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
)
100 err
= btrfs_init_acl(trans
, inode
, dir
);
102 err
= btrfs_xattr_security_init(trans
, inode
, dir
);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
112 struct btrfs_root
*root
, struct inode
*inode
,
113 u64 start
, size_t size
, size_t compressed_size
,
114 struct page
**compressed_pages
)
116 struct btrfs_key key
;
117 struct btrfs_path
*path
;
118 struct extent_buffer
*leaf
;
119 struct page
*page
= NULL
;
122 struct btrfs_file_extent_item
*ei
;
125 size_t cur_size
= size
;
127 unsigned long offset
;
128 int compress_type
= BTRFS_COMPRESS_NONE
;
130 if (compressed_size
&& compressed_pages
) {
131 compress_type
= root
->fs_info
->compress_type
;
132 cur_size
= compressed_size
;
135 path
= btrfs_alloc_path();
139 path
->leave_spinning
= 1;
140 btrfs_set_trans_block_group(trans
, inode
);
142 key
.objectid
= inode
->i_ino
;
144 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
145 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
147 inode_add_bytes(inode
, size
);
148 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
155 leaf
= path
->nodes
[0];
156 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
157 struct btrfs_file_extent_item
);
158 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
159 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
160 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
161 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
162 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
163 ptr
= btrfs_file_extent_inline_start(ei
);
165 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
168 while (compressed_size
> 0) {
169 cpage
= compressed_pages
[i
];
170 cur_size
= min_t(unsigned long, compressed_size
,
173 kaddr
= kmap_atomic(cpage
, KM_USER0
);
174 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
175 kunmap_atomic(kaddr
, KM_USER0
);
179 compressed_size
-= cur_size
;
181 btrfs_set_file_extent_compression(leaf
, ei
,
184 page
= find_get_page(inode
->i_mapping
,
185 start
>> PAGE_CACHE_SHIFT
);
186 btrfs_set_file_extent_compression(leaf
, ei
, 0);
187 kaddr
= kmap_atomic(page
, KM_USER0
);
188 offset
= start
& (PAGE_CACHE_SIZE
- 1);
189 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
190 kunmap_atomic(kaddr
, KM_USER0
);
191 page_cache_release(page
);
193 btrfs_mark_buffer_dirty(leaf
);
194 btrfs_free_path(path
);
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
205 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
206 btrfs_update_inode(trans
, root
, inode
);
210 btrfs_free_path(path
);
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
220 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
221 struct btrfs_root
*root
,
222 struct inode
*inode
, u64 start
, u64 end
,
223 size_t compressed_size
,
224 struct page
**compressed_pages
)
226 u64 isize
= i_size_read(inode
);
227 u64 actual_end
= min(end
+ 1, isize
);
228 u64 inline_len
= actual_end
- start
;
229 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
230 ~((u64
)root
->sectorsize
- 1);
232 u64 data_len
= inline_len
;
236 data_len
= compressed_size
;
239 actual_end
>= PAGE_CACHE_SIZE
||
240 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
242 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
244 data_len
> root
->fs_info
->max_inline
) {
248 ret
= btrfs_drop_extents(trans
, inode
, start
, aligned_end
,
252 if (isize
> actual_end
)
253 inline_len
= min_t(u64
, isize
, actual_end
);
254 ret
= insert_inline_extent(trans
, root
, inode
, start
,
255 inline_len
, compressed_size
,
258 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
259 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
263 struct async_extent
{
268 unsigned long nr_pages
;
270 struct list_head list
;
275 struct btrfs_root
*root
;
276 struct page
*locked_page
;
279 struct list_head extents
;
280 struct btrfs_work work
;
283 static noinline
int add_async_extent(struct async_cow
*cow
,
284 u64 start
, u64 ram_size
,
287 unsigned long nr_pages
,
290 struct async_extent
*async_extent
;
292 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
293 BUG_ON(!async_extent
);
294 async_extent
->start
= start
;
295 async_extent
->ram_size
= ram_size
;
296 async_extent
->compressed_size
= compressed_size
;
297 async_extent
->pages
= pages
;
298 async_extent
->nr_pages
= nr_pages
;
299 async_extent
->compress_type
= compress_type
;
300 list_add_tail(&async_extent
->list
, &cow
->extents
);
305 * we create compressed extents in two phases. The first
306 * phase compresses a range of pages that have already been
307 * locked (both pages and state bits are locked).
309 * This is done inside an ordered work queue, and the compression
310 * is spread across many cpus. The actual IO submission is step
311 * two, and the ordered work queue takes care of making sure that
312 * happens in the same order things were put onto the queue by
313 * writepages and friends.
315 * If this code finds it can't get good compression, it puts an
316 * entry onto the work queue to write the uncompressed bytes. This
317 * makes sure that both compressed inodes and uncompressed inodes
318 * are written in the same order that pdflush sent them down.
320 static noinline
int compress_file_range(struct inode
*inode
,
321 struct page
*locked_page
,
323 struct async_cow
*async_cow
,
326 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
327 struct btrfs_trans_handle
*trans
;
329 u64 blocksize
= root
->sectorsize
;
331 u64 isize
= i_size_read(inode
);
333 struct page
**pages
= NULL
;
334 unsigned long nr_pages
;
335 unsigned long nr_pages_ret
= 0;
336 unsigned long total_compressed
= 0;
337 unsigned long total_in
= 0;
338 unsigned long max_compressed
= 128 * 1024;
339 unsigned long max_uncompressed
= 128 * 1024;
342 int compress_type
= root
->fs_info
->compress_type
;
344 actual_end
= min_t(u64
, isize
, end
+ 1);
347 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
348 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
351 * we don't want to send crud past the end of i_size through
352 * compression, that's just a waste of CPU time. So, if the
353 * end of the file is before the start of our current
354 * requested range of bytes, we bail out to the uncompressed
355 * cleanup code that can deal with all of this.
357 * It isn't really the fastest way to fix things, but this is a
358 * very uncommon corner.
360 if (actual_end
<= start
)
361 goto cleanup_and_bail_uncompressed
;
363 total_compressed
= actual_end
- start
;
365 /* we want to make sure that amount of ram required to uncompress
366 * an extent is reasonable, so we limit the total size in ram
367 * of a compressed extent to 128k. This is a crucial number
368 * because it also controls how easily we can spread reads across
369 * cpus for decompression.
371 * We also want to make sure the amount of IO required to do
372 * a random read is reasonably small, so we limit the size of
373 * a compressed extent to 128k.
375 total_compressed
= min(total_compressed
, max_uncompressed
);
376 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
377 num_bytes
= max(blocksize
, num_bytes
);
382 * we do compression for mount -o compress and when the
383 * inode has not been flagged as nocompress. This flag can
384 * change at any time if we discover bad compression ratios.
386 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
387 (btrfs_test_opt(root
, COMPRESS
) ||
388 (BTRFS_I(inode
)->force_compress
) ||
389 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
391 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
394 if (BTRFS_I(inode
)->force_compress
)
395 compress_type
= BTRFS_I(inode
)->force_compress
;
397 ret
= btrfs_compress_pages(compress_type
,
398 inode
->i_mapping
, start
,
399 total_compressed
, pages
,
400 nr_pages
, &nr_pages_ret
,
406 unsigned long offset
= total_compressed
&
407 (PAGE_CACHE_SIZE
- 1);
408 struct page
*page
= pages
[nr_pages_ret
- 1];
411 /* zero the tail end of the last page, we might be
412 * sending it down to disk
415 kaddr
= kmap_atomic(page
, KM_USER0
);
416 memset(kaddr
+ offset
, 0,
417 PAGE_CACHE_SIZE
- offset
);
418 kunmap_atomic(kaddr
, KM_USER0
);
424 trans
= btrfs_join_transaction(root
, 1);
425 BUG_ON(IS_ERR(trans
));
426 btrfs_set_trans_block_group(trans
, inode
);
427 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
429 /* lets try to make an inline extent */
430 if (ret
|| total_in
< (actual_end
- start
)) {
431 /* we didn't compress the entire range, try
432 * to make an uncompressed inline extent.
434 ret
= cow_file_range_inline(trans
, root
, inode
,
435 start
, end
, 0, NULL
);
437 /* try making a compressed inline extent */
438 ret
= cow_file_range_inline(trans
, root
, inode
,
440 total_compressed
, 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
, 1);
621 BUG_ON(IS_ERR(trans
));
622 ret
= btrfs_reserve_extent(trans
, root
,
623 async_extent
->compressed_size
,
624 async_extent
->compressed_size
,
627 btrfs_end_transaction(trans
, root
);
631 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
632 WARN_ON(async_extent
->pages
[i
]->mapping
);
633 page_cache_release(async_extent
->pages
[i
]);
635 kfree(async_extent
->pages
);
636 async_extent
->nr_pages
= 0;
637 async_extent
->pages
= NULL
;
638 unlock_extent(io_tree
, async_extent
->start
,
639 async_extent
->start
+
640 async_extent
->ram_size
- 1, GFP_NOFS
);
645 * here we're doing allocation and writeback of the
648 btrfs_drop_extent_cache(inode
, async_extent
->start
,
649 async_extent
->start
+
650 async_extent
->ram_size
- 1, 0);
652 em
= alloc_extent_map(GFP_NOFS
);
654 em
->start
= async_extent
->start
;
655 em
->len
= async_extent
->ram_size
;
656 em
->orig_start
= em
->start
;
658 em
->block_start
= ins
.objectid
;
659 em
->block_len
= ins
.offset
;
660 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
661 em
->compress_type
= async_extent
->compress_type
;
662 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
663 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
666 write_lock(&em_tree
->lock
);
667 ret
= add_extent_mapping(em_tree
, em
);
668 write_unlock(&em_tree
->lock
);
669 if (ret
!= -EEXIST
) {
673 btrfs_drop_extent_cache(inode
, async_extent
->start
,
674 async_extent
->start
+
675 async_extent
->ram_size
- 1, 0);
678 ret
= btrfs_add_ordered_extent_compress(inode
,
681 async_extent
->ram_size
,
683 BTRFS_ORDERED_COMPRESSED
,
684 async_extent
->compress_type
);
688 * clear dirty, set writeback and unlock the pages.
690 extent_clear_unlock_delalloc(inode
,
691 &BTRFS_I(inode
)->io_tree
,
693 async_extent
->start
+
694 async_extent
->ram_size
- 1,
695 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
696 EXTENT_CLEAR_UNLOCK
|
697 EXTENT_CLEAR_DELALLOC
|
698 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
700 ret
= btrfs_submit_compressed_write(inode
,
702 async_extent
->ram_size
,
704 ins
.offset
, async_extent
->pages
,
705 async_extent
->nr_pages
);
708 alloc_hint
= ins
.objectid
+ ins
.offset
;
716 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
719 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
720 struct extent_map
*em
;
723 read_lock(&em_tree
->lock
);
724 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
727 * if block start isn't an actual block number then find the
728 * first block in this inode and use that as a hint. If that
729 * block is also bogus then just don't worry about it.
731 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
733 em
= search_extent_mapping(em_tree
, 0, 0);
734 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
735 alloc_hint
= em
->block_start
;
739 alloc_hint
= em
->block_start
;
743 read_unlock(&em_tree
->lock
);
749 * when extent_io.c finds a delayed allocation range in the file,
750 * the call backs end up in this code. The basic idea is to
751 * allocate extents on disk for the range, and create ordered data structs
752 * in ram to track those extents.
754 * locked_page is the page that writepage had locked already. We use
755 * it to make sure we don't do extra locks or unlocks.
757 * *page_started is set to one if we unlock locked_page and do everything
758 * required to start IO on it. It may be clean and already done with
761 static noinline
int cow_file_range(struct inode
*inode
,
762 struct page
*locked_page
,
763 u64 start
, u64 end
, int *page_started
,
764 unsigned long *nr_written
,
767 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
768 struct btrfs_trans_handle
*trans
;
771 unsigned long ram_size
;
774 u64 blocksize
= root
->sectorsize
;
775 struct btrfs_key ins
;
776 struct extent_map
*em
;
777 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
780 BUG_ON(root
== root
->fs_info
->tree_root
);
781 trans
= btrfs_join_transaction(root
, 1);
782 BUG_ON(IS_ERR(trans
));
783 btrfs_set_trans_block_group(trans
, inode
);
784 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
786 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
787 num_bytes
= max(blocksize
, num_bytes
);
788 disk_num_bytes
= num_bytes
;
792 /* lets try to make an inline extent */
793 ret
= cow_file_range_inline(trans
, root
, inode
,
794 start
, end
, 0, NULL
);
796 extent_clear_unlock_delalloc(inode
,
797 &BTRFS_I(inode
)->io_tree
,
799 EXTENT_CLEAR_UNLOCK_PAGE
|
800 EXTENT_CLEAR_UNLOCK
|
801 EXTENT_CLEAR_DELALLOC
|
803 EXTENT_SET_WRITEBACK
|
804 EXTENT_END_WRITEBACK
);
806 *nr_written
= *nr_written
+
807 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
814 BUG_ON(disk_num_bytes
>
815 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
817 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
818 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
820 while (disk_num_bytes
> 0) {
823 cur_alloc_size
= disk_num_bytes
;
824 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
825 root
->sectorsize
, 0, alloc_hint
,
829 em
= alloc_extent_map(GFP_NOFS
);
832 em
->orig_start
= em
->start
;
833 ram_size
= ins
.offset
;
834 em
->len
= ins
.offset
;
836 em
->block_start
= ins
.objectid
;
837 em
->block_len
= ins
.offset
;
838 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
839 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
842 write_lock(&em_tree
->lock
);
843 ret
= add_extent_mapping(em_tree
, em
);
844 write_unlock(&em_tree
->lock
);
845 if (ret
!= -EEXIST
) {
849 btrfs_drop_extent_cache(inode
, start
,
850 start
+ ram_size
- 1, 0);
853 cur_alloc_size
= ins
.offset
;
854 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
855 ram_size
, cur_alloc_size
, 0);
858 if (root
->root_key
.objectid
==
859 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
860 ret
= btrfs_reloc_clone_csums(inode
, start
,
865 if (disk_num_bytes
< cur_alloc_size
)
868 /* we're not doing compressed IO, don't unlock the first
869 * page (which the caller expects to stay locked), don't
870 * clear any dirty bits and don't set any writeback bits
872 * Do set the Private2 bit so we know this page was properly
873 * setup for writepage
875 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
876 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
879 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
880 start
, start
+ ram_size
- 1,
882 disk_num_bytes
-= cur_alloc_size
;
883 num_bytes
-= cur_alloc_size
;
884 alloc_hint
= ins
.objectid
+ ins
.offset
;
885 start
+= cur_alloc_size
;
889 btrfs_end_transaction(trans
, root
);
895 * work queue call back to started compression on a file and pages
897 static noinline
void async_cow_start(struct btrfs_work
*work
)
899 struct async_cow
*async_cow
;
901 async_cow
= container_of(work
, struct async_cow
, work
);
903 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
904 async_cow
->start
, async_cow
->end
, async_cow
,
907 async_cow
->inode
= NULL
;
911 * work queue call back to submit previously compressed pages
913 static noinline
void async_cow_submit(struct btrfs_work
*work
)
915 struct async_cow
*async_cow
;
916 struct btrfs_root
*root
;
917 unsigned long nr_pages
;
919 async_cow
= container_of(work
, struct async_cow
, work
);
921 root
= async_cow
->root
;
922 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
925 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
927 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
929 waitqueue_active(&root
->fs_info
->async_submit_wait
))
930 wake_up(&root
->fs_info
->async_submit_wait
);
932 if (async_cow
->inode
)
933 submit_compressed_extents(async_cow
->inode
, async_cow
);
936 static noinline
void async_cow_free(struct btrfs_work
*work
)
938 struct async_cow
*async_cow
;
939 async_cow
= container_of(work
, struct async_cow
, work
);
943 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
944 u64 start
, u64 end
, int *page_started
,
945 unsigned long *nr_written
)
947 struct async_cow
*async_cow
;
948 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
949 unsigned long nr_pages
;
951 int limit
= 10 * 1024 * 1042;
953 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
954 1, 0, NULL
, GFP_NOFS
);
955 while (start
< end
) {
956 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
957 async_cow
->inode
= inode
;
958 async_cow
->root
= root
;
959 async_cow
->locked_page
= locked_page
;
960 async_cow
->start
= start
;
962 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
965 cur_end
= min(end
, start
+ 512 * 1024 - 1);
967 async_cow
->end
= cur_end
;
968 INIT_LIST_HEAD(&async_cow
->extents
);
970 async_cow
->work
.func
= async_cow_start
;
971 async_cow
->work
.ordered_func
= async_cow_submit
;
972 async_cow
->work
.ordered_free
= async_cow_free
;
973 async_cow
->work
.flags
= 0;
975 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
977 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
979 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
982 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
983 wait_event(root
->fs_info
->async_submit_wait
,
984 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
988 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
989 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
990 wait_event(root
->fs_info
->async_submit_wait
,
991 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
995 *nr_written
+= nr_pages
;
1002 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1003 u64 bytenr
, u64 num_bytes
)
1006 struct btrfs_ordered_sum
*sums
;
1009 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1010 bytenr
+ num_bytes
- 1, &list
);
1011 if (ret
== 0 && list_empty(&list
))
1014 while (!list_empty(&list
)) {
1015 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1016 list_del(&sums
->list
);
1023 * when nowcow writeback call back. This checks for snapshots or COW copies
1024 * of the extents that exist in the file, and COWs the file as required.
1026 * If no cow copies or snapshots exist, we write directly to the existing
1029 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1030 struct page
*locked_page
,
1031 u64 start
, u64 end
, int *page_started
, int force
,
1032 unsigned long *nr_written
)
1034 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1035 struct btrfs_trans_handle
*trans
;
1036 struct extent_buffer
*leaf
;
1037 struct btrfs_path
*path
;
1038 struct btrfs_file_extent_item
*fi
;
1039 struct btrfs_key found_key
;
1051 bool nolock
= false;
1053 path
= btrfs_alloc_path();
1055 if (root
== root
->fs_info
->tree_root
) {
1057 trans
= btrfs_join_transaction_nolock(root
, 1);
1059 trans
= btrfs_join_transaction(root
, 1);
1061 BUG_ON(IS_ERR(trans
));
1063 cow_start
= (u64
)-1;
1066 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
1069 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1070 leaf
= path
->nodes
[0];
1071 btrfs_item_key_to_cpu(leaf
, &found_key
,
1072 path
->slots
[0] - 1);
1073 if (found_key
.objectid
== inode
->i_ino
&&
1074 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1079 leaf
= path
->nodes
[0];
1080 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1081 ret
= btrfs_next_leaf(root
, path
);
1086 leaf
= path
->nodes
[0];
1092 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1094 if (found_key
.objectid
> inode
->i_ino
||
1095 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1096 found_key
.offset
> end
)
1099 if (found_key
.offset
> cur_offset
) {
1100 extent_end
= found_key
.offset
;
1105 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1106 struct btrfs_file_extent_item
);
1107 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1109 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1110 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1111 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1112 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1113 extent_end
= found_key
.offset
+
1114 btrfs_file_extent_num_bytes(leaf
, fi
);
1115 if (extent_end
<= start
) {
1119 if (disk_bytenr
== 0)
1121 if (btrfs_file_extent_compression(leaf
, fi
) ||
1122 btrfs_file_extent_encryption(leaf
, fi
) ||
1123 btrfs_file_extent_other_encoding(leaf
, fi
))
1125 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1127 if (btrfs_extent_readonly(root
, disk_bytenr
))
1129 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1131 extent_offset
, disk_bytenr
))
1133 disk_bytenr
+= extent_offset
;
1134 disk_bytenr
+= cur_offset
- found_key
.offset
;
1135 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1137 * force cow if csum exists in the range.
1138 * this ensure that csum for a given extent are
1139 * either valid or do not exist.
1141 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1144 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1145 extent_end
= found_key
.offset
+
1146 btrfs_file_extent_inline_len(leaf
, fi
);
1147 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1152 if (extent_end
<= start
) {
1157 if (cow_start
== (u64
)-1)
1158 cow_start
= cur_offset
;
1159 cur_offset
= extent_end
;
1160 if (cur_offset
> end
)
1166 btrfs_release_path(root
, path
);
1167 if (cow_start
!= (u64
)-1) {
1168 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1169 found_key
.offset
- 1, page_started
,
1172 cow_start
= (u64
)-1;
1175 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1176 struct extent_map
*em
;
1177 struct extent_map_tree
*em_tree
;
1178 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1179 em
= alloc_extent_map(GFP_NOFS
);
1181 em
->start
= cur_offset
;
1182 em
->orig_start
= em
->start
;
1183 em
->len
= num_bytes
;
1184 em
->block_len
= num_bytes
;
1185 em
->block_start
= disk_bytenr
;
1186 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1187 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1189 write_lock(&em_tree
->lock
);
1190 ret
= add_extent_mapping(em_tree
, em
);
1191 write_unlock(&em_tree
->lock
);
1192 if (ret
!= -EEXIST
) {
1193 free_extent_map(em
);
1196 btrfs_drop_extent_cache(inode
, em
->start
,
1197 em
->start
+ em
->len
- 1, 0);
1199 type
= BTRFS_ORDERED_PREALLOC
;
1201 type
= BTRFS_ORDERED_NOCOW
;
1204 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1205 num_bytes
, num_bytes
, type
);
1208 if (root
->root_key
.objectid
==
1209 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1210 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1215 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1216 cur_offset
, cur_offset
+ num_bytes
- 1,
1217 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1218 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1219 EXTENT_SET_PRIVATE2
);
1220 cur_offset
= extent_end
;
1221 if (cur_offset
> end
)
1224 btrfs_release_path(root
, path
);
1226 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1227 cow_start
= cur_offset
;
1228 if (cow_start
!= (u64
)-1) {
1229 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1230 page_started
, nr_written
, 1);
1235 ret
= btrfs_end_transaction_nolock(trans
, root
);
1238 ret
= btrfs_end_transaction(trans
, root
);
1241 btrfs_free_path(path
);
1246 * extent_io.c call back to do delayed allocation processing
1248 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1249 u64 start
, u64 end
, int *page_started
,
1250 unsigned long *nr_written
)
1253 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1255 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1256 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1257 page_started
, 1, nr_written
);
1258 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1259 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1260 page_started
, 0, nr_written
);
1261 else if (!btrfs_test_opt(root
, COMPRESS
) &&
1262 !(BTRFS_I(inode
)->force_compress
) &&
1263 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))
1264 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1265 page_started
, nr_written
, 1);
1267 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1268 page_started
, nr_written
);
1272 static int btrfs_split_extent_hook(struct inode
*inode
,
1273 struct extent_state
*orig
, u64 split
)
1275 /* not delalloc, ignore it */
1276 if (!(orig
->state
& EXTENT_DELALLOC
))
1279 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1284 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1285 * extents so we can keep track of new extents that are just merged onto old
1286 * extents, such as when we are doing sequential writes, so we can properly
1287 * account for the metadata space we'll need.
1289 static int btrfs_merge_extent_hook(struct inode
*inode
,
1290 struct extent_state
*new,
1291 struct extent_state
*other
)
1293 /* not delalloc, ignore it */
1294 if (!(other
->state
& EXTENT_DELALLOC
))
1297 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1302 * extent_io.c set_bit_hook, used to track delayed allocation
1303 * bytes in this file, and to maintain the list of inodes that
1304 * have pending delalloc work to be done.
1306 static int btrfs_set_bit_hook(struct inode
*inode
,
1307 struct extent_state
*state
, int *bits
)
1311 * set_bit and clear bit hooks normally require _irqsave/restore
1312 * but in this case, we are only testeing for the DELALLOC
1313 * bit, which is only set or cleared with irqs on
1315 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1316 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1317 u64 len
= state
->end
+ 1 - state
->start
;
1318 int do_list
= (root
->root_key
.objectid
!=
1319 BTRFS_ROOT_TREE_OBJECTID
);
1321 if (*bits
& EXTENT_FIRST_DELALLOC
)
1322 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1324 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1326 spin_lock(&root
->fs_info
->delalloc_lock
);
1327 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1328 root
->fs_info
->delalloc_bytes
+= len
;
1329 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1330 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1331 &root
->fs_info
->delalloc_inodes
);
1333 spin_unlock(&root
->fs_info
->delalloc_lock
);
1339 * extent_io.c clear_bit_hook, see set_bit_hook for why
1341 static int btrfs_clear_bit_hook(struct inode
*inode
,
1342 struct extent_state
*state
, int *bits
)
1345 * set_bit and clear bit hooks normally require _irqsave/restore
1346 * but in this case, we are only testeing for the DELALLOC
1347 * bit, which is only set or cleared with irqs on
1349 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1350 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1351 u64 len
= state
->end
+ 1 - state
->start
;
1352 int do_list
= (root
->root_key
.objectid
!=
1353 BTRFS_ROOT_TREE_OBJECTID
);
1355 if (*bits
& EXTENT_FIRST_DELALLOC
)
1356 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1357 else if (!(*bits
& EXTENT_DO_ACCOUNTING
))
1358 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1360 if (*bits
& EXTENT_DO_ACCOUNTING
)
1361 btrfs_delalloc_release_metadata(inode
, len
);
1363 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1365 btrfs_free_reserved_data_space(inode
, len
);
1367 spin_lock(&root
->fs_info
->delalloc_lock
);
1368 root
->fs_info
->delalloc_bytes
-= len
;
1369 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1371 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1372 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1373 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1375 spin_unlock(&root
->fs_info
->delalloc_lock
);
1381 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1382 * we don't create bios that span stripes or chunks
1384 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1385 size_t size
, struct bio
*bio
,
1386 unsigned long bio_flags
)
1388 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1389 struct btrfs_mapping_tree
*map_tree
;
1390 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1395 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1398 length
= bio
->bi_size
;
1399 map_tree
= &root
->fs_info
->mapping_tree
;
1400 map_length
= length
;
1401 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1402 &map_length
, NULL
, 0);
1404 if (map_length
< length
+ size
)
1410 * in order to insert checksums into the metadata in large chunks,
1411 * we wait until bio submission time. All the pages in the bio are
1412 * checksummed and sums are attached onto the ordered extent record.
1414 * At IO completion time the cums attached on the ordered extent record
1415 * are inserted into the btree
1417 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1418 struct bio
*bio
, int mirror_num
,
1419 unsigned long bio_flags
,
1422 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1425 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1431 * in order to insert checksums into the metadata in large chunks,
1432 * we wait until bio submission time. All the pages in the bio are
1433 * checksummed and sums are attached onto the ordered extent record.
1435 * At IO completion time the cums attached on the ordered extent record
1436 * are inserted into the btree
1438 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1439 int mirror_num
, unsigned long bio_flags
,
1442 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1443 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1447 * extent_io.c submission hook. This does the right thing for csum calculation
1448 * on write, or reading the csums from the tree before a read
1450 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1451 int mirror_num
, unsigned long bio_flags
,
1454 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1458 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1460 if (root
== root
->fs_info
->tree_root
)
1461 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 2);
1463 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1466 if (!(rw
& REQ_WRITE
)) {
1467 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1468 return btrfs_submit_compressed_read(inode
, bio
,
1469 mirror_num
, bio_flags
);
1470 } else if (!skip_sum
) {
1471 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1476 } else if (!skip_sum
) {
1477 /* csum items have already been cloned */
1478 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1480 /* we're doing a write, do the async checksumming */
1481 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1482 inode
, rw
, bio
, mirror_num
,
1483 bio_flags
, bio_offset
,
1484 __btrfs_submit_bio_start
,
1485 __btrfs_submit_bio_done
);
1489 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1493 * given a list of ordered sums record them in the inode. This happens
1494 * at IO completion time based on sums calculated at bio submission time.
1496 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1497 struct inode
*inode
, u64 file_offset
,
1498 struct list_head
*list
)
1500 struct btrfs_ordered_sum
*sum
;
1502 btrfs_set_trans_block_group(trans
, inode
);
1504 list_for_each_entry(sum
, list
, list
) {
1505 btrfs_csum_file_blocks(trans
,
1506 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1511 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1512 struct extent_state
**cached_state
)
1514 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1516 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1517 cached_state
, GFP_NOFS
);
1520 /* see btrfs_writepage_start_hook for details on why this is required */
1521 struct btrfs_writepage_fixup
{
1523 struct btrfs_work work
;
1526 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1528 struct btrfs_writepage_fixup
*fixup
;
1529 struct btrfs_ordered_extent
*ordered
;
1530 struct extent_state
*cached_state
= NULL
;
1532 struct inode
*inode
;
1536 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1540 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1541 ClearPageChecked(page
);
1545 inode
= page
->mapping
->host
;
1546 page_start
= page_offset(page
);
1547 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1549 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1550 &cached_state
, GFP_NOFS
);
1552 /* already ordered? We're done */
1553 if (PagePrivate2(page
))
1556 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1558 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1559 page_end
, &cached_state
, GFP_NOFS
);
1561 btrfs_start_ordered_extent(inode
, ordered
, 1);
1566 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1567 ClearPageChecked(page
);
1569 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1570 &cached_state
, GFP_NOFS
);
1573 page_cache_release(page
);
1578 * There are a few paths in the higher layers of the kernel that directly
1579 * set the page dirty bit without asking the filesystem if it is a
1580 * good idea. This causes problems because we want to make sure COW
1581 * properly happens and the data=ordered rules are followed.
1583 * In our case any range that doesn't have the ORDERED bit set
1584 * hasn't been properly setup for IO. We kick off an async process
1585 * to fix it up. The async helper will wait for ordered extents, set
1586 * the delalloc bit and make it safe to write the page.
1588 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1590 struct inode
*inode
= page
->mapping
->host
;
1591 struct btrfs_writepage_fixup
*fixup
;
1592 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1594 /* this page is properly in the ordered list */
1595 if (TestClearPagePrivate2(page
))
1598 if (PageChecked(page
))
1601 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1605 SetPageChecked(page
);
1606 page_cache_get(page
);
1607 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1609 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1613 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1614 struct inode
*inode
, u64 file_pos
,
1615 u64 disk_bytenr
, u64 disk_num_bytes
,
1616 u64 num_bytes
, u64 ram_bytes
,
1617 u8 compression
, u8 encryption
,
1618 u16 other_encoding
, int extent_type
)
1620 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1621 struct btrfs_file_extent_item
*fi
;
1622 struct btrfs_path
*path
;
1623 struct extent_buffer
*leaf
;
1624 struct btrfs_key ins
;
1628 path
= btrfs_alloc_path();
1631 path
->leave_spinning
= 1;
1634 * we may be replacing one extent in the tree with another.
1635 * The new extent is pinned in the extent map, and we don't want
1636 * to drop it from the cache until it is completely in the btree.
1638 * So, tell btrfs_drop_extents to leave this extent in the cache.
1639 * the caller is expected to unpin it and allow it to be merged
1642 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1646 ins
.objectid
= inode
->i_ino
;
1647 ins
.offset
= file_pos
;
1648 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1649 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1651 leaf
= path
->nodes
[0];
1652 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1653 struct btrfs_file_extent_item
);
1654 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1655 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1656 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1657 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1658 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1659 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1660 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1661 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1662 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1663 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1665 btrfs_unlock_up_safe(path
, 1);
1666 btrfs_set_lock_blocking(leaf
);
1668 btrfs_mark_buffer_dirty(leaf
);
1670 inode_add_bytes(inode
, num_bytes
);
1672 ins
.objectid
= disk_bytenr
;
1673 ins
.offset
= disk_num_bytes
;
1674 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1675 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1676 root
->root_key
.objectid
,
1677 inode
->i_ino
, file_pos
, &ins
);
1679 btrfs_free_path(path
);
1685 * helper function for btrfs_finish_ordered_io, this
1686 * just reads in some of the csum leaves to prime them into ram
1687 * before we start the transaction. It limits the amount of btree
1688 * reads required while inside the transaction.
1690 /* as ordered data IO finishes, this gets called so we can finish
1691 * an ordered extent if the range of bytes in the file it covers are
1694 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1696 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1697 struct btrfs_trans_handle
*trans
= NULL
;
1698 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1699 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1700 struct extent_state
*cached_state
= NULL
;
1701 int compress_type
= 0;
1703 bool nolock
= false;
1705 ret
= btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1709 BUG_ON(!ordered_extent
);
1711 nolock
= (root
== root
->fs_info
->tree_root
);
1713 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1714 BUG_ON(!list_empty(&ordered_extent
->list
));
1715 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1718 trans
= btrfs_join_transaction_nolock(root
, 1);
1720 trans
= btrfs_join_transaction(root
, 1);
1721 BUG_ON(IS_ERR(trans
));
1722 btrfs_set_trans_block_group(trans
, inode
);
1723 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1724 ret
= btrfs_update_inode(trans
, root
, inode
);
1730 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1731 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1732 0, &cached_state
, GFP_NOFS
);
1735 trans
= btrfs_join_transaction_nolock(root
, 1);
1737 trans
= btrfs_join_transaction(root
, 1);
1738 BUG_ON(IS_ERR(trans
));
1739 btrfs_set_trans_block_group(trans
, inode
);
1740 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1742 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1743 compress_type
= ordered_extent
->compress_type
;
1744 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1745 BUG_ON(compress_type
);
1746 ret
= btrfs_mark_extent_written(trans
, inode
,
1747 ordered_extent
->file_offset
,
1748 ordered_extent
->file_offset
+
1749 ordered_extent
->len
);
1752 BUG_ON(root
== root
->fs_info
->tree_root
);
1753 ret
= insert_reserved_file_extent(trans
, inode
,
1754 ordered_extent
->file_offset
,
1755 ordered_extent
->start
,
1756 ordered_extent
->disk_len
,
1757 ordered_extent
->len
,
1758 ordered_extent
->len
,
1759 compress_type
, 0, 0,
1760 BTRFS_FILE_EXTENT_REG
);
1761 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1762 ordered_extent
->file_offset
,
1763 ordered_extent
->len
);
1766 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1767 ordered_extent
->file_offset
+
1768 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1770 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1771 &ordered_extent
->list
);
1773 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1774 ret
= btrfs_update_inode(trans
, root
, inode
);
1779 btrfs_end_transaction_nolock(trans
, root
);
1781 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1783 btrfs_end_transaction(trans
, root
);
1787 btrfs_put_ordered_extent(ordered_extent
);
1788 /* once for the tree */
1789 btrfs_put_ordered_extent(ordered_extent
);
1794 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1795 struct extent_state
*state
, int uptodate
)
1797 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1799 ClearPagePrivate2(page
);
1800 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1804 * When IO fails, either with EIO or csum verification fails, we
1805 * try other mirrors that might have a good copy of the data. This
1806 * io_failure_record is used to record state as we go through all the
1807 * mirrors. If another mirror has good data, the page is set up to date
1808 * and things continue. If a good mirror can't be found, the original
1809 * bio end_io callback is called to indicate things have failed.
1811 struct io_failure_record
{
1816 unsigned long bio_flags
;
1820 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1821 struct page
*page
, u64 start
, u64 end
,
1822 struct extent_state
*state
)
1824 struct io_failure_record
*failrec
= NULL
;
1826 struct extent_map
*em
;
1827 struct inode
*inode
= page
->mapping
->host
;
1828 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1829 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1836 ret
= get_state_private(failure_tree
, start
, &private);
1838 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1841 failrec
->start
= start
;
1842 failrec
->len
= end
- start
+ 1;
1843 failrec
->last_mirror
= 0;
1844 failrec
->bio_flags
= 0;
1846 read_lock(&em_tree
->lock
);
1847 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1848 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1849 free_extent_map(em
);
1852 read_unlock(&em_tree
->lock
);
1854 if (!em
|| IS_ERR(em
)) {
1858 logical
= start
- em
->start
;
1859 logical
= em
->block_start
+ logical
;
1860 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1861 logical
= em
->block_start
;
1862 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1863 extent_set_compress_type(&failrec
->bio_flags
,
1866 failrec
->logical
= logical
;
1867 free_extent_map(em
);
1868 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1869 EXTENT_DIRTY
, GFP_NOFS
);
1870 set_state_private(failure_tree
, start
,
1871 (u64
)(unsigned long)failrec
);
1873 failrec
= (struct io_failure_record
*)(unsigned long)private;
1875 num_copies
= btrfs_num_copies(
1876 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1877 failrec
->logical
, failrec
->len
);
1878 failrec
->last_mirror
++;
1880 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1881 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1884 if (state
&& state
->start
!= failrec
->start
)
1886 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1888 if (!state
|| failrec
->last_mirror
> num_copies
) {
1889 set_state_private(failure_tree
, failrec
->start
, 0);
1890 clear_extent_bits(failure_tree
, failrec
->start
,
1891 failrec
->start
+ failrec
->len
- 1,
1892 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1896 bio
= bio_alloc(GFP_NOFS
, 1);
1897 bio
->bi_private
= state
;
1898 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1899 bio
->bi_sector
= failrec
->logical
>> 9;
1900 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1903 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1904 if (failed_bio
->bi_rw
& REQ_WRITE
)
1909 ret
= BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1910 failrec
->last_mirror
,
1911 failrec
->bio_flags
, 0);
1916 * each time an IO finishes, we do a fast check in the IO failure tree
1917 * to see if we need to process or clean up an io_failure_record
1919 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1922 u64 private_failure
;
1923 struct io_failure_record
*failure
;
1927 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1928 (u64
)-1, 1, EXTENT_DIRTY
, 0)) {
1929 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1930 start
, &private_failure
);
1932 failure
= (struct io_failure_record
*)(unsigned long)
1934 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1936 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1938 failure
->start
+ failure
->len
- 1,
1939 EXTENT_DIRTY
| EXTENT_LOCKED
,
1948 * when reads are done, we need to check csums to verify the data is correct
1949 * if there's a match, we allow the bio to finish. If not, we go through
1950 * the io_failure_record routines to find good copies
1952 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1953 struct extent_state
*state
)
1955 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1956 struct inode
*inode
= page
->mapping
->host
;
1957 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1959 u64
private = ~(u32
)0;
1961 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1964 if (PageChecked(page
)) {
1965 ClearPageChecked(page
);
1969 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1972 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1973 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1974 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1979 if (state
&& state
->start
== start
) {
1980 private = state
->private;
1983 ret
= get_state_private(io_tree
, start
, &private);
1985 kaddr
= kmap_atomic(page
, KM_USER0
);
1989 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1990 btrfs_csum_final(csum
, (char *)&csum
);
1991 if (csum
!= private)
1994 kunmap_atomic(kaddr
, KM_USER0
);
1996 /* if the io failure tree for this inode is non-empty,
1997 * check to see if we've recovered from a failed IO
1999 btrfs_clean_io_failures(inode
, start
);
2003 if (printk_ratelimit()) {
2004 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
2005 "private %llu\n", page
->mapping
->host
->i_ino
,
2006 (unsigned long long)start
, csum
,
2007 (unsigned long long)private);
2009 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2010 flush_dcache_page(page
);
2011 kunmap_atomic(kaddr
, KM_USER0
);
2017 struct delayed_iput
{
2018 struct list_head list
;
2019 struct inode
*inode
;
2022 void btrfs_add_delayed_iput(struct inode
*inode
)
2024 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2025 struct delayed_iput
*delayed
;
2027 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2030 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2031 delayed
->inode
= inode
;
2033 spin_lock(&fs_info
->delayed_iput_lock
);
2034 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2035 spin_unlock(&fs_info
->delayed_iput_lock
);
2038 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2041 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2042 struct delayed_iput
*delayed
;
2045 spin_lock(&fs_info
->delayed_iput_lock
);
2046 empty
= list_empty(&fs_info
->delayed_iputs
);
2047 spin_unlock(&fs_info
->delayed_iput_lock
);
2051 down_read(&root
->fs_info
->cleanup_work_sem
);
2052 spin_lock(&fs_info
->delayed_iput_lock
);
2053 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2054 spin_unlock(&fs_info
->delayed_iput_lock
);
2056 while (!list_empty(&list
)) {
2057 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2058 list_del(&delayed
->list
);
2059 iput(delayed
->inode
);
2062 up_read(&root
->fs_info
->cleanup_work_sem
);
2066 * calculate extra metadata reservation when snapshotting a subvolume
2067 * contains orphan files.
2069 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle
*trans
,
2070 struct btrfs_pending_snapshot
*pending
,
2071 u64
*bytes_to_reserve
)
2073 struct btrfs_root
*root
;
2074 struct btrfs_block_rsv
*block_rsv
;
2078 root
= pending
->root
;
2079 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2082 block_rsv
= root
->orphan_block_rsv
;
2084 /* orphan block reservation for the snapshot */
2085 num_bytes
= block_rsv
->size
;
2088 * after the snapshot is created, COWing tree blocks may use more
2089 * space than it frees. So we should make sure there is enough
2092 index
= trans
->transid
& 0x1;
2093 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2094 num_bytes
+= block_rsv
->size
-
2095 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2098 *bytes_to_reserve
+= num_bytes
;
2101 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle
*trans
,
2102 struct btrfs_pending_snapshot
*pending
)
2104 struct btrfs_root
*root
= pending
->root
;
2105 struct btrfs_root
*snap
= pending
->snap
;
2106 struct btrfs_block_rsv
*block_rsv
;
2111 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2114 /* refill source subvolume's orphan block reservation */
2115 block_rsv
= root
->orphan_block_rsv
;
2116 index
= trans
->transid
& 0x1;
2117 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2118 num_bytes
= block_rsv
->size
-
2119 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2120 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2121 root
->orphan_block_rsv
,
2126 /* setup orphan block reservation for the snapshot */
2127 block_rsv
= btrfs_alloc_block_rsv(snap
);
2130 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2131 snap
->orphan_block_rsv
= block_rsv
;
2133 num_bytes
= root
->orphan_block_rsv
->size
;
2134 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2135 block_rsv
, num_bytes
);
2139 /* insert orphan item for the snapshot */
2140 WARN_ON(!root
->orphan_item_inserted
);
2141 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2142 snap
->root_key
.objectid
);
2144 snap
->orphan_item_inserted
= 1;
2148 enum btrfs_orphan_cleanup_state
{
2149 ORPHAN_CLEANUP_STARTED
= 1,
2150 ORPHAN_CLEANUP_DONE
= 2,
2154 * This is called in transaction commmit time. If there are no orphan
2155 * files in the subvolume, it removes orphan item and frees block_rsv
2158 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2159 struct btrfs_root
*root
)
2163 if (!list_empty(&root
->orphan_list
) ||
2164 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2167 if (root
->orphan_item_inserted
&&
2168 btrfs_root_refs(&root
->root_item
) > 0) {
2169 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2170 root
->root_key
.objectid
);
2172 root
->orphan_item_inserted
= 0;
2175 if (root
->orphan_block_rsv
) {
2176 WARN_ON(root
->orphan_block_rsv
->size
> 0);
2177 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
2178 root
->orphan_block_rsv
= NULL
;
2183 * This creates an orphan entry for the given inode in case something goes
2184 * wrong in the middle of an unlink/truncate.
2186 * NOTE: caller of this function should reserve 5 units of metadata for
2189 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2191 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2192 struct btrfs_block_rsv
*block_rsv
= NULL
;
2197 if (!root
->orphan_block_rsv
) {
2198 block_rsv
= btrfs_alloc_block_rsv(root
);
2202 spin_lock(&root
->orphan_lock
);
2203 if (!root
->orphan_block_rsv
) {
2204 root
->orphan_block_rsv
= block_rsv
;
2205 } else if (block_rsv
) {
2206 btrfs_free_block_rsv(root
, block_rsv
);
2210 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2211 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2214 * For proper ENOSPC handling, we should do orphan
2215 * cleanup when mounting. But this introduces backward
2216 * compatibility issue.
2218 if (!xchg(&root
->orphan_item_inserted
, 1))
2225 WARN_ON(!BTRFS_I(inode
)->orphan_meta_reserved
);
2228 if (!BTRFS_I(inode
)->orphan_meta_reserved
) {
2229 BTRFS_I(inode
)->orphan_meta_reserved
= 1;
2232 spin_unlock(&root
->orphan_lock
);
2235 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2237 /* grab metadata reservation from transaction handle */
2239 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2243 /* insert an orphan item to track this unlinked/truncated file */
2245 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
2249 /* insert an orphan item to track subvolume contains orphan files */
2251 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2252 root
->root_key
.objectid
);
2259 * We have done the truncate/delete so we can go ahead and remove the orphan
2260 * item for this particular inode.
2262 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2264 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2265 int delete_item
= 0;
2266 int release_rsv
= 0;
2269 spin_lock(&root
->orphan_lock
);
2270 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2271 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2275 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2276 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2279 spin_unlock(&root
->orphan_lock
);
2281 if (trans
&& delete_item
) {
2282 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
2287 btrfs_orphan_release_metadata(inode
);
2293 * this cleans up any orphans that may be left on the list from the last use
2296 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2298 struct btrfs_path
*path
;
2299 struct extent_buffer
*leaf
;
2300 struct btrfs_key key
, found_key
;
2301 struct btrfs_trans_handle
*trans
;
2302 struct inode
*inode
;
2303 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2305 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2308 path
= btrfs_alloc_path();
2315 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2316 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2317 key
.offset
= (u64
)-1;
2320 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2325 * if ret == 0 means we found what we were searching for, which
2326 * is weird, but possible, so only screw with path if we didnt
2327 * find the key and see if we have stuff that matches
2331 if (path
->slots
[0] == 0)
2336 /* pull out the item */
2337 leaf
= path
->nodes
[0];
2338 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2340 /* make sure the item matches what we want */
2341 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2343 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2346 /* release the path since we're done with it */
2347 btrfs_release_path(root
, path
);
2350 * this is where we are basically btrfs_lookup, without the
2351 * crossing root thing. we store the inode number in the
2352 * offset of the orphan item.
2354 found_key
.objectid
= found_key
.offset
;
2355 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2356 found_key
.offset
= 0;
2357 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2358 if (IS_ERR(inode
)) {
2359 ret
= PTR_ERR(inode
);
2364 * add this inode to the orphan list so btrfs_orphan_del does
2365 * the proper thing when we hit it
2367 spin_lock(&root
->orphan_lock
);
2368 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2369 spin_unlock(&root
->orphan_lock
);
2372 * if this is a bad inode, means we actually succeeded in
2373 * removing the inode, but not the orphan record, which means
2374 * we need to manually delete the orphan since iput will just
2375 * do a destroy_inode
2377 if (is_bad_inode(inode
)) {
2378 trans
= btrfs_start_transaction(root
, 0);
2379 if (IS_ERR(trans
)) {
2380 ret
= PTR_ERR(trans
);
2383 btrfs_orphan_del(trans
, inode
);
2384 btrfs_end_transaction(trans
, root
);
2389 /* if we have links, this was a truncate, lets do that */
2390 if (inode
->i_nlink
) {
2391 if (!S_ISREG(inode
->i_mode
)) {
2397 ret
= btrfs_truncate(inode
);
2402 /* this will do delete_inode and everything for us */
2407 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2409 if (root
->orphan_block_rsv
)
2410 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2413 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2414 trans
= btrfs_join_transaction(root
, 1);
2416 btrfs_end_transaction(trans
, root
);
2420 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2422 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2426 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2427 btrfs_free_path(path
);
2432 * very simple check to peek ahead in the leaf looking for xattrs. If we
2433 * don't find any xattrs, we know there can't be any acls.
2435 * slot is the slot the inode is in, objectid is the objectid of the inode
2437 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2438 int slot
, u64 objectid
)
2440 u32 nritems
= btrfs_header_nritems(leaf
);
2441 struct btrfs_key found_key
;
2445 while (slot
< nritems
) {
2446 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2448 /* we found a different objectid, there must not be acls */
2449 if (found_key
.objectid
!= objectid
)
2452 /* we found an xattr, assume we've got an acl */
2453 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2457 * we found a key greater than an xattr key, there can't
2458 * be any acls later on
2460 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2467 * it goes inode, inode backrefs, xattrs, extents,
2468 * so if there are a ton of hard links to an inode there can
2469 * be a lot of backrefs. Don't waste time searching too hard,
2470 * this is just an optimization
2475 /* we hit the end of the leaf before we found an xattr or
2476 * something larger than an xattr. We have to assume the inode
2483 * read an inode from the btree into the in-memory inode
2485 static void btrfs_read_locked_inode(struct inode
*inode
)
2487 struct btrfs_path
*path
;
2488 struct extent_buffer
*leaf
;
2489 struct btrfs_inode_item
*inode_item
;
2490 struct btrfs_timespec
*tspec
;
2491 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2492 struct btrfs_key location
;
2494 u64 alloc_group_block
;
2498 path
= btrfs_alloc_path();
2500 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2502 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2506 leaf
= path
->nodes
[0];
2507 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2508 struct btrfs_inode_item
);
2510 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2511 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2512 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2513 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2514 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2516 tspec
= btrfs_inode_atime(inode_item
);
2517 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2518 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2520 tspec
= btrfs_inode_mtime(inode_item
);
2521 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2522 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2524 tspec
= btrfs_inode_ctime(inode_item
);
2525 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2526 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2528 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2529 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2530 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2531 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2533 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2535 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2536 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2538 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2541 * try to precache a NULL acl entry for files that don't have
2542 * any xattrs or acls
2544 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0], inode
->i_ino
);
2546 cache_no_acl(inode
);
2548 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2549 alloc_group_block
, 0);
2550 btrfs_free_path(path
);
2553 switch (inode
->i_mode
& S_IFMT
) {
2555 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2556 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2557 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2558 inode
->i_fop
= &btrfs_file_operations
;
2559 inode
->i_op
= &btrfs_file_inode_operations
;
2562 inode
->i_fop
= &btrfs_dir_file_operations
;
2563 if (root
== root
->fs_info
->tree_root
)
2564 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2566 inode
->i_op
= &btrfs_dir_inode_operations
;
2569 inode
->i_op
= &btrfs_symlink_inode_operations
;
2570 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2571 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2574 inode
->i_op
= &btrfs_special_inode_operations
;
2575 init_special_inode(inode
, inode
->i_mode
, rdev
);
2579 btrfs_update_iflags(inode
);
2583 btrfs_free_path(path
);
2584 make_bad_inode(inode
);
2588 * given a leaf and an inode, copy the inode fields into the leaf
2590 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2591 struct extent_buffer
*leaf
,
2592 struct btrfs_inode_item
*item
,
2593 struct inode
*inode
)
2595 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2596 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2597 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2598 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2599 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2601 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2602 inode
->i_atime
.tv_sec
);
2603 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2604 inode
->i_atime
.tv_nsec
);
2606 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2607 inode
->i_mtime
.tv_sec
);
2608 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2609 inode
->i_mtime
.tv_nsec
);
2611 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2612 inode
->i_ctime
.tv_sec
);
2613 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2614 inode
->i_ctime
.tv_nsec
);
2616 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2617 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2618 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2619 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2620 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2621 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2622 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2626 * copy everything in the in-memory inode into the btree.
2628 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2629 struct btrfs_root
*root
, struct inode
*inode
)
2631 struct btrfs_inode_item
*inode_item
;
2632 struct btrfs_path
*path
;
2633 struct extent_buffer
*leaf
;
2636 path
= btrfs_alloc_path();
2638 path
->leave_spinning
= 1;
2639 ret
= btrfs_lookup_inode(trans
, root
, path
,
2640 &BTRFS_I(inode
)->location
, 1);
2647 btrfs_unlock_up_safe(path
, 1);
2648 leaf
= path
->nodes
[0];
2649 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2650 struct btrfs_inode_item
);
2652 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2653 btrfs_mark_buffer_dirty(leaf
);
2654 btrfs_set_inode_last_trans(trans
, inode
);
2657 btrfs_free_path(path
);
2663 * unlink helper that gets used here in inode.c and in the tree logging
2664 * recovery code. It remove a link in a directory with a given name, and
2665 * also drops the back refs in the inode to the directory
2667 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2668 struct btrfs_root
*root
,
2669 struct inode
*dir
, struct inode
*inode
,
2670 const char *name
, int name_len
)
2672 struct btrfs_path
*path
;
2674 struct extent_buffer
*leaf
;
2675 struct btrfs_dir_item
*di
;
2676 struct btrfs_key key
;
2679 path
= btrfs_alloc_path();
2685 path
->leave_spinning
= 1;
2686 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2687 name
, name_len
, -1);
2696 leaf
= path
->nodes
[0];
2697 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2698 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2701 btrfs_release_path(root
, path
);
2703 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2705 dir
->i_ino
, &index
);
2707 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2708 "inode %lu parent %lu\n", name_len
, name
,
2709 inode
->i_ino
, dir
->i_ino
);
2713 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2714 index
, name
, name_len
, -1);
2723 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2724 btrfs_release_path(root
, path
);
2726 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2728 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2730 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2735 btrfs_free_path(path
);
2739 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2740 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2741 btrfs_update_inode(trans
, root
, dir
);
2742 btrfs_drop_nlink(inode
);
2743 ret
= btrfs_update_inode(trans
, root
, inode
);
2748 /* helper to check if there is any shared block in the path */
2749 static int check_path_shared(struct btrfs_root
*root
,
2750 struct btrfs_path
*path
)
2752 struct extent_buffer
*eb
;
2756 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2759 if (!path
->nodes
[level
])
2761 eb
= path
->nodes
[level
];
2762 if (!btrfs_block_can_be_shared(root
, eb
))
2764 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2773 * helper to start transaction for unlink and rmdir.
2775 * unlink and rmdir are special in btrfs, they do not always free space.
2776 * so in enospc case, we should make sure they will free space before
2777 * allowing them to use the global metadata reservation.
2779 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2780 struct dentry
*dentry
)
2782 struct btrfs_trans_handle
*trans
;
2783 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2784 struct btrfs_path
*path
;
2785 struct btrfs_inode_ref
*ref
;
2786 struct btrfs_dir_item
*di
;
2787 struct inode
*inode
= dentry
->d_inode
;
2793 trans
= btrfs_start_transaction(root
, 10);
2794 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2797 if (inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2798 return ERR_PTR(-ENOSPC
);
2800 /* check if there is someone else holds reference */
2801 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2802 return ERR_PTR(-ENOSPC
);
2804 if (atomic_read(&inode
->i_count
) > 2)
2805 return ERR_PTR(-ENOSPC
);
2807 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2808 return ERR_PTR(-ENOSPC
);
2810 path
= btrfs_alloc_path();
2812 root
->fs_info
->enospc_unlink
= 0;
2813 return ERR_PTR(-ENOMEM
);
2816 trans
= btrfs_start_transaction(root
, 0);
2817 if (IS_ERR(trans
)) {
2818 btrfs_free_path(path
);
2819 root
->fs_info
->enospc_unlink
= 0;
2823 path
->skip_locking
= 1;
2824 path
->search_commit_root
= 1;
2826 ret
= btrfs_lookup_inode(trans
, root
, path
,
2827 &BTRFS_I(dir
)->location
, 0);
2833 if (check_path_shared(root
, path
))
2838 btrfs_release_path(root
, path
);
2840 ret
= btrfs_lookup_inode(trans
, root
, path
,
2841 &BTRFS_I(inode
)->location
, 0);
2847 if (check_path_shared(root
, path
))
2852 btrfs_release_path(root
, path
);
2854 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2855 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2856 inode
->i_ino
, (u64
)-1, 0);
2862 if (check_path_shared(root
, path
))
2864 btrfs_release_path(root
, path
);
2872 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2873 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2879 if (check_path_shared(root
, path
))
2885 btrfs_release_path(root
, path
);
2887 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2888 dentry
->d_name
.name
, dentry
->d_name
.len
,
2889 inode
->i_ino
, dir
->i_ino
, 0);
2895 if (check_path_shared(root
, path
))
2897 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2898 btrfs_release_path(root
, path
);
2900 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
, index
,
2901 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2906 BUG_ON(ret
== -ENOENT
);
2907 if (check_path_shared(root
, path
))
2912 btrfs_free_path(path
);
2914 btrfs_end_transaction(trans
, root
);
2915 root
->fs_info
->enospc_unlink
= 0;
2916 return ERR_PTR(err
);
2919 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2923 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2924 struct btrfs_root
*root
)
2926 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2927 BUG_ON(!root
->fs_info
->enospc_unlink
);
2928 root
->fs_info
->enospc_unlink
= 0;
2930 btrfs_end_transaction_throttle(trans
, root
);
2933 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2935 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2936 struct btrfs_trans_handle
*trans
;
2937 struct inode
*inode
= dentry
->d_inode
;
2939 unsigned long nr
= 0;
2941 trans
= __unlink_start_trans(dir
, dentry
);
2943 return PTR_ERR(trans
);
2945 btrfs_set_trans_block_group(trans
, dir
);
2947 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2949 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2950 dentry
->d_name
.name
, dentry
->d_name
.len
);
2953 if (inode
->i_nlink
== 0) {
2954 ret
= btrfs_orphan_add(trans
, inode
);
2958 nr
= trans
->blocks_used
;
2959 __unlink_end_trans(trans
, root
);
2960 btrfs_btree_balance_dirty(root
, nr
);
2964 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2965 struct btrfs_root
*root
,
2966 struct inode
*dir
, u64 objectid
,
2967 const char *name
, int name_len
)
2969 struct btrfs_path
*path
;
2970 struct extent_buffer
*leaf
;
2971 struct btrfs_dir_item
*di
;
2972 struct btrfs_key key
;
2976 path
= btrfs_alloc_path();
2980 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2981 name
, name_len
, -1);
2982 BUG_ON(!di
|| IS_ERR(di
));
2984 leaf
= path
->nodes
[0];
2985 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2986 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
2987 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2989 btrfs_release_path(root
, path
);
2991 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
2992 objectid
, root
->root_key
.objectid
,
2993 dir
->i_ino
, &index
, name
, name_len
);
2995 BUG_ON(ret
!= -ENOENT
);
2996 di
= btrfs_search_dir_index_item(root
, path
, dir
->i_ino
,
2998 BUG_ON(!di
|| IS_ERR(di
));
3000 leaf
= path
->nodes
[0];
3001 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3002 btrfs_release_path(root
, path
);
3006 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
3007 index
, name
, name_len
, -1);
3008 BUG_ON(!di
|| IS_ERR(di
));
3010 leaf
= path
->nodes
[0];
3011 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3012 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3013 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3015 btrfs_release_path(root
, path
);
3017 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3018 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3019 ret
= btrfs_update_inode(trans
, root
, dir
);
3022 btrfs_free_path(path
);
3026 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3028 struct inode
*inode
= dentry
->d_inode
;
3030 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3031 struct btrfs_trans_handle
*trans
;
3032 unsigned long nr
= 0;
3034 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3035 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
3038 trans
= __unlink_start_trans(dir
, dentry
);
3040 return PTR_ERR(trans
);
3042 btrfs_set_trans_block_group(trans
, dir
);
3044 if (unlikely(inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3045 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3046 BTRFS_I(inode
)->location
.objectid
,
3047 dentry
->d_name
.name
,
3048 dentry
->d_name
.len
);
3052 err
= btrfs_orphan_add(trans
, inode
);
3056 /* now the directory is empty */
3057 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3058 dentry
->d_name
.name
, dentry
->d_name
.len
);
3060 btrfs_i_size_write(inode
, 0);
3062 nr
= trans
->blocks_used
;
3063 __unlink_end_trans(trans
, root
);
3064 btrfs_btree_balance_dirty(root
, nr
);
3071 * when truncating bytes in a file, it is possible to avoid reading
3072 * the leaves that contain only checksum items. This can be the
3073 * majority of the IO required to delete a large file, but it must
3074 * be done carefully.
3076 * The keys in the level just above the leaves are checked to make sure
3077 * the lowest key in a given leaf is a csum key, and starts at an offset
3078 * after the new size.
3080 * Then the key for the next leaf is checked to make sure it also has
3081 * a checksum item for the same file. If it does, we know our target leaf
3082 * contains only checksum items, and it can be safely freed without reading
3085 * This is just an optimization targeted at large files. It may do
3086 * nothing. It will return 0 unless things went badly.
3088 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
3089 struct btrfs_root
*root
,
3090 struct btrfs_path
*path
,
3091 struct inode
*inode
, u64 new_size
)
3093 struct btrfs_key key
;
3096 struct btrfs_key found_key
;
3097 struct btrfs_key other_key
;
3098 struct btrfs_leaf_ref
*ref
;
3102 path
->lowest_level
= 1;
3103 key
.objectid
= inode
->i_ino
;
3104 key
.type
= BTRFS_CSUM_ITEM_KEY
;
3105 key
.offset
= new_size
;
3107 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3111 if (path
->nodes
[1] == NULL
) {
3116 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
3117 nritems
= btrfs_header_nritems(path
->nodes
[1]);
3122 if (path
->slots
[1] >= nritems
)
3125 /* did we find a key greater than anything we want to delete? */
3126 if (found_key
.objectid
> inode
->i_ino
||
3127 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
3130 /* we check the next key in the node to make sure the leave contains
3131 * only checksum items. This comparison doesn't work if our
3132 * leaf is the last one in the node
3134 if (path
->slots
[1] + 1 >= nritems
) {
3136 /* search forward from the last key in the node, this
3137 * will bring us into the next node in the tree
3139 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
3141 /* unlikely, but we inc below, so check to be safe */
3142 if (found_key
.offset
== (u64
)-1)
3145 /* search_forward needs a path with locks held, do the
3146 * search again for the original key. It is possible
3147 * this will race with a balance and return a path that
3148 * we could modify, but this drop is just an optimization
3149 * and is allowed to miss some leaves.
3151 btrfs_release_path(root
, path
);
3154 /* setup a max key for search_forward */
3155 other_key
.offset
= (u64
)-1;
3156 other_key
.type
= key
.type
;
3157 other_key
.objectid
= key
.objectid
;
3159 path
->keep_locks
= 1;
3160 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
3162 path
->keep_locks
= 0;
3163 if (ret
|| found_key
.objectid
!= key
.objectid
||
3164 found_key
.type
!= key
.type
) {
3169 key
.offset
= found_key
.offset
;
3170 btrfs_release_path(root
, path
);
3175 /* we know there's one more slot after us in the tree,
3176 * read that key so we can verify it is also a checksum item
3178 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
3180 if (found_key
.objectid
< inode
->i_ino
)
3183 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
3187 * if the key for the next leaf isn't a csum key from this objectid,
3188 * we can't be sure there aren't good items inside this leaf.
3191 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
3194 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
3195 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
3197 * it is safe to delete this leaf, it contains only
3198 * csum items from this inode at an offset >= new_size
3200 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
3203 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
3204 ref
= btrfs_alloc_leaf_ref(root
, 0);
3206 ref
->root_gen
= root
->root_key
.offset
;
3207 ref
->bytenr
= leaf_start
;
3209 ref
->generation
= leaf_gen
;
3212 btrfs_sort_leaf_ref(ref
);
3214 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
3216 btrfs_free_leaf_ref(root
, ref
);
3222 btrfs_release_path(root
, path
);
3224 if (other_key
.objectid
== inode
->i_ino
&&
3225 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
3226 key
.offset
= other_key
.offset
;
3232 /* fixup any changes we've made to the path */
3233 path
->lowest_level
= 0;
3234 path
->keep_locks
= 0;
3235 btrfs_release_path(root
, path
);
3242 * this can truncate away extent items, csum items and directory items.
3243 * It starts at a high offset and removes keys until it can't find
3244 * any higher than new_size
3246 * csum items that cross the new i_size are truncated to the new size
3249 * min_type is the minimum key type to truncate down to. If set to 0, this
3250 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3252 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3253 struct btrfs_root
*root
,
3254 struct inode
*inode
,
3255 u64 new_size
, u32 min_type
)
3257 struct btrfs_path
*path
;
3258 struct extent_buffer
*leaf
;
3259 struct btrfs_file_extent_item
*fi
;
3260 struct btrfs_key key
;
3261 struct btrfs_key found_key
;
3262 u64 extent_start
= 0;
3263 u64 extent_num_bytes
= 0;
3264 u64 extent_offset
= 0;
3266 u64 mask
= root
->sectorsize
- 1;
3267 u32 found_type
= (u8
)-1;
3270 int pending_del_nr
= 0;
3271 int pending_del_slot
= 0;
3272 int extent_type
= -1;
3277 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3279 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3280 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3282 path
= btrfs_alloc_path();
3286 key
.objectid
= inode
->i_ino
;
3287 key
.offset
= (u64
)-1;
3291 path
->leave_spinning
= 1;
3292 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3299 /* there are no items in the tree for us to truncate, we're
3302 if (path
->slots
[0] == 0)
3309 leaf
= path
->nodes
[0];
3310 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3311 found_type
= btrfs_key_type(&found_key
);
3314 if (found_key
.objectid
!= inode
->i_ino
)
3317 if (found_type
< min_type
)
3320 item_end
= found_key
.offset
;
3321 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3322 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3323 struct btrfs_file_extent_item
);
3324 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3325 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3326 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3327 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3329 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3331 btrfs_file_extent_num_bytes(leaf
, fi
);
3332 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3333 item_end
+= btrfs_file_extent_inline_len(leaf
,
3338 if (found_type
> min_type
) {
3341 if (item_end
< new_size
)
3343 if (found_key
.offset
>= new_size
)
3349 /* FIXME, shrink the extent if the ref count is only 1 */
3350 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3353 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3355 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3356 if (!del_item
&& !encoding
) {
3357 u64 orig_num_bytes
=
3358 btrfs_file_extent_num_bytes(leaf
, fi
);
3359 extent_num_bytes
= new_size
-
3360 found_key
.offset
+ root
->sectorsize
- 1;
3361 extent_num_bytes
= extent_num_bytes
&
3362 ~((u64
)root
->sectorsize
- 1);
3363 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3365 num_dec
= (orig_num_bytes
-
3367 if (root
->ref_cows
&& extent_start
!= 0)
3368 inode_sub_bytes(inode
, num_dec
);
3369 btrfs_mark_buffer_dirty(leaf
);
3372 btrfs_file_extent_disk_num_bytes(leaf
,
3374 extent_offset
= found_key
.offset
-
3375 btrfs_file_extent_offset(leaf
, fi
);
3377 /* FIXME blocksize != 4096 */
3378 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3379 if (extent_start
!= 0) {
3382 inode_sub_bytes(inode
, num_dec
);
3385 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3387 * we can't truncate inline items that have had
3391 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3392 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3393 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3394 u32 size
= new_size
- found_key
.offset
;
3396 if (root
->ref_cows
) {
3397 inode_sub_bytes(inode
, item_end
+ 1 -
3401 btrfs_file_extent_calc_inline_size(size
);
3402 ret
= btrfs_truncate_item(trans
, root
, path
,
3405 } else if (root
->ref_cows
) {
3406 inode_sub_bytes(inode
, item_end
+ 1 -
3412 if (!pending_del_nr
) {
3413 /* no pending yet, add ourselves */
3414 pending_del_slot
= path
->slots
[0];
3416 } else if (pending_del_nr
&&
3417 path
->slots
[0] + 1 == pending_del_slot
) {
3418 /* hop on the pending chunk */
3420 pending_del_slot
= path
->slots
[0];
3427 if (found_extent
&& (root
->ref_cows
||
3428 root
== root
->fs_info
->tree_root
)) {
3429 btrfs_set_path_blocking(path
);
3430 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3431 extent_num_bytes
, 0,
3432 btrfs_header_owner(leaf
),
3433 inode
->i_ino
, extent_offset
);
3437 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3440 if (path
->slots
[0] == 0 ||
3441 path
->slots
[0] != pending_del_slot
) {
3442 if (root
->ref_cows
) {
3446 if (pending_del_nr
) {
3447 ret
= btrfs_del_items(trans
, root
, path
,
3453 btrfs_release_path(root
, path
);
3460 if (pending_del_nr
) {
3461 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3465 btrfs_free_path(path
);
3470 * taken from block_truncate_page, but does cow as it zeros out
3471 * any bytes left in the last page in the file.
3473 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3475 struct inode
*inode
= mapping
->host
;
3476 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3477 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3478 struct btrfs_ordered_extent
*ordered
;
3479 struct extent_state
*cached_state
= NULL
;
3481 u32 blocksize
= root
->sectorsize
;
3482 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3483 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3489 if ((offset
& (blocksize
- 1)) == 0)
3491 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3497 page
= grab_cache_page(mapping
, index
);
3499 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3503 page_start
= page_offset(page
);
3504 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3506 if (!PageUptodate(page
)) {
3507 ret
= btrfs_readpage(NULL
, page
);
3509 if (page
->mapping
!= mapping
) {
3511 page_cache_release(page
);
3514 if (!PageUptodate(page
)) {
3519 wait_on_page_writeback(page
);
3521 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3523 set_page_extent_mapped(page
);
3525 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3527 unlock_extent_cached(io_tree
, page_start
, page_end
,
3528 &cached_state
, GFP_NOFS
);
3530 page_cache_release(page
);
3531 btrfs_start_ordered_extent(inode
, ordered
, 1);
3532 btrfs_put_ordered_extent(ordered
);
3536 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3537 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3538 0, 0, &cached_state
, GFP_NOFS
);
3540 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3543 unlock_extent_cached(io_tree
, page_start
, page_end
,
3544 &cached_state
, GFP_NOFS
);
3549 if (offset
!= PAGE_CACHE_SIZE
) {
3551 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3552 flush_dcache_page(page
);
3555 ClearPageChecked(page
);
3556 set_page_dirty(page
);
3557 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3562 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3564 page_cache_release(page
);
3570 * This function puts in dummy file extents for the area we're creating a hole
3571 * for. So if we are truncating this file to a larger size we need to insert
3572 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3573 * the range between oldsize and size
3575 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3577 struct btrfs_trans_handle
*trans
;
3578 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3579 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3580 struct extent_map
*em
= NULL
;
3581 struct extent_state
*cached_state
= NULL
;
3582 u64 mask
= root
->sectorsize
- 1;
3583 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3584 u64 block_end
= (size
+ mask
) & ~mask
;
3590 if (size
<= hole_start
)
3594 struct btrfs_ordered_extent
*ordered
;
3595 btrfs_wait_ordered_range(inode
, hole_start
,
3596 block_end
- hole_start
);
3597 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3598 &cached_state
, GFP_NOFS
);
3599 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3602 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3603 &cached_state
, GFP_NOFS
);
3604 btrfs_put_ordered_extent(ordered
);
3607 cur_offset
= hole_start
;
3609 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3610 block_end
- cur_offset
, 0);
3611 BUG_ON(IS_ERR(em
) || !em
);
3612 last_byte
= min(extent_map_end(em
), block_end
);
3613 last_byte
= (last_byte
+ mask
) & ~mask
;
3614 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3616 hole_size
= last_byte
- cur_offset
;
3618 trans
= btrfs_start_transaction(root
, 2);
3619 if (IS_ERR(trans
)) {
3620 err
= PTR_ERR(trans
);
3623 btrfs_set_trans_block_group(trans
, inode
);
3625 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3626 cur_offset
+ hole_size
,
3631 err
= btrfs_insert_file_extent(trans
, root
,
3632 inode
->i_ino
, cur_offset
, 0,
3633 0, hole_size
, 0, hole_size
,
3638 btrfs_drop_extent_cache(inode
, hole_start
,
3641 btrfs_end_transaction(trans
, root
);
3643 free_extent_map(em
);
3645 cur_offset
= last_byte
;
3646 if (cur_offset
>= block_end
)
3650 free_extent_map(em
);
3651 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3656 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3658 loff_t oldsize
= i_size_read(inode
);
3661 if (newsize
== oldsize
)
3664 if (newsize
> oldsize
) {
3665 i_size_write(inode
, newsize
);
3666 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3667 truncate_pagecache(inode
, oldsize
, newsize
);
3668 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3670 btrfs_setsize(inode
, oldsize
);
3674 mark_inode_dirty(inode
);
3678 * We're truncating a file that used to have good data down to
3679 * zero. Make sure it gets into the ordered flush list so that
3680 * any new writes get down to disk quickly.
3683 BTRFS_I(inode
)->ordered_data_close
= 1;
3685 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3686 truncate_setsize(inode
, newsize
);
3687 ret
= btrfs_truncate(inode
);
3693 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3695 struct inode
*inode
= dentry
->d_inode
;
3696 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3699 if (btrfs_root_readonly(root
))
3702 err
= inode_change_ok(inode
, attr
);
3706 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3707 err
= btrfs_setsize(inode
, attr
->ia_size
);
3712 if (attr
->ia_valid
) {
3713 setattr_copy(inode
, attr
);
3714 mark_inode_dirty(inode
);
3716 if (attr
->ia_valid
& ATTR_MODE
)
3717 err
= btrfs_acl_chmod(inode
);
3723 void btrfs_evict_inode(struct inode
*inode
)
3725 struct btrfs_trans_handle
*trans
;
3726 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3730 trace_btrfs_inode_evict(inode
);
3732 truncate_inode_pages(&inode
->i_data
, 0);
3733 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3734 root
== root
->fs_info
->tree_root
))
3737 if (is_bad_inode(inode
)) {
3738 btrfs_orphan_del(NULL
, inode
);
3741 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3742 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3744 if (root
->fs_info
->log_root_recovering
) {
3745 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3749 if (inode
->i_nlink
> 0) {
3750 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3754 btrfs_i_size_write(inode
, 0);
3757 trans
= btrfs_start_transaction(root
, 0);
3758 BUG_ON(IS_ERR(trans
));
3759 btrfs_set_trans_block_group(trans
, inode
);
3760 trans
->block_rsv
= root
->orphan_block_rsv
;
3762 ret
= btrfs_block_rsv_check(trans
, root
,
3763 root
->orphan_block_rsv
, 0, 5);
3765 BUG_ON(ret
!= -EAGAIN
);
3766 ret
= btrfs_commit_transaction(trans
, root
);
3771 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3775 nr
= trans
->blocks_used
;
3776 btrfs_end_transaction(trans
, root
);
3778 btrfs_btree_balance_dirty(root
, nr
);
3783 ret
= btrfs_orphan_del(trans
, inode
);
3787 nr
= trans
->blocks_used
;
3788 btrfs_end_transaction(trans
, root
);
3789 btrfs_btree_balance_dirty(root
, nr
);
3791 end_writeback(inode
);
3796 * this returns the key found in the dir entry in the location pointer.
3797 * If no dir entries were found, location->objectid is 0.
3799 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3800 struct btrfs_key
*location
)
3802 const char *name
= dentry
->d_name
.name
;
3803 int namelen
= dentry
->d_name
.len
;
3804 struct btrfs_dir_item
*di
;
3805 struct btrfs_path
*path
;
3806 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3809 path
= btrfs_alloc_path();
3812 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
3817 if (!di
|| IS_ERR(di
))
3820 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3822 btrfs_free_path(path
);
3825 location
->objectid
= 0;
3830 * when we hit a tree root in a directory, the btrfs part of the inode
3831 * needs to be changed to reflect the root directory of the tree root. This
3832 * is kind of like crossing a mount point.
3834 static int fixup_tree_root_location(struct btrfs_root
*root
,
3836 struct dentry
*dentry
,
3837 struct btrfs_key
*location
,
3838 struct btrfs_root
**sub_root
)
3840 struct btrfs_path
*path
;
3841 struct btrfs_root
*new_root
;
3842 struct btrfs_root_ref
*ref
;
3843 struct extent_buffer
*leaf
;
3847 path
= btrfs_alloc_path();
3854 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3855 BTRFS_I(dir
)->root
->root_key
.objectid
,
3856 location
->objectid
);
3863 leaf
= path
->nodes
[0];
3864 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3865 if (btrfs_root_ref_dirid(leaf
, ref
) != dir
->i_ino
||
3866 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3869 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3870 (unsigned long)(ref
+ 1),
3871 dentry
->d_name
.len
);
3875 btrfs_release_path(root
->fs_info
->tree_root
, path
);
3877 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3878 if (IS_ERR(new_root
)) {
3879 err
= PTR_ERR(new_root
);
3883 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3888 *sub_root
= new_root
;
3889 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3890 location
->type
= BTRFS_INODE_ITEM_KEY
;
3891 location
->offset
= 0;
3894 btrfs_free_path(path
);
3898 static void inode_tree_add(struct inode
*inode
)
3900 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3901 struct btrfs_inode
*entry
;
3903 struct rb_node
*parent
;
3905 p
= &root
->inode_tree
.rb_node
;
3908 if (inode_unhashed(inode
))
3911 spin_lock(&root
->inode_lock
);
3914 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3916 if (inode
->i_ino
< entry
->vfs_inode
.i_ino
)
3917 p
= &parent
->rb_left
;
3918 else if (inode
->i_ino
> entry
->vfs_inode
.i_ino
)
3919 p
= &parent
->rb_right
;
3921 WARN_ON(!(entry
->vfs_inode
.i_state
&
3922 (I_WILL_FREE
| I_FREEING
)));
3923 rb_erase(parent
, &root
->inode_tree
);
3924 RB_CLEAR_NODE(parent
);
3925 spin_unlock(&root
->inode_lock
);
3929 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3930 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3931 spin_unlock(&root
->inode_lock
);
3934 static void inode_tree_del(struct inode
*inode
)
3936 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3939 spin_lock(&root
->inode_lock
);
3940 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3941 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3942 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3943 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3945 spin_unlock(&root
->inode_lock
);
3948 * Free space cache has inodes in the tree root, but the tree root has a
3949 * root_refs of 0, so this could end up dropping the tree root as a
3950 * snapshot, so we need the extra !root->fs_info->tree_root check to
3951 * make sure we don't drop it.
3953 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3954 root
!= root
->fs_info
->tree_root
) {
3955 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3956 spin_lock(&root
->inode_lock
);
3957 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3958 spin_unlock(&root
->inode_lock
);
3960 btrfs_add_dead_root(root
);
3964 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3966 struct rb_node
*node
;
3967 struct rb_node
*prev
;
3968 struct btrfs_inode
*entry
;
3969 struct inode
*inode
;
3972 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3974 spin_lock(&root
->inode_lock
);
3976 node
= root
->inode_tree
.rb_node
;
3980 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3982 if (objectid
< entry
->vfs_inode
.i_ino
)
3983 node
= node
->rb_left
;
3984 else if (objectid
> entry
->vfs_inode
.i_ino
)
3985 node
= node
->rb_right
;
3991 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
3992 if (objectid
<= entry
->vfs_inode
.i_ino
) {
3996 prev
= rb_next(prev
);
4000 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4001 objectid
= entry
->vfs_inode
.i_ino
+ 1;
4002 inode
= igrab(&entry
->vfs_inode
);
4004 spin_unlock(&root
->inode_lock
);
4005 if (atomic_read(&inode
->i_count
) > 1)
4006 d_prune_aliases(inode
);
4008 * btrfs_drop_inode will have it removed from
4009 * the inode cache when its usage count
4014 spin_lock(&root
->inode_lock
);
4018 if (cond_resched_lock(&root
->inode_lock
))
4021 node
= rb_next(node
);
4023 spin_unlock(&root
->inode_lock
);
4027 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4029 struct btrfs_iget_args
*args
= p
;
4030 inode
->i_ino
= args
->ino
;
4031 BTRFS_I(inode
)->root
= args
->root
;
4032 btrfs_set_inode_space_info(args
->root
, inode
);
4036 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4038 struct btrfs_iget_args
*args
= opaque
;
4039 return args
->ino
== inode
->i_ino
&&
4040 args
->root
== BTRFS_I(inode
)->root
;
4043 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4045 struct btrfs_root
*root
)
4047 struct inode
*inode
;
4048 struct btrfs_iget_args args
;
4049 args
.ino
= objectid
;
4052 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4053 btrfs_init_locked_inode
,
4058 /* Get an inode object given its location and corresponding root.
4059 * Returns in *is_new if the inode was read from disk
4061 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4062 struct btrfs_root
*root
, int *new)
4064 struct inode
*inode
;
4066 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4068 return ERR_PTR(-ENOMEM
);
4070 if (inode
->i_state
& I_NEW
) {
4071 BTRFS_I(inode
)->root
= root
;
4072 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4073 btrfs_read_locked_inode(inode
);
4075 inode_tree_add(inode
);
4076 unlock_new_inode(inode
);
4084 static struct inode
*new_simple_dir(struct super_block
*s
,
4085 struct btrfs_key
*key
,
4086 struct btrfs_root
*root
)
4088 struct inode
*inode
= new_inode(s
);
4091 return ERR_PTR(-ENOMEM
);
4093 BTRFS_I(inode
)->root
= root
;
4094 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4095 BTRFS_I(inode
)->dummy_inode
= 1;
4097 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4098 inode
->i_op
= &simple_dir_inode_operations
;
4099 inode
->i_fop
= &simple_dir_operations
;
4100 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4101 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4106 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4108 struct inode
*inode
;
4109 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4110 struct btrfs_root
*sub_root
= root
;
4111 struct btrfs_key location
;
4115 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4116 return ERR_PTR(-ENAMETOOLONG
);
4118 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4121 return ERR_PTR(ret
);
4123 if (location
.objectid
== 0)
4126 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4127 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4131 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4133 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4134 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4135 &location
, &sub_root
);
4138 inode
= ERR_PTR(ret
);
4140 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4142 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4144 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4146 if (!IS_ERR(inode
) && root
!= sub_root
) {
4147 down_read(&root
->fs_info
->cleanup_work_sem
);
4148 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4149 ret
= btrfs_orphan_cleanup(sub_root
);
4150 up_read(&root
->fs_info
->cleanup_work_sem
);
4152 inode
= ERR_PTR(ret
);
4158 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4160 struct btrfs_root
*root
;
4162 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
4163 dentry
= dentry
->d_parent
;
4165 if (dentry
->d_inode
) {
4166 root
= BTRFS_I(dentry
->d_inode
)->root
;
4167 if (btrfs_root_refs(&root
->root_item
) == 0)
4173 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4174 struct nameidata
*nd
)
4176 struct inode
*inode
;
4178 inode
= btrfs_lookup_dentry(dir
, dentry
);
4180 return ERR_CAST(inode
);
4182 return d_splice_alias(inode
, dentry
);
4185 static unsigned char btrfs_filetype_table
[] = {
4186 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4189 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4192 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4193 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4194 struct btrfs_item
*item
;
4195 struct btrfs_dir_item
*di
;
4196 struct btrfs_key key
;
4197 struct btrfs_key found_key
;
4198 struct btrfs_path
*path
;
4201 struct extent_buffer
*leaf
;
4204 unsigned char d_type
;
4209 int key_type
= BTRFS_DIR_INDEX_KEY
;
4214 /* FIXME, use a real flag for deciding about the key type */
4215 if (root
->fs_info
->tree_root
== root
)
4216 key_type
= BTRFS_DIR_ITEM_KEY
;
4218 /* special case for "." */
4219 if (filp
->f_pos
== 0) {
4220 over
= filldir(dirent
, ".", 1,
4227 /* special case for .., just use the back ref */
4228 if (filp
->f_pos
== 1) {
4229 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4230 over
= filldir(dirent
, "..", 2,
4236 path
= btrfs_alloc_path();
4239 btrfs_set_key_type(&key
, key_type
);
4240 key
.offset
= filp
->f_pos
;
4241 key
.objectid
= inode
->i_ino
;
4243 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4249 leaf
= path
->nodes
[0];
4250 nritems
= btrfs_header_nritems(leaf
);
4251 slot
= path
->slots
[0];
4252 if (advance
|| slot
>= nritems
) {
4253 if (slot
>= nritems
- 1) {
4254 ret
= btrfs_next_leaf(root
, path
);
4257 leaf
= path
->nodes
[0];
4258 nritems
= btrfs_header_nritems(leaf
);
4259 slot
= path
->slots
[0];
4267 item
= btrfs_item_nr(leaf
, slot
);
4268 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4270 if (found_key
.objectid
!= key
.objectid
)
4272 if (btrfs_key_type(&found_key
) != key_type
)
4274 if (found_key
.offset
< filp
->f_pos
)
4277 filp
->f_pos
= found_key
.offset
;
4279 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4281 di_total
= btrfs_item_size(leaf
, item
);
4283 while (di_cur
< di_total
) {
4284 struct btrfs_key location
;
4286 if (verify_dir_item(root
, leaf
, di
))
4289 name_len
= btrfs_dir_name_len(leaf
, di
);
4290 if (name_len
<= sizeof(tmp_name
)) {
4291 name_ptr
= tmp_name
;
4293 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4299 read_extent_buffer(leaf
, name_ptr
,
4300 (unsigned long)(di
+ 1), name_len
);
4302 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4303 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4305 /* is this a reference to our own snapshot? If so
4308 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4309 location
.objectid
== root
->root_key
.objectid
) {
4313 over
= filldir(dirent
, name_ptr
, name_len
,
4314 found_key
.offset
, location
.objectid
,
4318 if (name_ptr
!= tmp_name
)
4323 di_len
= btrfs_dir_name_len(leaf
, di
) +
4324 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4326 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4330 /* Reached end of directory/root. Bump pos past the last item. */
4331 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4333 * 32-bit glibc will use getdents64, but then strtol -
4334 * so the last number we can serve is this.
4336 filp
->f_pos
= 0x7fffffff;
4342 btrfs_free_path(path
);
4346 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4348 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4349 struct btrfs_trans_handle
*trans
;
4351 bool nolock
= false;
4353 if (BTRFS_I(inode
)->dummy_inode
)
4357 nolock
= (root
->fs_info
->closing
&& root
== root
->fs_info
->tree_root
);
4359 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4361 trans
= btrfs_join_transaction_nolock(root
, 1);
4363 trans
= btrfs_join_transaction(root
, 1);
4365 return PTR_ERR(trans
);
4366 btrfs_set_trans_block_group(trans
, inode
);
4368 ret
= btrfs_end_transaction_nolock(trans
, root
);
4370 ret
= btrfs_commit_transaction(trans
, root
);
4376 * This is somewhat expensive, updating the tree every time the
4377 * inode changes. But, it is most likely to find the inode in cache.
4378 * FIXME, needs more benchmarking...there are no reasons other than performance
4379 * to keep or drop this code.
4381 void btrfs_dirty_inode(struct inode
*inode
)
4383 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4384 struct btrfs_trans_handle
*trans
;
4387 if (BTRFS_I(inode
)->dummy_inode
)
4390 trans
= btrfs_join_transaction(root
, 1);
4391 BUG_ON(IS_ERR(trans
));
4392 btrfs_set_trans_block_group(trans
, inode
);
4394 ret
= btrfs_update_inode(trans
, root
, inode
);
4395 if (ret
&& ret
== -ENOSPC
) {
4396 /* whoops, lets try again with the full transaction */
4397 btrfs_end_transaction(trans
, root
);
4398 trans
= btrfs_start_transaction(root
, 1);
4399 if (IS_ERR(trans
)) {
4400 if (printk_ratelimit()) {
4401 printk(KERN_ERR
"btrfs: fail to "
4402 "dirty inode %lu error %ld\n",
4403 inode
->i_ino
, PTR_ERR(trans
));
4407 btrfs_set_trans_block_group(trans
, inode
);
4409 ret
= btrfs_update_inode(trans
, root
, inode
);
4411 if (printk_ratelimit()) {
4412 printk(KERN_ERR
"btrfs: fail to "
4413 "dirty inode %lu error %d\n",
4418 btrfs_end_transaction(trans
, root
);
4422 * find the highest existing sequence number in a directory
4423 * and then set the in-memory index_cnt variable to reflect
4424 * free sequence numbers
4426 static int btrfs_set_inode_index_count(struct inode
*inode
)
4428 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4429 struct btrfs_key key
, found_key
;
4430 struct btrfs_path
*path
;
4431 struct extent_buffer
*leaf
;
4434 key
.objectid
= inode
->i_ino
;
4435 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4436 key
.offset
= (u64
)-1;
4438 path
= btrfs_alloc_path();
4442 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4445 /* FIXME: we should be able to handle this */
4451 * MAGIC NUMBER EXPLANATION:
4452 * since we search a directory based on f_pos we have to start at 2
4453 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4454 * else has to start at 2
4456 if (path
->slots
[0] == 0) {
4457 BTRFS_I(inode
)->index_cnt
= 2;
4463 leaf
= path
->nodes
[0];
4464 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4466 if (found_key
.objectid
!= inode
->i_ino
||
4467 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4468 BTRFS_I(inode
)->index_cnt
= 2;
4472 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4474 btrfs_free_path(path
);
4479 * helper to find a free sequence number in a given directory. This current
4480 * code is very simple, later versions will do smarter things in the btree
4482 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4486 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4487 ret
= btrfs_set_inode_index_count(dir
);
4492 *index
= BTRFS_I(dir
)->index_cnt
;
4493 BTRFS_I(dir
)->index_cnt
++;
4498 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4499 struct btrfs_root
*root
,
4501 const char *name
, int name_len
,
4502 u64 ref_objectid
, u64 objectid
,
4503 u64 alloc_hint
, int mode
, u64
*index
)
4505 struct inode
*inode
;
4506 struct btrfs_inode_item
*inode_item
;
4507 struct btrfs_key
*location
;
4508 struct btrfs_path
*path
;
4509 struct btrfs_inode_ref
*ref
;
4510 struct btrfs_key key
[2];
4516 path
= btrfs_alloc_path();
4519 inode
= new_inode(root
->fs_info
->sb
);
4521 return ERR_PTR(-ENOMEM
);
4524 trace_btrfs_inode_request(dir
);
4526 ret
= btrfs_set_inode_index(dir
, index
);
4529 return ERR_PTR(ret
);
4533 * index_cnt is ignored for everything but a dir,
4534 * btrfs_get_inode_index_count has an explanation for the magic
4537 BTRFS_I(inode
)->index_cnt
= 2;
4538 BTRFS_I(inode
)->root
= root
;
4539 BTRFS_I(inode
)->generation
= trans
->transid
;
4540 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4541 btrfs_set_inode_space_info(root
, inode
);
4547 BTRFS_I(inode
)->block_group
=
4548 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
4550 key
[0].objectid
= objectid
;
4551 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4554 key
[1].objectid
= objectid
;
4555 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4556 key
[1].offset
= ref_objectid
;
4558 sizes
[0] = sizeof(struct btrfs_inode_item
);
4559 sizes
[1] = name_len
+ sizeof(*ref
);
4561 path
->leave_spinning
= 1;
4562 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4566 inode_init_owner(inode
, dir
, mode
);
4567 inode
->i_ino
= objectid
;
4568 inode_set_bytes(inode
, 0);
4569 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4570 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4571 struct btrfs_inode_item
);
4572 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4574 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4575 struct btrfs_inode_ref
);
4576 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4577 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4578 ptr
= (unsigned long)(ref
+ 1);
4579 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4581 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4582 btrfs_free_path(path
);
4584 location
= &BTRFS_I(inode
)->location
;
4585 location
->objectid
= objectid
;
4586 location
->offset
= 0;
4587 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4589 btrfs_inherit_iflags(inode
, dir
);
4591 if ((mode
& S_IFREG
)) {
4592 if (btrfs_test_opt(root
, NODATASUM
))
4593 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4594 if (btrfs_test_opt(root
, NODATACOW
) ||
4595 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4596 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4599 insert_inode_hash(inode
);
4600 inode_tree_add(inode
);
4602 trace_btrfs_inode_new(inode
);
4607 BTRFS_I(dir
)->index_cnt
--;
4608 btrfs_free_path(path
);
4610 return ERR_PTR(ret
);
4613 static inline u8
btrfs_inode_type(struct inode
*inode
)
4615 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4619 * utility function to add 'inode' into 'parent_inode' with
4620 * a give name and a given sequence number.
4621 * if 'add_backref' is true, also insert a backref from the
4622 * inode to the parent directory.
4624 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4625 struct inode
*parent_inode
, struct inode
*inode
,
4626 const char *name
, int name_len
, int add_backref
, u64 index
)
4629 struct btrfs_key key
;
4630 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4632 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4633 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4635 key
.objectid
= inode
->i_ino
;
4636 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4640 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4641 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4642 key
.objectid
, root
->root_key
.objectid
,
4643 parent_inode
->i_ino
,
4644 index
, name
, name_len
);
4645 } else if (add_backref
) {
4646 ret
= btrfs_insert_inode_ref(trans
, root
,
4647 name
, name_len
, inode
->i_ino
,
4648 parent_inode
->i_ino
, index
);
4652 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4653 parent_inode
->i_ino
, &key
,
4654 btrfs_inode_type(inode
), index
);
4657 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4659 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4660 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4665 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4666 struct inode
*dir
, struct dentry
*dentry
,
4667 struct inode
*inode
, int backref
, u64 index
)
4669 int err
= btrfs_add_link(trans
, dir
, inode
,
4670 dentry
->d_name
.name
, dentry
->d_name
.len
,
4673 d_instantiate(dentry
, inode
);
4681 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4682 int mode
, dev_t rdev
)
4684 struct btrfs_trans_handle
*trans
;
4685 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4686 struct inode
*inode
= NULL
;
4690 unsigned long nr
= 0;
4693 if (!new_valid_dev(rdev
))
4696 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4701 * 2 for inode item and ref
4703 * 1 for xattr if selinux is on
4705 trans
= btrfs_start_transaction(root
, 5);
4707 return PTR_ERR(trans
);
4709 btrfs_set_trans_block_group(trans
, dir
);
4711 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4712 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4713 BTRFS_I(dir
)->block_group
, mode
, &index
);
4714 err
= PTR_ERR(inode
);
4718 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4724 btrfs_set_trans_block_group(trans
, inode
);
4725 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4729 inode
->i_op
= &btrfs_special_inode_operations
;
4730 init_special_inode(inode
, inode
->i_mode
, rdev
);
4731 btrfs_update_inode(trans
, root
, inode
);
4733 btrfs_update_inode_block_group(trans
, inode
);
4734 btrfs_update_inode_block_group(trans
, dir
);
4736 nr
= trans
->blocks_used
;
4737 btrfs_end_transaction_throttle(trans
, root
);
4738 btrfs_btree_balance_dirty(root
, nr
);
4740 inode_dec_link_count(inode
);
4746 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4747 int mode
, struct nameidata
*nd
)
4749 struct btrfs_trans_handle
*trans
;
4750 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4751 struct inode
*inode
= NULL
;
4754 unsigned long nr
= 0;
4758 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4762 * 2 for inode item and ref
4764 * 1 for xattr if selinux is on
4766 trans
= btrfs_start_transaction(root
, 5);
4768 return PTR_ERR(trans
);
4770 btrfs_set_trans_block_group(trans
, dir
);
4772 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4773 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4774 BTRFS_I(dir
)->block_group
, mode
, &index
);
4775 err
= PTR_ERR(inode
);
4779 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4785 btrfs_set_trans_block_group(trans
, inode
);
4786 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4790 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4791 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4792 inode
->i_fop
= &btrfs_file_operations
;
4793 inode
->i_op
= &btrfs_file_inode_operations
;
4794 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4796 btrfs_update_inode_block_group(trans
, inode
);
4797 btrfs_update_inode_block_group(trans
, dir
);
4799 nr
= trans
->blocks_used
;
4800 btrfs_end_transaction_throttle(trans
, root
);
4802 inode_dec_link_count(inode
);
4805 btrfs_btree_balance_dirty(root
, nr
);
4809 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4810 struct dentry
*dentry
)
4812 struct btrfs_trans_handle
*trans
;
4813 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4814 struct inode
*inode
= old_dentry
->d_inode
;
4816 unsigned long nr
= 0;
4820 if (inode
->i_nlink
== 0)
4823 /* do not allow sys_link's with other subvols of the same device */
4824 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4827 btrfs_inc_nlink(inode
);
4828 inode
->i_ctime
= CURRENT_TIME
;
4830 err
= btrfs_set_inode_index(dir
, &index
);
4835 * 2 items for inode and inode ref
4836 * 2 items for dir items
4837 * 1 item for parent inode
4839 trans
= btrfs_start_transaction(root
, 5);
4840 if (IS_ERR(trans
)) {
4841 err
= PTR_ERR(trans
);
4845 btrfs_set_trans_block_group(trans
, dir
);
4848 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4853 struct dentry
*parent
= dget_parent(dentry
);
4854 btrfs_update_inode_block_group(trans
, dir
);
4855 err
= btrfs_update_inode(trans
, root
, inode
);
4857 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4861 nr
= trans
->blocks_used
;
4862 btrfs_end_transaction_throttle(trans
, root
);
4865 inode_dec_link_count(inode
);
4868 btrfs_btree_balance_dirty(root
, nr
);
4872 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4874 struct inode
*inode
= NULL
;
4875 struct btrfs_trans_handle
*trans
;
4876 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4878 int drop_on_err
= 0;
4881 unsigned long nr
= 1;
4883 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4888 * 2 items for inode and ref
4889 * 2 items for dir items
4890 * 1 for xattr if selinux is on
4892 trans
= btrfs_start_transaction(root
, 5);
4894 return PTR_ERR(trans
);
4895 btrfs_set_trans_block_group(trans
, dir
);
4897 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4898 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4899 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
4901 if (IS_ERR(inode
)) {
4902 err
= PTR_ERR(inode
);
4908 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4912 inode
->i_op
= &btrfs_dir_inode_operations
;
4913 inode
->i_fop
= &btrfs_dir_file_operations
;
4914 btrfs_set_trans_block_group(trans
, inode
);
4916 btrfs_i_size_write(inode
, 0);
4917 err
= btrfs_update_inode(trans
, root
, inode
);
4921 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4922 dentry
->d_name
.len
, 0, index
);
4926 d_instantiate(dentry
, inode
);
4928 btrfs_update_inode_block_group(trans
, inode
);
4929 btrfs_update_inode_block_group(trans
, dir
);
4932 nr
= trans
->blocks_used
;
4933 btrfs_end_transaction_throttle(trans
, root
);
4936 btrfs_btree_balance_dirty(root
, nr
);
4940 /* helper for btfs_get_extent. Given an existing extent in the tree,
4941 * and an extent that you want to insert, deal with overlap and insert
4942 * the new extent into the tree.
4944 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4945 struct extent_map
*existing
,
4946 struct extent_map
*em
,
4947 u64 map_start
, u64 map_len
)
4951 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4952 start_diff
= map_start
- em
->start
;
4953 em
->start
= map_start
;
4955 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4956 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4957 em
->block_start
+= start_diff
;
4958 em
->block_len
-= start_diff
;
4960 return add_extent_mapping(em_tree
, em
);
4963 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4964 struct inode
*inode
, struct page
*page
,
4965 size_t pg_offset
, u64 extent_offset
,
4966 struct btrfs_file_extent_item
*item
)
4969 struct extent_buffer
*leaf
= path
->nodes
[0];
4972 unsigned long inline_size
;
4976 WARN_ON(pg_offset
!= 0);
4977 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4978 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4979 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4980 btrfs_item_nr(leaf
, path
->slots
[0]));
4981 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4982 ptr
= btrfs_file_extent_inline_start(item
);
4984 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4986 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4987 ret
= btrfs_decompress(compress_type
, tmp
, page
,
4988 extent_offset
, inline_size
, max_size
);
4990 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4991 unsigned long copy_size
= min_t(u64
,
4992 PAGE_CACHE_SIZE
- pg_offset
,
4993 max_size
- extent_offset
);
4994 memset(kaddr
+ pg_offset
, 0, copy_size
);
4995 kunmap_atomic(kaddr
, KM_USER0
);
5002 * a bit scary, this does extent mapping from logical file offset to the disk.
5003 * the ugly parts come from merging extents from the disk with the in-ram
5004 * representation. This gets more complex because of the data=ordered code,
5005 * where the in-ram extents might be locked pending data=ordered completion.
5007 * This also copies inline extents directly into the page.
5010 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
5011 size_t pg_offset
, u64 start
, u64 len
,
5017 u64 extent_start
= 0;
5019 u64 objectid
= inode
->i_ino
;
5021 struct btrfs_path
*path
= NULL
;
5022 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5023 struct btrfs_file_extent_item
*item
;
5024 struct extent_buffer
*leaf
;
5025 struct btrfs_key found_key
;
5026 struct extent_map
*em
= NULL
;
5027 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5028 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5029 struct btrfs_trans_handle
*trans
= NULL
;
5033 read_lock(&em_tree
->lock
);
5034 em
= lookup_extent_mapping(em_tree
, start
, len
);
5036 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5037 read_unlock(&em_tree
->lock
);
5040 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5041 free_extent_map(em
);
5042 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5043 free_extent_map(em
);
5047 em
= alloc_extent_map(GFP_NOFS
);
5052 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5053 em
->start
= EXTENT_MAP_HOLE
;
5054 em
->orig_start
= EXTENT_MAP_HOLE
;
5056 em
->block_len
= (u64
)-1;
5059 path
= btrfs_alloc_path();
5063 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5064 objectid
, start
, trans
!= NULL
);
5071 if (path
->slots
[0] == 0)
5076 leaf
= path
->nodes
[0];
5077 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5078 struct btrfs_file_extent_item
);
5079 /* are we inside the extent that was found? */
5080 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5081 found_type
= btrfs_key_type(&found_key
);
5082 if (found_key
.objectid
!= objectid
||
5083 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5087 found_type
= btrfs_file_extent_type(leaf
, item
);
5088 extent_start
= found_key
.offset
;
5089 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5090 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5091 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5092 extent_end
= extent_start
+
5093 btrfs_file_extent_num_bytes(leaf
, item
);
5094 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5096 size
= btrfs_file_extent_inline_len(leaf
, item
);
5097 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5098 ~((u64
)root
->sectorsize
- 1);
5101 if (start
>= extent_end
) {
5103 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5104 ret
= btrfs_next_leaf(root
, path
);
5111 leaf
= path
->nodes
[0];
5113 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5114 if (found_key
.objectid
!= objectid
||
5115 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5117 if (start
+ len
<= found_key
.offset
)
5120 em
->len
= found_key
.offset
- start
;
5124 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5125 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5126 em
->start
= extent_start
;
5127 em
->len
= extent_end
- extent_start
;
5128 em
->orig_start
= extent_start
-
5129 btrfs_file_extent_offset(leaf
, item
);
5130 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5132 em
->block_start
= EXTENT_MAP_HOLE
;
5135 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5136 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5137 em
->compress_type
= compress_type
;
5138 em
->block_start
= bytenr
;
5139 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5142 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5143 em
->block_start
= bytenr
;
5144 em
->block_len
= em
->len
;
5145 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5146 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5149 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5153 size_t extent_offset
;
5156 em
->block_start
= EXTENT_MAP_INLINE
;
5157 if (!page
|| create
) {
5158 em
->start
= extent_start
;
5159 em
->len
= extent_end
- extent_start
;
5163 size
= btrfs_file_extent_inline_len(leaf
, item
);
5164 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5165 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5166 size
- extent_offset
);
5167 em
->start
= extent_start
+ extent_offset
;
5168 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5169 ~((u64
)root
->sectorsize
- 1);
5170 em
->orig_start
= EXTENT_MAP_INLINE
;
5171 if (compress_type
) {
5172 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5173 em
->compress_type
= compress_type
;
5175 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5176 if (create
== 0 && !PageUptodate(page
)) {
5177 if (btrfs_file_extent_compression(leaf
, item
) !=
5178 BTRFS_COMPRESS_NONE
) {
5179 ret
= uncompress_inline(path
, inode
, page
,
5181 extent_offset
, item
);
5185 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5187 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5188 memset(map
+ pg_offset
+ copy_size
, 0,
5189 PAGE_CACHE_SIZE
- pg_offset
-
5194 flush_dcache_page(page
);
5195 } else if (create
&& PageUptodate(page
)) {
5199 free_extent_map(em
);
5201 btrfs_release_path(root
, path
);
5202 trans
= btrfs_join_transaction(root
, 1);
5204 return ERR_CAST(trans
);
5208 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5211 btrfs_mark_buffer_dirty(leaf
);
5213 set_extent_uptodate(io_tree
, em
->start
,
5214 extent_map_end(em
) - 1, GFP_NOFS
);
5217 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5224 em
->block_start
= EXTENT_MAP_HOLE
;
5225 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5227 btrfs_release_path(root
, path
);
5228 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5229 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5230 "[%llu %llu]\n", (unsigned long long)em
->start
,
5231 (unsigned long long)em
->len
,
5232 (unsigned long long)start
,
5233 (unsigned long long)len
);
5239 write_lock(&em_tree
->lock
);
5240 ret
= add_extent_mapping(em_tree
, em
);
5241 /* it is possible that someone inserted the extent into the tree
5242 * while we had the lock dropped. It is also possible that
5243 * an overlapping map exists in the tree
5245 if (ret
== -EEXIST
) {
5246 struct extent_map
*existing
;
5250 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5251 if (existing
&& (existing
->start
> start
||
5252 existing
->start
+ existing
->len
<= start
)) {
5253 free_extent_map(existing
);
5257 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5260 err
= merge_extent_mapping(em_tree
, existing
,
5263 free_extent_map(existing
);
5265 free_extent_map(em
);
5270 free_extent_map(em
);
5274 free_extent_map(em
);
5279 write_unlock(&em_tree
->lock
);
5282 trace_btrfs_get_extent(root
, em
);
5285 btrfs_free_path(path
);
5287 ret
= btrfs_end_transaction(trans
, root
);
5292 free_extent_map(em
);
5293 return ERR_PTR(err
);
5298 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5299 size_t pg_offset
, u64 start
, u64 len
,
5302 struct extent_map
*em
;
5303 struct extent_map
*hole_em
= NULL
;
5304 u64 range_start
= start
;
5310 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5315 * if our em maps to a hole, there might
5316 * actually be delalloc bytes behind it
5318 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5324 /* check to see if we've wrapped (len == -1 or similar) */
5333 /* ok, we didn't find anything, lets look for delalloc */
5334 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5335 end
, len
, EXTENT_DELALLOC
, 1);
5336 found_end
= range_start
+ found
;
5337 if (found_end
< range_start
)
5338 found_end
= (u64
)-1;
5341 * we didn't find anything useful, return
5342 * the original results from get_extent()
5344 if (range_start
> end
|| found_end
<= start
) {
5350 /* adjust the range_start to make sure it doesn't
5351 * go backwards from the start they passed in
5353 range_start
= max(start
,range_start
);
5354 found
= found_end
- range_start
;
5357 u64 hole_start
= start
;
5360 em
= alloc_extent_map(GFP_NOFS
);
5366 * when btrfs_get_extent can't find anything it
5367 * returns one huge hole
5369 * make sure what it found really fits our range, and
5370 * adjust to make sure it is based on the start from
5374 u64 calc_end
= extent_map_end(hole_em
);
5376 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5377 free_extent_map(hole_em
);
5380 hole_start
= max(hole_em
->start
, start
);
5381 hole_len
= calc_end
- hole_start
;
5385 if (hole_em
&& range_start
> hole_start
) {
5386 /* our hole starts before our delalloc, so we
5387 * have to return just the parts of the hole
5388 * that go until the delalloc starts
5390 em
->len
= min(hole_len
,
5391 range_start
- hole_start
);
5392 em
->start
= hole_start
;
5393 em
->orig_start
= hole_start
;
5395 * don't adjust block start at all,
5396 * it is fixed at EXTENT_MAP_HOLE
5398 em
->block_start
= hole_em
->block_start
;
5399 em
->block_len
= hole_len
;
5401 em
->start
= range_start
;
5403 em
->orig_start
= range_start
;
5404 em
->block_start
= EXTENT_MAP_DELALLOC
;
5405 em
->block_len
= found
;
5407 } else if (hole_em
) {
5412 free_extent_map(hole_em
);
5414 free_extent_map(em
);
5415 return ERR_PTR(err
);
5420 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5423 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5424 struct btrfs_trans_handle
*trans
;
5425 struct extent_map
*em
;
5426 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5427 struct btrfs_key ins
;
5431 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5433 trans
= btrfs_join_transaction(root
, 0);
5435 return ERR_CAST(trans
);
5437 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5439 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5440 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5441 alloc_hint
, (u64
)-1, &ins
, 1);
5447 em
= alloc_extent_map(GFP_NOFS
);
5449 em
= ERR_PTR(-ENOMEM
);
5454 em
->orig_start
= em
->start
;
5455 em
->len
= ins
.offset
;
5457 em
->block_start
= ins
.objectid
;
5458 em
->block_len
= ins
.offset
;
5459 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5460 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5463 write_lock(&em_tree
->lock
);
5464 ret
= add_extent_mapping(em_tree
, em
);
5465 write_unlock(&em_tree
->lock
);
5468 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5471 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5472 ins
.offset
, ins
.offset
, 0);
5474 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5478 btrfs_end_transaction(trans
, root
);
5483 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5484 * block must be cow'd
5486 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5487 struct inode
*inode
, u64 offset
, u64 len
)
5489 struct btrfs_path
*path
;
5491 struct extent_buffer
*leaf
;
5492 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5493 struct btrfs_file_extent_item
*fi
;
5494 struct btrfs_key key
;
5502 path
= btrfs_alloc_path();
5506 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
5511 slot
= path
->slots
[0];
5514 /* can't find the item, must cow */
5521 leaf
= path
->nodes
[0];
5522 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5523 if (key
.objectid
!= inode
->i_ino
||
5524 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5525 /* not our file or wrong item type, must cow */
5529 if (key
.offset
> offset
) {
5530 /* Wrong offset, must cow */
5534 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5535 found_type
= btrfs_file_extent_type(leaf
, fi
);
5536 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5537 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5538 /* not a regular extent, must cow */
5541 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5542 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5544 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5545 if (extent_end
< offset
+ len
) {
5546 /* extent doesn't include our full range, must cow */
5550 if (btrfs_extent_readonly(root
, disk_bytenr
))
5554 * look for other files referencing this extent, if we
5555 * find any we must cow
5557 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
5558 key
.offset
- backref_offset
, disk_bytenr
))
5562 * adjust disk_bytenr and num_bytes to cover just the bytes
5563 * in this extent we are about to write. If there
5564 * are any csums in that range we have to cow in order
5565 * to keep the csums correct
5567 disk_bytenr
+= backref_offset
;
5568 disk_bytenr
+= offset
- key
.offset
;
5569 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5570 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5573 * all of the above have passed, it is safe to overwrite this extent
5578 btrfs_free_path(path
);
5582 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5583 struct buffer_head
*bh_result
, int create
)
5585 struct extent_map
*em
;
5586 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5587 u64 start
= iblock
<< inode
->i_blkbits
;
5588 u64 len
= bh_result
->b_size
;
5589 struct btrfs_trans_handle
*trans
;
5591 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5596 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5597 * io. INLINE is special, and we could probably kludge it in here, but
5598 * it's still buffered so for safety lets just fall back to the generic
5601 * For COMPRESSED we _have_ to read the entire extent in so we can
5602 * decompress it, so there will be buffering required no matter what we
5603 * do, so go ahead and fallback to buffered.
5605 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5606 * to buffered IO. Don't blame me, this is the price we pay for using
5609 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5610 em
->block_start
== EXTENT_MAP_INLINE
) {
5611 free_extent_map(em
);
5615 /* Just a good old fashioned hole, return */
5616 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5617 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5618 free_extent_map(em
);
5619 /* DIO will do one hole at a time, so just unlock a sector */
5620 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5621 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5626 * We don't allocate a new extent in the following cases
5628 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5630 * 2) The extent is marked as PREALLOC. We're good to go here and can
5631 * just use the extent.
5635 len
= em
->len
- (start
- em
->start
);
5639 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5640 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5641 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5646 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5647 type
= BTRFS_ORDERED_PREALLOC
;
5649 type
= BTRFS_ORDERED_NOCOW
;
5650 len
= min(len
, em
->len
- (start
- em
->start
));
5651 block_start
= em
->block_start
+ (start
- em
->start
);
5654 * we're not going to log anything, but we do need
5655 * to make sure the current transaction stays open
5656 * while we look for nocow cross refs
5658 trans
= btrfs_join_transaction(root
, 0);
5662 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5663 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5664 block_start
, len
, len
, type
);
5665 btrfs_end_transaction(trans
, root
);
5667 free_extent_map(em
);
5672 btrfs_end_transaction(trans
, root
);
5676 * this will cow the extent, reset the len in case we changed
5679 len
= bh_result
->b_size
;
5680 free_extent_map(em
);
5681 em
= btrfs_new_extent_direct(inode
, start
, len
);
5684 len
= min(len
, em
->len
- (start
- em
->start
));
5686 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5687 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5690 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5692 bh_result
->b_size
= len
;
5693 bh_result
->b_bdev
= em
->bdev
;
5694 set_buffer_mapped(bh_result
);
5695 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5696 set_buffer_new(bh_result
);
5698 free_extent_map(em
);
5703 struct btrfs_dio_private
{
5704 struct inode
*inode
;
5711 /* number of bios pending for this dio */
5712 atomic_t pending_bios
;
5717 struct bio
*orig_bio
;
5720 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5722 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5723 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5724 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5725 struct inode
*inode
= dip
->inode
;
5726 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5728 u32
*private = dip
->csums
;
5730 start
= dip
->logical_offset
;
5732 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5733 struct page
*page
= bvec
->bv_page
;
5736 unsigned long flags
;
5738 local_irq_save(flags
);
5739 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5740 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5741 csum
, bvec
->bv_len
);
5742 btrfs_csum_final(csum
, (char *)&csum
);
5743 kunmap_atomic(kaddr
, KM_IRQ0
);
5744 local_irq_restore(flags
);
5746 flush_dcache_page(bvec
->bv_page
);
5747 if (csum
!= *private) {
5748 printk(KERN_ERR
"btrfs csum failed ino %lu off"
5749 " %llu csum %u private %u\n",
5750 inode
->i_ino
, (unsigned long long)start
,
5756 start
+= bvec
->bv_len
;
5759 } while (bvec
<= bvec_end
);
5761 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5762 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5763 bio
->bi_private
= dip
->private;
5768 /* If we had a csum failure make sure to clear the uptodate flag */
5770 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5771 dio_end_io(bio
, err
);
5774 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5776 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5777 struct inode
*inode
= dip
->inode
;
5778 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5779 struct btrfs_trans_handle
*trans
;
5780 struct btrfs_ordered_extent
*ordered
= NULL
;
5781 struct extent_state
*cached_state
= NULL
;
5782 u64 ordered_offset
= dip
->logical_offset
;
5783 u64 ordered_bytes
= dip
->bytes
;
5789 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5797 trans
= btrfs_join_transaction(root
, 1);
5798 if (IS_ERR(trans
)) {
5802 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5804 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5805 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5807 ret
= btrfs_update_inode(trans
, root
, inode
);
5812 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5813 ordered
->file_offset
+ ordered
->len
- 1, 0,
5814 &cached_state
, GFP_NOFS
);
5816 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5817 ret
= btrfs_mark_extent_written(trans
, inode
,
5818 ordered
->file_offset
,
5819 ordered
->file_offset
+
5826 ret
= insert_reserved_file_extent(trans
, inode
,
5827 ordered
->file_offset
,
5833 BTRFS_FILE_EXTENT_REG
);
5834 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5835 ordered
->file_offset
, ordered
->len
);
5843 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5844 btrfs_ordered_update_i_size(inode
, 0, ordered
);
5845 btrfs_update_inode(trans
, root
, inode
);
5847 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5848 ordered
->file_offset
+ ordered
->len
- 1,
5849 &cached_state
, GFP_NOFS
);
5851 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5852 btrfs_end_transaction(trans
, root
);
5853 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5854 btrfs_put_ordered_extent(ordered
);
5855 btrfs_put_ordered_extent(ordered
);
5859 * our bio might span multiple ordered extents. If we haven't
5860 * completed the accounting for the whole dio, go back and try again
5862 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5863 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5868 bio
->bi_private
= dip
->private;
5873 /* If we had an error make sure to clear the uptodate flag */
5875 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5876 dio_end_io(bio
, err
);
5879 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5880 struct bio
*bio
, int mirror_num
,
5881 unsigned long bio_flags
, u64 offset
)
5884 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5885 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5890 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5892 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5895 printk(KERN_ERR
"btrfs direct IO failed ino %lu rw %lu "
5896 "sector %#Lx len %u err no %d\n",
5897 dip
->inode
->i_ino
, bio
->bi_rw
,
5898 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5902 * before atomic variable goto zero, we must make sure
5903 * dip->errors is perceived to be set.
5905 smp_mb__before_atomic_dec();
5908 /* if there are more bios still pending for this dio, just exit */
5909 if (!atomic_dec_and_test(&dip
->pending_bios
))
5913 bio_io_error(dip
->orig_bio
);
5915 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5916 bio_endio(dip
->orig_bio
, 0);
5922 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5923 u64 first_sector
, gfp_t gfp_flags
)
5925 int nr_vecs
= bio_get_nr_vecs(bdev
);
5926 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5929 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5930 int rw
, u64 file_offset
, int skip_sum
,
5933 int write
= rw
& REQ_WRITE
;
5934 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5938 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5942 if (write
&& !skip_sum
) {
5943 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5944 inode
, rw
, bio
, 0, 0,
5946 __btrfs_submit_bio_start_direct_io
,
5947 __btrfs_submit_bio_done
);
5949 } else if (!skip_sum
) {
5950 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5951 file_offset
, csums
);
5956 ret
= btrfs_map_bio(root
, rw
, bio
, 0, 1);
5962 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5965 struct inode
*inode
= dip
->inode
;
5966 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5967 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5969 struct bio
*orig_bio
= dip
->orig_bio
;
5970 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5971 u64 start_sector
= orig_bio
->bi_sector
;
5972 u64 file_offset
= dip
->logical_offset
;
5976 u32
*csums
= dip
->csums
;
5978 int write
= rw
& REQ_WRITE
;
5980 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
5983 bio
->bi_private
= dip
;
5984 bio
->bi_end_io
= btrfs_end_dio_bio
;
5985 atomic_inc(&dip
->pending_bios
);
5987 map_length
= orig_bio
->bi_size
;
5988 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5989 &map_length
, NULL
, 0);
5995 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
5996 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
5997 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5998 bvec
->bv_offset
) < bvec
->bv_len
)) {
6000 * inc the count before we submit the bio so
6001 * we know the end IO handler won't happen before
6002 * we inc the count. Otherwise, the dip might get freed
6003 * before we're done setting it up
6005 atomic_inc(&dip
->pending_bios
);
6006 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
6007 file_offset
, skip_sum
,
6011 atomic_dec(&dip
->pending_bios
);
6015 /* Write's use the ordered csums */
6016 if (!write
&& !skip_sum
)
6017 csums
= csums
+ nr_pages
;
6018 start_sector
+= submit_len
>> 9;
6019 file_offset
+= submit_len
;
6024 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6025 start_sector
, GFP_NOFS
);
6028 bio
->bi_private
= dip
;
6029 bio
->bi_end_io
= btrfs_end_dio_bio
;
6031 map_length
= orig_bio
->bi_size
;
6032 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6033 &map_length
, NULL
, 0);
6039 submit_len
+= bvec
->bv_len
;
6045 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6054 * before atomic variable goto zero, we must
6055 * make sure dip->errors is perceived to be set.
6057 smp_mb__before_atomic_dec();
6058 if (atomic_dec_and_test(&dip
->pending_bios
))
6059 bio_io_error(dip
->orig_bio
);
6061 /* bio_end_io() will handle error, so we needn't return it */
6065 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6068 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6069 struct btrfs_dio_private
*dip
;
6070 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6072 int write
= rw
& REQ_WRITE
;
6075 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6077 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6084 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6085 if (!write
&& !skip_sum
) {
6086 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6094 dip
->private = bio
->bi_private
;
6096 dip
->logical_offset
= file_offset
;
6100 dip
->bytes
+= bvec
->bv_len
;
6102 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6104 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6105 bio
->bi_private
= dip
;
6107 dip
->orig_bio
= bio
;
6108 atomic_set(&dip
->pending_bios
, 0);
6111 bio
->bi_end_io
= btrfs_endio_direct_write
;
6113 bio
->bi_end_io
= btrfs_endio_direct_read
;
6115 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6120 * If this is a write, we need to clean up the reserved space and kill
6121 * the ordered extent.
6124 struct btrfs_ordered_extent
*ordered
;
6125 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6126 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6127 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6128 btrfs_free_reserved_extent(root
, ordered
->start
,
6130 btrfs_put_ordered_extent(ordered
);
6131 btrfs_put_ordered_extent(ordered
);
6133 bio_endio(bio
, ret
);
6136 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6137 const struct iovec
*iov
, loff_t offset
,
6138 unsigned long nr_segs
)
6143 unsigned blocksize_mask
= root
->sectorsize
- 1;
6144 ssize_t retval
= -EINVAL
;
6145 loff_t end
= offset
;
6147 if (offset
& blocksize_mask
)
6150 /* Check the memory alignment. Blocks cannot straddle pages */
6151 for (seg
= 0; seg
< nr_segs
; seg
++) {
6152 addr
= (unsigned long)iov
[seg
].iov_base
;
6153 size
= iov
[seg
].iov_len
;
6155 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6162 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6163 const struct iovec
*iov
, loff_t offset
,
6164 unsigned long nr_segs
)
6166 struct file
*file
= iocb
->ki_filp
;
6167 struct inode
*inode
= file
->f_mapping
->host
;
6168 struct btrfs_ordered_extent
*ordered
;
6169 struct extent_state
*cached_state
= NULL
;
6170 u64 lockstart
, lockend
;
6172 int writing
= rw
& WRITE
;
6174 size_t count
= iov_length(iov
, nr_segs
);
6176 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6182 lockend
= offset
+ count
- 1;
6185 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6191 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6192 0, &cached_state
, GFP_NOFS
);
6194 * We're concerned with the entire range that we're going to be
6195 * doing DIO to, so we need to make sure theres no ordered
6196 * extents in this range.
6198 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6199 lockend
- lockstart
+ 1);
6202 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6203 &cached_state
, GFP_NOFS
);
6204 btrfs_start_ordered_extent(inode
, ordered
, 1);
6205 btrfs_put_ordered_extent(ordered
);
6210 * we don't use btrfs_set_extent_delalloc because we don't want
6211 * the dirty or uptodate bits
6214 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6215 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6216 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6219 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6220 lockend
, EXTENT_LOCKED
| write_bits
,
6221 1, 0, &cached_state
, GFP_NOFS
);
6226 free_extent_state(cached_state
);
6227 cached_state
= NULL
;
6229 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6230 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6231 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6232 btrfs_submit_direct
, 0);
6234 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6235 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6236 offset
+ iov_length(iov
, nr_segs
) - 1,
6237 EXTENT_LOCKED
| write_bits
, 1, 0,
6238 &cached_state
, GFP_NOFS
);
6239 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6241 * We're falling back to buffered, unlock the section we didn't
6244 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6245 offset
+ iov_length(iov
, nr_segs
) - 1,
6246 EXTENT_LOCKED
| write_bits
, 1, 0,
6247 &cached_state
, GFP_NOFS
);
6250 free_extent_state(cached_state
);
6254 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6255 __u64 start
, __u64 len
)
6257 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6260 int btrfs_readpage(struct file
*file
, struct page
*page
)
6262 struct extent_io_tree
*tree
;
6263 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6264 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6267 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6269 struct extent_io_tree
*tree
;
6272 if (current
->flags
& PF_MEMALLOC
) {
6273 redirty_page_for_writepage(wbc
, page
);
6277 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6278 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6281 int btrfs_writepages(struct address_space
*mapping
,
6282 struct writeback_control
*wbc
)
6284 struct extent_io_tree
*tree
;
6286 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6287 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6291 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6292 struct list_head
*pages
, unsigned nr_pages
)
6294 struct extent_io_tree
*tree
;
6295 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6296 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6299 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6301 struct extent_io_tree
*tree
;
6302 struct extent_map_tree
*map
;
6305 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6306 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6307 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6309 ClearPagePrivate(page
);
6310 set_page_private(page
, 0);
6311 page_cache_release(page
);
6316 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6318 if (PageWriteback(page
) || PageDirty(page
))
6320 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6323 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6325 struct extent_io_tree
*tree
;
6326 struct btrfs_ordered_extent
*ordered
;
6327 struct extent_state
*cached_state
= NULL
;
6328 u64 page_start
= page_offset(page
);
6329 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6333 * we have the page locked, so new writeback can't start,
6334 * and the dirty bit won't be cleared while we are here.
6336 * Wait for IO on this page so that we can safely clear
6337 * the PagePrivate2 bit and do ordered accounting
6339 wait_on_page_writeback(page
);
6341 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6343 btrfs_releasepage(page
, GFP_NOFS
);
6346 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6348 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6352 * IO on this page will never be started, so we need
6353 * to account for any ordered extents now
6355 clear_extent_bit(tree
, page_start
, page_end
,
6356 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6357 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6358 &cached_state
, GFP_NOFS
);
6360 * whoever cleared the private bit is responsible
6361 * for the finish_ordered_io
6363 if (TestClearPagePrivate2(page
)) {
6364 btrfs_finish_ordered_io(page
->mapping
->host
,
6365 page_start
, page_end
);
6367 btrfs_put_ordered_extent(ordered
);
6368 cached_state
= NULL
;
6369 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6372 clear_extent_bit(tree
, page_start
, page_end
,
6373 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6374 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6375 __btrfs_releasepage(page
, GFP_NOFS
);
6377 ClearPageChecked(page
);
6378 if (PagePrivate(page
)) {
6379 ClearPagePrivate(page
);
6380 set_page_private(page
, 0);
6381 page_cache_release(page
);
6386 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6387 * called from a page fault handler when a page is first dirtied. Hence we must
6388 * be careful to check for EOF conditions here. We set the page up correctly
6389 * for a written page which means we get ENOSPC checking when writing into
6390 * holes and correct delalloc and unwritten extent mapping on filesystems that
6391 * support these features.
6393 * We are not allowed to take the i_mutex here so we have to play games to
6394 * protect against truncate races as the page could now be beyond EOF. Because
6395 * vmtruncate() writes the inode size before removing pages, once we have the
6396 * page lock we can determine safely if the page is beyond EOF. If it is not
6397 * beyond EOF, then the page is guaranteed safe against truncation until we
6400 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6402 struct page
*page
= vmf
->page
;
6403 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6404 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6405 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6406 struct btrfs_ordered_extent
*ordered
;
6407 struct extent_state
*cached_state
= NULL
;
6409 unsigned long zero_start
;
6415 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6419 else /* -ENOSPC, -EIO, etc */
6420 ret
= VM_FAULT_SIGBUS
;
6424 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6427 size
= i_size_read(inode
);
6428 page_start
= page_offset(page
);
6429 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6431 if ((page
->mapping
!= inode
->i_mapping
) ||
6432 (page_start
>= size
)) {
6433 /* page got truncated out from underneath us */
6436 wait_on_page_writeback(page
);
6438 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6440 set_page_extent_mapped(page
);
6443 * we can't set the delalloc bits if there are pending ordered
6444 * extents. Drop our locks and wait for them to finish
6446 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6448 unlock_extent_cached(io_tree
, page_start
, page_end
,
6449 &cached_state
, GFP_NOFS
);
6451 btrfs_start_ordered_extent(inode
, ordered
, 1);
6452 btrfs_put_ordered_extent(ordered
);
6457 * XXX - page_mkwrite gets called every time the page is dirtied, even
6458 * if it was already dirty, so for space accounting reasons we need to
6459 * clear any delalloc bits for the range we are fixing to save. There
6460 * is probably a better way to do this, but for now keep consistent with
6461 * prepare_pages in the normal write path.
6463 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6464 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6465 0, 0, &cached_state
, GFP_NOFS
);
6467 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6470 unlock_extent_cached(io_tree
, page_start
, page_end
,
6471 &cached_state
, GFP_NOFS
);
6472 ret
= VM_FAULT_SIGBUS
;
6477 /* page is wholly or partially inside EOF */
6478 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6479 zero_start
= size
& ~PAGE_CACHE_MASK
;
6481 zero_start
= PAGE_CACHE_SIZE
;
6483 if (zero_start
!= PAGE_CACHE_SIZE
) {
6485 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6486 flush_dcache_page(page
);
6489 ClearPageChecked(page
);
6490 set_page_dirty(page
);
6491 SetPageUptodate(page
);
6493 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6494 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6496 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6500 return VM_FAULT_LOCKED
;
6502 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6507 static int btrfs_truncate(struct inode
*inode
)
6509 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6512 struct btrfs_trans_handle
*trans
;
6514 u64 mask
= root
->sectorsize
- 1;
6516 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6520 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6521 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6523 trans
= btrfs_start_transaction(root
, 5);
6525 return PTR_ERR(trans
);
6527 btrfs_set_trans_block_group(trans
, inode
);
6529 ret
= btrfs_orphan_add(trans
, inode
);
6531 btrfs_end_transaction(trans
, root
);
6535 nr
= trans
->blocks_used
;
6536 btrfs_end_transaction(trans
, root
);
6537 btrfs_btree_balance_dirty(root
, nr
);
6539 /* Now start a transaction for the truncate */
6540 trans
= btrfs_start_transaction(root
, 0);
6542 return PTR_ERR(trans
);
6543 btrfs_set_trans_block_group(trans
, inode
);
6544 trans
->block_rsv
= root
->orphan_block_rsv
;
6547 * setattr is responsible for setting the ordered_data_close flag,
6548 * but that is only tested during the last file release. That
6549 * could happen well after the next commit, leaving a great big
6550 * window where new writes may get lost if someone chooses to write
6551 * to this file after truncating to zero
6553 * The inode doesn't have any dirty data here, and so if we commit
6554 * this is a noop. If someone immediately starts writing to the inode
6555 * it is very likely we'll catch some of their writes in this
6556 * transaction, and the commit will find this file on the ordered
6557 * data list with good things to send down.
6559 * This is a best effort solution, there is still a window where
6560 * using truncate to replace the contents of the file will
6561 * end up with a zero length file after a crash.
6563 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6564 btrfs_add_ordered_operation(trans
, root
, inode
);
6568 trans
= btrfs_start_transaction(root
, 0);
6570 return PTR_ERR(trans
);
6571 btrfs_set_trans_block_group(trans
, inode
);
6572 trans
->block_rsv
= root
->orphan_block_rsv
;
6575 ret
= btrfs_block_rsv_check(trans
, root
,
6576 root
->orphan_block_rsv
, 0, 5);
6577 if (ret
== -EAGAIN
) {
6578 ret
= btrfs_commit_transaction(trans
, root
);
6588 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6590 BTRFS_EXTENT_DATA_KEY
);
6591 if (ret
!= -EAGAIN
) {
6596 ret
= btrfs_update_inode(trans
, root
, inode
);
6602 nr
= trans
->blocks_used
;
6603 btrfs_end_transaction(trans
, root
);
6605 btrfs_btree_balance_dirty(root
, nr
);
6608 if (ret
== 0 && inode
->i_nlink
> 0) {
6609 ret
= btrfs_orphan_del(trans
, inode
);
6612 } else if (ret
&& inode
->i_nlink
> 0) {
6614 * Failed to do the truncate, remove us from the in memory
6617 ret
= btrfs_orphan_del(NULL
, inode
);
6620 ret
= btrfs_update_inode(trans
, root
, inode
);
6624 nr
= trans
->blocks_used
;
6625 ret
= btrfs_end_transaction_throttle(trans
, root
);
6628 btrfs_btree_balance_dirty(root
, nr
);
6634 * create a new subvolume directory/inode (helper for the ioctl).
6636 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6637 struct btrfs_root
*new_root
,
6638 u64 new_dirid
, u64 alloc_hint
)
6640 struct inode
*inode
;
6644 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6645 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
6647 return PTR_ERR(inode
);
6648 inode
->i_op
= &btrfs_dir_inode_operations
;
6649 inode
->i_fop
= &btrfs_dir_file_operations
;
6652 btrfs_i_size_write(inode
, 0);
6654 err
= btrfs_update_inode(trans
, new_root
, inode
);
6661 /* helper function for file defrag and space balancing. This
6662 * forces readahead on a given range of bytes in an inode
6664 unsigned long btrfs_force_ra(struct address_space
*mapping
,
6665 struct file_ra_state
*ra
, struct file
*file
,
6666 pgoff_t offset
, pgoff_t last_index
)
6668 pgoff_t req_size
= last_index
- offset
+ 1;
6670 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
6671 return offset
+ req_size
;
6674 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6676 struct btrfs_inode
*ei
;
6677 struct inode
*inode
;
6679 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6684 ei
->space_info
= NULL
;
6688 ei
->last_sub_trans
= 0;
6689 ei
->logged_trans
= 0;
6690 ei
->delalloc_bytes
= 0;
6691 ei
->reserved_bytes
= 0;
6692 ei
->disk_i_size
= 0;
6694 ei
->index_cnt
= (u64
)-1;
6695 ei
->last_unlink_trans
= 0;
6697 atomic_set(&ei
->outstanding_extents
, 0);
6698 atomic_set(&ei
->reserved_extents
, 0);
6700 ei
->ordered_data_close
= 0;
6701 ei
->orphan_meta_reserved
= 0;
6702 ei
->dummy_inode
= 0;
6703 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6705 inode
= &ei
->vfs_inode
;
6706 extent_map_tree_init(&ei
->extent_tree
, GFP_NOFS
);
6707 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
, GFP_NOFS
);
6708 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
, GFP_NOFS
);
6709 mutex_init(&ei
->log_mutex
);
6710 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6711 INIT_LIST_HEAD(&ei
->i_orphan
);
6712 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6713 INIT_LIST_HEAD(&ei
->ordered_operations
);
6714 RB_CLEAR_NODE(&ei
->rb_node
);
6719 static void btrfs_i_callback(struct rcu_head
*head
)
6721 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6722 INIT_LIST_HEAD(&inode
->i_dentry
);
6723 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6726 void btrfs_destroy_inode(struct inode
*inode
)
6728 struct btrfs_ordered_extent
*ordered
;
6729 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6731 WARN_ON(!list_empty(&inode
->i_dentry
));
6732 WARN_ON(inode
->i_data
.nrpages
);
6733 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
));
6734 WARN_ON(atomic_read(&BTRFS_I(inode
)->reserved_extents
));
6737 * This can happen where we create an inode, but somebody else also
6738 * created the same inode and we need to destroy the one we already
6745 * Make sure we're properly removed from the ordered operation
6749 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6750 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6751 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6752 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6755 if (root
== root
->fs_info
->tree_root
) {
6756 struct btrfs_block_group_cache
*block_group
;
6758 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6759 BTRFS_I(inode
)->block_group
);
6760 if (block_group
&& block_group
->inode
== inode
) {
6761 spin_lock(&block_group
->lock
);
6762 block_group
->inode
= NULL
;
6763 spin_unlock(&block_group
->lock
);
6764 btrfs_put_block_group(block_group
);
6765 } else if (block_group
) {
6766 btrfs_put_block_group(block_group
);
6770 spin_lock(&root
->orphan_lock
);
6771 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6772 printk(KERN_INFO
"BTRFS: inode %lu still on the orphan list\n",
6774 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6776 spin_unlock(&root
->orphan_lock
);
6779 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6783 printk(KERN_ERR
"btrfs found ordered "
6784 "extent %llu %llu on inode cleanup\n",
6785 (unsigned long long)ordered
->file_offset
,
6786 (unsigned long long)ordered
->len
);
6787 btrfs_remove_ordered_extent(inode
, ordered
);
6788 btrfs_put_ordered_extent(ordered
);
6789 btrfs_put_ordered_extent(ordered
);
6792 inode_tree_del(inode
);
6793 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6795 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6798 int btrfs_drop_inode(struct inode
*inode
)
6800 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6802 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6803 root
!= root
->fs_info
->tree_root
)
6806 return generic_drop_inode(inode
);
6809 static void init_once(void *foo
)
6811 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6813 inode_init_once(&ei
->vfs_inode
);
6816 void btrfs_destroy_cachep(void)
6818 if (btrfs_inode_cachep
)
6819 kmem_cache_destroy(btrfs_inode_cachep
);
6820 if (btrfs_trans_handle_cachep
)
6821 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6822 if (btrfs_transaction_cachep
)
6823 kmem_cache_destroy(btrfs_transaction_cachep
);
6824 if (btrfs_path_cachep
)
6825 kmem_cache_destroy(btrfs_path_cachep
);
6826 if (btrfs_free_space_cachep
)
6827 kmem_cache_destroy(btrfs_free_space_cachep
);
6830 int btrfs_init_cachep(void)
6832 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6833 sizeof(struct btrfs_inode
), 0,
6834 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6835 if (!btrfs_inode_cachep
)
6838 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6839 sizeof(struct btrfs_trans_handle
), 0,
6840 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6841 if (!btrfs_trans_handle_cachep
)
6844 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6845 sizeof(struct btrfs_transaction
), 0,
6846 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6847 if (!btrfs_transaction_cachep
)
6850 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6851 sizeof(struct btrfs_path
), 0,
6852 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6853 if (!btrfs_path_cachep
)
6856 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6857 sizeof(struct btrfs_free_space
), 0,
6858 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6859 if (!btrfs_free_space_cachep
)
6864 btrfs_destroy_cachep();
6868 static int btrfs_getattr(struct vfsmount
*mnt
,
6869 struct dentry
*dentry
, struct kstat
*stat
)
6871 struct inode
*inode
= dentry
->d_inode
;
6872 generic_fillattr(inode
, stat
);
6873 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
6874 stat
->blksize
= PAGE_CACHE_SIZE
;
6875 stat
->blocks
= (inode_get_bytes(inode
) +
6876 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
6881 * If a file is moved, it will inherit the cow and compression flags of the new
6884 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
6886 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
6887 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
6889 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
6890 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
6892 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
6894 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
)
6895 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
6897 b_inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6900 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6901 struct inode
*new_dir
, struct dentry
*new_dentry
)
6903 struct btrfs_trans_handle
*trans
;
6904 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6905 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6906 struct inode
*new_inode
= new_dentry
->d_inode
;
6907 struct inode
*old_inode
= old_dentry
->d_inode
;
6908 struct timespec ctime
= CURRENT_TIME
;
6913 if (new_dir
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
6916 /* we only allow rename subvolume link between subvolumes */
6917 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
6920 if (old_inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
6921 (new_inode
&& new_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
))
6924 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
6925 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
6928 * we're using rename to replace one file with another.
6929 * and the replacement file is large. Start IO on it now so
6930 * we don't add too much work to the end of the transaction
6932 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
6933 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
6934 filemap_flush(old_inode
->i_mapping
);
6936 /* close the racy window with snapshot create/destroy ioctl */
6937 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
6938 down_read(&root
->fs_info
->subvol_sem
);
6940 * We want to reserve the absolute worst case amount of items. So if
6941 * both inodes are subvols and we need to unlink them then that would
6942 * require 4 item modifications, but if they are both normal inodes it
6943 * would require 5 item modifications, so we'll assume their normal
6944 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6945 * should cover the worst case number of items we'll modify.
6947 trans
= btrfs_start_transaction(root
, 20);
6949 return PTR_ERR(trans
);
6951 btrfs_set_trans_block_group(trans
, new_dir
);
6954 btrfs_record_root_in_trans(trans
, dest
);
6956 ret
= btrfs_set_inode_index(new_dir
, &index
);
6960 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6961 /* force full log commit if subvolume involved. */
6962 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
6964 ret
= btrfs_insert_inode_ref(trans
, dest
,
6965 new_dentry
->d_name
.name
,
6966 new_dentry
->d_name
.len
,
6968 new_dir
->i_ino
, index
);
6972 * this is an ugly little race, but the rename is required
6973 * to make sure that if we crash, the inode is either at the
6974 * old name or the new one. pinning the log transaction lets
6975 * us make sure we don't allow a log commit to come in after
6976 * we unlink the name but before we add the new name back in.
6978 btrfs_pin_log_trans(root
);
6981 * make sure the inode gets flushed if it is replacing
6984 if (new_inode
&& new_inode
->i_size
&&
6985 old_inode
&& S_ISREG(old_inode
->i_mode
)) {
6986 btrfs_add_ordered_operation(trans
, root
, old_inode
);
6989 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
6990 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
6991 old_inode
->i_ctime
= ctime
;
6993 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
6994 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
6996 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6997 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
6998 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
6999 old_dentry
->d_name
.name
,
7000 old_dentry
->d_name
.len
);
7002 btrfs_inc_nlink(old_dentry
->d_inode
);
7003 ret
= btrfs_unlink_inode(trans
, root
, old_dir
,
7004 old_dentry
->d_inode
,
7005 old_dentry
->d_name
.name
,
7006 old_dentry
->d_name
.len
);
7011 new_inode
->i_ctime
= CURRENT_TIME
;
7012 if (unlikely(new_inode
->i_ino
==
7013 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7014 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7015 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7017 new_dentry
->d_name
.name
,
7018 new_dentry
->d_name
.len
);
7019 BUG_ON(new_inode
->i_nlink
== 0);
7021 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7022 new_dentry
->d_inode
,
7023 new_dentry
->d_name
.name
,
7024 new_dentry
->d_name
.len
);
7027 if (new_inode
->i_nlink
== 0) {
7028 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7033 fixup_inode_flags(new_dir
, old_inode
);
7035 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7036 new_dentry
->d_name
.name
,
7037 new_dentry
->d_name
.len
, 0, index
);
7040 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7041 struct dentry
*parent
= dget_parent(new_dentry
);
7042 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7044 btrfs_end_log_trans(root
);
7047 btrfs_end_transaction_throttle(trans
, root
);
7049 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7050 up_read(&root
->fs_info
->subvol_sem
);
7056 * some fairly slow code that needs optimization. This walks the list
7057 * of all the inodes with pending delalloc and forces them to disk.
7059 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7061 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7062 struct btrfs_inode
*binode
;
7063 struct inode
*inode
;
7065 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7068 spin_lock(&root
->fs_info
->delalloc_lock
);
7069 while (!list_empty(head
)) {
7070 binode
= list_entry(head
->next
, struct btrfs_inode
,
7072 inode
= igrab(&binode
->vfs_inode
);
7074 list_del_init(&binode
->delalloc_inodes
);
7075 spin_unlock(&root
->fs_info
->delalloc_lock
);
7077 filemap_flush(inode
->i_mapping
);
7079 btrfs_add_delayed_iput(inode
);
7084 spin_lock(&root
->fs_info
->delalloc_lock
);
7086 spin_unlock(&root
->fs_info
->delalloc_lock
);
7088 /* the filemap_flush will queue IO into the worker threads, but
7089 * we have to make sure the IO is actually started and that
7090 * ordered extents get created before we return
7092 atomic_inc(&root
->fs_info
->async_submit_draining
);
7093 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7094 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7095 wait_event(root
->fs_info
->async_submit_wait
,
7096 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7097 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7099 atomic_dec(&root
->fs_info
->async_submit_draining
);
7103 int btrfs_start_one_delalloc_inode(struct btrfs_root
*root
, int delay_iput
,
7106 struct btrfs_inode
*binode
;
7107 struct inode
*inode
= NULL
;
7109 spin_lock(&root
->fs_info
->delalloc_lock
);
7110 while (!list_empty(&root
->fs_info
->delalloc_inodes
)) {
7111 binode
= list_entry(root
->fs_info
->delalloc_inodes
.next
,
7112 struct btrfs_inode
, delalloc_inodes
);
7113 inode
= igrab(&binode
->vfs_inode
);
7115 list_move_tail(&binode
->delalloc_inodes
,
7116 &root
->fs_info
->delalloc_inodes
);
7120 list_del_init(&binode
->delalloc_inodes
);
7121 cond_resched_lock(&root
->fs_info
->delalloc_lock
);
7123 spin_unlock(&root
->fs_info
->delalloc_lock
);
7127 filemap_write_and_wait(inode
->i_mapping
);
7129 * We have to do this because compression doesn't
7130 * actually set PG_writeback until it submits the pages
7131 * for IO, which happens in an async thread, so we could
7132 * race and not actually wait for any writeback pages
7133 * because they've not been submitted yet. Technically
7134 * this could still be the case for the ordered stuff
7135 * since the async thread may not have started to do its
7136 * work yet. If this becomes the case then we need to
7137 * figure out a way to make sure that in writepage we
7138 * wait for any async pages to be submitted before
7139 * returning so that fdatawait does what its supposed to
7142 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
7144 filemap_flush(inode
->i_mapping
);
7147 btrfs_add_delayed_iput(inode
);
7155 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7156 const char *symname
)
7158 struct btrfs_trans_handle
*trans
;
7159 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7160 struct btrfs_path
*path
;
7161 struct btrfs_key key
;
7162 struct inode
*inode
= NULL
;
7170 struct btrfs_file_extent_item
*ei
;
7171 struct extent_buffer
*leaf
;
7172 unsigned long nr
= 0;
7174 name_len
= strlen(symname
) + 1;
7175 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7176 return -ENAMETOOLONG
;
7178 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
7182 * 2 items for inode item and ref
7183 * 2 items for dir items
7184 * 1 item for xattr if selinux is on
7186 trans
= btrfs_start_transaction(root
, 5);
7188 return PTR_ERR(trans
);
7190 btrfs_set_trans_block_group(trans
, dir
);
7192 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7193 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
7194 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
7196 err
= PTR_ERR(inode
);
7200 err
= btrfs_init_inode_security(trans
, inode
, dir
);
7206 btrfs_set_trans_block_group(trans
, inode
);
7207 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7211 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7212 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7213 inode
->i_fop
= &btrfs_file_operations
;
7214 inode
->i_op
= &btrfs_file_inode_operations
;
7215 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7217 btrfs_update_inode_block_group(trans
, inode
);
7218 btrfs_update_inode_block_group(trans
, dir
);
7222 path
= btrfs_alloc_path();
7224 key
.objectid
= inode
->i_ino
;
7226 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7227 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7228 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7234 leaf
= path
->nodes
[0];
7235 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7236 struct btrfs_file_extent_item
);
7237 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7238 btrfs_set_file_extent_type(leaf
, ei
,
7239 BTRFS_FILE_EXTENT_INLINE
);
7240 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7241 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7242 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7243 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7245 ptr
= btrfs_file_extent_inline_start(ei
);
7246 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7247 btrfs_mark_buffer_dirty(leaf
);
7248 btrfs_free_path(path
);
7250 inode
->i_op
= &btrfs_symlink_inode_operations
;
7251 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7252 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7253 inode_set_bytes(inode
, name_len
);
7254 btrfs_i_size_write(inode
, name_len
- 1);
7255 err
= btrfs_update_inode(trans
, root
, inode
);
7260 nr
= trans
->blocks_used
;
7261 btrfs_end_transaction_throttle(trans
, root
);
7263 inode_dec_link_count(inode
);
7266 btrfs_btree_balance_dirty(root
, nr
);
7270 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7271 u64 start
, u64 num_bytes
, u64 min_size
,
7272 loff_t actual_len
, u64
*alloc_hint
,
7273 struct btrfs_trans_handle
*trans
)
7275 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7276 struct btrfs_key ins
;
7277 u64 cur_offset
= start
;
7280 bool own_trans
= true;
7284 while (num_bytes
> 0) {
7286 trans
= btrfs_start_transaction(root
, 3);
7287 if (IS_ERR(trans
)) {
7288 ret
= PTR_ERR(trans
);
7293 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7294 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7297 btrfs_end_transaction(trans
, root
);
7301 ret
= insert_reserved_file_extent(trans
, inode
,
7302 cur_offset
, ins
.objectid
,
7303 ins
.offset
, ins
.offset
,
7304 ins
.offset
, 0, 0, 0,
7305 BTRFS_FILE_EXTENT_PREALLOC
);
7307 btrfs_drop_extent_cache(inode
, cur_offset
,
7308 cur_offset
+ ins
.offset
-1, 0);
7310 num_bytes
-= ins
.offset
;
7311 cur_offset
+= ins
.offset
;
7312 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7314 inode
->i_ctime
= CURRENT_TIME
;
7315 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7316 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7317 (actual_len
> inode
->i_size
) &&
7318 (cur_offset
> inode
->i_size
)) {
7319 if (cur_offset
> actual_len
)
7320 i_size
= actual_len
;
7322 i_size
= cur_offset
;
7323 i_size_write(inode
, i_size
);
7324 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7327 ret
= btrfs_update_inode(trans
, root
, inode
);
7331 btrfs_end_transaction(trans
, root
);
7336 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7337 u64 start
, u64 num_bytes
, u64 min_size
,
7338 loff_t actual_len
, u64
*alloc_hint
)
7340 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7341 min_size
, actual_len
, alloc_hint
,
7345 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7346 struct btrfs_trans_handle
*trans
, int mode
,
7347 u64 start
, u64 num_bytes
, u64 min_size
,
7348 loff_t actual_len
, u64
*alloc_hint
)
7350 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7351 min_size
, actual_len
, alloc_hint
, trans
);
7354 static int btrfs_set_page_dirty(struct page
*page
)
7356 return __set_page_dirty_nobuffers(page
);
7359 static int btrfs_permission(struct inode
*inode
, int mask
, unsigned int flags
)
7361 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7363 if (btrfs_root_readonly(root
) && (mask
& MAY_WRITE
))
7365 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
7367 return generic_permission(inode
, mask
, flags
, btrfs_check_acl
);
7370 static const struct inode_operations btrfs_dir_inode_operations
= {
7371 .getattr
= btrfs_getattr
,
7372 .lookup
= btrfs_lookup
,
7373 .create
= btrfs_create
,
7374 .unlink
= btrfs_unlink
,
7376 .mkdir
= btrfs_mkdir
,
7377 .rmdir
= btrfs_rmdir
,
7378 .rename
= btrfs_rename
,
7379 .symlink
= btrfs_symlink
,
7380 .setattr
= btrfs_setattr
,
7381 .mknod
= btrfs_mknod
,
7382 .setxattr
= btrfs_setxattr
,
7383 .getxattr
= btrfs_getxattr
,
7384 .listxattr
= btrfs_listxattr
,
7385 .removexattr
= btrfs_removexattr
,
7386 .permission
= btrfs_permission
,
7388 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7389 .lookup
= btrfs_lookup
,
7390 .permission
= btrfs_permission
,
7393 static const struct file_operations btrfs_dir_file_operations
= {
7394 .llseek
= generic_file_llseek
,
7395 .read
= generic_read_dir
,
7396 .readdir
= btrfs_real_readdir
,
7397 .unlocked_ioctl
= btrfs_ioctl
,
7398 #ifdef CONFIG_COMPAT
7399 .compat_ioctl
= btrfs_ioctl
,
7401 .release
= btrfs_release_file
,
7402 .fsync
= btrfs_sync_file
,
7405 static struct extent_io_ops btrfs_extent_io_ops
= {
7406 .fill_delalloc
= run_delalloc_range
,
7407 .submit_bio_hook
= btrfs_submit_bio_hook
,
7408 .merge_bio_hook
= btrfs_merge_bio_hook
,
7409 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7410 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7411 .writepage_start_hook
= btrfs_writepage_start_hook
,
7412 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
7413 .set_bit_hook
= btrfs_set_bit_hook
,
7414 .clear_bit_hook
= btrfs_clear_bit_hook
,
7415 .merge_extent_hook
= btrfs_merge_extent_hook
,
7416 .split_extent_hook
= btrfs_split_extent_hook
,
7420 * btrfs doesn't support the bmap operation because swapfiles
7421 * use bmap to make a mapping of extents in the file. They assume
7422 * these extents won't change over the life of the file and they
7423 * use the bmap result to do IO directly to the drive.
7425 * the btrfs bmap call would return logical addresses that aren't
7426 * suitable for IO and they also will change frequently as COW
7427 * operations happen. So, swapfile + btrfs == corruption.
7429 * For now we're avoiding this by dropping bmap.
7431 static const struct address_space_operations btrfs_aops
= {
7432 .readpage
= btrfs_readpage
,
7433 .writepage
= btrfs_writepage
,
7434 .writepages
= btrfs_writepages
,
7435 .readpages
= btrfs_readpages
,
7436 .sync_page
= block_sync_page
,
7437 .direct_IO
= btrfs_direct_IO
,
7438 .invalidatepage
= btrfs_invalidatepage
,
7439 .releasepage
= btrfs_releasepage
,
7440 .set_page_dirty
= btrfs_set_page_dirty
,
7441 .error_remove_page
= generic_error_remove_page
,
7444 static const struct address_space_operations btrfs_symlink_aops
= {
7445 .readpage
= btrfs_readpage
,
7446 .writepage
= btrfs_writepage
,
7447 .invalidatepage
= btrfs_invalidatepage
,
7448 .releasepage
= btrfs_releasepage
,
7451 static const struct inode_operations btrfs_file_inode_operations
= {
7452 .getattr
= btrfs_getattr
,
7453 .setattr
= btrfs_setattr
,
7454 .setxattr
= btrfs_setxattr
,
7455 .getxattr
= btrfs_getxattr
,
7456 .listxattr
= btrfs_listxattr
,
7457 .removexattr
= btrfs_removexattr
,
7458 .permission
= btrfs_permission
,
7459 .fiemap
= btrfs_fiemap
,
7461 static const struct inode_operations btrfs_special_inode_operations
= {
7462 .getattr
= btrfs_getattr
,
7463 .setattr
= btrfs_setattr
,
7464 .permission
= btrfs_permission
,
7465 .setxattr
= btrfs_setxattr
,
7466 .getxattr
= btrfs_getxattr
,
7467 .listxattr
= btrfs_listxattr
,
7468 .removexattr
= btrfs_removexattr
,
7470 static const struct inode_operations btrfs_symlink_inode_operations
= {
7471 .readlink
= generic_readlink
,
7472 .follow_link
= page_follow_link_light
,
7473 .put_link
= page_put_link
,
7474 .getattr
= btrfs_getattr
,
7475 .permission
= btrfs_permission
,
7476 .setxattr
= btrfs_setxattr
,
7477 .getxattr
= btrfs_getxattr
,
7478 .listxattr
= btrfs_listxattr
,
7479 .removexattr
= btrfs_removexattr
,
7482 const struct dentry_operations btrfs_dentry_operations
= {
7483 .d_delete
= btrfs_dentry_delete
,