= btrfs_start_transaction(root
, 5);
5043 return PTR_ERR(trans
);
5045 err
= btrfs_find_free_ino(root
, &objectid
);
5049 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5050 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5051 S_IFDIR
| mode
, &index
);
5052 if (IS_ERR(inode
)) {
5053 err
= PTR_ERR(inode
);
5059 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5063 inode
->i_op
= &btrfs_dir_inode_operations
;
5064 inode
->i_fop
= &btrfs_dir_file_operations
;
5066 btrfs_i_size_write(inode
, 0);
5067 err
= btrfs_update_inode(trans
, root
, inode
);
5071 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
5072 dentry
->d_name
.len
, 0, index
);
5076 d_instantiate(dentry
, inode
);
5080 nr
= trans
->blocks_used
;
5081 btrfs_end_transaction(trans
, root
);
5084 btrfs_btree_balance_dirty(root
, nr
);
5088 /* helper for btfs_get_extent. Given an existing extent in the tree,
5089 * and an extent that you want to insert, deal with overlap and insert
5090 * the new extent into the tree.
5092 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
5093 struct extent_map
*existing
,
5094 struct extent_map
*em
,
5095 u64 map_start
, u64 map_len
)
5099 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
5100 start_diff
= map_start
- em
->start
;
5101 em
->start
= map_start
;
5103 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
5104 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
5105 em
->block_start
+= start_diff
;
5106 em
->block_len
-= start_diff
;
5108 return add_extent_mapping(em_tree
, em
);
5111 static noinline
int uncompress_inline(struct btrfs_path
*path
,
5112 struct inode
*inode
, struct page
*page
,
5113 size_t pg_offset
, u64 extent_offset
,
5114 struct btrfs_file_extent_item
*item
)
5117 struct extent_buffer
*leaf
= path
->nodes
[0];
5120 unsigned long inline_size
;
5124 WARN_ON(pg_offset
!= 0);
5125 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5126 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
5127 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
5128 btrfs_item_nr(leaf
, path
->slots
[0]));
5129 tmp
= kmalloc(inline_size
, GFP_NOFS
);
5132 ptr
= btrfs_file_extent_inline_start(item
);
5134 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
5136 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
5137 ret
= btrfs_decompress(compress_type
, tmp
, page
,
5138 extent_offset
, inline_size
, max_size
);
5140 char *kaddr
= kmap_atomic(page
);
5141 unsigned long copy_size
= min_t(u64
,
5142 PAGE_CACHE_SIZE
- pg_offset
,
5143 max_size
- extent_offset
);
5144 memset(kaddr
+ pg_offset
, 0, copy_size
);
5145 kunmap_atomic(kaddr
);
5152 * a bit scary, this does extent mapping from logical file offset to the disk.
5153 * the ugly parts come from merging extents from the disk with the in-ram
5154 * representation. This gets more complex because of the data=ordered code,
5155 * where the in-ram extents might be locked pending data=ordered completion.
5157 * This also copies inline extents directly into the page.
5160 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
5161 size_t pg_offset
, u64 start
, u64 len
,
5167 u64 extent_start
= 0;
5169 u64 objectid
= btrfs_ino(inode
);
5171 struct btrfs_path
*path
= NULL
;
5172 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5173 struct btrfs_file_extent_item
*item
;
5174 struct extent_buffer
*leaf
;
5175 struct btrfs_key found_key
;
5176 struct extent_map
*em
= NULL
;
5177 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5178 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5179 struct btrfs_trans_handle
*trans
= NULL
;
5183 read_lock(&em_tree
->lock
);
5184 em
= lookup_extent_mapping(em_tree
, start
, len
);
5186 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5187 read_unlock(&em_tree
->lock
);
5190 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5191 free_extent_map(em
);
5192 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5193 free_extent_map(em
);
5197 em
= alloc_extent_map();
5202 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5203 em
->start
= EXTENT_MAP_HOLE
;
5204 em
->orig_start
= EXTENT_MAP_HOLE
;
5206 em
->block_len
= (u64
)-1;
5209 path
= btrfs_alloc_path();
5215 * Chances are we'll be called again, so go ahead and do
5221 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5222 objectid
, start
, trans
!= NULL
);
5229 if (path
->slots
[0] == 0)
5234 leaf
= path
->nodes
[0];
5235 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5236 struct btrfs_file_extent_item
);
5237 /* are we inside the extent that was found? */
5238 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5239 found_type
= btrfs_key_type(&found_key
);
5240 if (found_key
.objectid
!= objectid
||
5241 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5245 found_type
= btrfs_file_extent_type(leaf
, item
);
5246 extent_start
= found_key
.offset
;
5247 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5248 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5249 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5250 extent_end
= extent_start
+
5251 btrfs_file_extent_num_bytes(leaf
, item
);
5252 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5254 size
= btrfs_file_extent_inline_len(leaf
, item
);
5255 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5256 ~((u64
)root
->sectorsize
- 1);
5259 if (start
>= extent_end
) {
5261 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5262 ret
= btrfs_next_leaf(root
, path
);
5269 leaf
= path
->nodes
[0];
5271 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5272 if (found_key
.objectid
!= objectid
||
5273 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5275 if (start
+ len
<= found_key
.offset
)
5278 em
->len
= found_key
.offset
- start
;
5282 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5283 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5284 em
->start
= extent_start
;
5285 em
->len
= extent_end
- extent_start
;
5286 em
->orig_start
= extent_start
-
5287 btrfs_file_extent_offset(leaf
, item
);
5288 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5290 em
->block_start
= EXTENT_MAP_HOLE
;
5293 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5294 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5295 em
->compress_type
= compress_type
;
5296 em
->block_start
= bytenr
;
5297 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5300 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5301 em
->block_start
= bytenr
;
5302 em
->block_len
= em
->len
;
5303 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5304 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5307 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5311 size_t extent_offset
;
5314 em
->block_start
= EXTENT_MAP_INLINE
;
5315 if (!page
|| create
) {
5316 em
->start
= extent_start
;
5317 em
->len
= extent_end
- extent_start
;
5321 size
= btrfs_file_extent_inline_len(leaf
, item
);
5322 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5323 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5324 size
- extent_offset
);
5325 em
->start
= extent_start
+ extent_offset
;
5326 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5327 ~((u64
)root
->sectorsize
- 1);
5328 em
->orig_start
= EXTENT_MAP_INLINE
;
5329 if (compress_type
) {
5330 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5331 em
->compress_type
= compress_type
;
5333 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5334 if (create
== 0 && !PageUptodate(page
)) {
5335 if (btrfs_file_extent_compression(leaf
, item
) !=
5336 BTRFS_COMPRESS_NONE
) {
5337 ret
= uncompress_inline(path
, inode
, page
,
5339 extent_offset
, item
);
5340 BUG_ON(ret
); /* -ENOMEM */
5343 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5345 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5346 memset(map
+ pg_offset
+ copy_size
, 0,
5347 PAGE_CACHE_SIZE
- pg_offset
-
5352 flush_dcache_page(page
);
5353 } else if (create
&& PageUptodate(page
)) {
5357 free_extent_map(em
);
5360 btrfs_release_path(path
);
5361 trans
= btrfs_join_transaction(root
);
5364 return ERR_CAST(trans
);
5368 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5371 btrfs_mark_buffer_dirty(leaf
);
5373 set_extent_uptodate(io_tree
, em
->start
,
5374 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5377 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5384 em
->block_start
= EXTENT_MAP_HOLE
;
5385 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5387 btrfs_release_path(path
);
5388 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5389 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5390 "[%llu %llu]\n", (unsigned long long)em
->start
,
5391 (unsigned long long)em
->len
,
5392 (unsigned long long)start
,
5393 (unsigned long long)len
);
5399 write_lock(&em_tree
->lock
);
5400 ret
= add_extent_mapping(em_tree
, em
);
5401 /* it is possible that someone inserted the extent into the tree
5402 * while we had the lock dropped. It is also possible that
5403 * an overlapping map exists in the tree
5405 if (ret
== -EEXIST
) {
5406 struct extent_map
*existing
;
5410 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5411 if (existing
&& (existing
->start
> start
||
5412 existing
->start
+ existing
->len
<= start
)) {
5413 free_extent_map(existing
);
5417 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5420 err
= merge_extent_mapping(em_tree
, existing
,
5423 free_extent_map(existing
);
5425 free_extent_map(em
);
5430 free_extent_map(em
);
5434 free_extent_map(em
);
5439 write_unlock(&em_tree
->lock
);
5442 trace_btrfs_get_extent(root
, em
);
5445 btrfs_free_path(path
);
5447 ret
= btrfs_end_transaction(trans
, root
);
5452 free_extent_map(em
);
5453 return ERR_PTR(err
);
5455 BUG_ON(!em
); /* Error is always set */
5459 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5460 size_t pg_offset
, u64 start
, u64 len
,
5463 struct extent_map
*em
;
5464 struct extent_map
*hole_em
= NULL
;
5465 u64 range_start
= start
;
5471 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5476 * if our em maps to a hole, there might
5477 * actually be delalloc bytes behind it
5479 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5485 /* check to see if we've wrapped (len == -1 or similar) */
5494 /* ok, we didn't find anything, lets look for delalloc */
5495 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5496 end
, len
, EXTENT_DELALLOC
, 1);
5497 found_end
= range_start
+ found
;
5498 if (found_end
< range_start
)
5499 found_end
= (u64
)-1;
5502 * we didn't find anything useful, return
5503 * the original results from get_extent()
5505 if (range_start
> end
|| found_end
<= start
) {
5511 /* adjust the range_start to make sure it doesn't
5512 * go backwards from the start they passed in
5514 range_start
= max(start
,range_start
);
5515 found
= found_end
- range_start
;
5518 u64 hole_start
= start
;
5521 em
= alloc_extent_map();
5527 * when btrfs_get_extent can't find anything it
5528 * returns one huge hole
5530 * make sure what it found really fits our range, and
5531 * adjust to make sure it is based on the start from
5535 u64 calc_end
= extent_map_end(hole_em
);
5537 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5538 free_extent_map(hole_em
);
5541 hole_start
= max(hole_em
->start
, start
);
5542 hole_len
= calc_end
- hole_start
;
5546 if (hole_em
&& range_start
> hole_start
) {
5547 /* our hole starts before our delalloc, so we
5548 * have to return just the parts of the hole
5549 * that go until the delalloc starts
5551 em
->len
= min(hole_len
,
5552 range_start
- hole_start
);
5553 em
->start
= hole_start
;
5554 em
->orig_start
= hole_start
;
5556 * don't adjust block start at all,
5557 * it is fixed at EXTENT_MAP_HOLE
5559 em
->block_start
= hole_em
->block_start
;
5560 em
->block_len
= hole_len
;
5562 em
->start
= range_start
;
5564 em
->orig_start
= range_start
;
5565 em
->block_start
= EXTENT_MAP_DELALLOC
;
5566 em
->block_len
= found
;
5568 } else if (hole_em
) {
5573 free_extent_map(hole_em
);
5575 free_extent_map(em
);
5576 return ERR_PTR(err
);
5581 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5582 struct extent_map
*em
,
5585 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5586 struct btrfs_trans_handle
*trans
;
5587 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5588 struct btrfs_key ins
;
5591 bool insert
= false;
5594 * Ok if the extent map we looked up is a hole and is for the exact
5595 * range we want, there is no reason to allocate a new one, however if
5596 * it is not right then we need to free this one and drop the cache for
5599 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5601 free_extent_map(em
);
5604 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5607 trans
= btrfs_join_transaction(root
);
5609 return ERR_CAST(trans
);
5611 if (start
<= BTRFS_I(inode
)->disk_i_size
&& len
< 64 * 1024)
5612 btrfs_add_inode_defrag(trans
, inode
);
5614 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5616 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5617 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5618 alloc_hint
, &ins
, 1);
5625 em
= alloc_extent_map();
5627 em
= ERR_PTR(-ENOMEM
);
5633 em
->orig_start
= em
->start
;
5634 em
->len
= ins
.offset
;
5636 em
->block_start
= ins
.objectid
;
5637 em
->block_len
= ins
.offset
;
5638 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5641 * We need to do this because if we're using the original em we searched
5642 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5645 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5648 write_lock(&em_tree
->lock
);
5649 ret
= add_extent_mapping(em_tree
, em
);
5650 write_unlock(&em_tree
->lock
);
5653 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5656 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5657 ins
.offset
, ins
.offset
, 0);
5659 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5663 btrfs_end_transaction(trans
, root
);
5668 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5669 * block must be cow'd
5671 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5672 struct inode
*inode
, u64 offset
, u64 len
)
5674 struct btrfs_path
*path
;
5676 struct extent_buffer
*leaf
;
5677 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5678 struct btrfs_file_extent_item
*fi
;
5679 struct btrfs_key key
;
5687 path
= btrfs_alloc_path();
5691 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
5696 slot
= path
->slots
[0];
5699 /* can't find the item, must cow */
5706 leaf
= path
->nodes
[0];
5707 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5708 if (key
.objectid
!= btrfs_ino(inode
) ||
5709 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5710 /* not our file or wrong item type, must cow */
5714 if (key
.offset
> offset
) {
5715 /* Wrong offset, must cow */
5719 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5720 found_type
= btrfs_file_extent_type(leaf
, fi
);
5721 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5722 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5723 /* not a regular extent, must cow */
5726 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5727 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5729 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5730 if (extent_end
< offset
+ len
) {
5731 /* extent doesn't include our full range, must cow */
5735 if (btrfs_extent_readonly(root
, disk_bytenr
))
5739 * look for other files referencing this extent, if we
5740 * find any we must cow
5742 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
5743 key
.offset
- backref_offset
, disk_bytenr
))
5747 * adjust disk_bytenr and num_bytes to cover just the bytes
5748 * in this extent we are about to write. If there
5749 * are any csums in that range we have to cow in order
5750 * to keep the csums correct
5752 disk_bytenr
+= backref_offset
;
5753 disk_bytenr
+= offset
- key
.offset
;
5754 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5755 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5758 * all of the above have passed, it is safe to overwrite this extent
5763 btrfs_free_path(path
);
5767 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5768 struct buffer_head
*bh_result
, int create
)
5770 struct extent_map
*em
;
5771 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5772 u64 start
= iblock
<< inode
->i_blkbits
;
5773 u64 len
= bh_result
->b_size
;
5774 struct btrfs_trans_handle
*trans
;
5776 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5781 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5782 * io. INLINE is special, and we could probably kludge it in here, but
5783 * it's still buffered so for safety lets just fall back to the generic
5786 * For COMPRESSED we _have_ to read the entire extent in so we can
5787 * decompress it, so there will be buffering required no matter what we
5788 * do, so go ahead and fallback to buffered.
5790 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5791 * to buffered IO. Don't blame me, this is the price we pay for using
5794 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5795 em
->block_start
== EXTENT_MAP_INLINE
) {
5796 free_extent_map(em
);
5800 /* Just a good old fashioned hole, return */
5801 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5802 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5803 free_extent_map(em
);
5804 /* DIO will do one hole at a time, so just unlock a sector */
5805 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5806 start
+ root
->sectorsize
- 1);
5811 * We don't allocate a new extent in the following cases
5813 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5815 * 2) The extent is marked as PREALLOC. We're good to go here and can
5816 * just use the extent.
5820 len
= em
->len
- (start
- em
->start
);
5824 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5825 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5826 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5831 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5832 type
= BTRFS_ORDERED_PREALLOC
;
5834 type
= BTRFS_ORDERED_NOCOW
;
5835 len
= min(len
, em
->len
- (start
- em
->start
));
5836 block_start
= em
->block_start
+ (start
- em
->start
);
5839 * we're not going to log anything, but we do need
5840 * to make sure the current transaction stays open
5841 * while we look for nocow cross refs
5843 trans
= btrfs_join_transaction(root
);
5847 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5848 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5849 block_start
, len
, len
, type
);
5850 btrfs_end_transaction(trans
, root
);
5852 free_extent_map(em
);
5857 btrfs_end_transaction(trans
, root
);
5861 * this will cow the extent, reset the len in case we changed
5864 len
= bh_result
->b_size
;
5865 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
5868 len
= min(len
, em
->len
- (start
- em
->start
));
5870 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5871 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5874 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5876 bh_result
->b_size
= len
;
5877 bh_result
->b_bdev
= em
->bdev
;
5878 set_buffer_mapped(bh_result
);
5879 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5880 set_buffer_new(bh_result
);
5882 free_extent_map(em
);
5887 struct btrfs_dio_private
{
5888 struct inode
*inode
;
5895 /* number of bios pending for this dio */
5896 atomic_t pending_bios
;
5901 struct bio
*orig_bio
;
5904 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5906 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5907 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5908 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5909 struct inode
*inode
= dip
->inode
;
5910 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5912 u32
*private = dip
->csums
;
5914 start
= dip
->logical_offset
;
5916 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5917 struct page
*page
= bvec
->bv_page
;
5920 unsigned long flags
;
5922 local_irq_save(flags
);
5923 kaddr
= kmap_atomic(page
);
5924 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5925 csum
, bvec
->bv_len
);
5926 btrfs_csum_final(csum
, (char *)&csum
);
5927 kunmap_atomic(kaddr
);
5928 local_irq_restore(flags
);
5930 flush_dcache_page(bvec
->bv_page
);
5931 if (csum
!= *private) {
5932 printk(KERN_ERR
"btrfs csum failed ino %llu off"
5933 " %llu csum %u private %u\n",
5934 (unsigned long long)btrfs_ino(inode
),
5935 (unsigned long long)start
,
5941 start
+= bvec
->bv_len
;
5944 } while (bvec
<= bvec_end
);
5946 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5947 dip
->logical_offset
+ dip
->bytes
- 1);
5948 bio
->bi_private
= dip
->private;
5953 /* If we had a csum failure make sure to clear the uptodate flag */
5955 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5956 dio_end_io(bio
, err
);
5959 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5961 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5962 struct inode
*inode
= dip
->inode
;
5963 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5964 struct btrfs_ordered_extent
*ordered
= NULL
;
5965 u64 ordered_offset
= dip
->logical_offset
;
5966 u64 ordered_bytes
= dip
->bytes
;
5972 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5974 ordered_bytes
, !err
);
5978 ordered
->work
.func
= finish_ordered_fn
;
5979 ordered
->work
.flags
= 0;
5980 btrfs_queue_worker(&root
->fs_info
->endio_write_workers
,
5984 * our bio might span multiple ordered extents. If we haven't
5985 * completed the accounting for the whole dio, go back and try again
5987 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5988 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5994 bio
->bi_private
= dip
->private;
5998 /* If we had an error make sure to clear the uptodate flag */
6000 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
6001 dio_end_io(bio
, err
);
6004 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
6005 struct bio
*bio
, int mirror_num
,
6006 unsigned long bio_flags
, u64 offset
)
6009 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6010 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
6011 BUG_ON(ret
); /* -ENOMEM */
6015 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
6017 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6020 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
6021 "sector %#Lx len %u err no %d\n",
6022 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
6023 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
6027 * before atomic variable goto zero, we must make sure
6028 * dip->errors is perceived to be set.
6030 smp_mb__before_atomic_dec();
6033 /* if there are more bios still pending for this dio, just exit */
6034 if (!atomic_dec_and_test(&dip
->pending_bios
))
6038 bio_io_error(dip
->orig_bio
);
6040 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
6041 bio_endio(dip
->orig_bio
, 0);
6047 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
6048 u64 first_sector
, gfp_t gfp_flags
)
6050 int nr_vecs
= bio_get_nr_vecs(bdev
);
6051 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
6054 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
6055 int rw
, u64 file_offset
, int skip_sum
,
6056 u32
*csums
, int async_submit
)
6058 int write
= rw
& REQ_WRITE
;
6059 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6065 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
6073 if (write
&& async_submit
) {
6074 ret
= btrfs_wq_submit_bio(root
->fs_info
,
6075 inode
, rw
, bio
, 0, 0,
6077 __btrfs_submit_bio_start_direct_io
,
6078 __btrfs_submit_bio_done
);
6082 * If we aren't doing async submit, calculate the csum of the
6085 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
6088 } else if (!skip_sum
) {
6089 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
6090 file_offset
, csums
);
6096 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
6102 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
6105 struct inode
*inode
= dip
->inode
;
6106 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6107 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
6109 struct bio
*orig_bio
= dip
->orig_bio
;
6110 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
6111 u64 start_sector
= orig_bio
->bi_sector
;
6112 u64 file_offset
= dip
->logical_offset
;
6116 u32
*csums
= dip
->csums
;
6118 int async_submit
= 0;
6119 int write
= rw
& REQ_WRITE
;
6121 map_length
= orig_bio
->bi_size
;
6122 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6123 &map_length
, NULL
, 0);
6129 if (map_length
>= orig_bio
->bi_size
) {
6135 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
6138 bio
->bi_private
= dip
;
6139 bio
->bi_end_io
= btrfs_end_dio_bio
;
6140 atomic_inc(&dip
->pending_bios
);
6142 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
6143 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
6144 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
6145 bvec
->bv_offset
) < bvec
->bv_len
)) {
6147 * inc the count before we submit the bio so
6148 * we know the end IO handler won't happen before
6149 * we inc the count. Otherwise, the dip might get freed
6150 * before we're done setting it up
6152 atomic_inc(&dip
->pending_bios
);
6153 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
6154 file_offset
, skip_sum
,
6155 csums
, async_submit
);
6158 atomic_dec(&dip
->pending_bios
);
6162 /* Write's use the ordered csums */
6163 if (!write
&& !skip_sum
)
6164 csums
= csums
+ nr_pages
;
6165 start_sector
+= submit_len
>> 9;
6166 file_offset
+= submit_len
;
6171 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6172 start_sector
, GFP_NOFS
);
6175 bio
->bi_private
= dip
;
6176 bio
->bi_end_io
= btrfs_end_dio_bio
;
6178 map_length
= orig_bio
->bi_size
;
6179 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6180 &map_length
, NULL
, 0);
6186 submit_len
+= bvec
->bv_len
;
6193 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6194 csums
, async_submit
);
6202 * before atomic variable goto zero, we must
6203 * make sure dip->errors is perceived to be set.
6205 smp_mb__before_atomic_dec();
6206 if (atomic_dec_and_test(&dip
->pending_bios
))
6207 bio_io_error(dip
->orig_bio
);
6209 /* bio_end_io() will handle error, so we needn't return it */
6213 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6216 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6217 struct btrfs_dio_private
*dip
;
6218 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6220 int write
= rw
& REQ_WRITE
;
6223 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6225 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6232 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6233 if (!write
&& !skip_sum
) {
6234 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6242 dip
->private = bio
->bi_private
;
6244 dip
->logical_offset
= file_offset
;
6248 dip
->bytes
+= bvec
->bv_len
;
6250 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6252 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6253 bio
->bi_private
= dip
;
6255 dip
->orig_bio
= bio
;
6256 atomic_set(&dip
->pending_bios
, 0);
6259 bio
->bi_end_io
= btrfs_endio_direct_write
;
6261 bio
->bi_end_io
= btrfs_endio_direct_read
;
6263 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6268 * If this is a write, we need to clean up the reserved space and kill
6269 * the ordered extent.
6272 struct btrfs_ordered_extent
*ordered
;
6273 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6274 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6275 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6276 btrfs_free_reserved_extent(root
, ordered
->start
,
6278 btrfs_put_ordered_extent(ordered
);
6279 btrfs_put_ordered_extent(ordered
);
6281 bio_endio(bio
, ret
);
6284 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6285 const struct iovec
*iov
, loff_t offset
,
6286 unsigned long nr_segs
)
6292 unsigned blocksize_mask
= root
->sectorsize
- 1;
6293 ssize_t retval
= -EINVAL
;
6294 loff_t end
= offset
;
6296 if (offset
& blocksize_mask
)
6299 /* Check the memory alignment. Blocks cannot straddle pages */
6300 for (seg
= 0; seg
< nr_segs
; seg
++) {
6301 addr
= (unsigned long)iov
[seg
].iov_base
;
6302 size
= iov
[seg
].iov_len
;
6304 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6307 /* If this is a write we don't need to check anymore */
6312 * Check to make sure we don't have duplicate iov_base's in this
6313 * iovec, if so return EINVAL, otherwise we'll get csum errors
6314 * when reading back.
6316 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6317 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6325 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6326 const struct iovec
*iov
, loff_t offset
,
6327 unsigned long nr_segs
)
6329 struct file
*file
= iocb
->ki_filp
;
6330 struct inode
*inode
= file
->f_mapping
->host
;
6331 struct btrfs_ordered_extent
*ordered
;
6332 struct extent_state
*cached_state
= NULL
;
6333 u64 lockstart
, lockend
;
6335 int writing
= rw
& WRITE
;
6337 size_t count
= iov_length(iov
, nr_segs
);
6339 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6345 lockend
= offset
+ count
- 1;
6348 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6354 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6357 * We're concerned with the entire range that we're going to be
6358 * doing DIO to, so we need to make sure theres no ordered
6359 * extents in this range.
6361 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6362 lockend
- lockstart
+ 1);
6365 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6366 &cached_state
, GFP_NOFS
);
6367 btrfs_start_ordered_extent(inode
, ordered
, 1);
6368 btrfs_put_ordered_extent(ordered
);
6373 * we don't use btrfs_set_extent_delalloc because we don't want
6374 * the dirty or uptodate bits
6377 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6378 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6379 EXTENT_DELALLOC
, NULL
, &cached_state
,
6382 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6383 lockend
, EXTENT_LOCKED
| write_bits
,
6384 1, 0, &cached_state
, GFP_NOFS
);
6389 free_extent_state(cached_state
);
6390 cached_state
= NULL
;
6392 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6393 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6394 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6395 btrfs_submit_direct
, 0);
6397 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6398 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6399 offset
+ iov_length(iov
, nr_segs
) - 1,
6400 EXTENT_LOCKED
| write_bits
, 1, 0,
6401 &cached_state
, GFP_NOFS
);
6402 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6404 * We're falling back to buffered, unlock the section we didn't
6407 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6408 offset
+ iov_length(iov
, nr_segs
) - 1,
6409 EXTENT_LOCKED
| write_bits
, 1, 0,
6410 &cached_state
, GFP_NOFS
);
6413 free_extent_state(cached_state
);
6417 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6418 __u64 start
, __u64 len
)
6420 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6423 int btrfs_readpage(struct file
*file
, struct page
*page
)
6425 struct extent_io_tree
*tree
;
6426 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6427 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
6430 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6432 struct extent_io_tree
*tree
;
6435 if (current
->flags
& PF_MEMALLOC
) {
6436 redirty_page_for_writepage(wbc
, page
);
6440 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6441 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6444 int btrfs_writepages(struct address_space
*mapping
,
6445 struct writeback_control
*wbc
)
6447 struct extent_io_tree
*tree
;
6449 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6450 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6454 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6455 struct list_head
*pages
, unsigned nr_pages
)
6457 struct extent_io_tree
*tree
;
6458 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6459 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6462 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6464 struct extent_io_tree
*tree
;
6465 struct extent_map_tree
*map
;
6468 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6469 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6470 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6472 ClearPagePrivate(page
);
6473 set_page_private(page
, 0);
6474 page_cache_release(page
);
6479 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6481 if (PageWriteback(page
) || PageDirty(page
))
6483 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6486 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6488 struct inode
*inode
= page
->mapping
->host
;
6489 struct extent_io_tree
*tree
;
6490 struct btrfs_ordered_extent
*ordered
;
6491 struct extent_state
*cached_state
= NULL
;
6492 u64 page_start
= page_offset(page
);
6493 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6496 * we have the page locked, so new writeback can't start,
6497 * and the dirty bit won't be cleared while we are here.
6499 * Wait for IO on this page so that we can safely clear
6500 * the PagePrivate2 bit and do ordered accounting
6502 wait_on_page_writeback(page
);
6504 tree
= &BTRFS_I(inode
)->io_tree
;
6506 btrfs_releasepage(page
, GFP_NOFS
);
6509 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
6510 ordered
= btrfs_lookup_ordered_extent(inode
,
6514 * IO on this page will never be started, so we need
6515 * to account for any ordered extents now
6517 clear_extent_bit(tree
, page_start
, page_end
,
6518 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6519 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6520 &cached_state
, GFP_NOFS
);
6522 * whoever cleared the private bit is responsible
6523 * for the finish_ordered_io
6525 if (TestClearPagePrivate2(page
) &&
6526 btrfs_dec_test_ordered_pending(inode
, &ordered
, page_start
,
6527 PAGE_CACHE_SIZE
, 1)) {
6528 btrfs_finish_ordered_io(ordered
);
6530 btrfs_put_ordered_extent(ordered
);
6531 cached_state
= NULL
;
6532 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
6534 clear_extent_bit(tree
, page_start
, page_end
,
6535 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6536 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6537 __btrfs_releasepage(page
, GFP_NOFS
);
6539 ClearPageChecked(page
);
6540 if (PagePrivate(page
)) {
6541 ClearPagePrivate(page
);
6542 set_page_private(page
, 0);
6543 page_cache_release(page
);
6548 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6549 * called from a page fault handler when a page is first dirtied. Hence we must
6550 * be careful to check for EOF conditions here. We set the page up correctly
6551 * for a written page which means we get ENOSPC checking when writing into
6552 * holes and correct delalloc and unwritten extent mapping on filesystems that
6553 * support these features.
6555 * We are not allowed to take the i_mutex here so we have to play games to
6556 * protect against truncate races as the page could now be beyond EOF. Because
6557 * vmtruncate() writes the inode size before removing pages, once we have the
6558 * page lock we can determine safely if the page is beyond EOF. If it is not
6559 * beyond EOF, then the page is guaranteed safe against truncation until we
6562 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6564 struct page
*page
= vmf
->page
;
6565 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6566 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6567 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6568 struct btrfs_ordered_extent
*ordered
;
6569 struct extent_state
*cached_state
= NULL
;
6571 unsigned long zero_start
;
6578 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6580 ret
= file_update_time(vma
->vm_file
);
6586 else /* -ENOSPC, -EIO, etc */
6587 ret
= VM_FAULT_SIGBUS
;
6593 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6596 size
= i_size_read(inode
);
6597 page_start
= page_offset(page
);
6598 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6600 if ((page
->mapping
!= inode
->i_mapping
) ||
6601 (page_start
>= size
)) {
6602 /* page got truncated out from underneath us */
6605 wait_on_page_writeback(page
);
6607 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
6608 set_page_extent_mapped(page
);
6611 * we can't set the delalloc bits if there are pending ordered
6612 * extents. Drop our locks and wait for them to finish
6614 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6616 unlock_extent_cached(io_tree
, page_start
, page_end
,
6617 &cached_state
, GFP_NOFS
);
6619 btrfs_start_ordered_extent(inode
, ordered
, 1);
6620 btrfs_put_ordered_extent(ordered
);
6625 * XXX - page_mkwrite gets called every time the page is dirtied, even
6626 * if it was already dirty, so for space accounting reasons we need to
6627 * clear any delalloc bits for the range we are fixing to save. There
6628 * is probably a better way to do this, but for now keep consistent with
6629 * prepare_pages in the normal write path.
6631 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6632 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6633 0, 0, &cached_state
, GFP_NOFS
);
6635 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6638 unlock_extent_cached(io_tree
, page_start
, page_end
,
6639 &cached_state
, GFP_NOFS
);
6640 ret
= VM_FAULT_SIGBUS
;
6645 /* page is wholly or partially inside EOF */
6646 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6647 zero_start
= size
& ~PAGE_CACHE_MASK
;
6649 zero_start
= PAGE_CACHE_SIZE
;
6651 if (zero_start
!= PAGE_CACHE_SIZE
) {
6653 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6654 flush_dcache_page(page
);
6657 ClearPageChecked(page
);
6658 set_page_dirty(page
);
6659 SetPageUptodate(page
);
6661 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6662 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6664 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6668 return VM_FAULT_LOCKED
;
6671 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6676 static int btrfs_truncate(struct inode
*inode
)
6678 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6679 struct btrfs_block_rsv
*rsv
;
6682 struct btrfs_trans_handle
*trans
;
6684 u64 mask
= root
->sectorsize
- 1;
6685 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
6687 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6691 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6692 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6695 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6696 * 3 things going on here
6698 * 1) We need to reserve space for our orphan item and the space to
6699 * delete our orphan item. Lord knows we don't want to have a dangling
6700 * orphan item because we didn't reserve space to remove it.
6702 * 2) We need to reserve space to update our inode.
6704 * 3) We need to have something to cache all the space that is going to
6705 * be free'd up by the truncate operation, but also have some slack
6706 * space reserved in case it uses space during the truncate (thank you
6707 * very much snapshotting).
6709 * And we need these to all be seperate. The fact is we can use alot of
6710 * space doing the truncate, and we have no earthly idea how much space
6711 * we will use, so we need the truncate reservation to be seperate so it
6712 * doesn't end up using space reserved for updating the inode or
6713 * removing the orphan item. We also need to be able to stop the
6714 * transaction and start a new one, which means we need to be able to
6715 * update the inode several times, and we have no idea of knowing how
6716 * many times that will be, so we can't just reserve 1 item for the
6717 * entirety of the opration, so that has to be done seperately as well.
6718 * Then there is the orphan item, which does indeed need to be held on
6719 * to for the whole operation, and we need nobody to touch this reserved
6720 * space except the orphan code.
6722 * So that leaves us with
6724 * 1) root->orphan_block_rsv - for the orphan deletion.
6725 * 2) rsv - for the truncate reservation, which we will steal from the
6726 * transaction reservation.
6727 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6728 * updating the inode.
6730 rsv
= btrfs_alloc_block_rsv(root
);
6733 rsv
->size
= min_size
;
6736 * 1 for the truncate slack space
6737 * 1 for the orphan item we're going to add
6738 * 1 for the orphan item deletion
6739 * 1 for updating the inode.
6741 trans
= btrfs_start_transaction(root
, 4);
6742 if (IS_ERR(trans
)) {
6743 err
= PTR_ERR(trans
);
6747 /* Migrate the slack space for the truncate to our reserve */
6748 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
6752 ret
= btrfs_orphan_add(trans
, inode
);
6754 btrfs_end_transaction(trans
, root
);
6759 * setattr is responsible for setting the ordered_data_close flag,
6760 * but that is only tested during the last file release. That
6761 * could happen well after the next commit, leaving a great big
6762 * window where new writes may get lost if someone chooses to write
6763 * to this file after truncating to zero
6765 * The inode doesn't have any dirty data here, and so if we commit
6766 * this is a noop. If someone immediately starts writing to the inode
6767 * it is very likely we'll catch some of their writes in this
6768 * transaction, and the commit will find this file on the ordered
6769 * data list with good things to send down.
6771 * This is a best effort solution, there is still a window where
6772 * using truncate to replace the contents of the file will
6773 * end up with a zero length file after a crash.
6775 if (inode
->i_size
== 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
6776 &BTRFS_I(inode
)->runtime_flags
))
6777 btrfs_add_ordered_operation(trans
, root
, inode
);
6780 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
);
6783 * This can only happen with the original transaction we
6784 * started above, every other time we shouldn't have a
6785 * transaction started yet.
6794 /* Just need the 1 for updating the inode */
6795 trans
= btrfs_start_transaction(root
, 1);
6796 if (IS_ERR(trans
)) {
6797 ret
= err
= PTR_ERR(trans
);
6803 trans
->block_rsv
= rsv
;
6805 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6807 BTRFS_EXTENT_DATA_KEY
);
6808 if (ret
!= -EAGAIN
) {
6813 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6814 ret
= btrfs_update_inode(trans
, root
, inode
);
6820 nr
= trans
->blocks_used
;
6821 btrfs_end_transaction(trans
, root
);
6823 btrfs_btree_balance_dirty(root
, nr
);
6826 if (ret
== 0 && inode
->i_nlink
> 0) {
6827 trans
->block_rsv
= root
->orphan_block_rsv
;
6828 ret
= btrfs_orphan_del(trans
, inode
);
6831 } else if (ret
&& inode
->i_nlink
> 0) {
6833 * Failed to do the truncate, remove us from the in memory
6836 ret
= btrfs_orphan_del(NULL
, inode
);
6840 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6841 ret
= btrfs_update_inode(trans
, root
, inode
);
6845 nr
= trans
->blocks_used
;
6846 ret
= btrfs_end_transaction(trans
, root
);
6847 btrfs_btree_balance_dirty(root
, nr
);
6851 btrfs_free_block_rsv(root
, rsv
);
6860 * create a new subvolume directory/inode (helper for the ioctl).
6862 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6863 struct btrfs_root
*new_root
, u64 new_dirid
)
6865 struct inode
*inode
;
6869 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
6870 new_dirid
, new_dirid
,
6871 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
6874 return PTR_ERR(inode
);
6875 inode
->i_op
= &btrfs_dir_inode_operations
;
6876 inode
->i_fop
= &btrfs_dir_file_operations
;
6878 set_nlink(inode
, 1);
6879 btrfs_i_size_write(inode
, 0);
6881 err
= btrfs_update_inode(trans
, new_root
, inode
);
6887 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6889 struct btrfs_inode
*ei
;
6890 struct inode
*inode
;
6892 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6897 ei
->space_info
= NULL
;
6900 ei
->last_sub_trans
= 0;
6901 ei
->logged_trans
= 0;
6902 ei
->delalloc_bytes
= 0;
6903 ei
->disk_i_size
= 0;
6906 ei
->index_cnt
= (u64
)-1;
6907 ei
->last_unlink_trans
= 0;
6909 spin_lock_init(&ei
->lock
);
6910 ei
->outstanding_extents
= 0;
6911 ei
->reserved_extents
= 0;
6913 ei
->runtime_flags
= 0;
6914 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6916 ei
->delayed_node
= NULL
;
6918 inode
= &ei
->vfs_inode
;
6919 extent_map_tree_init(&ei
->extent_tree
);
6920 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
6921 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
6922 ei
->io_tree
.track_uptodate
= 1;
6923 ei
->io_failure_tree
.track_uptodate
= 1;
6924 mutex_init(&ei
->log_mutex
);
6925 mutex_init(&ei
->delalloc_mutex
);
6926 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6927 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6928 INIT_LIST_HEAD(&ei
->ordered_operations
);
6929 RB_CLEAR_NODE(&ei
->rb_node
);
6934 static void btrfs_i_callback(struct rcu_head
*head
)
6936 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6937 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6940 void btrfs_destroy_inode(struct inode
*inode
)
6942 struct btrfs_ordered_extent
*ordered
;
6943 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6945 WARN_ON(!list_empty(&inode
->i_dentry
));
6946 WARN_ON(inode
->i_data
.nrpages
);
6947 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
6948 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
6949 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
6950 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
6953 * This can happen where we create an inode, but somebody else also
6954 * created the same inode and we need to destroy the one we already
6961 * Make sure we're properly removed from the ordered operation
6965 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6966 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6967 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6968 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6971 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
6972 &BTRFS_I(inode
)->runtime_flags
)) {
6973 printk(KERN_INFO
"BTRFS: inode %llu still on the orphan list\n",
6974 (unsigned long long)btrfs_ino(inode
));
6975 atomic_dec(&root
->orphan_inodes
);
6979 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6983 printk(KERN_ERR
"btrfs found ordered "
6984 "extent %llu %llu on inode cleanup\n",
6985 (unsigned long long)ordered
->file_offset
,
6986 (unsigned long long)ordered
->len
);
6987 btrfs_remove_ordered_extent(inode
, ordered
);
6988 btrfs_put_ordered_extent(ordered
);
6989 btrfs_put_ordered_extent(ordered
);
6992 inode_tree_del(inode
);
6993 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6995 btrfs_remove_delayed_node(inode
);
6996 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6999 int btrfs_drop_inode(struct inode
*inode
)
7001 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7003 if (btrfs_root_refs(&root
->root_item
) == 0 &&
7004 !btrfs_is_free_space_inode(root
, inode
))
7007 return generic_drop_inode(inode
);
7010 static void init_once(void *foo
)
7012 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
7014 inode_init_once(&ei
->vfs_inode
);
7017 void btrfs_destroy_cachep(void)
7019 if (btrfs_inode_cachep
)
7020 kmem_cache_destroy(btrfs_inode_cachep
);
7021 if (btrfs_trans_handle_cachep
)
7022 kmem_cache_destroy(btrfs_trans_handle_cachep
);
7023 if (btrfs_transaction_cachep
)
7024 kmem_cache_destroy(btrfs_transaction_cachep
);
7025 if (btrfs_path_cachep
)
7026 kmem_cache_destroy(btrfs_path_cachep
);
7027 if (btrfs_free_space_cachep
)
7028 kmem_cache_destroy(btrfs_free_space_cachep
);
7031 int btrfs_init_cachep(void)
7033 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
7034 sizeof(struct btrfs_inode
), 0,
7035 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
7036 if (!btrfs_inode_cachep
)
7039 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
7040 sizeof(struct btrfs_trans_handle
), 0,
7041 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7042 if (!btrfs_trans_handle_cachep
)
7045 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
7046 sizeof(struct btrfs_transaction
), 0,
7047 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7048 if (!btrfs_transaction_cachep
)
7051 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
7052 sizeof(struct btrfs_path
), 0,
7053 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7054 if (!btrfs_path_cachep
)
7057 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
7058 sizeof(struct btrfs_free_space
), 0,
7059 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7060 if (!btrfs_free_space_cachep
)
7065 btrfs_destroy_cachep();
7069 static int btrfs_getattr(struct vfsmount
*mnt
,
7070 struct dentry
*dentry
, struct kstat
*stat
)
7072 struct inode
*inode
= dentry
->d_inode
;
7073 u32 blocksize
= inode
->i_sb
->s_blocksize
;
7075 generic_fillattr(inode
, stat
);
7076 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
7077 stat
->blksize
= PAGE_CACHE_SIZE
;
7078 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
7079 ALIGN(BTRFS_I(inode
)->delalloc_bytes
, blocksize
)) >> 9;
7084 * If a file is moved, it will inherit the cow and compression flags of the new
7087 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
7089 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
7090 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
7092 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
7093 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
7095 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
7097 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
) {
7098 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
7099 b_inode
->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
7101 b_inode
->flags
&= ~(BTRFS_INODE_COMPRESS
|
7102 BTRFS_INODE_NOCOMPRESS
);
7106 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
7107 struct inode
*new_dir
, struct dentry
*new_dentry
)
7109 struct btrfs_trans_handle
*trans
;
7110 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
7111 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
7112 struct inode
*new_inode
= new_dentry
->d_inode
;
7113 struct inode
*old_inode
= old_dentry
->d_inode
;
7114 struct timespec ctime
= CURRENT_TIME
;
7118 u64 old_ino
= btrfs_ino(old_inode
);
7120 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
7123 /* we only allow rename subvolume link between subvolumes */
7124 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
7127 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
7128 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
7131 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
7132 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
7135 * we're using rename to replace one file with another.
7136 * and the replacement file is large. Start IO on it now so
7137 * we don't add too much work to the end of the transaction
7139 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
7140 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
7141 filemap_flush(old_inode
->i_mapping
);
7143 /* close the racy window with snapshot create/destroy ioctl */
7144 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7145 down_read(&root
->fs_info
->subvol_sem
);
7147 * We want to reserve the absolute worst case amount of items. So if
7148 * both inodes are subvols and we need to unlink them then that would
7149 * require 4 item modifications, but if they are both normal inodes it
7150 * would require 5 item modifications, so we'll assume their normal
7151 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7152 * should cover the worst case number of items we'll modify.
7154 trans
= btrfs_start_transaction(root
, 20);
7155 if (IS_ERR(trans
)) {
7156 ret
= PTR_ERR(trans
);
7161 btrfs_record_root_in_trans(trans
, dest
);
7163 ret
= btrfs_set_inode_index(new_dir
, &index
);
7167 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7168 /* force full log commit if subvolume involved. */
7169 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7171 ret
= btrfs_insert_inode_ref(trans
, dest
,
7172 new_dentry
->d_name
.name
,
7173 new_dentry
->d_name
.len
,
7175 btrfs_ino(new_dir
), index
);
7179 * this is an ugly little race, but the rename is required
7180 * to make sure that if we crash, the inode is either at the
7181 * old name or the new one. pinning the log transaction lets
7182 * us make sure we don't allow a log commit to come in after
7183 * we unlink the name but before we add the new name back in.
7185 btrfs_pin_log_trans(root
);
7188 * make sure the inode gets flushed if it is replacing
7191 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
7192 btrfs_add_ordered_operation(trans
, root
, old_inode
);
7194 inode_inc_iversion(old_dir
);
7195 inode_inc_iversion(new_dir
);
7196 inode_inc_iversion(old_inode
);
7197 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
7198 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
7199 old_inode
->i_ctime
= ctime
;
7201 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
7202 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
7204 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7205 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
7206 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
7207 old_dentry
->d_name
.name
,
7208 old_dentry
->d_name
.len
);
7210 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7211 old_dentry
->d_inode
,
7212 old_dentry
->d_name
.name
,
7213 old_dentry
->d_name
.len
);
7215 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7218 btrfs_abort_transaction(trans
, root
, ret
);
7223 inode_inc_iversion(new_inode
);
7224 new_inode
->i_ctime
= CURRENT_TIME
;
7225 if (unlikely(btrfs_ino(new_inode
) ==
7226 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7227 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7228 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7230 new_dentry
->d_name
.name
,
7231 new_dentry
->d_name
.len
);
7232 BUG_ON(new_inode
->i_nlink
== 0);
7234 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7235 new_dentry
->d_inode
,
7236 new_dentry
->d_name
.name
,
7237 new_dentry
->d_name
.len
);
7239 if (!ret
&& new_inode
->i_nlink
== 0) {
7240 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7244 btrfs_abort_transaction(trans
, root
, ret
);
7249 fixup_inode_flags(new_dir
, old_inode
);
7251 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7252 new_dentry
->d_name
.name
,
7253 new_dentry
->d_name
.len
, 0, index
);
7255 btrfs_abort_transaction(trans
, root
, ret
);
7259 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7260 struct dentry
*parent
= new_dentry
->d_parent
;
7261 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7262 btrfs_end_log_trans(root
);
7265 btrfs_end_transaction(trans
, root
);
7267 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7268 up_read(&root
->fs_info
->subvol_sem
);
7274 * some fairly slow code that needs optimization. This walks the list
7275 * of all the inodes with pending delalloc and forces them to disk.
7277 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7279 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7280 struct btrfs_inode
*binode
;
7281 struct inode
*inode
;
7283 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7286 spin_lock(&root
->fs_info
->delalloc_lock
);
7287 while (!list_empty(head
)) {
7288 binode
= list_entry(head
->next
, struct btrfs_inode
,
7290 inode
= igrab(&binode
->vfs_inode
);
7292 list_del_init(&binode
->delalloc_inodes
);
7293 spin_unlock(&root
->fs_info
->delalloc_lock
);
7295 filemap_flush(inode
->i_mapping
);
7297 btrfs_add_delayed_iput(inode
);
7302 spin_lock(&root
->fs_info
->delalloc_lock
);
7304 spin_unlock(&root
->fs_info
->delalloc_lock
);
7306 /* the filemap_flush will queue IO into the worker threads, but
7307 * we have to make sure the IO is actually started and that
7308 * ordered extents get created before we return
7310 atomic_inc(&root
->fs_info
->async_submit_draining
);
7311 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7312 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7313 wait_event(root
->fs_info
->async_submit_wait
,
7314 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7315 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7317 atomic_dec(&root
->fs_info
->async_submit_draining
);
7321 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7322 const char *symname
)
7324 struct btrfs_trans_handle
*trans
;
7325 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7326 struct btrfs_path
*path
;
7327 struct btrfs_key key
;
7328 struct inode
*inode
= NULL
;
7336 struct btrfs_file_extent_item
*ei
;
7337 struct extent_buffer
*leaf
;
7338 unsigned long nr
= 0;
7340 name_len
= strlen(symname
) + 1;
7341 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7342 return -ENAMETOOLONG
;
7345 * 2 items for inode item and ref
7346 * 2 items for dir items
7347 * 1 item for xattr if selinux is on
7349 trans
= btrfs_start_transaction(root
, 5);
7351 return PTR_ERR(trans
);
7353 err
= btrfs_find_free_ino(root
, &objectid
);
7357 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7358 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
7359 S_IFLNK
|S_IRWXUGO
, &index
);
7360 if (IS_ERR(inode
)) {
7361 err
= PTR_ERR(inode
);
7365 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7372 * If the active LSM wants to access the inode during
7373 * d_instantiate it needs these. Smack checks to see
7374 * if the filesystem supports xattrs by looking at the
7377 inode
->i_fop
= &btrfs_file_operations
;
7378 inode
->i_op
= &btrfs_file_inode_operations
;
7380 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7384 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7385 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7386 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7391 path
= btrfs_alloc_path();
7397 key
.objectid
= btrfs_ino(inode
);
7399 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7400 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7401 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7405 btrfs_free_path(path
);
7408 leaf
= path
->nodes
[0];
7409 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7410 struct btrfs_file_extent_item
);
7411 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7412 btrfs_set_file_extent_type(leaf
, ei
,
7413 BTRFS_FILE_EXTENT_INLINE
);
7414 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7415 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7416 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7417 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7419 ptr
= btrfs_file_extent_inline_start(ei
);
7420 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7421 btrfs_mark_buffer_dirty(leaf
);
7422 btrfs_free_path(path
);
7424 inode
->i_op
= &btrfs_symlink_inode_operations
;
7425 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7426 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7427 inode_set_bytes(inode
, name_len
);
7428 btrfs_i_size_write(inode
, name_len
- 1);
7429 err
= btrfs_update_inode(trans
, root
, inode
);
7435 d_instantiate(dentry
, inode
);
7436 nr
= trans
->blocks_used
;
7437 btrfs_end_transaction(trans
, root
);
7439 inode_dec_link_count(inode
);
7442 btrfs_btree_balance_dirty(root
, nr
);
7446 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7447 u64 start
, u64 num_bytes
, u64 min_size
,
7448 loff_t actual_len
, u64
*alloc_hint
,
7449 struct btrfs_trans_handle
*trans
)
7451 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7452 struct btrfs_key ins
;
7453 u64 cur_offset
= start
;
7456 bool own_trans
= true;
7460 while (num_bytes
> 0) {
7462 trans
= btrfs_start_transaction(root
, 3);
7463 if (IS_ERR(trans
)) {
7464 ret
= PTR_ERR(trans
);
7469 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7470 0, *alloc_hint
, &ins
, 1);
7473 btrfs_end_transaction(trans
, root
);
7477 ret
= insert_reserved_file_extent(trans
, inode
,
7478 cur_offset
, ins
.objectid
,
7479 ins
.offset
, ins
.offset
,
7480 ins
.offset
, 0, 0, 0,
7481 BTRFS_FILE_EXTENT_PREALLOC
);
7483 btrfs_abort_transaction(trans
, root
, ret
);
7485 btrfs_end_transaction(trans
, root
);
7488 btrfs_drop_extent_cache(inode
, cur_offset
,
7489 cur_offset
+ ins
.offset
-1, 0);
7491 num_bytes
-= ins
.offset
;
7492 cur_offset
+= ins
.offset
;
7493 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7495 inode_inc_iversion(inode
);
7496 inode
->i_ctime
= CURRENT_TIME
;
7497 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7498 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7499 (actual_len
> inode
->i_size
) &&
7500 (cur_offset
> inode
->i_size
)) {
7501 if (cur_offset
> actual_len
)
7502 i_size
= actual_len
;
7504 i_size
= cur_offset
;
7505 i_size_write(inode
, i_size
);
7506 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7509 ret
= btrfs_update_inode(trans
, root
, inode
);
7512 btrfs_abort_transaction(trans
, root
, ret
);
7514 btrfs_end_transaction(trans
, root
);
7519 btrfs_end_transaction(trans
, root
);
7524 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7525 u64 start
, u64 num_bytes
, u64 min_size
,
7526 loff_t actual_len
, u64
*alloc_hint
)
7528 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7529 min_size
, actual_len
, alloc_hint
,
7533 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7534 struct btrfs_trans_handle
*trans
, int mode
,
7535 u64 start
, u64 num_bytes
, u64 min_size
,
7536 loff_t actual_len
, u64
*alloc_hint
)
7538 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7539 min_size
, actual_len
, alloc_hint
, trans
);
7542 static int btrfs_set_page_dirty(struct page
*page
)
7544 return __set_page_dirty_nobuffers(page
);
7547 static int btrfs_permission(struct inode
*inode
, int mask
)
7549 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7550 umode_t mode
= inode
->i_mode
;
7552 if (mask
& MAY_WRITE
&&
7553 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
7554 if (btrfs_root_readonly(root
))
7556 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
7559 return generic_permission(inode
, mask
);
7562 static const struct inode_operations btrfs_dir_inode_operations
= {
7563 .getattr
= btrfs_getattr
,
7564 .lookup
= btrfs_lookup
,
7565 .create
= btrfs_create
,
7566 .unlink
= btrfs_unlink
,
7568 .mkdir
= btrfs_mkdir
,
7569 .rmdir
= btrfs_rmdir
,
7570 .rename
= btrfs_rename
,
7571 .symlink
= btrfs_symlink
,
7572 .setattr
= btrfs_setattr
,
7573 .mknod
= btrfs_mknod
,
7574 .setxattr
= btrfs_setxattr
,
7575 .getxattr
= btrfs_getxattr
,
7576 .listxattr
= btrfs_listxattr
,
7577 .removexattr
= btrfs_removexattr
,
7578 .permission
= btrfs_permission
,
7579 .get_acl
= btrfs_get_acl
,
7581 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7582 .lookup
= btrfs_lookup
,
7583 .permission
= btrfs_permission
,
7584 .get_acl
= btrfs_get_acl
,
7587 static const struct file_operations btrfs_dir_file_operations
= {
7588 .llseek
= generic_file_llseek
,
7589 .read
= generic_read_dir
,
7590 .readdir
= btrfs_real_readdir
,
7591 .unlocked_ioctl
= btrfs_ioctl
,
7592 #ifdef CONFIG_COMPAT
7593 .compat_ioctl
= btrfs_ioctl
,
7595 .release
= btrfs_release_file
,
7596 .fsync
= btrfs_sync_file
,
7599 static struct extent_io_ops btrfs_extent_io_ops
= {
7600 .fill_delalloc
= run_delalloc_range
,
7601 .submit_bio_hook
= btrfs_submit_bio_hook
,
7602 .merge_bio_hook
= btrfs_merge_bio_hook
,
7603 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7604 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7605 .writepage_start_hook
= btrfs_writepage_start_hook
,
7606 .set_bit_hook
= btrfs_set_bit_hook
,
7607 .clear_bit_hook
= btrfs_clear_bit_hook
,
7608 .merge_extent_hook
= btrfs_merge_extent_hook
,
7609 .split_extent_hook
= btrfs_split_extent_hook
,
7613 * btrfs doesn't support the bmap operation because swapfiles
7614 * use bmap to make a mapping of extents in the file. They assume
7615 * these extents won't change over the life of the file and they
7616 * use the bmap result to do IO directly to the drive.
7618 * the btrfs bmap call would return logical addresses that aren't
7619 * suitable for IO and they also will change frequently as COW
7620 * operations happen. So, swapfile + btrfs == corruption.
7622 * For now we're avoiding this by dropping bmap.
7624 static const struct address_space_operations btrfs_aops
= {
7625 .readpage
= btrfs_readpage
,
7626 .writepage
= btrfs_writepage
,
7627 .writepages
= btrfs_writepages
,
7628 .readpages
= btrfs_readpages
,
7629 .direct_IO
= btrfs_direct_IO
,
7630 .invalidatepage
= btrfs_invalidatepage
,
7631 .releasepage
= btrfs_releasepage
,
7632 .set_page_dirty
= btrfs_set_page_dirty
,
7633 .error_remove_page
= generic_error_remove_page
,
7636 static const struct address_space_operations btrfs_symlink_aops
= {
7637 .readpage
= btrfs_readpage
,
7638 .writepage
= btrfs_writepage
,
7639 .invalidatepage
= btrfs_invalidatepage
,
7640 .releasepage
= btrfs_releasepage
,
7643 static const struct inode_operations btrfs_file_inode_operations
= {
7644 .getattr
= btrfs_getattr
,
7645 .setattr
= btrfs_setattr
,
7646 .setxattr
= btrfs_setxattr
,
7647 .getxattr
= btrfs_getxattr
,
7648 .listxattr
= btrfs_listxattr
,
7649 .removexattr
= btrfs_removexattr
,
7650 .permission
= btrfs_permission
,
7651 .fiemap
= btrfs_fiemap
,
7652 .get_acl
= btrfs_get_acl
,
7653 .update_time
= btrfs_update_time
,
7655 static const struct inode_operations btrfs_special_inode_operations
= {
7656 .getattr
= btrfs_getattr
,
7657 .setattr
= btrfs_setattr
,
7658 .permission
= btrfs_permission
,
7659 .setxattr
= btrfs_setxattr
,
7660 .getxattr
= btrfs_getxattr
,
7661 .listxattr
= btrfs_listxattr
,
7662 .removexattr
= btrfs_removexattr
,
7663 .get_acl
= btrfs_get_acl
,
7664 .update_time
= btrfs_update_time
,
7666 static const struct inode_operations btrfs_symlink_inode_operations
= {
7667 .readlink
= generic_readlink
,
7668 .follow_link
= page_follow_link_light
,
7669 .put_link
= page_put_link
,
7670 .getattr
= btrfs_getattr
,
7671 .setattr
= btrfs_setattr
,
7672 .permission
= btrfs_permission
,
7673 .setxattr
= btrfs_setxattr
,
7674 .getxattr
= btrfs_getxattr
,
7675 .listxattr
= btrfs_listxattr
,
7676 .removexattr
= btrfs_removexattr
,
7677 .get_acl
= btrfs_get_acl
,
7678 .update_time
= btrfs_update_time
,
7681 const struct dentry_operations btrfs_dentry_operations
= {
7682 .d_delete
= btrfs_dentry_delete
,
7683 .d_release
= btrfs_dentry_release
,