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>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args
{
57 struct btrfs_root
*root
;
60 static const struct inode_operations btrfs_dir_inode_operations
;
61 static const struct inode_operations btrfs_symlink_inode_operations
;
62 static const struct inode_operations btrfs_dir_ro_inode_operations
;
63 static const struct inode_operations btrfs_special_inode_operations
;
64 static const struct inode_operations btrfs_file_inode_operations
;
65 static const struct address_space_operations btrfs_aops
;
66 static const struct address_space_operations btrfs_symlink_aops
;
67 static const struct file_operations btrfs_dir_file_operations
;
68 static struct extent_io_ops btrfs_extent_io_ops
;
70 static struct kmem_cache
*btrfs_inode_cachep
;
71 struct kmem_cache
*btrfs_trans_handle_cachep
;
72 struct kmem_cache
*btrfs_transaction_cachep
;
73 struct kmem_cache
*btrfs_path_cachep
;
74 struct kmem_cache
*btrfs_free_space_cachep
;
77 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
78 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
79 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
80 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
81 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
82 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
83 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
84 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
87 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
);
88 static int btrfs_truncate(struct inode
*inode
);
89 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
90 static noinline
int cow_file_range(struct inode
*inode
,
91 struct page
*locked_page
,
92 u64 start
, u64 end
, int *page_started
,
93 unsigned long *nr_written
, int unlock
);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
96 struct inode
*inode
, struct inode
*dir
,
97 const struct qstr
*qstr
)
101 err
= btrfs_init_acl(trans
, inode
, dir
);
103 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
108 * this does all the hard work for inserting an inline extent into
109 * the btree. The caller should have done a btrfs_drop_extents so that
110 * no overlapping inline items exist in the btree
112 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
113 struct btrfs_root
*root
, struct inode
*inode
,
114 u64 start
, size_t size
, size_t compressed_size
,
116 struct page
**compressed_pages
)
118 struct btrfs_key key
;
119 struct btrfs_path
*path
;
120 struct extent_buffer
*leaf
;
121 struct page
*page
= NULL
;
124 struct btrfs_file_extent_item
*ei
;
127 size_t cur_size
= size
;
129 unsigned long offset
;
131 if (compressed_size
&& compressed_pages
)
132 cur_size
= compressed_size
;
134 path
= btrfs_alloc_path();
138 path
->leave_spinning
= 1;
139 btrfs_set_trans_block_group(trans
, inode
);
141 key
.objectid
= inode
->i_ino
;
143 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
144 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
146 inode_add_bytes(inode
, size
);
147 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
154 leaf
= path
->nodes
[0];
155 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
156 struct btrfs_file_extent_item
);
157 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
158 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
159 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
160 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
161 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
162 ptr
= btrfs_file_extent_inline_start(ei
);
164 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
167 while (compressed_size
> 0) {
168 cpage
= compressed_pages
[i
];
169 cur_size
= min_t(unsigned long, compressed_size
,
172 kaddr
= kmap_atomic(cpage
, KM_USER0
);
173 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
174 kunmap_atomic(kaddr
, KM_USER0
);
178 compressed_size
-= cur_size
;
180 btrfs_set_file_extent_compression(leaf
, ei
,
183 page
= find_get_page(inode
->i_mapping
,
184 start
>> PAGE_CACHE_SHIFT
);
185 btrfs_set_file_extent_compression(leaf
, ei
, 0);
186 kaddr
= kmap_atomic(page
, KM_USER0
);
187 offset
= start
& (PAGE_CACHE_SIZE
- 1);
188 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
189 kunmap_atomic(kaddr
, KM_USER0
);
190 page_cache_release(page
);
192 btrfs_mark_buffer_dirty(leaf
);
193 btrfs_free_path(path
);
196 * we're an inline extent, so nobody can
197 * extend the file past i_size without locking
198 * a page we already have locked.
200 * We must do any isize and inode updates
201 * before we unlock the pages. Otherwise we
202 * could end up racing with unlink.
204 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
205 btrfs_update_inode(trans
, root
, inode
);
209 btrfs_free_path(path
);
215 * conditionally insert an inline extent into the file. This
216 * does the checks required to make sure the data is small enough
217 * to fit as an inline extent.
219 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
220 struct btrfs_root
*root
,
221 struct inode
*inode
, u64 start
, u64 end
,
222 size_t compressed_size
, int compress_type
,
223 struct page
**compressed_pages
)
225 u64 isize
= i_size_read(inode
);
226 u64 actual_end
= min(end
+ 1, isize
);
227 u64 inline_len
= actual_end
- start
;
228 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
229 ~((u64
)root
->sectorsize
- 1);
231 u64 data_len
= inline_len
;
235 data_len
= compressed_size
;
238 actual_end
>= PAGE_CACHE_SIZE
||
239 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
241 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
243 data_len
> root
->fs_info
->max_inline
) {
247 ret
= btrfs_drop_extents(trans
, inode
, start
, aligned_end
,
251 if (isize
> actual_end
)
252 inline_len
= min_t(u64
, isize
, actual_end
);
253 ret
= insert_inline_extent(trans
, root
, inode
, start
,
254 inline_len
, compressed_size
,
255 compress_type
, compressed_pages
);
257 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
258 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
262 struct async_extent
{
267 unsigned long nr_pages
;
269 struct list_head list
;
274 struct btrfs_root
*root
;
275 struct page
*locked_page
;
278 struct list_head extents
;
279 struct btrfs_work work
;
282 static noinline
int add_async_extent(struct async_cow
*cow
,
283 u64 start
, u64 ram_size
,
286 unsigned long nr_pages
,
289 struct async_extent
*async_extent
;
291 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
292 BUG_ON(!async_extent
);
293 async_extent
->start
= start
;
294 async_extent
->ram_size
= ram_size
;
295 async_extent
->compressed_size
= compressed_size
;
296 async_extent
->pages
= pages
;
297 async_extent
->nr_pages
= nr_pages
;
298 async_extent
->compress_type
= compress_type
;
299 list_add_tail(&async_extent
->list
, &cow
->extents
);
304 * we create compressed extents in two phases. The first
305 * phase compresses a range of pages that have already been
306 * locked (both pages and state bits are locked).
308 * This is done inside an ordered work queue, and the compression
309 * is spread across many cpus. The actual IO submission is step
310 * two, and the ordered work queue takes care of making sure that
311 * happens in the same order things were put onto the queue by
312 * writepages and friends.
314 * If this code finds it can't get good compression, it puts an
315 * entry onto the work queue to write the uncompressed bytes. This
316 * makes sure that both compressed inodes and uncompressed inodes
317 * are written in the same order that pdflush sent them down.
319 static noinline
int compress_file_range(struct inode
*inode
,
320 struct page
*locked_page
,
322 struct async_cow
*async_cow
,
325 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
326 struct btrfs_trans_handle
*trans
;
328 u64 blocksize
= root
->sectorsize
;
330 u64 isize
= i_size_read(inode
);
332 struct page
**pages
= NULL
;
333 unsigned long nr_pages
;
334 unsigned long nr_pages_ret
= 0;
335 unsigned long total_compressed
= 0;
336 unsigned long total_in
= 0;
337 unsigned long max_compressed
= 128 * 1024;
338 unsigned long max_uncompressed
= 128 * 1024;
341 int compress_type
= root
->fs_info
->compress_type
;
343 actual_end
= min_t(u64
, isize
, end
+ 1);
346 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
347 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
350 * we don't want to send crud past the end of i_size through
351 * compression, that's just a waste of CPU time. So, if the
352 * end of the file is before the start of our current
353 * requested range of bytes, we bail out to the uncompressed
354 * cleanup code that can deal with all of this.
356 * It isn't really the fastest way to fix things, but this is a
357 * very uncommon corner.
359 if (actual_end
<= start
)
360 goto cleanup_and_bail_uncompressed
;
362 total_compressed
= actual_end
- start
;
364 /* we want to make sure that amount of ram required to uncompress
365 * an extent is reasonable, so we limit the total size in ram
366 * of a compressed extent to 128k. This is a crucial number
367 * because it also controls how easily we can spread reads across
368 * cpus for decompression.
370 * We also want to make sure the amount of IO required to do
371 * a random read is reasonably small, so we limit the size of
372 * a compressed extent to 128k.
374 total_compressed
= min(total_compressed
, max_uncompressed
);
375 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
376 num_bytes
= max(blocksize
, num_bytes
);
381 * we do compression for mount -o compress and when the
382 * inode has not been flagged as nocompress. This flag can
383 * change at any time if we discover bad compression ratios.
385 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
386 (btrfs_test_opt(root
, COMPRESS
) ||
387 (BTRFS_I(inode
)->force_compress
) ||
388 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
390 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
393 if (BTRFS_I(inode
)->force_compress
)
394 compress_type
= BTRFS_I(inode
)->force_compress
;
396 ret
= btrfs_compress_pages(compress_type
,
397 inode
->i_mapping
, start
,
398 total_compressed
, pages
,
399 nr_pages
, &nr_pages_ret
,
405 unsigned long offset
= total_compressed
&
406 (PAGE_CACHE_SIZE
- 1);
407 struct page
*page
= pages
[nr_pages_ret
- 1];
410 /* zero the tail end of the last page, we might be
411 * sending it down to disk
414 kaddr
= kmap_atomic(page
, KM_USER0
);
415 memset(kaddr
+ offset
, 0,
416 PAGE_CACHE_SIZE
- offset
);
417 kunmap_atomic(kaddr
, KM_USER0
);
423 trans
= btrfs_join_transaction(root
);
424 BUG_ON(IS_ERR(trans
));
425 btrfs_set_trans_block_group(trans
, inode
);
426 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
428 /* lets try to make an inline extent */
429 if (ret
|| total_in
< (actual_end
- start
)) {
430 /* we didn't compress the entire range, try
431 * to make an uncompressed inline extent.
433 ret
= cow_file_range_inline(trans
, root
, inode
,
434 start
, end
, 0, 0, NULL
);
436 /* try making a compressed inline extent */
437 ret
= cow_file_range_inline(trans
, root
, inode
,
440 compress_type
, pages
);
444 * inline extent creation worked, we don't need
445 * to create any more async work items. Unlock
446 * and free up our temp pages.
448 extent_clear_unlock_delalloc(inode
,
449 &BTRFS_I(inode
)->io_tree
,
451 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
452 EXTENT_CLEAR_DELALLOC
|
453 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
455 btrfs_end_transaction(trans
, root
);
458 btrfs_end_transaction(trans
, root
);
463 * we aren't doing an inline extent round the compressed size
464 * up to a block size boundary so the allocator does sane
467 total_compressed
= (total_compressed
+ blocksize
- 1) &
471 * one last check to make sure the compression is really a
472 * win, compare the page count read with the blocks on disk
474 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
475 ~(PAGE_CACHE_SIZE
- 1);
476 if (total_compressed
>= total_in
) {
479 num_bytes
= total_in
;
482 if (!will_compress
&& pages
) {
484 * the compression code ran but failed to make things smaller,
485 * free any pages it allocated and our page pointer array
487 for (i
= 0; i
< nr_pages_ret
; i
++) {
488 WARN_ON(pages
[i
]->mapping
);
489 page_cache_release(pages
[i
]);
493 total_compressed
= 0;
496 /* flag the file so we don't compress in the future */
497 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
498 !(BTRFS_I(inode
)->force_compress
)) {
499 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
505 /* the async work queues will take care of doing actual
506 * allocation on disk for these compressed pages,
507 * and will submit them to the elevator.
509 add_async_extent(async_cow
, start
, num_bytes
,
510 total_compressed
, pages
, nr_pages_ret
,
513 if (start
+ num_bytes
< end
) {
520 cleanup_and_bail_uncompressed
:
522 * No compression, but we still need to write the pages in
523 * the file we've been given so far. redirty the locked
524 * page if it corresponds to our extent and set things up
525 * for the async work queue to run cow_file_range to do
526 * the normal delalloc dance
528 if (page_offset(locked_page
) >= start
&&
529 page_offset(locked_page
) <= end
) {
530 __set_page_dirty_nobuffers(locked_page
);
531 /* unlocked later on in the async handlers */
533 add_async_extent(async_cow
, start
, end
- start
+ 1,
534 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
542 for (i
= 0; i
< nr_pages_ret
; i
++) {
543 WARN_ON(pages
[i
]->mapping
);
544 page_cache_release(pages
[i
]);
552 * phase two of compressed writeback. This is the ordered portion
553 * of the code, which only gets called in the order the work was
554 * queued. We walk all the async extents created by compress_file_range
555 * and send them down to the disk.
557 static noinline
int submit_compressed_extents(struct inode
*inode
,
558 struct async_cow
*async_cow
)
560 struct async_extent
*async_extent
;
562 struct btrfs_trans_handle
*trans
;
563 struct btrfs_key ins
;
564 struct extent_map
*em
;
565 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
566 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
567 struct extent_io_tree
*io_tree
;
570 if (list_empty(&async_cow
->extents
))
574 while (!list_empty(&async_cow
->extents
)) {
575 async_extent
= list_entry(async_cow
->extents
.next
,
576 struct async_extent
, list
);
577 list_del(&async_extent
->list
);
579 io_tree
= &BTRFS_I(inode
)->io_tree
;
582 /* did the compression code fall back to uncompressed IO? */
583 if (!async_extent
->pages
) {
584 int page_started
= 0;
585 unsigned long nr_written
= 0;
587 lock_extent(io_tree
, async_extent
->start
,
588 async_extent
->start
+
589 async_extent
->ram_size
- 1, GFP_NOFS
);
591 /* allocate blocks */
592 ret
= cow_file_range(inode
, async_cow
->locked_page
,
594 async_extent
->start
+
595 async_extent
->ram_size
- 1,
596 &page_started
, &nr_written
, 0);
599 * if page_started, cow_file_range inserted an
600 * inline extent and took care of all the unlocking
601 * and IO for us. Otherwise, we need to submit
602 * all those pages down to the drive.
604 if (!page_started
&& !ret
)
605 extent_write_locked_range(io_tree
,
606 inode
, async_extent
->start
,
607 async_extent
->start
+
608 async_extent
->ram_size
- 1,
616 lock_extent(io_tree
, async_extent
->start
,
617 async_extent
->start
+ async_extent
->ram_size
- 1,
620 trans
= btrfs_join_transaction(root
);
621 BUG_ON(IS_ERR(trans
));
622 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
623 ret
= btrfs_reserve_extent(trans
, root
,
624 async_extent
->compressed_size
,
625 async_extent
->compressed_size
,
628 btrfs_end_transaction(trans
, root
);
632 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
633 WARN_ON(async_extent
->pages
[i
]->mapping
);
634 page_cache_release(async_extent
->pages
[i
]);
636 kfree(async_extent
->pages
);
637 async_extent
->nr_pages
= 0;
638 async_extent
->pages
= NULL
;
639 unlock_extent(io_tree
, async_extent
->start
,
640 async_extent
->start
+
641 async_extent
->ram_size
- 1, GFP_NOFS
);
646 * here we're doing allocation and writeback of the
649 btrfs_drop_extent_cache(inode
, async_extent
->start
,
650 async_extent
->start
+
651 async_extent
->ram_size
- 1, 0);
653 em
= alloc_extent_map(GFP_NOFS
);
655 em
->start
= async_extent
->start
;
656 em
->len
= async_extent
->ram_size
;
657 em
->orig_start
= em
->start
;
659 em
->block_start
= ins
.objectid
;
660 em
->block_len
= ins
.offset
;
661 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
662 em
->compress_type
= async_extent
->compress_type
;
663 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
664 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
667 write_lock(&em_tree
->lock
);
668 ret
= add_extent_mapping(em_tree
, em
);
669 write_unlock(&em_tree
->lock
);
670 if (ret
!= -EEXIST
) {
674 btrfs_drop_extent_cache(inode
, async_extent
->start
,
675 async_extent
->start
+
676 async_extent
->ram_size
- 1, 0);
679 ret
= btrfs_add_ordered_extent_compress(inode
,
682 async_extent
->ram_size
,
684 BTRFS_ORDERED_COMPRESSED
,
685 async_extent
->compress_type
);
689 * clear dirty, set writeback and unlock the pages.
691 extent_clear_unlock_delalloc(inode
,
692 &BTRFS_I(inode
)->io_tree
,
694 async_extent
->start
+
695 async_extent
->ram_size
- 1,
696 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
697 EXTENT_CLEAR_UNLOCK
|
698 EXTENT_CLEAR_DELALLOC
|
699 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
701 ret
= btrfs_submit_compressed_write(inode
,
703 async_extent
->ram_size
,
705 ins
.offset
, async_extent
->pages
,
706 async_extent
->nr_pages
);
709 alloc_hint
= ins
.objectid
+ ins
.offset
;
717 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
720 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
721 struct extent_map
*em
;
724 read_lock(&em_tree
->lock
);
725 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
728 * if block start isn't an actual block number then find the
729 * first block in this inode and use that as a hint. If that
730 * block is also bogus then just don't worry about it.
732 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
734 em
= search_extent_mapping(em_tree
, 0, 0);
735 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
736 alloc_hint
= em
->block_start
;
740 alloc_hint
= em
->block_start
;
744 read_unlock(&em_tree
->lock
);
750 * when extent_io.c finds a delayed allocation range in the file,
751 * the call backs end up in this code. The basic idea is to
752 * allocate extents on disk for the range, and create ordered data structs
753 * in ram to track those extents.
755 * locked_page is the page that writepage had locked already. We use
756 * it to make sure we don't do extra locks or unlocks.
758 * *page_started is set to one if we unlock locked_page and do everything
759 * required to start IO on it. It may be clean and already done with
762 static noinline
int cow_file_range(struct inode
*inode
,
763 struct page
*locked_page
,
764 u64 start
, u64 end
, int *page_started
,
765 unsigned long *nr_written
,
768 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
769 struct btrfs_trans_handle
*trans
;
772 unsigned long ram_size
;
775 u64 blocksize
= root
->sectorsize
;
776 struct btrfs_key ins
;
777 struct extent_map
*em
;
778 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
781 BUG_ON(root
== root
->fs_info
->tree_root
);
782 trans
= btrfs_join_transaction(root
);
783 BUG_ON(IS_ERR(trans
));
784 btrfs_set_trans_block_group(trans
, inode
);
785 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
787 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
788 num_bytes
= max(blocksize
, num_bytes
);
789 disk_num_bytes
= num_bytes
;
793 /* lets try to make an inline extent */
794 ret
= cow_file_range_inline(trans
, root
, inode
,
795 start
, end
, 0, 0, NULL
);
797 extent_clear_unlock_delalloc(inode
,
798 &BTRFS_I(inode
)->io_tree
,
800 EXTENT_CLEAR_UNLOCK_PAGE
|
801 EXTENT_CLEAR_UNLOCK
|
802 EXTENT_CLEAR_DELALLOC
|
804 EXTENT_SET_WRITEBACK
|
805 EXTENT_END_WRITEBACK
);
807 *nr_written
= *nr_written
+
808 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
815 BUG_ON(disk_num_bytes
>
816 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
818 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
819 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
821 while (disk_num_bytes
> 0) {
824 cur_alloc_size
= disk_num_bytes
;
825 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
826 root
->sectorsize
, 0, alloc_hint
,
830 em
= alloc_extent_map(GFP_NOFS
);
833 em
->orig_start
= em
->start
;
834 ram_size
= ins
.offset
;
835 em
->len
= ins
.offset
;
837 em
->block_start
= ins
.objectid
;
838 em
->block_len
= ins
.offset
;
839 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
840 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
843 write_lock(&em_tree
->lock
);
844 ret
= add_extent_mapping(em_tree
, em
);
845 write_unlock(&em_tree
->lock
);
846 if (ret
!= -EEXIST
) {
850 btrfs_drop_extent_cache(inode
, start
,
851 start
+ ram_size
- 1, 0);
854 cur_alloc_size
= ins
.offset
;
855 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
856 ram_size
, cur_alloc_size
, 0);
859 if (root
->root_key
.objectid
==
860 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
861 ret
= btrfs_reloc_clone_csums(inode
, start
,
866 if (disk_num_bytes
< cur_alloc_size
)
869 /* we're not doing compressed IO, don't unlock the first
870 * page (which the caller expects to stay locked), don't
871 * clear any dirty bits and don't set any writeback bits
873 * Do set the Private2 bit so we know this page was properly
874 * setup for writepage
876 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
877 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
880 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
881 start
, start
+ ram_size
- 1,
883 disk_num_bytes
-= cur_alloc_size
;
884 num_bytes
-= cur_alloc_size
;
885 alloc_hint
= ins
.objectid
+ ins
.offset
;
886 start
+= cur_alloc_size
;
890 btrfs_end_transaction(trans
, root
);
896 * work queue call back to started compression on a file and pages
898 static noinline
void async_cow_start(struct btrfs_work
*work
)
900 struct async_cow
*async_cow
;
902 async_cow
= container_of(work
, struct async_cow
, work
);
904 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
905 async_cow
->start
, async_cow
->end
, async_cow
,
908 async_cow
->inode
= NULL
;
912 * work queue call back to submit previously compressed pages
914 static noinline
void async_cow_submit(struct btrfs_work
*work
)
916 struct async_cow
*async_cow
;
917 struct btrfs_root
*root
;
918 unsigned long nr_pages
;
920 async_cow
= container_of(work
, struct async_cow
, work
);
922 root
= async_cow
->root
;
923 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
926 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
928 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
930 waitqueue_active(&root
->fs_info
->async_submit_wait
))
931 wake_up(&root
->fs_info
->async_submit_wait
);
933 if (async_cow
->inode
)
934 submit_compressed_extents(async_cow
->inode
, async_cow
);
937 static noinline
void async_cow_free(struct btrfs_work
*work
)
939 struct async_cow
*async_cow
;
940 async_cow
= container_of(work
, struct async_cow
, work
);
944 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
945 u64 start
, u64 end
, int *page_started
,
946 unsigned long *nr_written
)
948 struct async_cow
*async_cow
;
949 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
950 unsigned long nr_pages
;
952 int limit
= 10 * 1024 * 1042;
954 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
955 1, 0, NULL
, GFP_NOFS
);
956 while (start
< end
) {
957 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
959 async_cow
->inode
= inode
;
960 async_cow
->root
= root
;
961 async_cow
->locked_page
= locked_page
;
962 async_cow
->start
= start
;
964 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
967 cur_end
= min(end
, start
+ 512 * 1024 - 1);
969 async_cow
->end
= cur_end
;
970 INIT_LIST_HEAD(&async_cow
->extents
);
972 async_cow
->work
.func
= async_cow_start
;
973 async_cow
->work
.ordered_func
= async_cow_submit
;
974 async_cow
->work
.ordered_free
= async_cow_free
;
975 async_cow
->work
.flags
= 0;
977 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
979 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
981 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
984 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
985 wait_event(root
->fs_info
->async_submit_wait
,
986 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
990 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
991 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
992 wait_event(root
->fs_info
->async_submit_wait
,
993 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
997 *nr_written
+= nr_pages
;
1004 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1005 u64 bytenr
, u64 num_bytes
)
1008 struct btrfs_ordered_sum
*sums
;
1011 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1012 bytenr
+ num_bytes
- 1, &list
);
1013 if (ret
== 0 && list_empty(&list
))
1016 while (!list_empty(&list
)) {
1017 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1018 list_del(&sums
->list
);
1025 * when nowcow writeback call back. This checks for snapshots or COW copies
1026 * of the extents that exist in the file, and COWs the file as required.
1028 * If no cow copies or snapshots exist, we write directly to the existing
1031 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1032 struct page
*locked_page
,
1033 u64 start
, u64 end
, int *page_started
, int force
,
1034 unsigned long *nr_written
)
1036 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1037 struct btrfs_trans_handle
*trans
;
1038 struct extent_buffer
*leaf
;
1039 struct btrfs_path
*path
;
1040 struct btrfs_file_extent_item
*fi
;
1041 struct btrfs_key found_key
;
1053 bool nolock
= false;
1055 path
= btrfs_alloc_path();
1057 if (root
== root
->fs_info
->tree_root
) {
1059 trans
= btrfs_join_transaction_nolock(root
);
1061 trans
= btrfs_join_transaction(root
);
1063 BUG_ON(IS_ERR(trans
));
1064 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1066 cow_start
= (u64
)-1;
1069 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
1072 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1073 leaf
= path
->nodes
[0];
1074 btrfs_item_key_to_cpu(leaf
, &found_key
,
1075 path
->slots
[0] - 1);
1076 if (found_key
.objectid
== inode
->i_ino
&&
1077 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1082 leaf
= path
->nodes
[0];
1083 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1084 ret
= btrfs_next_leaf(root
, path
);
1089 leaf
= path
->nodes
[0];
1095 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1097 if (found_key
.objectid
> inode
->i_ino
||
1098 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1099 found_key
.offset
> end
)
1102 if (found_key
.offset
> cur_offset
) {
1103 extent_end
= found_key
.offset
;
1108 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1109 struct btrfs_file_extent_item
);
1110 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1112 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1113 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1114 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1115 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1116 extent_end
= found_key
.offset
+
1117 btrfs_file_extent_num_bytes(leaf
, fi
);
1118 if (extent_end
<= start
) {
1122 if (disk_bytenr
== 0)
1124 if (btrfs_file_extent_compression(leaf
, fi
) ||
1125 btrfs_file_extent_encryption(leaf
, fi
) ||
1126 btrfs_file_extent_other_encoding(leaf
, fi
))
1128 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1130 if (btrfs_extent_readonly(root
, disk_bytenr
))
1132 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1134 extent_offset
, disk_bytenr
))
1136 disk_bytenr
+= extent_offset
;
1137 disk_bytenr
+= cur_offset
- found_key
.offset
;
1138 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1140 * force cow if csum exists in the range.
1141 * this ensure that csum for a given extent are
1142 * either valid or do not exist.
1144 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1147 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1148 extent_end
= found_key
.offset
+
1149 btrfs_file_extent_inline_len(leaf
, fi
);
1150 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1155 if (extent_end
<= start
) {
1160 if (cow_start
== (u64
)-1)
1161 cow_start
= cur_offset
;
1162 cur_offset
= extent_end
;
1163 if (cur_offset
> end
)
1169 btrfs_release_path(root
, path
);
1170 if (cow_start
!= (u64
)-1) {
1171 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1172 found_key
.offset
- 1, page_started
,
1175 cow_start
= (u64
)-1;
1178 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1179 struct extent_map
*em
;
1180 struct extent_map_tree
*em_tree
;
1181 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1182 em
= alloc_extent_map(GFP_NOFS
);
1184 em
->start
= cur_offset
;
1185 em
->orig_start
= em
->start
;
1186 em
->len
= num_bytes
;
1187 em
->block_len
= num_bytes
;
1188 em
->block_start
= disk_bytenr
;
1189 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1190 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1192 write_lock(&em_tree
->lock
);
1193 ret
= add_extent_mapping(em_tree
, em
);
1194 write_unlock(&em_tree
->lock
);
1195 if (ret
!= -EEXIST
) {
1196 free_extent_map(em
);
1199 btrfs_drop_extent_cache(inode
, em
->start
,
1200 em
->start
+ em
->len
- 1, 0);
1202 type
= BTRFS_ORDERED_PREALLOC
;
1204 type
= BTRFS_ORDERED_NOCOW
;
1207 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1208 num_bytes
, num_bytes
, type
);
1211 if (root
->root_key
.objectid
==
1212 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1213 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1218 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1219 cur_offset
, cur_offset
+ num_bytes
- 1,
1220 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1221 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1222 EXTENT_SET_PRIVATE2
);
1223 cur_offset
= extent_end
;
1224 if (cur_offset
> end
)
1227 btrfs_release_path(root
, path
);
1229 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1230 cow_start
= cur_offset
;
1231 if (cow_start
!= (u64
)-1) {
1232 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1233 page_started
, nr_written
, 1);
1238 ret
= btrfs_end_transaction_nolock(trans
, root
);
1241 ret
= btrfs_end_transaction(trans
, root
);
1244 btrfs_free_path(path
);
1249 * extent_io.c call back to do delayed allocation processing
1251 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1252 u64 start
, u64 end
, int *page_started
,
1253 unsigned long *nr_written
)
1256 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1258 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1259 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1260 page_started
, 1, nr_written
);
1261 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1262 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1263 page_started
, 0, nr_written
);
1264 else if (!btrfs_test_opt(root
, COMPRESS
) &&
1265 !(BTRFS_I(inode
)->force_compress
) &&
1266 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))
1267 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1268 page_started
, nr_written
, 1);
1270 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1271 page_started
, nr_written
);
1275 static int btrfs_split_extent_hook(struct inode
*inode
,
1276 struct extent_state
*orig
, u64 split
)
1278 /* not delalloc, ignore it */
1279 if (!(orig
->state
& EXTENT_DELALLOC
))
1282 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1287 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1288 * extents so we can keep track of new extents that are just merged onto old
1289 * extents, such as when we are doing sequential writes, so we can properly
1290 * account for the metadata space we'll need.
1292 static int btrfs_merge_extent_hook(struct inode
*inode
,
1293 struct extent_state
*new,
1294 struct extent_state
*other
)
1296 /* not delalloc, ignore it */
1297 if (!(other
->state
& EXTENT_DELALLOC
))
1300 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1305 * extent_io.c set_bit_hook, used to track delayed allocation
1306 * bytes in this file, and to maintain the list of inodes that
1307 * have pending delalloc work to be done.
1309 static int btrfs_set_bit_hook(struct inode
*inode
,
1310 struct extent_state
*state
, int *bits
)
1314 * set_bit and clear bit hooks normally require _irqsave/restore
1315 * but in this case, we are only testeing for the DELALLOC
1316 * bit, which is only set or cleared with irqs on
1318 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1319 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1320 u64 len
= state
->end
+ 1 - state
->start
;
1321 int do_list
= (root
->root_key
.objectid
!=
1322 BTRFS_ROOT_TREE_OBJECTID
);
1324 if (*bits
& EXTENT_FIRST_DELALLOC
)
1325 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1327 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1329 spin_lock(&root
->fs_info
->delalloc_lock
);
1330 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1331 root
->fs_info
->delalloc_bytes
+= len
;
1332 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1333 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1334 &root
->fs_info
->delalloc_inodes
);
1336 spin_unlock(&root
->fs_info
->delalloc_lock
);
1342 * extent_io.c clear_bit_hook, see set_bit_hook for why
1344 static int btrfs_clear_bit_hook(struct inode
*inode
,
1345 struct extent_state
*state
, int *bits
)
1348 * set_bit and clear bit hooks normally require _irqsave/restore
1349 * but in this case, we are only testeing for the DELALLOC
1350 * bit, which is only set or cleared with irqs on
1352 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1353 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1354 u64 len
= state
->end
+ 1 - state
->start
;
1355 int do_list
= (root
->root_key
.objectid
!=
1356 BTRFS_ROOT_TREE_OBJECTID
);
1358 if (*bits
& EXTENT_FIRST_DELALLOC
)
1359 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1360 else if (!(*bits
& EXTENT_DO_ACCOUNTING
))
1361 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1363 if (*bits
& EXTENT_DO_ACCOUNTING
)
1364 btrfs_delalloc_release_metadata(inode
, len
);
1366 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1368 btrfs_free_reserved_data_space(inode
, len
);
1370 spin_lock(&root
->fs_info
->delalloc_lock
);
1371 root
->fs_info
->delalloc_bytes
-= len
;
1372 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1374 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1375 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1376 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1378 spin_unlock(&root
->fs_info
->delalloc_lock
);
1384 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1385 * we don't create bios that span stripes or chunks
1387 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1388 size_t size
, struct bio
*bio
,
1389 unsigned long bio_flags
)
1391 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1392 struct btrfs_mapping_tree
*map_tree
;
1393 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1398 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1401 length
= bio
->bi_size
;
1402 map_tree
= &root
->fs_info
->mapping_tree
;
1403 map_length
= length
;
1404 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1405 &map_length
, NULL
, 0);
1407 if (map_length
< length
+ size
)
1413 * in order to insert checksums into the metadata in large chunks,
1414 * we wait until bio submission time. All the pages in the bio are
1415 * checksummed and sums are attached onto the ordered extent record.
1417 * At IO completion time the cums attached on the ordered extent record
1418 * are inserted into the btree
1420 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1421 struct bio
*bio
, int mirror_num
,
1422 unsigned long bio_flags
,
1425 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1428 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1434 * in order to insert checksums into the metadata in large chunks,
1435 * we wait until bio submission time. All the pages in the bio are
1436 * checksummed and sums are attached onto the ordered extent record.
1438 * At IO completion time the cums attached on the ordered extent record
1439 * are inserted into the btree
1441 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1442 int mirror_num
, unsigned long bio_flags
,
1445 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1446 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1450 * extent_io.c submission hook. This does the right thing for csum calculation
1451 * on write, or reading the csums from the tree before a read
1453 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1454 int mirror_num
, unsigned long bio_flags
,
1457 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1461 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1463 if (root
== root
->fs_info
->tree_root
)
1464 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 2);
1466 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1469 if (!(rw
& REQ_WRITE
)) {
1470 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1471 return btrfs_submit_compressed_read(inode
, bio
,
1472 mirror_num
, bio_flags
);
1473 } else if (!skip_sum
) {
1474 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1479 } else if (!skip_sum
) {
1480 /* csum items have already been cloned */
1481 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1483 /* we're doing a write, do the async checksumming */
1484 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1485 inode
, rw
, bio
, mirror_num
,
1486 bio_flags
, bio_offset
,
1487 __btrfs_submit_bio_start
,
1488 __btrfs_submit_bio_done
);
1492 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1496 * given a list of ordered sums record them in the inode. This happens
1497 * at IO completion time based on sums calculated at bio submission time.
1499 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1500 struct inode
*inode
, u64 file_offset
,
1501 struct list_head
*list
)
1503 struct btrfs_ordered_sum
*sum
;
1505 btrfs_set_trans_block_group(trans
, inode
);
1507 list_for_each_entry(sum
, list
, list
) {
1508 btrfs_csum_file_blocks(trans
,
1509 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1514 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1515 struct extent_state
**cached_state
)
1517 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1519 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1520 cached_state
, GFP_NOFS
);
1523 /* see btrfs_writepage_start_hook for details on why this is required */
1524 struct btrfs_writepage_fixup
{
1526 struct btrfs_work work
;
1529 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1531 struct btrfs_writepage_fixup
*fixup
;
1532 struct btrfs_ordered_extent
*ordered
;
1533 struct extent_state
*cached_state
= NULL
;
1535 struct inode
*inode
;
1539 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1543 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1544 ClearPageChecked(page
);
1548 inode
= page
->mapping
->host
;
1549 page_start
= page_offset(page
);
1550 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1552 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1553 &cached_state
, GFP_NOFS
);
1555 /* already ordered? We're done */
1556 if (PagePrivate2(page
))
1559 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1561 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1562 page_end
, &cached_state
, GFP_NOFS
);
1564 btrfs_start_ordered_extent(inode
, ordered
, 1);
1569 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1570 ClearPageChecked(page
);
1572 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1573 &cached_state
, GFP_NOFS
);
1576 page_cache_release(page
);
1581 * There are a few paths in the higher layers of the kernel that directly
1582 * set the page dirty bit without asking the filesystem if it is a
1583 * good idea. This causes problems because we want to make sure COW
1584 * properly happens and the data=ordered rules are followed.
1586 * In our case any range that doesn't have the ORDERED bit set
1587 * hasn't been properly setup for IO. We kick off an async process
1588 * to fix it up. The async helper will wait for ordered extents, set
1589 * the delalloc bit and make it safe to write the page.
1591 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1593 struct inode
*inode
= page
->mapping
->host
;
1594 struct btrfs_writepage_fixup
*fixup
;
1595 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1597 /* this page is properly in the ordered list */
1598 if (TestClearPagePrivate2(page
))
1601 if (PageChecked(page
))
1604 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1608 SetPageChecked(page
);
1609 page_cache_get(page
);
1610 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1612 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1616 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1617 struct inode
*inode
, u64 file_pos
,
1618 u64 disk_bytenr
, u64 disk_num_bytes
,
1619 u64 num_bytes
, u64 ram_bytes
,
1620 u8 compression
, u8 encryption
,
1621 u16 other_encoding
, int extent_type
)
1623 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1624 struct btrfs_file_extent_item
*fi
;
1625 struct btrfs_path
*path
;
1626 struct extent_buffer
*leaf
;
1627 struct btrfs_key ins
;
1631 path
= btrfs_alloc_path();
1634 path
->leave_spinning
= 1;
1637 * we may be replacing one extent in the tree with another.
1638 * The new extent is pinned in the extent map, and we don't want
1639 * to drop it from the cache until it is completely in the btree.
1641 * So, tell btrfs_drop_extents to leave this extent in the cache.
1642 * the caller is expected to unpin it and allow it to be merged
1645 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1649 ins
.objectid
= inode
->i_ino
;
1650 ins
.offset
= file_pos
;
1651 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1652 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1654 leaf
= path
->nodes
[0];
1655 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1656 struct btrfs_file_extent_item
);
1657 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1658 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1659 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1660 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1661 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1662 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1663 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1664 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1665 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1666 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1668 btrfs_unlock_up_safe(path
, 1);
1669 btrfs_set_lock_blocking(leaf
);
1671 btrfs_mark_buffer_dirty(leaf
);
1673 inode_add_bytes(inode
, num_bytes
);
1675 ins
.objectid
= disk_bytenr
;
1676 ins
.offset
= disk_num_bytes
;
1677 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1678 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1679 root
->root_key
.objectid
,
1680 inode
->i_ino
, file_pos
, &ins
);
1682 btrfs_free_path(path
);
1688 * helper function for btrfs_finish_ordered_io, this
1689 * just reads in some of the csum leaves to prime them into ram
1690 * before we start the transaction. It limits the amount of btree
1691 * reads required while inside the transaction.
1693 /* as ordered data IO finishes, this gets called so we can finish
1694 * an ordered extent if the range of bytes in the file it covers are
1697 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1699 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1700 struct btrfs_trans_handle
*trans
= NULL
;
1701 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1702 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1703 struct extent_state
*cached_state
= NULL
;
1704 int compress_type
= 0;
1706 bool nolock
= false;
1708 ret
= btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1712 BUG_ON(!ordered_extent
);
1714 nolock
= (root
== root
->fs_info
->tree_root
);
1716 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1717 BUG_ON(!list_empty(&ordered_extent
->list
));
1718 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1721 trans
= btrfs_join_transaction_nolock(root
);
1723 trans
= btrfs_join_transaction(root
);
1724 BUG_ON(IS_ERR(trans
));
1725 btrfs_set_trans_block_group(trans
, inode
);
1726 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1727 ret
= btrfs_update_inode(trans
, root
, inode
);
1733 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1734 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1735 0, &cached_state
, GFP_NOFS
);
1738 trans
= btrfs_join_transaction_nolock(root
);
1740 trans
= btrfs_join_transaction(root
);
1741 BUG_ON(IS_ERR(trans
));
1742 btrfs_set_trans_block_group(trans
, inode
);
1743 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1745 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1746 compress_type
= ordered_extent
->compress_type
;
1747 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1748 BUG_ON(compress_type
);
1749 ret
= btrfs_mark_extent_written(trans
, inode
,
1750 ordered_extent
->file_offset
,
1751 ordered_extent
->file_offset
+
1752 ordered_extent
->len
);
1755 BUG_ON(root
== root
->fs_info
->tree_root
);
1756 ret
= insert_reserved_file_extent(trans
, inode
,
1757 ordered_extent
->file_offset
,
1758 ordered_extent
->start
,
1759 ordered_extent
->disk_len
,
1760 ordered_extent
->len
,
1761 ordered_extent
->len
,
1762 compress_type
, 0, 0,
1763 BTRFS_FILE_EXTENT_REG
);
1764 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1765 ordered_extent
->file_offset
,
1766 ordered_extent
->len
);
1769 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1770 ordered_extent
->file_offset
+
1771 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1773 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1774 &ordered_extent
->list
);
1776 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1778 ret
= btrfs_update_inode(trans
, root
, inode
);
1785 btrfs_end_transaction_nolock(trans
, root
);
1787 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1789 btrfs_end_transaction(trans
, root
);
1793 btrfs_put_ordered_extent(ordered_extent
);
1794 /* once for the tree */
1795 btrfs_put_ordered_extent(ordered_extent
);
1800 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1801 struct extent_state
*state
, int uptodate
)
1803 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1805 ClearPagePrivate2(page
);
1806 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1810 * When IO fails, either with EIO or csum verification fails, we
1811 * try other mirrors that might have a good copy of the data. This
1812 * io_failure_record is used to record state as we go through all the
1813 * mirrors. If another mirror has good data, the page is set up to date
1814 * and things continue. If a good mirror can't be found, the original
1815 * bio end_io callback is called to indicate things have failed.
1817 struct io_failure_record
{
1822 unsigned long bio_flags
;
1826 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1827 struct page
*page
, u64 start
, u64 end
,
1828 struct extent_state
*state
)
1830 struct io_failure_record
*failrec
= NULL
;
1832 struct extent_map
*em
;
1833 struct inode
*inode
= page
->mapping
->host
;
1834 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1835 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1842 ret
= get_state_private(failure_tree
, start
, &private);
1844 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1847 failrec
->start
= start
;
1848 failrec
->len
= end
- start
+ 1;
1849 failrec
->last_mirror
= 0;
1850 failrec
->bio_flags
= 0;
1852 read_lock(&em_tree
->lock
);
1853 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1854 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1855 free_extent_map(em
);
1858 read_unlock(&em_tree
->lock
);
1860 if (!em
|| IS_ERR(em
)) {
1864 logical
= start
- em
->start
;
1865 logical
= em
->block_start
+ logical
;
1866 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1867 logical
= em
->block_start
;
1868 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1869 extent_set_compress_type(&failrec
->bio_flags
,
1872 failrec
->logical
= logical
;
1873 free_extent_map(em
);
1874 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1875 EXTENT_DIRTY
, GFP_NOFS
);
1876 set_state_private(failure_tree
, start
,
1877 (u64
)(unsigned long)failrec
);
1879 failrec
= (struct io_failure_record
*)(unsigned long)private;
1881 num_copies
= btrfs_num_copies(
1882 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1883 failrec
->logical
, failrec
->len
);
1884 failrec
->last_mirror
++;
1886 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1887 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1890 if (state
&& state
->start
!= failrec
->start
)
1892 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1894 if (!state
|| failrec
->last_mirror
> num_copies
) {
1895 set_state_private(failure_tree
, failrec
->start
, 0);
1896 clear_extent_bits(failure_tree
, failrec
->start
,
1897 failrec
->start
+ failrec
->len
- 1,
1898 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1902 bio
= bio_alloc(GFP_NOFS
, 1);
1903 bio
->bi_private
= state
;
1904 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1905 bio
->bi_sector
= failrec
->logical
>> 9;
1906 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1909 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1910 if (failed_bio
->bi_rw
& REQ_WRITE
)
1915 ret
= BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1916 failrec
->last_mirror
,
1917 failrec
->bio_flags
, 0);
1922 * each time an IO finishes, we do a fast check in the IO failure tree
1923 * to see if we need to process or clean up an io_failure_record
1925 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1928 u64 private_failure
;
1929 struct io_failure_record
*failure
;
1933 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1934 (u64
)-1, 1, EXTENT_DIRTY
, 0)) {
1935 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1936 start
, &private_failure
);
1938 failure
= (struct io_failure_record
*)(unsigned long)
1940 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1942 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1944 failure
->start
+ failure
->len
- 1,
1945 EXTENT_DIRTY
| EXTENT_LOCKED
,
1954 * when reads are done, we need to check csums to verify the data is correct
1955 * if there's a match, we allow the bio to finish. If not, we go through
1956 * the io_failure_record routines to find good copies
1958 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1959 struct extent_state
*state
)
1961 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1962 struct inode
*inode
= page
->mapping
->host
;
1963 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1965 u64
private = ~(u32
)0;
1967 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1970 if (PageChecked(page
)) {
1971 ClearPageChecked(page
);
1975 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1978 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1979 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1980 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1985 if (state
&& state
->start
== start
) {
1986 private = state
->private;
1989 ret
= get_state_private(io_tree
, start
, &private);
1991 kaddr
= kmap_atomic(page
, KM_USER0
);
1995 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1996 btrfs_csum_final(csum
, (char *)&csum
);
1997 if (csum
!= private)
2000 kunmap_atomic(kaddr
, KM_USER0
);
2002 /* if the io failure tree for this inode is non-empty,
2003 * check to see if we've recovered from a failed IO
2005 btrfs_clean_io_failures(inode
, start
);
2009 if (printk_ratelimit()) {
2010 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
2011 "private %llu\n", page
->mapping
->host
->i_ino
,
2012 (unsigned long long)start
, csum
,
2013 (unsigned long long)private);
2015 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2016 flush_dcache_page(page
);
2017 kunmap_atomic(kaddr
, KM_USER0
);
2023 struct delayed_iput
{
2024 struct list_head list
;
2025 struct inode
*inode
;
2028 void btrfs_add_delayed_iput(struct inode
*inode
)
2030 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2031 struct delayed_iput
*delayed
;
2033 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2036 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2037 delayed
->inode
= inode
;
2039 spin_lock(&fs_info
->delayed_iput_lock
);
2040 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2041 spin_unlock(&fs_info
->delayed_iput_lock
);
2044 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2047 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2048 struct delayed_iput
*delayed
;
2051 spin_lock(&fs_info
->delayed_iput_lock
);
2052 empty
= list_empty(&fs_info
->delayed_iputs
);
2053 spin_unlock(&fs_info
->delayed_iput_lock
);
2057 down_read(&root
->fs_info
->cleanup_work_sem
);
2058 spin_lock(&fs_info
->delayed_iput_lock
);
2059 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2060 spin_unlock(&fs_info
->delayed_iput_lock
);
2062 while (!list_empty(&list
)) {
2063 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2064 list_del(&delayed
->list
);
2065 iput(delayed
->inode
);
2068 up_read(&root
->fs_info
->cleanup_work_sem
);
2072 * calculate extra metadata reservation when snapshotting a subvolume
2073 * contains orphan files.
2075 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle
*trans
,
2076 struct btrfs_pending_snapshot
*pending
,
2077 u64
*bytes_to_reserve
)
2079 struct btrfs_root
*root
;
2080 struct btrfs_block_rsv
*block_rsv
;
2084 root
= pending
->root
;
2085 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2088 block_rsv
= root
->orphan_block_rsv
;
2090 /* orphan block reservation for the snapshot */
2091 num_bytes
= block_rsv
->size
;
2094 * after the snapshot is created, COWing tree blocks may use more
2095 * space than it frees. So we should make sure there is enough
2098 index
= trans
->transid
& 0x1;
2099 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2100 num_bytes
+= block_rsv
->size
-
2101 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2104 *bytes_to_reserve
+= num_bytes
;
2107 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle
*trans
,
2108 struct btrfs_pending_snapshot
*pending
)
2110 struct btrfs_root
*root
= pending
->root
;
2111 struct btrfs_root
*snap
= pending
->snap
;
2112 struct btrfs_block_rsv
*block_rsv
;
2117 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2120 /* refill source subvolume's orphan block reservation */
2121 block_rsv
= root
->orphan_block_rsv
;
2122 index
= trans
->transid
& 0x1;
2123 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2124 num_bytes
= block_rsv
->size
-
2125 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2126 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2127 root
->orphan_block_rsv
,
2132 /* setup orphan block reservation for the snapshot */
2133 block_rsv
= btrfs_alloc_block_rsv(snap
);
2136 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2137 snap
->orphan_block_rsv
= block_rsv
;
2139 num_bytes
= root
->orphan_block_rsv
->size
;
2140 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2141 block_rsv
, num_bytes
);
2145 /* insert orphan item for the snapshot */
2146 WARN_ON(!root
->orphan_item_inserted
);
2147 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2148 snap
->root_key
.objectid
);
2150 snap
->orphan_item_inserted
= 1;
2154 enum btrfs_orphan_cleanup_state
{
2155 ORPHAN_CLEANUP_STARTED
= 1,
2156 ORPHAN_CLEANUP_DONE
= 2,
2160 * This is called in transaction commmit time. If there are no orphan
2161 * files in the subvolume, it removes orphan item and frees block_rsv
2164 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2165 struct btrfs_root
*root
)
2169 if (!list_empty(&root
->orphan_list
) ||
2170 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2173 if (root
->orphan_item_inserted
&&
2174 btrfs_root_refs(&root
->root_item
) > 0) {
2175 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2176 root
->root_key
.objectid
);
2178 root
->orphan_item_inserted
= 0;
2181 if (root
->orphan_block_rsv
) {
2182 WARN_ON(root
->orphan_block_rsv
->size
> 0);
2183 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
2184 root
->orphan_block_rsv
= NULL
;
2189 * This creates an orphan entry for the given inode in case something goes
2190 * wrong in the middle of an unlink/truncate.
2192 * NOTE: caller of this function should reserve 5 units of metadata for
2195 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2197 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2198 struct btrfs_block_rsv
*block_rsv
= NULL
;
2203 if (!root
->orphan_block_rsv
) {
2204 block_rsv
= btrfs_alloc_block_rsv(root
);
2208 spin_lock(&root
->orphan_lock
);
2209 if (!root
->orphan_block_rsv
) {
2210 root
->orphan_block_rsv
= block_rsv
;
2211 } else if (block_rsv
) {
2212 btrfs_free_block_rsv(root
, block_rsv
);
2216 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2217 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2220 * For proper ENOSPC handling, we should do orphan
2221 * cleanup when mounting. But this introduces backward
2222 * compatibility issue.
2224 if (!xchg(&root
->orphan_item_inserted
, 1))
2232 if (!BTRFS_I(inode
)->orphan_meta_reserved
) {
2233 BTRFS_I(inode
)->orphan_meta_reserved
= 1;
2236 spin_unlock(&root
->orphan_lock
);
2239 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2241 /* grab metadata reservation from transaction handle */
2243 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2247 /* insert an orphan item to track this unlinked/truncated file */
2249 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
2253 /* insert an orphan item to track subvolume contains orphan files */
2255 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2256 root
->root_key
.objectid
);
2263 * We have done the truncate/delete so we can go ahead and remove the orphan
2264 * item for this particular inode.
2266 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2268 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2269 int delete_item
= 0;
2270 int release_rsv
= 0;
2273 spin_lock(&root
->orphan_lock
);
2274 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2275 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2279 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2280 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2283 spin_unlock(&root
->orphan_lock
);
2285 if (trans
&& delete_item
) {
2286 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
2291 btrfs_orphan_release_metadata(inode
);
2297 * this cleans up any orphans that may be left on the list from the last use
2300 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2302 struct btrfs_path
*path
;
2303 struct extent_buffer
*leaf
;
2304 struct btrfs_key key
, found_key
;
2305 struct btrfs_trans_handle
*trans
;
2306 struct inode
*inode
;
2307 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2309 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2312 path
= btrfs_alloc_path();
2319 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2320 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2321 key
.offset
= (u64
)-1;
2324 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2329 * if ret == 0 means we found what we were searching for, which
2330 * is weird, but possible, so only screw with path if we didn't
2331 * find the key and see if we have stuff that matches
2335 if (path
->slots
[0] == 0)
2340 /* pull out the item */
2341 leaf
= path
->nodes
[0];
2342 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2344 /* make sure the item matches what we want */
2345 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2347 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2350 /* release the path since we're done with it */
2351 btrfs_release_path(root
, path
);
2354 * this is where we are basically btrfs_lookup, without the
2355 * crossing root thing. we store the inode number in the
2356 * offset of the orphan item.
2358 found_key
.objectid
= found_key
.offset
;
2359 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2360 found_key
.offset
= 0;
2361 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2362 if (IS_ERR(inode
)) {
2363 ret
= PTR_ERR(inode
);
2368 * add this inode to the orphan list so btrfs_orphan_del does
2369 * the proper thing when we hit it
2371 spin_lock(&root
->orphan_lock
);
2372 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2373 spin_unlock(&root
->orphan_lock
);
2376 * if this is a bad inode, means we actually succeeded in
2377 * removing the inode, but not the orphan record, which means
2378 * we need to manually delete the orphan since iput will just
2379 * do a destroy_inode
2381 if (is_bad_inode(inode
)) {
2382 trans
= btrfs_start_transaction(root
, 0);
2383 if (IS_ERR(trans
)) {
2384 ret
= PTR_ERR(trans
);
2387 btrfs_orphan_del(trans
, inode
);
2388 btrfs_end_transaction(trans
, root
);
2393 /* if we have links, this was a truncate, lets do that */
2394 if (inode
->i_nlink
) {
2395 if (!S_ISREG(inode
->i_mode
)) {
2401 ret
= btrfs_truncate(inode
);
2406 /* this will do delete_inode and everything for us */
2411 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2413 if (root
->orphan_block_rsv
)
2414 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2417 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2418 trans
= btrfs_join_transaction(root
);
2420 btrfs_end_transaction(trans
, root
);
2424 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2426 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2430 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2431 btrfs_free_path(path
);
2436 * very simple check to peek ahead in the leaf looking for xattrs. If we
2437 * don't find any xattrs, we know there can't be any acls.
2439 * slot is the slot the inode is in, objectid is the objectid of the inode
2441 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2442 int slot
, u64 objectid
)
2444 u32 nritems
= btrfs_header_nritems(leaf
);
2445 struct btrfs_key found_key
;
2449 while (slot
< nritems
) {
2450 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2452 /* we found a different objectid, there must not be acls */
2453 if (found_key
.objectid
!= objectid
)
2456 /* we found an xattr, assume we've got an acl */
2457 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2461 * we found a key greater than an xattr key, there can't
2462 * be any acls later on
2464 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2471 * it goes inode, inode backrefs, xattrs, extents,
2472 * so if there are a ton of hard links to an inode there can
2473 * be a lot of backrefs. Don't waste time searching too hard,
2474 * this is just an optimization
2479 /* we hit the end of the leaf before we found an xattr or
2480 * something larger than an xattr. We have to assume the inode
2487 * read an inode from the btree into the in-memory inode
2489 static void btrfs_read_locked_inode(struct inode
*inode
)
2491 struct btrfs_path
*path
;
2492 struct extent_buffer
*leaf
;
2493 struct btrfs_inode_item
*inode_item
;
2494 struct btrfs_timespec
*tspec
;
2495 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2496 struct btrfs_key location
;
2498 u64 alloc_group_block
;
2502 path
= btrfs_alloc_path();
2504 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2506 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2510 leaf
= path
->nodes
[0];
2511 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2512 struct btrfs_inode_item
);
2514 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2515 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2516 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2517 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2518 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2520 tspec
= btrfs_inode_atime(inode_item
);
2521 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2522 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2524 tspec
= btrfs_inode_mtime(inode_item
);
2525 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2526 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2528 tspec
= btrfs_inode_ctime(inode_item
);
2529 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2530 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2532 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2533 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2534 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2535 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2537 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2539 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2540 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2542 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2545 * try to precache a NULL acl entry for files that don't have
2546 * any xattrs or acls
2548 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0], inode
->i_ino
);
2550 cache_no_acl(inode
);
2552 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2553 alloc_group_block
, 0);
2554 btrfs_free_path(path
);
2557 switch (inode
->i_mode
& S_IFMT
) {
2559 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2560 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2561 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2562 inode
->i_fop
= &btrfs_file_operations
;
2563 inode
->i_op
= &btrfs_file_inode_operations
;
2566 inode
->i_fop
= &btrfs_dir_file_operations
;
2567 if (root
== root
->fs_info
->tree_root
)
2568 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2570 inode
->i_op
= &btrfs_dir_inode_operations
;
2573 inode
->i_op
= &btrfs_symlink_inode_operations
;
2574 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2575 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2578 inode
->i_op
= &btrfs_special_inode_operations
;
2579 init_special_inode(inode
, inode
->i_mode
, rdev
);
2583 btrfs_update_iflags(inode
);
2587 btrfs_free_path(path
);
2588 make_bad_inode(inode
);
2592 * given a leaf and an inode, copy the inode fields into the leaf
2594 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2595 struct extent_buffer
*leaf
,
2596 struct btrfs_inode_item
*item
,
2597 struct inode
*inode
)
2599 if (!leaf
->map_token
)
2600 map_private_extent_buffer(leaf
, (unsigned long)item
,
2601 sizeof(struct btrfs_inode_item
),
2602 &leaf
->map_token
, &leaf
->kaddr
,
2603 &leaf
->map_start
, &leaf
->map_len
,
2606 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2607 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2608 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2609 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2610 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2612 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2613 inode
->i_atime
.tv_sec
);
2614 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2615 inode
->i_atime
.tv_nsec
);
2617 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2618 inode
->i_mtime
.tv_sec
);
2619 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2620 inode
->i_mtime
.tv_nsec
);
2622 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2623 inode
->i_ctime
.tv_sec
);
2624 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2625 inode
->i_ctime
.tv_nsec
);
2627 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2628 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2629 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2630 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2631 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2632 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2633 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2635 if (leaf
->map_token
) {
2636 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
2637 leaf
->map_token
= NULL
;
2642 * copy everything in the in-memory inode into the btree.
2644 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2645 struct btrfs_root
*root
, struct inode
*inode
)
2647 struct btrfs_inode_item
*inode_item
;
2648 struct btrfs_path
*path
;
2649 struct extent_buffer
*leaf
;
2652 path
= btrfs_alloc_path();
2654 path
->leave_spinning
= 1;
2655 ret
= btrfs_lookup_inode(trans
, root
, path
,
2656 &BTRFS_I(inode
)->location
, 1);
2663 btrfs_unlock_up_safe(path
, 1);
2664 leaf
= path
->nodes
[0];
2665 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2666 struct btrfs_inode_item
);
2668 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2669 btrfs_mark_buffer_dirty(leaf
);
2670 btrfs_set_inode_last_trans(trans
, inode
);
2673 btrfs_free_path(path
);
2679 * unlink helper that gets used here in inode.c and in the tree logging
2680 * recovery code. It remove a link in a directory with a given name, and
2681 * also drops the back refs in the inode to the directory
2683 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2684 struct btrfs_root
*root
,
2685 struct inode
*dir
, struct inode
*inode
,
2686 const char *name
, int name_len
)
2688 struct btrfs_path
*path
;
2690 struct extent_buffer
*leaf
;
2691 struct btrfs_dir_item
*di
;
2692 struct btrfs_key key
;
2695 path
= btrfs_alloc_path();
2701 path
->leave_spinning
= 1;
2702 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2703 name
, name_len
, -1);
2712 leaf
= path
->nodes
[0];
2713 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2714 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2717 btrfs_release_path(root
, path
);
2719 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2721 dir
->i_ino
, &index
);
2723 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2724 "inode %lu parent %lu\n", name_len
, name
,
2725 inode
->i_ino
, dir
->i_ino
);
2729 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2730 index
, name
, name_len
, -1);
2739 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2740 btrfs_release_path(root
, path
);
2742 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2744 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2746 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2751 btrfs_free_path(path
);
2755 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2756 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2757 btrfs_update_inode(trans
, root
, dir
);
2762 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
)
2768 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2770 btrfs_drop_nlink(inode
);
2771 ret
= btrfs_update_inode(trans
, root
, inode
);
2777 /* helper to check if there is any shared block in the path */
2778 static int check_path_shared(struct btrfs_root
*root
,
2779 struct btrfs_path
*path
)
2781 struct extent_buffer
*eb
;
2785 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2788 if (!path
->nodes
[level
])
2790 eb
= path
->nodes
[level
];
2791 if (!btrfs_block_can_be_shared(root
, eb
))
2793 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2802 * helper to start transaction for unlink and rmdir.
2804 * unlink and rmdir are special in btrfs, they do not always free space.
2805 * so in enospc case, we should make sure they will free space before
2806 * allowing them to use the global metadata reservation.
2808 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2809 struct dentry
*dentry
)
2811 struct btrfs_trans_handle
*trans
;
2812 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2813 struct btrfs_path
*path
;
2814 struct btrfs_inode_ref
*ref
;
2815 struct btrfs_dir_item
*di
;
2816 struct inode
*inode
= dentry
->d_inode
;
2822 trans
= btrfs_start_transaction(root
, 10);
2823 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2826 if (inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2827 return ERR_PTR(-ENOSPC
);
2829 /* check if there is someone else holds reference */
2830 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2831 return ERR_PTR(-ENOSPC
);
2833 if (atomic_read(&inode
->i_count
) > 2)
2834 return ERR_PTR(-ENOSPC
);
2836 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2837 return ERR_PTR(-ENOSPC
);
2839 path
= btrfs_alloc_path();
2841 root
->fs_info
->enospc_unlink
= 0;
2842 return ERR_PTR(-ENOMEM
);
2845 trans
= btrfs_start_transaction(root
, 0);
2846 if (IS_ERR(trans
)) {
2847 btrfs_free_path(path
);
2848 root
->fs_info
->enospc_unlink
= 0;
2852 path
->skip_locking
= 1;
2853 path
->search_commit_root
= 1;
2855 ret
= btrfs_lookup_inode(trans
, root
, path
,
2856 &BTRFS_I(dir
)->location
, 0);
2862 if (check_path_shared(root
, path
))
2867 btrfs_release_path(root
, path
);
2869 ret
= btrfs_lookup_inode(trans
, root
, path
,
2870 &BTRFS_I(inode
)->location
, 0);
2876 if (check_path_shared(root
, path
))
2881 btrfs_release_path(root
, path
);
2883 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2884 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2885 inode
->i_ino
, (u64
)-1, 0);
2891 if (check_path_shared(root
, path
))
2893 btrfs_release_path(root
, path
);
2901 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2902 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2908 if (check_path_shared(root
, path
))
2914 btrfs_release_path(root
, path
);
2916 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2917 dentry
->d_name
.name
, dentry
->d_name
.len
,
2918 inode
->i_ino
, dir
->i_ino
, 0);
2924 if (check_path_shared(root
, path
))
2926 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2927 btrfs_release_path(root
, path
);
2929 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
, index
,
2930 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2935 BUG_ON(ret
== -ENOENT
);
2936 if (check_path_shared(root
, path
))
2941 btrfs_free_path(path
);
2943 btrfs_end_transaction(trans
, root
);
2944 root
->fs_info
->enospc_unlink
= 0;
2945 return ERR_PTR(err
);
2948 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2952 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2953 struct btrfs_root
*root
)
2955 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2956 BUG_ON(!root
->fs_info
->enospc_unlink
);
2957 root
->fs_info
->enospc_unlink
= 0;
2959 btrfs_end_transaction_throttle(trans
, root
);
2962 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2964 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2965 struct btrfs_trans_handle
*trans
;
2966 struct inode
*inode
= dentry
->d_inode
;
2968 unsigned long nr
= 0;
2970 trans
= __unlink_start_trans(dir
, dentry
);
2972 return PTR_ERR(trans
);
2974 btrfs_set_trans_block_group(trans
, dir
);
2976 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2978 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2979 dentry
->d_name
.name
, dentry
->d_name
.len
);
2982 if (inode
->i_nlink
== 0) {
2983 ret
= btrfs_orphan_add(trans
, inode
);
2987 nr
= trans
->blocks_used
;
2988 __unlink_end_trans(trans
, root
);
2989 btrfs_btree_balance_dirty(root
, nr
);
2993 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2994 struct btrfs_root
*root
,
2995 struct inode
*dir
, u64 objectid
,
2996 const char *name
, int name_len
)
2998 struct btrfs_path
*path
;
2999 struct extent_buffer
*leaf
;
3000 struct btrfs_dir_item
*di
;
3001 struct btrfs_key key
;
3005 path
= btrfs_alloc_path();
3009 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
3010 name
, name_len
, -1);
3011 BUG_ON(!di
|| IS_ERR(di
));
3013 leaf
= path
->nodes
[0];
3014 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3015 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3016 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3018 btrfs_release_path(root
, path
);
3020 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3021 objectid
, root
->root_key
.objectid
,
3022 dir
->i_ino
, &index
, name
, name_len
);
3024 BUG_ON(ret
!= -ENOENT
);
3025 di
= btrfs_search_dir_index_item(root
, path
, dir
->i_ino
,
3027 BUG_ON(!di
|| IS_ERR(di
));
3029 leaf
= path
->nodes
[0];
3030 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3031 btrfs_release_path(root
, path
);
3035 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
3036 index
, name
, name_len
, -1);
3037 BUG_ON(!di
|| IS_ERR(di
));
3039 leaf
= path
->nodes
[0];
3040 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3041 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3042 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3044 btrfs_release_path(root
, path
);
3046 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3047 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3048 ret
= btrfs_update_inode(trans
, root
, dir
);
3051 btrfs_free_path(path
);
3055 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3057 struct inode
*inode
= dentry
->d_inode
;
3059 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3060 struct btrfs_trans_handle
*trans
;
3061 unsigned long nr
= 0;
3063 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3064 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
3067 trans
= __unlink_start_trans(dir
, dentry
);
3069 return PTR_ERR(trans
);
3071 btrfs_set_trans_block_group(trans
, dir
);
3073 if (unlikely(inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3074 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3075 BTRFS_I(inode
)->location
.objectid
,
3076 dentry
->d_name
.name
,
3077 dentry
->d_name
.len
);
3081 err
= btrfs_orphan_add(trans
, inode
);
3085 /* now the directory is empty */
3086 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3087 dentry
->d_name
.name
, dentry
->d_name
.len
);
3089 btrfs_i_size_write(inode
, 0);
3091 nr
= trans
->blocks_used
;
3092 __unlink_end_trans(trans
, root
);
3093 btrfs_btree_balance_dirty(root
, nr
);
3100 * when truncating bytes in a file, it is possible to avoid reading
3101 * the leaves that contain only checksum items. This can be the
3102 * majority of the IO required to delete a large file, but it must
3103 * be done carefully.
3105 * The keys in the level just above the leaves are checked to make sure
3106 * the lowest key in a given leaf is a csum key, and starts at an offset
3107 * after the new size.
3109 * Then the key for the next leaf is checked to make sure it also has
3110 * a checksum item for the same file. If it does, we know our target leaf
3111 * contains only checksum items, and it can be safely freed without reading
3114 * This is just an optimization targeted at large files. It may do
3115 * nothing. It will return 0 unless things went badly.
3117 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
3118 struct btrfs_root
*root
,
3119 struct btrfs_path
*path
,
3120 struct inode
*inode
, u64 new_size
)
3122 struct btrfs_key key
;
3125 struct btrfs_key found_key
;
3126 struct btrfs_key other_key
;
3127 struct btrfs_leaf_ref
*ref
;
3131 path
->lowest_level
= 1;
3132 key
.objectid
= inode
->i_ino
;
3133 key
.type
= BTRFS_CSUM_ITEM_KEY
;
3134 key
.offset
= new_size
;
3136 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3140 if (path
->nodes
[1] == NULL
) {
3145 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
3146 nritems
= btrfs_header_nritems(path
->nodes
[1]);
3151 if (path
->slots
[1] >= nritems
)
3154 /* did we find a key greater than anything we want to delete? */
3155 if (found_key
.objectid
> inode
->i_ino
||
3156 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
3159 /* we check the next key in the node to make sure the leave contains
3160 * only checksum items. This comparison doesn't work if our
3161 * leaf is the last one in the node
3163 if (path
->slots
[1] + 1 >= nritems
) {
3165 /* search forward from the last key in the node, this
3166 * will bring us into the next node in the tree
3168 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
3170 /* unlikely, but we inc below, so check to be safe */
3171 if (found_key
.offset
== (u64
)-1)
3174 /* search_forward needs a path with locks held, do the
3175 * search again for the original key. It is possible
3176 * this will race with a balance and return a path that
3177 * we could modify, but this drop is just an optimization
3178 * and is allowed to miss some leaves.
3180 btrfs_release_path(root
, path
);
3183 /* setup a max key for search_forward */
3184 other_key
.offset
= (u64
)-1;
3185 other_key
.type
= key
.type
;
3186 other_key
.objectid
= key
.objectid
;
3188 path
->keep_locks
= 1;
3189 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
3191 path
->keep_locks
= 0;
3192 if (ret
|| found_key
.objectid
!= key
.objectid
||
3193 found_key
.type
!= key
.type
) {
3198 key
.offset
= found_key
.offset
;
3199 btrfs_release_path(root
, path
);
3204 /* we know there's one more slot after us in the tree,
3205 * read that key so we can verify it is also a checksum item
3207 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
3209 if (found_key
.objectid
< inode
->i_ino
)
3212 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
3216 * if the key for the next leaf isn't a csum key from this objectid,
3217 * we can't be sure there aren't good items inside this leaf.
3220 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
3223 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
3224 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
3226 * it is safe to delete this leaf, it contains only
3227 * csum items from this inode at an offset >= new_size
3229 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
3232 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
3233 ref
= btrfs_alloc_leaf_ref(root
, 0);
3235 ref
->root_gen
= root
->root_key
.offset
;
3236 ref
->bytenr
= leaf_start
;
3238 ref
->generation
= leaf_gen
;
3241 btrfs_sort_leaf_ref(ref
);
3243 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
3245 btrfs_free_leaf_ref(root
, ref
);
3251 btrfs_release_path(root
, path
);
3253 if (other_key
.objectid
== inode
->i_ino
&&
3254 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
3255 key
.offset
= other_key
.offset
;
3261 /* fixup any changes we've made to the path */
3262 path
->lowest_level
= 0;
3263 path
->keep_locks
= 0;
3264 btrfs_release_path(root
, path
);
3271 * this can truncate away extent items, csum items and directory items.
3272 * It starts at a high offset and removes keys until it can't find
3273 * any higher than new_size
3275 * csum items that cross the new i_size are truncated to the new size
3278 * min_type is the minimum key type to truncate down to. If set to 0, this
3279 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3281 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3282 struct btrfs_root
*root
,
3283 struct inode
*inode
,
3284 u64 new_size
, u32 min_type
)
3286 struct btrfs_path
*path
;
3287 struct extent_buffer
*leaf
;
3288 struct btrfs_file_extent_item
*fi
;
3289 struct btrfs_key key
;
3290 struct btrfs_key found_key
;
3291 u64 extent_start
= 0;
3292 u64 extent_num_bytes
= 0;
3293 u64 extent_offset
= 0;
3295 u64 mask
= root
->sectorsize
- 1;
3296 u32 found_type
= (u8
)-1;
3299 int pending_del_nr
= 0;
3300 int pending_del_slot
= 0;
3301 int extent_type
= -1;
3306 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3308 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3309 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3311 path
= btrfs_alloc_path();
3315 key
.objectid
= inode
->i_ino
;
3316 key
.offset
= (u64
)-1;
3320 path
->leave_spinning
= 1;
3321 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3328 /* there are no items in the tree for us to truncate, we're
3331 if (path
->slots
[0] == 0)
3338 leaf
= path
->nodes
[0];
3339 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3340 found_type
= btrfs_key_type(&found_key
);
3343 if (found_key
.objectid
!= inode
->i_ino
)
3346 if (found_type
< min_type
)
3349 item_end
= found_key
.offset
;
3350 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3351 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3352 struct btrfs_file_extent_item
);
3353 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3354 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3355 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3356 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3358 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3360 btrfs_file_extent_num_bytes(leaf
, fi
);
3361 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3362 item_end
+= btrfs_file_extent_inline_len(leaf
,
3367 if (found_type
> min_type
) {
3370 if (item_end
< new_size
)
3372 if (found_key
.offset
>= new_size
)
3378 /* FIXME, shrink the extent if the ref count is only 1 */
3379 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3382 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3384 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3385 if (!del_item
&& !encoding
) {
3386 u64 orig_num_bytes
=
3387 btrfs_file_extent_num_bytes(leaf
, fi
);
3388 extent_num_bytes
= new_size
-
3389 found_key
.offset
+ root
->sectorsize
- 1;
3390 extent_num_bytes
= extent_num_bytes
&
3391 ~((u64
)root
->sectorsize
- 1);
3392 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3394 num_dec
= (orig_num_bytes
-
3396 if (root
->ref_cows
&& extent_start
!= 0)
3397 inode_sub_bytes(inode
, num_dec
);
3398 btrfs_mark_buffer_dirty(leaf
);
3401 btrfs_file_extent_disk_num_bytes(leaf
,
3403 extent_offset
= found_key
.offset
-
3404 btrfs_file_extent_offset(leaf
, fi
);
3406 /* FIXME blocksize != 4096 */
3407 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3408 if (extent_start
!= 0) {
3411 inode_sub_bytes(inode
, num_dec
);
3414 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3416 * we can't truncate inline items that have had
3420 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3421 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3422 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3423 u32 size
= new_size
- found_key
.offset
;
3425 if (root
->ref_cows
) {
3426 inode_sub_bytes(inode
, item_end
+ 1 -
3430 btrfs_file_extent_calc_inline_size(size
);
3431 ret
= btrfs_truncate_item(trans
, root
, path
,
3434 } else if (root
->ref_cows
) {
3435 inode_sub_bytes(inode
, item_end
+ 1 -
3441 if (!pending_del_nr
) {
3442 /* no pending yet, add ourselves */
3443 pending_del_slot
= path
->slots
[0];
3445 } else if (pending_del_nr
&&
3446 path
->slots
[0] + 1 == pending_del_slot
) {
3447 /* hop on the pending chunk */
3449 pending_del_slot
= path
->slots
[0];
3456 if (found_extent
&& (root
->ref_cows
||
3457 root
== root
->fs_info
->tree_root
)) {
3458 btrfs_set_path_blocking(path
);
3459 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3460 extent_num_bytes
, 0,
3461 btrfs_header_owner(leaf
),
3462 inode
->i_ino
, extent_offset
);
3466 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3469 if (path
->slots
[0] == 0 ||
3470 path
->slots
[0] != pending_del_slot
) {
3471 if (root
->ref_cows
) {
3475 if (pending_del_nr
) {
3476 ret
= btrfs_del_items(trans
, root
, path
,
3482 btrfs_release_path(root
, path
);
3489 if (pending_del_nr
) {
3490 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3494 btrfs_free_path(path
);
3499 * taken from block_truncate_page, but does cow as it zeros out
3500 * any bytes left in the last page in the file.
3502 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3504 struct inode
*inode
= mapping
->host
;
3505 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3506 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3507 struct btrfs_ordered_extent
*ordered
;
3508 struct extent_state
*cached_state
= NULL
;
3510 u32 blocksize
= root
->sectorsize
;
3511 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3512 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3518 if ((offset
& (blocksize
- 1)) == 0)
3520 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3526 page
= grab_cache_page(mapping
, index
);
3528 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3532 page_start
= page_offset(page
);
3533 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3535 if (!PageUptodate(page
)) {
3536 ret
= btrfs_readpage(NULL
, page
);
3538 if (page
->mapping
!= mapping
) {
3540 page_cache_release(page
);
3543 if (!PageUptodate(page
)) {
3548 wait_on_page_writeback(page
);
3550 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3552 set_page_extent_mapped(page
);
3554 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3556 unlock_extent_cached(io_tree
, page_start
, page_end
,
3557 &cached_state
, GFP_NOFS
);
3559 page_cache_release(page
);
3560 btrfs_start_ordered_extent(inode
, ordered
, 1);
3561 btrfs_put_ordered_extent(ordered
);
3565 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3566 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3567 0, 0, &cached_state
, GFP_NOFS
);
3569 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3572 unlock_extent_cached(io_tree
, page_start
, page_end
,
3573 &cached_state
, GFP_NOFS
);
3578 if (offset
!= PAGE_CACHE_SIZE
) {
3580 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3581 flush_dcache_page(page
);
3584 ClearPageChecked(page
);
3585 set_page_dirty(page
);
3586 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3591 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3593 page_cache_release(page
);
3599 * This function puts in dummy file extents for the area we're creating a hole
3600 * for. So if we are truncating this file to a larger size we need to insert
3601 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3602 * the range between oldsize and size
3604 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3606 struct btrfs_trans_handle
*trans
;
3607 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3608 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3609 struct extent_map
*em
= NULL
;
3610 struct extent_state
*cached_state
= NULL
;
3611 u64 mask
= root
->sectorsize
- 1;
3612 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3613 u64 block_end
= (size
+ mask
) & ~mask
;
3619 if (size
<= hole_start
)
3623 struct btrfs_ordered_extent
*ordered
;
3624 btrfs_wait_ordered_range(inode
, hole_start
,
3625 block_end
- hole_start
);
3626 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3627 &cached_state
, GFP_NOFS
);
3628 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3631 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3632 &cached_state
, GFP_NOFS
);
3633 btrfs_put_ordered_extent(ordered
);
3636 cur_offset
= hole_start
;
3638 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3639 block_end
- cur_offset
, 0);
3640 BUG_ON(IS_ERR(em
) || !em
);
3641 last_byte
= min(extent_map_end(em
), block_end
);
3642 last_byte
= (last_byte
+ mask
) & ~mask
;
3643 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3645 hole_size
= last_byte
- cur_offset
;
3647 trans
= btrfs_start_transaction(root
, 2);
3648 if (IS_ERR(trans
)) {
3649 err
= PTR_ERR(trans
);
3652 btrfs_set_trans_block_group(trans
, inode
);
3654 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3655 cur_offset
+ hole_size
,
3660 err
= btrfs_insert_file_extent(trans
, root
,
3661 inode
->i_ino
, cur_offset
, 0,
3662 0, hole_size
, 0, hole_size
,
3667 btrfs_drop_extent_cache(inode
, hole_start
,
3670 btrfs_end_transaction(trans
, root
);
3672 free_extent_map(em
);
3674 cur_offset
= last_byte
;
3675 if (cur_offset
>= block_end
)
3679 free_extent_map(em
);
3680 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3685 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3687 loff_t oldsize
= i_size_read(inode
);
3690 if (newsize
== oldsize
)
3693 if (newsize
> oldsize
) {
3694 i_size_write(inode
, newsize
);
3695 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3696 truncate_pagecache(inode
, oldsize
, newsize
);
3697 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3699 btrfs_setsize(inode
, oldsize
);
3703 mark_inode_dirty(inode
);
3707 * We're truncating a file that used to have good data down to
3708 * zero. Make sure it gets into the ordered flush list so that
3709 * any new writes get down to disk quickly.
3712 BTRFS_I(inode
)->ordered_data_close
= 1;
3714 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3715 truncate_setsize(inode
, newsize
);
3716 ret
= btrfs_truncate(inode
);
3722 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3724 struct inode
*inode
= dentry
->d_inode
;
3725 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3728 if (btrfs_root_readonly(root
))
3731 err
= inode_change_ok(inode
, attr
);
3735 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3736 err
= btrfs_setsize(inode
, attr
->ia_size
);
3741 if (attr
->ia_valid
) {
3742 setattr_copy(inode
, attr
);
3743 mark_inode_dirty(inode
);
3745 if (attr
->ia_valid
& ATTR_MODE
)
3746 err
= btrfs_acl_chmod(inode
);
3752 void btrfs_evict_inode(struct inode
*inode
)
3754 struct btrfs_trans_handle
*trans
;
3755 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3759 trace_btrfs_inode_evict(inode
);
3761 truncate_inode_pages(&inode
->i_data
, 0);
3762 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3763 root
== root
->fs_info
->tree_root
))
3766 if (is_bad_inode(inode
)) {
3767 btrfs_orphan_del(NULL
, inode
);
3770 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3771 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3773 if (root
->fs_info
->log_root_recovering
) {
3774 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3778 if (inode
->i_nlink
> 0) {
3779 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3783 btrfs_i_size_write(inode
, 0);
3786 trans
= btrfs_start_transaction(root
, 0);
3787 BUG_ON(IS_ERR(trans
));
3788 btrfs_set_trans_block_group(trans
, inode
);
3789 trans
->block_rsv
= root
->orphan_block_rsv
;
3791 ret
= btrfs_block_rsv_check(trans
, root
,
3792 root
->orphan_block_rsv
, 0, 5);
3794 BUG_ON(ret
!= -EAGAIN
);
3795 ret
= btrfs_commit_transaction(trans
, root
);
3800 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3804 nr
= trans
->blocks_used
;
3805 btrfs_end_transaction(trans
, root
);
3807 btrfs_btree_balance_dirty(root
, nr
);
3812 ret
= btrfs_orphan_del(trans
, inode
);
3816 nr
= trans
->blocks_used
;
3817 btrfs_end_transaction(trans
, root
);
3818 btrfs_btree_balance_dirty(root
, nr
);
3820 end_writeback(inode
);
3825 * this returns the key found in the dir entry in the location pointer.
3826 * If no dir entries were found, location->objectid is 0.
3828 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3829 struct btrfs_key
*location
)
3831 const char *name
= dentry
->d_name
.name
;
3832 int namelen
= dentry
->d_name
.len
;
3833 struct btrfs_dir_item
*di
;
3834 struct btrfs_path
*path
;
3835 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3838 path
= btrfs_alloc_path();
3841 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
3846 if (!di
|| IS_ERR(di
))
3849 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3851 btrfs_free_path(path
);
3854 location
->objectid
= 0;
3859 * when we hit a tree root in a directory, the btrfs part of the inode
3860 * needs to be changed to reflect the root directory of the tree root. This
3861 * is kind of like crossing a mount point.
3863 static int fixup_tree_root_location(struct btrfs_root
*root
,
3865 struct dentry
*dentry
,
3866 struct btrfs_key
*location
,
3867 struct btrfs_root
**sub_root
)
3869 struct btrfs_path
*path
;
3870 struct btrfs_root
*new_root
;
3871 struct btrfs_root_ref
*ref
;
3872 struct extent_buffer
*leaf
;
3876 path
= btrfs_alloc_path();
3883 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3884 BTRFS_I(dir
)->root
->root_key
.objectid
,
3885 location
->objectid
);
3892 leaf
= path
->nodes
[0];
3893 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3894 if (btrfs_root_ref_dirid(leaf
, ref
) != dir
->i_ino
||
3895 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3898 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3899 (unsigned long)(ref
+ 1),
3900 dentry
->d_name
.len
);
3904 btrfs_release_path(root
->fs_info
->tree_root
, path
);
3906 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3907 if (IS_ERR(new_root
)) {
3908 err
= PTR_ERR(new_root
);
3912 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3917 *sub_root
= new_root
;
3918 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3919 location
->type
= BTRFS_INODE_ITEM_KEY
;
3920 location
->offset
= 0;
3923 btrfs_free_path(path
);
3927 static void inode_tree_add(struct inode
*inode
)
3929 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3930 struct btrfs_inode
*entry
;
3932 struct rb_node
*parent
;
3934 p
= &root
->inode_tree
.rb_node
;
3937 if (inode_unhashed(inode
))
3940 spin_lock(&root
->inode_lock
);
3943 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3945 if (inode
->i_ino
< entry
->vfs_inode
.i_ino
)
3946 p
= &parent
->rb_left
;
3947 else if (inode
->i_ino
> entry
->vfs_inode
.i_ino
)
3948 p
= &parent
->rb_right
;
3950 WARN_ON(!(entry
->vfs_inode
.i_state
&
3951 (I_WILL_FREE
| I_FREEING
)));
3952 rb_erase(parent
, &root
->inode_tree
);
3953 RB_CLEAR_NODE(parent
);
3954 spin_unlock(&root
->inode_lock
);
3958 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3959 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3960 spin_unlock(&root
->inode_lock
);
3963 static void inode_tree_del(struct inode
*inode
)
3965 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3968 spin_lock(&root
->inode_lock
);
3969 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3970 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3971 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3972 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3974 spin_unlock(&root
->inode_lock
);
3977 * Free space cache has inodes in the tree root, but the tree root has a
3978 * root_refs of 0, so this could end up dropping the tree root as a
3979 * snapshot, so we need the extra !root->fs_info->tree_root check to
3980 * make sure we don't drop it.
3982 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3983 root
!= root
->fs_info
->tree_root
) {
3984 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3985 spin_lock(&root
->inode_lock
);
3986 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3987 spin_unlock(&root
->inode_lock
);
3989 btrfs_add_dead_root(root
);
3993 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3995 struct rb_node
*node
;
3996 struct rb_node
*prev
;
3997 struct btrfs_inode
*entry
;
3998 struct inode
*inode
;
4001 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4003 spin_lock(&root
->inode_lock
);
4005 node
= root
->inode_tree
.rb_node
;
4009 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4011 if (objectid
< entry
->vfs_inode
.i_ino
)
4012 node
= node
->rb_left
;
4013 else if (objectid
> entry
->vfs_inode
.i_ino
)
4014 node
= node
->rb_right
;
4020 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4021 if (objectid
<= entry
->vfs_inode
.i_ino
) {
4025 prev
= rb_next(prev
);
4029 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4030 objectid
= entry
->vfs_inode
.i_ino
+ 1;
4031 inode
= igrab(&entry
->vfs_inode
);
4033 spin_unlock(&root
->inode_lock
);
4034 if (atomic_read(&inode
->i_count
) > 1)
4035 d_prune_aliases(inode
);
4037 * btrfs_drop_inode will have it removed from
4038 * the inode cache when its usage count
4043 spin_lock(&root
->inode_lock
);
4047 if (cond_resched_lock(&root
->inode_lock
))
4050 node
= rb_next(node
);
4052 spin_unlock(&root
->inode_lock
);
4056 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4058 struct btrfs_iget_args
*args
= p
;
4059 inode
->i_ino
= args
->ino
;
4060 BTRFS_I(inode
)->root
= args
->root
;
4061 btrfs_set_inode_space_info(args
->root
, inode
);
4065 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4067 struct btrfs_iget_args
*args
= opaque
;
4068 return args
->ino
== inode
->i_ino
&&
4069 args
->root
== BTRFS_I(inode
)->root
;
4072 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4074 struct btrfs_root
*root
)
4076 struct inode
*inode
;
4077 struct btrfs_iget_args args
;
4078 args
.ino
= objectid
;
4081 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4082 btrfs_init_locked_inode
,
4087 /* Get an inode object given its location and corresponding root.
4088 * Returns in *is_new if the inode was read from disk
4090 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4091 struct btrfs_root
*root
, int *new)
4093 struct inode
*inode
;
4095 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4097 return ERR_PTR(-ENOMEM
);
4099 if (inode
->i_state
& I_NEW
) {
4100 BTRFS_I(inode
)->root
= root
;
4101 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4102 btrfs_read_locked_inode(inode
);
4103 inode_tree_add(inode
);
4104 unlock_new_inode(inode
);
4112 static struct inode
*new_simple_dir(struct super_block
*s
,
4113 struct btrfs_key
*key
,
4114 struct btrfs_root
*root
)
4116 struct inode
*inode
= new_inode(s
);
4119 return ERR_PTR(-ENOMEM
);
4121 BTRFS_I(inode
)->root
= root
;
4122 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4123 BTRFS_I(inode
)->dummy_inode
= 1;
4125 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4126 inode
->i_op
= &simple_dir_inode_operations
;
4127 inode
->i_fop
= &simple_dir_operations
;
4128 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4129 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4134 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4136 struct inode
*inode
;
4137 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4138 struct btrfs_root
*sub_root
= root
;
4139 struct btrfs_key location
;
4143 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4144 return ERR_PTR(-ENAMETOOLONG
);
4146 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4149 return ERR_PTR(ret
);
4151 if (location
.objectid
== 0)
4154 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4155 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4159 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4161 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4162 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4163 &location
, &sub_root
);
4166 inode
= ERR_PTR(ret
);
4168 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4170 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4172 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4174 if (!IS_ERR(inode
) && root
!= sub_root
) {
4175 down_read(&root
->fs_info
->cleanup_work_sem
);
4176 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4177 ret
= btrfs_orphan_cleanup(sub_root
);
4178 up_read(&root
->fs_info
->cleanup_work_sem
);
4180 inode
= ERR_PTR(ret
);
4186 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4188 struct btrfs_root
*root
;
4190 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
4191 dentry
= dentry
->d_parent
;
4193 if (dentry
->d_inode
) {
4194 root
= BTRFS_I(dentry
->d_inode
)->root
;
4195 if (btrfs_root_refs(&root
->root_item
) == 0)
4201 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4202 struct nameidata
*nd
)
4204 struct inode
*inode
;
4206 inode
= btrfs_lookup_dentry(dir
, dentry
);
4208 return ERR_CAST(inode
);
4210 return d_splice_alias(inode
, dentry
);
4213 static unsigned char btrfs_filetype_table
[] = {
4214 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4217 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4220 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4221 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4222 struct btrfs_item
*item
;
4223 struct btrfs_dir_item
*di
;
4224 struct btrfs_key key
;
4225 struct btrfs_key found_key
;
4226 struct btrfs_path
*path
;
4228 struct extent_buffer
*leaf
;
4230 unsigned char d_type
;
4235 int key_type
= BTRFS_DIR_INDEX_KEY
;
4240 /* FIXME, use a real flag for deciding about the key type */
4241 if (root
->fs_info
->tree_root
== root
)
4242 key_type
= BTRFS_DIR_ITEM_KEY
;
4244 /* special case for "." */
4245 if (filp
->f_pos
== 0) {
4246 over
= filldir(dirent
, ".", 1,
4253 /* special case for .., just use the back ref */
4254 if (filp
->f_pos
== 1) {
4255 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4256 over
= filldir(dirent
, "..", 2,
4262 path
= btrfs_alloc_path();
4265 btrfs_set_key_type(&key
, key_type
);
4266 key
.offset
= filp
->f_pos
;
4267 key
.objectid
= inode
->i_ino
;
4269 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4274 leaf
= path
->nodes
[0];
4275 slot
= path
->slots
[0];
4276 if (slot
>= btrfs_header_nritems(leaf
)) {
4277 ret
= btrfs_next_leaf(root
, path
);
4285 item
= btrfs_item_nr(leaf
, slot
);
4286 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4288 if (found_key
.objectid
!= key
.objectid
)
4290 if (btrfs_key_type(&found_key
) != key_type
)
4292 if (found_key
.offset
< filp
->f_pos
)
4295 filp
->f_pos
= found_key
.offset
;
4297 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4299 di_total
= btrfs_item_size(leaf
, item
);
4301 while (di_cur
< di_total
) {
4302 struct btrfs_key location
;
4304 if (verify_dir_item(root
, leaf
, di
))
4307 name_len
= btrfs_dir_name_len(leaf
, di
);
4308 if (name_len
<= sizeof(tmp_name
)) {
4309 name_ptr
= tmp_name
;
4311 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4317 read_extent_buffer(leaf
, name_ptr
,
4318 (unsigned long)(di
+ 1), name_len
);
4320 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4321 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4323 /* is this a reference to our own snapshot? If so
4326 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4327 location
.objectid
== root
->root_key
.objectid
) {
4331 over
= filldir(dirent
, name_ptr
, name_len
,
4332 found_key
.offset
, location
.objectid
,
4336 if (name_ptr
!= tmp_name
)
4341 di_len
= btrfs_dir_name_len(leaf
, di
) +
4342 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4344 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4350 /* Reached end of directory/root. Bump pos past the last item. */
4351 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4353 * 32-bit glibc will use getdents64, but then strtol -
4354 * so the last number we can serve is this.
4356 filp
->f_pos
= 0x7fffffff;
4362 btrfs_free_path(path
);
4366 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4368 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4369 struct btrfs_trans_handle
*trans
;
4371 bool nolock
= false;
4373 if (BTRFS_I(inode
)->dummy_inode
)
4377 nolock
= (root
->fs_info
->closing
&& root
== root
->fs_info
->tree_root
);
4379 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4381 trans
= btrfs_join_transaction_nolock(root
);
4383 trans
= btrfs_join_transaction(root
);
4385 return PTR_ERR(trans
);
4386 btrfs_set_trans_block_group(trans
, inode
);
4388 ret
= btrfs_end_transaction_nolock(trans
, root
);
4390 ret
= btrfs_commit_transaction(trans
, root
);
4396 * This is somewhat expensive, updating the tree every time the
4397 * inode changes. But, it is most likely to find the inode in cache.
4398 * FIXME, needs more benchmarking...there are no reasons other than performance
4399 * to keep or drop this code.
4401 void btrfs_dirty_inode(struct inode
*inode
)
4403 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4404 struct btrfs_trans_handle
*trans
;
4407 if (BTRFS_I(inode
)->dummy_inode
)
4410 trans
= btrfs_join_transaction(root
);
4411 BUG_ON(IS_ERR(trans
));
4412 btrfs_set_trans_block_group(trans
, inode
);
4414 ret
= btrfs_update_inode(trans
, root
, inode
);
4415 if (ret
&& ret
== -ENOSPC
) {
4416 /* whoops, lets try again with the full transaction */
4417 btrfs_end_transaction(trans
, root
);
4418 trans
= btrfs_start_transaction(root
, 1);
4419 if (IS_ERR(trans
)) {
4420 if (printk_ratelimit()) {
4421 printk(KERN_ERR
"btrfs: fail to "
4422 "dirty inode %lu error %ld\n",
4423 inode
->i_ino
, PTR_ERR(trans
));
4427 btrfs_set_trans_block_group(trans
, inode
);
4429 ret
= btrfs_update_inode(trans
, root
, inode
);
4431 if (printk_ratelimit()) {
4432 printk(KERN_ERR
"btrfs: fail to "
4433 "dirty inode %lu error %d\n",
4438 btrfs_end_transaction(trans
, root
);
4442 * find the highest existing sequence number in a directory
4443 * and then set the in-memory index_cnt variable to reflect
4444 * free sequence numbers
4446 static int btrfs_set_inode_index_count(struct inode
*inode
)
4448 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4449 struct btrfs_key key
, found_key
;
4450 struct btrfs_path
*path
;
4451 struct extent_buffer
*leaf
;
4454 key
.objectid
= inode
->i_ino
;
4455 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4456 key
.offset
= (u64
)-1;
4458 path
= btrfs_alloc_path();
4462 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4465 /* FIXME: we should be able to handle this */
4471 * MAGIC NUMBER EXPLANATION:
4472 * since we search a directory based on f_pos we have to start at 2
4473 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4474 * else has to start at 2
4476 if (path
->slots
[0] == 0) {
4477 BTRFS_I(inode
)->index_cnt
= 2;
4483 leaf
= path
->nodes
[0];
4484 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4486 if (found_key
.objectid
!= inode
->i_ino
||
4487 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4488 BTRFS_I(inode
)->index_cnt
= 2;
4492 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4494 btrfs_free_path(path
);
4499 * helper to find a free sequence number in a given directory. This current
4500 * code is very simple, later versions will do smarter things in the btree
4502 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4506 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4507 ret
= btrfs_set_inode_index_count(dir
);
4512 *index
= BTRFS_I(dir
)->index_cnt
;
4513 BTRFS_I(dir
)->index_cnt
++;
4518 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4519 struct btrfs_root
*root
,
4521 const char *name
, int name_len
,
4522 u64 ref_objectid
, u64 objectid
,
4523 u64 alloc_hint
, int mode
, u64
*index
)
4525 struct inode
*inode
;
4526 struct btrfs_inode_item
*inode_item
;
4527 struct btrfs_key
*location
;
4528 struct btrfs_path
*path
;
4529 struct btrfs_inode_ref
*ref
;
4530 struct btrfs_key key
[2];
4536 path
= btrfs_alloc_path();
4539 inode
= new_inode(root
->fs_info
->sb
);
4541 btrfs_free_path(path
);
4542 return ERR_PTR(-ENOMEM
);
4546 trace_btrfs_inode_request(dir
);
4548 ret
= btrfs_set_inode_index(dir
, index
);
4550 btrfs_free_path(path
);
4552 return ERR_PTR(ret
);
4556 * index_cnt is ignored for everything but a dir,
4557 * btrfs_get_inode_index_count has an explanation for the magic
4560 BTRFS_I(inode
)->index_cnt
= 2;
4561 BTRFS_I(inode
)->root
= root
;
4562 BTRFS_I(inode
)->generation
= trans
->transid
;
4563 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4564 btrfs_set_inode_space_info(root
, inode
);
4570 BTRFS_I(inode
)->block_group
=
4571 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
4573 key
[0].objectid
= objectid
;
4574 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4577 key
[1].objectid
= objectid
;
4578 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4579 key
[1].offset
= ref_objectid
;
4581 sizes
[0] = sizeof(struct btrfs_inode_item
);
4582 sizes
[1] = name_len
+ sizeof(*ref
);
4584 path
->leave_spinning
= 1;
4585 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4589 inode_init_owner(inode
, dir
, mode
);
4590 inode
->i_ino
= objectid
;
4591 inode_set_bytes(inode
, 0);
4592 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4593 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4594 struct btrfs_inode_item
);
4595 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4597 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4598 struct btrfs_inode_ref
);
4599 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4600 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4601 ptr
= (unsigned long)(ref
+ 1);
4602 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4604 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4605 btrfs_free_path(path
);
4607 location
= &BTRFS_I(inode
)->location
;
4608 location
->objectid
= objectid
;
4609 location
->offset
= 0;
4610 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4612 btrfs_inherit_iflags(inode
, dir
);
4614 if ((mode
& S_IFREG
)) {
4615 if (btrfs_test_opt(root
, NODATASUM
))
4616 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4617 if (btrfs_test_opt(root
, NODATACOW
) ||
4618 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4619 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4622 insert_inode_hash(inode
);
4623 inode_tree_add(inode
);
4625 trace_btrfs_inode_new(inode
);
4630 BTRFS_I(dir
)->index_cnt
--;
4631 btrfs_free_path(path
);
4633 return ERR_PTR(ret
);
4636 static inline u8
btrfs_inode_type(struct inode
*inode
)
4638 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4642 * utility function to add 'inode' into 'parent_inode' with
4643 * a give name and a given sequence number.
4644 * if 'add_backref' is true, also insert a backref from the
4645 * inode to the parent directory.
4647 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4648 struct inode
*parent_inode
, struct inode
*inode
,
4649 const char *name
, int name_len
, int add_backref
, u64 index
)
4652 struct btrfs_key key
;
4653 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4655 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4656 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4658 key
.objectid
= inode
->i_ino
;
4659 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4663 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4664 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4665 key
.objectid
, root
->root_key
.objectid
,
4666 parent_inode
->i_ino
,
4667 index
, name
, name_len
);
4668 } else if (add_backref
) {
4669 ret
= btrfs_insert_inode_ref(trans
, root
,
4670 name
, name_len
, inode
->i_ino
,
4671 parent_inode
->i_ino
, index
);
4675 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4676 parent_inode
->i_ino
, &key
,
4677 btrfs_inode_type(inode
), index
);
4680 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4682 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4683 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4688 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4689 struct inode
*dir
, struct dentry
*dentry
,
4690 struct inode
*inode
, int backref
, u64 index
)
4692 int err
= btrfs_add_link(trans
, dir
, inode
,
4693 dentry
->d_name
.name
, dentry
->d_name
.len
,
4696 d_instantiate(dentry
, inode
);
4704 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4705 int mode
, dev_t rdev
)
4707 struct btrfs_trans_handle
*trans
;
4708 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4709 struct inode
*inode
= NULL
;
4713 unsigned long nr
= 0;
4716 if (!new_valid_dev(rdev
))
4719 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4724 * 2 for inode item and ref
4726 * 1 for xattr if selinux is on
4728 trans
= btrfs_start_transaction(root
, 5);
4730 return PTR_ERR(trans
);
4732 btrfs_set_trans_block_group(trans
, dir
);
4734 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4735 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4736 BTRFS_I(dir
)->block_group
, mode
, &index
);
4737 if (IS_ERR(inode
)) {
4738 err
= PTR_ERR(inode
);
4742 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4748 btrfs_set_trans_block_group(trans
, inode
);
4749 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4753 inode
->i_op
= &btrfs_special_inode_operations
;
4754 init_special_inode(inode
, inode
->i_mode
, rdev
);
4755 btrfs_update_inode(trans
, root
, inode
);
4757 btrfs_update_inode_block_group(trans
, inode
);
4758 btrfs_update_inode_block_group(trans
, dir
);
4760 nr
= trans
->blocks_used
;
4761 btrfs_end_transaction_throttle(trans
, root
);
4762 btrfs_btree_balance_dirty(root
, nr
);
4764 inode_dec_link_count(inode
);
4770 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4771 int mode
, struct nameidata
*nd
)
4773 struct btrfs_trans_handle
*trans
;
4774 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4775 struct inode
*inode
= NULL
;
4778 unsigned long nr
= 0;
4782 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4786 * 2 for inode item and ref
4788 * 1 for xattr if selinux is on
4790 trans
= btrfs_start_transaction(root
, 5);
4792 return PTR_ERR(trans
);
4794 btrfs_set_trans_block_group(trans
, dir
);
4796 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4797 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4798 BTRFS_I(dir
)->block_group
, mode
, &index
);
4799 if (IS_ERR(inode
)) {
4800 err
= PTR_ERR(inode
);
4804 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4810 btrfs_set_trans_block_group(trans
, inode
);
4811 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4815 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4816 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4817 inode
->i_fop
= &btrfs_file_operations
;
4818 inode
->i_op
= &btrfs_file_inode_operations
;
4819 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4821 btrfs_update_inode_block_group(trans
, inode
);
4822 btrfs_update_inode_block_group(trans
, dir
);
4824 nr
= trans
->blocks_used
;
4825 btrfs_end_transaction_throttle(trans
, root
);
4827 inode_dec_link_count(inode
);
4830 btrfs_btree_balance_dirty(root
, nr
);
4834 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4835 struct dentry
*dentry
)
4837 struct btrfs_trans_handle
*trans
;
4838 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4839 struct inode
*inode
= old_dentry
->d_inode
;
4841 unsigned long nr
= 0;
4845 /* do not allow sys_link's with other subvols of the same device */
4846 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4849 if (inode
->i_nlink
== ~0U)
4852 err
= btrfs_set_inode_index(dir
, &index
);
4857 * 2 items for inode and inode ref
4858 * 2 items for dir items
4859 * 1 item for parent inode
4861 trans
= btrfs_start_transaction(root
, 5);
4862 if (IS_ERR(trans
)) {
4863 err
= PTR_ERR(trans
);
4867 btrfs_inc_nlink(inode
);
4868 inode
->i_ctime
= CURRENT_TIME
;
4870 btrfs_set_trans_block_group(trans
, dir
);
4873 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4878 struct dentry
*parent
= dget_parent(dentry
);
4879 btrfs_update_inode_block_group(trans
, dir
);
4880 err
= btrfs_update_inode(trans
, root
, inode
);
4882 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4886 nr
= trans
->blocks_used
;
4887 btrfs_end_transaction_throttle(trans
, root
);
4890 inode_dec_link_count(inode
);
4893 btrfs_btree_balance_dirty(root
, nr
);
4897 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4899 struct inode
*inode
= NULL
;
4900 struct btrfs_trans_handle
*trans
;
4901 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4903 int drop_on_err
= 0;
4906 unsigned long nr
= 1;
4908 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4913 * 2 items for inode and ref
4914 * 2 items for dir items
4915 * 1 for xattr if selinux is on
4917 trans
= btrfs_start_transaction(root
, 5);
4919 return PTR_ERR(trans
);
4920 btrfs_set_trans_block_group(trans
, dir
);
4922 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4923 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4924 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
4926 if (IS_ERR(inode
)) {
4927 err
= PTR_ERR(inode
);
4933 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4937 inode
->i_op
= &btrfs_dir_inode_operations
;
4938 inode
->i_fop
= &btrfs_dir_file_operations
;
4939 btrfs_set_trans_block_group(trans
, inode
);
4941 btrfs_i_size_write(inode
, 0);
4942 err
= btrfs_update_inode(trans
, root
, inode
);
4946 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4947 dentry
->d_name
.len
, 0, index
);
4951 d_instantiate(dentry
, inode
);
4953 btrfs_update_inode_block_group(trans
, inode
);
4954 btrfs_update_inode_block_group(trans
, dir
);
4957 nr
= trans
->blocks_used
;
4958 btrfs_end_transaction_throttle(trans
, root
);
4961 btrfs_btree_balance_dirty(root
, nr
);
4965 /* helper for btfs_get_extent. Given an existing extent in the tree,
4966 * and an extent that you want to insert, deal with overlap and insert
4967 * the new extent into the tree.
4969 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4970 struct extent_map
*existing
,
4971 struct extent_map
*em
,
4972 u64 map_start
, u64 map_len
)
4976 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4977 start_diff
= map_start
- em
->start
;
4978 em
->start
= map_start
;
4980 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4981 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4982 em
->block_start
+= start_diff
;
4983 em
->block_len
-= start_diff
;
4985 return add_extent_mapping(em_tree
, em
);
4988 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4989 struct inode
*inode
, struct page
*page
,
4990 size_t pg_offset
, u64 extent_offset
,
4991 struct btrfs_file_extent_item
*item
)
4994 struct extent_buffer
*leaf
= path
->nodes
[0];
4997 unsigned long inline_size
;
5001 WARN_ON(pg_offset
!= 0);
5002 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5003 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
5004 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
5005 btrfs_item_nr(leaf
, path
->slots
[0]));
5006 tmp
= kmalloc(inline_size
, GFP_NOFS
);
5009 ptr
= btrfs_file_extent_inline_start(item
);
5011 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
5013 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
5014 ret
= btrfs_decompress(compress_type
, tmp
, page
,
5015 extent_offset
, inline_size
, max_size
);
5017 char *kaddr
= kmap_atomic(page
, KM_USER0
);
5018 unsigned long copy_size
= min_t(u64
,
5019 PAGE_CACHE_SIZE
- pg_offset
,
5020 max_size
- extent_offset
);
5021 memset(kaddr
+ pg_offset
, 0, copy_size
);
5022 kunmap_atomic(kaddr
, KM_USER0
);
5029 * a bit scary, this does extent mapping from logical file offset to the disk.
5030 * the ugly parts come from merging extents from the disk with the in-ram
5031 * representation. This gets more complex because of the data=ordered code,
5032 * where the in-ram extents might be locked pending data=ordered completion.
5034 * This also copies inline extents directly into the page.
5037 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
5038 size_t pg_offset
, u64 start
, u64 len
,
5044 u64 extent_start
= 0;
5046 u64 objectid
= inode
->i_ino
;
5048 struct btrfs_path
*path
= NULL
;
5049 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5050 struct btrfs_file_extent_item
*item
;
5051 struct extent_buffer
*leaf
;
5052 struct btrfs_key found_key
;
5053 struct extent_map
*em
= NULL
;
5054 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5055 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5056 struct btrfs_trans_handle
*trans
= NULL
;
5060 read_lock(&em_tree
->lock
);
5061 em
= lookup_extent_mapping(em_tree
, start
, len
);
5063 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5064 read_unlock(&em_tree
->lock
);
5067 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5068 free_extent_map(em
);
5069 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5070 free_extent_map(em
);
5074 em
= alloc_extent_map(GFP_NOFS
);
5079 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5080 em
->start
= EXTENT_MAP_HOLE
;
5081 em
->orig_start
= EXTENT_MAP_HOLE
;
5083 em
->block_len
= (u64
)-1;
5086 path
= btrfs_alloc_path();
5090 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5091 objectid
, start
, trans
!= NULL
);
5098 if (path
->slots
[0] == 0)
5103 leaf
= path
->nodes
[0];
5104 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5105 struct btrfs_file_extent_item
);
5106 /* are we inside the extent that was found? */
5107 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5108 found_type
= btrfs_key_type(&found_key
);
5109 if (found_key
.objectid
!= objectid
||
5110 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5114 found_type
= btrfs_file_extent_type(leaf
, item
);
5115 extent_start
= found_key
.offset
;
5116 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5117 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5118 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5119 extent_end
= extent_start
+
5120 btrfs_file_extent_num_bytes(leaf
, item
);
5121 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5123 size
= btrfs_file_extent_inline_len(leaf
, item
);
5124 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5125 ~((u64
)root
->sectorsize
- 1);
5128 if (start
>= extent_end
) {
5130 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5131 ret
= btrfs_next_leaf(root
, path
);
5138 leaf
= path
->nodes
[0];
5140 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5141 if (found_key
.objectid
!= objectid
||
5142 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5144 if (start
+ len
<= found_key
.offset
)
5147 em
->len
= found_key
.offset
- start
;
5151 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5152 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5153 em
->start
= extent_start
;
5154 em
->len
= extent_end
- extent_start
;
5155 em
->orig_start
= extent_start
-
5156 btrfs_file_extent_offset(leaf
, item
);
5157 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5159 em
->block_start
= EXTENT_MAP_HOLE
;
5162 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5163 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5164 em
->compress_type
= compress_type
;
5165 em
->block_start
= bytenr
;
5166 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5169 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5170 em
->block_start
= bytenr
;
5171 em
->block_len
= em
->len
;
5172 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5173 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5176 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5180 size_t extent_offset
;
5183 em
->block_start
= EXTENT_MAP_INLINE
;
5184 if (!page
|| create
) {
5185 em
->start
= extent_start
;
5186 em
->len
= extent_end
- extent_start
;
5190 size
= btrfs_file_extent_inline_len(leaf
, item
);
5191 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5192 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5193 size
- extent_offset
);
5194 em
->start
= extent_start
+ extent_offset
;
5195 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5196 ~((u64
)root
->sectorsize
- 1);
5197 em
->orig_start
= EXTENT_MAP_INLINE
;
5198 if (compress_type
) {
5199 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5200 em
->compress_type
= compress_type
;
5202 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5203 if (create
== 0 && !PageUptodate(page
)) {
5204 if (btrfs_file_extent_compression(leaf
, item
) !=
5205 BTRFS_COMPRESS_NONE
) {
5206 ret
= uncompress_inline(path
, inode
, page
,
5208 extent_offset
, item
);
5212 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5214 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5215 memset(map
+ pg_offset
+ copy_size
, 0,
5216 PAGE_CACHE_SIZE
- pg_offset
-
5221 flush_dcache_page(page
);
5222 } else if (create
&& PageUptodate(page
)) {
5226 free_extent_map(em
);
5228 btrfs_release_path(root
, path
);
5229 trans
= btrfs_join_transaction(root
);
5231 return ERR_CAST(trans
);
5235 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5238 btrfs_mark_buffer_dirty(leaf
);
5240 set_extent_uptodate(io_tree
, em
->start
,
5241 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5244 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5251 em
->block_start
= EXTENT_MAP_HOLE
;
5252 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5254 btrfs_release_path(root
, path
);
5255 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5256 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5257 "[%llu %llu]\n", (unsigned long long)em
->start
,
5258 (unsigned long long)em
->len
,
5259 (unsigned long long)start
,
5260 (unsigned long long)len
);
5266 write_lock(&em_tree
->lock
);
5267 ret
= add_extent_mapping(em_tree
, em
);
5268 /* it is possible that someone inserted the extent into the tree
5269 * while we had the lock dropped. It is also possible that
5270 * an overlapping map exists in the tree
5272 if (ret
== -EEXIST
) {
5273 struct extent_map
*existing
;
5277 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5278 if (existing
&& (existing
->start
> start
||
5279 existing
->start
+ existing
->len
<= start
)) {
5280 free_extent_map(existing
);
5284 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5287 err
= merge_extent_mapping(em_tree
, existing
,
5290 free_extent_map(existing
);
5292 free_extent_map(em
);
5297 free_extent_map(em
);
5301 free_extent_map(em
);
5306 write_unlock(&em_tree
->lock
);
5309 trace_btrfs_get_extent(root
, em
);
5312 btrfs_free_path(path
);
5314 ret
= btrfs_end_transaction(trans
, root
);
5319 free_extent_map(em
);
5320 return ERR_PTR(err
);
5325 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5326 size_t pg_offset
, u64 start
, u64 len
,
5329 struct extent_map
*em
;
5330 struct extent_map
*hole_em
= NULL
;
5331 u64 range_start
= start
;
5337 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5342 * if our em maps to a hole, there might
5343 * actually be delalloc bytes behind it
5345 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5351 /* check to see if we've wrapped (len == -1 or similar) */
5360 /* ok, we didn't find anything, lets look for delalloc */
5361 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5362 end
, len
, EXTENT_DELALLOC
, 1);
5363 found_end
= range_start
+ found
;
5364 if (found_end
< range_start
)
5365 found_end
= (u64
)-1;
5368 * we didn't find anything useful, return
5369 * the original results from get_extent()
5371 if (range_start
> end
|| found_end
<= start
) {
5377 /* adjust the range_start to make sure it doesn't
5378 * go backwards from the start they passed in
5380 range_start
= max(start
,range_start
);
5381 found
= found_end
- range_start
;
5384 u64 hole_start
= start
;
5387 em
= alloc_extent_map(GFP_NOFS
);
5393 * when btrfs_get_extent can't find anything it
5394 * returns one huge hole
5396 * make sure what it found really fits our range, and
5397 * adjust to make sure it is based on the start from
5401 u64 calc_end
= extent_map_end(hole_em
);
5403 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5404 free_extent_map(hole_em
);
5407 hole_start
= max(hole_em
->start
, start
);
5408 hole_len
= calc_end
- hole_start
;
5412 if (hole_em
&& range_start
> hole_start
) {
5413 /* our hole starts before our delalloc, so we
5414 * have to return just the parts of the hole
5415 * that go until the delalloc starts
5417 em
->len
= min(hole_len
,
5418 range_start
- hole_start
);
5419 em
->start
= hole_start
;
5420 em
->orig_start
= hole_start
;
5422 * don't adjust block start at all,
5423 * it is fixed at EXTENT_MAP_HOLE
5425 em
->block_start
= hole_em
->block_start
;
5426 em
->block_len
= hole_len
;
5428 em
->start
= range_start
;
5430 em
->orig_start
= range_start
;
5431 em
->block_start
= EXTENT_MAP_DELALLOC
;
5432 em
->block_len
= found
;
5434 } else if (hole_em
) {
5439 free_extent_map(hole_em
);
5441 free_extent_map(em
);
5442 return ERR_PTR(err
);
5447 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5448 struct extent_map
*em
,
5451 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5452 struct btrfs_trans_handle
*trans
;
5453 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5454 struct btrfs_key ins
;
5457 bool insert
= false;
5460 * Ok if the extent map we looked up is a hole and is for the exact
5461 * range we want, there is no reason to allocate a new one, however if
5462 * it is not right then we need to free this one and drop the cache for
5465 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5467 free_extent_map(em
);
5470 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5473 trans
= btrfs_join_transaction(root
);
5475 return ERR_CAST(trans
);
5477 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5479 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5480 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5481 alloc_hint
, (u64
)-1, &ins
, 1);
5488 em
= alloc_extent_map(GFP_NOFS
);
5490 em
= ERR_PTR(-ENOMEM
);
5496 em
->orig_start
= em
->start
;
5497 em
->len
= ins
.offset
;
5499 em
->block_start
= ins
.objectid
;
5500 em
->block_len
= ins
.offset
;
5501 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5504 * We need to do this because if we're using the original em we searched
5505 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5508 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5511 write_lock(&em_tree
->lock
);
5512 ret
= add_extent_mapping(em_tree
, em
);
5513 write_unlock(&em_tree
->lock
);
5516 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5519 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5520 ins
.offset
, ins
.offset
, 0);
5522 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5526 btrfs_end_transaction(trans
, root
);
5531 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5532 * block must be cow'd
5534 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5535 struct inode
*inode
, u64 offset
, u64 len
)
5537 struct btrfs_path
*path
;
5539 struct extent_buffer
*leaf
;
5540 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5541 struct btrfs_file_extent_item
*fi
;
5542 struct btrfs_key key
;
5550 path
= btrfs_alloc_path();
5554 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
5559 slot
= path
->slots
[0];
5562 /* can't find the item, must cow */
5569 leaf
= path
->nodes
[0];
5570 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5571 if (key
.objectid
!= inode
->i_ino
||
5572 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5573 /* not our file or wrong item type, must cow */
5577 if (key
.offset
> offset
) {
5578 /* Wrong offset, must cow */
5582 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5583 found_type
= btrfs_file_extent_type(leaf
, fi
);
5584 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5585 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5586 /* not a regular extent, must cow */
5589 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5590 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5592 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5593 if (extent_end
< offset
+ len
) {
5594 /* extent doesn't include our full range, must cow */
5598 if (btrfs_extent_readonly(root
, disk_bytenr
))
5602 * look for other files referencing this extent, if we
5603 * find any we must cow
5605 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
5606 key
.offset
- backref_offset
, disk_bytenr
))
5610 * adjust disk_bytenr and num_bytes to cover just the bytes
5611 * in this extent we are about to write. If there
5612 * are any csums in that range we have to cow in order
5613 * to keep the csums correct
5615 disk_bytenr
+= backref_offset
;
5616 disk_bytenr
+= offset
- key
.offset
;
5617 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5618 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5621 * all of the above have passed, it is safe to overwrite this extent
5626 btrfs_free_path(path
);
5630 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5631 struct buffer_head
*bh_result
, int create
)
5633 struct extent_map
*em
;
5634 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5635 u64 start
= iblock
<< inode
->i_blkbits
;
5636 u64 len
= bh_result
->b_size
;
5637 struct btrfs_trans_handle
*trans
;
5639 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5644 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5645 * io. INLINE is special, and we could probably kludge it in here, but
5646 * it's still buffered so for safety lets just fall back to the generic
5649 * For COMPRESSED we _have_ to read the entire extent in so we can
5650 * decompress it, so there will be buffering required no matter what we
5651 * do, so go ahead and fallback to buffered.
5653 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5654 * to buffered IO. Don't blame me, this is the price we pay for using
5657 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5658 em
->block_start
== EXTENT_MAP_INLINE
) {
5659 free_extent_map(em
);
5663 /* Just a good old fashioned hole, return */
5664 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5665 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5666 free_extent_map(em
);
5667 /* DIO will do one hole at a time, so just unlock a sector */
5668 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5669 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5674 * We don't allocate a new extent in the following cases
5676 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5678 * 2) The extent is marked as PREALLOC. We're good to go here and can
5679 * just use the extent.
5683 len
= em
->len
- (start
- em
->start
);
5687 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5688 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5689 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5694 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5695 type
= BTRFS_ORDERED_PREALLOC
;
5697 type
= BTRFS_ORDERED_NOCOW
;
5698 len
= min(len
, em
->len
- (start
- em
->start
));
5699 block_start
= em
->block_start
+ (start
- em
->start
);
5702 * we're not going to log anything, but we do need
5703 * to make sure the current transaction stays open
5704 * while we look for nocow cross refs
5706 trans
= btrfs_join_transaction(root
);
5710 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5711 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5712 block_start
, len
, len
, type
);
5713 btrfs_end_transaction(trans
, root
);
5715 free_extent_map(em
);
5720 btrfs_end_transaction(trans
, root
);
5724 * this will cow the extent, reset the len in case we changed
5727 len
= bh_result
->b_size
;
5728 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
5731 len
= min(len
, em
->len
- (start
- em
->start
));
5733 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5734 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5737 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5739 bh_result
->b_size
= len
;
5740 bh_result
->b_bdev
= em
->bdev
;
5741 set_buffer_mapped(bh_result
);
5742 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5743 set_buffer_new(bh_result
);
5745 free_extent_map(em
);
5750 struct btrfs_dio_private
{
5751 struct inode
*inode
;
5758 /* number of bios pending for this dio */
5759 atomic_t pending_bios
;
5764 struct bio
*orig_bio
;
5767 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5769 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5770 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5771 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5772 struct inode
*inode
= dip
->inode
;
5773 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5775 u32
*private = dip
->csums
;
5777 start
= dip
->logical_offset
;
5779 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5780 struct page
*page
= bvec
->bv_page
;
5783 unsigned long flags
;
5785 local_irq_save(flags
);
5786 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5787 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5788 csum
, bvec
->bv_len
);
5789 btrfs_csum_final(csum
, (char *)&csum
);
5790 kunmap_atomic(kaddr
, KM_IRQ0
);
5791 local_irq_restore(flags
);
5793 flush_dcache_page(bvec
->bv_page
);
5794 if (csum
!= *private) {
5795 printk(KERN_ERR
"btrfs csum failed ino %lu off"
5796 " %llu csum %u private %u\n",
5797 inode
->i_ino
, (unsigned long long)start
,
5803 start
+= bvec
->bv_len
;
5806 } while (bvec
<= bvec_end
);
5808 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5809 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5810 bio
->bi_private
= dip
->private;
5815 /* If we had a csum failure make sure to clear the uptodate flag */
5817 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5818 dio_end_io(bio
, err
);
5821 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5823 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5824 struct inode
*inode
= dip
->inode
;
5825 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5826 struct btrfs_trans_handle
*trans
;
5827 struct btrfs_ordered_extent
*ordered
= NULL
;
5828 struct extent_state
*cached_state
= NULL
;
5829 u64 ordered_offset
= dip
->logical_offset
;
5830 u64 ordered_bytes
= dip
->bytes
;
5836 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5844 trans
= btrfs_join_transaction(root
);
5845 if (IS_ERR(trans
)) {
5849 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5851 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5852 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5854 ret
= btrfs_update_inode(trans
, root
, inode
);
5859 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5860 ordered
->file_offset
+ ordered
->len
- 1, 0,
5861 &cached_state
, GFP_NOFS
);
5863 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5864 ret
= btrfs_mark_extent_written(trans
, inode
,
5865 ordered
->file_offset
,
5866 ordered
->file_offset
+
5873 ret
= insert_reserved_file_extent(trans
, inode
,
5874 ordered
->file_offset
,
5880 BTRFS_FILE_EXTENT_REG
);
5881 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5882 ordered
->file_offset
, ordered
->len
);
5890 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5891 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5893 btrfs_update_inode(trans
, root
, inode
);
5896 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5897 ordered
->file_offset
+ ordered
->len
- 1,
5898 &cached_state
, GFP_NOFS
);
5900 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5901 btrfs_end_transaction(trans
, root
);
5902 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5903 btrfs_put_ordered_extent(ordered
);
5904 btrfs_put_ordered_extent(ordered
);
5908 * our bio might span multiple ordered extents. If we haven't
5909 * completed the accounting for the whole dio, go back and try again
5911 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5912 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5917 bio
->bi_private
= dip
->private;
5922 /* If we had an error make sure to clear the uptodate flag */
5924 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5925 dio_end_io(bio
, err
);
5928 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5929 struct bio
*bio
, int mirror_num
,
5930 unsigned long bio_flags
, u64 offset
)
5933 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5934 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5939 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5941 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5944 printk(KERN_ERR
"btrfs direct IO failed ino %lu rw %lu "
5945 "sector %#Lx len %u err no %d\n",
5946 dip
->inode
->i_ino
, bio
->bi_rw
,
5947 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5951 * before atomic variable goto zero, we must make sure
5952 * dip->errors is perceived to be set.
5954 smp_mb__before_atomic_dec();
5957 /* if there are more bios still pending for this dio, just exit */
5958 if (!atomic_dec_and_test(&dip
->pending_bios
))
5962 bio_io_error(dip
->orig_bio
);
5964 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5965 bio_endio(dip
->orig_bio
, 0);
5971 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5972 u64 first_sector
, gfp_t gfp_flags
)
5974 int nr_vecs
= bio_get_nr_vecs(bdev
);
5975 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5978 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5979 int rw
, u64 file_offset
, int skip_sum
,
5980 u32
*csums
, int async_submit
)
5982 int write
= rw
& REQ_WRITE
;
5983 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5987 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5994 if (write
&& async_submit
) {
5995 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5996 inode
, rw
, bio
, 0, 0,
5998 __btrfs_submit_bio_start_direct_io
,
5999 __btrfs_submit_bio_done
);
6003 * If we aren't doing async submit, calculate the csum of the
6006 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
6009 } else if (!skip_sum
) {
6010 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
6011 file_offset
, csums
);
6017 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
6023 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
6026 struct inode
*inode
= dip
->inode
;
6027 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6028 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
6030 struct bio
*orig_bio
= dip
->orig_bio
;
6031 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
6032 u64 start_sector
= orig_bio
->bi_sector
;
6033 u64 file_offset
= dip
->logical_offset
;
6037 u32
*csums
= dip
->csums
;
6039 int async_submit
= 0;
6040 int write
= rw
& REQ_WRITE
;
6042 map_length
= orig_bio
->bi_size
;
6043 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6044 &map_length
, NULL
, 0);
6050 if (map_length
>= orig_bio
->bi_size
) {
6056 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
6059 bio
->bi_private
= dip
;
6060 bio
->bi_end_io
= btrfs_end_dio_bio
;
6061 atomic_inc(&dip
->pending_bios
);
6063 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
6064 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
6065 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
6066 bvec
->bv_offset
) < bvec
->bv_len
)) {
6068 * inc the count before we submit the bio so
6069 * we know the end IO handler won't happen before
6070 * we inc the count. Otherwise, the dip might get freed
6071 * before we're done setting it up
6073 atomic_inc(&dip
->pending_bios
);
6074 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
6075 file_offset
, skip_sum
,
6076 csums
, async_submit
);
6079 atomic_dec(&dip
->pending_bios
);
6083 /* Write's use the ordered csums */
6084 if (!write
&& !skip_sum
)
6085 csums
= csums
+ nr_pages
;
6086 start_sector
+= submit_len
>> 9;
6087 file_offset
+= submit_len
;
6092 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6093 start_sector
, GFP_NOFS
);
6096 bio
->bi_private
= dip
;
6097 bio
->bi_end_io
= btrfs_end_dio_bio
;
6099 map_length
= orig_bio
->bi_size
;
6100 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6101 &map_length
, NULL
, 0);
6107 submit_len
+= bvec
->bv_len
;
6114 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6115 csums
, async_submit
);
6123 * before atomic variable goto zero, we must
6124 * make sure dip->errors is perceived to be set.
6126 smp_mb__before_atomic_dec();
6127 if (atomic_dec_and_test(&dip
->pending_bios
))
6128 bio_io_error(dip
->orig_bio
);
6130 /* bio_end_io() will handle error, so we needn't return it */
6134 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6137 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6138 struct btrfs_dio_private
*dip
;
6139 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6141 int write
= rw
& REQ_WRITE
;
6144 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6146 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6153 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6154 if (!write
&& !skip_sum
) {
6155 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6163 dip
->private = bio
->bi_private
;
6165 dip
->logical_offset
= file_offset
;
6169 dip
->bytes
+= bvec
->bv_len
;
6171 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6173 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6174 bio
->bi_private
= dip
;
6176 dip
->orig_bio
= bio
;
6177 atomic_set(&dip
->pending_bios
, 0);
6180 bio
->bi_end_io
= btrfs_endio_direct_write
;
6182 bio
->bi_end_io
= btrfs_endio_direct_read
;
6184 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6189 * If this is a write, we need to clean up the reserved space and kill
6190 * the ordered extent.
6193 struct btrfs_ordered_extent
*ordered
;
6194 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6195 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6196 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6197 btrfs_free_reserved_extent(root
, ordered
->start
,
6199 btrfs_put_ordered_extent(ordered
);
6200 btrfs_put_ordered_extent(ordered
);
6202 bio_endio(bio
, ret
);
6205 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6206 const struct iovec
*iov
, loff_t offset
,
6207 unsigned long nr_segs
)
6213 unsigned blocksize_mask
= root
->sectorsize
- 1;
6214 ssize_t retval
= -EINVAL
;
6215 loff_t end
= offset
;
6217 if (offset
& blocksize_mask
)
6220 /* Check the memory alignment. Blocks cannot straddle pages */
6221 for (seg
= 0; seg
< nr_segs
; seg
++) {
6222 addr
= (unsigned long)iov
[seg
].iov_base
;
6223 size
= iov
[seg
].iov_len
;
6225 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6228 /* If this is a write we don't need to check anymore */
6233 * Check to make sure we don't have duplicate iov_base's in this
6234 * iovec, if so return EINVAL, otherwise we'll get csum errors
6235 * when reading back.
6237 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6238 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6246 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6247 const struct iovec
*iov
, loff_t offset
,
6248 unsigned long nr_segs
)
6250 struct file
*file
= iocb
->ki_filp
;
6251 struct inode
*inode
= file
->f_mapping
->host
;
6252 struct btrfs_ordered_extent
*ordered
;
6253 struct extent_state
*cached_state
= NULL
;
6254 u64 lockstart
, lockend
;
6256 int writing
= rw
& WRITE
;
6258 size_t count
= iov_length(iov
, nr_segs
);
6260 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6266 lockend
= offset
+ count
- 1;
6269 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6275 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6276 0, &cached_state
, GFP_NOFS
);
6278 * We're concerned with the entire range that we're going to be
6279 * doing DIO to, so we need to make sure theres no ordered
6280 * extents in this range.
6282 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6283 lockend
- lockstart
+ 1);
6286 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6287 &cached_state
, GFP_NOFS
);
6288 btrfs_start_ordered_extent(inode
, ordered
, 1);
6289 btrfs_put_ordered_extent(ordered
);
6294 * we don't use btrfs_set_extent_delalloc because we don't want
6295 * the dirty or uptodate bits
6298 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6299 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6300 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6303 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6304 lockend
, EXTENT_LOCKED
| write_bits
,
6305 1, 0, &cached_state
, GFP_NOFS
);
6310 free_extent_state(cached_state
);
6311 cached_state
= NULL
;
6313 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6314 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6315 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6316 btrfs_submit_direct
, 0);
6318 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6319 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6320 offset
+ iov_length(iov
, nr_segs
) - 1,
6321 EXTENT_LOCKED
| write_bits
, 1, 0,
6322 &cached_state
, GFP_NOFS
);
6323 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6325 * We're falling back to buffered, unlock the section we didn't
6328 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6329 offset
+ iov_length(iov
, nr_segs
) - 1,
6330 EXTENT_LOCKED
| write_bits
, 1, 0,
6331 &cached_state
, GFP_NOFS
);
6334 free_extent_state(cached_state
);
6338 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6339 __u64 start
, __u64 len
)
6341 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6344 int btrfs_readpage(struct file
*file
, struct page
*page
)
6346 struct extent_io_tree
*tree
;
6347 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6348 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6351 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6353 struct extent_io_tree
*tree
;
6356 if (current
->flags
& PF_MEMALLOC
) {
6357 redirty_page_for_writepage(wbc
, page
);
6361 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6362 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6365 int btrfs_writepages(struct address_space
*mapping
,
6366 struct writeback_control
*wbc
)
6368 struct extent_io_tree
*tree
;
6370 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6371 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6375 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6376 struct list_head
*pages
, unsigned nr_pages
)
6378 struct extent_io_tree
*tree
;
6379 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6380 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6383 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6385 struct extent_io_tree
*tree
;
6386 struct extent_map_tree
*map
;
6389 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6390 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6391 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6393 ClearPagePrivate(page
);
6394 set_page_private(page
, 0);
6395 page_cache_release(page
);
6400 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6402 if (PageWriteback(page
) || PageDirty(page
))
6404 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6407 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6409 struct extent_io_tree
*tree
;
6410 struct btrfs_ordered_extent
*ordered
;
6411 struct extent_state
*cached_state
= NULL
;
6412 u64 page_start
= page_offset(page
);
6413 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6417 * we have the page locked, so new writeback can't start,
6418 * and the dirty bit won't be cleared while we are here.
6420 * Wait for IO on this page so that we can safely clear
6421 * the PagePrivate2 bit and do ordered accounting
6423 wait_on_page_writeback(page
);
6425 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6427 btrfs_releasepage(page
, GFP_NOFS
);
6430 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6432 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6436 * IO on this page will never be started, so we need
6437 * to account for any ordered extents now
6439 clear_extent_bit(tree
, page_start
, page_end
,
6440 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6441 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6442 &cached_state
, GFP_NOFS
);
6444 * whoever cleared the private bit is responsible
6445 * for the finish_ordered_io
6447 if (TestClearPagePrivate2(page
)) {
6448 btrfs_finish_ordered_io(page
->mapping
->host
,
6449 page_start
, page_end
);
6451 btrfs_put_ordered_extent(ordered
);
6452 cached_state
= NULL
;
6453 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6456 clear_extent_bit(tree
, page_start
, page_end
,
6457 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6458 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6459 __btrfs_releasepage(page
, GFP_NOFS
);
6461 ClearPageChecked(page
);
6462 if (PagePrivate(page
)) {
6463 ClearPagePrivate(page
);
6464 set_page_private(page
, 0);
6465 page_cache_release(page
);
6470 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6471 * called from a page fault handler when a page is first dirtied. Hence we must
6472 * be careful to check for EOF conditions here. We set the page up correctly
6473 * for a written page which means we get ENOSPC checking when writing into
6474 * holes and correct delalloc and unwritten extent mapping on filesystems that
6475 * support these features.
6477 * We are not allowed to take the i_mutex here so we have to play games to
6478 * protect against truncate races as the page could now be beyond EOF. Because
6479 * vmtruncate() writes the inode size before removing pages, once we have the
6480 * page lock we can determine safely if the page is beyond EOF. If it is not
6481 * beyond EOF, then the page is guaranteed safe against truncation until we
6484 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6486 struct page
*page
= vmf
->page
;
6487 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6488 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6489 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6490 struct btrfs_ordered_extent
*ordered
;
6491 struct extent_state
*cached_state
= NULL
;
6493 unsigned long zero_start
;
6499 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6503 else /* -ENOSPC, -EIO, etc */
6504 ret
= VM_FAULT_SIGBUS
;
6508 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6511 size
= i_size_read(inode
);
6512 page_start
= page_offset(page
);
6513 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6515 if ((page
->mapping
!= inode
->i_mapping
) ||
6516 (page_start
>= size
)) {
6517 /* page got truncated out from underneath us */
6520 wait_on_page_writeback(page
);
6522 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6524 set_page_extent_mapped(page
);
6527 * we can't set the delalloc bits if there are pending ordered
6528 * extents. Drop our locks and wait for them to finish
6530 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6532 unlock_extent_cached(io_tree
, page_start
, page_end
,
6533 &cached_state
, GFP_NOFS
);
6535 btrfs_start_ordered_extent(inode
, ordered
, 1);
6536 btrfs_put_ordered_extent(ordered
);
6541 * XXX - page_mkwrite gets called every time the page is dirtied, even
6542 * if it was already dirty, so for space accounting reasons we need to
6543 * clear any delalloc bits for the range we are fixing to save. There
6544 * is probably a better way to do this, but for now keep consistent with
6545 * prepare_pages in the normal write path.
6547 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6548 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6549 0, 0, &cached_state
, GFP_NOFS
);
6551 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6554 unlock_extent_cached(io_tree
, page_start
, page_end
,
6555 &cached_state
, GFP_NOFS
);
6556 ret
= VM_FAULT_SIGBUS
;
6561 /* page is wholly or partially inside EOF */
6562 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6563 zero_start
= size
& ~PAGE_CACHE_MASK
;
6565 zero_start
= PAGE_CACHE_SIZE
;
6567 if (zero_start
!= PAGE_CACHE_SIZE
) {
6569 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6570 flush_dcache_page(page
);
6573 ClearPageChecked(page
);
6574 set_page_dirty(page
);
6575 SetPageUptodate(page
);
6577 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6578 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6580 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6584 return VM_FAULT_LOCKED
;
6586 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6591 static int btrfs_truncate(struct inode
*inode
)
6593 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6596 struct btrfs_trans_handle
*trans
;
6598 u64 mask
= root
->sectorsize
- 1;
6600 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6604 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6605 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6607 trans
= btrfs_start_transaction(root
, 5);
6609 return PTR_ERR(trans
);
6611 btrfs_set_trans_block_group(trans
, inode
);
6613 ret
= btrfs_orphan_add(trans
, inode
);
6615 btrfs_end_transaction(trans
, root
);
6619 nr
= trans
->blocks_used
;
6620 btrfs_end_transaction(trans
, root
);
6621 btrfs_btree_balance_dirty(root
, nr
);
6623 /* Now start a transaction for the truncate */
6624 trans
= btrfs_start_transaction(root
, 0);
6626 return PTR_ERR(trans
);
6627 btrfs_set_trans_block_group(trans
, inode
);
6628 trans
->block_rsv
= root
->orphan_block_rsv
;
6631 * setattr is responsible for setting the ordered_data_close flag,
6632 * but that is only tested during the last file release. That
6633 * could happen well after the next commit, leaving a great big
6634 * window where new writes may get lost if someone chooses to write
6635 * to this file after truncating to zero
6637 * The inode doesn't have any dirty data here, and so if we commit
6638 * this is a noop. If someone immediately starts writing to the inode
6639 * it is very likely we'll catch some of their writes in this
6640 * transaction, and the commit will find this file on the ordered
6641 * data list with good things to send down.
6643 * This is a best effort solution, there is still a window where
6644 * using truncate to replace the contents of the file will
6645 * end up with a zero length file after a crash.
6647 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6648 btrfs_add_ordered_operation(trans
, root
, inode
);
6652 trans
= btrfs_start_transaction(root
, 0);
6654 return PTR_ERR(trans
);
6655 btrfs_set_trans_block_group(trans
, inode
);
6656 trans
->block_rsv
= root
->orphan_block_rsv
;
6659 ret
= btrfs_block_rsv_check(trans
, root
,
6660 root
->orphan_block_rsv
, 0, 5);
6661 if (ret
== -EAGAIN
) {
6662 ret
= btrfs_commit_transaction(trans
, root
);
6672 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6674 BTRFS_EXTENT_DATA_KEY
);
6675 if (ret
!= -EAGAIN
) {
6680 ret
= btrfs_update_inode(trans
, root
, inode
);
6686 nr
= trans
->blocks_used
;
6687 btrfs_end_transaction(trans
, root
);
6689 btrfs_btree_balance_dirty(root
, nr
);
6692 if (ret
== 0 && inode
->i_nlink
> 0) {
6693 ret
= btrfs_orphan_del(trans
, inode
);
6696 } else if (ret
&& inode
->i_nlink
> 0) {
6698 * Failed to do the truncate, remove us from the in memory
6701 ret
= btrfs_orphan_del(NULL
, inode
);
6704 ret
= btrfs_update_inode(trans
, root
, inode
);
6708 nr
= trans
->blocks_used
;
6709 ret
= btrfs_end_transaction_throttle(trans
, root
);
6712 btrfs_btree_balance_dirty(root
, nr
);
6718 * create a new subvolume directory/inode (helper for the ioctl).
6720 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6721 struct btrfs_root
*new_root
,
6722 u64 new_dirid
, u64 alloc_hint
)
6724 struct inode
*inode
;
6728 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6729 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
6731 return PTR_ERR(inode
);
6732 inode
->i_op
= &btrfs_dir_inode_operations
;
6733 inode
->i_fop
= &btrfs_dir_file_operations
;
6736 btrfs_i_size_write(inode
, 0);
6738 err
= btrfs_update_inode(trans
, new_root
, inode
);
6745 /* helper function for file defrag and space balancing. This
6746 * forces readahead on a given range of bytes in an inode
6748 unsigned long btrfs_force_ra(struct address_space
*mapping
,
6749 struct file_ra_state
*ra
, struct file
*file
,
6750 pgoff_t offset
, pgoff_t last_index
)
6752 pgoff_t req_size
= last_index
- offset
+ 1;
6754 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
6755 return offset
+ req_size
;
6758 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6760 struct btrfs_inode
*ei
;
6761 struct inode
*inode
;
6763 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6768 ei
->space_info
= NULL
;
6772 ei
->last_sub_trans
= 0;
6773 ei
->logged_trans
= 0;
6774 ei
->delalloc_bytes
= 0;
6775 ei
->reserved_bytes
= 0;
6776 ei
->disk_i_size
= 0;
6778 ei
->index_cnt
= (u64
)-1;
6779 ei
->last_unlink_trans
= 0;
6781 atomic_set(&ei
->outstanding_extents
, 0);
6782 atomic_set(&ei
->reserved_extents
, 0);
6784 ei
->ordered_data_close
= 0;
6785 ei
->orphan_meta_reserved
= 0;
6786 ei
->dummy_inode
= 0;
6787 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6789 inode
= &ei
->vfs_inode
;
6790 extent_map_tree_init(&ei
->extent_tree
, GFP_NOFS
);
6791 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
, GFP_NOFS
);
6792 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
, GFP_NOFS
);
6793 mutex_init(&ei
->log_mutex
);
6794 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6795 INIT_LIST_HEAD(&ei
->i_orphan
);
6796 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6797 INIT_LIST_HEAD(&ei
->ordered_operations
);
6798 RB_CLEAR_NODE(&ei
->rb_node
);
6803 static void btrfs_i_callback(struct rcu_head
*head
)
6805 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6806 INIT_LIST_HEAD(&inode
->i_dentry
);
6807 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6810 void btrfs_destroy_inode(struct inode
*inode
)
6812 struct btrfs_ordered_extent
*ordered
;
6813 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6815 WARN_ON(!list_empty(&inode
->i_dentry
));
6816 WARN_ON(inode
->i_data
.nrpages
);
6817 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
));
6818 WARN_ON(atomic_read(&BTRFS_I(inode
)->reserved_extents
));
6821 * This can happen where we create an inode, but somebody else also
6822 * created the same inode and we need to destroy the one we already
6829 * Make sure we're properly removed from the ordered operation
6833 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6834 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6835 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6836 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6839 if (root
== root
->fs_info
->tree_root
) {
6840 struct btrfs_block_group_cache
*block_group
;
6842 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6843 BTRFS_I(inode
)->block_group
);
6844 if (block_group
&& block_group
->inode
== inode
) {
6845 spin_lock(&block_group
->lock
);
6846 block_group
->inode
= NULL
;
6847 spin_unlock(&block_group
->lock
);
6848 btrfs_put_block_group(block_group
);
6849 } else if (block_group
) {
6850 btrfs_put_block_group(block_group
);
6854 spin_lock(&root
->orphan_lock
);
6855 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6856 printk(KERN_INFO
"BTRFS: inode %lu still on the orphan list\n",
6858 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6860 spin_unlock(&root
->orphan_lock
);
6863 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6867 printk(KERN_ERR
"btrfs found ordered "
6868 "extent %llu %llu on inode cleanup\n",
6869 (unsigned long long)ordered
->file_offset
,
6870 (unsigned long long)ordered
->len
);
6871 btrfs_remove_ordered_extent(inode
, ordered
);
6872 btrfs_put_ordered_extent(ordered
);
6873 btrfs_put_ordered_extent(ordered
);
6876 inode_tree_del(inode
);
6877 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6879 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6882 int btrfs_drop_inode(struct inode
*inode
)
6884 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6886 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6887 root
!= root
->fs_info
->tree_root
)
6890 return generic_drop_inode(inode
);
6893 static void init_once(void *foo
)
6895 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6897 inode_init_once(&ei
->vfs_inode
);
6900 void btrfs_destroy_cachep(void)
6902 if (btrfs_inode_cachep
)
6903 kmem_cache_destroy(btrfs_inode_cachep
);
6904 if (btrfs_trans_handle_cachep
)
6905 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6906 if (btrfs_transaction_cachep
)
6907 kmem_cache_destroy(btrfs_transaction_cachep
);
6908 if (btrfs_path_cachep
)
6909 kmem_cache_destroy(btrfs_path_cachep
);
6910 if (btrfs_free_space_cachep
)
6911 kmem_cache_destroy(btrfs_free_space_cachep
);
6914 int btrfs_init_cachep(void)
6916 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6917 sizeof(struct btrfs_inode
), 0,
6918 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6919 if (!btrfs_inode_cachep
)
6922 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6923 sizeof(struct btrfs_trans_handle
), 0,
6924 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6925 if (!btrfs_trans_handle_cachep
)
6928 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6929 sizeof(struct btrfs_transaction
), 0,
6930 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6931 if (!btrfs_transaction_cachep
)
6934 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6935 sizeof(struct btrfs_path
), 0,
6936 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6937 if (!btrfs_path_cachep
)
6940 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6941 sizeof(struct btrfs_free_space
), 0,
6942 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6943 if (!btrfs_free_space_cachep
)
6948 btrfs_destroy_cachep();
6952 static int btrfs_getattr(struct vfsmount
*mnt
,
6953 struct dentry
*dentry
, struct kstat
*stat
)
6955 struct inode
*inode
= dentry
->d_inode
;
6956 generic_fillattr(inode
, stat
);
6957 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
6958 stat
->blksize
= PAGE_CACHE_SIZE
;
6959 stat
->blocks
= (inode_get_bytes(inode
) +
6960 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
6965 * If a file is moved, it will inherit the cow and compression flags of the new
6968 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
6970 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
6971 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
6973 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
6974 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
6976 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
6978 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
)
6979 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
6981 b_inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6984 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6985 struct inode
*new_dir
, struct dentry
*new_dentry
)
6987 struct btrfs_trans_handle
*trans
;
6988 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6989 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6990 struct inode
*new_inode
= new_dentry
->d_inode
;
6991 struct inode
*old_inode
= old_dentry
->d_inode
;
6992 struct timespec ctime
= CURRENT_TIME
;
6997 if (new_dir
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
7000 /* we only allow rename subvolume link between subvolumes */
7001 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
7004 if (old_inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
7005 (new_inode
&& new_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
))
7008 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
7009 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
7012 * we're using rename to replace one file with another.
7013 * and the replacement file is large. Start IO on it now so
7014 * we don't add too much work to the end of the transaction
7016 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
7017 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
7018 filemap_flush(old_inode
->i_mapping
);
7020 /* close the racy window with snapshot create/destroy ioctl */
7021 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7022 down_read(&root
->fs_info
->subvol_sem
);
7024 * We want to reserve the absolute worst case amount of items. So if
7025 * both inodes are subvols and we need to unlink them then that would
7026 * require 4 item modifications, but if they are both normal inodes it
7027 * would require 5 item modifications, so we'll assume their normal
7028 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7029 * should cover the worst case number of items we'll modify.
7031 trans
= btrfs_start_transaction(root
, 20);
7032 if (IS_ERR(trans
)) {
7033 ret
= PTR_ERR(trans
);
7037 btrfs_set_trans_block_group(trans
, new_dir
);
7040 btrfs_record_root_in_trans(trans
, dest
);
7042 ret
= btrfs_set_inode_index(new_dir
, &index
);
7046 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7047 /* force full log commit if subvolume involved. */
7048 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7050 ret
= btrfs_insert_inode_ref(trans
, dest
,
7051 new_dentry
->d_name
.name
,
7052 new_dentry
->d_name
.len
,
7054 new_dir
->i_ino
, index
);
7058 * this is an ugly little race, but the rename is required
7059 * to make sure that if we crash, the inode is either at the
7060 * old name or the new one. pinning the log transaction lets
7061 * us make sure we don't allow a log commit to come in after
7062 * we unlink the name but before we add the new name back in.
7064 btrfs_pin_log_trans(root
);
7067 * make sure the inode gets flushed if it is replacing
7070 if (new_inode
&& new_inode
->i_size
&&
7071 old_inode
&& S_ISREG(old_inode
->i_mode
)) {
7072 btrfs_add_ordered_operation(trans
, root
, old_inode
);
7075 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
7076 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
7077 old_inode
->i_ctime
= ctime
;
7079 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
7080 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
7082 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7083 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
7084 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
7085 old_dentry
->d_name
.name
,
7086 old_dentry
->d_name
.len
);
7088 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7089 old_dentry
->d_inode
,
7090 old_dentry
->d_name
.name
,
7091 old_dentry
->d_name
.len
);
7093 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7098 new_inode
->i_ctime
= CURRENT_TIME
;
7099 if (unlikely(new_inode
->i_ino
==
7100 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7101 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7102 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7104 new_dentry
->d_name
.name
,
7105 new_dentry
->d_name
.len
);
7106 BUG_ON(new_inode
->i_nlink
== 0);
7108 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7109 new_dentry
->d_inode
,
7110 new_dentry
->d_name
.name
,
7111 new_dentry
->d_name
.len
);
7114 if (new_inode
->i_nlink
== 0) {
7115 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7120 fixup_inode_flags(new_dir
, old_inode
);
7122 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7123 new_dentry
->d_name
.name
,
7124 new_dentry
->d_name
.len
, 0, index
);
7127 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7128 struct dentry
*parent
= dget_parent(new_dentry
);
7129 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7131 btrfs_end_log_trans(root
);
7134 btrfs_end_transaction_throttle(trans
, root
);
7136 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7137 up_read(&root
->fs_info
->subvol_sem
);
7143 * some fairly slow code that needs optimization. This walks the list
7144 * of all the inodes with pending delalloc and forces them to disk.
7146 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7148 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7149 struct btrfs_inode
*binode
;
7150 struct inode
*inode
;
7152 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7155 spin_lock(&root
->fs_info
->delalloc_lock
);
7156 while (!list_empty(head
)) {
7157 binode
= list_entry(head
->next
, struct btrfs_inode
,
7159 inode
= igrab(&binode
->vfs_inode
);
7161 list_del_init(&binode
->delalloc_inodes
);
7162 spin_unlock(&root
->fs_info
->delalloc_lock
);
7164 filemap_flush(inode
->i_mapping
);
7166 btrfs_add_delayed_iput(inode
);
7171 spin_lock(&root
->fs_info
->delalloc_lock
);
7173 spin_unlock(&root
->fs_info
->delalloc_lock
);
7175 /* the filemap_flush will queue IO into the worker threads, but
7176 * we have to make sure the IO is actually started and that
7177 * ordered extents get created before we return
7179 atomic_inc(&root
->fs_info
->async_submit_draining
);
7180 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7181 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7182 wait_event(root
->fs_info
->async_submit_wait
,
7183 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7184 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7186 atomic_dec(&root
->fs_info
->async_submit_draining
);
7190 int btrfs_start_one_delalloc_inode(struct btrfs_root
*root
, int delay_iput
,
7193 struct btrfs_inode
*binode
;
7194 struct inode
*inode
= NULL
;
7196 spin_lock(&root
->fs_info
->delalloc_lock
);
7197 while (!list_empty(&root
->fs_info
->delalloc_inodes
)) {
7198 binode
= list_entry(root
->fs_info
->delalloc_inodes
.next
,
7199 struct btrfs_inode
, delalloc_inodes
);
7200 inode
= igrab(&binode
->vfs_inode
);
7202 list_move_tail(&binode
->delalloc_inodes
,
7203 &root
->fs_info
->delalloc_inodes
);
7207 list_del_init(&binode
->delalloc_inodes
);
7208 cond_resched_lock(&root
->fs_info
->delalloc_lock
);
7210 spin_unlock(&root
->fs_info
->delalloc_lock
);
7214 filemap_write_and_wait(inode
->i_mapping
);
7216 * We have to do this because compression doesn't
7217 * actually set PG_writeback until it submits the pages
7218 * for IO, which happens in an async thread, so we could
7219 * race and not actually wait for any writeback pages
7220 * because they've not been submitted yet. Technically
7221 * this could still be the case for the ordered stuff
7222 * since the async thread may not have started to do its
7223 * work yet. If this becomes the case then we need to
7224 * figure out a way to make sure that in writepage we
7225 * wait for any async pages to be submitted before
7226 * returning so that fdatawait does what its supposed to
7229 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
7231 filemap_flush(inode
->i_mapping
);
7234 btrfs_add_delayed_iput(inode
);
7242 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7243 const char *symname
)
7245 struct btrfs_trans_handle
*trans
;
7246 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7247 struct btrfs_path
*path
;
7248 struct btrfs_key key
;
7249 struct inode
*inode
= NULL
;
7257 struct btrfs_file_extent_item
*ei
;
7258 struct extent_buffer
*leaf
;
7259 unsigned long nr
= 0;
7261 name_len
= strlen(symname
) + 1;
7262 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7263 return -ENAMETOOLONG
;
7265 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
7269 * 2 items for inode item and ref
7270 * 2 items for dir items
7271 * 1 item for xattr if selinux is on
7273 trans
= btrfs_start_transaction(root
, 5);
7275 return PTR_ERR(trans
);
7277 btrfs_set_trans_block_group(trans
, dir
);
7279 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7280 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
7281 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
7283 if (IS_ERR(inode
)) {
7284 err
= PTR_ERR(inode
);
7288 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7294 btrfs_set_trans_block_group(trans
, inode
);
7295 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7299 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7300 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7301 inode
->i_fop
= &btrfs_file_operations
;
7302 inode
->i_op
= &btrfs_file_inode_operations
;
7303 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7305 btrfs_update_inode_block_group(trans
, inode
);
7306 btrfs_update_inode_block_group(trans
, dir
);
7310 path
= btrfs_alloc_path();
7312 key
.objectid
= inode
->i_ino
;
7314 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7315 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7316 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7322 leaf
= path
->nodes
[0];
7323 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7324 struct btrfs_file_extent_item
);
7325 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7326 btrfs_set_file_extent_type(leaf
, ei
,
7327 BTRFS_FILE_EXTENT_INLINE
);
7328 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7329 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7330 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7331 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7333 ptr
= btrfs_file_extent_inline_start(ei
);
7334 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7335 btrfs_mark_buffer_dirty(leaf
);
7336 btrfs_free_path(path
);
7338 inode
->i_op
= &btrfs_symlink_inode_operations
;
7339 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7340 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7341 inode_set_bytes(inode
, name_len
);
7342 btrfs_i_size_write(inode
, name_len
- 1);
7343 err
= btrfs_update_inode(trans
, root
, inode
);
7348 nr
= trans
->blocks_used
;
7349 btrfs_end_transaction_throttle(trans
, root
);
7351 inode_dec_link_count(inode
);
7354 btrfs_btree_balance_dirty(root
, nr
);
7358 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7359 u64 start
, u64 num_bytes
, u64 min_size
,
7360 loff_t actual_len
, u64
*alloc_hint
,
7361 struct btrfs_trans_handle
*trans
)
7363 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7364 struct btrfs_key ins
;
7365 u64 cur_offset
= start
;
7368 bool own_trans
= true;
7372 while (num_bytes
> 0) {
7374 trans
= btrfs_start_transaction(root
, 3);
7375 if (IS_ERR(trans
)) {
7376 ret
= PTR_ERR(trans
);
7381 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7382 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7385 btrfs_end_transaction(trans
, root
);
7389 ret
= insert_reserved_file_extent(trans
, inode
,
7390 cur_offset
, ins
.objectid
,
7391 ins
.offset
, ins
.offset
,
7392 ins
.offset
, 0, 0, 0,
7393 BTRFS_FILE_EXTENT_PREALLOC
);
7395 btrfs_drop_extent_cache(inode
, cur_offset
,
7396 cur_offset
+ ins
.offset
-1, 0);
7398 num_bytes
-= ins
.offset
;
7399 cur_offset
+= ins
.offset
;
7400 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7402 inode
->i_ctime
= CURRENT_TIME
;
7403 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7404 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7405 (actual_len
> inode
->i_size
) &&
7406 (cur_offset
> inode
->i_size
)) {
7407 if (cur_offset
> actual_len
)
7408 i_size
= actual_len
;
7410 i_size
= cur_offset
;
7411 i_size_write(inode
, i_size
);
7412 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7415 ret
= btrfs_update_inode(trans
, root
, inode
);
7419 btrfs_end_transaction(trans
, root
);
7424 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7425 u64 start
, u64 num_bytes
, u64 min_size
,
7426 loff_t actual_len
, u64
*alloc_hint
)
7428 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7429 min_size
, actual_len
, alloc_hint
,
7433 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7434 struct btrfs_trans_handle
*trans
, int mode
,
7435 u64 start
, u64 num_bytes
, u64 min_size
,
7436 loff_t actual_len
, u64
*alloc_hint
)
7438 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7439 min_size
, actual_len
, alloc_hint
, trans
);
7442 static int btrfs_set_page_dirty(struct page
*page
)
7444 return __set_page_dirty_nobuffers(page
);
7447 static int btrfs_permission(struct inode
*inode
, int mask
, unsigned int flags
)
7449 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7451 if (btrfs_root_readonly(root
) && (mask
& MAY_WRITE
))
7453 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
7455 return generic_permission(inode
, mask
, flags
, btrfs_check_acl
);
7458 static const struct inode_operations btrfs_dir_inode_operations
= {
7459 .getattr
= btrfs_getattr
,
7460 .lookup
= btrfs_lookup
,
7461 .create
= btrfs_create
,
7462 .unlink
= btrfs_unlink
,
7464 .mkdir
= btrfs_mkdir
,
7465 .rmdir
= btrfs_rmdir
,
7466 .rename
= btrfs_rename
,
7467 .symlink
= btrfs_symlink
,
7468 .setattr
= btrfs_setattr
,
7469 .mknod
= btrfs_mknod
,
7470 .setxattr
= btrfs_setxattr
,
7471 .getxattr
= btrfs_getxattr
,
7472 .listxattr
= btrfs_listxattr
,
7473 .removexattr
= btrfs_removexattr
,
7474 .permission
= btrfs_permission
,
7476 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7477 .lookup
= btrfs_lookup
,
7478 .permission
= btrfs_permission
,
7481 static const struct file_operations btrfs_dir_file_operations
= {
7482 .llseek
= generic_file_llseek
,
7483 .read
= generic_read_dir
,
7484 .readdir
= btrfs_real_readdir
,
7485 .unlocked_ioctl
= btrfs_ioctl
,
7486 #ifdef CONFIG_COMPAT
7487 .compat_ioctl
= btrfs_ioctl
,
7489 .release
= btrfs_release_file
,
7490 .fsync
= btrfs_sync_file
,
7493 static struct extent_io_ops btrfs_extent_io_ops
= {
7494 .fill_delalloc
= run_delalloc_range
,
7495 .submit_bio_hook
= btrfs_submit_bio_hook
,
7496 .merge_bio_hook
= btrfs_merge_bio_hook
,
7497 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7498 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7499 .writepage_start_hook
= btrfs_writepage_start_hook
,
7500 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
7501 .set_bit_hook
= btrfs_set_bit_hook
,
7502 .clear_bit_hook
= btrfs_clear_bit_hook
,
7503 .merge_extent_hook
= btrfs_merge_extent_hook
,
7504 .split_extent_hook
= btrfs_split_extent_hook
,
7508 * btrfs doesn't support the bmap operation because swapfiles
7509 * use bmap to make a mapping of extents in the file. They assume
7510 * these extents won't change over the life of the file and they
7511 * use the bmap result to do IO directly to the drive.
7513 * the btrfs bmap call would return logical addresses that aren't
7514 * suitable for IO and they also will change frequently as COW
7515 * operations happen. So, swapfile + btrfs == corruption.
7517 * For now we're avoiding this by dropping bmap.
7519 static const struct address_space_operations btrfs_aops
= {
7520 .readpage
= btrfs_readpage
,
7521 .writepage
= btrfs_writepage
,
7522 .writepages
= btrfs_writepages
,
7523 .readpages
= btrfs_readpages
,
7524 .direct_IO
= btrfs_direct_IO
,
7525 .invalidatepage
= btrfs_invalidatepage
,
7526 .releasepage
= btrfs_releasepage
,
7527 .set_page_dirty
= btrfs_set_page_dirty
,
7528 .error_remove_page
= generic_error_remove_page
,
7531 static const struct address_space_operations btrfs_symlink_aops
= {
7532 .readpage
= btrfs_readpage
,
7533 .writepage
= btrfs_writepage
,
7534 .invalidatepage
= btrfs_invalidatepage
,
7535 .releasepage
= btrfs_releasepage
,
7538 static const struct inode_operations btrfs_file_inode_operations
= {
7539 .getattr
= btrfs_getattr
,
7540 .setattr
= btrfs_setattr
,
7541 .setxattr
= btrfs_setxattr
,
7542 .getxattr
= btrfs_getxattr
,
7543 .listxattr
= btrfs_listxattr
,
7544 .removexattr
= btrfs_removexattr
,
7545 .permission
= btrfs_permission
,
7546 .fiemap
= btrfs_fiemap
,
7548 static const struct inode_operations btrfs_special_inode_operations
= {
7549 .getattr
= btrfs_getattr
,
7550 .setattr
= btrfs_setattr
,
7551 .permission
= btrfs_permission
,
7552 .setxattr
= btrfs_setxattr
,
7553 .getxattr
= btrfs_getxattr
,
7554 .listxattr
= btrfs_listxattr
,
7555 .removexattr
= btrfs_removexattr
,
7557 static const struct inode_operations btrfs_symlink_inode_operations
= {
7558 .readlink
= generic_readlink
,
7559 .follow_link
= page_follow_link_light
,
7560 .put_link
= page_put_link
,
7561 .getattr
= btrfs_getattr
,
7562 .permission
= btrfs_permission
,
7563 .setxattr
= btrfs_setxattr
,
7564 .getxattr
= btrfs_getxattr
,
7565 .listxattr
= btrfs_listxattr
,
7566 .removexattr
= btrfs_removexattr
,
7569 const struct dentry_operations btrfs_dentry_operations
= {
7570 .d_delete
= btrfs_dentry_delete
,