2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
44 #include "transaction.h"
45 #include "btrfs_inode.h"
47 #include "print-tree.h"
49 #include "ordered-data.h"
52 #include "compression.h"
54 #include "free-space-cache.h"
55 #include "inode-map.h"
57 struct btrfs_iget_args
{
59 struct btrfs_root
*root
;
62 static const struct inode_operations btrfs_dir_inode_operations
;
63 static const struct inode_operations btrfs_symlink_inode_operations
;
64 static const struct inode_operations btrfs_dir_ro_inode_operations
;
65 static const struct inode_operations btrfs_special_inode_operations
;
66 static const struct inode_operations btrfs_file_inode_operations
;
67 static const struct address_space_operations btrfs_aops
;
68 static const struct address_space_operations btrfs_symlink_aops
;
69 static const struct file_operations btrfs_dir_file_operations
;
70 static struct extent_io_ops btrfs_extent_io_ops
;
72 static struct kmem_cache
*btrfs_inode_cachep
;
73 struct kmem_cache
*btrfs_trans_handle_cachep
;
74 struct kmem_cache
*btrfs_transaction_cachep
;
75 struct kmem_cache
*btrfs_path_cachep
;
76 struct kmem_cache
*btrfs_free_space_cachep
;
79 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
80 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
81 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
82 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
83 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
84 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
85 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
86 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
89 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
);
90 static int btrfs_truncate(struct inode
*inode
);
91 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
92 static noinline
int cow_file_range(struct inode
*inode
,
93 struct page
*locked_page
,
94 u64 start
, u64 end
, int *page_started
,
95 unsigned long *nr_written
, int unlock
);
97 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
98 struct inode
*inode
, struct inode
*dir
,
99 const struct qstr
*qstr
)
103 err
= btrfs_init_acl(trans
, inode
, dir
);
105 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
110 * this does all the hard work for inserting an inline extent into
111 * the btree. The caller should have done a btrfs_drop_extents so that
112 * no overlapping inline items exist in the btree
114 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
115 struct btrfs_root
*root
, struct inode
*inode
,
116 u64 start
, size_t size
, size_t compressed_size
,
118 struct page
**compressed_pages
)
120 struct btrfs_key key
;
121 struct btrfs_path
*path
;
122 struct extent_buffer
*leaf
;
123 struct page
*page
= NULL
;
126 struct btrfs_file_extent_item
*ei
;
129 size_t cur_size
= size
;
131 unsigned long offset
;
133 if (compressed_size
&& compressed_pages
)
134 cur_size
= compressed_size
;
136 path
= btrfs_alloc_path();
140 path
->leave_spinning
= 1;
142 key
.objectid
= btrfs_ino(inode
);
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
, int compress_type
,
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
,
256 compress_type
, compressed_pages
);
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 /* if this is a small write inside eof, kick off a defragbot */
345 if (end
<= BTRFS_I(inode
)->disk_i_size
&& (end
- start
+ 1) < 16 * 1024)
346 btrfs_add_inode_defrag(NULL
, inode
);
348 actual_end
= min_t(u64
, isize
, end
+ 1);
351 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
352 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
355 * we don't want to send crud past the end of i_size through
356 * compression, that's just a waste of CPU time. So, if the
357 * end of the file is before the start of our current
358 * requested range of bytes, we bail out to the uncompressed
359 * cleanup code that can deal with all of this.
361 * It isn't really the fastest way to fix things, but this is a
362 * very uncommon corner.
364 if (actual_end
<= start
)
365 goto cleanup_and_bail_uncompressed
;
367 total_compressed
= actual_end
- start
;
369 /* we want to make sure that amount of ram required to uncompress
370 * an extent is reasonable, so we limit the total size in ram
371 * of a compressed extent to 128k. This is a crucial number
372 * because it also controls how easily we can spread reads across
373 * cpus for decompression.
375 * We also want to make sure the amount of IO required to do
376 * a random read is reasonably small, so we limit the size of
377 * a compressed extent to 128k.
379 total_compressed
= min(total_compressed
, max_uncompressed
);
380 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
381 num_bytes
= max(blocksize
, num_bytes
);
386 * we do compression for mount -o compress and when the
387 * inode has not been flagged as nocompress. This flag can
388 * change at any time if we discover bad compression ratios.
390 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
391 (btrfs_test_opt(root
, COMPRESS
) ||
392 (BTRFS_I(inode
)->force_compress
) ||
393 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
395 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
398 if (BTRFS_I(inode
)->force_compress
)
399 compress_type
= BTRFS_I(inode
)->force_compress
;
401 ret
= btrfs_compress_pages(compress_type
,
402 inode
->i_mapping
, start
,
403 total_compressed
, pages
,
404 nr_pages
, &nr_pages_ret
,
410 unsigned long offset
= total_compressed
&
411 (PAGE_CACHE_SIZE
- 1);
412 struct page
*page
= pages
[nr_pages_ret
- 1];
415 /* zero the tail end of the last page, we might be
416 * sending it down to disk
419 kaddr
= kmap_atomic(page
, KM_USER0
);
420 memset(kaddr
+ offset
, 0,
421 PAGE_CACHE_SIZE
- offset
);
422 kunmap_atomic(kaddr
, KM_USER0
);
428 trans
= btrfs_join_transaction(root
);
429 BUG_ON(IS_ERR(trans
));
430 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
432 /* lets try to make an inline extent */
433 if (ret
|| total_in
< (actual_end
- start
)) {
434 /* we didn't compress the entire range, try
435 * to make an uncompressed inline extent.
437 ret
= cow_file_range_inline(trans
, root
, inode
,
438 start
, end
, 0, 0, NULL
);
440 /* try making a compressed inline extent */
441 ret
= cow_file_range_inline(trans
, root
, inode
,
444 compress_type
, pages
);
448 * inline extent creation worked, we don't need
449 * to create any more async work items. Unlock
450 * and free up our temp pages.
452 extent_clear_unlock_delalloc(inode
,
453 &BTRFS_I(inode
)->io_tree
,
455 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
456 EXTENT_CLEAR_DELALLOC
|
457 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
459 btrfs_end_transaction(trans
, root
);
462 btrfs_end_transaction(trans
, root
);
467 * we aren't doing an inline extent round the compressed size
468 * up to a block size boundary so the allocator does sane
471 total_compressed
= (total_compressed
+ blocksize
- 1) &
475 * one last check to make sure the compression is really a
476 * win, compare the page count read with the blocks on disk
478 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
479 ~(PAGE_CACHE_SIZE
- 1);
480 if (total_compressed
>= total_in
) {
483 num_bytes
= total_in
;
486 if (!will_compress
&& pages
) {
488 * the compression code ran but failed to make things smaller,
489 * free any pages it allocated and our page pointer array
491 for (i
= 0; i
< nr_pages_ret
; i
++) {
492 WARN_ON(pages
[i
]->mapping
);
493 page_cache_release(pages
[i
]);
497 total_compressed
= 0;
500 /* flag the file so we don't compress in the future */
501 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
502 !(BTRFS_I(inode
)->force_compress
)) {
503 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
509 /* the async work queues will take care of doing actual
510 * allocation on disk for these compressed pages,
511 * and will submit them to the elevator.
513 add_async_extent(async_cow
, start
, num_bytes
,
514 total_compressed
, pages
, nr_pages_ret
,
517 if (start
+ num_bytes
< end
) {
524 cleanup_and_bail_uncompressed
:
526 * No compression, but we still need to write the pages in
527 * the file we've been given so far. redirty the locked
528 * page if it corresponds to our extent and set things up
529 * for the async work queue to run cow_file_range to do
530 * the normal delalloc dance
532 if (page_offset(locked_page
) >= start
&&
533 page_offset(locked_page
) <= end
) {
534 __set_page_dirty_nobuffers(locked_page
);
535 /* unlocked later on in the async handlers */
537 add_async_extent(async_cow
, start
, end
- start
+ 1,
538 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
546 for (i
= 0; i
< nr_pages_ret
; i
++) {
547 WARN_ON(pages
[i
]->mapping
);
548 page_cache_release(pages
[i
]);
556 * phase two of compressed writeback. This is the ordered portion
557 * of the code, which only gets called in the order the work was
558 * queued. We walk all the async extents created by compress_file_range
559 * and send them down to the disk.
561 static noinline
int submit_compressed_extents(struct inode
*inode
,
562 struct async_cow
*async_cow
)
564 struct async_extent
*async_extent
;
566 struct btrfs_trans_handle
*trans
;
567 struct btrfs_key ins
;
568 struct extent_map
*em
;
569 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
570 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
571 struct extent_io_tree
*io_tree
;
574 if (list_empty(&async_cow
->extents
))
578 while (!list_empty(&async_cow
->extents
)) {
579 async_extent
= list_entry(async_cow
->extents
.next
,
580 struct async_extent
, list
);
581 list_del(&async_extent
->list
);
583 io_tree
= &BTRFS_I(inode
)->io_tree
;
586 /* did the compression code fall back to uncompressed IO? */
587 if (!async_extent
->pages
) {
588 int page_started
= 0;
589 unsigned long nr_written
= 0;
591 lock_extent(io_tree
, async_extent
->start
,
592 async_extent
->start
+
593 async_extent
->ram_size
- 1, GFP_NOFS
);
595 /* allocate blocks */
596 ret
= cow_file_range(inode
, async_cow
->locked_page
,
598 async_extent
->start
+
599 async_extent
->ram_size
- 1,
600 &page_started
, &nr_written
, 0);
603 * if page_started, cow_file_range inserted an
604 * inline extent and took care of all the unlocking
605 * and IO for us. Otherwise, we need to submit
606 * all those pages down to the drive.
608 if (!page_started
&& !ret
)
609 extent_write_locked_range(io_tree
,
610 inode
, async_extent
->start
,
611 async_extent
->start
+
612 async_extent
->ram_size
- 1,
620 lock_extent(io_tree
, async_extent
->start
,
621 async_extent
->start
+ async_extent
->ram_size
- 1,
624 trans
= btrfs_join_transaction(root
);
625 BUG_ON(IS_ERR(trans
));
626 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
627 ret
= btrfs_reserve_extent(trans
, root
,
628 async_extent
->compressed_size
,
629 async_extent
->compressed_size
,
632 btrfs_end_transaction(trans
, root
);
636 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
637 WARN_ON(async_extent
->pages
[i
]->mapping
);
638 page_cache_release(async_extent
->pages
[i
]);
640 kfree(async_extent
->pages
);
641 async_extent
->nr_pages
= 0;
642 async_extent
->pages
= NULL
;
643 unlock_extent(io_tree
, async_extent
->start
,
644 async_extent
->start
+
645 async_extent
->ram_size
- 1, GFP_NOFS
);
650 * here we're doing allocation and writeback of the
653 btrfs_drop_extent_cache(inode
, async_extent
->start
,
654 async_extent
->start
+
655 async_extent
->ram_size
- 1, 0);
657 em
= alloc_extent_map();
659 em
->start
= async_extent
->start
;
660 em
->len
= async_extent
->ram_size
;
661 em
->orig_start
= em
->start
;
663 em
->block_start
= ins
.objectid
;
664 em
->block_len
= ins
.offset
;
665 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
666 em
->compress_type
= async_extent
->compress_type
;
667 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
668 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
671 write_lock(&em_tree
->lock
);
672 ret
= add_extent_mapping(em_tree
, em
);
673 write_unlock(&em_tree
->lock
);
674 if (ret
!= -EEXIST
) {
678 btrfs_drop_extent_cache(inode
, async_extent
->start
,
679 async_extent
->start
+
680 async_extent
->ram_size
- 1, 0);
683 ret
= btrfs_add_ordered_extent_compress(inode
,
686 async_extent
->ram_size
,
688 BTRFS_ORDERED_COMPRESSED
,
689 async_extent
->compress_type
);
693 * clear dirty, set writeback and unlock the pages.
695 extent_clear_unlock_delalloc(inode
,
696 &BTRFS_I(inode
)->io_tree
,
698 async_extent
->start
+
699 async_extent
->ram_size
- 1,
700 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
701 EXTENT_CLEAR_UNLOCK
|
702 EXTENT_CLEAR_DELALLOC
|
703 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
705 ret
= btrfs_submit_compressed_write(inode
,
707 async_extent
->ram_size
,
709 ins
.offset
, async_extent
->pages
,
710 async_extent
->nr_pages
);
713 alloc_hint
= ins
.objectid
+ ins
.offset
;
721 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
724 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
725 struct extent_map
*em
;
728 read_lock(&em_tree
->lock
);
729 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
732 * if block start isn't an actual block number then find the
733 * first block in this inode and use that as a hint. If that
734 * block is also bogus then just don't worry about it.
736 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
738 em
= search_extent_mapping(em_tree
, 0, 0);
739 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
740 alloc_hint
= em
->block_start
;
744 alloc_hint
= em
->block_start
;
748 read_unlock(&em_tree
->lock
);
753 static inline bool is_free_space_inode(struct btrfs_root
*root
,
756 if (root
== root
->fs_info
->tree_root
||
757 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
)
763 * when extent_io.c finds a delayed allocation range in the file,
764 * the call backs end up in this code. The basic idea is to
765 * allocate extents on disk for the range, and create ordered data structs
766 * in ram to track those extents.
768 * locked_page is the page that writepage had locked already. We use
769 * it to make sure we don't do extra locks or unlocks.
771 * *page_started is set to one if we unlock locked_page and do everything
772 * required to start IO on it. It may be clean and already done with
775 static noinline
int cow_file_range(struct inode
*inode
,
776 struct page
*locked_page
,
777 u64 start
, u64 end
, int *page_started
,
778 unsigned long *nr_written
,
781 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
782 struct btrfs_trans_handle
*trans
;
785 unsigned long ram_size
;
788 u64 blocksize
= root
->sectorsize
;
789 struct btrfs_key ins
;
790 struct extent_map
*em
;
791 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
794 BUG_ON(is_free_space_inode(root
, inode
));
795 trans
= btrfs_join_transaction(root
);
796 BUG_ON(IS_ERR(trans
));
797 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
799 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
800 num_bytes
= max(blocksize
, num_bytes
);
801 disk_num_bytes
= num_bytes
;
804 /* if this is a small write inside eof, kick off defrag */
805 if (end
<= BTRFS_I(inode
)->disk_i_size
&& num_bytes
< 64 * 1024)
806 btrfs_add_inode_defrag(trans
, inode
);
809 /* lets try to make an inline extent */
810 ret
= cow_file_range_inline(trans
, root
, inode
,
811 start
, end
, 0, 0, NULL
);
813 extent_clear_unlock_delalloc(inode
,
814 &BTRFS_I(inode
)->io_tree
,
816 EXTENT_CLEAR_UNLOCK_PAGE
|
817 EXTENT_CLEAR_UNLOCK
|
818 EXTENT_CLEAR_DELALLOC
|
820 EXTENT_SET_WRITEBACK
|
821 EXTENT_END_WRITEBACK
);
823 *nr_written
= *nr_written
+
824 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
831 BUG_ON(disk_num_bytes
>
832 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
834 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
835 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
837 while (disk_num_bytes
> 0) {
840 cur_alloc_size
= disk_num_bytes
;
841 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
842 root
->sectorsize
, 0, alloc_hint
,
846 em
= alloc_extent_map();
849 em
->orig_start
= em
->start
;
850 ram_size
= ins
.offset
;
851 em
->len
= ins
.offset
;
853 em
->block_start
= ins
.objectid
;
854 em
->block_len
= ins
.offset
;
855 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
856 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
859 write_lock(&em_tree
->lock
);
860 ret
= add_extent_mapping(em_tree
, em
);
861 write_unlock(&em_tree
->lock
);
862 if (ret
!= -EEXIST
) {
866 btrfs_drop_extent_cache(inode
, start
,
867 start
+ ram_size
- 1, 0);
870 cur_alloc_size
= ins
.offset
;
871 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
872 ram_size
, cur_alloc_size
, 0);
875 if (root
->root_key
.objectid
==
876 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
877 ret
= btrfs_reloc_clone_csums(inode
, start
,
882 if (disk_num_bytes
< cur_alloc_size
)
885 /* we're not doing compressed IO, don't unlock the first
886 * page (which the caller expects to stay locked), don't
887 * clear any dirty bits and don't set any writeback bits
889 * Do set the Private2 bit so we know this page was properly
890 * setup for writepage
892 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
893 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
896 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
897 start
, start
+ ram_size
- 1,
899 disk_num_bytes
-= cur_alloc_size
;
900 num_bytes
-= cur_alloc_size
;
901 alloc_hint
= ins
.objectid
+ ins
.offset
;
902 start
+= cur_alloc_size
;
906 btrfs_end_transaction(trans
, root
);
912 * work queue call back to started compression on a file and pages
914 static noinline
void async_cow_start(struct btrfs_work
*work
)
916 struct async_cow
*async_cow
;
918 async_cow
= container_of(work
, struct async_cow
, work
);
920 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
921 async_cow
->start
, async_cow
->end
, async_cow
,
924 async_cow
->inode
= NULL
;
928 * work queue call back to submit previously compressed pages
930 static noinline
void async_cow_submit(struct btrfs_work
*work
)
932 struct async_cow
*async_cow
;
933 struct btrfs_root
*root
;
934 unsigned long nr_pages
;
936 async_cow
= container_of(work
, struct async_cow
, work
);
938 root
= async_cow
->root
;
939 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
942 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
944 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
946 waitqueue_active(&root
->fs_info
->async_submit_wait
))
947 wake_up(&root
->fs_info
->async_submit_wait
);
949 if (async_cow
->inode
)
950 submit_compressed_extents(async_cow
->inode
, async_cow
);
953 static noinline
void async_cow_free(struct btrfs_work
*work
)
955 struct async_cow
*async_cow
;
956 async_cow
= container_of(work
, struct async_cow
, work
);
960 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
961 u64 start
, u64 end
, int *page_started
,
962 unsigned long *nr_written
)
964 struct async_cow
*async_cow
;
965 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
966 unsigned long nr_pages
;
968 int limit
= 10 * 1024 * 1042;
970 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
971 1, 0, NULL
, GFP_NOFS
);
972 while (start
< end
) {
973 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
975 async_cow
->inode
= inode
;
976 async_cow
->root
= root
;
977 async_cow
->locked_page
= locked_page
;
978 async_cow
->start
= start
;
980 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
983 cur_end
= min(end
, start
+ 512 * 1024 - 1);
985 async_cow
->end
= cur_end
;
986 INIT_LIST_HEAD(&async_cow
->extents
);
988 async_cow
->work
.func
= async_cow_start
;
989 async_cow
->work
.ordered_func
= async_cow_submit
;
990 async_cow
->work
.ordered_free
= async_cow_free
;
991 async_cow
->work
.flags
= 0;
993 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
995 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
997 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
1000 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1001 wait_event(root
->fs_info
->async_submit_wait
,
1002 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1006 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1007 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1008 wait_event(root
->fs_info
->async_submit_wait
,
1009 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1013 *nr_written
+= nr_pages
;
1014 start
= cur_end
+ 1;
1020 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1021 u64 bytenr
, u64 num_bytes
)
1024 struct btrfs_ordered_sum
*sums
;
1027 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1028 bytenr
+ num_bytes
- 1, &list
, 0);
1029 if (ret
== 0 && list_empty(&list
))
1032 while (!list_empty(&list
)) {
1033 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1034 list_del(&sums
->list
);
1041 * when nowcow writeback call back. This checks for snapshots or COW copies
1042 * of the extents that exist in the file, and COWs the file as required.
1044 * If no cow copies or snapshots exist, we write directly to the existing
1047 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1048 struct page
*locked_page
,
1049 u64 start
, u64 end
, int *page_started
, int force
,
1050 unsigned long *nr_written
)
1052 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1053 struct btrfs_trans_handle
*trans
;
1054 struct extent_buffer
*leaf
;
1055 struct btrfs_path
*path
;
1056 struct btrfs_file_extent_item
*fi
;
1057 struct btrfs_key found_key
;
1070 u64 ino
= btrfs_ino(inode
);
1072 path
= btrfs_alloc_path();
1075 nolock
= is_free_space_inode(root
, inode
);
1078 trans
= btrfs_join_transaction_nolock(root
);
1080 trans
= btrfs_join_transaction(root
);
1082 BUG_ON(IS_ERR(trans
));
1083 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1085 cow_start
= (u64
)-1;
1088 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1091 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1092 leaf
= path
->nodes
[0];
1093 btrfs_item_key_to_cpu(leaf
, &found_key
,
1094 path
->slots
[0] - 1);
1095 if (found_key
.objectid
== ino
&&
1096 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1101 leaf
= path
->nodes
[0];
1102 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1103 ret
= btrfs_next_leaf(root
, path
);
1108 leaf
= path
->nodes
[0];
1114 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1116 if (found_key
.objectid
> ino
||
1117 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1118 found_key
.offset
> end
)
1121 if (found_key
.offset
> cur_offset
) {
1122 extent_end
= found_key
.offset
;
1127 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1128 struct btrfs_file_extent_item
);
1129 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1131 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1132 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1133 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1134 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1135 extent_end
= found_key
.offset
+
1136 btrfs_file_extent_num_bytes(leaf
, fi
);
1137 if (extent_end
<= start
) {
1141 if (disk_bytenr
== 0)
1143 if (btrfs_file_extent_compression(leaf
, fi
) ||
1144 btrfs_file_extent_encryption(leaf
, fi
) ||
1145 btrfs_file_extent_other_encoding(leaf
, fi
))
1147 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1149 if (btrfs_extent_readonly(root
, disk_bytenr
))
1151 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1153 extent_offset
, disk_bytenr
))
1155 disk_bytenr
+= extent_offset
;
1156 disk_bytenr
+= cur_offset
- found_key
.offset
;
1157 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1159 * force cow if csum exists in the range.
1160 * this ensure that csum for a given extent are
1161 * either valid or do not exist.
1163 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1166 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1167 extent_end
= found_key
.offset
+
1168 btrfs_file_extent_inline_len(leaf
, fi
);
1169 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1174 if (extent_end
<= start
) {
1179 if (cow_start
== (u64
)-1)
1180 cow_start
= cur_offset
;
1181 cur_offset
= extent_end
;
1182 if (cur_offset
> end
)
1188 btrfs_release_path(path
);
1189 if (cow_start
!= (u64
)-1) {
1190 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1191 found_key
.offset
- 1, page_started
,
1194 cow_start
= (u64
)-1;
1197 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1198 struct extent_map
*em
;
1199 struct extent_map_tree
*em_tree
;
1200 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1201 em
= alloc_extent_map();
1203 em
->start
= cur_offset
;
1204 em
->orig_start
= em
->start
;
1205 em
->len
= num_bytes
;
1206 em
->block_len
= num_bytes
;
1207 em
->block_start
= disk_bytenr
;
1208 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1209 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1211 write_lock(&em_tree
->lock
);
1212 ret
= add_extent_mapping(em_tree
, em
);
1213 write_unlock(&em_tree
->lock
);
1214 if (ret
!= -EEXIST
) {
1215 free_extent_map(em
);
1218 btrfs_drop_extent_cache(inode
, em
->start
,
1219 em
->start
+ em
->len
- 1, 0);
1221 type
= BTRFS_ORDERED_PREALLOC
;
1223 type
= BTRFS_ORDERED_NOCOW
;
1226 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1227 num_bytes
, num_bytes
, type
);
1230 if (root
->root_key
.objectid
==
1231 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1232 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1237 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1238 cur_offset
, cur_offset
+ num_bytes
- 1,
1239 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1240 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1241 EXTENT_SET_PRIVATE2
);
1242 cur_offset
= extent_end
;
1243 if (cur_offset
> end
)
1246 btrfs_release_path(path
);
1248 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1249 cow_start
= cur_offset
;
1250 if (cow_start
!= (u64
)-1) {
1251 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1252 page_started
, nr_written
, 1);
1257 ret
= btrfs_end_transaction_nolock(trans
, root
);
1260 ret
= btrfs_end_transaction(trans
, root
);
1263 btrfs_free_path(path
);
1268 * extent_io.c call back to do delayed allocation processing
1270 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1271 u64 start
, u64 end
, int *page_started
,
1272 unsigned long *nr_written
)
1275 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1277 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1278 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1279 page_started
, 1, nr_written
);
1280 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1281 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1282 page_started
, 0, nr_written
);
1283 else if (!btrfs_test_opt(root
, COMPRESS
) &&
1284 !(BTRFS_I(inode
)->force_compress
) &&
1285 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))
1286 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1287 page_started
, nr_written
, 1);
1289 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1290 page_started
, nr_written
);
1294 static int btrfs_split_extent_hook(struct inode
*inode
,
1295 struct extent_state
*orig
, u64 split
)
1297 /* not delalloc, ignore it */
1298 if (!(orig
->state
& EXTENT_DELALLOC
))
1301 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1306 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307 * extents so we can keep track of new extents that are just merged onto old
1308 * extents, such as when we are doing sequential writes, so we can properly
1309 * account for the metadata space we'll need.
1311 static int btrfs_merge_extent_hook(struct inode
*inode
,
1312 struct extent_state
*new,
1313 struct extent_state
*other
)
1315 /* not delalloc, ignore it */
1316 if (!(other
->state
& EXTENT_DELALLOC
))
1319 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1324 * extent_io.c set_bit_hook, used to track delayed allocation
1325 * bytes in this file, and to maintain the list of inodes that
1326 * have pending delalloc work to be done.
1328 static int btrfs_set_bit_hook(struct inode
*inode
,
1329 struct extent_state
*state
, int *bits
)
1333 * set_bit and clear bit hooks normally require _irqsave/restore
1334 * but in this case, we are only testing for the DELALLOC
1335 * bit, which is only set or cleared with irqs on
1337 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1338 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1339 u64 len
= state
->end
+ 1 - state
->start
;
1340 bool do_list
= !is_free_space_inode(root
, inode
);
1342 if (*bits
& EXTENT_FIRST_DELALLOC
)
1343 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1345 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1347 spin_lock(&root
->fs_info
->delalloc_lock
);
1348 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1349 root
->fs_info
->delalloc_bytes
+= len
;
1350 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1351 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1352 &root
->fs_info
->delalloc_inodes
);
1354 spin_unlock(&root
->fs_info
->delalloc_lock
);
1360 * extent_io.c clear_bit_hook, see set_bit_hook for why
1362 static int btrfs_clear_bit_hook(struct inode
*inode
,
1363 struct extent_state
*state
, int *bits
)
1366 * set_bit and clear bit hooks normally require _irqsave/restore
1367 * but in this case, we are only testing for the DELALLOC
1368 * bit, which is only set or cleared with irqs on
1370 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1371 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1372 u64 len
= state
->end
+ 1 - state
->start
;
1373 bool do_list
= !is_free_space_inode(root
, inode
);
1375 if (*bits
& EXTENT_FIRST_DELALLOC
)
1376 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1377 else if (!(*bits
& EXTENT_DO_ACCOUNTING
))
1378 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1380 if (*bits
& EXTENT_DO_ACCOUNTING
)
1381 btrfs_delalloc_release_metadata(inode
, len
);
1383 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1385 btrfs_free_reserved_data_space(inode
, len
);
1387 spin_lock(&root
->fs_info
->delalloc_lock
);
1388 root
->fs_info
->delalloc_bytes
-= len
;
1389 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1391 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1392 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1393 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1395 spin_unlock(&root
->fs_info
->delalloc_lock
);
1401 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1402 * we don't create bios that span stripes or chunks
1404 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1405 size_t size
, struct bio
*bio
,
1406 unsigned long bio_flags
)
1408 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1409 struct btrfs_mapping_tree
*map_tree
;
1410 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1415 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1418 length
= bio
->bi_size
;
1419 map_tree
= &root
->fs_info
->mapping_tree
;
1420 map_length
= length
;
1421 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1422 &map_length
, NULL
, 0);
1424 if (map_length
< length
+ size
)
1430 * in order to insert checksums into the metadata in large chunks,
1431 * we wait until bio submission time. All the pages in the bio are
1432 * checksummed and sums are attached onto the ordered extent record.
1434 * At IO completion time the cums attached on the ordered extent record
1435 * are inserted into the btree
1437 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1438 struct bio
*bio
, int mirror_num
,
1439 unsigned long bio_flags
,
1442 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1445 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1451 * in order to insert checksums into the metadata in large chunks,
1452 * we wait until bio submission time. All the pages in the bio are
1453 * checksummed and sums are attached onto the ordered extent record.
1455 * At IO completion time the cums attached on the ordered extent record
1456 * are inserted into the btree
1458 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1459 int mirror_num
, unsigned long bio_flags
,
1462 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1463 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1467 * extent_io.c submission hook. This does the right thing for csum calculation
1468 * on write, or reading the csums from the tree before a read
1470 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1471 int mirror_num
, unsigned long bio_flags
,
1474 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1478 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1480 if (is_free_space_inode(root
, inode
))
1481 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 2);
1483 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1486 if (!(rw
& REQ_WRITE
)) {
1487 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1488 return btrfs_submit_compressed_read(inode
, bio
,
1489 mirror_num
, bio_flags
);
1490 } else if (!skip_sum
) {
1491 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1496 } else if (!skip_sum
) {
1497 /* csum items have already been cloned */
1498 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1500 /* we're doing a write, do the async checksumming */
1501 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1502 inode
, rw
, bio
, mirror_num
,
1503 bio_flags
, bio_offset
,
1504 __btrfs_submit_bio_start
,
1505 __btrfs_submit_bio_done
);
1509 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1513 * given a list of ordered sums record them in the inode. This happens
1514 * at IO completion time based on sums calculated at bio submission time.
1516 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1517 struct inode
*inode
, u64 file_offset
,
1518 struct list_head
*list
)
1520 struct btrfs_ordered_sum
*sum
;
1522 list_for_each_entry(sum
, list
, list
) {
1523 btrfs_csum_file_blocks(trans
,
1524 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1529 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1530 struct extent_state
**cached_state
)
1532 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1534 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1535 cached_state
, GFP_NOFS
);
1538 /* see btrfs_writepage_start_hook for details on why this is required */
1539 struct btrfs_writepage_fixup
{
1541 struct btrfs_work work
;
1544 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1546 struct btrfs_writepage_fixup
*fixup
;
1547 struct btrfs_ordered_extent
*ordered
;
1548 struct extent_state
*cached_state
= NULL
;
1550 struct inode
*inode
;
1554 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1558 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1559 ClearPageChecked(page
);
1563 inode
= page
->mapping
->host
;
1564 page_start
= page_offset(page
);
1565 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1567 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1568 &cached_state
, GFP_NOFS
);
1570 /* already ordered? We're done */
1571 if (PagePrivate2(page
))
1574 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1576 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1577 page_end
, &cached_state
, GFP_NOFS
);
1579 btrfs_start_ordered_extent(inode
, ordered
, 1);
1584 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1585 ClearPageChecked(page
);
1587 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1588 &cached_state
, GFP_NOFS
);
1591 page_cache_release(page
);
1596 * There are a few paths in the higher layers of the kernel that directly
1597 * set the page dirty bit without asking the filesystem if it is a
1598 * good idea. This causes problems because we want to make sure COW
1599 * properly happens and the data=ordered rules are followed.
1601 * In our case any range that doesn't have the ORDERED bit set
1602 * hasn't been properly setup for IO. We kick off an async process
1603 * to fix it up. The async helper will wait for ordered extents, set
1604 * the delalloc bit and make it safe to write the page.
1606 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1608 struct inode
*inode
= page
->mapping
->host
;
1609 struct btrfs_writepage_fixup
*fixup
;
1610 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1612 /* this page is properly in the ordered list */
1613 if (TestClearPagePrivate2(page
))
1616 if (PageChecked(page
))
1619 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1623 SetPageChecked(page
);
1624 page_cache_get(page
);
1625 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1627 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1631 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1632 struct inode
*inode
, u64 file_pos
,
1633 u64 disk_bytenr
, u64 disk_num_bytes
,
1634 u64 num_bytes
, u64 ram_bytes
,
1635 u8 compression
, u8 encryption
,
1636 u16 other_encoding
, int extent_type
)
1638 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1639 struct btrfs_file_extent_item
*fi
;
1640 struct btrfs_path
*path
;
1641 struct extent_buffer
*leaf
;
1642 struct btrfs_key ins
;
1646 path
= btrfs_alloc_path();
1649 path
->leave_spinning
= 1;
1652 * we may be replacing one extent in the tree with another.
1653 * The new extent is pinned in the extent map, and we don't want
1654 * to drop it from the cache until it is completely in the btree.
1656 * So, tell btrfs_drop_extents to leave this extent in the cache.
1657 * the caller is expected to unpin it and allow it to be merged
1660 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1664 ins
.objectid
= btrfs_ino(inode
);
1665 ins
.offset
= file_pos
;
1666 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1667 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1669 leaf
= path
->nodes
[0];
1670 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1671 struct btrfs_file_extent_item
);
1672 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1673 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1674 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1675 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1676 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1677 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1678 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1679 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1680 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1681 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1683 btrfs_unlock_up_safe(path
, 1);
1684 btrfs_set_lock_blocking(leaf
);
1686 btrfs_mark_buffer_dirty(leaf
);
1688 inode_add_bytes(inode
, num_bytes
);
1690 ins
.objectid
= disk_bytenr
;
1691 ins
.offset
= disk_num_bytes
;
1692 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1693 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1694 root
->root_key
.objectid
,
1695 btrfs_ino(inode
), file_pos
, &ins
);
1697 btrfs_free_path(path
);
1703 * helper function for btrfs_finish_ordered_io, this
1704 * just reads in some of the csum leaves to prime them into ram
1705 * before we start the transaction. It limits the amount of btree
1706 * reads required while inside the transaction.
1708 /* as ordered data IO finishes, this gets called so we can finish
1709 * an ordered extent if the range of bytes in the file it covers are
1712 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1714 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1715 struct btrfs_trans_handle
*trans
= NULL
;
1716 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1717 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1718 struct extent_state
*cached_state
= NULL
;
1719 int compress_type
= 0;
1723 ret
= btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1727 BUG_ON(!ordered_extent
);
1729 nolock
= is_free_space_inode(root
, inode
);
1731 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1732 BUG_ON(!list_empty(&ordered_extent
->list
));
1733 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1736 trans
= btrfs_join_transaction_nolock(root
);
1738 trans
= btrfs_join_transaction(root
);
1739 BUG_ON(IS_ERR(trans
));
1740 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1741 ret
= btrfs_update_inode(trans
, root
, inode
);
1747 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1748 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1749 0, &cached_state
, GFP_NOFS
);
1752 trans
= btrfs_join_transaction_nolock(root
);
1754 trans
= btrfs_join_transaction(root
);
1755 BUG_ON(IS_ERR(trans
));
1756 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1758 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1759 compress_type
= ordered_extent
->compress_type
;
1760 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1761 BUG_ON(compress_type
);
1762 ret
= btrfs_mark_extent_written(trans
, inode
,
1763 ordered_extent
->file_offset
,
1764 ordered_extent
->file_offset
+
1765 ordered_extent
->len
);
1768 BUG_ON(root
== root
->fs_info
->tree_root
);
1769 ret
= insert_reserved_file_extent(trans
, inode
,
1770 ordered_extent
->file_offset
,
1771 ordered_extent
->start
,
1772 ordered_extent
->disk_len
,
1773 ordered_extent
->len
,
1774 ordered_extent
->len
,
1775 compress_type
, 0, 0,
1776 BTRFS_FILE_EXTENT_REG
);
1777 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1778 ordered_extent
->file_offset
,
1779 ordered_extent
->len
);
1782 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1783 ordered_extent
->file_offset
+
1784 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1786 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1787 &ordered_extent
->list
);
1789 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1791 ret
= btrfs_update_inode(trans
, root
, inode
);
1798 btrfs_end_transaction_nolock(trans
, root
);
1800 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1802 btrfs_end_transaction(trans
, root
);
1806 btrfs_put_ordered_extent(ordered_extent
);
1807 /* once for the tree */
1808 btrfs_put_ordered_extent(ordered_extent
);
1813 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1814 struct extent_state
*state
, int uptodate
)
1816 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1818 ClearPagePrivate2(page
);
1819 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1823 * When IO fails, either with EIO or csum verification fails, we
1824 * try other mirrors that might have a good copy of the data. This
1825 * io_failure_record is used to record state as we go through all the
1826 * mirrors. If another mirror has good data, the page is set up to date
1827 * and things continue. If a good mirror can't be found, the original
1828 * bio end_io callback is called to indicate things have failed.
1830 struct io_failure_record
{
1835 unsigned long bio_flags
;
1839 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1840 struct page
*page
, u64 start
, u64 end
,
1841 struct extent_state
*state
)
1843 struct io_failure_record
*failrec
= NULL
;
1845 struct extent_map
*em
;
1846 struct inode
*inode
= page
->mapping
->host
;
1847 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1848 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1855 ret
= get_state_private(failure_tree
, start
, &private);
1857 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1860 failrec
->start
= start
;
1861 failrec
->len
= end
- start
+ 1;
1862 failrec
->last_mirror
= 0;
1863 failrec
->bio_flags
= 0;
1865 read_lock(&em_tree
->lock
);
1866 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1867 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1868 free_extent_map(em
);
1871 read_unlock(&em_tree
->lock
);
1873 if (IS_ERR_OR_NULL(em
)) {
1877 logical
= start
- em
->start
;
1878 logical
= em
->block_start
+ logical
;
1879 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1880 logical
= em
->block_start
;
1881 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1882 extent_set_compress_type(&failrec
->bio_flags
,
1885 failrec
->logical
= logical
;
1886 free_extent_map(em
);
1887 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1888 EXTENT_DIRTY
, GFP_NOFS
);
1889 set_state_private(failure_tree
, start
,
1890 (u64
)(unsigned long)failrec
);
1892 failrec
= (struct io_failure_record
*)(unsigned long)private;
1894 num_copies
= btrfs_num_copies(
1895 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1896 failrec
->logical
, failrec
->len
);
1897 failrec
->last_mirror
++;
1899 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1900 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1903 if (state
&& state
->start
!= failrec
->start
)
1905 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1907 if (!state
|| failrec
->last_mirror
> num_copies
) {
1908 set_state_private(failure_tree
, failrec
->start
, 0);
1909 clear_extent_bits(failure_tree
, failrec
->start
,
1910 failrec
->start
+ failrec
->len
- 1,
1911 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1915 bio
= bio_alloc(GFP_NOFS
, 1);
1916 bio
->bi_private
= state
;
1917 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1918 bio
->bi_sector
= failrec
->logical
>> 9;
1919 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1922 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1923 if (failed_bio
->bi_rw
& REQ_WRITE
)
1928 ret
= BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1929 failrec
->last_mirror
,
1930 failrec
->bio_flags
, 0);
1935 * each time an IO finishes, we do a fast check in the IO failure tree
1936 * to see if we need to process or clean up an io_failure_record
1938 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1941 u64 private_failure
;
1942 struct io_failure_record
*failure
;
1946 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1947 (u64
)-1, 1, EXTENT_DIRTY
, 0)) {
1948 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1949 start
, &private_failure
);
1951 failure
= (struct io_failure_record
*)(unsigned long)
1953 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1955 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1957 failure
->start
+ failure
->len
- 1,
1958 EXTENT_DIRTY
| EXTENT_LOCKED
,
1967 * when reads are done, we need to check csums to verify the data is correct
1968 * if there's a match, we allow the bio to finish. If not, we go through
1969 * the io_failure_record routines to find good copies
1971 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1972 struct extent_state
*state
)
1974 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1975 struct inode
*inode
= page
->mapping
->host
;
1976 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1978 u64
private = ~(u32
)0;
1980 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1983 if (PageChecked(page
)) {
1984 ClearPageChecked(page
);
1988 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1991 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1992 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1993 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1998 if (state
&& state
->start
== start
) {
1999 private = state
->private;
2002 ret
= get_state_private(io_tree
, start
, &private);
2004 kaddr
= kmap_atomic(page
, KM_USER0
);
2008 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
2009 btrfs_csum_final(csum
, (char *)&csum
);
2010 if (csum
!= private)
2013 kunmap_atomic(kaddr
, KM_USER0
);
2015 /* if the io failure tree for this inode is non-empty,
2016 * check to see if we've recovered from a failed IO
2018 btrfs_clean_io_failures(inode
, start
);
2022 printk_ratelimited(KERN_INFO
"btrfs csum failed ino %llu off %llu csum %u "
2024 (unsigned long long)btrfs_ino(page
->mapping
->host
),
2025 (unsigned long long)start
, csum
,
2026 (unsigned long long)private);
2027 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2028 flush_dcache_page(page
);
2029 kunmap_atomic(kaddr
, KM_USER0
);
2035 struct delayed_iput
{
2036 struct list_head list
;
2037 struct inode
*inode
;
2040 void btrfs_add_delayed_iput(struct inode
*inode
)
2042 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2043 struct delayed_iput
*delayed
;
2045 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2048 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2049 delayed
->inode
= inode
;
2051 spin_lock(&fs_info
->delayed_iput_lock
);
2052 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2053 spin_unlock(&fs_info
->delayed_iput_lock
);
2056 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2059 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2060 struct delayed_iput
*delayed
;
2063 spin_lock(&fs_info
->delayed_iput_lock
);
2064 empty
= list_empty(&fs_info
->delayed_iputs
);
2065 spin_unlock(&fs_info
->delayed_iput_lock
);
2069 down_read(&root
->fs_info
->cleanup_work_sem
);
2070 spin_lock(&fs_info
->delayed_iput_lock
);
2071 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2072 spin_unlock(&fs_info
->delayed_iput_lock
);
2074 while (!list_empty(&list
)) {
2075 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2076 list_del(&delayed
->list
);
2077 iput(delayed
->inode
);
2080 up_read(&root
->fs_info
->cleanup_work_sem
);
2084 * calculate extra metadata reservation when snapshotting a subvolume
2085 * contains orphan files.
2087 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle
*trans
,
2088 struct btrfs_pending_snapshot
*pending
,
2089 u64
*bytes_to_reserve
)
2091 struct btrfs_root
*root
;
2092 struct btrfs_block_rsv
*block_rsv
;
2096 root
= pending
->root
;
2097 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2100 block_rsv
= root
->orphan_block_rsv
;
2102 /* orphan block reservation for the snapshot */
2103 num_bytes
= block_rsv
->size
;
2106 * after the snapshot is created, COWing tree blocks may use more
2107 * space than it frees. So we should make sure there is enough
2110 index
= trans
->transid
& 0x1;
2111 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2112 num_bytes
+= block_rsv
->size
-
2113 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2116 *bytes_to_reserve
+= num_bytes
;
2119 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle
*trans
,
2120 struct btrfs_pending_snapshot
*pending
)
2122 struct btrfs_root
*root
= pending
->root
;
2123 struct btrfs_root
*snap
= pending
->snap
;
2124 struct btrfs_block_rsv
*block_rsv
;
2129 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2132 /* refill source subvolume's orphan block reservation */
2133 block_rsv
= root
->orphan_block_rsv
;
2134 index
= trans
->transid
& 0x1;
2135 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2136 num_bytes
= block_rsv
->size
-
2137 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2138 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2139 root
->orphan_block_rsv
,
2144 /* setup orphan block reservation for the snapshot */
2145 block_rsv
= btrfs_alloc_block_rsv(snap
);
2148 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2149 snap
->orphan_block_rsv
= block_rsv
;
2151 num_bytes
= root
->orphan_block_rsv
->size
;
2152 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2153 block_rsv
, num_bytes
);
2157 /* insert orphan item for the snapshot */
2158 WARN_ON(!root
->orphan_item_inserted
);
2159 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2160 snap
->root_key
.objectid
);
2162 snap
->orphan_item_inserted
= 1;
2166 enum btrfs_orphan_cleanup_state
{
2167 ORPHAN_CLEANUP_STARTED
= 1,
2168 ORPHAN_CLEANUP_DONE
= 2,
2172 * This is called in transaction commmit time. If there are no orphan
2173 * files in the subvolume, it removes orphan item and frees block_rsv
2176 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2177 struct btrfs_root
*root
)
2181 if (!list_empty(&root
->orphan_list
) ||
2182 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2185 if (root
->orphan_item_inserted
&&
2186 btrfs_root_refs(&root
->root_item
) > 0) {
2187 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2188 root
->root_key
.objectid
);
2190 root
->orphan_item_inserted
= 0;
2193 if (root
->orphan_block_rsv
) {
2194 WARN_ON(root
->orphan_block_rsv
->size
> 0);
2195 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
2196 root
->orphan_block_rsv
= NULL
;
2201 * This creates an orphan entry for the given inode in case something goes
2202 * wrong in the middle of an unlink/truncate.
2204 * NOTE: caller of this function should reserve 5 units of metadata for
2207 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2209 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2210 struct btrfs_block_rsv
*block_rsv
= NULL
;
2215 if (!root
->orphan_block_rsv
) {
2216 block_rsv
= btrfs_alloc_block_rsv(root
);
2220 spin_lock(&root
->orphan_lock
);
2221 if (!root
->orphan_block_rsv
) {
2222 root
->orphan_block_rsv
= block_rsv
;
2223 } else if (block_rsv
) {
2224 btrfs_free_block_rsv(root
, block_rsv
);
2228 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2229 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2232 * For proper ENOSPC handling, we should do orphan
2233 * cleanup when mounting. But this introduces backward
2234 * compatibility issue.
2236 if (!xchg(&root
->orphan_item_inserted
, 1))
2244 if (!BTRFS_I(inode
)->orphan_meta_reserved
) {
2245 BTRFS_I(inode
)->orphan_meta_reserved
= 1;
2248 spin_unlock(&root
->orphan_lock
);
2251 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2253 /* grab metadata reservation from transaction handle */
2255 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2259 /* insert an orphan item to track this unlinked/truncated file */
2261 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
2265 /* insert an orphan item to track subvolume contains orphan files */
2267 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2268 root
->root_key
.objectid
);
2275 * We have done the truncate/delete so we can go ahead and remove the orphan
2276 * item for this particular inode.
2278 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2280 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2281 int delete_item
= 0;
2282 int release_rsv
= 0;
2285 spin_lock(&root
->orphan_lock
);
2286 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2287 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2291 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2292 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2295 spin_unlock(&root
->orphan_lock
);
2297 if (trans
&& delete_item
) {
2298 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
2303 btrfs_orphan_release_metadata(inode
);
2309 * this cleans up any orphans that may be left on the list from the last use
2312 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2314 struct btrfs_path
*path
;
2315 struct extent_buffer
*leaf
;
2316 struct btrfs_key key
, found_key
;
2317 struct btrfs_trans_handle
*trans
;
2318 struct inode
*inode
;
2319 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2321 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2324 path
= btrfs_alloc_path();
2331 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2332 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2333 key
.offset
= (u64
)-1;
2336 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2341 * if ret == 0 means we found what we were searching for, which
2342 * is weird, but possible, so only screw with path if we didn't
2343 * find the key and see if we have stuff that matches
2347 if (path
->slots
[0] == 0)
2352 /* pull out the item */
2353 leaf
= path
->nodes
[0];
2354 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2356 /* make sure the item matches what we want */
2357 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2359 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2362 /* release the path since we're done with it */
2363 btrfs_release_path(path
);
2366 * this is where we are basically btrfs_lookup, without the
2367 * crossing root thing. we store the inode number in the
2368 * offset of the orphan item.
2370 found_key
.objectid
= found_key
.offset
;
2371 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2372 found_key
.offset
= 0;
2373 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2374 if (IS_ERR(inode
)) {
2375 ret
= PTR_ERR(inode
);
2380 * add this inode to the orphan list so btrfs_orphan_del does
2381 * the proper thing when we hit it
2383 spin_lock(&root
->orphan_lock
);
2384 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2385 spin_unlock(&root
->orphan_lock
);
2388 * if this is a bad inode, means we actually succeeded in
2389 * removing the inode, but not the orphan record, which means
2390 * we need to manually delete the orphan since iput will just
2391 * do a destroy_inode
2393 if (is_bad_inode(inode
)) {
2394 trans
= btrfs_start_transaction(root
, 0);
2395 if (IS_ERR(trans
)) {
2396 ret
= PTR_ERR(trans
);
2399 btrfs_orphan_del(trans
, inode
);
2400 btrfs_end_transaction(trans
, root
);
2405 /* if we have links, this was a truncate, lets do that */
2406 if (inode
->i_nlink
) {
2407 if (!S_ISREG(inode
->i_mode
)) {
2413 ret
= btrfs_truncate(inode
);
2418 /* this will do delete_inode and everything for us */
2423 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2425 if (root
->orphan_block_rsv
)
2426 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2429 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2430 trans
= btrfs_join_transaction(root
);
2432 btrfs_end_transaction(trans
, root
);
2436 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2438 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2442 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2443 btrfs_free_path(path
);
2448 * very simple check to peek ahead in the leaf looking for xattrs. If we
2449 * don't find any xattrs, we know there can't be any acls.
2451 * slot is the slot the inode is in, objectid is the objectid of the inode
2453 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2454 int slot
, u64 objectid
)
2456 u32 nritems
= btrfs_header_nritems(leaf
);
2457 struct btrfs_key found_key
;
2461 while (slot
< nritems
) {
2462 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2464 /* we found a different objectid, there must not be acls */
2465 if (found_key
.objectid
!= objectid
)
2468 /* we found an xattr, assume we've got an acl */
2469 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2473 * we found a key greater than an xattr key, there can't
2474 * be any acls later on
2476 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2483 * it goes inode, inode backrefs, xattrs, extents,
2484 * so if there are a ton of hard links to an inode there can
2485 * be a lot of backrefs. Don't waste time searching too hard,
2486 * this is just an optimization
2491 /* we hit the end of the leaf before we found an xattr or
2492 * something larger than an xattr. We have to assume the inode
2499 * read an inode from the btree into the in-memory inode
2501 static void btrfs_read_locked_inode(struct inode
*inode
)
2503 struct btrfs_path
*path
;
2504 struct extent_buffer
*leaf
;
2505 struct btrfs_inode_item
*inode_item
;
2506 struct btrfs_timespec
*tspec
;
2507 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2508 struct btrfs_key location
;
2512 bool filled
= false;
2514 ret
= btrfs_fill_inode(inode
, &rdev
);
2518 path
= btrfs_alloc_path();
2520 path
->leave_spinning
= 1;
2521 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2523 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2527 leaf
= path
->nodes
[0];
2532 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2533 struct btrfs_inode_item
);
2534 if (!leaf
->map_token
)
2535 map_private_extent_buffer(leaf
, (unsigned long)inode_item
,
2536 sizeof(struct btrfs_inode_item
),
2537 &leaf
->map_token
, &leaf
->kaddr
,
2538 &leaf
->map_start
, &leaf
->map_len
,
2541 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2542 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2543 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2544 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2545 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2547 tspec
= btrfs_inode_atime(inode_item
);
2548 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2549 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2551 tspec
= btrfs_inode_mtime(inode_item
);
2552 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2553 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2555 tspec
= btrfs_inode_ctime(inode_item
);
2556 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2557 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2559 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2560 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2561 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2562 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2564 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2566 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2567 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2570 * try to precache a NULL acl entry for files that don't have
2571 * any xattrs or acls
2573 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
2576 cache_no_acl(inode
);
2578 if (leaf
->map_token
) {
2579 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
2580 leaf
->map_token
= NULL
;
2583 btrfs_free_path(path
);
2585 switch (inode
->i_mode
& S_IFMT
) {
2587 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2588 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2589 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2590 inode
->i_fop
= &btrfs_file_operations
;
2591 inode
->i_op
= &btrfs_file_inode_operations
;
2594 inode
->i_fop
= &btrfs_dir_file_operations
;
2595 if (root
== root
->fs_info
->tree_root
)
2596 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2598 inode
->i_op
= &btrfs_dir_inode_operations
;
2601 inode
->i_op
= &btrfs_symlink_inode_operations
;
2602 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2603 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2606 inode
->i_op
= &btrfs_special_inode_operations
;
2607 init_special_inode(inode
, inode
->i_mode
, rdev
);
2611 btrfs_update_iflags(inode
);
2615 btrfs_free_path(path
);
2616 make_bad_inode(inode
);
2620 * given a leaf and an inode, copy the inode fields into the leaf
2622 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2623 struct extent_buffer
*leaf
,
2624 struct btrfs_inode_item
*item
,
2625 struct inode
*inode
)
2627 if (!leaf
->map_token
)
2628 map_private_extent_buffer(leaf
, (unsigned long)item
,
2629 sizeof(struct btrfs_inode_item
),
2630 &leaf
->map_token
, &leaf
->kaddr
,
2631 &leaf
->map_start
, &leaf
->map_len
,
2634 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2635 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2636 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2637 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2638 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2640 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2641 inode
->i_atime
.tv_sec
);
2642 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2643 inode
->i_atime
.tv_nsec
);
2645 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2646 inode
->i_mtime
.tv_sec
);
2647 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2648 inode
->i_mtime
.tv_nsec
);
2650 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2651 inode
->i_ctime
.tv_sec
);
2652 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2653 inode
->i_ctime
.tv_nsec
);
2655 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2656 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2657 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2658 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2659 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2660 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2661 btrfs_set_inode_block_group(leaf
, item
, 0);
2663 if (leaf
->map_token
) {
2664 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
2665 leaf
->map_token
= NULL
;
2670 * copy everything in the in-memory inode into the btree.
2672 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2673 struct btrfs_root
*root
, struct inode
*inode
)
2675 struct btrfs_inode_item
*inode_item
;
2676 struct btrfs_path
*path
;
2677 struct extent_buffer
*leaf
;
2681 * If the inode is a free space inode, we can deadlock during commit
2682 * if we put it into the delayed code.
2684 * The data relocation inode should also be directly updated
2687 if (!is_free_space_inode(root
, inode
)
2688 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
2689 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
2691 btrfs_set_inode_last_trans(trans
, inode
);
2695 path
= btrfs_alloc_path();
2699 path
->leave_spinning
= 1;
2700 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
2708 btrfs_unlock_up_safe(path
, 1);
2709 leaf
= path
->nodes
[0];
2710 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2711 struct btrfs_inode_item
);
2713 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2714 btrfs_mark_buffer_dirty(leaf
);
2715 btrfs_set_inode_last_trans(trans
, inode
);
2718 btrfs_free_path(path
);
2723 * unlink helper that gets used here in inode.c and in the tree logging
2724 * recovery code. It remove a link in a directory with a given name, and
2725 * also drops the back refs in the inode to the directory
2727 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2728 struct btrfs_root
*root
,
2729 struct inode
*dir
, struct inode
*inode
,
2730 const char *name
, int name_len
)
2732 struct btrfs_path
*path
;
2734 struct extent_buffer
*leaf
;
2735 struct btrfs_dir_item
*di
;
2736 struct btrfs_key key
;
2738 u64 ino
= btrfs_ino(inode
);
2739 u64 dir_ino
= btrfs_ino(dir
);
2741 path
= btrfs_alloc_path();
2747 path
->leave_spinning
= 1;
2748 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2749 name
, name_len
, -1);
2758 leaf
= path
->nodes
[0];
2759 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2760 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2763 btrfs_release_path(path
);
2765 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
2768 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2769 "inode %llu parent %llu\n", name_len
, name
,
2770 (unsigned long long)ino
, (unsigned long long)dir_ino
);
2774 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2778 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2780 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2782 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2787 btrfs_free_path(path
);
2791 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2792 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2793 btrfs_update_inode(trans
, root
, dir
);
2798 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2799 struct btrfs_root
*root
,
2800 struct inode
*dir
, struct inode
*inode
,
2801 const char *name
, int name_len
)
2804 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2806 btrfs_drop_nlink(inode
);
2807 ret
= btrfs_update_inode(trans
, root
, inode
);
2813 /* helper to check if there is any shared block in the path */
2814 static int check_path_shared(struct btrfs_root
*root
,
2815 struct btrfs_path
*path
)
2817 struct extent_buffer
*eb
;
2821 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2824 if (!path
->nodes
[level
])
2826 eb
= path
->nodes
[level
];
2827 if (!btrfs_block_can_be_shared(root
, eb
))
2829 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2838 * helper to start transaction for unlink and rmdir.
2840 * unlink and rmdir are special in btrfs, they do not always free space.
2841 * so in enospc case, we should make sure they will free space before
2842 * allowing them to use the global metadata reservation.
2844 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2845 struct dentry
*dentry
)
2847 struct btrfs_trans_handle
*trans
;
2848 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2849 struct btrfs_path
*path
;
2850 struct btrfs_inode_ref
*ref
;
2851 struct btrfs_dir_item
*di
;
2852 struct inode
*inode
= dentry
->d_inode
;
2857 u64 ino
= btrfs_ino(inode
);
2858 u64 dir_ino
= btrfs_ino(dir
);
2860 trans
= btrfs_start_transaction(root
, 10);
2861 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2864 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2865 return ERR_PTR(-ENOSPC
);
2867 /* check if there is someone else holds reference */
2868 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2869 return ERR_PTR(-ENOSPC
);
2871 if (atomic_read(&inode
->i_count
) > 2)
2872 return ERR_PTR(-ENOSPC
);
2874 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2875 return ERR_PTR(-ENOSPC
);
2877 path
= btrfs_alloc_path();
2879 root
->fs_info
->enospc_unlink
= 0;
2880 return ERR_PTR(-ENOMEM
);
2883 trans
= btrfs_start_transaction(root
, 0);
2884 if (IS_ERR(trans
)) {
2885 btrfs_free_path(path
);
2886 root
->fs_info
->enospc_unlink
= 0;
2890 path
->skip_locking
= 1;
2891 path
->search_commit_root
= 1;
2893 ret
= btrfs_lookup_inode(trans
, root
, path
,
2894 &BTRFS_I(dir
)->location
, 0);
2900 if (check_path_shared(root
, path
))
2905 btrfs_release_path(path
);
2907 ret
= btrfs_lookup_inode(trans
, root
, path
,
2908 &BTRFS_I(inode
)->location
, 0);
2914 if (check_path_shared(root
, path
))
2919 btrfs_release_path(path
);
2921 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2922 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2929 if (check_path_shared(root
, path
))
2931 btrfs_release_path(path
);
2939 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2940 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2946 if (check_path_shared(root
, path
))
2952 btrfs_release_path(path
);
2954 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2955 dentry
->d_name
.name
, dentry
->d_name
.len
,
2962 if (check_path_shared(root
, path
))
2964 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2965 btrfs_release_path(path
);
2968 * This is a commit root search, if we can lookup inode item and other
2969 * relative items in the commit root, it means the transaction of
2970 * dir/file creation has been committed, and the dir index item that we
2971 * delay to insert has also been inserted into the commit root. So
2972 * we needn't worry about the delayed insertion of the dir index item
2975 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
2976 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2981 BUG_ON(ret
== -ENOENT
);
2982 if (check_path_shared(root
, path
))
2987 btrfs_free_path(path
);
2989 btrfs_end_transaction(trans
, root
);
2990 root
->fs_info
->enospc_unlink
= 0;
2991 return ERR_PTR(err
);
2994 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2998 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2999 struct btrfs_root
*root
)
3001 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
3002 BUG_ON(!root
->fs_info
->enospc_unlink
);
3003 root
->fs_info
->enospc_unlink
= 0;
3005 btrfs_end_transaction_throttle(trans
, root
);
3008 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3010 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3011 struct btrfs_trans_handle
*trans
;
3012 struct inode
*inode
= dentry
->d_inode
;
3014 unsigned long nr
= 0;
3016 trans
= __unlink_start_trans(dir
, dentry
);
3018 return PTR_ERR(trans
);
3020 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
3022 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3023 dentry
->d_name
.name
, dentry
->d_name
.len
);
3026 if (inode
->i_nlink
== 0) {
3027 ret
= btrfs_orphan_add(trans
, inode
);
3031 nr
= trans
->blocks_used
;
3032 __unlink_end_trans(trans
, root
);
3033 btrfs_btree_balance_dirty(root
, nr
);
3037 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
3038 struct btrfs_root
*root
,
3039 struct inode
*dir
, u64 objectid
,
3040 const char *name
, int name_len
)
3042 struct btrfs_path
*path
;
3043 struct extent_buffer
*leaf
;
3044 struct btrfs_dir_item
*di
;
3045 struct btrfs_key key
;
3048 u64 dir_ino
= btrfs_ino(dir
);
3050 path
= btrfs_alloc_path();
3054 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3055 name
, name_len
, -1);
3056 BUG_ON(IS_ERR_OR_NULL(di
));
3058 leaf
= path
->nodes
[0];
3059 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3060 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3061 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3063 btrfs_release_path(path
);
3065 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3066 objectid
, root
->root_key
.objectid
,
3067 dir_ino
, &index
, name
, name_len
);
3069 BUG_ON(ret
!= -ENOENT
);
3070 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
3072 BUG_ON(IS_ERR_OR_NULL(di
));
3074 leaf
= path
->nodes
[0];
3075 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3076 btrfs_release_path(path
);
3079 btrfs_release_path(path
);
3081 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3084 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3085 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3086 ret
= btrfs_update_inode(trans
, root
, dir
);
3089 btrfs_free_path(path
);
3093 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3095 struct inode
*inode
= dentry
->d_inode
;
3097 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3098 struct btrfs_trans_handle
*trans
;
3099 unsigned long nr
= 0;
3101 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3102 btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3105 trans
= __unlink_start_trans(dir
, dentry
);
3107 return PTR_ERR(trans
);
3109 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3110 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3111 BTRFS_I(inode
)->location
.objectid
,
3112 dentry
->d_name
.name
,
3113 dentry
->d_name
.len
);
3117 err
= btrfs_orphan_add(trans
, inode
);
3121 /* now the directory is empty */
3122 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3123 dentry
->d_name
.name
, dentry
->d_name
.len
);
3125 btrfs_i_size_write(inode
, 0);
3127 nr
= trans
->blocks_used
;
3128 __unlink_end_trans(trans
, root
);
3129 btrfs_btree_balance_dirty(root
, nr
);
3135 * this can truncate away extent items, csum items and directory items.
3136 * It starts at a high offset and removes keys until it can't find
3137 * any higher than new_size
3139 * csum items that cross the new i_size are truncated to the new size
3142 * min_type is the minimum key type to truncate down to. If set to 0, this
3143 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3145 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3146 struct btrfs_root
*root
,
3147 struct inode
*inode
,
3148 u64 new_size
, u32 min_type
)
3150 struct btrfs_path
*path
;
3151 struct extent_buffer
*leaf
;
3152 struct btrfs_file_extent_item
*fi
;
3153 struct btrfs_key key
;
3154 struct btrfs_key found_key
;
3155 u64 extent_start
= 0;
3156 u64 extent_num_bytes
= 0;
3157 u64 extent_offset
= 0;
3159 u64 mask
= root
->sectorsize
- 1;
3160 u32 found_type
= (u8
)-1;
3163 int pending_del_nr
= 0;
3164 int pending_del_slot
= 0;
3165 int extent_type
= -1;
3169 u64 ino
= btrfs_ino(inode
);
3171 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3173 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3174 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3177 * This function is also used to drop the items in the log tree before
3178 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3179 * it is used to drop the loged items. So we shouldn't kill the delayed
3182 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3183 btrfs_kill_delayed_inode_items(inode
);
3185 path
= btrfs_alloc_path();
3190 key
.offset
= (u64
)-1;
3194 path
->leave_spinning
= 1;
3195 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3202 /* there are no items in the tree for us to truncate, we're
3205 if (path
->slots
[0] == 0)
3212 leaf
= path
->nodes
[0];
3213 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3214 found_type
= btrfs_key_type(&found_key
);
3217 if (found_key
.objectid
!= ino
)
3220 if (found_type
< min_type
)
3223 item_end
= found_key
.offset
;
3224 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3225 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3226 struct btrfs_file_extent_item
);
3227 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3228 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3229 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3230 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3232 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3234 btrfs_file_extent_num_bytes(leaf
, fi
);
3235 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3236 item_end
+= btrfs_file_extent_inline_len(leaf
,
3241 if (found_type
> min_type
) {
3244 if (item_end
< new_size
)
3246 if (found_key
.offset
>= new_size
)
3252 /* FIXME, shrink the extent if the ref count is only 1 */
3253 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3256 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3258 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3259 if (!del_item
&& !encoding
) {
3260 u64 orig_num_bytes
=
3261 btrfs_file_extent_num_bytes(leaf
, fi
);
3262 extent_num_bytes
= new_size
-
3263 found_key
.offset
+ root
->sectorsize
- 1;
3264 extent_num_bytes
= extent_num_bytes
&
3265 ~((u64
)root
->sectorsize
- 1);
3266 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3268 num_dec
= (orig_num_bytes
-
3270 if (root
->ref_cows
&& extent_start
!= 0)
3271 inode_sub_bytes(inode
, num_dec
);
3272 btrfs_mark_buffer_dirty(leaf
);
3275 btrfs_file_extent_disk_num_bytes(leaf
,
3277 extent_offset
= found_key
.offset
-
3278 btrfs_file_extent_offset(leaf
, fi
);
3280 /* FIXME blocksize != 4096 */
3281 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3282 if (extent_start
!= 0) {
3285 inode_sub_bytes(inode
, num_dec
);
3288 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3290 * we can't truncate inline items that have had
3294 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3295 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3296 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3297 u32 size
= new_size
- found_key
.offset
;
3299 if (root
->ref_cows
) {
3300 inode_sub_bytes(inode
, item_end
+ 1 -
3304 btrfs_file_extent_calc_inline_size(size
);
3305 ret
= btrfs_truncate_item(trans
, root
, path
,
3307 } else if (root
->ref_cows
) {
3308 inode_sub_bytes(inode
, item_end
+ 1 -
3314 if (!pending_del_nr
) {
3315 /* no pending yet, add ourselves */
3316 pending_del_slot
= path
->slots
[0];
3318 } else if (pending_del_nr
&&
3319 path
->slots
[0] + 1 == pending_del_slot
) {
3320 /* hop on the pending chunk */
3322 pending_del_slot
= path
->slots
[0];
3329 if (found_extent
&& (root
->ref_cows
||
3330 root
== root
->fs_info
->tree_root
)) {
3331 btrfs_set_path_blocking(path
);
3332 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3333 extent_num_bytes
, 0,
3334 btrfs_header_owner(leaf
),
3335 ino
, extent_offset
);
3339 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3342 if (path
->slots
[0] == 0 ||
3343 path
->slots
[0] != pending_del_slot
) {
3344 if (root
->ref_cows
&&
3345 BTRFS_I(inode
)->location
.objectid
!=
3346 BTRFS_FREE_INO_OBJECTID
) {
3350 if (pending_del_nr
) {
3351 ret
= btrfs_del_items(trans
, root
, path
,
3357 btrfs_release_path(path
);
3364 if (pending_del_nr
) {
3365 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3369 btrfs_free_path(path
);
3374 * taken from block_truncate_page, but does cow as it zeros out
3375 * any bytes left in the last page in the file.
3377 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3379 struct inode
*inode
= mapping
->host
;
3380 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3381 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3382 struct btrfs_ordered_extent
*ordered
;
3383 struct extent_state
*cached_state
= NULL
;
3385 u32 blocksize
= root
->sectorsize
;
3386 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3387 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3393 if ((offset
& (blocksize
- 1)) == 0)
3395 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3401 page
= grab_cache_page(mapping
, index
);
3403 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3407 page_start
= page_offset(page
);
3408 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3410 if (!PageUptodate(page
)) {
3411 ret
= btrfs_readpage(NULL
, page
);
3413 if (page
->mapping
!= mapping
) {
3415 page_cache_release(page
);
3418 if (!PageUptodate(page
)) {
3423 wait_on_page_writeback(page
);
3425 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3427 set_page_extent_mapped(page
);
3429 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3431 unlock_extent_cached(io_tree
, page_start
, page_end
,
3432 &cached_state
, GFP_NOFS
);
3434 page_cache_release(page
);
3435 btrfs_start_ordered_extent(inode
, ordered
, 1);
3436 btrfs_put_ordered_extent(ordered
);
3440 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3441 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3442 0, 0, &cached_state
, GFP_NOFS
);
3444 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3447 unlock_extent_cached(io_tree
, page_start
, page_end
,
3448 &cached_state
, GFP_NOFS
);
3453 if (offset
!= PAGE_CACHE_SIZE
) {
3455 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3456 flush_dcache_page(page
);
3459 ClearPageChecked(page
);
3460 set_page_dirty(page
);
3461 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3466 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3468 page_cache_release(page
);
3474 * This function puts in dummy file extents for the area we're creating a hole
3475 * for. So if we are truncating this file to a larger size we need to insert
3476 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3477 * the range between oldsize and size
3479 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3481 struct btrfs_trans_handle
*trans
;
3482 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3483 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3484 struct extent_map
*em
= NULL
;
3485 struct extent_state
*cached_state
= NULL
;
3486 u64 mask
= root
->sectorsize
- 1;
3487 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3488 u64 block_end
= (size
+ mask
) & ~mask
;
3494 if (size
<= hole_start
)
3498 struct btrfs_ordered_extent
*ordered
;
3499 btrfs_wait_ordered_range(inode
, hole_start
,
3500 block_end
- hole_start
);
3501 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3502 &cached_state
, GFP_NOFS
);
3503 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3506 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3507 &cached_state
, GFP_NOFS
);
3508 btrfs_put_ordered_extent(ordered
);
3511 cur_offset
= hole_start
;
3513 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3514 block_end
- cur_offset
, 0);
3515 BUG_ON(IS_ERR_OR_NULL(em
));
3516 last_byte
= min(extent_map_end(em
), block_end
);
3517 last_byte
= (last_byte
+ mask
) & ~mask
;
3518 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3520 hole_size
= last_byte
- cur_offset
;
3522 trans
= btrfs_start_transaction(root
, 2);
3523 if (IS_ERR(trans
)) {
3524 err
= PTR_ERR(trans
);
3528 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3529 cur_offset
+ hole_size
,
3534 err
= btrfs_insert_file_extent(trans
, root
,
3535 btrfs_ino(inode
), cur_offset
, 0,
3536 0, hole_size
, 0, hole_size
,
3541 btrfs_drop_extent_cache(inode
, hole_start
,
3544 btrfs_end_transaction(trans
, root
);
3546 free_extent_map(em
);
3548 cur_offset
= last_byte
;
3549 if (cur_offset
>= block_end
)
3553 free_extent_map(em
);
3554 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3559 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3561 loff_t oldsize
= i_size_read(inode
);
3564 if (newsize
== oldsize
)
3567 if (newsize
> oldsize
) {
3568 i_size_write(inode
, newsize
);
3569 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3570 truncate_pagecache(inode
, oldsize
, newsize
);
3571 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3573 btrfs_setsize(inode
, oldsize
);
3577 mark_inode_dirty(inode
);
3581 * We're truncating a file that used to have good data down to
3582 * zero. Make sure it gets into the ordered flush list so that
3583 * any new writes get down to disk quickly.
3586 BTRFS_I(inode
)->ordered_data_close
= 1;
3588 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3589 truncate_setsize(inode
, newsize
);
3590 ret
= btrfs_truncate(inode
);
3596 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3598 struct inode
*inode
= dentry
->d_inode
;
3599 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3602 if (btrfs_root_readonly(root
))
3605 err
= inode_change_ok(inode
, attr
);
3609 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3610 err
= btrfs_setsize(inode
, attr
->ia_size
);
3615 if (attr
->ia_valid
) {
3616 setattr_copy(inode
, attr
);
3617 mark_inode_dirty(inode
);
3619 if (attr
->ia_valid
& ATTR_MODE
)
3620 err
= btrfs_acl_chmod(inode
);
3626 void btrfs_evict_inode(struct inode
*inode
)
3628 struct btrfs_trans_handle
*trans
;
3629 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3633 trace_btrfs_inode_evict(inode
);
3635 truncate_inode_pages(&inode
->i_data
, 0);
3636 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3637 is_free_space_inode(root
, inode
)))
3640 if (is_bad_inode(inode
)) {
3641 btrfs_orphan_del(NULL
, inode
);
3644 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3645 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3647 if (root
->fs_info
->log_root_recovering
) {
3648 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3652 if (inode
->i_nlink
> 0) {
3653 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3657 btrfs_i_size_write(inode
, 0);
3660 trans
= btrfs_join_transaction(root
);
3661 BUG_ON(IS_ERR(trans
));
3662 trans
->block_rsv
= root
->orphan_block_rsv
;
3664 ret
= btrfs_block_rsv_check(trans
, root
,
3665 root
->orphan_block_rsv
, 0, 5);
3667 BUG_ON(ret
!= -EAGAIN
);
3668 ret
= btrfs_commit_transaction(trans
, root
);
3673 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3677 nr
= trans
->blocks_used
;
3678 btrfs_end_transaction(trans
, root
);
3680 btrfs_btree_balance_dirty(root
, nr
);
3685 ret
= btrfs_orphan_del(trans
, inode
);
3689 if (!(root
== root
->fs_info
->tree_root
||
3690 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
3691 btrfs_return_ino(root
, btrfs_ino(inode
));
3693 nr
= trans
->blocks_used
;
3694 btrfs_end_transaction(trans
, root
);
3695 btrfs_btree_balance_dirty(root
, nr
);
3697 end_writeback(inode
);
3702 * this returns the key found in the dir entry in the location pointer.
3703 * If no dir entries were found, location->objectid is 0.
3705 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3706 struct btrfs_key
*location
)
3708 const char *name
= dentry
->d_name
.name
;
3709 int namelen
= dentry
->d_name
.len
;
3710 struct btrfs_dir_item
*di
;
3711 struct btrfs_path
*path
;
3712 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3715 path
= btrfs_alloc_path();
3718 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
3723 if (IS_ERR_OR_NULL(di
))
3726 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3728 btrfs_free_path(path
);
3731 location
->objectid
= 0;
3736 * when we hit a tree root in a directory, the btrfs part of the inode
3737 * needs to be changed to reflect the root directory of the tree root. This
3738 * is kind of like crossing a mount point.
3740 static int fixup_tree_root_location(struct btrfs_root
*root
,
3742 struct dentry
*dentry
,
3743 struct btrfs_key
*location
,
3744 struct btrfs_root
**sub_root
)
3746 struct btrfs_path
*path
;
3747 struct btrfs_root
*new_root
;
3748 struct btrfs_root_ref
*ref
;
3749 struct extent_buffer
*leaf
;
3753 path
= btrfs_alloc_path();
3760 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3761 BTRFS_I(dir
)->root
->root_key
.objectid
,
3762 location
->objectid
);
3769 leaf
= path
->nodes
[0];
3770 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3771 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
3772 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3775 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3776 (unsigned long)(ref
+ 1),
3777 dentry
->d_name
.len
);
3781 btrfs_release_path(path
);
3783 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3784 if (IS_ERR(new_root
)) {
3785 err
= PTR_ERR(new_root
);
3789 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3794 *sub_root
= new_root
;
3795 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3796 location
->type
= BTRFS_INODE_ITEM_KEY
;
3797 location
->offset
= 0;
3800 btrfs_free_path(path
);
3804 static void inode_tree_add(struct inode
*inode
)
3806 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3807 struct btrfs_inode
*entry
;
3809 struct rb_node
*parent
;
3810 u64 ino
= btrfs_ino(inode
);
3812 p
= &root
->inode_tree
.rb_node
;
3815 if (inode_unhashed(inode
))
3818 spin_lock(&root
->inode_lock
);
3821 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3823 if (ino
< btrfs_ino(&entry
->vfs_inode
))
3824 p
= &parent
->rb_left
;
3825 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
3826 p
= &parent
->rb_right
;
3828 WARN_ON(!(entry
->vfs_inode
.i_state
&
3829 (I_WILL_FREE
| I_FREEING
)));
3830 rb_erase(parent
, &root
->inode_tree
);
3831 RB_CLEAR_NODE(parent
);
3832 spin_unlock(&root
->inode_lock
);
3836 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3837 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3838 spin_unlock(&root
->inode_lock
);
3841 static void inode_tree_del(struct inode
*inode
)
3843 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3846 spin_lock(&root
->inode_lock
);
3847 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3848 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3849 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3850 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3852 spin_unlock(&root
->inode_lock
);
3855 * Free space cache has inodes in the tree root, but the tree root has a
3856 * root_refs of 0, so this could end up dropping the tree root as a
3857 * snapshot, so we need the extra !root->fs_info->tree_root check to
3858 * make sure we don't drop it.
3860 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3861 root
!= root
->fs_info
->tree_root
) {
3862 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3863 spin_lock(&root
->inode_lock
);
3864 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3865 spin_unlock(&root
->inode_lock
);
3867 btrfs_add_dead_root(root
);
3871 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3873 struct rb_node
*node
;
3874 struct rb_node
*prev
;
3875 struct btrfs_inode
*entry
;
3876 struct inode
*inode
;
3879 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3881 spin_lock(&root
->inode_lock
);
3883 node
= root
->inode_tree
.rb_node
;
3887 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3889 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
3890 node
= node
->rb_left
;
3891 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
3892 node
= node
->rb_right
;
3898 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
3899 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
3903 prev
= rb_next(prev
);
3907 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3908 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
3909 inode
= igrab(&entry
->vfs_inode
);
3911 spin_unlock(&root
->inode_lock
);
3912 if (atomic_read(&inode
->i_count
) > 1)
3913 d_prune_aliases(inode
);
3915 * btrfs_drop_inode will have it removed from
3916 * the inode cache when its usage count
3921 spin_lock(&root
->inode_lock
);
3925 if (cond_resched_lock(&root
->inode_lock
))
3928 node
= rb_next(node
);
3930 spin_unlock(&root
->inode_lock
);
3934 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3936 struct btrfs_iget_args
*args
= p
;
3937 inode
->i_ino
= args
->ino
;
3938 BTRFS_I(inode
)->root
= args
->root
;
3939 btrfs_set_inode_space_info(args
->root
, inode
);
3943 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3945 struct btrfs_iget_args
*args
= opaque
;
3946 return args
->ino
== btrfs_ino(inode
) &&
3947 args
->root
== BTRFS_I(inode
)->root
;
3950 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
3952 struct btrfs_root
*root
)
3954 struct inode
*inode
;
3955 struct btrfs_iget_args args
;
3956 args
.ino
= objectid
;
3959 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3960 btrfs_init_locked_inode
,
3965 /* Get an inode object given its location and corresponding root.
3966 * Returns in *is_new if the inode was read from disk
3968 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3969 struct btrfs_root
*root
, int *new)
3971 struct inode
*inode
;
3973 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3975 return ERR_PTR(-ENOMEM
);
3977 if (inode
->i_state
& I_NEW
) {
3978 BTRFS_I(inode
)->root
= root
;
3979 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3980 btrfs_read_locked_inode(inode
);
3981 inode_tree_add(inode
);
3982 unlock_new_inode(inode
);
3990 static struct inode
*new_simple_dir(struct super_block
*s
,
3991 struct btrfs_key
*key
,
3992 struct btrfs_root
*root
)
3994 struct inode
*inode
= new_inode(s
);
3997 return ERR_PTR(-ENOMEM
);
3999 BTRFS_I(inode
)->root
= root
;
4000 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4001 BTRFS_I(inode
)->dummy_inode
= 1;
4003 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4004 inode
->i_op
= &simple_dir_inode_operations
;
4005 inode
->i_fop
= &simple_dir_operations
;
4006 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4007 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4012 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4014 struct inode
*inode
;
4015 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4016 struct btrfs_root
*sub_root
= root
;
4017 struct btrfs_key location
;
4021 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4022 return ERR_PTR(-ENAMETOOLONG
);
4024 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4027 return ERR_PTR(ret
);
4029 if (location
.objectid
== 0)
4032 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4033 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4037 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4039 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4040 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4041 &location
, &sub_root
);
4044 inode
= ERR_PTR(ret
);
4046 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4048 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4050 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4052 if (!IS_ERR(inode
) && root
!= sub_root
) {
4053 down_read(&root
->fs_info
->cleanup_work_sem
);
4054 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4055 ret
= btrfs_orphan_cleanup(sub_root
);
4056 up_read(&root
->fs_info
->cleanup_work_sem
);
4058 inode
= ERR_PTR(ret
);
4064 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4066 struct btrfs_root
*root
;
4068 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
4069 dentry
= dentry
->d_parent
;
4071 if (dentry
->d_inode
) {
4072 root
= BTRFS_I(dentry
->d_inode
)->root
;
4073 if (btrfs_root_refs(&root
->root_item
) == 0)
4079 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4080 struct nameidata
*nd
)
4082 struct inode
*inode
;
4084 inode
= btrfs_lookup_dentry(dir
, dentry
);
4086 return ERR_CAST(inode
);
4088 return d_splice_alias(inode
, dentry
);
4091 unsigned char btrfs_filetype_table
[] = {
4092 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4095 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4098 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4099 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4100 struct btrfs_item
*item
;
4101 struct btrfs_dir_item
*di
;
4102 struct btrfs_key key
;
4103 struct btrfs_key found_key
;
4104 struct btrfs_path
*path
;
4105 struct list_head ins_list
;
4106 struct list_head del_list
;
4108 struct extent_buffer
*leaf
;
4110 unsigned char d_type
;
4115 int key_type
= BTRFS_DIR_INDEX_KEY
;
4119 int is_curr
= 0; /* filp->f_pos points to the current index? */
4121 /* FIXME, use a real flag for deciding about the key type */
4122 if (root
->fs_info
->tree_root
== root
)
4123 key_type
= BTRFS_DIR_ITEM_KEY
;
4125 /* special case for "." */
4126 if (filp
->f_pos
== 0) {
4127 over
= filldir(dirent
, ".", 1, 1, btrfs_ino(inode
), DT_DIR
);
4132 /* special case for .., just use the back ref */
4133 if (filp
->f_pos
== 1) {
4134 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4135 over
= filldir(dirent
, "..", 2,
4141 path
= btrfs_alloc_path();
4147 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4148 INIT_LIST_HEAD(&ins_list
);
4149 INIT_LIST_HEAD(&del_list
);
4150 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
4153 btrfs_set_key_type(&key
, key_type
);
4154 key
.offset
= filp
->f_pos
;
4155 key
.objectid
= btrfs_ino(inode
);
4157 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4162 leaf
= path
->nodes
[0];
4163 slot
= path
->slots
[0];
4164 if (slot
>= btrfs_header_nritems(leaf
)) {
4165 ret
= btrfs_next_leaf(root
, path
);
4173 item
= btrfs_item_nr(leaf
, slot
);
4174 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4176 if (found_key
.objectid
!= key
.objectid
)
4178 if (btrfs_key_type(&found_key
) != key_type
)
4180 if (found_key
.offset
< filp
->f_pos
)
4182 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
4183 btrfs_should_delete_dir_index(&del_list
,
4187 filp
->f_pos
= found_key
.offset
;
4190 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4192 di_total
= btrfs_item_size(leaf
, item
);
4194 while (di_cur
< di_total
) {
4195 struct btrfs_key location
;
4197 if (verify_dir_item(root
, leaf
, di
))
4200 name_len
= btrfs_dir_name_len(leaf
, di
);
4201 if (name_len
<= sizeof(tmp_name
)) {
4202 name_ptr
= tmp_name
;
4204 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4210 read_extent_buffer(leaf
, name_ptr
,
4211 (unsigned long)(di
+ 1), name_len
);
4213 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4214 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4216 /* is this a reference to our own snapshot? If so
4219 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4220 location
.objectid
== root
->root_key
.objectid
) {
4224 over
= filldir(dirent
, name_ptr
, name_len
,
4225 found_key
.offset
, location
.objectid
,
4229 if (name_ptr
!= tmp_name
)
4234 di_len
= btrfs_dir_name_len(leaf
, di
) +
4235 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4237 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4243 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4246 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
4252 /* Reached end of directory/root. Bump pos past the last item. */
4253 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4255 * 32-bit glibc will use getdents64, but then strtol -
4256 * so the last number we can serve is this.
4258 filp
->f_pos
= 0x7fffffff;
4264 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4265 btrfs_put_delayed_items(&ins_list
, &del_list
);
4266 btrfs_free_path(path
);
4270 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4272 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4273 struct btrfs_trans_handle
*trans
;
4275 bool nolock
= false;
4277 if (BTRFS_I(inode
)->dummy_inode
)
4280 if (btrfs_fs_closing(root
->fs_info
) && is_free_space_inode(root
, inode
))
4283 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4285 trans
= btrfs_join_transaction_nolock(root
);
4287 trans
= btrfs_join_transaction(root
);
4289 return PTR_ERR(trans
);
4291 ret
= btrfs_end_transaction_nolock(trans
, root
);
4293 ret
= btrfs_commit_transaction(trans
, root
);
4299 * This is somewhat expensive, updating the tree every time the
4300 * inode changes. But, it is most likely to find the inode in cache.
4301 * FIXME, needs more benchmarking...there are no reasons other than performance
4302 * to keep or drop this code.
4304 void btrfs_dirty_inode(struct inode
*inode
, int flags
)
4306 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4307 struct btrfs_trans_handle
*trans
;
4310 if (BTRFS_I(inode
)->dummy_inode
)
4313 trans
= btrfs_join_transaction(root
);
4314 BUG_ON(IS_ERR(trans
));
4316 ret
= btrfs_update_inode(trans
, root
, inode
);
4317 if (ret
&& ret
== -ENOSPC
) {
4318 /* whoops, lets try again with the full transaction */
4319 btrfs_end_transaction(trans
, root
);
4320 trans
= btrfs_start_transaction(root
, 1);
4321 if (IS_ERR(trans
)) {
4322 printk_ratelimited(KERN_ERR
"btrfs: fail to "
4323 "dirty inode %llu error %ld\n",
4324 (unsigned long long)btrfs_ino(inode
),
4329 ret
= btrfs_update_inode(trans
, root
, inode
);
4331 printk_ratelimited(KERN_ERR
"btrfs: fail to "
4332 "dirty inode %llu error %d\n",
4333 (unsigned long long)btrfs_ino(inode
),
4337 btrfs_end_transaction(trans
, root
);
4338 if (BTRFS_I(inode
)->delayed_node
)
4339 btrfs_balance_delayed_items(root
);
4343 * find the highest existing sequence number in a directory
4344 * and then set the in-memory index_cnt variable to reflect
4345 * free sequence numbers
4347 static int btrfs_set_inode_index_count(struct inode
*inode
)
4349 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4350 struct btrfs_key key
, found_key
;
4351 struct btrfs_path
*path
;
4352 struct extent_buffer
*leaf
;
4355 key
.objectid
= btrfs_ino(inode
);
4356 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4357 key
.offset
= (u64
)-1;
4359 path
= btrfs_alloc_path();
4363 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4366 /* FIXME: we should be able to handle this */
4372 * MAGIC NUMBER EXPLANATION:
4373 * since we search a directory based on f_pos we have to start at 2
4374 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4375 * else has to start at 2
4377 if (path
->slots
[0] == 0) {
4378 BTRFS_I(inode
)->index_cnt
= 2;
4384 leaf
= path
->nodes
[0];
4385 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4387 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4388 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4389 BTRFS_I(inode
)->index_cnt
= 2;
4393 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4395 btrfs_free_path(path
);
4400 * helper to find a free sequence number in a given directory. This current
4401 * code is very simple, later versions will do smarter things in the btree
4403 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4407 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4408 ret
= btrfs_inode_delayed_dir_index_count(dir
);
4410 ret
= btrfs_set_inode_index_count(dir
);
4416 *index
= BTRFS_I(dir
)->index_cnt
;
4417 BTRFS_I(dir
)->index_cnt
++;
4422 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4423 struct btrfs_root
*root
,
4425 const char *name
, int name_len
,
4426 u64 ref_objectid
, u64 objectid
, int mode
,
4429 struct inode
*inode
;
4430 struct btrfs_inode_item
*inode_item
;
4431 struct btrfs_key
*location
;
4432 struct btrfs_path
*path
;
4433 struct btrfs_inode_ref
*ref
;
4434 struct btrfs_key key
[2];
4440 path
= btrfs_alloc_path();
4443 inode
= new_inode(root
->fs_info
->sb
);
4445 btrfs_free_path(path
);
4446 return ERR_PTR(-ENOMEM
);
4450 * we have to initialize this early, so we can reclaim the inode
4451 * number if we fail afterwards in this function.
4453 inode
->i_ino
= objectid
;
4456 trace_btrfs_inode_request(dir
);
4458 ret
= btrfs_set_inode_index(dir
, index
);
4460 btrfs_free_path(path
);
4462 return ERR_PTR(ret
);
4466 * index_cnt is ignored for everything but a dir,
4467 * btrfs_get_inode_index_count has an explanation for the magic
4470 BTRFS_I(inode
)->index_cnt
= 2;
4471 BTRFS_I(inode
)->root
= root
;
4472 BTRFS_I(inode
)->generation
= trans
->transid
;
4473 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4474 btrfs_set_inode_space_info(root
, inode
);
4481 key
[0].objectid
= objectid
;
4482 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4485 key
[1].objectid
= objectid
;
4486 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4487 key
[1].offset
= ref_objectid
;
4489 sizes
[0] = sizeof(struct btrfs_inode_item
);
4490 sizes
[1] = name_len
+ sizeof(*ref
);
4492 path
->leave_spinning
= 1;
4493 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4497 inode_init_owner(inode
, dir
, mode
);
4498 inode_set_bytes(inode
, 0);
4499 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4500 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4501 struct btrfs_inode_item
);
4502 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4504 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4505 struct btrfs_inode_ref
);
4506 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4507 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4508 ptr
= (unsigned long)(ref
+ 1);
4509 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4511 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4512 btrfs_free_path(path
);
4514 location
= &BTRFS_I(inode
)->location
;
4515 location
->objectid
= objectid
;
4516 location
->offset
= 0;
4517 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4519 btrfs_inherit_iflags(inode
, dir
);
4521 if ((mode
& S_IFREG
)) {
4522 if (btrfs_test_opt(root
, NODATASUM
))
4523 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4524 if (btrfs_test_opt(root
, NODATACOW
) ||
4525 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4526 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4529 insert_inode_hash(inode
);
4530 inode_tree_add(inode
);
4532 trace_btrfs_inode_new(inode
);
4533 btrfs_set_inode_last_trans(trans
, inode
);
4538 BTRFS_I(dir
)->index_cnt
--;
4539 btrfs_free_path(path
);
4541 return ERR_PTR(ret
);
4544 static inline u8
btrfs_inode_type(struct inode
*inode
)
4546 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4550 * utility function to add 'inode' into 'parent_inode' with
4551 * a give name and a given sequence number.
4552 * if 'add_backref' is true, also insert a backref from the
4553 * inode to the parent directory.
4555 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4556 struct inode
*parent_inode
, struct inode
*inode
,
4557 const char *name
, int name_len
, int add_backref
, u64 index
)
4560 struct btrfs_key key
;
4561 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4562 u64 ino
= btrfs_ino(inode
);
4563 u64 parent_ino
= btrfs_ino(parent_inode
);
4565 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4566 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4569 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4573 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4574 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4575 key
.objectid
, root
->root_key
.objectid
,
4576 parent_ino
, index
, name
, name_len
);
4577 } else if (add_backref
) {
4578 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
4583 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4585 btrfs_inode_type(inode
), index
);
4588 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4590 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4591 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4596 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4597 struct inode
*dir
, struct dentry
*dentry
,
4598 struct inode
*inode
, int backref
, u64 index
)
4600 int err
= btrfs_add_link(trans
, dir
, inode
,
4601 dentry
->d_name
.name
, dentry
->d_name
.len
,
4604 d_instantiate(dentry
, inode
);
4612 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4613 int mode
, dev_t rdev
)
4615 struct btrfs_trans_handle
*trans
;
4616 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4617 struct inode
*inode
= NULL
;
4621 unsigned long nr
= 0;
4624 if (!new_valid_dev(rdev
))
4628 * 2 for inode item and ref
4630 * 1 for xattr if selinux is on
4632 trans
= btrfs_start_transaction(root
, 5);
4634 return PTR_ERR(trans
);
4636 err
= btrfs_find_free_ino(root
, &objectid
);
4640 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4641 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4643 if (IS_ERR(inode
)) {
4644 err
= PTR_ERR(inode
);
4648 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4654 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4658 inode
->i_op
= &btrfs_special_inode_operations
;
4659 init_special_inode(inode
, inode
->i_mode
, rdev
);
4660 btrfs_update_inode(trans
, root
, inode
);
4663 nr
= trans
->blocks_used
;
4664 btrfs_end_transaction_throttle(trans
, root
);
4665 btrfs_btree_balance_dirty(root
, nr
);
4667 inode_dec_link_count(inode
);
4673 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4674 int mode
, struct nameidata
*nd
)
4676 struct btrfs_trans_handle
*trans
;
4677 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4678 struct inode
*inode
= NULL
;
4681 unsigned long nr
= 0;
4686 * 2 for inode item and ref
4688 * 1 for xattr if selinux is on
4690 trans
= btrfs_start_transaction(root
, 5);
4692 return PTR_ERR(trans
);
4694 err
= btrfs_find_free_ino(root
, &objectid
);
4698 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4699 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4701 if (IS_ERR(inode
)) {
4702 err
= PTR_ERR(inode
);
4706 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4712 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4716 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4717 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4718 inode
->i_fop
= &btrfs_file_operations
;
4719 inode
->i_op
= &btrfs_file_inode_operations
;
4720 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4723 nr
= trans
->blocks_used
;
4724 btrfs_end_transaction_throttle(trans
, root
);
4726 inode_dec_link_count(inode
);
4729 btrfs_btree_balance_dirty(root
, nr
);
4733 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4734 struct dentry
*dentry
)
4736 struct btrfs_trans_handle
*trans
;
4737 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4738 struct inode
*inode
= old_dentry
->d_inode
;
4740 unsigned long nr
= 0;
4744 /* do not allow sys_link's with other subvols of the same device */
4745 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4748 if (inode
->i_nlink
== ~0U)
4751 err
= btrfs_set_inode_index(dir
, &index
);
4756 * 2 items for inode and inode ref
4757 * 2 items for dir items
4758 * 1 item for parent inode
4760 trans
= btrfs_start_transaction(root
, 5);
4761 if (IS_ERR(trans
)) {
4762 err
= PTR_ERR(trans
);
4766 btrfs_inc_nlink(inode
);
4767 inode
->i_ctime
= CURRENT_TIME
;
4770 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4775 struct dentry
*parent
= dget_parent(dentry
);
4776 err
= btrfs_update_inode(trans
, root
, inode
);
4778 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4782 nr
= trans
->blocks_used
;
4783 btrfs_end_transaction_throttle(trans
, root
);
4786 inode_dec_link_count(inode
);
4789 btrfs_btree_balance_dirty(root
, nr
);
4793 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4795 struct inode
*inode
= NULL
;
4796 struct btrfs_trans_handle
*trans
;
4797 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4799 int drop_on_err
= 0;
4802 unsigned long nr
= 1;
4805 * 2 items for inode and ref
4806 * 2 items for dir items
4807 * 1 for xattr if selinux is on
4809 trans
= btrfs_start_transaction(root
, 5);
4811 return PTR_ERR(trans
);
4813 err
= btrfs_find_free_ino(root
, &objectid
);
4817 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4818 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4819 S_IFDIR
| mode
, &index
);
4820 if (IS_ERR(inode
)) {
4821 err
= PTR_ERR(inode
);
4827 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4831 inode
->i_op
= &btrfs_dir_inode_operations
;
4832 inode
->i_fop
= &btrfs_dir_file_operations
;
4834 btrfs_i_size_write(inode
, 0);
4835 err
= btrfs_update_inode(trans
, root
, inode
);
4839 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4840 dentry
->d_name
.len
, 0, index
);
4844 d_instantiate(dentry
, inode
);
4848 nr
= trans
->blocks_used
;
4849 btrfs_end_transaction_throttle(trans
, root
);
4852 btrfs_btree_balance_dirty(root
, nr
);
4856 /* helper for btfs_get_extent. Given an existing extent in the tree,
4857 * and an extent that you want to insert, deal with overlap and insert
4858 * the new extent into the tree.
4860 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4861 struct extent_map
*existing
,
4862 struct extent_map
*em
,
4863 u64 map_start
, u64 map_len
)
4867 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4868 start_diff
= map_start
- em
->start
;
4869 em
->start
= map_start
;
4871 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4872 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4873 em
->block_start
+= start_diff
;
4874 em
->block_len
-= start_diff
;
4876 return add_extent_mapping(em_tree
, em
);
4879 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4880 struct inode
*inode
, struct page
*page
,
4881 size_t pg_offset
, u64 extent_offset
,
4882 struct btrfs_file_extent_item
*item
)
4885 struct extent_buffer
*leaf
= path
->nodes
[0];
4888 unsigned long inline_size
;
4892 WARN_ON(pg_offset
!= 0);
4893 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4894 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4895 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4896 btrfs_item_nr(leaf
, path
->slots
[0]));
4897 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4900 ptr
= btrfs_file_extent_inline_start(item
);
4902 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4904 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4905 ret
= btrfs_decompress(compress_type
, tmp
, page
,
4906 extent_offset
, inline_size
, max_size
);
4908 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4909 unsigned long copy_size
= min_t(u64
,
4910 PAGE_CACHE_SIZE
- pg_offset
,
4911 max_size
- extent_offset
);
4912 memset(kaddr
+ pg_offset
, 0, copy_size
);
4913 kunmap_atomic(kaddr
, KM_USER0
);
4920 * a bit scary, this does extent mapping from logical file offset to the disk.
4921 * the ugly parts come from merging extents from the disk with the in-ram
4922 * representation. This gets more complex because of the data=ordered code,
4923 * where the in-ram extents might be locked pending data=ordered completion.
4925 * This also copies inline extents directly into the page.
4928 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
4929 size_t pg_offset
, u64 start
, u64 len
,
4935 u64 extent_start
= 0;
4937 u64 objectid
= btrfs_ino(inode
);
4939 struct btrfs_path
*path
= NULL
;
4940 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4941 struct btrfs_file_extent_item
*item
;
4942 struct extent_buffer
*leaf
;
4943 struct btrfs_key found_key
;
4944 struct extent_map
*em
= NULL
;
4945 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4946 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4947 struct btrfs_trans_handle
*trans
= NULL
;
4951 read_lock(&em_tree
->lock
);
4952 em
= lookup_extent_mapping(em_tree
, start
, len
);
4954 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4955 read_unlock(&em_tree
->lock
);
4958 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
4959 free_extent_map(em
);
4960 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
4961 free_extent_map(em
);
4965 em
= alloc_extent_map();
4970 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4971 em
->start
= EXTENT_MAP_HOLE
;
4972 em
->orig_start
= EXTENT_MAP_HOLE
;
4974 em
->block_len
= (u64
)-1;
4977 path
= btrfs_alloc_path();
4983 * Chances are we'll be called again, so go ahead and do
4989 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
4990 objectid
, start
, trans
!= NULL
);
4997 if (path
->slots
[0] == 0)
5002 leaf
= path
->nodes
[0];
5003 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5004 struct btrfs_file_extent_item
);
5005 /* are we inside the extent that was found? */
5006 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5007 found_type
= btrfs_key_type(&found_key
);
5008 if (found_key
.objectid
!= objectid
||
5009 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5013 found_type
= btrfs_file_extent_type(leaf
, item
);
5014 extent_start
= found_key
.offset
;
5015 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5016 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5017 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5018 extent_end
= extent_start
+
5019 btrfs_file_extent_num_bytes(leaf
, item
);
5020 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5022 size
= btrfs_file_extent_inline_len(leaf
, item
);
5023 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5024 ~((u64
)root
->sectorsize
- 1);
5027 if (start
>= extent_end
) {
5029 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5030 ret
= btrfs_next_leaf(root
, path
);
5037 leaf
= path
->nodes
[0];
5039 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5040 if (found_key
.objectid
!= objectid
||
5041 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5043 if (start
+ len
<= found_key
.offset
)
5046 em
->len
= found_key
.offset
- start
;
5050 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5051 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5052 em
->start
= extent_start
;
5053 em
->len
= extent_end
- extent_start
;
5054 em
->orig_start
= extent_start
-
5055 btrfs_file_extent_offset(leaf
, item
);
5056 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5058 em
->block_start
= EXTENT_MAP_HOLE
;
5061 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5062 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5063 em
->compress_type
= compress_type
;
5064 em
->block_start
= bytenr
;
5065 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5068 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5069 em
->block_start
= bytenr
;
5070 em
->block_len
= em
->len
;
5071 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5072 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5075 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5079 size_t extent_offset
;
5082 em
->block_start
= EXTENT_MAP_INLINE
;
5083 if (!page
|| create
) {
5084 em
->start
= extent_start
;
5085 em
->len
= extent_end
- extent_start
;
5089 size
= btrfs_file_extent_inline_len(leaf
, item
);
5090 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5091 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5092 size
- extent_offset
);
5093 em
->start
= extent_start
+ extent_offset
;
5094 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5095 ~((u64
)root
->sectorsize
- 1);
5096 em
->orig_start
= EXTENT_MAP_INLINE
;
5097 if (compress_type
) {
5098 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5099 em
->compress_type
= compress_type
;
5101 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5102 if (create
== 0 && !PageUptodate(page
)) {
5103 if (btrfs_file_extent_compression(leaf
, item
) !=
5104 BTRFS_COMPRESS_NONE
) {
5105 ret
= uncompress_inline(path
, inode
, page
,
5107 extent_offset
, item
);
5111 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5113 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5114 memset(map
+ pg_offset
+ copy_size
, 0,
5115 PAGE_CACHE_SIZE
- pg_offset
-
5120 flush_dcache_page(page
);
5121 } else if (create
&& PageUptodate(page
)) {
5125 free_extent_map(em
);
5128 btrfs_release_path(path
);
5129 trans
= btrfs_join_transaction(root
);
5132 return ERR_CAST(trans
);
5136 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5139 btrfs_mark_buffer_dirty(leaf
);
5141 set_extent_uptodate(io_tree
, em
->start
,
5142 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5145 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5152 em
->block_start
= EXTENT_MAP_HOLE
;
5153 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5155 btrfs_release_path(path
);
5156 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5157 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5158 "[%llu %llu]\n", (unsigned long long)em
->start
,
5159 (unsigned long long)em
->len
,
5160 (unsigned long long)start
,
5161 (unsigned long long)len
);
5167 write_lock(&em_tree
->lock
);
5168 ret
= add_extent_mapping(em_tree
, em
);
5169 /* it is possible that someone inserted the extent into the tree
5170 * while we had the lock dropped. It is also possible that
5171 * an overlapping map exists in the tree
5173 if (ret
== -EEXIST
) {
5174 struct extent_map
*existing
;
5178 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5179 if (existing
&& (existing
->start
> start
||
5180 existing
->start
+ existing
->len
<= start
)) {
5181 free_extent_map(existing
);
5185 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5188 err
= merge_extent_mapping(em_tree
, existing
,
5191 free_extent_map(existing
);
5193 free_extent_map(em
);
5198 free_extent_map(em
);
5202 free_extent_map(em
);
5207 write_unlock(&em_tree
->lock
);
5210 trace_btrfs_get_extent(root
, em
);
5213 btrfs_free_path(path
);
5215 ret
= btrfs_end_transaction(trans
, root
);
5220 free_extent_map(em
);
5221 return ERR_PTR(err
);
5226 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5227 size_t pg_offset
, u64 start
, u64 len
,
5230 struct extent_map
*em
;
5231 struct extent_map
*hole_em
= NULL
;
5232 u64 range_start
= start
;
5238 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5243 * if our em maps to a hole, there might
5244 * actually be delalloc bytes behind it
5246 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5252 /* check to see if we've wrapped (len == -1 or similar) */
5261 /* ok, we didn't find anything, lets look for delalloc */
5262 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5263 end
, len
, EXTENT_DELALLOC
, 1);
5264 found_end
= range_start
+ found
;
5265 if (found_end
< range_start
)
5266 found_end
= (u64
)-1;
5269 * we didn't find anything useful, return
5270 * the original results from get_extent()
5272 if (range_start
> end
|| found_end
<= start
) {
5278 /* adjust the range_start to make sure it doesn't
5279 * go backwards from the start they passed in
5281 range_start
= max(start
,range_start
);
5282 found
= found_end
- range_start
;
5285 u64 hole_start
= start
;
5288 em
= alloc_extent_map();
5294 * when btrfs_get_extent can't find anything it
5295 * returns one huge hole
5297 * make sure what it found really fits our range, and
5298 * adjust to make sure it is based on the start from
5302 u64 calc_end
= extent_map_end(hole_em
);
5304 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5305 free_extent_map(hole_em
);
5308 hole_start
= max(hole_em
->start
, start
);
5309 hole_len
= calc_end
- hole_start
;
5313 if (hole_em
&& range_start
> hole_start
) {
5314 /* our hole starts before our delalloc, so we
5315 * have to return just the parts of the hole
5316 * that go until the delalloc starts
5318 em
->len
= min(hole_len
,
5319 range_start
- hole_start
);
5320 em
->start
= hole_start
;
5321 em
->orig_start
= hole_start
;
5323 * don't adjust block start at all,
5324 * it is fixed at EXTENT_MAP_HOLE
5326 em
->block_start
= hole_em
->block_start
;
5327 em
->block_len
= hole_len
;
5329 em
->start
= range_start
;
5331 em
->orig_start
= range_start
;
5332 em
->block_start
= EXTENT_MAP_DELALLOC
;
5333 em
->block_len
= found
;
5335 } else if (hole_em
) {
5340 free_extent_map(hole_em
);
5342 free_extent_map(em
);
5343 return ERR_PTR(err
);
5348 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5349 struct extent_map
*em
,
5352 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5353 struct btrfs_trans_handle
*trans
;
5354 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5355 struct btrfs_key ins
;
5358 bool insert
= false;
5361 * Ok if the extent map we looked up is a hole and is for the exact
5362 * range we want, there is no reason to allocate a new one, however if
5363 * it is not right then we need to free this one and drop the cache for
5366 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5368 free_extent_map(em
);
5371 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5374 trans
= btrfs_join_transaction(root
);
5376 return ERR_CAST(trans
);
5378 if (start
<= BTRFS_I(inode
)->disk_i_size
&& len
< 64 * 1024)
5379 btrfs_add_inode_defrag(trans
, inode
);
5381 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5383 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5384 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5385 alloc_hint
, (u64
)-1, &ins
, 1);
5392 em
= alloc_extent_map();
5394 em
= ERR_PTR(-ENOMEM
);
5400 em
->orig_start
= em
->start
;
5401 em
->len
= ins
.offset
;
5403 em
->block_start
= ins
.objectid
;
5404 em
->block_len
= ins
.offset
;
5405 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5408 * We need to do this because if we're using the original em we searched
5409 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5412 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5415 write_lock(&em_tree
->lock
);
5416 ret
= add_extent_mapping(em_tree
, em
);
5417 write_unlock(&em_tree
->lock
);
5420 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5423 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5424 ins
.offset
, ins
.offset
, 0);
5426 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5430 btrfs_end_transaction(trans
, root
);
5435 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5436 * block must be cow'd
5438 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5439 struct inode
*inode
, u64 offset
, u64 len
)
5441 struct btrfs_path
*path
;
5443 struct extent_buffer
*leaf
;
5444 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5445 struct btrfs_file_extent_item
*fi
;
5446 struct btrfs_key key
;
5454 path
= btrfs_alloc_path();
5458 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
5463 slot
= path
->slots
[0];
5466 /* can't find the item, must cow */
5473 leaf
= path
->nodes
[0];
5474 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5475 if (key
.objectid
!= btrfs_ino(inode
) ||
5476 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5477 /* not our file or wrong item type, must cow */
5481 if (key
.offset
> offset
) {
5482 /* Wrong offset, must cow */
5486 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5487 found_type
= btrfs_file_extent_type(leaf
, fi
);
5488 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5489 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5490 /* not a regular extent, must cow */
5493 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5494 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5496 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5497 if (extent_end
< offset
+ len
) {
5498 /* extent doesn't include our full range, must cow */
5502 if (btrfs_extent_readonly(root
, disk_bytenr
))
5506 * look for other files referencing this extent, if we
5507 * find any we must cow
5509 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
5510 key
.offset
- backref_offset
, disk_bytenr
))
5514 * adjust disk_bytenr and num_bytes to cover just the bytes
5515 * in this extent we are about to write. If there
5516 * are any csums in that range we have to cow in order
5517 * to keep the csums correct
5519 disk_bytenr
+= backref_offset
;
5520 disk_bytenr
+= offset
- key
.offset
;
5521 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5522 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5525 * all of the above have passed, it is safe to overwrite this extent
5530 btrfs_free_path(path
);
5534 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5535 struct buffer_head
*bh_result
, int create
)
5537 struct extent_map
*em
;
5538 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5539 u64 start
= iblock
<< inode
->i_blkbits
;
5540 u64 len
= bh_result
->b_size
;
5541 struct btrfs_trans_handle
*trans
;
5543 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5548 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5549 * io. INLINE is special, and we could probably kludge it in here, but
5550 * it's still buffered so for safety lets just fall back to the generic
5553 * For COMPRESSED we _have_ to read the entire extent in so we can
5554 * decompress it, so there will be buffering required no matter what we
5555 * do, so go ahead and fallback to buffered.
5557 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5558 * to buffered IO. Don't blame me, this is the price we pay for using
5561 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5562 em
->block_start
== EXTENT_MAP_INLINE
) {
5563 free_extent_map(em
);
5567 /* Just a good old fashioned hole, return */
5568 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5569 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5570 free_extent_map(em
);
5571 /* DIO will do one hole at a time, so just unlock a sector */
5572 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5573 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5578 * We don't allocate a new extent in the following cases
5580 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5582 * 2) The extent is marked as PREALLOC. We're good to go here and can
5583 * just use the extent.
5587 len
= em
->len
- (start
- em
->start
);
5591 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5592 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5593 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5598 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5599 type
= BTRFS_ORDERED_PREALLOC
;
5601 type
= BTRFS_ORDERED_NOCOW
;
5602 len
= min(len
, em
->len
- (start
- em
->start
));
5603 block_start
= em
->block_start
+ (start
- em
->start
);
5606 * we're not going to log anything, but we do need
5607 * to make sure the current transaction stays open
5608 * while we look for nocow cross refs
5610 trans
= btrfs_join_transaction(root
);
5614 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5615 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5616 block_start
, len
, len
, type
);
5617 btrfs_end_transaction(trans
, root
);
5619 free_extent_map(em
);
5624 btrfs_end_transaction(trans
, root
);
5628 * this will cow the extent, reset the len in case we changed
5631 len
= bh_result
->b_size
;
5632 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
5635 len
= min(len
, em
->len
- (start
- em
->start
));
5637 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5638 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5641 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5643 bh_result
->b_size
= len
;
5644 bh_result
->b_bdev
= em
->bdev
;
5645 set_buffer_mapped(bh_result
);
5646 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5647 set_buffer_new(bh_result
);
5649 free_extent_map(em
);
5654 struct btrfs_dio_private
{
5655 struct inode
*inode
;
5662 /* number of bios pending for this dio */
5663 atomic_t pending_bios
;
5668 struct bio
*orig_bio
;
5671 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5673 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5674 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5675 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5676 struct inode
*inode
= dip
->inode
;
5677 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5679 u32
*private = dip
->csums
;
5681 start
= dip
->logical_offset
;
5683 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5684 struct page
*page
= bvec
->bv_page
;
5687 unsigned long flags
;
5689 local_irq_save(flags
);
5690 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5691 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5692 csum
, bvec
->bv_len
);
5693 btrfs_csum_final(csum
, (char *)&csum
);
5694 kunmap_atomic(kaddr
, KM_IRQ0
);
5695 local_irq_restore(flags
);
5697 flush_dcache_page(bvec
->bv_page
);
5698 if (csum
!= *private) {
5699 printk(KERN_ERR
"btrfs csum failed ino %llu off"
5700 " %llu csum %u private %u\n",
5701 (unsigned long long)btrfs_ino(inode
),
5702 (unsigned long long)start
,
5708 start
+= bvec
->bv_len
;
5711 } while (bvec
<= bvec_end
);
5713 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5714 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5715 bio
->bi_private
= dip
->private;
5720 /* If we had a csum failure make sure to clear the uptodate flag */
5722 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5723 dio_end_io(bio
, err
);
5726 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5728 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5729 struct inode
*inode
= dip
->inode
;
5730 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5731 struct btrfs_trans_handle
*trans
;
5732 struct btrfs_ordered_extent
*ordered
= NULL
;
5733 struct extent_state
*cached_state
= NULL
;
5734 u64 ordered_offset
= dip
->logical_offset
;
5735 u64 ordered_bytes
= dip
->bytes
;
5741 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5749 trans
= btrfs_join_transaction(root
);
5750 if (IS_ERR(trans
)) {
5754 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5756 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5757 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5759 ret
= btrfs_update_inode(trans
, root
, inode
);
5764 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5765 ordered
->file_offset
+ ordered
->len
- 1, 0,
5766 &cached_state
, GFP_NOFS
);
5768 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5769 ret
= btrfs_mark_extent_written(trans
, inode
,
5770 ordered
->file_offset
,
5771 ordered
->file_offset
+
5778 ret
= insert_reserved_file_extent(trans
, inode
,
5779 ordered
->file_offset
,
5785 BTRFS_FILE_EXTENT_REG
);
5786 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5787 ordered
->file_offset
, ordered
->len
);
5795 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5796 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5798 btrfs_update_inode(trans
, root
, inode
);
5801 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5802 ordered
->file_offset
+ ordered
->len
- 1,
5803 &cached_state
, GFP_NOFS
);
5805 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5806 btrfs_end_transaction(trans
, root
);
5807 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5808 btrfs_put_ordered_extent(ordered
);
5809 btrfs_put_ordered_extent(ordered
);
5813 * our bio might span multiple ordered extents. If we haven't
5814 * completed the accounting for the whole dio, go back and try again
5816 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5817 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5822 bio
->bi_private
= dip
->private;
5827 /* If we had an error make sure to clear the uptodate flag */
5829 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5830 dio_end_io(bio
, err
);
5833 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5834 struct bio
*bio
, int mirror_num
,
5835 unsigned long bio_flags
, u64 offset
)
5838 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5839 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5844 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5846 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5849 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
5850 "sector %#Lx len %u err no %d\n",
5851 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
5852 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5856 * before atomic variable goto zero, we must make sure
5857 * dip->errors is perceived to be set.
5859 smp_mb__before_atomic_dec();
5862 /* if there are more bios still pending for this dio, just exit */
5863 if (!atomic_dec_and_test(&dip
->pending_bios
))
5867 bio_io_error(dip
->orig_bio
);
5869 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5870 bio_endio(dip
->orig_bio
, 0);
5876 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5877 u64 first_sector
, gfp_t gfp_flags
)
5879 int nr_vecs
= bio_get_nr_vecs(bdev
);
5880 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5883 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5884 int rw
, u64 file_offset
, int skip_sum
,
5885 u32
*csums
, int async_submit
)
5887 int write
= rw
& REQ_WRITE
;
5888 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5892 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5899 if (write
&& async_submit
) {
5900 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5901 inode
, rw
, bio
, 0, 0,
5903 __btrfs_submit_bio_start_direct_io
,
5904 __btrfs_submit_bio_done
);
5908 * If we aren't doing async submit, calculate the csum of the
5911 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
5914 } else if (!skip_sum
) {
5915 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5916 file_offset
, csums
);
5922 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
5928 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5931 struct inode
*inode
= dip
->inode
;
5932 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5933 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5935 struct bio
*orig_bio
= dip
->orig_bio
;
5936 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5937 u64 start_sector
= orig_bio
->bi_sector
;
5938 u64 file_offset
= dip
->logical_offset
;
5942 u32
*csums
= dip
->csums
;
5944 int async_submit
= 0;
5945 int write
= rw
& REQ_WRITE
;
5947 map_length
= orig_bio
->bi_size
;
5948 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5949 &map_length
, NULL
, 0);
5955 if (map_length
>= orig_bio
->bi_size
) {
5961 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
5964 bio
->bi_private
= dip
;
5965 bio
->bi_end_io
= btrfs_end_dio_bio
;
5966 atomic_inc(&dip
->pending_bios
);
5968 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
5969 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
5970 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5971 bvec
->bv_offset
) < bvec
->bv_len
)) {
5973 * inc the count before we submit the bio so
5974 * we know the end IO handler won't happen before
5975 * we inc the count. Otherwise, the dip might get freed
5976 * before we're done setting it up
5978 atomic_inc(&dip
->pending_bios
);
5979 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
5980 file_offset
, skip_sum
,
5981 csums
, async_submit
);
5984 atomic_dec(&dip
->pending_bios
);
5988 /* Write's use the ordered csums */
5989 if (!write
&& !skip_sum
)
5990 csums
= csums
+ nr_pages
;
5991 start_sector
+= submit_len
>> 9;
5992 file_offset
+= submit_len
;
5997 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
5998 start_sector
, GFP_NOFS
);
6001 bio
->bi_private
= dip
;
6002 bio
->bi_end_io
= btrfs_end_dio_bio
;
6004 map_length
= orig_bio
->bi_size
;
6005 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6006 &map_length
, NULL
, 0);
6012 submit_len
+= bvec
->bv_len
;
6019 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6020 csums
, async_submit
);
6028 * before atomic variable goto zero, we must
6029 * make sure dip->errors is perceived to be set.
6031 smp_mb__before_atomic_dec();
6032 if (atomic_dec_and_test(&dip
->pending_bios
))
6033 bio_io_error(dip
->orig_bio
);
6035 /* bio_end_io() will handle error, so we needn't return it */
6039 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6042 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6043 struct btrfs_dio_private
*dip
;
6044 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6046 int write
= rw
& REQ_WRITE
;
6049 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6051 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6058 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6059 if (!write
&& !skip_sum
) {
6060 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6068 dip
->private = bio
->bi_private
;
6070 dip
->logical_offset
= file_offset
;
6074 dip
->bytes
+= bvec
->bv_len
;
6076 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6078 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6079 bio
->bi_private
= dip
;
6081 dip
->orig_bio
= bio
;
6082 atomic_set(&dip
->pending_bios
, 0);
6085 bio
->bi_end_io
= btrfs_endio_direct_write
;
6087 bio
->bi_end_io
= btrfs_endio_direct_read
;
6089 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6094 * If this is a write, we need to clean up the reserved space and kill
6095 * the ordered extent.
6098 struct btrfs_ordered_extent
*ordered
;
6099 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6100 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6101 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6102 btrfs_free_reserved_extent(root
, ordered
->start
,
6104 btrfs_put_ordered_extent(ordered
);
6105 btrfs_put_ordered_extent(ordered
);
6107 bio_endio(bio
, ret
);
6110 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6111 const struct iovec
*iov
, loff_t offset
,
6112 unsigned long nr_segs
)
6118 unsigned blocksize_mask
= root
->sectorsize
- 1;
6119 ssize_t retval
= -EINVAL
;
6120 loff_t end
= offset
;
6122 if (offset
& blocksize_mask
)
6125 /* Check the memory alignment. Blocks cannot straddle pages */
6126 for (seg
= 0; seg
< nr_segs
; seg
++) {
6127 addr
= (unsigned long)iov
[seg
].iov_base
;
6128 size
= iov
[seg
].iov_len
;
6130 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6133 /* If this is a write we don't need to check anymore */
6138 * Check to make sure we don't have duplicate iov_base's in this
6139 * iovec, if so return EINVAL, otherwise we'll get csum errors
6140 * when reading back.
6142 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6143 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6151 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6152 const struct iovec
*iov
, loff_t offset
,
6153 unsigned long nr_segs
)
6155 struct file
*file
= iocb
->ki_filp
;
6156 struct inode
*inode
= file
->f_mapping
->host
;
6157 struct btrfs_ordered_extent
*ordered
;
6158 struct extent_state
*cached_state
= NULL
;
6159 u64 lockstart
, lockend
;
6161 int writing
= rw
& WRITE
;
6163 size_t count
= iov_length(iov
, nr_segs
);
6165 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6171 lockend
= offset
+ count
- 1;
6174 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6180 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6181 0, &cached_state
, GFP_NOFS
);
6183 * We're concerned with the entire range that we're going to be
6184 * doing DIO to, so we need to make sure theres no ordered
6185 * extents in this range.
6187 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6188 lockend
- lockstart
+ 1);
6191 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6192 &cached_state
, GFP_NOFS
);
6193 btrfs_start_ordered_extent(inode
, ordered
, 1);
6194 btrfs_put_ordered_extent(ordered
);
6199 * we don't use btrfs_set_extent_delalloc because we don't want
6200 * the dirty or uptodate bits
6203 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6204 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6205 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6208 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6209 lockend
, EXTENT_LOCKED
| write_bits
,
6210 1, 0, &cached_state
, GFP_NOFS
);
6215 free_extent_state(cached_state
);
6216 cached_state
= NULL
;
6218 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6219 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6220 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6221 btrfs_submit_direct
, 0);
6223 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6224 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6225 offset
+ iov_length(iov
, nr_segs
) - 1,
6226 EXTENT_LOCKED
| write_bits
, 1, 0,
6227 &cached_state
, GFP_NOFS
);
6228 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6230 * We're falling back to buffered, unlock the section we didn't
6233 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6234 offset
+ iov_length(iov
, nr_segs
) - 1,
6235 EXTENT_LOCKED
| write_bits
, 1, 0,
6236 &cached_state
, GFP_NOFS
);
6239 free_extent_state(cached_state
);
6243 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6244 __u64 start
, __u64 len
)
6246 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6249 int btrfs_readpage(struct file
*file
, struct page
*page
)
6251 struct extent_io_tree
*tree
;
6252 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6253 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6256 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6258 struct extent_io_tree
*tree
;
6261 if (current
->flags
& PF_MEMALLOC
) {
6262 redirty_page_for_writepage(wbc
, page
);
6266 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6267 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6270 int btrfs_writepages(struct address_space
*mapping
,
6271 struct writeback_control
*wbc
)
6273 struct extent_io_tree
*tree
;
6275 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6276 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6280 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6281 struct list_head
*pages
, unsigned nr_pages
)
6283 struct extent_io_tree
*tree
;
6284 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6285 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6288 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6290 struct extent_io_tree
*tree
;
6291 struct extent_map_tree
*map
;
6294 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6295 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6296 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6298 ClearPagePrivate(page
);
6299 set_page_private(page
, 0);
6300 page_cache_release(page
);
6305 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6307 if (PageWriteback(page
) || PageDirty(page
))
6309 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6312 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6314 struct extent_io_tree
*tree
;
6315 struct btrfs_ordered_extent
*ordered
;
6316 struct extent_state
*cached_state
= NULL
;
6317 u64 page_start
= page_offset(page
);
6318 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6322 * we have the page locked, so new writeback can't start,
6323 * and the dirty bit won't be cleared while we are here.
6325 * Wait for IO on this page so that we can safely clear
6326 * the PagePrivate2 bit and do ordered accounting
6328 wait_on_page_writeback(page
);
6330 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6332 btrfs_releasepage(page
, GFP_NOFS
);
6335 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6337 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6341 * IO on this page will never be started, so we need
6342 * to account for any ordered extents now
6344 clear_extent_bit(tree
, page_start
, page_end
,
6345 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6346 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6347 &cached_state
, GFP_NOFS
);
6349 * whoever cleared the private bit is responsible
6350 * for the finish_ordered_io
6352 if (TestClearPagePrivate2(page
)) {
6353 btrfs_finish_ordered_io(page
->mapping
->host
,
6354 page_start
, page_end
);
6356 btrfs_put_ordered_extent(ordered
);
6357 cached_state
= NULL
;
6358 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6361 clear_extent_bit(tree
, page_start
, page_end
,
6362 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6363 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6364 __btrfs_releasepage(page
, GFP_NOFS
);
6366 ClearPageChecked(page
);
6367 if (PagePrivate(page
)) {
6368 ClearPagePrivate(page
);
6369 set_page_private(page
, 0);
6370 page_cache_release(page
);
6375 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6376 * called from a page fault handler when a page is first dirtied. Hence we must
6377 * be careful to check for EOF conditions here. We set the page up correctly
6378 * for a written page which means we get ENOSPC checking when writing into
6379 * holes and correct delalloc and unwritten extent mapping on filesystems that
6380 * support these features.
6382 * We are not allowed to take the i_mutex here so we have to play games to
6383 * protect against truncate races as the page could now be beyond EOF. Because
6384 * vmtruncate() writes the inode size before removing pages, once we have the
6385 * page lock we can determine safely if the page is beyond EOF. If it is not
6386 * beyond EOF, then the page is guaranteed safe against truncation until we
6389 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6391 struct page
*page
= vmf
->page
;
6392 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6393 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6394 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6395 struct btrfs_ordered_extent
*ordered
;
6396 struct extent_state
*cached_state
= NULL
;
6398 unsigned long zero_start
;
6404 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6408 else /* -ENOSPC, -EIO, etc */
6409 ret
= VM_FAULT_SIGBUS
;
6413 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6416 size
= i_size_read(inode
);
6417 page_start
= page_offset(page
);
6418 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6420 if ((page
->mapping
!= inode
->i_mapping
) ||
6421 (page_start
>= size
)) {
6422 /* page got truncated out from underneath us */
6425 wait_on_page_writeback(page
);
6427 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6429 set_page_extent_mapped(page
);
6432 * we can't set the delalloc bits if there are pending ordered
6433 * extents. Drop our locks and wait for them to finish
6435 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6437 unlock_extent_cached(io_tree
, page_start
, page_end
,
6438 &cached_state
, GFP_NOFS
);
6440 btrfs_start_ordered_extent(inode
, ordered
, 1);
6441 btrfs_put_ordered_extent(ordered
);
6446 * XXX - page_mkwrite gets called every time the page is dirtied, even
6447 * if it was already dirty, so for space accounting reasons we need to
6448 * clear any delalloc bits for the range we are fixing to save. There
6449 * is probably a better way to do this, but for now keep consistent with
6450 * prepare_pages in the normal write path.
6452 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6453 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6454 0, 0, &cached_state
, GFP_NOFS
);
6456 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6459 unlock_extent_cached(io_tree
, page_start
, page_end
,
6460 &cached_state
, GFP_NOFS
);
6461 ret
= VM_FAULT_SIGBUS
;
6466 /* page is wholly or partially inside EOF */
6467 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6468 zero_start
= size
& ~PAGE_CACHE_MASK
;
6470 zero_start
= PAGE_CACHE_SIZE
;
6472 if (zero_start
!= PAGE_CACHE_SIZE
) {
6474 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6475 flush_dcache_page(page
);
6478 ClearPageChecked(page
);
6479 set_page_dirty(page
);
6480 SetPageUptodate(page
);
6482 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6483 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6485 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6489 return VM_FAULT_LOCKED
;
6491 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6496 static int btrfs_truncate(struct inode
*inode
)
6498 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6499 struct btrfs_block_rsv
*rsv
;
6502 struct btrfs_trans_handle
*trans
;
6504 u64 mask
= root
->sectorsize
- 1;
6506 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6510 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6511 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6514 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6515 * 3 things going on here
6517 * 1) We need to reserve space for our orphan item and the space to
6518 * delete our orphan item. Lord knows we don't want to have a dangling
6519 * orphan item because we didn't reserve space to remove it.
6521 * 2) We need to reserve space to update our inode.
6523 * 3) We need to have something to cache all the space that is going to
6524 * be free'd up by the truncate operation, but also have some slack
6525 * space reserved in case it uses space during the truncate (thank you
6526 * very much snapshotting).
6528 * And we need these to all be seperate. The fact is we can use alot of
6529 * space doing the truncate, and we have no earthly idea how much space
6530 * we will use, so we need the truncate reservation to be seperate so it
6531 * doesn't end up using space reserved for updating the inode or
6532 * removing the orphan item. We also need to be able to stop the
6533 * transaction and start a new one, which means we need to be able to
6534 * update the inode several times, and we have no idea of knowing how
6535 * many times that will be, so we can't just reserve 1 item for the
6536 * entirety of the opration, so that has to be done seperately as well.
6537 * Then there is the orphan item, which does indeed need to be held on
6538 * to for the whole operation, and we need nobody to touch this reserved
6539 * space except the orphan code.
6541 * So that leaves us with
6543 * 1) root->orphan_block_rsv - for the orphan deletion.
6544 * 2) rsv - for the truncate reservation, which we will steal from the
6545 * transaction reservation.
6546 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6547 * updating the inode.
6549 rsv
= btrfs_alloc_block_rsv(root
);
6552 btrfs_add_durable_block_rsv(root
->fs_info
, rsv
);
6554 trans
= btrfs_start_transaction(root
, 4);
6555 if (IS_ERR(trans
)) {
6556 err
= PTR_ERR(trans
);
6561 * Reserve space for the truncate process. Truncate should be adding
6562 * space, but if there are snapshots it may end up using space.
6564 ret
= btrfs_truncate_reserve_metadata(trans
, root
, rsv
);
6567 ret
= btrfs_orphan_add(trans
, inode
);
6569 btrfs_end_transaction(trans
, root
);
6573 nr
= trans
->blocks_used
;
6574 btrfs_end_transaction(trans
, root
);
6575 btrfs_btree_balance_dirty(root
, nr
);
6578 * Ok so we've already migrated our bytes over for the truncate, so here
6579 * just reserve the one slot we need for updating the inode.
6581 trans
= btrfs_start_transaction(root
, 1);
6582 if (IS_ERR(trans
)) {
6583 err
= PTR_ERR(trans
);
6586 trans
->block_rsv
= rsv
;
6589 * setattr is responsible for setting the ordered_data_close flag,
6590 * but that is only tested during the last file release. That
6591 * could happen well after the next commit, leaving a great big
6592 * window where new writes may get lost if someone chooses to write
6593 * to this file after truncating to zero
6595 * The inode doesn't have any dirty data here, and so if we commit
6596 * this is a noop. If someone immediately starts writing to the inode
6597 * it is very likely we'll catch some of their writes in this
6598 * transaction, and the commit will find this file on the ordered
6599 * data list with good things to send down.
6601 * This is a best effort solution, there is still a window where
6602 * using truncate to replace the contents of the file will
6603 * end up with a zero length file after a crash.
6605 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6606 btrfs_add_ordered_operation(trans
, root
, inode
);
6610 trans
= btrfs_start_transaction(root
, 3);
6611 if (IS_ERR(trans
)) {
6612 err
= PTR_ERR(trans
);
6616 ret
= btrfs_truncate_reserve_metadata(trans
, root
,
6620 trans
->block_rsv
= rsv
;
6623 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6625 BTRFS_EXTENT_DATA_KEY
);
6626 if (ret
!= -EAGAIN
) {
6631 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6632 ret
= btrfs_update_inode(trans
, root
, inode
);
6638 nr
= trans
->blocks_used
;
6639 btrfs_end_transaction(trans
, root
);
6641 btrfs_btree_balance_dirty(root
, nr
);
6644 if (ret
== 0 && inode
->i_nlink
> 0) {
6645 trans
->block_rsv
= root
->orphan_block_rsv
;
6646 ret
= btrfs_orphan_del(trans
, inode
);
6649 } else if (ret
&& inode
->i_nlink
> 0) {
6651 * Failed to do the truncate, remove us from the in memory
6654 ret
= btrfs_orphan_del(NULL
, inode
);
6657 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6658 ret
= btrfs_update_inode(trans
, root
, inode
);
6662 nr
= trans
->blocks_used
;
6663 ret
= btrfs_end_transaction_throttle(trans
, root
);
6664 btrfs_btree_balance_dirty(root
, nr
);
6667 btrfs_free_block_rsv(root
, rsv
);
6676 * create a new subvolume directory/inode (helper for the ioctl).
6678 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6679 struct btrfs_root
*new_root
, u64 new_dirid
)
6681 struct inode
*inode
;
6685 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6686 new_dirid
, S_IFDIR
| 0700, &index
);
6688 return PTR_ERR(inode
);
6689 inode
->i_op
= &btrfs_dir_inode_operations
;
6690 inode
->i_fop
= &btrfs_dir_file_operations
;
6693 btrfs_i_size_write(inode
, 0);
6695 err
= btrfs_update_inode(trans
, new_root
, inode
);
6702 /* helper function for file defrag and space balancing. This
6703 * forces readahead on a given range of bytes in an inode
6705 unsigned long btrfs_force_ra(struct address_space
*mapping
,
6706 struct file_ra_state
*ra
, struct file
*file
,
6707 pgoff_t offset
, pgoff_t last_index
)
6709 pgoff_t req_size
= last_index
- offset
+ 1;
6711 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
6712 return offset
+ req_size
;
6715 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6717 struct btrfs_inode
*ei
;
6718 struct inode
*inode
;
6720 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6725 ei
->space_info
= NULL
;
6729 ei
->last_sub_trans
= 0;
6730 ei
->logged_trans
= 0;
6731 ei
->delalloc_bytes
= 0;
6732 ei
->reserved_bytes
= 0;
6733 ei
->disk_i_size
= 0;
6735 ei
->index_cnt
= (u64
)-1;
6736 ei
->last_unlink_trans
= 0;
6738 atomic_set(&ei
->outstanding_extents
, 0);
6739 atomic_set(&ei
->reserved_extents
, 0);
6741 ei
->ordered_data_close
= 0;
6742 ei
->orphan_meta_reserved
= 0;
6743 ei
->dummy_inode
= 0;
6745 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6747 ei
->delayed_node
= NULL
;
6749 inode
= &ei
->vfs_inode
;
6750 extent_map_tree_init(&ei
->extent_tree
);
6751 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
6752 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
6753 mutex_init(&ei
->log_mutex
);
6754 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6755 INIT_LIST_HEAD(&ei
->i_orphan
);
6756 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6757 INIT_LIST_HEAD(&ei
->ordered_operations
);
6758 RB_CLEAR_NODE(&ei
->rb_node
);
6763 static void btrfs_i_callback(struct rcu_head
*head
)
6765 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6766 INIT_LIST_HEAD(&inode
->i_dentry
);
6767 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6770 void btrfs_destroy_inode(struct inode
*inode
)
6772 struct btrfs_ordered_extent
*ordered
;
6773 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6775 WARN_ON(!list_empty(&inode
->i_dentry
));
6776 WARN_ON(inode
->i_data
.nrpages
);
6777 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
));
6778 WARN_ON(atomic_read(&BTRFS_I(inode
)->reserved_extents
));
6781 * This can happen where we create an inode, but somebody else also
6782 * created the same inode and we need to destroy the one we already
6789 * Make sure we're properly removed from the ordered operation
6793 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6794 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6795 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6796 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6799 spin_lock(&root
->orphan_lock
);
6800 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6801 printk(KERN_INFO
"BTRFS: inode %llu still on the orphan list\n",
6802 (unsigned long long)btrfs_ino(inode
));
6803 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6805 spin_unlock(&root
->orphan_lock
);
6808 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6812 printk(KERN_ERR
"btrfs found ordered "
6813 "extent %llu %llu on inode cleanup\n",
6814 (unsigned long long)ordered
->file_offset
,
6815 (unsigned long long)ordered
->len
);
6816 btrfs_remove_ordered_extent(inode
, ordered
);
6817 btrfs_put_ordered_extent(ordered
);
6818 btrfs_put_ordered_extent(ordered
);
6821 inode_tree_del(inode
);
6822 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6824 btrfs_remove_delayed_node(inode
);
6825 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6828 int btrfs_drop_inode(struct inode
*inode
)
6830 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6832 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6833 !is_free_space_inode(root
, inode
))
6836 return generic_drop_inode(inode
);
6839 static void init_once(void *foo
)
6841 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6843 inode_init_once(&ei
->vfs_inode
);
6846 void btrfs_destroy_cachep(void)
6848 if (btrfs_inode_cachep
)
6849 kmem_cache_destroy(btrfs_inode_cachep
);
6850 if (btrfs_trans_handle_cachep
)
6851 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6852 if (btrfs_transaction_cachep
)
6853 kmem_cache_destroy(btrfs_transaction_cachep
);
6854 if (btrfs_path_cachep
)
6855 kmem_cache_destroy(btrfs_path_cachep
);
6856 if (btrfs_free_space_cachep
)
6857 kmem_cache_destroy(btrfs_free_space_cachep
);
6860 int btrfs_init_cachep(void)
6862 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6863 sizeof(struct btrfs_inode
), 0,
6864 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6865 if (!btrfs_inode_cachep
)
6868 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6869 sizeof(struct btrfs_trans_handle
), 0,
6870 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6871 if (!btrfs_trans_handle_cachep
)
6874 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6875 sizeof(struct btrfs_transaction
), 0,
6876 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6877 if (!btrfs_transaction_cachep
)
6880 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6881 sizeof(struct btrfs_path
), 0,
6882 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6883 if (!btrfs_path_cachep
)
6886 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6887 sizeof(struct btrfs_free_space
), 0,
6888 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6889 if (!btrfs_free_space_cachep
)
6894 btrfs_destroy_cachep();
6898 static int btrfs_getattr(struct vfsmount
*mnt
,
6899 struct dentry
*dentry
, struct kstat
*stat
)
6901 struct inode
*inode
= dentry
->d_inode
;
6902 generic_fillattr(inode
, stat
);
6903 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
6904 stat
->blksize
= PAGE_CACHE_SIZE
;
6905 stat
->blocks
= (inode_get_bytes(inode
) +
6906 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
6911 * If a file is moved, it will inherit the cow and compression flags of the new
6914 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
6916 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
6917 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
6919 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
6920 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
6922 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
6924 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
)
6925 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
6927 b_inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6930 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6931 struct inode
*new_dir
, struct dentry
*new_dentry
)
6933 struct btrfs_trans_handle
*trans
;
6934 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6935 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6936 struct inode
*new_inode
= new_dentry
->d_inode
;
6937 struct inode
*old_inode
= old_dentry
->d_inode
;
6938 struct timespec ctime
= CURRENT_TIME
;
6942 u64 old_ino
= btrfs_ino(old_inode
);
6944 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
6947 /* we only allow rename subvolume link between subvolumes */
6948 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
6951 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
6952 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
6955 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
6956 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
6959 * we're using rename to replace one file with another.
6960 * and the replacement file is large. Start IO on it now so
6961 * we don't add too much work to the end of the transaction
6963 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
6964 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
6965 filemap_flush(old_inode
->i_mapping
);
6967 /* close the racy window with snapshot create/destroy ioctl */
6968 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
6969 down_read(&root
->fs_info
->subvol_sem
);
6971 * We want to reserve the absolute worst case amount of items. So if
6972 * both inodes are subvols and we need to unlink them then that would
6973 * require 4 item modifications, but if they are both normal inodes it
6974 * would require 5 item modifications, so we'll assume their normal
6975 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6976 * should cover the worst case number of items we'll modify.
6978 trans
= btrfs_start_transaction(root
, 20);
6979 if (IS_ERR(trans
)) {
6980 ret
= PTR_ERR(trans
);
6985 btrfs_record_root_in_trans(trans
, dest
);
6987 ret
= btrfs_set_inode_index(new_dir
, &index
);
6991 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6992 /* force full log commit if subvolume involved. */
6993 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
6995 ret
= btrfs_insert_inode_ref(trans
, dest
,
6996 new_dentry
->d_name
.name
,
6997 new_dentry
->d_name
.len
,
6999 btrfs_ino(new_dir
), index
);
7003 * this is an ugly little race, but the rename is required
7004 * to make sure that if we crash, the inode is either at the
7005 * old name or the new one. pinning the log transaction lets
7006 * us make sure we don't allow a log commit to come in after
7007 * we unlink the name but before we add the new name back in.
7009 btrfs_pin_log_trans(root
);
7012 * make sure the inode gets flushed if it is replacing
7015 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
7016 btrfs_add_ordered_operation(trans
, root
, old_inode
);
7018 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
7019 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
7020 old_inode
->i_ctime
= ctime
;
7022 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
7023 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
7025 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7026 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
7027 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
7028 old_dentry
->d_name
.name
,
7029 old_dentry
->d_name
.len
);
7031 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7032 old_dentry
->d_inode
,
7033 old_dentry
->d_name
.name
,
7034 old_dentry
->d_name
.len
);
7036 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7041 new_inode
->i_ctime
= CURRENT_TIME
;
7042 if (unlikely(btrfs_ino(new_inode
) ==
7043 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7044 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7045 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7047 new_dentry
->d_name
.name
,
7048 new_dentry
->d_name
.len
);
7049 BUG_ON(new_inode
->i_nlink
== 0);
7051 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7052 new_dentry
->d_inode
,
7053 new_dentry
->d_name
.name
,
7054 new_dentry
->d_name
.len
);
7057 if (new_inode
->i_nlink
== 0) {
7058 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7063 fixup_inode_flags(new_dir
, old_inode
);
7065 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7066 new_dentry
->d_name
.name
,
7067 new_dentry
->d_name
.len
, 0, index
);
7070 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7071 struct dentry
*parent
= dget_parent(new_dentry
);
7072 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7074 btrfs_end_log_trans(root
);
7077 btrfs_end_transaction_throttle(trans
, root
);
7079 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7080 up_read(&root
->fs_info
->subvol_sem
);
7086 * some fairly slow code that needs optimization. This walks the list
7087 * of all the inodes with pending delalloc and forces them to disk.
7089 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7091 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7092 struct btrfs_inode
*binode
;
7093 struct inode
*inode
;
7095 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7098 spin_lock(&root
->fs_info
->delalloc_lock
);
7099 while (!list_empty(head
)) {
7100 binode
= list_entry(head
->next
, struct btrfs_inode
,
7102 inode
= igrab(&binode
->vfs_inode
);
7104 list_del_init(&binode
->delalloc_inodes
);
7105 spin_unlock(&root
->fs_info
->delalloc_lock
);
7107 filemap_flush(inode
->i_mapping
);
7109 btrfs_add_delayed_iput(inode
);
7114 spin_lock(&root
->fs_info
->delalloc_lock
);
7116 spin_unlock(&root
->fs_info
->delalloc_lock
);
7118 /* the filemap_flush will queue IO into the worker threads, but
7119 * we have to make sure the IO is actually started and that
7120 * ordered extents get created before we return
7122 atomic_inc(&root
->fs_info
->async_submit_draining
);
7123 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7124 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7125 wait_event(root
->fs_info
->async_submit_wait
,
7126 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7127 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7129 atomic_dec(&root
->fs_info
->async_submit_draining
);
7133 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7134 const char *symname
)
7136 struct btrfs_trans_handle
*trans
;
7137 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7138 struct btrfs_path
*path
;
7139 struct btrfs_key key
;
7140 struct inode
*inode
= NULL
;
7148 struct btrfs_file_extent_item
*ei
;
7149 struct extent_buffer
*leaf
;
7150 unsigned long nr
= 0;
7152 name_len
= strlen(symname
) + 1;
7153 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7154 return -ENAMETOOLONG
;
7157 * 2 items for inode item and ref
7158 * 2 items for dir items
7159 * 1 item for xattr if selinux is on
7161 trans
= btrfs_start_transaction(root
, 5);
7163 return PTR_ERR(trans
);
7165 err
= btrfs_find_free_ino(root
, &objectid
);
7169 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7170 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
7171 S_IFLNK
|S_IRWXUGO
, &index
);
7172 if (IS_ERR(inode
)) {
7173 err
= PTR_ERR(inode
);
7177 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7183 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7187 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7188 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7189 inode
->i_fop
= &btrfs_file_operations
;
7190 inode
->i_op
= &btrfs_file_inode_operations
;
7191 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7196 path
= btrfs_alloc_path();
7198 key
.objectid
= btrfs_ino(inode
);
7200 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7201 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7202 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7206 btrfs_free_path(path
);
7209 leaf
= path
->nodes
[0];
7210 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7211 struct btrfs_file_extent_item
);
7212 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7213 btrfs_set_file_extent_type(leaf
, ei
,
7214 BTRFS_FILE_EXTENT_INLINE
);
7215 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7216 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7217 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7218 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7220 ptr
= btrfs_file_extent_inline_start(ei
);
7221 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7222 btrfs_mark_buffer_dirty(leaf
);
7223 btrfs_free_path(path
);
7225 inode
->i_op
= &btrfs_symlink_inode_operations
;
7226 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7227 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7228 inode_set_bytes(inode
, name_len
);
7229 btrfs_i_size_write(inode
, name_len
- 1);
7230 err
= btrfs_update_inode(trans
, root
, inode
);
7235 nr
= trans
->blocks_used
;
7236 btrfs_end_transaction_throttle(trans
, root
);
7238 inode_dec_link_count(inode
);
7241 btrfs_btree_balance_dirty(root
, nr
);
7245 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7246 u64 start
, u64 num_bytes
, u64 min_size
,
7247 loff_t actual_len
, u64
*alloc_hint
,
7248 struct btrfs_trans_handle
*trans
)
7250 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7251 struct btrfs_key ins
;
7252 u64 cur_offset
= start
;
7255 bool own_trans
= true;
7259 while (num_bytes
> 0) {
7261 trans
= btrfs_start_transaction(root
, 3);
7262 if (IS_ERR(trans
)) {
7263 ret
= PTR_ERR(trans
);
7268 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7269 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7272 btrfs_end_transaction(trans
, root
);
7276 ret
= insert_reserved_file_extent(trans
, inode
,
7277 cur_offset
, ins
.objectid
,
7278 ins
.offset
, ins
.offset
,
7279 ins
.offset
, 0, 0, 0,
7280 BTRFS_FILE_EXTENT_PREALLOC
);
7282 btrfs_drop_extent_cache(inode
, cur_offset
,
7283 cur_offset
+ ins
.offset
-1, 0);
7285 num_bytes
-= ins
.offset
;
7286 cur_offset
+= ins
.offset
;
7287 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7289 inode
->i_ctime
= CURRENT_TIME
;
7290 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7291 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7292 (actual_len
> inode
->i_size
) &&
7293 (cur_offset
> inode
->i_size
)) {
7294 if (cur_offset
> actual_len
)
7295 i_size
= actual_len
;
7297 i_size
= cur_offset
;
7298 i_size_write(inode
, i_size
);
7299 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7302 ret
= btrfs_update_inode(trans
, root
, inode
);
7306 btrfs_end_transaction(trans
, root
);
7311 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7312 u64 start
, u64 num_bytes
, u64 min_size
,
7313 loff_t actual_len
, u64
*alloc_hint
)
7315 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7316 min_size
, actual_len
, alloc_hint
,
7320 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7321 struct btrfs_trans_handle
*trans
, int mode
,
7322 u64 start
, u64 num_bytes
, u64 min_size
,
7323 loff_t actual_len
, u64
*alloc_hint
)
7325 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7326 min_size
, actual_len
, alloc_hint
, trans
);
7329 static int btrfs_set_page_dirty(struct page
*page
)
7331 return __set_page_dirty_nobuffers(page
);
7334 static int btrfs_permission(struct inode
*inode
, int mask
, unsigned int flags
)
7336 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7338 if (btrfs_root_readonly(root
) && (mask
& MAY_WRITE
))
7340 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
7342 return generic_permission(inode
, mask
, flags
, btrfs_check_acl
);
7345 static const struct inode_operations btrfs_dir_inode_operations
= {
7346 .getattr
= btrfs_getattr
,
7347 .lookup
= btrfs_lookup
,
7348 .create
= btrfs_create
,
7349 .unlink
= btrfs_unlink
,
7351 .mkdir
= btrfs_mkdir
,
7352 .rmdir
= btrfs_rmdir
,
7353 .rename
= btrfs_rename
,
7354 .symlink
= btrfs_symlink
,
7355 .setattr
= btrfs_setattr
,
7356 .mknod
= btrfs_mknod
,
7357 .setxattr
= btrfs_setxattr
,
7358 .getxattr
= btrfs_getxattr
,
7359 .listxattr
= btrfs_listxattr
,
7360 .removexattr
= btrfs_removexattr
,
7361 .permission
= btrfs_permission
,
7363 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7364 .lookup
= btrfs_lookup
,
7365 .permission
= btrfs_permission
,
7368 static const struct file_operations btrfs_dir_file_operations
= {
7369 .llseek
= generic_file_llseek
,
7370 .read
= generic_read_dir
,
7371 .readdir
= btrfs_real_readdir
,
7372 .unlocked_ioctl
= btrfs_ioctl
,
7373 #ifdef CONFIG_COMPAT
7374 .compat_ioctl
= btrfs_ioctl
,
7376 .release
= btrfs_release_file
,
7377 .fsync
= btrfs_sync_file
,
7380 static struct extent_io_ops btrfs_extent_io_ops
= {
7381 .fill_delalloc
= run_delalloc_range
,
7382 .submit_bio_hook
= btrfs_submit_bio_hook
,
7383 .merge_bio_hook
= btrfs_merge_bio_hook
,
7384 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7385 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7386 .writepage_start_hook
= btrfs_writepage_start_hook
,
7387 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
7388 .set_bit_hook
= btrfs_set_bit_hook
,
7389 .clear_bit_hook
= btrfs_clear_bit_hook
,
7390 .merge_extent_hook
= btrfs_merge_extent_hook
,
7391 .split_extent_hook
= btrfs_split_extent_hook
,
7395 * btrfs doesn't support the bmap operation because swapfiles
7396 * use bmap to make a mapping of extents in the file. They assume
7397 * these extents won't change over the life of the file and they
7398 * use the bmap result to do IO directly to the drive.
7400 * the btrfs bmap call would return logical addresses that aren't
7401 * suitable for IO and they also will change frequently as COW
7402 * operations happen. So, swapfile + btrfs == corruption.
7404 * For now we're avoiding this by dropping bmap.
7406 static const struct address_space_operations btrfs_aops
= {
7407 .readpage
= btrfs_readpage
,
7408 .writepage
= btrfs_writepage
,
7409 .writepages
= btrfs_writepages
,
7410 .readpages
= btrfs_readpages
,
7411 .direct_IO
= btrfs_direct_IO
,
7412 .invalidatepage
= btrfs_invalidatepage
,
7413 .releasepage
= btrfs_releasepage
,
7414 .set_page_dirty
= btrfs_set_page_dirty
,
7415 .error_remove_page
= generic_error_remove_page
,
7418 static const struct address_space_operations btrfs_symlink_aops
= {
7419 .readpage
= btrfs_readpage
,
7420 .writepage
= btrfs_writepage
,
7421 .invalidatepage
= btrfs_invalidatepage
,
7422 .releasepage
= btrfs_releasepage
,
7425 static const struct inode_operations btrfs_file_inode_operations
= {
7426 .getattr
= btrfs_getattr
,
7427 .setattr
= btrfs_setattr
,
7428 .setxattr
= btrfs_setxattr
,
7429 .getxattr
= btrfs_getxattr
,
7430 .listxattr
= btrfs_listxattr
,
7431 .removexattr
= btrfs_removexattr
,
7432 .permission
= btrfs_permission
,
7433 .fiemap
= btrfs_fiemap
,
7435 static const struct inode_operations btrfs_special_inode_operations
= {
7436 .getattr
= btrfs_getattr
,
7437 .setattr
= btrfs_setattr
,
7438 .permission
= btrfs_permission
,
7439 .setxattr
= btrfs_setxattr
,
7440 .getxattr
= btrfs_getxattr
,
7441 .listxattr
= btrfs_listxattr
,
7442 .removexattr
= btrfs_removexattr
,
7444 static const struct inode_operations btrfs_symlink_inode_operations
= {
7445 .readlink
= generic_readlink
,
7446 .follow_link
= page_follow_link_light
,
7447 .put_link
= page_put_link
,
7448 .getattr
= btrfs_getattr
,
7449 .permission
= btrfs_permission
,
7450 .setxattr
= btrfs_setxattr
,
7451 .getxattr
= btrfs_getxattr
,
7452 .listxattr
= btrfs_listxattr
,
7453 .removexattr
= btrfs_removexattr
,
7456 const struct dentry_operations btrfs_dentry_operations
= {
7457 .d_delete
= btrfs_dentry_delete
,