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
);
2539 if (location
.objectid
== BTRFS_FREE_SPACE_OBJECTID
)
2540 inode
->i_mapping
->flags
&= ~__GFP_FS
;
2543 * try to precache a NULL acl entry for files that don't have
2544 * any xattrs or acls
2546 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0], inode
->i_ino
);
2548 cache_no_acl(inode
);
2550 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2551 alloc_group_block
, 0);
2552 btrfs_free_path(path
);
2555 switch (inode
->i_mode
& S_IFMT
) {
2557 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2558 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2559 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2560 inode
->i_fop
= &btrfs_file_operations
;
2561 inode
->i_op
= &btrfs_file_inode_operations
;
2564 inode
->i_fop
= &btrfs_dir_file_operations
;
2565 if (root
== root
->fs_info
->tree_root
)
2566 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2568 inode
->i_op
= &btrfs_dir_inode_operations
;
2571 inode
->i_op
= &btrfs_symlink_inode_operations
;
2572 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2573 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2576 inode
->i_op
= &btrfs_special_inode_operations
;
2577 init_special_inode(inode
, inode
->i_mode
, rdev
);
2581 btrfs_update_iflags(inode
);
2585 btrfs_free_path(path
);
2586 make_bad_inode(inode
);
2590 * given a leaf and an inode, copy the inode fields into the leaf
2592 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2593 struct extent_buffer
*leaf
,
2594 struct btrfs_inode_item
*item
,
2595 struct inode
*inode
)
2597 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2598 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2599 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2600 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2601 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2603 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2604 inode
->i_atime
.tv_sec
);
2605 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2606 inode
->i_atime
.tv_nsec
);
2608 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2609 inode
->i_mtime
.tv_sec
);
2610 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2611 inode
->i_mtime
.tv_nsec
);
2613 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2614 inode
->i_ctime
.tv_sec
);
2615 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2616 inode
->i_ctime
.tv_nsec
);
2618 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2619 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2620 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2621 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2622 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2623 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2624 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2628 * copy everything in the in-memory inode into the btree.
2630 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2631 struct btrfs_root
*root
, struct inode
*inode
)
2633 struct btrfs_inode_item
*inode_item
;
2634 struct btrfs_path
*path
;
2635 struct extent_buffer
*leaf
;
2638 path
= btrfs_alloc_path();
2640 path
->leave_spinning
= 1;
2641 ret
= btrfs_lookup_inode(trans
, root
, path
,
2642 &BTRFS_I(inode
)->location
, 1);
2649 btrfs_unlock_up_safe(path
, 1);
2650 leaf
= path
->nodes
[0];
2651 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2652 struct btrfs_inode_item
);
2654 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2655 btrfs_mark_buffer_dirty(leaf
);
2656 btrfs_set_inode_last_trans(trans
, inode
);
2659 btrfs_free_path(path
);
2665 * unlink helper that gets used here in inode.c and in the tree logging
2666 * recovery code. It remove a link in a directory with a given name, and
2667 * also drops the back refs in the inode to the directory
2669 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2670 struct btrfs_root
*root
,
2671 struct inode
*dir
, struct inode
*inode
,
2672 const char *name
, int name_len
)
2674 struct btrfs_path
*path
;
2676 struct extent_buffer
*leaf
;
2677 struct btrfs_dir_item
*di
;
2678 struct btrfs_key key
;
2681 path
= btrfs_alloc_path();
2687 path
->leave_spinning
= 1;
2688 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2689 name
, name_len
, -1);
2698 leaf
= path
->nodes
[0];
2699 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2700 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2703 btrfs_release_path(root
, path
);
2705 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2707 dir
->i_ino
, &index
);
2709 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2710 "inode %lu parent %lu\n", name_len
, name
,
2711 inode
->i_ino
, dir
->i_ino
);
2715 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2716 index
, name
, name_len
, -1);
2725 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2726 btrfs_release_path(root
, path
);
2728 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2730 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2732 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2737 btrfs_free_path(path
);
2741 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2742 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2743 btrfs_update_inode(trans
, root
, dir
);
2748 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2749 struct btrfs_root
*root
,
2750 struct inode
*dir
, struct inode
*inode
,
2751 const char *name
, int name_len
)
2754 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2756 btrfs_drop_nlink(inode
);
2757 ret
= btrfs_update_inode(trans
, root
, inode
);
2763 /* helper to check if there is any shared block in the path */
2764 static int check_path_shared(struct btrfs_root
*root
,
2765 struct btrfs_path
*path
)
2767 struct extent_buffer
*eb
;
2771 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2774 if (!path
->nodes
[level
])
2776 eb
= path
->nodes
[level
];
2777 if (!btrfs_block_can_be_shared(root
, eb
))
2779 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2788 * helper to start transaction for unlink and rmdir.
2790 * unlink and rmdir are special in btrfs, they do not always free space.
2791 * so in enospc case, we should make sure they will free space before
2792 * allowing them to use the global metadata reservation.
2794 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2795 struct dentry
*dentry
)
2797 struct btrfs_trans_handle
*trans
;
2798 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2799 struct btrfs_path
*path
;
2800 struct btrfs_inode_ref
*ref
;
2801 struct btrfs_dir_item
*di
;
2802 struct inode
*inode
= dentry
->d_inode
;
2808 trans
= btrfs_start_transaction(root
, 10);
2809 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2812 if (inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2813 return ERR_PTR(-ENOSPC
);
2815 /* check if there is someone else holds reference */
2816 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2817 return ERR_PTR(-ENOSPC
);
2819 if (atomic_read(&inode
->i_count
) > 2)
2820 return ERR_PTR(-ENOSPC
);
2822 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2823 return ERR_PTR(-ENOSPC
);
2825 path
= btrfs_alloc_path();
2827 root
->fs_info
->enospc_unlink
= 0;
2828 return ERR_PTR(-ENOMEM
);
2831 trans
= btrfs_start_transaction(root
, 0);
2832 if (IS_ERR(trans
)) {
2833 btrfs_free_path(path
);
2834 root
->fs_info
->enospc_unlink
= 0;
2838 path
->skip_locking
= 1;
2839 path
->search_commit_root
= 1;
2841 ret
= btrfs_lookup_inode(trans
, root
, path
,
2842 &BTRFS_I(dir
)->location
, 0);
2848 if (check_path_shared(root
, path
))
2853 btrfs_release_path(root
, path
);
2855 ret
= btrfs_lookup_inode(trans
, root
, path
,
2856 &BTRFS_I(inode
)->location
, 0);
2862 if (check_path_shared(root
, path
))
2867 btrfs_release_path(root
, path
);
2869 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2870 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2871 inode
->i_ino
, (u64
)-1, 0);
2877 if (check_path_shared(root
, path
))
2879 btrfs_release_path(root
, path
);
2887 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2888 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2894 if (check_path_shared(root
, path
))
2900 btrfs_release_path(root
, path
);
2902 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2903 dentry
->d_name
.name
, dentry
->d_name
.len
,
2904 inode
->i_ino
, dir
->i_ino
, 0);
2910 if (check_path_shared(root
, path
))
2912 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2913 btrfs_release_path(root
, path
);
2915 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
, index
,
2916 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2921 BUG_ON(ret
== -ENOENT
);
2922 if (check_path_shared(root
, path
))
2927 btrfs_free_path(path
);
2929 btrfs_end_transaction(trans
, root
);
2930 root
->fs_info
->enospc_unlink
= 0;
2931 return ERR_PTR(err
);
2934 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2938 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2939 struct btrfs_root
*root
)
2941 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2942 BUG_ON(!root
->fs_info
->enospc_unlink
);
2943 root
->fs_info
->enospc_unlink
= 0;
2945 btrfs_end_transaction_throttle(trans
, root
);
2948 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2950 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2951 struct btrfs_trans_handle
*trans
;
2952 struct inode
*inode
= dentry
->d_inode
;
2954 unsigned long nr
= 0;
2956 trans
= __unlink_start_trans(dir
, dentry
);
2958 return PTR_ERR(trans
);
2960 btrfs_set_trans_block_group(trans
, dir
);
2962 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2964 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2965 dentry
->d_name
.name
, dentry
->d_name
.len
);
2968 if (inode
->i_nlink
== 0) {
2969 ret
= btrfs_orphan_add(trans
, inode
);
2973 nr
= trans
->blocks_used
;
2974 __unlink_end_trans(trans
, root
);
2975 btrfs_btree_balance_dirty(root
, nr
);
2979 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2980 struct btrfs_root
*root
,
2981 struct inode
*dir
, u64 objectid
,
2982 const char *name
, int name_len
)
2984 struct btrfs_path
*path
;
2985 struct extent_buffer
*leaf
;
2986 struct btrfs_dir_item
*di
;
2987 struct btrfs_key key
;
2991 path
= btrfs_alloc_path();
2995 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2996 name
, name_len
, -1);
2997 BUG_ON(!di
|| IS_ERR(di
));
2999 leaf
= path
->nodes
[0];
3000 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3001 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3002 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3004 btrfs_release_path(root
, path
);
3006 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3007 objectid
, root
->root_key
.objectid
,
3008 dir
->i_ino
, &index
, name
, name_len
);
3010 BUG_ON(ret
!= -ENOENT
);
3011 di
= btrfs_search_dir_index_item(root
, path
, dir
->i_ino
,
3013 BUG_ON(!di
|| IS_ERR(di
));
3015 leaf
= path
->nodes
[0];
3016 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3017 btrfs_release_path(root
, path
);
3021 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
3022 index
, name
, name_len
, -1);
3023 BUG_ON(!di
|| IS_ERR(di
));
3025 leaf
= path
->nodes
[0];
3026 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3027 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3028 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3030 btrfs_release_path(root
, path
);
3032 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3033 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3034 ret
= btrfs_update_inode(trans
, root
, dir
);
3037 btrfs_free_path(path
);
3041 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3043 struct inode
*inode
= dentry
->d_inode
;
3045 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3046 struct btrfs_trans_handle
*trans
;
3047 unsigned long nr
= 0;
3049 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3050 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
3053 trans
= __unlink_start_trans(dir
, dentry
);
3055 return PTR_ERR(trans
);
3057 btrfs_set_trans_block_group(trans
, dir
);
3059 if (unlikely(inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3060 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3061 BTRFS_I(inode
)->location
.objectid
,
3062 dentry
->d_name
.name
,
3063 dentry
->d_name
.len
);
3067 err
= btrfs_orphan_add(trans
, inode
);
3071 /* now the directory is empty */
3072 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3073 dentry
->d_name
.name
, dentry
->d_name
.len
);
3075 btrfs_i_size_write(inode
, 0);
3077 nr
= trans
->blocks_used
;
3078 __unlink_end_trans(trans
, root
);
3079 btrfs_btree_balance_dirty(root
, nr
);
3086 * when truncating bytes in a file, it is possible to avoid reading
3087 * the leaves that contain only checksum items. This can be the
3088 * majority of the IO required to delete a large file, but it must
3089 * be done carefully.
3091 * The keys in the level just above the leaves are checked to make sure
3092 * the lowest key in a given leaf is a csum key, and starts at an offset
3093 * after the new size.
3095 * Then the key for the next leaf is checked to make sure it also has
3096 * a checksum item for the same file. If it does, we know our target leaf
3097 * contains only checksum items, and it can be safely freed without reading
3100 * This is just an optimization targeted at large files. It may do
3101 * nothing. It will return 0 unless things went badly.
3103 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
3104 struct btrfs_root
*root
,
3105 struct btrfs_path
*path
,
3106 struct inode
*inode
, u64 new_size
)
3108 struct btrfs_key key
;
3111 struct btrfs_key found_key
;
3112 struct btrfs_key other_key
;
3113 struct btrfs_leaf_ref
*ref
;
3117 path
->lowest_level
= 1;
3118 key
.objectid
= inode
->i_ino
;
3119 key
.type
= BTRFS_CSUM_ITEM_KEY
;
3120 key
.offset
= new_size
;
3122 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3126 if (path
->nodes
[1] == NULL
) {
3131 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
3132 nritems
= btrfs_header_nritems(path
->nodes
[1]);
3137 if (path
->slots
[1] >= nritems
)
3140 /* did we find a key greater than anything we want to delete? */
3141 if (found_key
.objectid
> inode
->i_ino
||
3142 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
3145 /* we check the next key in the node to make sure the leave contains
3146 * only checksum items. This comparison doesn't work if our
3147 * leaf is the last one in the node
3149 if (path
->slots
[1] + 1 >= nritems
) {
3151 /* search forward from the last key in the node, this
3152 * will bring us into the next node in the tree
3154 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
3156 /* unlikely, but we inc below, so check to be safe */
3157 if (found_key
.offset
== (u64
)-1)
3160 /* search_forward needs a path with locks held, do the
3161 * search again for the original key. It is possible
3162 * this will race with a balance and return a path that
3163 * we could modify, but this drop is just an optimization
3164 * and is allowed to miss some leaves.
3166 btrfs_release_path(root
, path
);
3169 /* setup a max key for search_forward */
3170 other_key
.offset
= (u64
)-1;
3171 other_key
.type
= key
.type
;
3172 other_key
.objectid
= key
.objectid
;
3174 path
->keep_locks
= 1;
3175 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
3177 path
->keep_locks
= 0;
3178 if (ret
|| found_key
.objectid
!= key
.objectid
||
3179 found_key
.type
!= key
.type
) {
3184 key
.offset
= found_key
.offset
;
3185 btrfs_release_path(root
, path
);
3190 /* we know there's one more slot after us in the tree,
3191 * read that key so we can verify it is also a checksum item
3193 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
3195 if (found_key
.objectid
< inode
->i_ino
)
3198 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
3202 * if the key for the next leaf isn't a csum key from this objectid,
3203 * we can't be sure there aren't good items inside this leaf.
3206 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
3209 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
3210 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
3212 * it is safe to delete this leaf, it contains only
3213 * csum items from this inode at an offset >= new_size
3215 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
3218 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
3219 ref
= btrfs_alloc_leaf_ref(root
, 0);
3221 ref
->root_gen
= root
->root_key
.offset
;
3222 ref
->bytenr
= leaf_start
;
3224 ref
->generation
= leaf_gen
;
3227 btrfs_sort_leaf_ref(ref
);
3229 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
3231 btrfs_free_leaf_ref(root
, ref
);
3237 btrfs_release_path(root
, path
);
3239 if (other_key
.objectid
== inode
->i_ino
&&
3240 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
3241 key
.offset
= other_key
.offset
;
3247 /* fixup any changes we've made to the path */
3248 path
->lowest_level
= 0;
3249 path
->keep_locks
= 0;
3250 btrfs_release_path(root
, path
);
3257 * this can truncate away extent items, csum items and directory items.
3258 * It starts at a high offset and removes keys until it can't find
3259 * any higher than new_size
3261 * csum items that cross the new i_size are truncated to the new size
3264 * min_type is the minimum key type to truncate down to. If set to 0, this
3265 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3267 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3268 struct btrfs_root
*root
,
3269 struct inode
*inode
,
3270 u64 new_size
, u32 min_type
)
3272 struct btrfs_path
*path
;
3273 struct extent_buffer
*leaf
;
3274 struct btrfs_file_extent_item
*fi
;
3275 struct btrfs_key key
;
3276 struct btrfs_key found_key
;
3277 u64 extent_start
= 0;
3278 u64 extent_num_bytes
= 0;
3279 u64 extent_offset
= 0;
3281 u64 mask
= root
->sectorsize
- 1;
3282 u32 found_type
= (u8
)-1;
3285 int pending_del_nr
= 0;
3286 int pending_del_slot
= 0;
3287 int extent_type
= -1;
3292 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3294 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3295 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3297 path
= btrfs_alloc_path();
3301 key
.objectid
= inode
->i_ino
;
3302 key
.offset
= (u64
)-1;
3306 path
->leave_spinning
= 1;
3307 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3314 /* there are no items in the tree for us to truncate, we're
3317 if (path
->slots
[0] == 0)
3324 leaf
= path
->nodes
[0];
3325 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3326 found_type
= btrfs_key_type(&found_key
);
3329 if (found_key
.objectid
!= inode
->i_ino
)
3332 if (found_type
< min_type
)
3335 item_end
= found_key
.offset
;
3336 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3337 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3338 struct btrfs_file_extent_item
);
3339 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3340 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3341 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3342 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3344 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3346 btrfs_file_extent_num_bytes(leaf
, fi
);
3347 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3348 item_end
+= btrfs_file_extent_inline_len(leaf
,
3353 if (found_type
> min_type
) {
3356 if (item_end
< new_size
)
3358 if (found_key
.offset
>= new_size
)
3364 /* FIXME, shrink the extent if the ref count is only 1 */
3365 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3368 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3370 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3371 if (!del_item
&& !encoding
) {
3372 u64 orig_num_bytes
=
3373 btrfs_file_extent_num_bytes(leaf
, fi
);
3374 extent_num_bytes
= new_size
-
3375 found_key
.offset
+ root
->sectorsize
- 1;
3376 extent_num_bytes
= extent_num_bytes
&
3377 ~((u64
)root
->sectorsize
- 1);
3378 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3380 num_dec
= (orig_num_bytes
-
3382 if (root
->ref_cows
&& extent_start
!= 0)
3383 inode_sub_bytes(inode
, num_dec
);
3384 btrfs_mark_buffer_dirty(leaf
);
3387 btrfs_file_extent_disk_num_bytes(leaf
,
3389 extent_offset
= found_key
.offset
-
3390 btrfs_file_extent_offset(leaf
, fi
);
3392 /* FIXME blocksize != 4096 */
3393 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3394 if (extent_start
!= 0) {
3397 inode_sub_bytes(inode
, num_dec
);
3400 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3402 * we can't truncate inline items that have had
3406 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3407 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3408 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3409 u32 size
= new_size
- found_key
.offset
;
3411 if (root
->ref_cows
) {
3412 inode_sub_bytes(inode
, item_end
+ 1 -
3416 btrfs_file_extent_calc_inline_size(size
);
3417 ret
= btrfs_truncate_item(trans
, root
, path
,
3420 } else if (root
->ref_cows
) {
3421 inode_sub_bytes(inode
, item_end
+ 1 -
3427 if (!pending_del_nr
) {
3428 /* no pending yet, add ourselves */
3429 pending_del_slot
= path
->slots
[0];
3431 } else if (pending_del_nr
&&
3432 path
->slots
[0] + 1 == pending_del_slot
) {
3433 /* hop on the pending chunk */
3435 pending_del_slot
= path
->slots
[0];
3442 if (found_extent
&& (root
->ref_cows
||
3443 root
== root
->fs_info
->tree_root
)) {
3444 btrfs_set_path_blocking(path
);
3445 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3446 extent_num_bytes
, 0,
3447 btrfs_header_owner(leaf
),
3448 inode
->i_ino
, extent_offset
);
3452 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3455 if (path
->slots
[0] == 0 ||
3456 path
->slots
[0] != pending_del_slot
) {
3457 if (root
->ref_cows
) {
3461 if (pending_del_nr
) {
3462 ret
= btrfs_del_items(trans
, root
, path
,
3468 btrfs_release_path(root
, path
);
3475 if (pending_del_nr
) {
3476 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3480 btrfs_free_path(path
);
3485 * taken from block_truncate_page, but does cow as it zeros out
3486 * any bytes left in the last page in the file.
3488 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3490 struct inode
*inode
= mapping
->host
;
3491 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3492 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3493 struct btrfs_ordered_extent
*ordered
;
3494 struct extent_state
*cached_state
= NULL
;
3496 u32 blocksize
= root
->sectorsize
;
3497 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3498 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3504 if ((offset
& (blocksize
- 1)) == 0)
3506 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3512 page
= grab_cache_page(mapping
, index
);
3514 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3518 page_start
= page_offset(page
);
3519 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3521 if (!PageUptodate(page
)) {
3522 ret
= btrfs_readpage(NULL
, page
);
3524 if (page
->mapping
!= mapping
) {
3526 page_cache_release(page
);
3529 if (!PageUptodate(page
)) {
3534 wait_on_page_writeback(page
);
3536 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3538 set_page_extent_mapped(page
);
3540 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3542 unlock_extent_cached(io_tree
, page_start
, page_end
,
3543 &cached_state
, GFP_NOFS
);
3545 page_cache_release(page
);
3546 btrfs_start_ordered_extent(inode
, ordered
, 1);
3547 btrfs_put_ordered_extent(ordered
);
3551 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3552 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3553 0, 0, &cached_state
, GFP_NOFS
);
3555 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3558 unlock_extent_cached(io_tree
, page_start
, page_end
,
3559 &cached_state
, GFP_NOFS
);
3564 if (offset
!= PAGE_CACHE_SIZE
) {
3566 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3567 flush_dcache_page(page
);
3570 ClearPageChecked(page
);
3571 set_page_dirty(page
);
3572 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3577 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3579 page_cache_release(page
);
3585 * This function puts in dummy file extents for the area we're creating a hole
3586 * for. So if we are truncating this file to a larger size we need to insert
3587 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3588 * the range between oldsize and size
3590 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3592 struct btrfs_trans_handle
*trans
;
3593 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3594 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3595 struct extent_map
*em
= NULL
;
3596 struct extent_state
*cached_state
= NULL
;
3597 u64 mask
= root
->sectorsize
- 1;
3598 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3599 u64 block_end
= (size
+ mask
) & ~mask
;
3605 if (size
<= hole_start
)
3609 struct btrfs_ordered_extent
*ordered
;
3610 btrfs_wait_ordered_range(inode
, hole_start
,
3611 block_end
- hole_start
);
3612 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3613 &cached_state
, GFP_NOFS
);
3614 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3617 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3618 &cached_state
, GFP_NOFS
);
3619 btrfs_put_ordered_extent(ordered
);
3622 cur_offset
= hole_start
;
3624 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3625 block_end
- cur_offset
, 0);
3626 BUG_ON(IS_ERR(em
) || !em
);
3627 last_byte
= min(extent_map_end(em
), block_end
);
3628 last_byte
= (last_byte
+ mask
) & ~mask
;
3629 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3631 hole_size
= last_byte
- cur_offset
;
3633 trans
= btrfs_start_transaction(root
, 2);
3634 if (IS_ERR(trans
)) {
3635 err
= PTR_ERR(trans
);
3638 btrfs_set_trans_block_group(trans
, inode
);
3640 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3641 cur_offset
+ hole_size
,
3646 err
= btrfs_insert_file_extent(trans
, root
,
3647 inode
->i_ino
, cur_offset
, 0,
3648 0, hole_size
, 0, hole_size
,
3653 btrfs_drop_extent_cache(inode
, hole_start
,
3656 btrfs_end_transaction(trans
, root
);
3658 free_extent_map(em
);
3660 cur_offset
= last_byte
;
3661 if (cur_offset
>= block_end
)
3665 free_extent_map(em
);
3666 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3671 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3673 loff_t oldsize
= i_size_read(inode
);
3676 if (newsize
== oldsize
)
3679 if (newsize
> oldsize
) {
3680 i_size_write(inode
, newsize
);
3681 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3682 truncate_pagecache(inode
, oldsize
, newsize
);
3683 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3685 btrfs_setsize(inode
, oldsize
);
3689 mark_inode_dirty(inode
);
3693 * We're truncating a file that used to have good data down to
3694 * zero. Make sure it gets into the ordered flush list so that
3695 * any new writes get down to disk quickly.
3698 BTRFS_I(inode
)->ordered_data_close
= 1;
3700 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3701 truncate_setsize(inode
, newsize
);
3702 ret
= btrfs_truncate(inode
);
3708 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3710 struct inode
*inode
= dentry
->d_inode
;
3711 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3714 if (btrfs_root_readonly(root
))
3717 err
= inode_change_ok(inode
, attr
);
3721 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3722 err
= btrfs_setsize(inode
, attr
->ia_size
);
3727 if (attr
->ia_valid
) {
3728 setattr_copy(inode
, attr
);
3729 mark_inode_dirty(inode
);
3731 if (attr
->ia_valid
& ATTR_MODE
)
3732 err
= btrfs_acl_chmod(inode
);
3738 void btrfs_evict_inode(struct inode
*inode
)
3740 struct btrfs_trans_handle
*trans
;
3741 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3745 trace_btrfs_inode_evict(inode
);
3747 truncate_inode_pages(&inode
->i_data
, 0);
3748 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3749 root
== root
->fs_info
->tree_root
))
3752 if (is_bad_inode(inode
)) {
3753 btrfs_orphan_del(NULL
, inode
);
3756 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3757 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3759 if (root
->fs_info
->log_root_recovering
) {
3760 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3764 if (inode
->i_nlink
> 0) {
3765 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3769 btrfs_i_size_write(inode
, 0);
3772 trans
= btrfs_start_transaction(root
, 0);
3773 BUG_ON(IS_ERR(trans
));
3774 btrfs_set_trans_block_group(trans
, inode
);
3775 trans
->block_rsv
= root
->orphan_block_rsv
;
3777 ret
= btrfs_block_rsv_check(trans
, root
,
3778 root
->orphan_block_rsv
, 0, 5);
3780 BUG_ON(ret
!= -EAGAIN
);
3781 ret
= btrfs_commit_transaction(trans
, root
);
3786 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3790 nr
= trans
->blocks_used
;
3791 btrfs_end_transaction(trans
, root
);
3793 btrfs_btree_balance_dirty(root
, nr
);
3798 ret
= btrfs_orphan_del(trans
, inode
);
3802 nr
= trans
->blocks_used
;
3803 btrfs_end_transaction(trans
, root
);
3804 btrfs_btree_balance_dirty(root
, nr
);
3806 end_writeback(inode
);
3811 * this returns the key found in the dir entry in the location pointer.
3812 * If no dir entries were found, location->objectid is 0.
3814 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3815 struct btrfs_key
*location
)
3817 const char *name
= dentry
->d_name
.name
;
3818 int namelen
= dentry
->d_name
.len
;
3819 struct btrfs_dir_item
*di
;
3820 struct btrfs_path
*path
;
3821 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3824 path
= btrfs_alloc_path();
3827 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
3832 if (!di
|| IS_ERR(di
))
3835 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3837 btrfs_free_path(path
);
3840 location
->objectid
= 0;
3845 * when we hit a tree root in a directory, the btrfs part of the inode
3846 * needs to be changed to reflect the root directory of the tree root. This
3847 * is kind of like crossing a mount point.
3849 static int fixup_tree_root_location(struct btrfs_root
*root
,
3851 struct dentry
*dentry
,
3852 struct btrfs_key
*location
,
3853 struct btrfs_root
**sub_root
)
3855 struct btrfs_path
*path
;
3856 struct btrfs_root
*new_root
;
3857 struct btrfs_root_ref
*ref
;
3858 struct extent_buffer
*leaf
;
3862 path
= btrfs_alloc_path();
3869 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3870 BTRFS_I(dir
)->root
->root_key
.objectid
,
3871 location
->objectid
);
3878 leaf
= path
->nodes
[0];
3879 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3880 if (btrfs_root_ref_dirid(leaf
, ref
) != dir
->i_ino
||
3881 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3884 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3885 (unsigned long)(ref
+ 1),
3886 dentry
->d_name
.len
);
3890 btrfs_release_path(root
->fs_info
->tree_root
, path
);
3892 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3893 if (IS_ERR(new_root
)) {
3894 err
= PTR_ERR(new_root
);
3898 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3903 *sub_root
= new_root
;
3904 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3905 location
->type
= BTRFS_INODE_ITEM_KEY
;
3906 location
->offset
= 0;
3909 btrfs_free_path(path
);
3913 static void inode_tree_add(struct inode
*inode
)
3915 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3916 struct btrfs_inode
*entry
;
3918 struct rb_node
*parent
;
3920 p
= &root
->inode_tree
.rb_node
;
3923 if (inode_unhashed(inode
))
3926 spin_lock(&root
->inode_lock
);
3929 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3931 if (inode
->i_ino
< entry
->vfs_inode
.i_ino
)
3932 p
= &parent
->rb_left
;
3933 else if (inode
->i_ino
> entry
->vfs_inode
.i_ino
)
3934 p
= &parent
->rb_right
;
3936 WARN_ON(!(entry
->vfs_inode
.i_state
&
3937 (I_WILL_FREE
| I_FREEING
)));
3938 rb_erase(parent
, &root
->inode_tree
);
3939 RB_CLEAR_NODE(parent
);
3940 spin_unlock(&root
->inode_lock
);
3944 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3945 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3946 spin_unlock(&root
->inode_lock
);
3949 static void inode_tree_del(struct inode
*inode
)
3951 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3954 spin_lock(&root
->inode_lock
);
3955 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3956 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3957 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3958 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3960 spin_unlock(&root
->inode_lock
);
3963 * Free space cache has inodes in the tree root, but the tree root has a
3964 * root_refs of 0, so this could end up dropping the tree root as a
3965 * snapshot, so we need the extra !root->fs_info->tree_root check to
3966 * make sure we don't drop it.
3968 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3969 root
!= root
->fs_info
->tree_root
) {
3970 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3971 spin_lock(&root
->inode_lock
);
3972 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3973 spin_unlock(&root
->inode_lock
);
3975 btrfs_add_dead_root(root
);
3979 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3981 struct rb_node
*node
;
3982 struct rb_node
*prev
;
3983 struct btrfs_inode
*entry
;
3984 struct inode
*inode
;
3987 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3989 spin_lock(&root
->inode_lock
);
3991 node
= root
->inode_tree
.rb_node
;
3995 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3997 if (objectid
< entry
->vfs_inode
.i_ino
)
3998 node
= node
->rb_left
;
3999 else if (objectid
> entry
->vfs_inode
.i_ino
)
4000 node
= node
->rb_right
;
4006 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4007 if (objectid
<= entry
->vfs_inode
.i_ino
) {
4011 prev
= rb_next(prev
);
4015 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4016 objectid
= entry
->vfs_inode
.i_ino
+ 1;
4017 inode
= igrab(&entry
->vfs_inode
);
4019 spin_unlock(&root
->inode_lock
);
4020 if (atomic_read(&inode
->i_count
) > 1)
4021 d_prune_aliases(inode
);
4023 * btrfs_drop_inode will have it removed from
4024 * the inode cache when its usage count
4029 spin_lock(&root
->inode_lock
);
4033 if (cond_resched_lock(&root
->inode_lock
))
4036 node
= rb_next(node
);
4038 spin_unlock(&root
->inode_lock
);
4042 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4044 struct btrfs_iget_args
*args
= p
;
4045 inode
->i_ino
= args
->ino
;
4046 BTRFS_I(inode
)->root
= args
->root
;
4047 btrfs_set_inode_space_info(args
->root
, inode
);
4051 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4053 struct btrfs_iget_args
*args
= opaque
;
4054 return args
->ino
== inode
->i_ino
&&
4055 args
->root
== BTRFS_I(inode
)->root
;
4058 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4060 struct btrfs_root
*root
)
4062 struct inode
*inode
;
4063 struct btrfs_iget_args args
;
4064 args
.ino
= objectid
;
4067 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4068 btrfs_init_locked_inode
,
4073 /* Get an inode object given its location and corresponding root.
4074 * Returns in *is_new if the inode was read from disk
4076 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4077 struct btrfs_root
*root
, int *new)
4079 struct inode
*inode
;
4081 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4083 return ERR_PTR(-ENOMEM
);
4085 if (inode
->i_state
& I_NEW
) {
4086 BTRFS_I(inode
)->root
= root
;
4087 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4088 btrfs_read_locked_inode(inode
);
4089 inode_tree_add(inode
);
4090 unlock_new_inode(inode
);
4098 static struct inode
*new_simple_dir(struct super_block
*s
,
4099 struct btrfs_key
*key
,
4100 struct btrfs_root
*root
)
4102 struct inode
*inode
= new_inode(s
);
4105 return ERR_PTR(-ENOMEM
);
4107 BTRFS_I(inode
)->root
= root
;
4108 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4109 BTRFS_I(inode
)->dummy_inode
= 1;
4111 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4112 inode
->i_op
= &simple_dir_inode_operations
;
4113 inode
->i_fop
= &simple_dir_operations
;
4114 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4115 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4120 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4122 struct inode
*inode
;
4123 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4124 struct btrfs_root
*sub_root
= root
;
4125 struct btrfs_key location
;
4129 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4130 return ERR_PTR(-ENAMETOOLONG
);
4132 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4135 return ERR_PTR(ret
);
4137 if (location
.objectid
== 0)
4140 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4141 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4145 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4147 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4148 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4149 &location
, &sub_root
);
4152 inode
= ERR_PTR(ret
);
4154 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4156 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4158 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4160 if (!IS_ERR(inode
) && root
!= sub_root
) {
4161 down_read(&root
->fs_info
->cleanup_work_sem
);
4162 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4163 ret
= btrfs_orphan_cleanup(sub_root
);
4164 up_read(&root
->fs_info
->cleanup_work_sem
);
4166 inode
= ERR_PTR(ret
);
4172 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4174 struct btrfs_root
*root
;
4176 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
4177 dentry
= dentry
->d_parent
;
4179 if (dentry
->d_inode
) {
4180 root
= BTRFS_I(dentry
->d_inode
)->root
;
4181 if (btrfs_root_refs(&root
->root_item
) == 0)
4187 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4188 struct nameidata
*nd
)
4190 struct inode
*inode
;
4192 inode
= btrfs_lookup_dentry(dir
, dentry
);
4194 return ERR_CAST(inode
);
4196 return d_splice_alias(inode
, dentry
);
4199 static unsigned char btrfs_filetype_table
[] = {
4200 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4203 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4206 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4207 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4208 struct btrfs_item
*item
;
4209 struct btrfs_dir_item
*di
;
4210 struct btrfs_key key
;
4211 struct btrfs_key found_key
;
4212 struct btrfs_path
*path
;
4215 struct extent_buffer
*leaf
;
4218 unsigned char d_type
;
4223 int key_type
= BTRFS_DIR_INDEX_KEY
;
4228 /* FIXME, use a real flag for deciding about the key type */
4229 if (root
->fs_info
->tree_root
== root
)
4230 key_type
= BTRFS_DIR_ITEM_KEY
;
4232 /* special case for "." */
4233 if (filp
->f_pos
== 0) {
4234 over
= filldir(dirent
, ".", 1,
4241 /* special case for .., just use the back ref */
4242 if (filp
->f_pos
== 1) {
4243 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4244 over
= filldir(dirent
, "..", 2,
4250 path
= btrfs_alloc_path();
4253 btrfs_set_key_type(&key
, key_type
);
4254 key
.offset
= filp
->f_pos
;
4255 key
.objectid
= inode
->i_ino
;
4257 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4263 leaf
= path
->nodes
[0];
4264 nritems
= btrfs_header_nritems(leaf
);
4265 slot
= path
->slots
[0];
4266 if (advance
|| slot
>= nritems
) {
4267 if (slot
>= nritems
- 1) {
4268 ret
= btrfs_next_leaf(root
, path
);
4271 leaf
= path
->nodes
[0];
4272 nritems
= btrfs_header_nritems(leaf
);
4273 slot
= path
->slots
[0];
4281 item
= btrfs_item_nr(leaf
, slot
);
4282 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4284 if (found_key
.objectid
!= key
.objectid
)
4286 if (btrfs_key_type(&found_key
) != key_type
)
4288 if (found_key
.offset
< filp
->f_pos
)
4291 filp
->f_pos
= found_key
.offset
;
4293 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4295 di_total
= btrfs_item_size(leaf
, item
);
4297 while (di_cur
< di_total
) {
4298 struct btrfs_key location
;
4300 if (verify_dir_item(root
, leaf
, di
))
4303 name_len
= btrfs_dir_name_len(leaf
, di
);
4304 if (name_len
<= sizeof(tmp_name
)) {
4305 name_ptr
= tmp_name
;
4307 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4313 read_extent_buffer(leaf
, name_ptr
,
4314 (unsigned long)(di
+ 1), name_len
);
4316 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4317 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4319 /* is this a reference to our own snapshot? If so
4322 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4323 location
.objectid
== root
->root_key
.objectid
) {
4327 over
= filldir(dirent
, name_ptr
, name_len
,
4328 found_key
.offset
, location
.objectid
,
4332 if (name_ptr
!= tmp_name
)
4337 di_len
= btrfs_dir_name_len(leaf
, di
) +
4338 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4340 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4344 /* Reached end of directory/root. Bump pos past the last item. */
4345 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4347 * 32-bit glibc will use getdents64, but then strtol -
4348 * so the last number we can serve is this.
4350 filp
->f_pos
= 0x7fffffff;
4356 btrfs_free_path(path
);
4360 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4362 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4363 struct btrfs_trans_handle
*trans
;
4365 bool nolock
= false;
4367 if (BTRFS_I(inode
)->dummy_inode
)
4371 nolock
= (root
->fs_info
->closing
&& root
== root
->fs_info
->tree_root
);
4373 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4375 trans
= btrfs_join_transaction_nolock(root
, 1);
4377 trans
= btrfs_join_transaction(root
, 1);
4379 return PTR_ERR(trans
);
4380 btrfs_set_trans_block_group(trans
, inode
);
4382 ret
= btrfs_end_transaction_nolock(trans
, root
);
4384 ret
= btrfs_commit_transaction(trans
, root
);
4390 * This is somewhat expensive, updating the tree every time the
4391 * inode changes. But, it is most likely to find the inode in cache.
4392 * FIXME, needs more benchmarking...there are no reasons other than performance
4393 * to keep or drop this code.
4395 void btrfs_dirty_inode(struct inode
*inode
)
4397 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4398 struct btrfs_trans_handle
*trans
;
4401 if (BTRFS_I(inode
)->dummy_inode
)
4404 trans
= btrfs_join_transaction(root
, 1);
4405 BUG_ON(IS_ERR(trans
));
4406 btrfs_set_trans_block_group(trans
, inode
);
4408 ret
= btrfs_update_inode(trans
, root
, inode
);
4409 if (ret
&& ret
== -ENOSPC
) {
4410 /* whoops, lets try again with the full transaction */
4411 btrfs_end_transaction(trans
, root
);
4412 trans
= btrfs_start_transaction(root
, 1);
4413 if (IS_ERR(trans
)) {
4414 if (printk_ratelimit()) {
4415 printk(KERN_ERR
"btrfs: fail to "
4416 "dirty inode %lu error %ld\n",
4417 inode
->i_ino
, PTR_ERR(trans
));
4421 btrfs_set_trans_block_group(trans
, inode
);
4423 ret
= btrfs_update_inode(trans
, root
, inode
);
4425 if (printk_ratelimit()) {
4426 printk(KERN_ERR
"btrfs: fail to "
4427 "dirty inode %lu error %d\n",
4432 btrfs_end_transaction(trans
, root
);
4436 * find the highest existing sequence number in a directory
4437 * and then set the in-memory index_cnt variable to reflect
4438 * free sequence numbers
4440 static int btrfs_set_inode_index_count(struct inode
*inode
)
4442 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4443 struct btrfs_key key
, found_key
;
4444 struct btrfs_path
*path
;
4445 struct extent_buffer
*leaf
;
4448 key
.objectid
= inode
->i_ino
;
4449 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4450 key
.offset
= (u64
)-1;
4452 path
= btrfs_alloc_path();
4456 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4459 /* FIXME: we should be able to handle this */
4465 * MAGIC NUMBER EXPLANATION:
4466 * since we search a directory based on f_pos we have to start at 2
4467 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4468 * else has to start at 2
4470 if (path
->slots
[0] == 0) {
4471 BTRFS_I(inode
)->index_cnt
= 2;
4477 leaf
= path
->nodes
[0];
4478 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4480 if (found_key
.objectid
!= inode
->i_ino
||
4481 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4482 BTRFS_I(inode
)->index_cnt
= 2;
4486 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4488 btrfs_free_path(path
);
4493 * helper to find a free sequence number in a given directory. This current
4494 * code is very simple, later versions will do smarter things in the btree
4496 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4500 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4501 ret
= btrfs_set_inode_index_count(dir
);
4506 *index
= BTRFS_I(dir
)->index_cnt
;
4507 BTRFS_I(dir
)->index_cnt
++;
4512 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4513 struct btrfs_root
*root
,
4515 const char *name
, int name_len
,
4516 u64 ref_objectid
, u64 objectid
,
4517 u64 alloc_hint
, int mode
, u64
*index
)
4519 struct inode
*inode
;
4520 struct btrfs_inode_item
*inode_item
;
4521 struct btrfs_key
*location
;
4522 struct btrfs_path
*path
;
4523 struct btrfs_inode_ref
*ref
;
4524 struct btrfs_key key
[2];
4530 path
= btrfs_alloc_path();
4533 inode
= new_inode(root
->fs_info
->sb
);
4535 return ERR_PTR(-ENOMEM
);
4538 trace_btrfs_inode_request(dir
);
4540 ret
= btrfs_set_inode_index(dir
, index
);
4543 return ERR_PTR(ret
);
4547 * index_cnt is ignored for everything but a dir,
4548 * btrfs_get_inode_index_count has an explanation for the magic
4551 BTRFS_I(inode
)->index_cnt
= 2;
4552 BTRFS_I(inode
)->root
= root
;
4553 BTRFS_I(inode
)->generation
= trans
->transid
;
4554 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4555 btrfs_set_inode_space_info(root
, inode
);
4561 BTRFS_I(inode
)->block_group
=
4562 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
4564 key
[0].objectid
= objectid
;
4565 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4568 key
[1].objectid
= objectid
;
4569 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4570 key
[1].offset
= ref_objectid
;
4572 sizes
[0] = sizeof(struct btrfs_inode_item
);
4573 sizes
[1] = name_len
+ sizeof(*ref
);
4575 path
->leave_spinning
= 1;
4576 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4580 inode_init_owner(inode
, dir
, mode
);
4581 inode
->i_ino
= objectid
;
4582 inode_set_bytes(inode
, 0);
4583 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4584 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4585 struct btrfs_inode_item
);
4586 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4588 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4589 struct btrfs_inode_ref
);
4590 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4591 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4592 ptr
= (unsigned long)(ref
+ 1);
4593 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4595 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4596 btrfs_free_path(path
);
4598 location
= &BTRFS_I(inode
)->location
;
4599 location
->objectid
= objectid
;
4600 location
->offset
= 0;
4601 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4603 btrfs_inherit_iflags(inode
, dir
);
4605 if ((mode
& S_IFREG
)) {
4606 if (btrfs_test_opt(root
, NODATASUM
))
4607 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4608 if (btrfs_test_opt(root
, NODATACOW
) ||
4609 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4610 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4613 insert_inode_hash(inode
);
4614 inode_tree_add(inode
);
4616 trace_btrfs_inode_new(inode
);
4621 BTRFS_I(dir
)->index_cnt
--;
4622 btrfs_free_path(path
);
4624 return ERR_PTR(ret
);
4627 static inline u8
btrfs_inode_type(struct inode
*inode
)
4629 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4633 * utility function to add 'inode' into 'parent_inode' with
4634 * a give name and a given sequence number.
4635 * if 'add_backref' is true, also insert a backref from the
4636 * inode to the parent directory.
4638 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4639 struct inode
*parent_inode
, struct inode
*inode
,
4640 const char *name
, int name_len
, int add_backref
, u64 index
)
4643 struct btrfs_key key
;
4644 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4646 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4647 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4649 key
.objectid
= inode
->i_ino
;
4650 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4654 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4655 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4656 key
.objectid
, root
->root_key
.objectid
,
4657 parent_inode
->i_ino
,
4658 index
, name
, name_len
);
4659 } else if (add_backref
) {
4660 ret
= btrfs_insert_inode_ref(trans
, root
,
4661 name
, name_len
, inode
->i_ino
,
4662 parent_inode
->i_ino
, index
);
4666 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4667 parent_inode
->i_ino
, &key
,
4668 btrfs_inode_type(inode
), index
);
4671 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4673 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4674 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4679 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4680 struct inode
*dir
, struct dentry
*dentry
,
4681 struct inode
*inode
, int backref
, u64 index
)
4683 int err
= btrfs_add_link(trans
, dir
, inode
,
4684 dentry
->d_name
.name
, dentry
->d_name
.len
,
4687 d_instantiate(dentry
, inode
);
4695 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4696 int mode
, dev_t rdev
)
4698 struct btrfs_trans_handle
*trans
;
4699 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4700 struct inode
*inode
= NULL
;
4704 unsigned long nr
= 0;
4707 if (!new_valid_dev(rdev
))
4710 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4715 * 2 for inode item and ref
4717 * 1 for xattr if selinux is on
4719 trans
= btrfs_start_transaction(root
, 5);
4721 return PTR_ERR(trans
);
4723 btrfs_set_trans_block_group(trans
, dir
);
4725 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4726 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4727 BTRFS_I(dir
)->block_group
, mode
, &index
);
4728 err
= PTR_ERR(inode
);
4732 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4738 btrfs_set_trans_block_group(trans
, inode
);
4739 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4743 inode
->i_op
= &btrfs_special_inode_operations
;
4744 init_special_inode(inode
, inode
->i_mode
, rdev
);
4745 btrfs_update_inode(trans
, root
, inode
);
4747 btrfs_update_inode_block_group(trans
, inode
);
4748 btrfs_update_inode_block_group(trans
, dir
);
4750 nr
= trans
->blocks_used
;
4751 btrfs_end_transaction_throttle(trans
, root
);
4752 btrfs_btree_balance_dirty(root
, nr
);
4754 inode_dec_link_count(inode
);
4760 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4761 int mode
, struct nameidata
*nd
)
4763 struct btrfs_trans_handle
*trans
;
4764 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4765 struct inode
*inode
= NULL
;
4768 unsigned long nr
= 0;
4772 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4776 * 2 for inode item and ref
4778 * 1 for xattr if selinux is on
4780 trans
= btrfs_start_transaction(root
, 5);
4782 return PTR_ERR(trans
);
4784 btrfs_set_trans_block_group(trans
, dir
);
4786 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4787 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4788 BTRFS_I(dir
)->block_group
, mode
, &index
);
4789 err
= PTR_ERR(inode
);
4793 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4799 btrfs_set_trans_block_group(trans
, inode
);
4800 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4804 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4805 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4806 inode
->i_fop
= &btrfs_file_operations
;
4807 inode
->i_op
= &btrfs_file_inode_operations
;
4808 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4810 btrfs_update_inode_block_group(trans
, inode
);
4811 btrfs_update_inode_block_group(trans
, dir
);
4813 nr
= trans
->blocks_used
;
4814 btrfs_end_transaction_throttle(trans
, root
);
4816 inode_dec_link_count(inode
);
4819 btrfs_btree_balance_dirty(root
, nr
);
4823 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4824 struct dentry
*dentry
)
4826 struct btrfs_trans_handle
*trans
;
4827 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4828 struct inode
*inode
= old_dentry
->d_inode
;
4830 unsigned long nr
= 0;
4834 if (inode
->i_nlink
== 0)
4837 /* do not allow sys_link's with other subvols of the same device */
4838 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4841 if (inode
->i_nlink
== ~0U)
4844 btrfs_inc_nlink(inode
);
4845 inode
->i_ctime
= CURRENT_TIME
;
4847 err
= btrfs_set_inode_index(dir
, &index
);
4852 * 2 items for inode and inode ref
4853 * 2 items for dir items
4854 * 1 item for parent inode
4856 trans
= btrfs_start_transaction(root
, 5);
4857 if (IS_ERR(trans
)) {
4858 err
= PTR_ERR(trans
);
4862 btrfs_set_trans_block_group(trans
, dir
);
4865 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4870 struct dentry
*parent
= dget_parent(dentry
);
4871 btrfs_update_inode_block_group(trans
, dir
);
4872 err
= btrfs_update_inode(trans
, root
, inode
);
4874 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4878 nr
= trans
->blocks_used
;
4879 btrfs_end_transaction_throttle(trans
, root
);
4882 inode_dec_link_count(inode
);
4885 btrfs_btree_balance_dirty(root
, nr
);
4889 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4891 struct inode
*inode
= NULL
;
4892 struct btrfs_trans_handle
*trans
;
4893 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4895 int drop_on_err
= 0;
4898 unsigned long nr
= 1;
4900 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4905 * 2 items for inode and ref
4906 * 2 items for dir items
4907 * 1 for xattr if selinux is on
4909 trans
= btrfs_start_transaction(root
, 5);
4911 return PTR_ERR(trans
);
4912 btrfs_set_trans_block_group(trans
, dir
);
4914 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4915 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4916 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
4918 if (IS_ERR(inode
)) {
4919 err
= PTR_ERR(inode
);
4925 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4929 inode
->i_op
= &btrfs_dir_inode_operations
;
4930 inode
->i_fop
= &btrfs_dir_file_operations
;
4931 btrfs_set_trans_block_group(trans
, inode
);
4933 btrfs_i_size_write(inode
, 0);
4934 err
= btrfs_update_inode(trans
, root
, inode
);
4938 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4939 dentry
->d_name
.len
, 0, index
);
4943 d_instantiate(dentry
, inode
);
4945 btrfs_update_inode_block_group(trans
, inode
);
4946 btrfs_update_inode_block_group(trans
, dir
);
4949 nr
= trans
->blocks_used
;
4950 btrfs_end_transaction_throttle(trans
, root
);
4953 btrfs_btree_balance_dirty(root
, nr
);
4957 /* helper for btfs_get_extent. Given an existing extent in the tree,
4958 * and an extent that you want to insert, deal with overlap and insert
4959 * the new extent into the tree.
4961 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4962 struct extent_map
*existing
,
4963 struct extent_map
*em
,
4964 u64 map_start
, u64 map_len
)
4968 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4969 start_diff
= map_start
- em
->start
;
4970 em
->start
= map_start
;
4972 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4973 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4974 em
->block_start
+= start_diff
;
4975 em
->block_len
-= start_diff
;
4977 return add_extent_mapping(em_tree
, em
);
4980 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4981 struct inode
*inode
, struct page
*page
,
4982 size_t pg_offset
, u64 extent_offset
,
4983 struct btrfs_file_extent_item
*item
)
4986 struct extent_buffer
*leaf
= path
->nodes
[0];
4989 unsigned long inline_size
;
4993 WARN_ON(pg_offset
!= 0);
4994 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4995 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4996 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4997 btrfs_item_nr(leaf
, path
->slots
[0]));
4998 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4999 ptr
= btrfs_file_extent_inline_start(item
);
5001 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
5003 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
5004 ret
= btrfs_decompress(compress_type
, tmp
, page
,
5005 extent_offset
, inline_size
, max_size
);
5007 char *kaddr
= kmap_atomic(page
, KM_USER0
);
5008 unsigned long copy_size
= min_t(u64
,
5009 PAGE_CACHE_SIZE
- pg_offset
,
5010 max_size
- extent_offset
);
5011 memset(kaddr
+ pg_offset
, 0, copy_size
);
5012 kunmap_atomic(kaddr
, KM_USER0
);
5019 * a bit scary, this does extent mapping from logical file offset to the disk.
5020 * the ugly parts come from merging extents from the disk with the in-ram
5021 * representation. This gets more complex because of the data=ordered code,
5022 * where the in-ram extents might be locked pending data=ordered completion.
5024 * This also copies inline extents directly into the page.
5027 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
5028 size_t pg_offset
, u64 start
, u64 len
,
5034 u64 extent_start
= 0;
5036 u64 objectid
= inode
->i_ino
;
5038 struct btrfs_path
*path
= NULL
;
5039 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5040 struct btrfs_file_extent_item
*item
;
5041 struct extent_buffer
*leaf
;
5042 struct btrfs_key found_key
;
5043 struct extent_map
*em
= NULL
;
5044 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5045 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5046 struct btrfs_trans_handle
*trans
= NULL
;
5050 read_lock(&em_tree
->lock
);
5051 em
= lookup_extent_mapping(em_tree
, start
, len
);
5053 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5054 read_unlock(&em_tree
->lock
);
5057 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5058 free_extent_map(em
);
5059 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5060 free_extent_map(em
);
5064 em
= alloc_extent_map(GFP_NOFS
);
5069 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5070 em
->start
= EXTENT_MAP_HOLE
;
5071 em
->orig_start
= EXTENT_MAP_HOLE
;
5073 em
->block_len
= (u64
)-1;
5076 path
= btrfs_alloc_path();
5080 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5081 objectid
, start
, trans
!= NULL
);
5088 if (path
->slots
[0] == 0)
5093 leaf
= path
->nodes
[0];
5094 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5095 struct btrfs_file_extent_item
);
5096 /* are we inside the extent that was found? */
5097 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5098 found_type
= btrfs_key_type(&found_key
);
5099 if (found_key
.objectid
!= objectid
||
5100 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5104 found_type
= btrfs_file_extent_type(leaf
, item
);
5105 extent_start
= found_key
.offset
;
5106 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5107 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5108 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5109 extent_end
= extent_start
+
5110 btrfs_file_extent_num_bytes(leaf
, item
);
5111 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5113 size
= btrfs_file_extent_inline_len(leaf
, item
);
5114 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5115 ~((u64
)root
->sectorsize
- 1);
5118 if (start
>= extent_end
) {
5120 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5121 ret
= btrfs_next_leaf(root
, path
);
5128 leaf
= path
->nodes
[0];
5130 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5131 if (found_key
.objectid
!= objectid
||
5132 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5134 if (start
+ len
<= found_key
.offset
)
5137 em
->len
= found_key
.offset
- start
;
5141 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5142 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5143 em
->start
= extent_start
;
5144 em
->len
= extent_end
- extent_start
;
5145 em
->orig_start
= extent_start
-
5146 btrfs_file_extent_offset(leaf
, item
);
5147 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5149 em
->block_start
= EXTENT_MAP_HOLE
;
5152 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5153 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5154 em
->compress_type
= compress_type
;
5155 em
->block_start
= bytenr
;
5156 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5159 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5160 em
->block_start
= bytenr
;
5161 em
->block_len
= em
->len
;
5162 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5163 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5166 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5170 size_t extent_offset
;
5173 em
->block_start
= EXTENT_MAP_INLINE
;
5174 if (!page
|| create
) {
5175 em
->start
= extent_start
;
5176 em
->len
= extent_end
- extent_start
;
5180 size
= btrfs_file_extent_inline_len(leaf
, item
);
5181 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5182 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5183 size
- extent_offset
);
5184 em
->start
= extent_start
+ extent_offset
;
5185 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5186 ~((u64
)root
->sectorsize
- 1);
5187 em
->orig_start
= EXTENT_MAP_INLINE
;
5188 if (compress_type
) {
5189 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5190 em
->compress_type
= compress_type
;
5192 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5193 if (create
== 0 && !PageUptodate(page
)) {
5194 if (btrfs_file_extent_compression(leaf
, item
) !=
5195 BTRFS_COMPRESS_NONE
) {
5196 ret
= uncompress_inline(path
, inode
, page
,
5198 extent_offset
, item
);
5202 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5204 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5205 memset(map
+ pg_offset
+ copy_size
, 0,
5206 PAGE_CACHE_SIZE
- pg_offset
-
5211 flush_dcache_page(page
);
5212 } else if (create
&& PageUptodate(page
)) {
5216 free_extent_map(em
);
5218 btrfs_release_path(root
, path
);
5219 trans
= btrfs_join_transaction(root
, 1);
5221 return ERR_CAST(trans
);
5225 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5228 btrfs_mark_buffer_dirty(leaf
);
5230 set_extent_uptodate(io_tree
, em
->start
,
5231 extent_map_end(em
) - 1, GFP_NOFS
);
5234 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5241 em
->block_start
= EXTENT_MAP_HOLE
;
5242 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5244 btrfs_release_path(root
, path
);
5245 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5246 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5247 "[%llu %llu]\n", (unsigned long long)em
->start
,
5248 (unsigned long long)em
->len
,
5249 (unsigned long long)start
,
5250 (unsigned long long)len
);
5256 write_lock(&em_tree
->lock
);
5257 ret
= add_extent_mapping(em_tree
, em
);
5258 /* it is possible that someone inserted the extent into the tree
5259 * while we had the lock dropped. It is also possible that
5260 * an overlapping map exists in the tree
5262 if (ret
== -EEXIST
) {
5263 struct extent_map
*existing
;
5267 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5268 if (existing
&& (existing
->start
> start
||
5269 existing
->start
+ existing
->len
<= start
)) {
5270 free_extent_map(existing
);
5274 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5277 err
= merge_extent_mapping(em_tree
, existing
,
5280 free_extent_map(existing
);
5282 free_extent_map(em
);
5287 free_extent_map(em
);
5291 free_extent_map(em
);
5296 write_unlock(&em_tree
->lock
);
5299 trace_btrfs_get_extent(root
, em
);
5302 btrfs_free_path(path
);
5304 ret
= btrfs_end_transaction(trans
, root
);
5309 free_extent_map(em
);
5310 return ERR_PTR(err
);
5315 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5316 size_t pg_offset
, u64 start
, u64 len
,
5319 struct extent_map
*em
;
5320 struct extent_map
*hole_em
= NULL
;
5321 u64 range_start
= start
;
5327 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5332 * if our em maps to a hole, there might
5333 * actually be delalloc bytes behind it
5335 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5341 /* check to see if we've wrapped (len == -1 or similar) */
5350 /* ok, we didn't find anything, lets look for delalloc */
5351 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5352 end
, len
, EXTENT_DELALLOC
, 1);
5353 found_end
= range_start
+ found
;
5354 if (found_end
< range_start
)
5355 found_end
= (u64
)-1;
5358 * we didn't find anything useful, return
5359 * the original results from get_extent()
5361 if (range_start
> end
|| found_end
<= start
) {
5367 /* adjust the range_start to make sure it doesn't
5368 * go backwards from the start they passed in
5370 range_start
= max(start
,range_start
);
5371 found
= found_end
- range_start
;
5374 u64 hole_start
= start
;
5377 em
= alloc_extent_map(GFP_NOFS
);
5383 * when btrfs_get_extent can't find anything it
5384 * returns one huge hole
5386 * make sure what it found really fits our range, and
5387 * adjust to make sure it is based on the start from
5391 u64 calc_end
= extent_map_end(hole_em
);
5393 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5394 free_extent_map(hole_em
);
5397 hole_start
= max(hole_em
->start
, start
);
5398 hole_len
= calc_end
- hole_start
;
5402 if (hole_em
&& range_start
> hole_start
) {
5403 /* our hole starts before our delalloc, so we
5404 * have to return just the parts of the hole
5405 * that go until the delalloc starts
5407 em
->len
= min(hole_len
,
5408 range_start
- hole_start
);
5409 em
->start
= hole_start
;
5410 em
->orig_start
= hole_start
;
5412 * don't adjust block start at all,
5413 * it is fixed at EXTENT_MAP_HOLE
5415 em
->block_start
= hole_em
->block_start
;
5416 em
->block_len
= hole_len
;
5418 em
->start
= range_start
;
5420 em
->orig_start
= range_start
;
5421 em
->block_start
= EXTENT_MAP_DELALLOC
;
5422 em
->block_len
= found
;
5424 } else if (hole_em
) {
5429 free_extent_map(hole_em
);
5431 free_extent_map(em
);
5432 return ERR_PTR(err
);
5437 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5440 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5441 struct btrfs_trans_handle
*trans
;
5442 struct extent_map
*em
;
5443 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5444 struct btrfs_key ins
;
5448 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5450 trans
= btrfs_join_transaction(root
, 0);
5452 return ERR_CAST(trans
);
5454 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5456 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5457 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5458 alloc_hint
, (u64
)-1, &ins
, 1);
5464 em
= alloc_extent_map(GFP_NOFS
);
5466 em
= ERR_PTR(-ENOMEM
);
5471 em
->orig_start
= em
->start
;
5472 em
->len
= ins
.offset
;
5474 em
->block_start
= ins
.objectid
;
5475 em
->block_len
= ins
.offset
;
5476 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5477 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5480 write_lock(&em_tree
->lock
);
5481 ret
= add_extent_mapping(em_tree
, em
);
5482 write_unlock(&em_tree
->lock
);
5485 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5488 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5489 ins
.offset
, ins
.offset
, 0);
5491 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5495 btrfs_end_transaction(trans
, root
);
5500 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5501 * block must be cow'd
5503 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5504 struct inode
*inode
, u64 offset
, u64 len
)
5506 struct btrfs_path
*path
;
5508 struct extent_buffer
*leaf
;
5509 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5510 struct btrfs_file_extent_item
*fi
;
5511 struct btrfs_key key
;
5519 path
= btrfs_alloc_path();
5523 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
5528 slot
= path
->slots
[0];
5531 /* can't find the item, must cow */
5538 leaf
= path
->nodes
[0];
5539 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5540 if (key
.objectid
!= inode
->i_ino
||
5541 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5542 /* not our file or wrong item type, must cow */
5546 if (key
.offset
> offset
) {
5547 /* Wrong offset, must cow */
5551 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5552 found_type
= btrfs_file_extent_type(leaf
, fi
);
5553 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5554 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5555 /* not a regular extent, must cow */
5558 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5559 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5561 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5562 if (extent_end
< offset
+ len
) {
5563 /* extent doesn't include our full range, must cow */
5567 if (btrfs_extent_readonly(root
, disk_bytenr
))
5571 * look for other files referencing this extent, if we
5572 * find any we must cow
5574 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
5575 key
.offset
- backref_offset
, disk_bytenr
))
5579 * adjust disk_bytenr and num_bytes to cover just the bytes
5580 * in this extent we are about to write. If there
5581 * are any csums in that range we have to cow in order
5582 * to keep the csums correct
5584 disk_bytenr
+= backref_offset
;
5585 disk_bytenr
+= offset
- key
.offset
;
5586 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5587 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5590 * all of the above have passed, it is safe to overwrite this extent
5595 btrfs_free_path(path
);
5599 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5600 struct buffer_head
*bh_result
, int create
)
5602 struct extent_map
*em
;
5603 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5604 u64 start
= iblock
<< inode
->i_blkbits
;
5605 u64 len
= bh_result
->b_size
;
5606 struct btrfs_trans_handle
*trans
;
5608 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5613 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5614 * io. INLINE is special, and we could probably kludge it in here, but
5615 * it's still buffered so for safety lets just fall back to the generic
5618 * For COMPRESSED we _have_ to read the entire extent in so we can
5619 * decompress it, so there will be buffering required no matter what we
5620 * do, so go ahead and fallback to buffered.
5622 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5623 * to buffered IO. Don't blame me, this is the price we pay for using
5626 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5627 em
->block_start
== EXTENT_MAP_INLINE
) {
5628 free_extent_map(em
);
5632 /* Just a good old fashioned hole, return */
5633 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5634 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5635 free_extent_map(em
);
5636 /* DIO will do one hole at a time, so just unlock a sector */
5637 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5638 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5643 * We don't allocate a new extent in the following cases
5645 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5647 * 2) The extent is marked as PREALLOC. We're good to go here and can
5648 * just use the extent.
5652 len
= em
->len
- (start
- em
->start
);
5656 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5657 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5658 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5663 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5664 type
= BTRFS_ORDERED_PREALLOC
;
5666 type
= BTRFS_ORDERED_NOCOW
;
5667 len
= min(len
, em
->len
- (start
- em
->start
));
5668 block_start
= em
->block_start
+ (start
- em
->start
);
5671 * we're not going to log anything, but we do need
5672 * to make sure the current transaction stays open
5673 * while we look for nocow cross refs
5675 trans
= btrfs_join_transaction(root
, 0);
5679 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5680 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5681 block_start
, len
, len
, type
);
5682 btrfs_end_transaction(trans
, root
);
5684 free_extent_map(em
);
5689 btrfs_end_transaction(trans
, root
);
5693 * this will cow the extent, reset the len in case we changed
5696 len
= bh_result
->b_size
;
5697 free_extent_map(em
);
5698 em
= btrfs_new_extent_direct(inode
, start
, len
);
5701 len
= min(len
, em
->len
- (start
- em
->start
));
5703 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5704 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5707 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5709 bh_result
->b_size
= len
;
5710 bh_result
->b_bdev
= em
->bdev
;
5711 set_buffer_mapped(bh_result
);
5712 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5713 set_buffer_new(bh_result
);
5715 free_extent_map(em
);
5720 struct btrfs_dio_private
{
5721 struct inode
*inode
;
5728 /* number of bios pending for this dio */
5729 atomic_t pending_bios
;
5734 struct bio
*orig_bio
;
5737 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5739 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5740 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5741 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5742 struct inode
*inode
= dip
->inode
;
5743 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5745 u32
*private = dip
->csums
;
5747 start
= dip
->logical_offset
;
5749 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5750 struct page
*page
= bvec
->bv_page
;
5753 unsigned long flags
;
5755 local_irq_save(flags
);
5756 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5757 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5758 csum
, bvec
->bv_len
);
5759 btrfs_csum_final(csum
, (char *)&csum
);
5760 kunmap_atomic(kaddr
, KM_IRQ0
);
5761 local_irq_restore(flags
);
5763 flush_dcache_page(bvec
->bv_page
);
5764 if (csum
!= *private) {
5765 printk(KERN_ERR
"btrfs csum failed ino %lu off"
5766 " %llu csum %u private %u\n",
5767 inode
->i_ino
, (unsigned long long)start
,
5773 start
+= bvec
->bv_len
;
5776 } while (bvec
<= bvec_end
);
5778 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5779 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5780 bio
->bi_private
= dip
->private;
5785 /* If we had a csum failure make sure to clear the uptodate flag */
5787 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5788 dio_end_io(bio
, err
);
5791 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5793 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5794 struct inode
*inode
= dip
->inode
;
5795 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5796 struct btrfs_trans_handle
*trans
;
5797 struct btrfs_ordered_extent
*ordered
= NULL
;
5798 struct extent_state
*cached_state
= NULL
;
5799 u64 ordered_offset
= dip
->logical_offset
;
5800 u64 ordered_bytes
= dip
->bytes
;
5806 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5814 trans
= btrfs_join_transaction(root
, 1);
5815 if (IS_ERR(trans
)) {
5819 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5821 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5822 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5824 ret
= btrfs_update_inode(trans
, root
, inode
);
5829 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5830 ordered
->file_offset
+ ordered
->len
- 1, 0,
5831 &cached_state
, GFP_NOFS
);
5833 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5834 ret
= btrfs_mark_extent_written(trans
, inode
,
5835 ordered
->file_offset
,
5836 ordered
->file_offset
+
5843 ret
= insert_reserved_file_extent(trans
, inode
,
5844 ordered
->file_offset
,
5850 BTRFS_FILE_EXTENT_REG
);
5851 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5852 ordered
->file_offset
, ordered
->len
);
5860 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5861 btrfs_ordered_update_i_size(inode
, 0, ordered
);
5862 btrfs_update_inode(trans
, root
, inode
);
5864 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5865 ordered
->file_offset
+ ordered
->len
- 1,
5866 &cached_state
, GFP_NOFS
);
5868 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5869 btrfs_end_transaction(trans
, root
);
5870 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5871 btrfs_put_ordered_extent(ordered
);
5872 btrfs_put_ordered_extent(ordered
);
5876 * our bio might span multiple ordered extents. If we haven't
5877 * completed the accounting for the whole dio, go back and try again
5879 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5880 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5885 bio
->bi_private
= dip
->private;
5890 /* If we had an error make sure to clear the uptodate flag */
5892 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5893 dio_end_io(bio
, err
);
5896 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5897 struct bio
*bio
, int mirror_num
,
5898 unsigned long bio_flags
, u64 offset
)
5901 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5902 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5907 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5909 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5912 printk(KERN_ERR
"btrfs direct IO failed ino %lu rw %lu "
5913 "sector %#Lx len %u err no %d\n",
5914 dip
->inode
->i_ino
, bio
->bi_rw
,
5915 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5919 * before atomic variable goto zero, we must make sure
5920 * dip->errors is perceived to be set.
5922 smp_mb__before_atomic_dec();
5925 /* if there are more bios still pending for this dio, just exit */
5926 if (!atomic_dec_and_test(&dip
->pending_bios
))
5930 bio_io_error(dip
->orig_bio
);
5932 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5933 bio_endio(dip
->orig_bio
, 0);
5939 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5940 u64 first_sector
, gfp_t gfp_flags
)
5942 int nr_vecs
= bio_get_nr_vecs(bdev
);
5943 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5946 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5947 int rw
, u64 file_offset
, int skip_sum
,
5950 int write
= rw
& REQ_WRITE
;
5951 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5955 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5959 if (write
&& !skip_sum
) {
5960 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5961 inode
, rw
, bio
, 0, 0,
5963 __btrfs_submit_bio_start_direct_io
,
5964 __btrfs_submit_bio_done
);
5966 } else if (!skip_sum
) {
5967 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5968 file_offset
, csums
);
5973 ret
= btrfs_map_bio(root
, rw
, bio
, 0, 1);
5979 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5982 struct inode
*inode
= dip
->inode
;
5983 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5984 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5986 struct bio
*orig_bio
= dip
->orig_bio
;
5987 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5988 u64 start_sector
= orig_bio
->bi_sector
;
5989 u64 file_offset
= dip
->logical_offset
;
5993 u32
*csums
= dip
->csums
;
5995 int write
= rw
& REQ_WRITE
;
5997 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
6000 bio
->bi_private
= dip
;
6001 bio
->bi_end_io
= btrfs_end_dio_bio
;
6002 atomic_inc(&dip
->pending_bios
);
6004 map_length
= orig_bio
->bi_size
;
6005 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6006 &map_length
, NULL
, 0);
6012 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
6013 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
6014 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
6015 bvec
->bv_offset
) < bvec
->bv_len
)) {
6017 * inc the count before we submit the bio so
6018 * we know the end IO handler won't happen before
6019 * we inc the count. Otherwise, the dip might get freed
6020 * before we're done setting it up
6022 atomic_inc(&dip
->pending_bios
);
6023 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
6024 file_offset
, skip_sum
,
6028 atomic_dec(&dip
->pending_bios
);
6032 /* Write's use the ordered csums */
6033 if (!write
&& !skip_sum
)
6034 csums
= csums
+ nr_pages
;
6035 start_sector
+= submit_len
>> 9;
6036 file_offset
+= submit_len
;
6041 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6042 start_sector
, GFP_NOFS
);
6045 bio
->bi_private
= dip
;
6046 bio
->bi_end_io
= btrfs_end_dio_bio
;
6048 map_length
= orig_bio
->bi_size
;
6049 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6050 &map_length
, NULL
, 0);
6056 submit_len
+= bvec
->bv_len
;
6062 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6071 * before atomic variable goto zero, we must
6072 * make sure dip->errors is perceived to be set.
6074 smp_mb__before_atomic_dec();
6075 if (atomic_dec_and_test(&dip
->pending_bios
))
6076 bio_io_error(dip
->orig_bio
);
6078 /* bio_end_io() will handle error, so we needn't return it */
6082 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6085 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6086 struct btrfs_dio_private
*dip
;
6087 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6089 int write
= rw
& REQ_WRITE
;
6092 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6094 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6101 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6102 if (!write
&& !skip_sum
) {
6103 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6111 dip
->private = bio
->bi_private
;
6113 dip
->logical_offset
= file_offset
;
6117 dip
->bytes
+= bvec
->bv_len
;
6119 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6121 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6122 bio
->bi_private
= dip
;
6124 dip
->orig_bio
= bio
;
6125 atomic_set(&dip
->pending_bios
, 0);
6128 bio
->bi_end_io
= btrfs_endio_direct_write
;
6130 bio
->bi_end_io
= btrfs_endio_direct_read
;
6132 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6137 * If this is a write, we need to clean up the reserved space and kill
6138 * the ordered extent.
6141 struct btrfs_ordered_extent
*ordered
;
6142 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6143 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6144 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6145 btrfs_free_reserved_extent(root
, ordered
->start
,
6147 btrfs_put_ordered_extent(ordered
);
6148 btrfs_put_ordered_extent(ordered
);
6150 bio_endio(bio
, ret
);
6153 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6154 const struct iovec
*iov
, loff_t offset
,
6155 unsigned long nr_segs
)
6160 unsigned blocksize_mask
= root
->sectorsize
- 1;
6161 ssize_t retval
= -EINVAL
;
6162 loff_t end
= offset
;
6164 if (offset
& blocksize_mask
)
6167 /* Check the memory alignment. Blocks cannot straddle pages */
6168 for (seg
= 0; seg
< nr_segs
; seg
++) {
6169 addr
= (unsigned long)iov
[seg
].iov_base
;
6170 size
= iov
[seg
].iov_len
;
6172 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6179 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6180 const struct iovec
*iov
, loff_t offset
,
6181 unsigned long nr_segs
)
6183 struct file
*file
= iocb
->ki_filp
;
6184 struct inode
*inode
= file
->f_mapping
->host
;
6185 struct btrfs_ordered_extent
*ordered
;
6186 struct extent_state
*cached_state
= NULL
;
6187 u64 lockstart
, lockend
;
6189 int writing
= rw
& WRITE
;
6191 size_t count
= iov_length(iov
, nr_segs
);
6193 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6199 lockend
= offset
+ count
- 1;
6202 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6208 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6209 0, &cached_state
, GFP_NOFS
);
6211 * We're concerned with the entire range that we're going to be
6212 * doing DIO to, so we need to make sure theres no ordered
6213 * extents in this range.
6215 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6216 lockend
- lockstart
+ 1);
6219 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6220 &cached_state
, GFP_NOFS
);
6221 btrfs_start_ordered_extent(inode
, ordered
, 1);
6222 btrfs_put_ordered_extent(ordered
);
6227 * we don't use btrfs_set_extent_delalloc because we don't want
6228 * the dirty or uptodate bits
6231 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6232 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6233 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6236 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6237 lockend
, EXTENT_LOCKED
| write_bits
,
6238 1, 0, &cached_state
, GFP_NOFS
);
6243 free_extent_state(cached_state
);
6244 cached_state
= NULL
;
6246 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6247 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6248 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6249 btrfs_submit_direct
, 0);
6251 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6252 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6253 offset
+ iov_length(iov
, nr_segs
) - 1,
6254 EXTENT_LOCKED
| write_bits
, 1, 0,
6255 &cached_state
, GFP_NOFS
);
6256 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6258 * We're falling back to buffered, unlock the section we didn't
6261 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6262 offset
+ iov_length(iov
, nr_segs
) - 1,
6263 EXTENT_LOCKED
| write_bits
, 1, 0,
6264 &cached_state
, GFP_NOFS
);
6267 free_extent_state(cached_state
);
6271 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6272 __u64 start
, __u64 len
)
6274 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6277 int btrfs_readpage(struct file
*file
, struct page
*page
)
6279 struct extent_io_tree
*tree
;
6280 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6281 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6284 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6286 struct extent_io_tree
*tree
;
6289 if (current
->flags
& PF_MEMALLOC
) {
6290 redirty_page_for_writepage(wbc
, page
);
6294 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6295 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6298 int btrfs_writepages(struct address_space
*mapping
,
6299 struct writeback_control
*wbc
)
6301 struct extent_io_tree
*tree
;
6303 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6304 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6308 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6309 struct list_head
*pages
, unsigned nr_pages
)
6311 struct extent_io_tree
*tree
;
6312 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6313 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6316 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6318 struct extent_io_tree
*tree
;
6319 struct extent_map_tree
*map
;
6322 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6323 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6324 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6326 ClearPagePrivate(page
);
6327 set_page_private(page
, 0);
6328 page_cache_release(page
);
6333 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6335 if (PageWriteback(page
) || PageDirty(page
))
6337 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6340 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6342 struct extent_io_tree
*tree
;
6343 struct btrfs_ordered_extent
*ordered
;
6344 struct extent_state
*cached_state
= NULL
;
6345 u64 page_start
= page_offset(page
);
6346 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6350 * we have the page locked, so new writeback can't start,
6351 * and the dirty bit won't be cleared while we are here.
6353 * Wait for IO on this page so that we can safely clear
6354 * the PagePrivate2 bit and do ordered accounting
6356 wait_on_page_writeback(page
);
6358 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6360 btrfs_releasepage(page
, GFP_NOFS
);
6363 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6365 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6369 * IO on this page will never be started, so we need
6370 * to account for any ordered extents now
6372 clear_extent_bit(tree
, page_start
, page_end
,
6373 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6374 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6375 &cached_state
, GFP_NOFS
);
6377 * whoever cleared the private bit is responsible
6378 * for the finish_ordered_io
6380 if (TestClearPagePrivate2(page
)) {
6381 btrfs_finish_ordered_io(page
->mapping
->host
,
6382 page_start
, page_end
);
6384 btrfs_put_ordered_extent(ordered
);
6385 cached_state
= NULL
;
6386 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6389 clear_extent_bit(tree
, page_start
, page_end
,
6390 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6391 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6392 __btrfs_releasepage(page
, GFP_NOFS
);
6394 ClearPageChecked(page
);
6395 if (PagePrivate(page
)) {
6396 ClearPagePrivate(page
);
6397 set_page_private(page
, 0);
6398 page_cache_release(page
);
6403 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6404 * called from a page fault handler when a page is first dirtied. Hence we must
6405 * be careful to check for EOF conditions here. We set the page up correctly
6406 * for a written page which means we get ENOSPC checking when writing into
6407 * holes and correct delalloc and unwritten extent mapping on filesystems that
6408 * support these features.
6410 * We are not allowed to take the i_mutex here so we have to play games to
6411 * protect against truncate races as the page could now be beyond EOF. Because
6412 * vmtruncate() writes the inode size before removing pages, once we have the
6413 * page lock we can determine safely if the page is beyond EOF. If it is not
6414 * beyond EOF, then the page is guaranteed safe against truncation until we
6417 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6419 struct page
*page
= vmf
->page
;
6420 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6421 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6422 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6423 struct btrfs_ordered_extent
*ordered
;
6424 struct extent_state
*cached_state
= NULL
;
6426 unsigned long zero_start
;
6432 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6436 else /* -ENOSPC, -EIO, etc */
6437 ret
= VM_FAULT_SIGBUS
;
6441 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6444 size
= i_size_read(inode
);
6445 page_start
= page_offset(page
);
6446 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6448 if ((page
->mapping
!= inode
->i_mapping
) ||
6449 (page_start
>= size
)) {
6450 /* page got truncated out from underneath us */
6453 wait_on_page_writeback(page
);
6455 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6457 set_page_extent_mapped(page
);
6460 * we can't set the delalloc bits if there are pending ordered
6461 * extents. Drop our locks and wait for them to finish
6463 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6465 unlock_extent_cached(io_tree
, page_start
, page_end
,
6466 &cached_state
, GFP_NOFS
);
6468 btrfs_start_ordered_extent(inode
, ordered
, 1);
6469 btrfs_put_ordered_extent(ordered
);
6474 * XXX - page_mkwrite gets called every time the page is dirtied, even
6475 * if it was already dirty, so for space accounting reasons we need to
6476 * clear any delalloc bits for the range we are fixing to save. There
6477 * is probably a better way to do this, but for now keep consistent with
6478 * prepare_pages in the normal write path.
6480 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6481 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6482 0, 0, &cached_state
, GFP_NOFS
);
6484 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6487 unlock_extent_cached(io_tree
, page_start
, page_end
,
6488 &cached_state
, GFP_NOFS
);
6489 ret
= VM_FAULT_SIGBUS
;
6494 /* page is wholly or partially inside EOF */
6495 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6496 zero_start
= size
& ~PAGE_CACHE_MASK
;
6498 zero_start
= PAGE_CACHE_SIZE
;
6500 if (zero_start
!= PAGE_CACHE_SIZE
) {
6502 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6503 flush_dcache_page(page
);
6506 ClearPageChecked(page
);
6507 set_page_dirty(page
);
6508 SetPageUptodate(page
);
6510 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6511 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6513 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6517 return VM_FAULT_LOCKED
;
6519 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6524 static int btrfs_truncate(struct inode
*inode
)
6526 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6529 struct btrfs_trans_handle
*trans
;
6531 u64 mask
= root
->sectorsize
- 1;
6533 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6537 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6538 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6540 trans
= btrfs_start_transaction(root
, 5);
6542 return PTR_ERR(trans
);
6544 btrfs_set_trans_block_group(trans
, inode
);
6546 ret
= btrfs_orphan_add(trans
, inode
);
6548 btrfs_end_transaction(trans
, root
);
6552 nr
= trans
->blocks_used
;
6553 btrfs_end_transaction(trans
, root
);
6554 btrfs_btree_balance_dirty(root
, nr
);
6556 /* Now start a transaction for the truncate */
6557 trans
= btrfs_start_transaction(root
, 0);
6559 return PTR_ERR(trans
);
6560 btrfs_set_trans_block_group(trans
, inode
);
6561 trans
->block_rsv
= root
->orphan_block_rsv
;
6564 * setattr is responsible for setting the ordered_data_close flag,
6565 * but that is only tested during the last file release. That
6566 * could happen well after the next commit, leaving a great big
6567 * window where new writes may get lost if someone chooses to write
6568 * to this file after truncating to zero
6570 * The inode doesn't have any dirty data here, and so if we commit
6571 * this is a noop. If someone immediately starts writing to the inode
6572 * it is very likely we'll catch some of their writes in this
6573 * transaction, and the commit will find this file on the ordered
6574 * data list with good things to send down.
6576 * This is a best effort solution, there is still a window where
6577 * using truncate to replace the contents of the file will
6578 * end up with a zero length file after a crash.
6580 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6581 btrfs_add_ordered_operation(trans
, root
, inode
);
6585 trans
= btrfs_start_transaction(root
, 0);
6587 return PTR_ERR(trans
);
6588 btrfs_set_trans_block_group(trans
, inode
);
6589 trans
->block_rsv
= root
->orphan_block_rsv
;
6592 ret
= btrfs_block_rsv_check(trans
, root
,
6593 root
->orphan_block_rsv
, 0, 5);
6594 if (ret
== -EAGAIN
) {
6595 ret
= btrfs_commit_transaction(trans
, root
);
6605 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6607 BTRFS_EXTENT_DATA_KEY
);
6608 if (ret
!= -EAGAIN
) {
6613 ret
= btrfs_update_inode(trans
, root
, inode
);
6619 nr
= trans
->blocks_used
;
6620 btrfs_end_transaction(trans
, root
);
6622 btrfs_btree_balance_dirty(root
, nr
);
6625 if (ret
== 0 && inode
->i_nlink
> 0) {
6626 ret
= btrfs_orphan_del(trans
, inode
);
6629 } else if (ret
&& inode
->i_nlink
> 0) {
6631 * Failed to do the truncate, remove us from the in memory
6634 ret
= btrfs_orphan_del(NULL
, inode
);
6637 ret
= btrfs_update_inode(trans
, root
, inode
);
6641 nr
= trans
->blocks_used
;
6642 ret
= btrfs_end_transaction_throttle(trans
, root
);
6645 btrfs_btree_balance_dirty(root
, nr
);
6651 * create a new subvolume directory/inode (helper for the ioctl).
6653 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6654 struct btrfs_root
*new_root
,
6655 u64 new_dirid
, u64 alloc_hint
)
6657 struct inode
*inode
;
6661 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6662 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
6664 return PTR_ERR(inode
);
6665 inode
->i_op
= &btrfs_dir_inode_operations
;
6666 inode
->i_fop
= &btrfs_dir_file_operations
;
6669 btrfs_i_size_write(inode
, 0);
6671 err
= btrfs_update_inode(trans
, new_root
, inode
);
6678 /* helper function for file defrag and space balancing. This
6679 * forces readahead on a given range of bytes in an inode
6681 unsigned long btrfs_force_ra(struct address_space
*mapping
,
6682 struct file_ra_state
*ra
, struct file
*file
,
6683 pgoff_t offset
, pgoff_t last_index
)
6685 pgoff_t req_size
= last_index
- offset
+ 1;
6687 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
6688 return offset
+ req_size
;
6691 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6693 struct btrfs_inode
*ei
;
6694 struct inode
*inode
;
6696 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6701 ei
->space_info
= NULL
;
6705 ei
->last_sub_trans
= 0;
6706 ei
->logged_trans
= 0;
6707 ei
->delalloc_bytes
= 0;
6708 ei
->reserved_bytes
= 0;
6709 ei
->disk_i_size
= 0;
6711 ei
->index_cnt
= (u64
)-1;
6712 ei
->last_unlink_trans
= 0;
6714 atomic_set(&ei
->outstanding_extents
, 0);
6715 atomic_set(&ei
->reserved_extents
, 0);
6717 ei
->ordered_data_close
= 0;
6718 ei
->orphan_meta_reserved
= 0;
6719 ei
->dummy_inode
= 0;
6720 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6722 inode
= &ei
->vfs_inode
;
6723 extent_map_tree_init(&ei
->extent_tree
, GFP_NOFS
);
6724 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
, GFP_NOFS
);
6725 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
, GFP_NOFS
);
6726 mutex_init(&ei
->log_mutex
);
6727 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6728 INIT_LIST_HEAD(&ei
->i_orphan
);
6729 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6730 INIT_LIST_HEAD(&ei
->ordered_operations
);
6731 RB_CLEAR_NODE(&ei
->rb_node
);
6736 static void btrfs_i_callback(struct rcu_head
*head
)
6738 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6739 INIT_LIST_HEAD(&inode
->i_dentry
);
6740 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6743 void btrfs_destroy_inode(struct inode
*inode
)
6745 struct btrfs_ordered_extent
*ordered
;
6746 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6748 WARN_ON(!list_empty(&inode
->i_dentry
));
6749 WARN_ON(inode
->i_data
.nrpages
);
6750 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
));
6751 WARN_ON(atomic_read(&BTRFS_I(inode
)->reserved_extents
));
6754 * This can happen where we create an inode, but somebody else also
6755 * created the same inode and we need to destroy the one we already
6762 * Make sure we're properly removed from the ordered operation
6766 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6767 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6768 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6769 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6772 if (root
== root
->fs_info
->tree_root
) {
6773 struct btrfs_block_group_cache
*block_group
;
6775 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6776 BTRFS_I(inode
)->block_group
);
6777 if (block_group
&& block_group
->inode
== inode
) {
6778 spin_lock(&block_group
->lock
);
6779 block_group
->inode
= NULL
;
6780 spin_unlock(&block_group
->lock
);
6781 btrfs_put_block_group(block_group
);
6782 } else if (block_group
) {
6783 btrfs_put_block_group(block_group
);
6787 spin_lock(&root
->orphan_lock
);
6788 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6789 printk(KERN_INFO
"BTRFS: inode %lu still on the orphan list\n",
6791 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6793 spin_unlock(&root
->orphan_lock
);
6796 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6800 printk(KERN_ERR
"btrfs found ordered "
6801 "extent %llu %llu on inode cleanup\n",
6802 (unsigned long long)ordered
->file_offset
,
6803 (unsigned long long)ordered
->len
);
6804 btrfs_remove_ordered_extent(inode
, ordered
);
6805 btrfs_put_ordered_extent(ordered
);
6806 btrfs_put_ordered_extent(ordered
);
6809 inode_tree_del(inode
);
6810 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6812 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6815 int btrfs_drop_inode(struct inode
*inode
)
6817 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6819 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6820 root
!= root
->fs_info
->tree_root
)
6823 return generic_drop_inode(inode
);
6826 static void init_once(void *foo
)
6828 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6830 inode_init_once(&ei
->vfs_inode
);
6833 void btrfs_destroy_cachep(void)
6835 if (btrfs_inode_cachep
)
6836 kmem_cache_destroy(btrfs_inode_cachep
);
6837 if (btrfs_trans_handle_cachep
)
6838 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6839 if (btrfs_transaction_cachep
)
6840 kmem_cache_destroy(btrfs_transaction_cachep
);
6841 if (btrfs_path_cachep
)
6842 kmem_cache_destroy(btrfs_path_cachep
);
6843 if (btrfs_free_space_cachep
)
6844 kmem_cache_destroy(btrfs_free_space_cachep
);
6847 int btrfs_init_cachep(void)
6849 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6850 sizeof(struct btrfs_inode
), 0,
6851 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6852 if (!btrfs_inode_cachep
)
6855 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6856 sizeof(struct btrfs_trans_handle
), 0,
6857 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6858 if (!btrfs_trans_handle_cachep
)
6861 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6862 sizeof(struct btrfs_transaction
), 0,
6863 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6864 if (!btrfs_transaction_cachep
)
6867 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6868 sizeof(struct btrfs_path
), 0,
6869 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6870 if (!btrfs_path_cachep
)
6873 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6874 sizeof(struct btrfs_free_space
), 0,
6875 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6876 if (!btrfs_free_space_cachep
)
6881 btrfs_destroy_cachep();
6885 static int btrfs_getattr(struct vfsmount
*mnt
,
6886 struct dentry
*dentry
, struct kstat
*stat
)
6888 struct inode
*inode
= dentry
->d_inode
;
6889 generic_fillattr(inode
, stat
);
6890 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
6891 stat
->blksize
= PAGE_CACHE_SIZE
;
6892 stat
->blocks
= (inode_get_bytes(inode
) +
6893 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
6898 * If a file is moved, it will inherit the cow and compression flags of the new
6901 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
6903 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
6904 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
6906 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
6907 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
6909 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
6911 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
)
6912 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
6914 b_inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6917 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6918 struct inode
*new_dir
, struct dentry
*new_dentry
)
6920 struct btrfs_trans_handle
*trans
;
6921 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6922 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6923 struct inode
*new_inode
= new_dentry
->d_inode
;
6924 struct inode
*old_inode
= old_dentry
->d_inode
;
6925 struct timespec ctime
= CURRENT_TIME
;
6930 if (new_dir
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
6933 /* we only allow rename subvolume link between subvolumes */
6934 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
6937 if (old_inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
6938 (new_inode
&& new_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
))
6941 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
6942 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
6945 * we're using rename to replace one file with another.
6946 * and the replacement file is large. Start IO on it now so
6947 * we don't add too much work to the end of the transaction
6949 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
6950 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
6951 filemap_flush(old_inode
->i_mapping
);
6953 /* close the racy window with snapshot create/destroy ioctl */
6954 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
6955 down_read(&root
->fs_info
->subvol_sem
);
6957 * We want to reserve the absolute worst case amount of items. So if
6958 * both inodes are subvols and we need to unlink them then that would
6959 * require 4 item modifications, but if they are both normal inodes it
6960 * would require 5 item modifications, so we'll assume their normal
6961 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6962 * should cover the worst case number of items we'll modify.
6964 trans
= btrfs_start_transaction(root
, 20);
6966 return PTR_ERR(trans
);
6968 btrfs_set_trans_block_group(trans
, new_dir
);
6971 btrfs_record_root_in_trans(trans
, dest
);
6973 ret
= btrfs_set_inode_index(new_dir
, &index
);
6977 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6978 /* force full log commit if subvolume involved. */
6979 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
6981 ret
= btrfs_insert_inode_ref(trans
, dest
,
6982 new_dentry
->d_name
.name
,
6983 new_dentry
->d_name
.len
,
6985 new_dir
->i_ino
, index
);
6989 * this is an ugly little race, but the rename is required
6990 * to make sure that if we crash, the inode is either at the
6991 * old name or the new one. pinning the log transaction lets
6992 * us make sure we don't allow a log commit to come in after
6993 * we unlink the name but before we add the new name back in.
6995 btrfs_pin_log_trans(root
);
6998 * make sure the inode gets flushed if it is replacing
7001 if (new_inode
&& new_inode
->i_size
&&
7002 old_inode
&& S_ISREG(old_inode
->i_mode
)) {
7003 btrfs_add_ordered_operation(trans
, root
, old_inode
);
7006 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
7007 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
7008 old_inode
->i_ctime
= ctime
;
7010 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
7011 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
7013 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7014 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
7015 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
7016 old_dentry
->d_name
.name
,
7017 old_dentry
->d_name
.len
);
7019 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7020 old_dentry
->d_inode
,
7021 old_dentry
->d_name
.name
,
7022 old_dentry
->d_name
.len
);
7024 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7029 new_inode
->i_ctime
= CURRENT_TIME
;
7030 if (unlikely(new_inode
->i_ino
==
7031 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7032 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7033 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7035 new_dentry
->d_name
.name
,
7036 new_dentry
->d_name
.len
);
7037 BUG_ON(new_inode
->i_nlink
== 0);
7039 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7040 new_dentry
->d_inode
,
7041 new_dentry
->d_name
.name
,
7042 new_dentry
->d_name
.len
);
7045 if (new_inode
->i_nlink
== 0) {
7046 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7051 fixup_inode_flags(new_dir
, old_inode
);
7053 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7054 new_dentry
->d_name
.name
,
7055 new_dentry
->d_name
.len
, 0, index
);
7058 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7059 struct dentry
*parent
= dget_parent(new_dentry
);
7060 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7062 btrfs_end_log_trans(root
);
7065 btrfs_end_transaction_throttle(trans
, root
);
7067 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7068 up_read(&root
->fs_info
->subvol_sem
);
7074 * some fairly slow code that needs optimization. This walks the list
7075 * of all the inodes with pending delalloc and forces them to disk.
7077 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7079 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7080 struct btrfs_inode
*binode
;
7081 struct inode
*inode
;
7083 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7086 spin_lock(&root
->fs_info
->delalloc_lock
);
7087 while (!list_empty(head
)) {
7088 binode
= list_entry(head
->next
, struct btrfs_inode
,
7090 inode
= igrab(&binode
->vfs_inode
);
7092 list_del_init(&binode
->delalloc_inodes
);
7093 spin_unlock(&root
->fs_info
->delalloc_lock
);
7095 filemap_flush(inode
->i_mapping
);
7097 btrfs_add_delayed_iput(inode
);
7102 spin_lock(&root
->fs_info
->delalloc_lock
);
7104 spin_unlock(&root
->fs_info
->delalloc_lock
);
7106 /* the filemap_flush will queue IO into the worker threads, but
7107 * we have to make sure the IO is actually started and that
7108 * ordered extents get created before we return
7110 atomic_inc(&root
->fs_info
->async_submit_draining
);
7111 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7112 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7113 wait_event(root
->fs_info
->async_submit_wait
,
7114 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7115 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7117 atomic_dec(&root
->fs_info
->async_submit_draining
);
7121 int btrfs_start_one_delalloc_inode(struct btrfs_root
*root
, int delay_iput
,
7124 struct btrfs_inode
*binode
;
7125 struct inode
*inode
= NULL
;
7127 spin_lock(&root
->fs_info
->delalloc_lock
);
7128 while (!list_empty(&root
->fs_info
->delalloc_inodes
)) {
7129 binode
= list_entry(root
->fs_info
->delalloc_inodes
.next
,
7130 struct btrfs_inode
, delalloc_inodes
);
7131 inode
= igrab(&binode
->vfs_inode
);
7133 list_move_tail(&binode
->delalloc_inodes
,
7134 &root
->fs_info
->delalloc_inodes
);
7138 list_del_init(&binode
->delalloc_inodes
);
7139 cond_resched_lock(&root
->fs_info
->delalloc_lock
);
7141 spin_unlock(&root
->fs_info
->delalloc_lock
);
7145 filemap_write_and_wait(inode
->i_mapping
);
7147 * We have to do this because compression doesn't
7148 * actually set PG_writeback until it submits the pages
7149 * for IO, which happens in an async thread, so we could
7150 * race and not actually wait for any writeback pages
7151 * because they've not been submitted yet. Technically
7152 * this could still be the case for the ordered stuff
7153 * since the async thread may not have started to do its
7154 * work yet. If this becomes the case then we need to
7155 * figure out a way to make sure that in writepage we
7156 * wait for any async pages to be submitted before
7157 * returning so that fdatawait does what its supposed to
7160 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
7162 filemap_flush(inode
->i_mapping
);
7165 btrfs_add_delayed_iput(inode
);
7173 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7174 const char *symname
)
7176 struct btrfs_trans_handle
*trans
;
7177 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7178 struct btrfs_path
*path
;
7179 struct btrfs_key key
;
7180 struct inode
*inode
= NULL
;
7188 struct btrfs_file_extent_item
*ei
;
7189 struct extent_buffer
*leaf
;
7190 unsigned long nr
= 0;
7192 name_len
= strlen(symname
) + 1;
7193 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7194 return -ENAMETOOLONG
;
7196 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
7200 * 2 items for inode item and ref
7201 * 2 items for dir items
7202 * 1 item for xattr if selinux is on
7204 trans
= btrfs_start_transaction(root
, 5);
7206 return PTR_ERR(trans
);
7208 btrfs_set_trans_block_group(trans
, dir
);
7210 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7211 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
7212 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
7214 err
= PTR_ERR(inode
);
7218 err
= btrfs_init_inode_security(trans
, inode
, dir
);
7224 btrfs_set_trans_block_group(trans
, inode
);
7225 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7229 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7230 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7231 inode
->i_fop
= &btrfs_file_operations
;
7232 inode
->i_op
= &btrfs_file_inode_operations
;
7233 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7235 btrfs_update_inode_block_group(trans
, inode
);
7236 btrfs_update_inode_block_group(trans
, dir
);
7240 path
= btrfs_alloc_path();
7242 key
.objectid
= inode
->i_ino
;
7244 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7245 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7246 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7252 leaf
= path
->nodes
[0];
7253 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7254 struct btrfs_file_extent_item
);
7255 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7256 btrfs_set_file_extent_type(leaf
, ei
,
7257 BTRFS_FILE_EXTENT_INLINE
);
7258 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7259 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7260 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7261 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7263 ptr
= btrfs_file_extent_inline_start(ei
);
7264 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7265 btrfs_mark_buffer_dirty(leaf
);
7266 btrfs_free_path(path
);
7268 inode
->i_op
= &btrfs_symlink_inode_operations
;
7269 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7270 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7271 inode_set_bytes(inode
, name_len
);
7272 btrfs_i_size_write(inode
, name_len
- 1);
7273 err
= btrfs_update_inode(trans
, root
, inode
);
7278 nr
= trans
->blocks_used
;
7279 btrfs_end_transaction_throttle(trans
, root
);
7281 inode_dec_link_count(inode
);
7284 btrfs_btree_balance_dirty(root
, nr
);
7288 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7289 u64 start
, u64 num_bytes
, u64 min_size
,
7290 loff_t actual_len
, u64
*alloc_hint
,
7291 struct btrfs_trans_handle
*trans
)
7293 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7294 struct btrfs_key ins
;
7295 u64 cur_offset
= start
;
7298 bool own_trans
= true;
7302 while (num_bytes
> 0) {
7304 trans
= btrfs_start_transaction(root
, 3);
7305 if (IS_ERR(trans
)) {
7306 ret
= PTR_ERR(trans
);
7311 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7312 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7315 btrfs_end_transaction(trans
, root
);
7319 ret
= insert_reserved_file_extent(trans
, inode
,
7320 cur_offset
, ins
.objectid
,
7321 ins
.offset
, ins
.offset
,
7322 ins
.offset
, 0, 0, 0,
7323 BTRFS_FILE_EXTENT_PREALLOC
);
7325 btrfs_drop_extent_cache(inode
, cur_offset
,
7326 cur_offset
+ ins
.offset
-1, 0);
7328 num_bytes
-= ins
.offset
;
7329 cur_offset
+= ins
.offset
;
7330 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7332 inode
->i_ctime
= CURRENT_TIME
;
7333 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7334 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7335 (actual_len
> inode
->i_size
) &&
7336 (cur_offset
> inode
->i_size
)) {
7337 if (cur_offset
> actual_len
)
7338 i_size
= actual_len
;
7340 i_size
= cur_offset
;
7341 i_size_write(inode
, i_size
);
7342 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7345 ret
= btrfs_update_inode(trans
, root
, inode
);
7349 btrfs_end_transaction(trans
, root
);
7354 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7355 u64 start
, u64 num_bytes
, u64 min_size
,
7356 loff_t actual_len
, u64
*alloc_hint
)
7358 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7359 min_size
, actual_len
, alloc_hint
,
7363 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7364 struct btrfs_trans_handle
*trans
, int mode
,
7365 u64 start
, u64 num_bytes
, u64 min_size
,
7366 loff_t actual_len
, u64
*alloc_hint
)
7368 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7369 min_size
, actual_len
, alloc_hint
, trans
);
7372 static int btrfs_set_page_dirty(struct page
*page
)
7374 return __set_page_dirty_nobuffers(page
);
7377 static int btrfs_permission(struct inode
*inode
, int mask
, unsigned int flags
)
7379 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7381 if (btrfs_root_readonly(root
) && (mask
& MAY_WRITE
))
7383 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
7385 return generic_permission(inode
, mask
, flags
, btrfs_check_acl
);
7388 static const struct inode_operations btrfs_dir_inode_operations
= {
7389 .getattr
= btrfs_getattr
,
7390 .lookup
= btrfs_lookup
,
7391 .create
= btrfs_create
,
7392 .unlink
= btrfs_unlink
,
7394 .mkdir
= btrfs_mkdir
,
7395 .rmdir
= btrfs_rmdir
,
7396 .rename
= btrfs_rename
,
7397 .symlink
= btrfs_symlink
,
7398 .setattr
= btrfs_setattr
,
7399 .mknod
= btrfs_mknod
,
7400 .setxattr
= btrfs_setxattr
,
7401 .getxattr
= btrfs_getxattr
,
7402 .listxattr
= btrfs_listxattr
,
7403 .removexattr
= btrfs_removexattr
,
7404 .permission
= btrfs_permission
,
7406 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7407 .lookup
= btrfs_lookup
,
7408 .permission
= btrfs_permission
,
7411 static const struct file_operations btrfs_dir_file_operations
= {
7412 .llseek
= generic_file_llseek
,
7413 .read
= generic_read_dir
,
7414 .readdir
= btrfs_real_readdir
,
7415 .unlocked_ioctl
= btrfs_ioctl
,
7416 #ifdef CONFIG_COMPAT
7417 .compat_ioctl
= btrfs_ioctl
,
7419 .release
= btrfs_release_file
,
7420 .fsync
= btrfs_sync_file
,
7423 static struct extent_io_ops btrfs_extent_io_ops
= {
7424 .fill_delalloc
= run_delalloc_range
,
7425 .submit_bio_hook
= btrfs_submit_bio_hook
,
7426 .merge_bio_hook
= btrfs_merge_bio_hook
,
7427 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7428 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7429 .writepage_start_hook
= btrfs_writepage_start_hook
,
7430 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
7431 .set_bit_hook
= btrfs_set_bit_hook
,
7432 .clear_bit_hook
= btrfs_clear_bit_hook
,
7433 .merge_extent_hook
= btrfs_merge_extent_hook
,
7434 .split_extent_hook
= btrfs_split_extent_hook
,
7438 * btrfs doesn't support the bmap operation because swapfiles
7439 * use bmap to make a mapping of extents in the file. They assume
7440 * these extents won't change over the life of the file and they
7441 * use the bmap result to do IO directly to the drive.
7443 * the btrfs bmap call would return logical addresses that aren't
7444 * suitable for IO and they also will change frequently as COW
7445 * operations happen. So, swapfile + btrfs == corruption.
7447 * For now we're avoiding this by dropping bmap.
7449 static const struct address_space_operations btrfs_aops
= {
7450 .readpage
= btrfs_readpage
,
7451 .writepage
= btrfs_writepage
,
7452 .writepages
= btrfs_writepages
,
7453 .readpages
= btrfs_readpages
,
7454 .sync_page
= block_sync_page
,
7455 .direct_IO
= btrfs_direct_IO
,
7456 .invalidatepage
= btrfs_invalidatepage
,
7457 .releasepage
= btrfs_releasepage
,
7458 .set_page_dirty
= btrfs_set_page_dirty
,
7459 .error_remove_page
= generic_error_remove_page
,
7462 static const struct address_space_operations btrfs_symlink_aops
= {
7463 .readpage
= btrfs_readpage
,
7464 .writepage
= btrfs_writepage
,
7465 .invalidatepage
= btrfs_invalidatepage
,
7466 .releasepage
= btrfs_releasepage
,
7469 static const struct inode_operations btrfs_file_inode_operations
= {
7470 .getattr
= btrfs_getattr
,
7471 .setattr
= btrfs_setattr
,
7472 .setxattr
= btrfs_setxattr
,
7473 .getxattr
= btrfs_getxattr
,
7474 .listxattr
= btrfs_listxattr
,
7475 .removexattr
= btrfs_removexattr
,
7476 .permission
= btrfs_permission
,
7477 .fiemap
= btrfs_fiemap
,
7479 static const struct inode_operations btrfs_special_inode_operations
= {
7480 .getattr
= btrfs_getattr
,
7481 .setattr
= btrfs_setattr
,
7482 .permission
= btrfs_permission
,
7483 .setxattr
= btrfs_setxattr
,
7484 .getxattr
= btrfs_getxattr
,
7485 .listxattr
= btrfs_listxattr
,
7486 .removexattr
= btrfs_removexattr
,
7488 static const struct inode_operations btrfs_symlink_inode_operations
= {
7489 .readlink
= generic_readlink
,
7490 .follow_link
= page_follow_link_light
,
7491 .put_link
= page_put_link
,
7492 .getattr
= btrfs_getattr
,
7493 .permission
= btrfs_permission
,
7494 .setxattr
= btrfs_setxattr
,
7495 .getxattr
= btrfs_getxattr
,
7496 .listxattr
= btrfs_listxattr
,
7497 .removexattr
= btrfs_removexattr
,
7500 const struct dentry_operations btrfs_dentry_operations
= {
7501 .d_delete
= btrfs_dentry_delete
,