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
);
754 * when extent_io.c finds a delayed allocation range in the file,
755 * the call backs end up in this code. The basic idea is to
756 * allocate extents on disk for the range, and create ordered data structs
757 * in ram to track those extents.
759 * locked_page is the page that writepage had locked already. We use
760 * it to make sure we don't do extra locks or unlocks.
762 * *page_started is set to one if we unlock locked_page and do everything
763 * required to start IO on it. It may be clean and already done with
766 static noinline
int cow_file_range(struct inode
*inode
,
767 struct page
*locked_page
,
768 u64 start
, u64 end
, int *page_started
,
769 unsigned long *nr_written
,
772 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
773 struct btrfs_trans_handle
*trans
;
776 unsigned long ram_size
;
779 u64 blocksize
= root
->sectorsize
;
780 struct btrfs_key ins
;
781 struct extent_map
*em
;
782 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
785 BUG_ON(btrfs_is_free_space_inode(root
, inode
));
786 trans
= btrfs_join_transaction(root
);
787 BUG_ON(IS_ERR(trans
));
788 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
790 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
791 num_bytes
= max(blocksize
, num_bytes
);
792 disk_num_bytes
= num_bytes
;
795 /* if this is a small write inside eof, kick off defrag */
796 if (end
<= BTRFS_I(inode
)->disk_i_size
&& num_bytes
< 64 * 1024)
797 btrfs_add_inode_defrag(trans
, inode
);
800 /* lets try to make an inline extent */
801 ret
= cow_file_range_inline(trans
, root
, inode
,
802 start
, end
, 0, 0, NULL
);
804 extent_clear_unlock_delalloc(inode
,
805 &BTRFS_I(inode
)->io_tree
,
807 EXTENT_CLEAR_UNLOCK_PAGE
|
808 EXTENT_CLEAR_UNLOCK
|
809 EXTENT_CLEAR_DELALLOC
|
811 EXTENT_SET_WRITEBACK
|
812 EXTENT_END_WRITEBACK
);
814 *nr_written
= *nr_written
+
815 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
822 BUG_ON(disk_num_bytes
>
823 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
825 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
826 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
828 while (disk_num_bytes
> 0) {
831 cur_alloc_size
= disk_num_bytes
;
832 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
833 root
->sectorsize
, 0, alloc_hint
,
837 em
= alloc_extent_map();
840 em
->orig_start
= em
->start
;
841 ram_size
= ins
.offset
;
842 em
->len
= ins
.offset
;
844 em
->block_start
= ins
.objectid
;
845 em
->block_len
= ins
.offset
;
846 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
847 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
850 write_lock(&em_tree
->lock
);
851 ret
= add_extent_mapping(em_tree
, em
);
852 write_unlock(&em_tree
->lock
);
853 if (ret
!= -EEXIST
) {
857 btrfs_drop_extent_cache(inode
, start
,
858 start
+ ram_size
- 1, 0);
861 cur_alloc_size
= ins
.offset
;
862 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
863 ram_size
, cur_alloc_size
, 0);
866 if (root
->root_key
.objectid
==
867 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
868 ret
= btrfs_reloc_clone_csums(inode
, start
,
873 if (disk_num_bytes
< cur_alloc_size
)
876 /* we're not doing compressed IO, don't unlock the first
877 * page (which the caller expects to stay locked), don't
878 * clear any dirty bits and don't set any writeback bits
880 * Do set the Private2 bit so we know this page was properly
881 * setup for writepage
883 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
884 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
887 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
888 start
, start
+ ram_size
- 1,
890 disk_num_bytes
-= cur_alloc_size
;
891 num_bytes
-= cur_alloc_size
;
892 alloc_hint
= ins
.objectid
+ ins
.offset
;
893 start
+= cur_alloc_size
;
897 btrfs_end_transaction(trans
, root
);
903 * work queue call back to started compression on a file and pages
905 static noinline
void async_cow_start(struct btrfs_work
*work
)
907 struct async_cow
*async_cow
;
909 async_cow
= container_of(work
, struct async_cow
, work
);
911 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
912 async_cow
->start
, async_cow
->end
, async_cow
,
915 async_cow
->inode
= NULL
;
919 * work queue call back to submit previously compressed pages
921 static noinline
void async_cow_submit(struct btrfs_work
*work
)
923 struct async_cow
*async_cow
;
924 struct btrfs_root
*root
;
925 unsigned long nr_pages
;
927 async_cow
= container_of(work
, struct async_cow
, work
);
929 root
= async_cow
->root
;
930 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
933 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
935 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
937 waitqueue_active(&root
->fs_info
->async_submit_wait
))
938 wake_up(&root
->fs_info
->async_submit_wait
);
940 if (async_cow
->inode
)
941 submit_compressed_extents(async_cow
->inode
, async_cow
);
944 static noinline
void async_cow_free(struct btrfs_work
*work
)
946 struct async_cow
*async_cow
;
947 async_cow
= container_of(work
, struct async_cow
, work
);
951 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
952 u64 start
, u64 end
, int *page_started
,
953 unsigned long *nr_written
)
955 struct async_cow
*async_cow
;
956 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
957 unsigned long nr_pages
;
959 int limit
= 10 * 1024 * 1042;
961 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
962 1, 0, NULL
, GFP_NOFS
);
963 while (start
< end
) {
964 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
966 async_cow
->inode
= inode
;
967 async_cow
->root
= root
;
968 async_cow
->locked_page
= locked_page
;
969 async_cow
->start
= start
;
971 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
974 cur_end
= min(end
, start
+ 512 * 1024 - 1);
976 async_cow
->end
= cur_end
;
977 INIT_LIST_HEAD(&async_cow
->extents
);
979 async_cow
->work
.func
= async_cow_start
;
980 async_cow
->work
.ordered_func
= async_cow_submit
;
981 async_cow
->work
.ordered_free
= async_cow_free
;
982 async_cow
->work
.flags
= 0;
984 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
986 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
988 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
991 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
992 wait_event(root
->fs_info
->async_submit_wait
,
993 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
997 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
998 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
999 wait_event(root
->fs_info
->async_submit_wait
,
1000 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1004 *nr_written
+= nr_pages
;
1005 start
= cur_end
+ 1;
1011 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1012 u64 bytenr
, u64 num_bytes
)
1015 struct btrfs_ordered_sum
*sums
;
1018 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1019 bytenr
+ num_bytes
- 1, &list
, 0);
1020 if (ret
== 0 && list_empty(&list
))
1023 while (!list_empty(&list
)) {
1024 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1025 list_del(&sums
->list
);
1032 * when nowcow writeback call back. This checks for snapshots or COW copies
1033 * of the extents that exist in the file, and COWs the file as required.
1035 * If no cow copies or snapshots exist, we write directly to the existing
1038 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1039 struct page
*locked_page
,
1040 u64 start
, u64 end
, int *page_started
, int force
,
1041 unsigned long *nr_written
)
1043 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1044 struct btrfs_trans_handle
*trans
;
1045 struct extent_buffer
*leaf
;
1046 struct btrfs_path
*path
;
1047 struct btrfs_file_extent_item
*fi
;
1048 struct btrfs_key found_key
;
1061 u64 ino
= btrfs_ino(inode
);
1063 path
= btrfs_alloc_path();
1067 nolock
= btrfs_is_free_space_inode(root
, inode
);
1070 trans
= btrfs_join_transaction_nolock(root
);
1072 trans
= btrfs_join_transaction(root
);
1074 BUG_ON(IS_ERR(trans
));
1075 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1077 cow_start
= (u64
)-1;
1080 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1083 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1084 leaf
= path
->nodes
[0];
1085 btrfs_item_key_to_cpu(leaf
, &found_key
,
1086 path
->slots
[0] - 1);
1087 if (found_key
.objectid
== ino
&&
1088 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1093 leaf
= path
->nodes
[0];
1094 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1095 ret
= btrfs_next_leaf(root
, path
);
1100 leaf
= path
->nodes
[0];
1106 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1108 if (found_key
.objectid
> ino
||
1109 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1110 found_key
.offset
> end
)
1113 if (found_key
.offset
> cur_offset
) {
1114 extent_end
= found_key
.offset
;
1119 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1120 struct btrfs_file_extent_item
);
1121 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1123 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1124 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1125 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1126 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1127 extent_end
= found_key
.offset
+
1128 btrfs_file_extent_num_bytes(leaf
, fi
);
1129 if (extent_end
<= start
) {
1133 if (disk_bytenr
== 0)
1135 if (btrfs_file_extent_compression(leaf
, fi
) ||
1136 btrfs_file_extent_encryption(leaf
, fi
) ||
1137 btrfs_file_extent_other_encoding(leaf
, fi
))
1139 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1141 if (btrfs_extent_readonly(root
, disk_bytenr
))
1143 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1145 extent_offset
, disk_bytenr
))
1147 disk_bytenr
+= extent_offset
;
1148 disk_bytenr
+= cur_offset
- found_key
.offset
;
1149 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1151 * force cow if csum exists in the range.
1152 * this ensure that csum for a given extent are
1153 * either valid or do not exist.
1155 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1158 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1159 extent_end
= found_key
.offset
+
1160 btrfs_file_extent_inline_len(leaf
, fi
);
1161 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1166 if (extent_end
<= start
) {
1171 if (cow_start
== (u64
)-1)
1172 cow_start
= cur_offset
;
1173 cur_offset
= extent_end
;
1174 if (cur_offset
> end
)
1180 btrfs_release_path(path
);
1181 if (cow_start
!= (u64
)-1) {
1182 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1183 found_key
.offset
- 1, page_started
,
1186 cow_start
= (u64
)-1;
1189 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1190 struct extent_map
*em
;
1191 struct extent_map_tree
*em_tree
;
1192 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1193 em
= alloc_extent_map();
1195 em
->start
= cur_offset
;
1196 em
->orig_start
= em
->start
;
1197 em
->len
= num_bytes
;
1198 em
->block_len
= num_bytes
;
1199 em
->block_start
= disk_bytenr
;
1200 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1201 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1203 write_lock(&em_tree
->lock
);
1204 ret
= add_extent_mapping(em_tree
, em
);
1205 write_unlock(&em_tree
->lock
);
1206 if (ret
!= -EEXIST
) {
1207 free_extent_map(em
);
1210 btrfs_drop_extent_cache(inode
, em
->start
,
1211 em
->start
+ em
->len
- 1, 0);
1213 type
= BTRFS_ORDERED_PREALLOC
;
1215 type
= BTRFS_ORDERED_NOCOW
;
1218 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1219 num_bytes
, num_bytes
, type
);
1222 if (root
->root_key
.objectid
==
1223 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1224 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1229 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1230 cur_offset
, cur_offset
+ num_bytes
- 1,
1231 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1232 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1233 EXTENT_SET_PRIVATE2
);
1234 cur_offset
= extent_end
;
1235 if (cur_offset
> end
)
1238 btrfs_release_path(path
);
1240 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1241 cow_start
= cur_offset
;
1242 if (cow_start
!= (u64
)-1) {
1243 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1244 page_started
, nr_written
, 1);
1249 ret
= btrfs_end_transaction_nolock(trans
, root
);
1252 ret
= btrfs_end_transaction(trans
, root
);
1255 btrfs_free_path(path
);
1260 * extent_io.c call back to do delayed allocation processing
1262 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1263 u64 start
, u64 end
, int *page_started
,
1264 unsigned long *nr_written
)
1267 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1269 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1270 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1271 page_started
, 1, nr_written
);
1272 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1273 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1274 page_started
, 0, nr_written
);
1275 else if (!btrfs_test_opt(root
, COMPRESS
) &&
1276 !(BTRFS_I(inode
)->force_compress
) &&
1277 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))
1278 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1279 page_started
, nr_written
, 1);
1281 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1282 page_started
, nr_written
);
1286 static void btrfs_split_extent_hook(struct inode
*inode
,
1287 struct extent_state
*orig
, u64 split
)
1289 /* not delalloc, ignore it */
1290 if (!(orig
->state
& EXTENT_DELALLOC
))
1293 spin_lock(&BTRFS_I(inode
)->lock
);
1294 BTRFS_I(inode
)->outstanding_extents
++;
1295 spin_unlock(&BTRFS_I(inode
)->lock
);
1299 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1300 * extents so we can keep track of new extents that are just merged onto old
1301 * extents, such as when we are doing sequential writes, so we can properly
1302 * account for the metadata space we'll need.
1304 static void btrfs_merge_extent_hook(struct inode
*inode
,
1305 struct extent_state
*new,
1306 struct extent_state
*other
)
1308 /* not delalloc, ignore it */
1309 if (!(other
->state
& EXTENT_DELALLOC
))
1312 spin_lock(&BTRFS_I(inode
)->lock
);
1313 BTRFS_I(inode
)->outstanding_extents
--;
1314 spin_unlock(&BTRFS_I(inode
)->lock
);
1318 * extent_io.c set_bit_hook, used to track delayed allocation
1319 * bytes in this file, and to maintain the list of inodes that
1320 * have pending delalloc work to be done.
1322 static void btrfs_set_bit_hook(struct inode
*inode
,
1323 struct extent_state
*state
, int *bits
)
1327 * set_bit and clear bit hooks normally require _irqsave/restore
1328 * but in this case, we are only testing for the DELALLOC
1329 * bit, which is only set or cleared with irqs on
1331 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1332 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1333 u64 len
= state
->end
+ 1 - state
->start
;
1334 bool do_list
= !btrfs_is_free_space_inode(root
, inode
);
1336 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1337 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1339 spin_lock(&BTRFS_I(inode
)->lock
);
1340 BTRFS_I(inode
)->outstanding_extents
++;
1341 spin_unlock(&BTRFS_I(inode
)->lock
);
1344 spin_lock(&root
->fs_info
->delalloc_lock
);
1345 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1346 root
->fs_info
->delalloc_bytes
+= len
;
1347 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1348 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1349 &root
->fs_info
->delalloc_inodes
);
1351 spin_unlock(&root
->fs_info
->delalloc_lock
);
1356 * extent_io.c clear_bit_hook, see set_bit_hook for why
1358 static void btrfs_clear_bit_hook(struct inode
*inode
,
1359 struct extent_state
*state
, int *bits
)
1362 * set_bit and clear bit hooks normally require _irqsave/restore
1363 * but in this case, we are only testing for the DELALLOC
1364 * bit, which is only set or cleared with irqs on
1366 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1367 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1368 u64 len
= state
->end
+ 1 - state
->start
;
1369 bool do_list
= !btrfs_is_free_space_inode(root
, inode
);
1371 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1372 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1373 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1374 spin_lock(&BTRFS_I(inode
)->lock
);
1375 BTRFS_I(inode
)->outstanding_extents
--;
1376 spin_unlock(&BTRFS_I(inode
)->lock
);
1379 if (*bits
& EXTENT_DO_ACCOUNTING
)
1380 btrfs_delalloc_release_metadata(inode
, len
);
1382 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1384 btrfs_free_reserved_data_space(inode
, len
);
1386 spin_lock(&root
->fs_info
->delalloc_lock
);
1387 root
->fs_info
->delalloc_bytes
-= len
;
1388 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1390 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1391 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1392 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1394 spin_unlock(&root
->fs_info
->delalloc_lock
);
1399 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1400 * we don't create bios that span stripes or chunks
1402 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1403 size_t size
, struct bio
*bio
,
1404 unsigned long bio_flags
)
1406 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1407 struct btrfs_mapping_tree
*map_tree
;
1408 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1413 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1416 length
= bio
->bi_size
;
1417 map_tree
= &root
->fs_info
->mapping_tree
;
1418 map_length
= length
;
1419 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1420 &map_length
, NULL
, 0);
1422 if (map_length
< length
+ size
)
1428 * in order to insert checksums into the metadata in large chunks,
1429 * we wait until bio submission time. All the pages in the bio are
1430 * checksummed and sums are attached onto the ordered extent record.
1432 * At IO completion time the cums attached on the ordered extent record
1433 * are inserted into the btree
1435 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1436 struct bio
*bio
, int mirror_num
,
1437 unsigned long bio_flags
,
1440 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1443 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1449 * in order to insert checksums into the metadata in large chunks,
1450 * we wait until bio submission time. All the pages in the bio are
1451 * checksummed and sums are attached onto the ordered extent record.
1453 * At IO completion time the cums attached on the ordered extent record
1454 * are inserted into the btree
1456 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1457 int mirror_num
, unsigned long bio_flags
,
1460 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1461 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1465 * extent_io.c submission hook. This does the right thing for csum calculation
1466 * on write, or reading the csums from the tree before a read
1468 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1469 int mirror_num
, unsigned long bio_flags
,
1472 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1476 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1478 if (btrfs_is_free_space_inode(root
, inode
))
1479 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 2);
1481 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1484 if (!(rw
& REQ_WRITE
)) {
1485 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1486 return btrfs_submit_compressed_read(inode
, bio
,
1487 mirror_num
, bio_flags
);
1488 } else if (!skip_sum
) {
1489 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1494 } else if (!skip_sum
) {
1495 /* csum items have already been cloned */
1496 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1498 /* we're doing a write, do the async checksumming */
1499 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1500 inode
, rw
, bio
, mirror_num
,
1501 bio_flags
, bio_offset
,
1502 __btrfs_submit_bio_start
,
1503 __btrfs_submit_bio_done
);
1507 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1511 * given a list of ordered sums record them in the inode. This happens
1512 * at IO completion time based on sums calculated at bio submission time.
1514 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1515 struct inode
*inode
, u64 file_offset
,
1516 struct list_head
*list
)
1518 struct btrfs_ordered_sum
*sum
;
1520 list_for_each_entry(sum
, list
, list
) {
1521 btrfs_csum_file_blocks(trans
,
1522 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1527 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1528 struct extent_state
**cached_state
)
1530 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1532 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1533 cached_state
, GFP_NOFS
);
1536 /* see btrfs_writepage_start_hook for details on why this is required */
1537 struct btrfs_writepage_fixup
{
1539 struct btrfs_work work
;
1542 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1544 struct btrfs_writepage_fixup
*fixup
;
1545 struct btrfs_ordered_extent
*ordered
;
1546 struct extent_state
*cached_state
= NULL
;
1548 struct inode
*inode
;
1552 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1556 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1557 ClearPageChecked(page
);
1561 inode
= page
->mapping
->host
;
1562 page_start
= page_offset(page
);
1563 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1565 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1566 &cached_state
, GFP_NOFS
);
1568 /* already ordered? We're done */
1569 if (PagePrivate2(page
))
1572 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1574 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1575 page_end
, &cached_state
, GFP_NOFS
);
1577 btrfs_start_ordered_extent(inode
, ordered
, 1);
1582 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1583 ClearPageChecked(page
);
1585 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1586 &cached_state
, GFP_NOFS
);
1589 page_cache_release(page
);
1594 * There are a few paths in the higher layers of the kernel that directly
1595 * set the page dirty bit without asking the filesystem if it is a
1596 * good idea. This causes problems because we want to make sure COW
1597 * properly happens and the data=ordered rules are followed.
1599 * In our case any range that doesn't have the ORDERED bit set
1600 * hasn't been properly setup for IO. We kick off an async process
1601 * to fix it up. The async helper will wait for ordered extents, set
1602 * the delalloc bit and make it safe to write the page.
1604 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1606 struct inode
*inode
= page
->mapping
->host
;
1607 struct btrfs_writepage_fixup
*fixup
;
1608 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1610 /* this page is properly in the ordered list */
1611 if (TestClearPagePrivate2(page
))
1614 if (PageChecked(page
))
1617 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1621 SetPageChecked(page
);
1622 page_cache_get(page
);
1623 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1625 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1629 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1630 struct inode
*inode
, u64 file_pos
,
1631 u64 disk_bytenr
, u64 disk_num_bytes
,
1632 u64 num_bytes
, u64 ram_bytes
,
1633 u8 compression
, u8 encryption
,
1634 u16 other_encoding
, int extent_type
)
1636 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1637 struct btrfs_file_extent_item
*fi
;
1638 struct btrfs_path
*path
;
1639 struct extent_buffer
*leaf
;
1640 struct btrfs_key ins
;
1644 path
= btrfs_alloc_path();
1648 path
->leave_spinning
= 1;
1651 * we may be replacing one extent in the tree with another.
1652 * The new extent is pinned in the extent map, and we don't want
1653 * to drop it from the cache until it is completely in the btree.
1655 * So, tell btrfs_drop_extents to leave this extent in the cache.
1656 * the caller is expected to unpin it and allow it to be merged
1659 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1663 ins
.objectid
= btrfs_ino(inode
);
1664 ins
.offset
= file_pos
;
1665 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1666 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1668 leaf
= path
->nodes
[0];
1669 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1670 struct btrfs_file_extent_item
);
1671 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1672 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1673 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1674 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1675 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1676 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1677 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1678 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1679 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1680 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1682 btrfs_unlock_up_safe(path
, 1);
1683 btrfs_set_lock_blocking(leaf
);
1685 btrfs_mark_buffer_dirty(leaf
);
1687 inode_add_bytes(inode
, num_bytes
);
1689 ins
.objectid
= disk_bytenr
;
1690 ins
.offset
= disk_num_bytes
;
1691 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1692 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1693 root
->root_key
.objectid
,
1694 btrfs_ino(inode
), file_pos
, &ins
);
1696 btrfs_free_path(path
);
1702 * helper function for btrfs_finish_ordered_io, this
1703 * just reads in some of the csum leaves to prime them into ram
1704 * before we start the transaction. It limits the amount of btree
1705 * reads required while inside the transaction.
1707 /* as ordered data IO finishes, this gets called so we can finish
1708 * an ordered extent if the range of bytes in the file it covers are
1711 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1713 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1714 struct btrfs_trans_handle
*trans
= NULL
;
1715 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1716 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1717 struct extent_state
*cached_state
= NULL
;
1718 int compress_type
= 0;
1722 ret
= btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1726 BUG_ON(!ordered_extent
);
1728 nolock
= btrfs_is_free_space_inode(root
, inode
);
1730 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1731 BUG_ON(!list_empty(&ordered_extent
->list
));
1732 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1735 trans
= btrfs_join_transaction_nolock(root
);
1737 trans
= btrfs_join_transaction(root
);
1738 BUG_ON(IS_ERR(trans
));
1739 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1740 ret
= btrfs_update_inode(trans
, root
, inode
);
1746 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1747 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1748 0, &cached_state
, GFP_NOFS
);
1751 trans
= btrfs_join_transaction_nolock(root
);
1753 trans
= btrfs_join_transaction(root
);
1754 BUG_ON(IS_ERR(trans
));
1755 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1757 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1758 compress_type
= ordered_extent
->compress_type
;
1759 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1760 BUG_ON(compress_type
);
1761 ret
= btrfs_mark_extent_written(trans
, inode
,
1762 ordered_extent
->file_offset
,
1763 ordered_extent
->file_offset
+
1764 ordered_extent
->len
);
1767 BUG_ON(root
== root
->fs_info
->tree_root
);
1768 ret
= insert_reserved_file_extent(trans
, inode
,
1769 ordered_extent
->file_offset
,
1770 ordered_extent
->start
,
1771 ordered_extent
->disk_len
,
1772 ordered_extent
->len
,
1773 ordered_extent
->len
,
1774 compress_type
, 0, 0,
1775 BTRFS_FILE_EXTENT_REG
);
1776 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1777 ordered_extent
->file_offset
,
1778 ordered_extent
->len
);
1781 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1782 ordered_extent
->file_offset
+
1783 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1785 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1786 &ordered_extent
->list
);
1788 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1789 if (!ret
|| !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1790 ret
= btrfs_update_inode(trans
, root
, inode
);
1795 if (root
!= root
->fs_info
->tree_root
)
1796 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1799 btrfs_end_transaction_nolock(trans
, root
);
1801 btrfs_end_transaction(trans
, root
);
1805 btrfs_put_ordered_extent(ordered_extent
);
1806 /* once for the tree */
1807 btrfs_put_ordered_extent(ordered_extent
);
1812 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1813 struct extent_state
*state
, int uptodate
)
1815 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1817 ClearPagePrivate2(page
);
1818 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1822 * When IO fails, either with EIO or csum verification fails, we
1823 * try other mirrors that might have a good copy of the data. This
1824 * io_failure_record is used to record state as we go through all the
1825 * mirrors. If another mirror has good data, the page is set up to date
1826 * and things continue. If a good mirror can't be found, the original
1827 * bio end_io callback is called to indicate things have failed.
1829 struct io_failure_record
{
1834 unsigned long bio_flags
;
1838 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1839 struct page
*page
, u64 start
, u64 end
,
1840 struct extent_state
*state
)
1842 struct io_failure_record
*failrec
= NULL
;
1844 struct extent_map
*em
;
1845 struct inode
*inode
= page
->mapping
->host
;
1846 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1847 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1854 ret
= get_state_private(failure_tree
, start
, &private);
1856 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1859 failrec
->start
= start
;
1860 failrec
->len
= end
- start
+ 1;
1861 failrec
->last_mirror
= 0;
1862 failrec
->bio_flags
= 0;
1864 read_lock(&em_tree
->lock
);
1865 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1866 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1867 free_extent_map(em
);
1870 read_unlock(&em_tree
->lock
);
1872 if (IS_ERR_OR_NULL(em
)) {
1876 logical
= start
- em
->start
;
1877 logical
= em
->block_start
+ logical
;
1878 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1879 logical
= em
->block_start
;
1880 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1881 extent_set_compress_type(&failrec
->bio_flags
,
1884 failrec
->logical
= logical
;
1885 free_extent_map(em
);
1886 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1887 EXTENT_DIRTY
, GFP_NOFS
);
1888 set_state_private(failure_tree
, start
,
1889 (u64
)(unsigned long)failrec
);
1891 failrec
= (struct io_failure_record
*)(unsigned long)private;
1893 num_copies
= btrfs_num_copies(
1894 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1895 failrec
->logical
, failrec
->len
);
1896 failrec
->last_mirror
++;
1898 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1899 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1902 if (state
&& state
->start
!= failrec
->start
)
1904 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1906 if (!state
|| failrec
->last_mirror
> num_copies
) {
1907 set_state_private(failure_tree
, failrec
->start
, 0);
1908 clear_extent_bits(failure_tree
, failrec
->start
,
1909 failrec
->start
+ failrec
->len
- 1,
1910 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1914 bio
= bio_alloc(GFP_NOFS
, 1);
1915 bio
->bi_private
= state
;
1916 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1917 bio
->bi_sector
= failrec
->logical
>> 9;
1918 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1921 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1922 if (failed_bio
->bi_rw
& REQ_WRITE
)
1927 ret
= BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1928 failrec
->last_mirror
,
1929 failrec
->bio_flags
, 0);
1934 * each time an IO finishes, we do a fast check in the IO failure tree
1935 * to see if we need to process or clean up an io_failure_record
1937 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1940 u64 private_failure
;
1941 struct io_failure_record
*failure
;
1945 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1946 (u64
)-1, 1, EXTENT_DIRTY
, 0)) {
1947 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1948 start
, &private_failure
);
1950 failure
= (struct io_failure_record
*)(unsigned long)
1952 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1954 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1956 failure
->start
+ failure
->len
- 1,
1957 EXTENT_DIRTY
| EXTENT_LOCKED
,
1966 * when reads are done, we need to check csums to verify the data is correct
1967 * if there's a match, we allow the bio to finish. If not, we go through
1968 * the io_failure_record routines to find good copies
1970 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1971 struct extent_state
*state
)
1973 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1974 struct inode
*inode
= page
->mapping
->host
;
1975 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1977 u64
private = ~(u32
)0;
1979 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1982 if (PageChecked(page
)) {
1983 ClearPageChecked(page
);
1987 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1990 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1991 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1992 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1997 if (state
&& state
->start
== start
) {
1998 private = state
->private;
2001 ret
= get_state_private(io_tree
, start
, &private);
2003 kaddr
= kmap_atomic(page
, KM_USER0
);
2007 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
2008 btrfs_csum_final(csum
, (char *)&csum
);
2009 if (csum
!= private)
2012 kunmap_atomic(kaddr
, KM_USER0
);
2014 /* if the io failure tree for this inode is non-empty,
2015 * check to see if we've recovered from a failed IO
2017 btrfs_clean_io_failures(inode
, start
);
2021 printk_ratelimited(KERN_INFO
"btrfs csum failed ino %llu off %llu csum %u "
2023 (unsigned long long)btrfs_ino(page
->mapping
->host
),
2024 (unsigned long long)start
, csum
,
2025 (unsigned long long)private);
2026 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2027 flush_dcache_page(page
);
2028 kunmap_atomic(kaddr
, KM_USER0
);
2034 struct delayed_iput
{
2035 struct list_head list
;
2036 struct inode
*inode
;
2039 void btrfs_add_delayed_iput(struct inode
*inode
)
2041 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2042 struct delayed_iput
*delayed
;
2044 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2047 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2048 delayed
->inode
= inode
;
2050 spin_lock(&fs_info
->delayed_iput_lock
);
2051 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2052 spin_unlock(&fs_info
->delayed_iput_lock
);
2055 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2058 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2059 struct delayed_iput
*delayed
;
2062 spin_lock(&fs_info
->delayed_iput_lock
);
2063 empty
= list_empty(&fs_info
->delayed_iputs
);
2064 spin_unlock(&fs_info
->delayed_iput_lock
);
2068 down_read(&root
->fs_info
->cleanup_work_sem
);
2069 spin_lock(&fs_info
->delayed_iput_lock
);
2070 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2071 spin_unlock(&fs_info
->delayed_iput_lock
);
2073 while (!list_empty(&list
)) {
2074 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2075 list_del(&delayed
->list
);
2076 iput(delayed
->inode
);
2079 up_read(&root
->fs_info
->cleanup_work_sem
);
2082 enum btrfs_orphan_cleanup_state
{
2083 ORPHAN_CLEANUP_STARTED
= 1,
2084 ORPHAN_CLEANUP_DONE
= 2,
2088 * This is called in transaction commmit time. If there are no orphan
2089 * files in the subvolume, it removes orphan item and frees block_rsv
2092 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2093 struct btrfs_root
*root
)
2097 if (!list_empty(&root
->orphan_list
) ||
2098 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2101 if (root
->orphan_item_inserted
&&
2102 btrfs_root_refs(&root
->root_item
) > 0) {
2103 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2104 root
->root_key
.objectid
);
2106 root
->orphan_item_inserted
= 0;
2109 if (root
->orphan_block_rsv
) {
2110 WARN_ON(root
->orphan_block_rsv
->size
> 0);
2111 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
2112 root
->orphan_block_rsv
= NULL
;
2117 * This creates an orphan entry for the given inode in case something goes
2118 * wrong in the middle of an unlink/truncate.
2120 * NOTE: caller of this function should reserve 5 units of metadata for
2123 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2125 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2126 struct btrfs_block_rsv
*block_rsv
= NULL
;
2131 if (!root
->orphan_block_rsv
) {
2132 block_rsv
= btrfs_alloc_block_rsv(root
);
2137 spin_lock(&root
->orphan_lock
);
2138 if (!root
->orphan_block_rsv
) {
2139 root
->orphan_block_rsv
= block_rsv
;
2140 } else if (block_rsv
) {
2141 btrfs_free_block_rsv(root
, block_rsv
);
2145 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2146 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2149 * For proper ENOSPC handling, we should do orphan
2150 * cleanup when mounting. But this introduces backward
2151 * compatibility issue.
2153 if (!xchg(&root
->orphan_item_inserted
, 1))
2161 if (!BTRFS_I(inode
)->orphan_meta_reserved
) {
2162 BTRFS_I(inode
)->orphan_meta_reserved
= 1;
2165 spin_unlock(&root
->orphan_lock
);
2167 /* grab metadata reservation from transaction handle */
2169 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2173 /* insert an orphan item to track this unlinked/truncated file */
2175 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
2179 /* insert an orphan item to track subvolume contains orphan files */
2181 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2182 root
->root_key
.objectid
);
2189 * We have done the truncate/delete so we can go ahead and remove the orphan
2190 * item for this particular inode.
2192 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2194 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2195 int delete_item
= 0;
2196 int release_rsv
= 0;
2199 spin_lock(&root
->orphan_lock
);
2200 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2201 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2205 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2206 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2209 spin_unlock(&root
->orphan_lock
);
2211 if (trans
&& delete_item
) {
2212 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
2217 btrfs_orphan_release_metadata(inode
);
2223 * this cleans up any orphans that may be left on the list from the last use
2226 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2228 struct btrfs_path
*path
;
2229 struct extent_buffer
*leaf
;
2230 struct btrfs_key key
, found_key
;
2231 struct btrfs_trans_handle
*trans
;
2232 struct inode
*inode
;
2233 u64 last_objectid
= 0;
2234 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2236 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2239 path
= btrfs_alloc_path();
2246 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2247 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2248 key
.offset
= (u64
)-1;
2251 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2256 * if ret == 0 means we found what we were searching for, which
2257 * is weird, but possible, so only screw with path if we didn't
2258 * find the key and see if we have stuff that matches
2262 if (path
->slots
[0] == 0)
2267 /* pull out the item */
2268 leaf
= path
->nodes
[0];
2269 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2271 /* make sure the item matches what we want */
2272 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2274 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2277 /* release the path since we're done with it */
2278 btrfs_release_path(path
);
2281 * this is where we are basically btrfs_lookup, without the
2282 * crossing root thing. we store the inode number in the
2283 * offset of the orphan item.
2286 if (found_key
.offset
== last_objectid
) {
2287 printk(KERN_ERR
"btrfs: Error removing orphan entry, "
2288 "stopping orphan cleanup\n");
2293 last_objectid
= found_key
.offset
;
2295 found_key
.objectid
= found_key
.offset
;
2296 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2297 found_key
.offset
= 0;
2298 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2299 ret
= PTR_RET(inode
);
2300 if (ret
&& ret
!= -ESTALE
)
2304 * Inode is already gone but the orphan item is still there,
2305 * kill the orphan item.
2307 if (ret
== -ESTALE
) {
2308 trans
= btrfs_start_transaction(root
, 1);
2309 if (IS_ERR(trans
)) {
2310 ret
= PTR_ERR(trans
);
2313 ret
= btrfs_del_orphan_item(trans
, root
,
2314 found_key
.objectid
);
2316 btrfs_end_transaction(trans
, root
);
2321 * add this inode to the orphan list so btrfs_orphan_del does
2322 * the proper thing when we hit it
2324 spin_lock(&root
->orphan_lock
);
2325 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2326 spin_unlock(&root
->orphan_lock
);
2328 /* if we have links, this was a truncate, lets do that */
2329 if (inode
->i_nlink
) {
2330 if (!S_ISREG(inode
->i_mode
)) {
2336 ret
= btrfs_truncate(inode
);
2341 /* this will do delete_inode and everything for us */
2346 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2348 if (root
->orphan_block_rsv
)
2349 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2352 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2353 trans
= btrfs_join_transaction(root
);
2355 btrfs_end_transaction(trans
, root
);
2359 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2361 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2365 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2366 btrfs_free_path(path
);
2371 * very simple check to peek ahead in the leaf looking for xattrs. If we
2372 * don't find any xattrs, we know there can't be any acls.
2374 * slot is the slot the inode is in, objectid is the objectid of the inode
2376 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2377 int slot
, u64 objectid
)
2379 u32 nritems
= btrfs_header_nritems(leaf
);
2380 struct btrfs_key found_key
;
2384 while (slot
< nritems
) {
2385 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2387 /* we found a different objectid, there must not be acls */
2388 if (found_key
.objectid
!= objectid
)
2391 /* we found an xattr, assume we've got an acl */
2392 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2396 * we found a key greater than an xattr key, there can't
2397 * be any acls later on
2399 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2406 * it goes inode, inode backrefs, xattrs, extents,
2407 * so if there are a ton of hard links to an inode there can
2408 * be a lot of backrefs. Don't waste time searching too hard,
2409 * this is just an optimization
2414 /* we hit the end of the leaf before we found an xattr or
2415 * something larger than an xattr. We have to assume the inode
2422 * read an inode from the btree into the in-memory inode
2424 static void btrfs_read_locked_inode(struct inode
*inode
)
2426 struct btrfs_path
*path
;
2427 struct extent_buffer
*leaf
;
2428 struct btrfs_inode_item
*inode_item
;
2429 struct btrfs_timespec
*tspec
;
2430 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2431 struct btrfs_key location
;
2435 bool filled
= false;
2437 ret
= btrfs_fill_inode(inode
, &rdev
);
2441 path
= btrfs_alloc_path();
2445 path
->leave_spinning
= 1;
2446 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2448 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2452 leaf
= path
->nodes
[0];
2457 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2458 struct btrfs_inode_item
);
2459 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2460 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2461 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2462 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2463 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2465 tspec
= btrfs_inode_atime(inode_item
);
2466 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2467 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2469 tspec
= btrfs_inode_mtime(inode_item
);
2470 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2471 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2473 tspec
= btrfs_inode_ctime(inode_item
);
2474 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2475 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2477 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2478 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2479 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2480 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2482 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2484 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2485 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2488 * try to precache a NULL acl entry for files that don't have
2489 * any xattrs or acls
2491 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
2494 cache_no_acl(inode
);
2496 btrfs_free_path(path
);
2498 switch (inode
->i_mode
& S_IFMT
) {
2500 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2501 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2502 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2503 inode
->i_fop
= &btrfs_file_operations
;
2504 inode
->i_op
= &btrfs_file_inode_operations
;
2507 inode
->i_fop
= &btrfs_dir_file_operations
;
2508 if (root
== root
->fs_info
->tree_root
)
2509 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2511 inode
->i_op
= &btrfs_dir_inode_operations
;
2514 inode
->i_op
= &btrfs_symlink_inode_operations
;
2515 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2516 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2519 inode
->i_op
= &btrfs_special_inode_operations
;
2520 init_special_inode(inode
, inode
->i_mode
, rdev
);
2524 btrfs_update_iflags(inode
);
2528 btrfs_free_path(path
);
2529 make_bad_inode(inode
);
2533 * given a leaf and an inode, copy the inode fields into the leaf
2535 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2536 struct extent_buffer
*leaf
,
2537 struct btrfs_inode_item
*item
,
2538 struct inode
*inode
)
2540 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2541 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2542 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2543 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2544 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2546 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2547 inode
->i_atime
.tv_sec
);
2548 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2549 inode
->i_atime
.tv_nsec
);
2551 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2552 inode
->i_mtime
.tv_sec
);
2553 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2554 inode
->i_mtime
.tv_nsec
);
2556 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2557 inode
->i_ctime
.tv_sec
);
2558 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2559 inode
->i_ctime
.tv_nsec
);
2561 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2562 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2563 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2564 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2565 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2566 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2567 btrfs_set_inode_block_group(leaf
, item
, 0);
2571 * copy everything in the in-memory inode into the btree.
2573 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2574 struct btrfs_root
*root
, struct inode
*inode
)
2576 struct btrfs_inode_item
*inode_item
;
2577 struct btrfs_path
*path
;
2578 struct extent_buffer
*leaf
;
2582 * If the inode is a free space inode, we can deadlock during commit
2583 * if we put it into the delayed code.
2585 * The data relocation inode should also be directly updated
2588 if (!btrfs_is_free_space_inode(root
, inode
)
2589 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
2590 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
2592 btrfs_set_inode_last_trans(trans
, inode
);
2596 path
= btrfs_alloc_path();
2600 path
->leave_spinning
= 1;
2601 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
2609 btrfs_unlock_up_safe(path
, 1);
2610 leaf
= path
->nodes
[0];
2611 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2612 struct btrfs_inode_item
);
2614 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2615 btrfs_mark_buffer_dirty(leaf
);
2616 btrfs_set_inode_last_trans(trans
, inode
);
2619 btrfs_free_path(path
);
2624 * unlink helper that gets used here in inode.c and in the tree logging
2625 * recovery code. It remove a link in a directory with a given name, and
2626 * also drops the back refs in the inode to the directory
2628 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2629 struct btrfs_root
*root
,
2630 struct inode
*dir
, struct inode
*inode
,
2631 const char *name
, int name_len
)
2633 struct btrfs_path
*path
;
2635 struct extent_buffer
*leaf
;
2636 struct btrfs_dir_item
*di
;
2637 struct btrfs_key key
;
2639 u64 ino
= btrfs_ino(inode
);
2640 u64 dir_ino
= btrfs_ino(dir
);
2642 path
= btrfs_alloc_path();
2648 path
->leave_spinning
= 1;
2649 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2650 name
, name_len
, -1);
2659 leaf
= path
->nodes
[0];
2660 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2661 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2664 btrfs_release_path(path
);
2666 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
2669 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2670 "inode %llu parent %llu\n", name_len
, name
,
2671 (unsigned long long)ino
, (unsigned long long)dir_ino
);
2675 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2679 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2681 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2683 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2688 btrfs_free_path(path
);
2692 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2693 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2694 btrfs_update_inode(trans
, root
, dir
);
2699 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2700 struct btrfs_root
*root
,
2701 struct inode
*dir
, struct inode
*inode
,
2702 const char *name
, int name_len
)
2705 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2707 btrfs_drop_nlink(inode
);
2708 ret
= btrfs_update_inode(trans
, root
, inode
);
2714 /* helper to check if there is any shared block in the path */
2715 static int check_path_shared(struct btrfs_root
*root
,
2716 struct btrfs_path
*path
)
2718 struct extent_buffer
*eb
;
2722 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2725 if (!path
->nodes
[level
])
2727 eb
= path
->nodes
[level
];
2728 if (!btrfs_block_can_be_shared(root
, eb
))
2730 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2739 * helper to start transaction for unlink and rmdir.
2741 * unlink and rmdir are special in btrfs, they do not always free space.
2742 * so in enospc case, we should make sure they will free space before
2743 * allowing them to use the global metadata reservation.
2745 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2746 struct dentry
*dentry
)
2748 struct btrfs_trans_handle
*trans
;
2749 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2750 struct btrfs_path
*path
;
2751 struct btrfs_inode_ref
*ref
;
2752 struct btrfs_dir_item
*di
;
2753 struct inode
*inode
= dentry
->d_inode
;
2758 u64 ino
= btrfs_ino(inode
);
2759 u64 dir_ino
= btrfs_ino(dir
);
2762 * 1 for the possible orphan item
2763 * 1 for the dir item
2764 * 1 for the dir index
2765 * 1 for the inode ref
2766 * 1 for the inode ref in the tree log
2767 * 2 for the dir entries in the log
2770 trans
= btrfs_start_transaction(root
, 8);
2771 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2774 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2775 return ERR_PTR(-ENOSPC
);
2777 /* check if there is someone else holds reference */
2778 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2779 return ERR_PTR(-ENOSPC
);
2781 if (atomic_read(&inode
->i_count
) > 2)
2782 return ERR_PTR(-ENOSPC
);
2784 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2785 return ERR_PTR(-ENOSPC
);
2787 path
= btrfs_alloc_path();
2789 root
->fs_info
->enospc_unlink
= 0;
2790 return ERR_PTR(-ENOMEM
);
2793 trans
= btrfs_start_transaction(root
, 0);
2794 if (IS_ERR(trans
)) {
2795 btrfs_free_path(path
);
2796 root
->fs_info
->enospc_unlink
= 0;
2800 path
->skip_locking
= 1;
2801 path
->search_commit_root
= 1;
2803 ret
= btrfs_lookup_inode(trans
, root
, path
,
2804 &BTRFS_I(dir
)->location
, 0);
2810 if (check_path_shared(root
, path
))
2815 btrfs_release_path(path
);
2817 ret
= btrfs_lookup_inode(trans
, root
, path
,
2818 &BTRFS_I(inode
)->location
, 0);
2824 if (check_path_shared(root
, path
))
2829 btrfs_release_path(path
);
2831 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2832 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2839 if (check_path_shared(root
, path
))
2841 btrfs_release_path(path
);
2849 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2850 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2856 if (check_path_shared(root
, path
))
2862 btrfs_release_path(path
);
2864 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2865 dentry
->d_name
.name
, dentry
->d_name
.len
,
2872 if (check_path_shared(root
, path
))
2874 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2875 btrfs_release_path(path
);
2878 * This is a commit root search, if we can lookup inode item and other
2879 * relative items in the commit root, it means the transaction of
2880 * dir/file creation has been committed, and the dir index item that we
2881 * delay to insert has also been inserted into the commit root. So
2882 * we needn't worry about the delayed insertion of the dir index item
2885 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
2886 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2891 BUG_ON(ret
== -ENOENT
);
2892 if (check_path_shared(root
, path
))
2897 btrfs_free_path(path
);
2899 btrfs_end_transaction(trans
, root
);
2900 root
->fs_info
->enospc_unlink
= 0;
2901 return ERR_PTR(err
);
2904 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2908 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2909 struct btrfs_root
*root
)
2911 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2912 BUG_ON(!root
->fs_info
->enospc_unlink
);
2913 root
->fs_info
->enospc_unlink
= 0;
2915 btrfs_end_transaction_throttle(trans
, root
);
2918 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2920 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2921 struct btrfs_trans_handle
*trans
;
2922 struct inode
*inode
= dentry
->d_inode
;
2924 unsigned long nr
= 0;
2926 trans
= __unlink_start_trans(dir
, dentry
);
2928 return PTR_ERR(trans
);
2930 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2932 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2933 dentry
->d_name
.name
, dentry
->d_name
.len
);
2937 if (inode
->i_nlink
== 0) {
2938 ret
= btrfs_orphan_add(trans
, inode
);
2944 nr
= trans
->blocks_used
;
2945 __unlink_end_trans(trans
, root
);
2946 btrfs_btree_balance_dirty(root
, nr
);
2950 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2951 struct btrfs_root
*root
,
2952 struct inode
*dir
, u64 objectid
,
2953 const char *name
, int name_len
)
2955 struct btrfs_path
*path
;
2956 struct extent_buffer
*leaf
;
2957 struct btrfs_dir_item
*di
;
2958 struct btrfs_key key
;
2961 u64 dir_ino
= btrfs_ino(dir
);
2963 path
= btrfs_alloc_path();
2967 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2968 name
, name_len
, -1);
2969 BUG_ON(IS_ERR_OR_NULL(di
));
2971 leaf
= path
->nodes
[0];
2972 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2973 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
2974 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2976 btrfs_release_path(path
);
2978 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
2979 objectid
, root
->root_key
.objectid
,
2980 dir_ino
, &index
, name
, name_len
);
2982 BUG_ON(ret
!= -ENOENT
);
2983 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
2985 BUG_ON(IS_ERR_OR_NULL(di
));
2987 leaf
= path
->nodes
[0];
2988 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2989 btrfs_release_path(path
);
2992 btrfs_release_path(path
);
2994 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2997 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2998 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2999 ret
= btrfs_update_inode(trans
, root
, dir
);
3002 btrfs_free_path(path
);
3006 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3008 struct inode
*inode
= dentry
->d_inode
;
3010 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3011 struct btrfs_trans_handle
*trans
;
3012 unsigned long nr
= 0;
3014 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3015 btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3018 trans
= __unlink_start_trans(dir
, dentry
);
3020 return PTR_ERR(trans
);
3022 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3023 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3024 BTRFS_I(inode
)->location
.objectid
,
3025 dentry
->d_name
.name
,
3026 dentry
->d_name
.len
);
3030 err
= btrfs_orphan_add(trans
, inode
);
3034 /* now the directory is empty */
3035 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3036 dentry
->d_name
.name
, dentry
->d_name
.len
);
3038 btrfs_i_size_write(inode
, 0);
3040 nr
= trans
->blocks_used
;
3041 __unlink_end_trans(trans
, root
);
3042 btrfs_btree_balance_dirty(root
, nr
);
3048 * this can truncate away extent items, csum items and directory items.
3049 * It starts at a high offset and removes keys until it can't find
3050 * any higher than new_size
3052 * csum items that cross the new i_size are truncated to the new size
3055 * min_type is the minimum key type to truncate down to. If set to 0, this
3056 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3058 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3059 struct btrfs_root
*root
,
3060 struct inode
*inode
,
3061 u64 new_size
, u32 min_type
)
3063 struct btrfs_path
*path
;
3064 struct extent_buffer
*leaf
;
3065 struct btrfs_file_extent_item
*fi
;
3066 struct btrfs_key key
;
3067 struct btrfs_key found_key
;
3068 u64 extent_start
= 0;
3069 u64 extent_num_bytes
= 0;
3070 u64 extent_offset
= 0;
3072 u64 mask
= root
->sectorsize
- 1;
3073 u32 found_type
= (u8
)-1;
3076 int pending_del_nr
= 0;
3077 int pending_del_slot
= 0;
3078 int extent_type
= -1;
3082 u64 ino
= btrfs_ino(inode
);
3084 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3086 path
= btrfs_alloc_path();
3091 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3092 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3095 * This function is also used to drop the items in the log tree before
3096 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3097 * it is used to drop the loged items. So we shouldn't kill the delayed
3100 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3101 btrfs_kill_delayed_inode_items(inode
);
3104 key
.offset
= (u64
)-1;
3108 path
->leave_spinning
= 1;
3109 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3116 /* there are no items in the tree for us to truncate, we're
3119 if (path
->slots
[0] == 0)
3126 leaf
= path
->nodes
[0];
3127 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3128 found_type
= btrfs_key_type(&found_key
);
3131 if (found_key
.objectid
!= ino
)
3134 if (found_type
< min_type
)
3137 item_end
= found_key
.offset
;
3138 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3139 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3140 struct btrfs_file_extent_item
);
3141 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3142 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3143 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3144 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3146 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3148 btrfs_file_extent_num_bytes(leaf
, fi
);
3149 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3150 item_end
+= btrfs_file_extent_inline_len(leaf
,
3155 if (found_type
> min_type
) {
3158 if (item_end
< new_size
)
3160 if (found_key
.offset
>= new_size
)
3166 /* FIXME, shrink the extent if the ref count is only 1 */
3167 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3170 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3172 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3173 if (!del_item
&& !encoding
) {
3174 u64 orig_num_bytes
=
3175 btrfs_file_extent_num_bytes(leaf
, fi
);
3176 extent_num_bytes
= new_size
-
3177 found_key
.offset
+ root
->sectorsize
- 1;
3178 extent_num_bytes
= extent_num_bytes
&
3179 ~((u64
)root
->sectorsize
- 1);
3180 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3182 num_dec
= (orig_num_bytes
-
3184 if (root
->ref_cows
&& extent_start
!= 0)
3185 inode_sub_bytes(inode
, num_dec
);
3186 btrfs_mark_buffer_dirty(leaf
);
3189 btrfs_file_extent_disk_num_bytes(leaf
,
3191 extent_offset
= found_key
.offset
-
3192 btrfs_file_extent_offset(leaf
, fi
);
3194 /* FIXME blocksize != 4096 */
3195 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3196 if (extent_start
!= 0) {
3199 inode_sub_bytes(inode
, num_dec
);
3202 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3204 * we can't truncate inline items that have had
3208 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3209 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3210 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3211 u32 size
= new_size
- found_key
.offset
;
3213 if (root
->ref_cows
) {
3214 inode_sub_bytes(inode
, item_end
+ 1 -
3218 btrfs_file_extent_calc_inline_size(size
);
3219 ret
= btrfs_truncate_item(trans
, root
, path
,
3221 } else if (root
->ref_cows
) {
3222 inode_sub_bytes(inode
, item_end
+ 1 -
3228 if (!pending_del_nr
) {
3229 /* no pending yet, add ourselves */
3230 pending_del_slot
= path
->slots
[0];
3232 } else if (pending_del_nr
&&
3233 path
->slots
[0] + 1 == pending_del_slot
) {
3234 /* hop on the pending chunk */
3236 pending_del_slot
= path
->slots
[0];
3243 if (found_extent
&& (root
->ref_cows
||
3244 root
== root
->fs_info
->tree_root
)) {
3245 btrfs_set_path_blocking(path
);
3246 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3247 extent_num_bytes
, 0,
3248 btrfs_header_owner(leaf
),
3249 ino
, extent_offset
);
3253 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3256 if (path
->slots
[0] == 0 ||
3257 path
->slots
[0] != pending_del_slot
) {
3258 if (root
->ref_cows
&&
3259 BTRFS_I(inode
)->location
.objectid
!=
3260 BTRFS_FREE_INO_OBJECTID
) {
3264 if (pending_del_nr
) {
3265 ret
= btrfs_del_items(trans
, root
, path
,
3271 btrfs_release_path(path
);
3278 if (pending_del_nr
) {
3279 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3283 btrfs_free_path(path
);
3288 * taken from block_truncate_page, but does cow as it zeros out
3289 * any bytes left in the last page in the file.
3291 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3293 struct inode
*inode
= mapping
->host
;
3294 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3295 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3296 struct btrfs_ordered_extent
*ordered
;
3297 struct extent_state
*cached_state
= NULL
;
3299 u32 blocksize
= root
->sectorsize
;
3300 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3301 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3303 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
3308 if ((offset
& (blocksize
- 1)) == 0)
3310 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3316 page
= find_or_create_page(mapping
, index
, mask
);
3318 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3322 page_start
= page_offset(page
);
3323 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3325 if (!PageUptodate(page
)) {
3326 ret
= btrfs_readpage(NULL
, page
);
3328 if (page
->mapping
!= mapping
) {
3330 page_cache_release(page
);
3333 if (!PageUptodate(page
)) {
3338 wait_on_page_writeback(page
);
3340 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3342 set_page_extent_mapped(page
);
3344 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3346 unlock_extent_cached(io_tree
, page_start
, page_end
,
3347 &cached_state
, GFP_NOFS
);
3349 page_cache_release(page
);
3350 btrfs_start_ordered_extent(inode
, ordered
, 1);
3351 btrfs_put_ordered_extent(ordered
);
3355 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3356 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3357 0, 0, &cached_state
, GFP_NOFS
);
3359 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3362 unlock_extent_cached(io_tree
, page_start
, page_end
,
3363 &cached_state
, GFP_NOFS
);
3368 if (offset
!= PAGE_CACHE_SIZE
) {
3370 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3371 flush_dcache_page(page
);
3374 ClearPageChecked(page
);
3375 set_page_dirty(page
);
3376 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3381 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3383 page_cache_release(page
);
3389 * This function puts in dummy file extents for the area we're creating a hole
3390 * for. So if we are truncating this file to a larger size we need to insert
3391 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3392 * the range between oldsize and size
3394 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3396 struct btrfs_trans_handle
*trans
;
3397 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3398 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3399 struct extent_map
*em
= NULL
;
3400 struct extent_state
*cached_state
= NULL
;
3401 u64 mask
= root
->sectorsize
- 1;
3402 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3403 u64 block_end
= (size
+ mask
) & ~mask
;
3409 if (size
<= hole_start
)
3413 struct btrfs_ordered_extent
*ordered
;
3414 btrfs_wait_ordered_range(inode
, hole_start
,
3415 block_end
- hole_start
);
3416 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3417 &cached_state
, GFP_NOFS
);
3418 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3421 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3422 &cached_state
, GFP_NOFS
);
3423 btrfs_put_ordered_extent(ordered
);
3426 cur_offset
= hole_start
;
3428 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3429 block_end
- cur_offset
, 0);
3430 BUG_ON(IS_ERR_OR_NULL(em
));
3431 last_byte
= min(extent_map_end(em
), block_end
);
3432 last_byte
= (last_byte
+ mask
) & ~mask
;
3433 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3435 hole_size
= last_byte
- cur_offset
;
3437 trans
= btrfs_start_transaction(root
, 2);
3438 if (IS_ERR(trans
)) {
3439 err
= PTR_ERR(trans
);
3443 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3444 cur_offset
+ hole_size
,
3447 btrfs_end_transaction(trans
, root
);
3451 err
= btrfs_insert_file_extent(trans
, root
,
3452 btrfs_ino(inode
), cur_offset
, 0,
3453 0, hole_size
, 0, hole_size
,
3456 btrfs_end_transaction(trans
, root
);
3460 btrfs_drop_extent_cache(inode
, hole_start
,
3463 btrfs_end_transaction(trans
, root
);
3465 free_extent_map(em
);
3467 cur_offset
= last_byte
;
3468 if (cur_offset
>= block_end
)
3472 free_extent_map(em
);
3473 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3478 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3480 loff_t oldsize
= i_size_read(inode
);
3483 if (newsize
== oldsize
)
3486 if (newsize
> oldsize
) {
3487 i_size_write(inode
, newsize
);
3488 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3489 truncate_pagecache(inode
, oldsize
, newsize
);
3490 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3492 btrfs_setsize(inode
, oldsize
);
3496 mark_inode_dirty(inode
);
3500 * We're truncating a file that used to have good data down to
3501 * zero. Make sure it gets into the ordered flush list so that
3502 * any new writes get down to disk quickly.
3505 BTRFS_I(inode
)->ordered_data_close
= 1;
3507 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3508 truncate_setsize(inode
, newsize
);
3509 ret
= btrfs_truncate(inode
);
3515 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3517 struct inode
*inode
= dentry
->d_inode
;
3518 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3521 if (btrfs_root_readonly(root
))
3524 err
= inode_change_ok(inode
, attr
);
3528 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3529 err
= btrfs_setsize(inode
, attr
->ia_size
);
3534 if (attr
->ia_valid
) {
3535 setattr_copy(inode
, attr
);
3536 mark_inode_dirty(inode
);
3538 if (attr
->ia_valid
& ATTR_MODE
)
3539 err
= btrfs_acl_chmod(inode
);
3545 void btrfs_evict_inode(struct inode
*inode
)
3547 struct btrfs_trans_handle
*trans
;
3548 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3549 struct btrfs_block_rsv
*rsv
, *global_rsv
;
3550 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
3554 trace_btrfs_inode_evict(inode
);
3556 truncate_inode_pages(&inode
->i_data
, 0);
3557 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3558 btrfs_is_free_space_inode(root
, inode
)))
3561 if (is_bad_inode(inode
)) {
3562 btrfs_orphan_del(NULL
, inode
);
3565 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3566 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3568 if (root
->fs_info
->log_root_recovering
) {
3569 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3573 if (inode
->i_nlink
> 0) {
3574 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3578 rsv
= btrfs_alloc_block_rsv(root
);
3580 btrfs_orphan_del(NULL
, inode
);
3583 rsv
->size
= min_size
;
3584 global_rsv
= &root
->fs_info
->global_block_rsv
;
3586 btrfs_i_size_write(inode
, 0);
3589 * This is a bit simpler than btrfs_truncate since
3591 * 1) We've already reserved our space for our orphan item in the
3593 * 2) We're going to delete the inode item, so we don't need to update
3596 * So we just need to reserve some slack space in case we add bytes when
3597 * doing the truncate.
3600 ret
= btrfs_block_rsv_check(root
, rsv
, min_size
, 0, 1);
3603 * Try and steal from the global reserve since we will
3604 * likely not use this space anyway, we want to try as
3605 * hard as possible to get this to work.
3608 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, min_size
);
3611 printk(KERN_WARNING
"Could not get space for a "
3612 "delete, will truncate on mount %d\n", ret
);
3613 btrfs_orphan_del(NULL
, inode
);
3614 btrfs_free_block_rsv(root
, rsv
);
3618 trans
= btrfs_start_transaction(root
, 0);
3619 if (IS_ERR(trans
)) {
3620 btrfs_orphan_del(NULL
, inode
);
3621 btrfs_free_block_rsv(root
, rsv
);
3625 trans
->block_rsv
= rsv
;
3627 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3631 nr
= trans
->blocks_used
;
3632 btrfs_end_transaction(trans
, root
);
3634 btrfs_btree_balance_dirty(root
, nr
);
3637 btrfs_free_block_rsv(root
, rsv
);
3640 trans
->block_rsv
= root
->orphan_block_rsv
;
3641 ret
= btrfs_orphan_del(trans
, inode
);
3645 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3646 if (!(root
== root
->fs_info
->tree_root
||
3647 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
3648 btrfs_return_ino(root
, btrfs_ino(inode
));
3650 nr
= trans
->blocks_used
;
3651 btrfs_end_transaction(trans
, root
);
3652 btrfs_btree_balance_dirty(root
, nr
);
3654 end_writeback(inode
);
3659 * this returns the key found in the dir entry in the location pointer.
3660 * If no dir entries were found, location->objectid is 0.
3662 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3663 struct btrfs_key
*location
)
3665 const char *name
= dentry
->d_name
.name
;
3666 int namelen
= dentry
->d_name
.len
;
3667 struct btrfs_dir_item
*di
;
3668 struct btrfs_path
*path
;
3669 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3672 path
= btrfs_alloc_path();
3676 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
3681 if (IS_ERR_OR_NULL(di
))
3684 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3686 btrfs_free_path(path
);
3689 location
->objectid
= 0;
3694 * when we hit a tree root in a directory, the btrfs part of the inode
3695 * needs to be changed to reflect the root directory of the tree root. This
3696 * is kind of like crossing a mount point.
3698 static int fixup_tree_root_location(struct btrfs_root
*root
,
3700 struct dentry
*dentry
,
3701 struct btrfs_key
*location
,
3702 struct btrfs_root
**sub_root
)
3704 struct btrfs_path
*path
;
3705 struct btrfs_root
*new_root
;
3706 struct btrfs_root_ref
*ref
;
3707 struct extent_buffer
*leaf
;
3711 path
= btrfs_alloc_path();
3718 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3719 BTRFS_I(dir
)->root
->root_key
.objectid
,
3720 location
->objectid
);
3727 leaf
= path
->nodes
[0];
3728 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3729 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
3730 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3733 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3734 (unsigned long)(ref
+ 1),
3735 dentry
->d_name
.len
);
3739 btrfs_release_path(path
);
3741 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3742 if (IS_ERR(new_root
)) {
3743 err
= PTR_ERR(new_root
);
3747 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3752 *sub_root
= new_root
;
3753 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3754 location
->type
= BTRFS_INODE_ITEM_KEY
;
3755 location
->offset
= 0;
3758 btrfs_free_path(path
);
3762 static void inode_tree_add(struct inode
*inode
)
3764 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3765 struct btrfs_inode
*entry
;
3767 struct rb_node
*parent
;
3768 u64 ino
= btrfs_ino(inode
);
3770 p
= &root
->inode_tree
.rb_node
;
3773 if (inode_unhashed(inode
))
3776 spin_lock(&root
->inode_lock
);
3779 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3781 if (ino
< btrfs_ino(&entry
->vfs_inode
))
3782 p
= &parent
->rb_left
;
3783 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
3784 p
= &parent
->rb_right
;
3786 WARN_ON(!(entry
->vfs_inode
.i_state
&
3787 (I_WILL_FREE
| I_FREEING
)));
3788 rb_erase(parent
, &root
->inode_tree
);
3789 RB_CLEAR_NODE(parent
);
3790 spin_unlock(&root
->inode_lock
);
3794 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3795 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3796 spin_unlock(&root
->inode_lock
);
3799 static void inode_tree_del(struct inode
*inode
)
3801 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3804 spin_lock(&root
->inode_lock
);
3805 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3806 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3807 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3808 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3810 spin_unlock(&root
->inode_lock
);
3813 * Free space cache has inodes in the tree root, but the tree root has a
3814 * root_refs of 0, so this could end up dropping the tree root as a
3815 * snapshot, so we need the extra !root->fs_info->tree_root check to
3816 * make sure we don't drop it.
3818 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3819 root
!= root
->fs_info
->tree_root
) {
3820 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3821 spin_lock(&root
->inode_lock
);
3822 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3823 spin_unlock(&root
->inode_lock
);
3825 btrfs_add_dead_root(root
);
3829 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3831 struct rb_node
*node
;
3832 struct rb_node
*prev
;
3833 struct btrfs_inode
*entry
;
3834 struct inode
*inode
;
3837 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3839 spin_lock(&root
->inode_lock
);
3841 node
= root
->inode_tree
.rb_node
;
3845 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3847 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
3848 node
= node
->rb_left
;
3849 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
3850 node
= node
->rb_right
;
3856 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
3857 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
3861 prev
= rb_next(prev
);
3865 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3866 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
3867 inode
= igrab(&entry
->vfs_inode
);
3869 spin_unlock(&root
->inode_lock
);
3870 if (atomic_read(&inode
->i_count
) > 1)
3871 d_prune_aliases(inode
);
3873 * btrfs_drop_inode will have it removed from
3874 * the inode cache when its usage count
3879 spin_lock(&root
->inode_lock
);
3883 if (cond_resched_lock(&root
->inode_lock
))
3886 node
= rb_next(node
);
3888 spin_unlock(&root
->inode_lock
);
3892 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3894 struct btrfs_iget_args
*args
= p
;
3895 inode
->i_ino
= args
->ino
;
3896 BTRFS_I(inode
)->root
= args
->root
;
3897 btrfs_set_inode_space_info(args
->root
, inode
);
3901 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3903 struct btrfs_iget_args
*args
= opaque
;
3904 return args
->ino
== btrfs_ino(inode
) &&
3905 args
->root
== BTRFS_I(inode
)->root
;
3908 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
3910 struct btrfs_root
*root
)
3912 struct inode
*inode
;
3913 struct btrfs_iget_args args
;
3914 args
.ino
= objectid
;
3917 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3918 btrfs_init_locked_inode
,
3923 /* Get an inode object given its location and corresponding root.
3924 * Returns in *is_new if the inode was read from disk
3926 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3927 struct btrfs_root
*root
, int *new)
3929 struct inode
*inode
;
3931 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3933 return ERR_PTR(-ENOMEM
);
3935 if (inode
->i_state
& I_NEW
) {
3936 BTRFS_I(inode
)->root
= root
;
3937 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3938 btrfs_read_locked_inode(inode
);
3939 if (!is_bad_inode(inode
)) {
3940 inode_tree_add(inode
);
3941 unlock_new_inode(inode
);
3945 unlock_new_inode(inode
);
3947 inode
= ERR_PTR(-ESTALE
);
3954 static struct inode
*new_simple_dir(struct super_block
*s
,
3955 struct btrfs_key
*key
,
3956 struct btrfs_root
*root
)
3958 struct inode
*inode
= new_inode(s
);
3961 return ERR_PTR(-ENOMEM
);
3963 BTRFS_I(inode
)->root
= root
;
3964 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
3965 BTRFS_I(inode
)->dummy_inode
= 1;
3967 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
3968 inode
->i_op
= &simple_dir_inode_operations
;
3969 inode
->i_fop
= &simple_dir_operations
;
3970 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
3971 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3976 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3978 struct inode
*inode
;
3979 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3980 struct btrfs_root
*sub_root
= root
;
3981 struct btrfs_key location
;
3985 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3986 return ERR_PTR(-ENAMETOOLONG
);
3988 if (unlikely(d_need_lookup(dentry
))) {
3989 memcpy(&location
, dentry
->d_fsdata
, sizeof(struct btrfs_key
));
3990 kfree(dentry
->d_fsdata
);
3991 dentry
->d_fsdata
= NULL
;
3992 /* This thing is hashed, drop it for now */
3995 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3999 return ERR_PTR(ret
);
4001 if (location
.objectid
== 0)
4004 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4005 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4009 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4011 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4012 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4013 &location
, &sub_root
);
4016 inode
= ERR_PTR(ret
);
4018 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4020 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4022 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4024 if (!IS_ERR(inode
) && root
!= sub_root
) {
4025 down_read(&root
->fs_info
->cleanup_work_sem
);
4026 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4027 ret
= btrfs_orphan_cleanup(sub_root
);
4028 up_read(&root
->fs_info
->cleanup_work_sem
);
4030 inode
= ERR_PTR(ret
);
4036 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4038 struct btrfs_root
*root
;
4040 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
4041 dentry
= dentry
->d_parent
;
4043 if (dentry
->d_inode
) {
4044 root
= BTRFS_I(dentry
->d_inode
)->root
;
4045 if (btrfs_root_refs(&root
->root_item
) == 0)
4051 static void btrfs_dentry_release(struct dentry
*dentry
)
4053 if (dentry
->d_fsdata
)
4054 kfree(dentry
->d_fsdata
);
4057 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4058 struct nameidata
*nd
)
4062 ret
= d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
4063 if (unlikely(d_need_lookup(dentry
))) {
4064 spin_lock(&dentry
->d_lock
);
4065 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
4066 spin_unlock(&dentry
->d_lock
);
4071 unsigned char btrfs_filetype_table
[] = {
4072 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4075 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4078 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4079 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4080 struct btrfs_item
*item
;
4081 struct btrfs_dir_item
*di
;
4082 struct btrfs_key key
;
4083 struct btrfs_key found_key
;
4084 struct btrfs_path
*path
;
4085 struct list_head ins_list
;
4086 struct list_head del_list
;
4089 struct extent_buffer
*leaf
;
4091 unsigned char d_type
;
4096 int key_type
= BTRFS_DIR_INDEX_KEY
;
4100 int is_curr
= 0; /* filp->f_pos points to the current index? */
4102 /* FIXME, use a real flag for deciding about the key type */
4103 if (root
->fs_info
->tree_root
== root
)
4104 key_type
= BTRFS_DIR_ITEM_KEY
;
4106 /* special case for "." */
4107 if (filp
->f_pos
== 0) {
4108 over
= filldir(dirent
, ".", 1,
4109 filp
->f_pos
, btrfs_ino(inode
), DT_DIR
);
4114 /* special case for .., just use the back ref */
4115 if (filp
->f_pos
== 1) {
4116 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4117 over
= filldir(dirent
, "..", 2,
4118 filp
->f_pos
, pino
, DT_DIR
);
4123 path
= btrfs_alloc_path();
4129 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4130 INIT_LIST_HEAD(&ins_list
);
4131 INIT_LIST_HEAD(&del_list
);
4132 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
4135 btrfs_set_key_type(&key
, key_type
);
4136 key
.offset
= filp
->f_pos
;
4137 key
.objectid
= btrfs_ino(inode
);
4139 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4144 leaf
= path
->nodes
[0];
4145 slot
= path
->slots
[0];
4146 if (slot
>= btrfs_header_nritems(leaf
)) {
4147 ret
= btrfs_next_leaf(root
, path
);
4155 item
= btrfs_item_nr(leaf
, slot
);
4156 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4158 if (found_key
.objectid
!= key
.objectid
)
4160 if (btrfs_key_type(&found_key
) != key_type
)
4162 if (found_key
.offset
< filp
->f_pos
)
4164 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
4165 btrfs_should_delete_dir_index(&del_list
,
4169 filp
->f_pos
= found_key
.offset
;
4172 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4174 di_total
= btrfs_item_size(leaf
, item
);
4176 while (di_cur
< di_total
) {
4177 struct btrfs_key location
;
4180 if (verify_dir_item(root
, leaf
, di
))
4183 name_len
= btrfs_dir_name_len(leaf
, di
);
4184 if (name_len
<= sizeof(tmp_name
)) {
4185 name_ptr
= tmp_name
;
4187 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4193 read_extent_buffer(leaf
, name_ptr
,
4194 (unsigned long)(di
+ 1), name_len
);
4196 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4197 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4201 q
.hash
= full_name_hash(q
.name
, q
.len
);
4202 tmp
= d_lookup(filp
->f_dentry
, &q
);
4204 struct btrfs_key
*newkey
;
4206 newkey
= kzalloc(sizeof(struct btrfs_key
),
4210 tmp
= d_alloc(filp
->f_dentry
, &q
);
4216 memcpy(newkey
, &location
,
4217 sizeof(struct btrfs_key
));
4218 tmp
->d_fsdata
= newkey
;
4219 tmp
->d_flags
|= DCACHE_NEED_LOOKUP
;
4226 /* is this a reference to our own snapshot? If so
4229 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4230 location
.objectid
== root
->root_key
.objectid
) {
4234 over
= filldir(dirent
, name_ptr
, name_len
,
4235 found_key
.offset
, location
.objectid
,
4239 if (name_ptr
!= tmp_name
)
4244 di_len
= btrfs_dir_name_len(leaf
, di
) +
4245 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4247 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4253 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4256 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
4262 /* Reached end of directory/root. Bump pos past the last item. */
4263 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4265 * 32-bit glibc will use getdents64, but then strtol -
4266 * so the last number we can serve is this.
4268 filp
->f_pos
= 0x7fffffff;
4274 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4275 btrfs_put_delayed_items(&ins_list
, &del_list
);
4276 btrfs_free_path(path
);
4280 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4282 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4283 struct btrfs_trans_handle
*trans
;
4285 bool nolock
= false;
4287 if (BTRFS_I(inode
)->dummy_inode
)
4290 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(root
, inode
))
4293 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4295 trans
= btrfs_join_transaction_nolock(root
);
4297 trans
= btrfs_join_transaction(root
);
4299 return PTR_ERR(trans
);
4301 ret
= btrfs_end_transaction_nolock(trans
, root
);
4303 ret
= btrfs_commit_transaction(trans
, root
);
4309 * This is somewhat expensive, updating the tree every time the
4310 * inode changes. But, it is most likely to find the inode in cache.
4311 * FIXME, needs more benchmarking...there are no reasons other than performance
4312 * to keep or drop this code.
4314 void btrfs_dirty_inode(struct inode
*inode
, int flags
)
4316 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4317 struct btrfs_trans_handle
*trans
;
4320 if (BTRFS_I(inode
)->dummy_inode
)
4323 trans
= btrfs_join_transaction(root
);
4324 BUG_ON(IS_ERR(trans
));
4326 ret
= btrfs_update_inode(trans
, root
, inode
);
4327 if (ret
&& ret
== -ENOSPC
) {
4328 /* whoops, lets try again with the full transaction */
4329 btrfs_end_transaction(trans
, root
);
4330 trans
= btrfs_start_transaction(root
, 1);
4331 if (IS_ERR(trans
)) {
4332 printk_ratelimited(KERN_ERR
"btrfs: fail to "
4333 "dirty inode %llu error %ld\n",
4334 (unsigned long long)btrfs_ino(inode
),
4339 ret
= btrfs_update_inode(trans
, root
, inode
);
4341 printk_ratelimited(KERN_ERR
"btrfs: fail to "
4342 "dirty inode %llu error %d\n",
4343 (unsigned long long)btrfs_ino(inode
),
4347 btrfs_end_transaction(trans
, root
);
4348 if (BTRFS_I(inode
)->delayed_node
)
4349 btrfs_balance_delayed_items(root
);
4353 * find the highest existing sequence number in a directory
4354 * and then set the in-memory index_cnt variable to reflect
4355 * free sequence numbers
4357 static int btrfs_set_inode_index_count(struct inode
*inode
)
4359 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4360 struct btrfs_key key
, found_key
;
4361 struct btrfs_path
*path
;
4362 struct extent_buffer
*leaf
;
4365 key
.objectid
= btrfs_ino(inode
);
4366 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4367 key
.offset
= (u64
)-1;
4369 path
= btrfs_alloc_path();
4373 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4376 /* FIXME: we should be able to handle this */
4382 * MAGIC NUMBER EXPLANATION:
4383 * since we search a directory based on f_pos we have to start at 2
4384 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4385 * else has to start at 2
4387 if (path
->slots
[0] == 0) {
4388 BTRFS_I(inode
)->index_cnt
= 2;
4394 leaf
= path
->nodes
[0];
4395 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4397 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4398 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4399 BTRFS_I(inode
)->index_cnt
= 2;
4403 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4405 btrfs_free_path(path
);
4410 * helper to find a free sequence number in a given directory. This current
4411 * code is very simple, later versions will do smarter things in the btree
4413 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4417 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4418 ret
= btrfs_inode_delayed_dir_index_count(dir
);
4420 ret
= btrfs_set_inode_index_count(dir
);
4426 *index
= BTRFS_I(dir
)->index_cnt
;
4427 BTRFS_I(dir
)->index_cnt
++;
4432 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4433 struct btrfs_root
*root
,
4435 const char *name
, int name_len
,
4436 u64 ref_objectid
, u64 objectid
, int mode
,
4439 struct inode
*inode
;
4440 struct btrfs_inode_item
*inode_item
;
4441 struct btrfs_key
*location
;
4442 struct btrfs_path
*path
;
4443 struct btrfs_inode_ref
*ref
;
4444 struct btrfs_key key
[2];
4450 path
= btrfs_alloc_path();
4452 return ERR_PTR(-ENOMEM
);
4454 inode
= new_inode(root
->fs_info
->sb
);
4456 btrfs_free_path(path
);
4457 return ERR_PTR(-ENOMEM
);
4461 * we have to initialize this early, so we can reclaim the inode
4462 * number if we fail afterwards in this function.
4464 inode
->i_ino
= objectid
;
4467 trace_btrfs_inode_request(dir
);
4469 ret
= btrfs_set_inode_index(dir
, index
);
4471 btrfs_free_path(path
);
4473 return ERR_PTR(ret
);
4477 * index_cnt is ignored for everything but a dir,
4478 * btrfs_get_inode_index_count has an explanation for the magic
4481 BTRFS_I(inode
)->index_cnt
= 2;
4482 BTRFS_I(inode
)->root
= root
;
4483 BTRFS_I(inode
)->generation
= trans
->transid
;
4484 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4485 btrfs_set_inode_space_info(root
, inode
);
4492 key
[0].objectid
= objectid
;
4493 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4496 key
[1].objectid
= objectid
;
4497 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4498 key
[1].offset
= ref_objectid
;
4500 sizes
[0] = sizeof(struct btrfs_inode_item
);
4501 sizes
[1] = name_len
+ sizeof(*ref
);
4503 path
->leave_spinning
= 1;
4504 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4508 inode_init_owner(inode
, dir
, mode
);
4509 inode_set_bytes(inode
, 0);
4510 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4511 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4512 struct btrfs_inode_item
);
4513 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4515 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4516 struct btrfs_inode_ref
);
4517 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4518 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4519 ptr
= (unsigned long)(ref
+ 1);
4520 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4522 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4523 btrfs_free_path(path
);
4525 location
= &BTRFS_I(inode
)->location
;
4526 location
->objectid
= objectid
;
4527 location
->offset
= 0;
4528 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4530 btrfs_inherit_iflags(inode
, dir
);
4532 if (S_ISREG(mode
)) {
4533 if (btrfs_test_opt(root
, NODATASUM
))
4534 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4535 if (btrfs_test_opt(root
, NODATACOW
) ||
4536 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4537 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4540 insert_inode_hash(inode
);
4541 inode_tree_add(inode
);
4543 trace_btrfs_inode_new(inode
);
4544 btrfs_set_inode_last_trans(trans
, inode
);
4549 BTRFS_I(dir
)->index_cnt
--;
4550 btrfs_free_path(path
);
4552 return ERR_PTR(ret
);
4555 static inline u8
btrfs_inode_type(struct inode
*inode
)
4557 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4561 * utility function to add 'inode' into 'parent_inode' with
4562 * a give name and a given sequence number.
4563 * if 'add_backref' is true, also insert a backref from the
4564 * inode to the parent directory.
4566 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4567 struct inode
*parent_inode
, struct inode
*inode
,
4568 const char *name
, int name_len
, int add_backref
, u64 index
)
4571 struct btrfs_key key
;
4572 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4573 u64 ino
= btrfs_ino(inode
);
4574 u64 parent_ino
= btrfs_ino(parent_inode
);
4576 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4577 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4580 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4584 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4585 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4586 key
.objectid
, root
->root_key
.objectid
,
4587 parent_ino
, index
, name
, name_len
);
4588 } else if (add_backref
) {
4589 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
4594 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4596 btrfs_inode_type(inode
), index
);
4599 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4601 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4602 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4607 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4608 struct inode
*dir
, struct dentry
*dentry
,
4609 struct inode
*inode
, int backref
, u64 index
)
4611 int err
= btrfs_add_link(trans
, dir
, inode
,
4612 dentry
->d_name
.name
, dentry
->d_name
.len
,
4615 d_instantiate(dentry
, inode
);
4623 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4624 int mode
, dev_t rdev
)
4626 struct btrfs_trans_handle
*trans
;
4627 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4628 struct inode
*inode
= NULL
;
4632 unsigned long nr
= 0;
4635 if (!new_valid_dev(rdev
))
4639 * 2 for inode item and ref
4641 * 1 for xattr if selinux is on
4643 trans
= btrfs_start_transaction(root
, 5);
4645 return PTR_ERR(trans
);
4647 err
= btrfs_find_free_ino(root
, &objectid
);
4651 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4652 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4654 if (IS_ERR(inode
)) {
4655 err
= PTR_ERR(inode
);
4659 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4665 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4669 inode
->i_op
= &btrfs_special_inode_operations
;
4670 init_special_inode(inode
, inode
->i_mode
, rdev
);
4671 btrfs_update_inode(trans
, root
, inode
);
4674 nr
= trans
->blocks_used
;
4675 btrfs_end_transaction_throttle(trans
, root
);
4676 btrfs_btree_balance_dirty(root
, nr
);
4678 inode_dec_link_count(inode
);
4684 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4685 int mode
, struct nameidata
*nd
)
4687 struct btrfs_trans_handle
*trans
;
4688 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4689 struct inode
*inode
= NULL
;
4692 unsigned long nr
= 0;
4697 * 2 for inode item and ref
4699 * 1 for xattr if selinux is on
4701 trans
= btrfs_start_transaction(root
, 5);
4703 return PTR_ERR(trans
);
4705 err
= btrfs_find_free_ino(root
, &objectid
);
4709 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4710 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4712 if (IS_ERR(inode
)) {
4713 err
= PTR_ERR(inode
);
4717 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4723 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4727 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4728 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4729 inode
->i_fop
= &btrfs_file_operations
;
4730 inode
->i_op
= &btrfs_file_inode_operations
;
4731 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4734 nr
= trans
->blocks_used
;
4735 btrfs_end_transaction_throttle(trans
, root
);
4737 inode_dec_link_count(inode
);
4740 btrfs_btree_balance_dirty(root
, nr
);
4744 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4745 struct dentry
*dentry
)
4747 struct btrfs_trans_handle
*trans
;
4748 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4749 struct inode
*inode
= old_dentry
->d_inode
;
4751 unsigned long nr
= 0;
4755 /* do not allow sys_link's with other subvols of the same device */
4756 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4759 if (inode
->i_nlink
== ~0U)
4762 err
= btrfs_set_inode_index(dir
, &index
);
4767 * 2 items for inode and inode ref
4768 * 2 items for dir items
4769 * 1 item for parent inode
4771 trans
= btrfs_start_transaction(root
, 5);
4772 if (IS_ERR(trans
)) {
4773 err
= PTR_ERR(trans
);
4777 btrfs_inc_nlink(inode
);
4778 inode
->i_ctime
= CURRENT_TIME
;
4781 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4786 struct dentry
*parent
= dentry
->d_parent
;
4787 err
= btrfs_update_inode(trans
, root
, inode
);
4789 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4792 nr
= trans
->blocks_used
;
4793 btrfs_end_transaction_throttle(trans
, root
);
4796 inode_dec_link_count(inode
);
4799 btrfs_btree_balance_dirty(root
, nr
);
4803 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4805 struct inode
*inode
= NULL
;
4806 struct btrfs_trans_handle
*trans
;
4807 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4809 int drop_on_err
= 0;
4812 unsigned long nr
= 1;
4815 * 2 items for inode and ref
4816 * 2 items for dir items
4817 * 1 for xattr if selinux is on
4819 trans
= btrfs_start_transaction(root
, 5);
4821 return PTR_ERR(trans
);
4823 err
= btrfs_find_free_ino(root
, &objectid
);
4827 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4828 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4829 S_IFDIR
| mode
, &index
);
4830 if (IS_ERR(inode
)) {
4831 err
= PTR_ERR(inode
);
4837 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4841 inode
->i_op
= &btrfs_dir_inode_operations
;
4842 inode
->i_fop
= &btrfs_dir_file_operations
;
4844 btrfs_i_size_write(inode
, 0);
4845 err
= btrfs_update_inode(trans
, root
, inode
);
4849 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4850 dentry
->d_name
.len
, 0, index
);
4854 d_instantiate(dentry
, inode
);
4858 nr
= trans
->blocks_used
;
4859 btrfs_end_transaction_throttle(trans
, root
);
4862 btrfs_btree_balance_dirty(root
, nr
);
4866 /* helper for btfs_get_extent. Given an existing extent in the tree,
4867 * and an extent that you want to insert, deal with overlap and insert
4868 * the new extent into the tree.
4870 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4871 struct extent_map
*existing
,
4872 struct extent_map
*em
,
4873 u64 map_start
, u64 map_len
)
4877 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4878 start_diff
= map_start
- em
->start
;
4879 em
->start
= map_start
;
4881 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4882 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4883 em
->block_start
+= start_diff
;
4884 em
->block_len
-= start_diff
;
4886 return add_extent_mapping(em_tree
, em
);
4889 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4890 struct inode
*inode
, struct page
*page
,
4891 size_t pg_offset
, u64 extent_offset
,
4892 struct btrfs_file_extent_item
*item
)
4895 struct extent_buffer
*leaf
= path
->nodes
[0];
4898 unsigned long inline_size
;
4902 WARN_ON(pg_offset
!= 0);
4903 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4904 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4905 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4906 btrfs_item_nr(leaf
, path
->slots
[0]));
4907 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4910 ptr
= btrfs_file_extent_inline_start(item
);
4912 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4914 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4915 ret
= btrfs_decompress(compress_type
, tmp
, page
,
4916 extent_offset
, inline_size
, max_size
);
4918 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4919 unsigned long copy_size
= min_t(u64
,
4920 PAGE_CACHE_SIZE
- pg_offset
,
4921 max_size
- extent_offset
);
4922 memset(kaddr
+ pg_offset
, 0, copy_size
);
4923 kunmap_atomic(kaddr
, KM_USER0
);
4930 * a bit scary, this does extent mapping from logical file offset to the disk.
4931 * the ugly parts come from merging extents from the disk with the in-ram
4932 * representation. This gets more complex because of the data=ordered code,
4933 * where the in-ram extents might be locked pending data=ordered completion.
4935 * This also copies inline extents directly into the page.
4938 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
4939 size_t pg_offset
, u64 start
, u64 len
,
4945 u64 extent_start
= 0;
4947 u64 objectid
= btrfs_ino(inode
);
4949 struct btrfs_path
*path
= NULL
;
4950 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4951 struct btrfs_file_extent_item
*item
;
4952 struct extent_buffer
*leaf
;
4953 struct btrfs_key found_key
;
4954 struct extent_map
*em
= NULL
;
4955 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4956 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4957 struct btrfs_trans_handle
*trans
= NULL
;
4961 read_lock(&em_tree
->lock
);
4962 em
= lookup_extent_mapping(em_tree
, start
, len
);
4964 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4965 read_unlock(&em_tree
->lock
);
4968 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
4969 free_extent_map(em
);
4970 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
4971 free_extent_map(em
);
4975 em
= alloc_extent_map();
4980 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4981 em
->start
= EXTENT_MAP_HOLE
;
4982 em
->orig_start
= EXTENT_MAP_HOLE
;
4984 em
->block_len
= (u64
)-1;
4987 path
= btrfs_alloc_path();
4993 * Chances are we'll be called again, so go ahead and do
4999 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5000 objectid
, start
, trans
!= NULL
);
5007 if (path
->slots
[0] == 0)
5012 leaf
= path
->nodes
[0];
5013 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5014 struct btrfs_file_extent_item
);
5015 /* are we inside the extent that was found? */
5016 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5017 found_type
= btrfs_key_type(&found_key
);
5018 if (found_key
.objectid
!= objectid
||
5019 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5023 found_type
= btrfs_file_extent_type(leaf
, item
);
5024 extent_start
= found_key
.offset
;
5025 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5026 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5027 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5028 extent_end
= extent_start
+
5029 btrfs_file_extent_num_bytes(leaf
, item
);
5030 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5032 size
= btrfs_file_extent_inline_len(leaf
, item
);
5033 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5034 ~((u64
)root
->sectorsize
- 1);
5037 if (start
>= extent_end
) {
5039 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5040 ret
= btrfs_next_leaf(root
, path
);
5047 leaf
= path
->nodes
[0];
5049 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5050 if (found_key
.objectid
!= objectid
||
5051 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5053 if (start
+ len
<= found_key
.offset
)
5056 em
->len
= found_key
.offset
- start
;
5060 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5061 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5062 em
->start
= extent_start
;
5063 em
->len
= extent_end
- extent_start
;
5064 em
->orig_start
= extent_start
-
5065 btrfs_file_extent_offset(leaf
, item
);
5066 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5068 em
->block_start
= EXTENT_MAP_HOLE
;
5071 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5072 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5073 em
->compress_type
= compress_type
;
5074 em
->block_start
= bytenr
;
5075 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5078 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5079 em
->block_start
= bytenr
;
5080 em
->block_len
= em
->len
;
5081 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5082 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5085 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5089 size_t extent_offset
;
5092 em
->block_start
= EXTENT_MAP_INLINE
;
5093 if (!page
|| create
) {
5094 em
->start
= extent_start
;
5095 em
->len
= extent_end
- extent_start
;
5099 size
= btrfs_file_extent_inline_len(leaf
, item
);
5100 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5101 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5102 size
- extent_offset
);
5103 em
->start
= extent_start
+ extent_offset
;
5104 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5105 ~((u64
)root
->sectorsize
- 1);
5106 em
->orig_start
= EXTENT_MAP_INLINE
;
5107 if (compress_type
) {
5108 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5109 em
->compress_type
= compress_type
;
5111 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5112 if (create
== 0 && !PageUptodate(page
)) {
5113 if (btrfs_file_extent_compression(leaf
, item
) !=
5114 BTRFS_COMPRESS_NONE
) {
5115 ret
= uncompress_inline(path
, inode
, page
,
5117 extent_offset
, item
);
5121 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5123 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5124 memset(map
+ pg_offset
+ copy_size
, 0,
5125 PAGE_CACHE_SIZE
- pg_offset
-
5130 flush_dcache_page(page
);
5131 } else if (create
&& PageUptodate(page
)) {
5135 free_extent_map(em
);
5138 btrfs_release_path(path
);
5139 trans
= btrfs_join_transaction(root
);
5142 return ERR_CAST(trans
);
5146 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5149 btrfs_mark_buffer_dirty(leaf
);
5151 set_extent_uptodate(io_tree
, em
->start
,
5152 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5155 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5162 em
->block_start
= EXTENT_MAP_HOLE
;
5163 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5165 btrfs_release_path(path
);
5166 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5167 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5168 "[%llu %llu]\n", (unsigned long long)em
->start
,
5169 (unsigned long long)em
->len
,
5170 (unsigned long long)start
,
5171 (unsigned long long)len
);
5177 write_lock(&em_tree
->lock
);
5178 ret
= add_extent_mapping(em_tree
, em
);
5179 /* it is possible that someone inserted the extent into the tree
5180 * while we had the lock dropped. It is also possible that
5181 * an overlapping map exists in the tree
5183 if (ret
== -EEXIST
) {
5184 struct extent_map
*existing
;
5188 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5189 if (existing
&& (existing
->start
> start
||
5190 existing
->start
+ existing
->len
<= start
)) {
5191 free_extent_map(existing
);
5195 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5198 err
= merge_extent_mapping(em_tree
, existing
,
5201 free_extent_map(existing
);
5203 free_extent_map(em
);
5208 free_extent_map(em
);
5212 free_extent_map(em
);
5217 write_unlock(&em_tree
->lock
);
5220 trace_btrfs_get_extent(root
, em
);
5223 btrfs_free_path(path
);
5225 ret
= btrfs_end_transaction(trans
, root
);
5230 free_extent_map(em
);
5231 return ERR_PTR(err
);
5236 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5237 size_t pg_offset
, u64 start
, u64 len
,
5240 struct extent_map
*em
;
5241 struct extent_map
*hole_em
= NULL
;
5242 u64 range_start
= start
;
5248 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5253 * if our em maps to a hole, there might
5254 * actually be delalloc bytes behind it
5256 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5262 /* check to see if we've wrapped (len == -1 or similar) */
5271 /* ok, we didn't find anything, lets look for delalloc */
5272 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5273 end
, len
, EXTENT_DELALLOC
, 1);
5274 found_end
= range_start
+ found
;
5275 if (found_end
< range_start
)
5276 found_end
= (u64
)-1;
5279 * we didn't find anything useful, return
5280 * the original results from get_extent()
5282 if (range_start
> end
|| found_end
<= start
) {
5288 /* adjust the range_start to make sure it doesn't
5289 * go backwards from the start they passed in
5291 range_start
= max(start
,range_start
);
5292 found
= found_end
- range_start
;
5295 u64 hole_start
= start
;
5298 em
= alloc_extent_map();
5304 * when btrfs_get_extent can't find anything it
5305 * returns one huge hole
5307 * make sure what it found really fits our range, and
5308 * adjust to make sure it is based on the start from
5312 u64 calc_end
= extent_map_end(hole_em
);
5314 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5315 free_extent_map(hole_em
);
5318 hole_start
= max(hole_em
->start
, start
);
5319 hole_len
= calc_end
- hole_start
;
5323 if (hole_em
&& range_start
> hole_start
) {
5324 /* our hole starts before our delalloc, so we
5325 * have to return just the parts of the hole
5326 * that go until the delalloc starts
5328 em
->len
= min(hole_len
,
5329 range_start
- hole_start
);
5330 em
->start
= hole_start
;
5331 em
->orig_start
= hole_start
;
5333 * don't adjust block start at all,
5334 * it is fixed at EXTENT_MAP_HOLE
5336 em
->block_start
= hole_em
->block_start
;
5337 em
->block_len
= hole_len
;
5339 em
->start
= range_start
;
5341 em
->orig_start
= range_start
;
5342 em
->block_start
= EXTENT_MAP_DELALLOC
;
5343 em
->block_len
= found
;
5345 } else if (hole_em
) {
5350 free_extent_map(hole_em
);
5352 free_extent_map(em
);
5353 return ERR_PTR(err
);
5358 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5359 struct extent_map
*em
,
5362 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5363 struct btrfs_trans_handle
*trans
;
5364 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5365 struct btrfs_key ins
;
5368 bool insert
= false;
5371 * Ok if the extent map we looked up is a hole and is for the exact
5372 * range we want, there is no reason to allocate a new one, however if
5373 * it is not right then we need to free this one and drop the cache for
5376 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5378 free_extent_map(em
);
5381 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5384 trans
= btrfs_join_transaction(root
);
5386 return ERR_CAST(trans
);
5388 if (start
<= BTRFS_I(inode
)->disk_i_size
&& len
< 64 * 1024)
5389 btrfs_add_inode_defrag(trans
, inode
);
5391 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5393 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5394 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5395 alloc_hint
, (u64
)-1, &ins
, 1);
5402 em
= alloc_extent_map();
5404 em
= ERR_PTR(-ENOMEM
);
5410 em
->orig_start
= em
->start
;
5411 em
->len
= ins
.offset
;
5413 em
->block_start
= ins
.objectid
;
5414 em
->block_len
= ins
.offset
;
5415 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5418 * We need to do this because if we're using the original em we searched
5419 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5422 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5425 write_lock(&em_tree
->lock
);
5426 ret
= add_extent_mapping(em_tree
, em
);
5427 write_unlock(&em_tree
->lock
);
5430 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5433 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5434 ins
.offset
, ins
.offset
, 0);
5436 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5440 btrfs_end_transaction(trans
, root
);
5445 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5446 * block must be cow'd
5448 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5449 struct inode
*inode
, u64 offset
, u64 len
)
5451 struct btrfs_path
*path
;
5453 struct extent_buffer
*leaf
;
5454 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5455 struct btrfs_file_extent_item
*fi
;
5456 struct btrfs_key key
;
5464 path
= btrfs_alloc_path();
5468 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
5473 slot
= path
->slots
[0];
5476 /* can't find the item, must cow */
5483 leaf
= path
->nodes
[0];
5484 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5485 if (key
.objectid
!= btrfs_ino(inode
) ||
5486 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5487 /* not our file or wrong item type, must cow */
5491 if (key
.offset
> offset
) {
5492 /* Wrong offset, must cow */
5496 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5497 found_type
= btrfs_file_extent_type(leaf
, fi
);
5498 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5499 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5500 /* not a regular extent, must cow */
5503 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5504 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5506 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5507 if (extent_end
< offset
+ len
) {
5508 /* extent doesn't include our full range, must cow */
5512 if (btrfs_extent_readonly(root
, disk_bytenr
))
5516 * look for other files referencing this extent, if we
5517 * find any we must cow
5519 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
5520 key
.offset
- backref_offset
, disk_bytenr
))
5524 * adjust disk_bytenr and num_bytes to cover just the bytes
5525 * in this extent we are about to write. If there
5526 * are any csums in that range we have to cow in order
5527 * to keep the csums correct
5529 disk_bytenr
+= backref_offset
;
5530 disk_bytenr
+= offset
- key
.offset
;
5531 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5532 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5535 * all of the above have passed, it is safe to overwrite this extent
5540 btrfs_free_path(path
);
5544 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5545 struct buffer_head
*bh_result
, int create
)
5547 struct extent_map
*em
;
5548 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5549 u64 start
= iblock
<< inode
->i_blkbits
;
5550 u64 len
= bh_result
->b_size
;
5551 struct btrfs_trans_handle
*trans
;
5553 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5558 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5559 * io. INLINE is special, and we could probably kludge it in here, but
5560 * it's still buffered so for safety lets just fall back to the generic
5563 * For COMPRESSED we _have_ to read the entire extent in so we can
5564 * decompress it, so there will be buffering required no matter what we
5565 * do, so go ahead and fallback to buffered.
5567 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5568 * to buffered IO. Don't blame me, this is the price we pay for using
5571 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5572 em
->block_start
== EXTENT_MAP_INLINE
) {
5573 free_extent_map(em
);
5577 /* Just a good old fashioned hole, return */
5578 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5579 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5580 free_extent_map(em
);
5581 /* DIO will do one hole at a time, so just unlock a sector */
5582 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5583 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5588 * We don't allocate a new extent in the following cases
5590 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5592 * 2) The extent is marked as PREALLOC. We're good to go here and can
5593 * just use the extent.
5597 len
= em
->len
- (start
- em
->start
);
5601 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5602 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5603 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5608 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5609 type
= BTRFS_ORDERED_PREALLOC
;
5611 type
= BTRFS_ORDERED_NOCOW
;
5612 len
= min(len
, em
->len
- (start
- em
->start
));
5613 block_start
= em
->block_start
+ (start
- em
->start
);
5616 * we're not going to log anything, but we do need
5617 * to make sure the current transaction stays open
5618 * while we look for nocow cross refs
5620 trans
= btrfs_join_transaction(root
);
5624 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5625 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5626 block_start
, len
, len
, type
);
5627 btrfs_end_transaction(trans
, root
);
5629 free_extent_map(em
);
5634 btrfs_end_transaction(trans
, root
);
5638 * this will cow the extent, reset the len in case we changed
5641 len
= bh_result
->b_size
;
5642 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
5645 len
= min(len
, em
->len
- (start
- em
->start
));
5647 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5648 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5651 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5653 bh_result
->b_size
= len
;
5654 bh_result
->b_bdev
= em
->bdev
;
5655 set_buffer_mapped(bh_result
);
5656 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5657 set_buffer_new(bh_result
);
5659 free_extent_map(em
);
5664 struct btrfs_dio_private
{
5665 struct inode
*inode
;
5672 /* number of bios pending for this dio */
5673 atomic_t pending_bios
;
5678 struct bio
*orig_bio
;
5681 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5683 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5684 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5685 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5686 struct inode
*inode
= dip
->inode
;
5687 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5689 u32
*private = dip
->csums
;
5691 start
= dip
->logical_offset
;
5693 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5694 struct page
*page
= bvec
->bv_page
;
5697 unsigned long flags
;
5699 local_irq_save(flags
);
5700 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5701 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5702 csum
, bvec
->bv_len
);
5703 btrfs_csum_final(csum
, (char *)&csum
);
5704 kunmap_atomic(kaddr
, KM_IRQ0
);
5705 local_irq_restore(flags
);
5707 flush_dcache_page(bvec
->bv_page
);
5708 if (csum
!= *private) {
5709 printk(KERN_ERR
"btrfs csum failed ino %llu off"
5710 " %llu csum %u private %u\n",
5711 (unsigned long long)btrfs_ino(inode
),
5712 (unsigned long long)start
,
5718 start
+= bvec
->bv_len
;
5721 } while (bvec
<= bvec_end
);
5723 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5724 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5725 bio
->bi_private
= dip
->private;
5730 /* If we had a csum failure make sure to clear the uptodate flag */
5732 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5733 dio_end_io(bio
, err
);
5736 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5738 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5739 struct inode
*inode
= dip
->inode
;
5740 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5741 struct btrfs_trans_handle
*trans
;
5742 struct btrfs_ordered_extent
*ordered
= NULL
;
5743 struct extent_state
*cached_state
= NULL
;
5744 u64 ordered_offset
= dip
->logical_offset
;
5745 u64 ordered_bytes
= dip
->bytes
;
5751 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5759 trans
= btrfs_join_transaction(root
);
5760 if (IS_ERR(trans
)) {
5764 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5766 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5767 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5769 ret
= btrfs_update_inode(trans
, root
, inode
);
5774 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5775 ordered
->file_offset
+ ordered
->len
- 1, 0,
5776 &cached_state
, GFP_NOFS
);
5778 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5779 ret
= btrfs_mark_extent_written(trans
, inode
,
5780 ordered
->file_offset
,
5781 ordered
->file_offset
+
5788 ret
= insert_reserved_file_extent(trans
, inode
,
5789 ordered
->file_offset
,
5795 BTRFS_FILE_EXTENT_REG
);
5796 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5797 ordered
->file_offset
, ordered
->len
);
5805 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5806 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5807 if (!ret
|| !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
))
5808 btrfs_update_inode(trans
, root
, inode
);
5811 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5812 ordered
->file_offset
+ ordered
->len
- 1,
5813 &cached_state
, GFP_NOFS
);
5815 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5816 btrfs_end_transaction(trans
, root
);
5817 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5818 btrfs_put_ordered_extent(ordered
);
5819 btrfs_put_ordered_extent(ordered
);
5823 * our bio might span multiple ordered extents. If we haven't
5824 * completed the accounting for the whole dio, go back and try again
5826 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5827 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5832 bio
->bi_private
= dip
->private;
5837 /* If we had an error make sure to clear the uptodate flag */
5839 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5840 dio_end_io(bio
, err
);
5843 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5844 struct bio
*bio
, int mirror_num
,
5845 unsigned long bio_flags
, u64 offset
)
5848 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5849 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5854 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5856 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5859 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
5860 "sector %#Lx len %u err no %d\n",
5861 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
5862 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5866 * before atomic variable goto zero, we must make sure
5867 * dip->errors is perceived to be set.
5869 smp_mb__before_atomic_dec();
5872 /* if there are more bios still pending for this dio, just exit */
5873 if (!atomic_dec_and_test(&dip
->pending_bios
))
5877 bio_io_error(dip
->orig_bio
);
5879 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5880 bio_endio(dip
->orig_bio
, 0);
5886 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5887 u64 first_sector
, gfp_t gfp_flags
)
5889 int nr_vecs
= bio_get_nr_vecs(bdev
);
5890 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5893 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5894 int rw
, u64 file_offset
, int skip_sum
,
5895 u32
*csums
, int async_submit
)
5897 int write
= rw
& REQ_WRITE
;
5898 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5902 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5909 if (write
&& async_submit
) {
5910 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5911 inode
, rw
, bio
, 0, 0,
5913 __btrfs_submit_bio_start_direct_io
,
5914 __btrfs_submit_bio_done
);
5918 * If we aren't doing async submit, calculate the csum of the
5921 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
5924 } else if (!skip_sum
) {
5925 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5926 file_offset
, csums
);
5932 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
5938 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5941 struct inode
*inode
= dip
->inode
;
5942 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5943 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5945 struct bio
*orig_bio
= dip
->orig_bio
;
5946 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5947 u64 start_sector
= orig_bio
->bi_sector
;
5948 u64 file_offset
= dip
->logical_offset
;
5952 u32
*csums
= dip
->csums
;
5954 int async_submit
= 0;
5955 int write
= rw
& REQ_WRITE
;
5957 map_length
= orig_bio
->bi_size
;
5958 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5959 &map_length
, NULL
, 0);
5965 if (map_length
>= orig_bio
->bi_size
) {
5971 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
5974 bio
->bi_private
= dip
;
5975 bio
->bi_end_io
= btrfs_end_dio_bio
;
5976 atomic_inc(&dip
->pending_bios
);
5978 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
5979 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
5980 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5981 bvec
->bv_offset
) < bvec
->bv_len
)) {
5983 * inc the count before we submit the bio so
5984 * we know the end IO handler won't happen before
5985 * we inc the count. Otherwise, the dip might get freed
5986 * before we're done setting it up
5988 atomic_inc(&dip
->pending_bios
);
5989 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
5990 file_offset
, skip_sum
,
5991 csums
, async_submit
);
5994 atomic_dec(&dip
->pending_bios
);
5998 /* Write's use the ordered csums */
5999 if (!write
&& !skip_sum
)
6000 csums
= csums
+ nr_pages
;
6001 start_sector
+= submit_len
>> 9;
6002 file_offset
+= submit_len
;
6007 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6008 start_sector
, GFP_NOFS
);
6011 bio
->bi_private
= dip
;
6012 bio
->bi_end_io
= btrfs_end_dio_bio
;
6014 map_length
= orig_bio
->bi_size
;
6015 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6016 &map_length
, NULL
, 0);
6022 submit_len
+= bvec
->bv_len
;
6029 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6030 csums
, async_submit
);
6038 * before atomic variable goto zero, we must
6039 * make sure dip->errors is perceived to be set.
6041 smp_mb__before_atomic_dec();
6042 if (atomic_dec_and_test(&dip
->pending_bios
))
6043 bio_io_error(dip
->orig_bio
);
6045 /* bio_end_io() will handle error, so we needn't return it */
6049 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6052 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6053 struct btrfs_dio_private
*dip
;
6054 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6056 int write
= rw
& REQ_WRITE
;
6059 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6061 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6068 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6069 if (!write
&& !skip_sum
) {
6070 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6078 dip
->private = bio
->bi_private
;
6080 dip
->logical_offset
= file_offset
;
6084 dip
->bytes
+= bvec
->bv_len
;
6086 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6088 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6089 bio
->bi_private
= dip
;
6091 dip
->orig_bio
= bio
;
6092 atomic_set(&dip
->pending_bios
, 0);
6095 bio
->bi_end_io
= btrfs_endio_direct_write
;
6097 bio
->bi_end_io
= btrfs_endio_direct_read
;
6099 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6104 * If this is a write, we need to clean up the reserved space and kill
6105 * the ordered extent.
6108 struct btrfs_ordered_extent
*ordered
;
6109 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6110 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6111 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6112 btrfs_free_reserved_extent(root
, ordered
->start
,
6114 btrfs_put_ordered_extent(ordered
);
6115 btrfs_put_ordered_extent(ordered
);
6117 bio_endio(bio
, ret
);
6120 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6121 const struct iovec
*iov
, loff_t offset
,
6122 unsigned long nr_segs
)
6128 unsigned blocksize_mask
= root
->sectorsize
- 1;
6129 ssize_t retval
= -EINVAL
;
6130 loff_t end
= offset
;
6132 if (offset
& blocksize_mask
)
6135 /* Check the memory alignment. Blocks cannot straddle pages */
6136 for (seg
= 0; seg
< nr_segs
; seg
++) {
6137 addr
= (unsigned long)iov
[seg
].iov_base
;
6138 size
= iov
[seg
].iov_len
;
6140 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6143 /* If this is a write we don't need to check anymore */
6148 * Check to make sure we don't have duplicate iov_base's in this
6149 * iovec, if so return EINVAL, otherwise we'll get csum errors
6150 * when reading back.
6152 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6153 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6161 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6162 const struct iovec
*iov
, loff_t offset
,
6163 unsigned long nr_segs
)
6165 struct file
*file
= iocb
->ki_filp
;
6166 struct inode
*inode
= file
->f_mapping
->host
;
6167 struct btrfs_ordered_extent
*ordered
;
6168 struct extent_state
*cached_state
= NULL
;
6169 u64 lockstart
, lockend
;
6171 int writing
= rw
& WRITE
;
6173 size_t count
= iov_length(iov
, nr_segs
);
6175 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6181 lockend
= offset
+ count
- 1;
6184 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6190 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6191 0, &cached_state
, GFP_NOFS
);
6193 * We're concerned with the entire range that we're going to be
6194 * doing DIO to, so we need to make sure theres no ordered
6195 * extents in this range.
6197 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6198 lockend
- lockstart
+ 1);
6201 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6202 &cached_state
, GFP_NOFS
);
6203 btrfs_start_ordered_extent(inode
, ordered
, 1);
6204 btrfs_put_ordered_extent(ordered
);
6209 * we don't use btrfs_set_extent_delalloc because we don't want
6210 * the dirty or uptodate bits
6213 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6214 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6215 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6218 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6219 lockend
, EXTENT_LOCKED
| write_bits
,
6220 1, 0, &cached_state
, GFP_NOFS
);
6225 free_extent_state(cached_state
);
6226 cached_state
= NULL
;
6228 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6229 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6230 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6231 btrfs_submit_direct
, 0);
6233 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6234 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6235 offset
+ iov_length(iov
, nr_segs
) - 1,
6236 EXTENT_LOCKED
| write_bits
, 1, 0,
6237 &cached_state
, GFP_NOFS
);
6238 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6240 * We're falling back to buffered, unlock the section we didn't
6243 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6244 offset
+ iov_length(iov
, nr_segs
) - 1,
6245 EXTENT_LOCKED
| write_bits
, 1, 0,
6246 &cached_state
, GFP_NOFS
);
6249 free_extent_state(cached_state
);
6253 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6254 __u64 start
, __u64 len
)
6256 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6259 int btrfs_readpage(struct file
*file
, struct page
*page
)
6261 struct extent_io_tree
*tree
;
6262 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6263 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6266 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6268 struct extent_io_tree
*tree
;
6271 if (current
->flags
& PF_MEMALLOC
) {
6272 redirty_page_for_writepage(wbc
, page
);
6276 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6277 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6280 int btrfs_writepages(struct address_space
*mapping
,
6281 struct writeback_control
*wbc
)
6283 struct extent_io_tree
*tree
;
6285 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6286 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6290 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6291 struct list_head
*pages
, unsigned nr_pages
)
6293 struct extent_io_tree
*tree
;
6294 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6295 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6298 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6300 struct extent_io_tree
*tree
;
6301 struct extent_map_tree
*map
;
6304 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6305 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6306 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6308 ClearPagePrivate(page
);
6309 set_page_private(page
, 0);
6310 page_cache_release(page
);
6315 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6317 if (PageWriteback(page
) || PageDirty(page
))
6319 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6322 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6324 struct extent_io_tree
*tree
;
6325 struct btrfs_ordered_extent
*ordered
;
6326 struct extent_state
*cached_state
= NULL
;
6327 u64 page_start
= page_offset(page
);
6328 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6332 * we have the page locked, so new writeback can't start,
6333 * and the dirty bit won't be cleared while we are here.
6335 * Wait for IO on this page so that we can safely clear
6336 * the PagePrivate2 bit and do ordered accounting
6338 wait_on_page_writeback(page
);
6340 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6342 btrfs_releasepage(page
, GFP_NOFS
);
6345 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6347 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6351 * IO on this page will never be started, so we need
6352 * to account for any ordered extents now
6354 clear_extent_bit(tree
, page_start
, page_end
,
6355 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6356 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6357 &cached_state
, GFP_NOFS
);
6359 * whoever cleared the private bit is responsible
6360 * for the finish_ordered_io
6362 if (TestClearPagePrivate2(page
)) {
6363 btrfs_finish_ordered_io(page
->mapping
->host
,
6364 page_start
, page_end
);
6366 btrfs_put_ordered_extent(ordered
);
6367 cached_state
= NULL
;
6368 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6371 clear_extent_bit(tree
, page_start
, page_end
,
6372 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6373 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6374 __btrfs_releasepage(page
, GFP_NOFS
);
6376 ClearPageChecked(page
);
6377 if (PagePrivate(page
)) {
6378 ClearPagePrivate(page
);
6379 set_page_private(page
, 0);
6380 page_cache_release(page
);
6385 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6386 * called from a page fault handler when a page is first dirtied. Hence we must
6387 * be careful to check for EOF conditions here. We set the page up correctly
6388 * for a written page which means we get ENOSPC checking when writing into
6389 * holes and correct delalloc and unwritten extent mapping on filesystems that
6390 * support these features.
6392 * We are not allowed to take the i_mutex here so we have to play games to
6393 * protect against truncate races as the page could now be beyond EOF. Because
6394 * vmtruncate() writes the inode size before removing pages, once we have the
6395 * page lock we can determine safely if the page is beyond EOF. If it is not
6396 * beyond EOF, then the page is guaranteed safe against truncation until we
6399 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6401 struct page
*page
= vmf
->page
;
6402 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6403 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6404 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6405 struct btrfs_ordered_extent
*ordered
;
6406 struct extent_state
*cached_state
= NULL
;
6408 unsigned long zero_start
;
6414 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6418 else /* -ENOSPC, -EIO, etc */
6419 ret
= VM_FAULT_SIGBUS
;
6423 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6426 size
= i_size_read(inode
);
6427 page_start
= page_offset(page
);
6428 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6430 if ((page
->mapping
!= inode
->i_mapping
) ||
6431 (page_start
>= size
)) {
6432 /* page got truncated out from underneath us */
6435 wait_on_page_writeback(page
);
6437 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6439 set_page_extent_mapped(page
);
6442 * we can't set the delalloc bits if there are pending ordered
6443 * extents. Drop our locks and wait for them to finish
6445 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6447 unlock_extent_cached(io_tree
, page_start
, page_end
,
6448 &cached_state
, GFP_NOFS
);
6450 btrfs_start_ordered_extent(inode
, ordered
, 1);
6451 btrfs_put_ordered_extent(ordered
);
6456 * XXX - page_mkwrite gets called every time the page is dirtied, even
6457 * if it was already dirty, so for space accounting reasons we need to
6458 * clear any delalloc bits for the range we are fixing to save. There
6459 * is probably a better way to do this, but for now keep consistent with
6460 * prepare_pages in the normal write path.
6462 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6463 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6464 0, 0, &cached_state
, GFP_NOFS
);
6466 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6469 unlock_extent_cached(io_tree
, page_start
, page_end
,
6470 &cached_state
, GFP_NOFS
);
6471 ret
= VM_FAULT_SIGBUS
;
6476 /* page is wholly or partially inside EOF */
6477 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6478 zero_start
= size
& ~PAGE_CACHE_MASK
;
6480 zero_start
= PAGE_CACHE_SIZE
;
6482 if (zero_start
!= PAGE_CACHE_SIZE
) {
6484 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6485 flush_dcache_page(page
);
6488 ClearPageChecked(page
);
6489 set_page_dirty(page
);
6490 SetPageUptodate(page
);
6492 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6493 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6495 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6499 return VM_FAULT_LOCKED
;
6501 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6506 static int btrfs_truncate(struct inode
*inode
)
6508 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6509 struct btrfs_block_rsv
*rsv
;
6512 struct btrfs_trans_handle
*trans
;
6514 u64 mask
= root
->sectorsize
- 1;
6515 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
6517 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6521 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6522 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6525 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6526 * 3 things going on here
6528 * 1) We need to reserve space for our orphan item and the space to
6529 * delete our orphan item. Lord knows we don't want to have a dangling
6530 * orphan item because we didn't reserve space to remove it.
6532 * 2) We need to reserve space to update our inode.
6534 * 3) We need to have something to cache all the space that is going to
6535 * be free'd up by the truncate operation, but also have some slack
6536 * space reserved in case it uses space during the truncate (thank you
6537 * very much snapshotting).
6539 * And we need these to all be seperate. The fact is we can use alot of
6540 * space doing the truncate, and we have no earthly idea how much space
6541 * we will use, so we need the truncate reservation to be seperate so it
6542 * doesn't end up using space reserved for updating the inode or
6543 * removing the orphan item. We also need to be able to stop the
6544 * transaction and start a new one, which means we need to be able to
6545 * update the inode several times, and we have no idea of knowing how
6546 * many times that will be, so we can't just reserve 1 item for the
6547 * entirety of the opration, so that has to be done seperately as well.
6548 * Then there is the orphan item, which does indeed need to be held on
6549 * to for the whole operation, and we need nobody to touch this reserved
6550 * space except the orphan code.
6552 * So that leaves us with
6554 * 1) root->orphan_block_rsv - for the orphan deletion.
6555 * 2) rsv - for the truncate reservation, which we will steal from the
6556 * transaction reservation.
6557 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6558 * updating the inode.
6560 rsv
= btrfs_alloc_block_rsv(root
);
6563 rsv
->size
= min_size
;
6566 * 1 for the truncate slack space
6567 * 1 for the orphan item we're going to add
6568 * 1 for the orphan item deletion
6569 * 1 for updating the inode.
6571 trans
= btrfs_start_transaction(root
, 4);
6572 if (IS_ERR(trans
)) {
6573 err
= PTR_ERR(trans
);
6577 /* Migrate the slack space for the truncate to our reserve */
6578 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
6582 ret
= btrfs_orphan_add(trans
, inode
);
6584 btrfs_end_transaction(trans
, root
);
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
);
6609 ret
= btrfs_block_rsv_check(root
, rsv
, min_size
, 0, 1);
6612 * This can only happen with the original transaction we
6613 * started above, every other time we shouldn't have a
6614 * transaction started yet.
6623 /* Just need the 1 for updating the inode */
6624 trans
= btrfs_start_transaction(root
, 1);
6625 if (IS_ERR(trans
)) {
6626 err
= PTR_ERR(trans
);
6631 trans
->block_rsv
= rsv
;
6633 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6635 BTRFS_EXTENT_DATA_KEY
);
6636 if (ret
!= -EAGAIN
) {
6641 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6642 ret
= btrfs_update_inode(trans
, root
, inode
);
6648 nr
= trans
->blocks_used
;
6649 btrfs_end_transaction(trans
, root
);
6651 btrfs_btree_balance_dirty(root
, nr
);
6654 if (ret
== 0 && inode
->i_nlink
> 0) {
6655 trans
->block_rsv
= root
->orphan_block_rsv
;
6656 ret
= btrfs_orphan_del(trans
, inode
);
6659 } else if (ret
&& inode
->i_nlink
> 0) {
6661 * Failed to do the truncate, remove us from the in memory
6664 ret
= btrfs_orphan_del(NULL
, inode
);
6667 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6668 ret
= btrfs_update_inode(trans
, root
, inode
);
6672 nr
= trans
->blocks_used
;
6673 ret
= btrfs_end_transaction_throttle(trans
, root
);
6674 btrfs_btree_balance_dirty(root
, nr
);
6677 btrfs_free_block_rsv(root
, rsv
);
6686 * create a new subvolume directory/inode (helper for the ioctl).
6688 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6689 struct btrfs_root
*new_root
, u64 new_dirid
)
6691 struct inode
*inode
;
6695 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6696 new_dirid
, S_IFDIR
| 0700, &index
);
6698 return PTR_ERR(inode
);
6699 inode
->i_op
= &btrfs_dir_inode_operations
;
6700 inode
->i_fop
= &btrfs_dir_file_operations
;
6703 btrfs_i_size_write(inode
, 0);
6705 err
= btrfs_update_inode(trans
, new_root
, inode
);
6712 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6714 struct btrfs_inode
*ei
;
6715 struct inode
*inode
;
6717 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6722 ei
->space_info
= NULL
;
6726 ei
->last_sub_trans
= 0;
6727 ei
->logged_trans
= 0;
6728 ei
->delalloc_bytes
= 0;
6729 ei
->disk_i_size
= 0;
6732 ei
->index_cnt
= (u64
)-1;
6733 ei
->last_unlink_trans
= 0;
6735 spin_lock_init(&ei
->lock
);
6736 ei
->outstanding_extents
= 0;
6737 ei
->reserved_extents
= 0;
6739 ei
->ordered_data_close
= 0;
6740 ei
->orphan_meta_reserved
= 0;
6741 ei
->dummy_inode
= 0;
6743 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6745 ei
->delayed_node
= NULL
;
6747 inode
= &ei
->vfs_inode
;
6748 extent_map_tree_init(&ei
->extent_tree
);
6749 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
6750 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
6751 mutex_init(&ei
->log_mutex
);
6752 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6753 INIT_LIST_HEAD(&ei
->i_orphan
);
6754 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6755 INIT_LIST_HEAD(&ei
->ordered_operations
);
6756 RB_CLEAR_NODE(&ei
->rb_node
);
6761 static void btrfs_i_callback(struct rcu_head
*head
)
6763 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6764 INIT_LIST_HEAD(&inode
->i_dentry
);
6765 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6768 void btrfs_destroy_inode(struct inode
*inode
)
6770 struct btrfs_ordered_extent
*ordered
;
6771 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6773 WARN_ON(!list_empty(&inode
->i_dentry
));
6774 WARN_ON(inode
->i_data
.nrpages
);
6775 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
6776 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
6777 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
6778 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
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 !btrfs_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_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
= new_dentry
->d_parent
;
7072 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7073 btrfs_end_log_trans(root
);
7076 btrfs_end_transaction_throttle(trans
, root
);
7078 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7079 up_read(&root
->fs_info
->subvol_sem
);
7085 * some fairly slow code that needs optimization. This walks the list
7086 * of all the inodes with pending delalloc and forces them to disk.
7088 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7090 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7091 struct btrfs_inode
*binode
;
7092 struct inode
*inode
;
7094 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7097 spin_lock(&root
->fs_info
->delalloc_lock
);
7098 while (!list_empty(head
)) {
7099 binode
= list_entry(head
->next
, struct btrfs_inode
,
7101 inode
= igrab(&binode
->vfs_inode
);
7103 list_del_init(&binode
->delalloc_inodes
);
7104 spin_unlock(&root
->fs_info
->delalloc_lock
);
7106 filemap_flush(inode
->i_mapping
);
7108 btrfs_add_delayed_iput(inode
);
7113 spin_lock(&root
->fs_info
->delalloc_lock
);
7115 spin_unlock(&root
->fs_info
->delalloc_lock
);
7117 /* the filemap_flush will queue IO into the worker threads, but
7118 * we have to make sure the IO is actually started and that
7119 * ordered extents get created before we return
7121 atomic_inc(&root
->fs_info
->async_submit_draining
);
7122 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7123 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7124 wait_event(root
->fs_info
->async_submit_wait
,
7125 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7126 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7128 atomic_dec(&root
->fs_info
->async_submit_draining
);
7132 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7133 const char *symname
)
7135 struct btrfs_trans_handle
*trans
;
7136 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7137 struct btrfs_path
*path
;
7138 struct btrfs_key key
;
7139 struct inode
*inode
= NULL
;
7147 struct btrfs_file_extent_item
*ei
;
7148 struct extent_buffer
*leaf
;
7149 unsigned long nr
= 0;
7151 name_len
= strlen(symname
) + 1;
7152 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7153 return -ENAMETOOLONG
;
7156 * 2 items for inode item and ref
7157 * 2 items for dir items
7158 * 1 item for xattr if selinux is on
7160 trans
= btrfs_start_transaction(root
, 5);
7162 return PTR_ERR(trans
);
7164 err
= btrfs_find_free_ino(root
, &objectid
);
7168 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7169 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
7170 S_IFLNK
|S_IRWXUGO
, &index
);
7171 if (IS_ERR(inode
)) {
7172 err
= PTR_ERR(inode
);
7176 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7182 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7186 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7187 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7188 inode
->i_fop
= &btrfs_file_operations
;
7189 inode
->i_op
= &btrfs_file_inode_operations
;
7190 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7195 path
= btrfs_alloc_path();
7201 key
.objectid
= btrfs_ino(inode
);
7203 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7204 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7205 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7209 btrfs_free_path(path
);
7212 leaf
= path
->nodes
[0];
7213 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7214 struct btrfs_file_extent_item
);
7215 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7216 btrfs_set_file_extent_type(leaf
, ei
,
7217 BTRFS_FILE_EXTENT_INLINE
);
7218 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7219 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7220 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7221 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7223 ptr
= btrfs_file_extent_inline_start(ei
);
7224 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7225 btrfs_mark_buffer_dirty(leaf
);
7226 btrfs_free_path(path
);
7228 inode
->i_op
= &btrfs_symlink_inode_operations
;
7229 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7230 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7231 inode_set_bytes(inode
, name_len
);
7232 btrfs_i_size_write(inode
, name_len
- 1);
7233 err
= btrfs_update_inode(trans
, root
, inode
);
7238 nr
= trans
->blocks_used
;
7239 btrfs_end_transaction_throttle(trans
, root
);
7241 inode_dec_link_count(inode
);
7244 btrfs_btree_balance_dirty(root
, nr
);
7248 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7249 u64 start
, u64 num_bytes
, u64 min_size
,
7250 loff_t actual_len
, u64
*alloc_hint
,
7251 struct btrfs_trans_handle
*trans
)
7253 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7254 struct btrfs_key ins
;
7255 u64 cur_offset
= start
;
7258 bool own_trans
= true;
7262 while (num_bytes
> 0) {
7264 trans
= btrfs_start_transaction(root
, 3);
7265 if (IS_ERR(trans
)) {
7266 ret
= PTR_ERR(trans
);
7271 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7272 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7275 btrfs_end_transaction(trans
, root
);
7279 ret
= insert_reserved_file_extent(trans
, inode
,
7280 cur_offset
, ins
.objectid
,
7281 ins
.offset
, ins
.offset
,
7282 ins
.offset
, 0, 0, 0,
7283 BTRFS_FILE_EXTENT_PREALLOC
);
7285 btrfs_drop_extent_cache(inode
, cur_offset
,
7286 cur_offset
+ ins
.offset
-1, 0);
7288 num_bytes
-= ins
.offset
;
7289 cur_offset
+= ins
.offset
;
7290 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7292 inode
->i_ctime
= CURRENT_TIME
;
7293 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7294 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7295 (actual_len
> inode
->i_size
) &&
7296 (cur_offset
> inode
->i_size
)) {
7297 if (cur_offset
> actual_len
)
7298 i_size
= actual_len
;
7300 i_size
= cur_offset
;
7301 i_size_write(inode
, i_size
);
7302 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7305 ret
= btrfs_update_inode(trans
, root
, inode
);
7309 btrfs_end_transaction(trans
, root
);
7314 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7315 u64 start
, u64 num_bytes
, u64 min_size
,
7316 loff_t actual_len
, u64
*alloc_hint
)
7318 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7319 min_size
, actual_len
, alloc_hint
,
7323 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7324 struct btrfs_trans_handle
*trans
, int mode
,
7325 u64 start
, u64 num_bytes
, u64 min_size
,
7326 loff_t actual_len
, u64
*alloc_hint
)
7328 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7329 min_size
, actual_len
, alloc_hint
, trans
);
7332 static int btrfs_set_page_dirty(struct page
*page
)
7334 return __set_page_dirty_nobuffers(page
);
7337 static int btrfs_permission(struct inode
*inode
, int mask
)
7339 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7340 umode_t mode
= inode
->i_mode
;
7342 if (mask
& MAY_WRITE
&&
7343 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
7344 if (btrfs_root_readonly(root
))
7346 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
7349 return generic_permission(inode
, mask
);
7352 static const struct inode_operations btrfs_dir_inode_operations
= {
7353 .getattr
= btrfs_getattr
,
7354 .lookup
= btrfs_lookup
,
7355 .create
= btrfs_create
,
7356 .unlink
= btrfs_unlink
,
7358 .mkdir
= btrfs_mkdir
,
7359 .rmdir
= btrfs_rmdir
,
7360 .rename
= btrfs_rename
,
7361 .symlink
= btrfs_symlink
,
7362 .setattr
= btrfs_setattr
,
7363 .mknod
= btrfs_mknod
,
7364 .setxattr
= btrfs_setxattr
,
7365 .getxattr
= btrfs_getxattr
,
7366 .listxattr
= btrfs_listxattr
,
7367 .removexattr
= btrfs_removexattr
,
7368 .permission
= btrfs_permission
,
7369 .get_acl
= btrfs_get_acl
,
7371 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7372 .lookup
= btrfs_lookup
,
7373 .permission
= btrfs_permission
,
7374 .get_acl
= btrfs_get_acl
,
7377 static const struct file_operations btrfs_dir_file_operations
= {
7378 .llseek
= generic_file_llseek
,
7379 .read
= generic_read_dir
,
7380 .readdir
= btrfs_real_readdir
,
7381 .unlocked_ioctl
= btrfs_ioctl
,
7382 #ifdef CONFIG_COMPAT
7383 .compat_ioctl
= btrfs_ioctl
,
7385 .release
= btrfs_release_file
,
7386 .fsync
= btrfs_sync_file
,
7389 static struct extent_io_ops btrfs_extent_io_ops
= {
7390 .fill_delalloc
= run_delalloc_range
,
7391 .submit_bio_hook
= btrfs_submit_bio_hook
,
7392 .merge_bio_hook
= btrfs_merge_bio_hook
,
7393 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7394 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7395 .writepage_start_hook
= btrfs_writepage_start_hook
,
7396 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
7397 .set_bit_hook
= btrfs_set_bit_hook
,
7398 .clear_bit_hook
= btrfs_clear_bit_hook
,
7399 .merge_extent_hook
= btrfs_merge_extent_hook
,
7400 .split_extent_hook
= btrfs_split_extent_hook
,
7404 * btrfs doesn't support the bmap operation because swapfiles
7405 * use bmap to make a mapping of extents in the file. They assume
7406 * these extents won't change over the life of the file and they
7407 * use the bmap result to do IO directly to the drive.
7409 * the btrfs bmap call would return logical addresses that aren't
7410 * suitable for IO and they also will change frequently as COW
7411 * operations happen. So, swapfile + btrfs == corruption.
7413 * For now we're avoiding this by dropping bmap.
7415 static const struct address_space_operations btrfs_aops
= {
7416 .readpage
= btrfs_readpage
,
7417 .writepage
= btrfs_writepage
,
7418 .writepages
= btrfs_writepages
,
7419 .readpages
= btrfs_readpages
,
7420 .direct_IO
= btrfs_direct_IO
,
7421 .invalidatepage
= btrfs_invalidatepage
,
7422 .releasepage
= btrfs_releasepage
,
7423 .set_page_dirty
= btrfs_set_page_dirty
,
7424 .error_remove_page
= generic_error_remove_page
,
7427 static const struct address_space_operations btrfs_symlink_aops
= {
7428 .readpage
= btrfs_readpage
,
7429 .writepage
= btrfs_writepage
,
7430 .invalidatepage
= btrfs_invalidatepage
,
7431 .releasepage
= btrfs_releasepage
,
7434 static const struct inode_operations btrfs_file_inode_operations
= {
7435 .getattr
= btrfs_getattr
,
7436 .setattr
= btrfs_setattr
,
7437 .setxattr
= btrfs_setxattr
,
7438 .getxattr
= btrfs_getxattr
,
7439 .listxattr
= btrfs_listxattr
,
7440 .removexattr
= btrfs_removexattr
,
7441 .permission
= btrfs_permission
,
7442 .fiemap
= btrfs_fiemap
,
7443 .get_acl
= btrfs_get_acl
,
7445 static const struct inode_operations btrfs_special_inode_operations
= {
7446 .getattr
= btrfs_getattr
,
7447 .setattr
= btrfs_setattr
,
7448 .permission
= btrfs_permission
,
7449 .setxattr
= btrfs_setxattr
,
7450 .getxattr
= btrfs_getxattr
,
7451 .listxattr
= btrfs_listxattr
,
7452 .removexattr
= btrfs_removexattr
,
7453 .get_acl
= btrfs_get_acl
,
7455 static const struct inode_operations btrfs_symlink_inode_operations
= {
7456 .readlink
= generic_readlink
,
7457 .follow_link
= page_follow_link_light
,
7458 .put_link
= page_put_link
,
7459 .getattr
= btrfs_getattr
,
7460 .permission
= btrfs_permission
,
7461 .setxattr
= btrfs_setxattr
,
7462 .getxattr
= btrfs_getxattr
,
7463 .listxattr
= btrfs_listxattr
,
7464 .removexattr
= btrfs_removexattr
,
7465 .get_acl
= btrfs_get_acl
,
7468 const struct dentry_operations btrfs_dentry_operations
= {
7469 .d_delete
= btrfs_dentry_delete
,
7470 .d_release
= btrfs_dentry_release
,