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
);
1797 btrfs_end_transaction_nolock(trans
, root
);
1799 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
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
);
2168 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2170 /* grab metadata reservation from transaction handle */
2172 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2176 /* insert an orphan item to track this unlinked/truncated file */
2178 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
2182 /* insert an orphan item to track subvolume contains orphan files */
2184 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2185 root
->root_key
.objectid
);
2192 * We have done the truncate/delete so we can go ahead and remove the orphan
2193 * item for this particular inode.
2195 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2197 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2198 int delete_item
= 0;
2199 int release_rsv
= 0;
2202 spin_lock(&root
->orphan_lock
);
2203 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2204 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2208 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2209 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2212 spin_unlock(&root
->orphan_lock
);
2214 if (trans
&& delete_item
) {
2215 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
2220 btrfs_orphan_release_metadata(inode
);
2226 * this cleans up any orphans that may be left on the list from the last use
2229 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2231 struct btrfs_path
*path
;
2232 struct extent_buffer
*leaf
;
2233 struct btrfs_key key
, found_key
;
2234 struct btrfs_trans_handle
*trans
;
2235 struct inode
*inode
;
2236 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2238 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2241 path
= btrfs_alloc_path();
2248 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2249 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2250 key
.offset
= (u64
)-1;
2253 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2258 * if ret == 0 means we found what we were searching for, which
2259 * is weird, but possible, so only screw with path if we didn't
2260 * find the key and see if we have stuff that matches
2264 if (path
->slots
[0] == 0)
2269 /* pull out the item */
2270 leaf
= path
->nodes
[0];
2271 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2273 /* make sure the item matches what we want */
2274 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2276 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2279 /* release the path since we're done with it */
2280 btrfs_release_path(path
);
2283 * this is where we are basically btrfs_lookup, without the
2284 * crossing root thing. we store the inode number in the
2285 * offset of the orphan item.
2287 found_key
.objectid
= found_key
.offset
;
2288 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2289 found_key
.offset
= 0;
2290 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2291 if (IS_ERR(inode
)) {
2292 ret
= PTR_ERR(inode
);
2297 * add this inode to the orphan list so btrfs_orphan_del does
2298 * the proper thing when we hit it
2300 spin_lock(&root
->orphan_lock
);
2301 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2302 spin_unlock(&root
->orphan_lock
);
2305 * if this is a bad inode, means we actually succeeded in
2306 * removing the inode, but not the orphan record, which means
2307 * we need to manually delete the orphan since iput will just
2308 * do a destroy_inode
2310 if (is_bad_inode(inode
)) {
2311 trans
= btrfs_start_transaction(root
, 0);
2312 if (IS_ERR(trans
)) {
2313 ret
= PTR_ERR(trans
);
2316 btrfs_orphan_del(trans
, inode
);
2317 btrfs_end_transaction(trans
, root
);
2322 /* if we have links, this was a truncate, lets do that */
2323 if (inode
->i_nlink
) {
2324 if (!S_ISREG(inode
->i_mode
)) {
2330 ret
= btrfs_truncate(inode
);
2335 /* this will do delete_inode and everything for us */
2340 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2342 if (root
->orphan_block_rsv
)
2343 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2346 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2347 trans
= btrfs_join_transaction(root
);
2349 btrfs_end_transaction(trans
, root
);
2353 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2355 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2359 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2360 btrfs_free_path(path
);
2365 * very simple check to peek ahead in the leaf looking for xattrs. If we
2366 * don't find any xattrs, we know there can't be any acls.
2368 * slot is the slot the inode is in, objectid is the objectid of the inode
2370 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2371 int slot
, u64 objectid
)
2373 u32 nritems
= btrfs_header_nritems(leaf
);
2374 struct btrfs_key found_key
;
2378 while (slot
< nritems
) {
2379 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2381 /* we found a different objectid, there must not be acls */
2382 if (found_key
.objectid
!= objectid
)
2385 /* we found an xattr, assume we've got an acl */
2386 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2390 * we found a key greater than an xattr key, there can't
2391 * be any acls later on
2393 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2400 * it goes inode, inode backrefs, xattrs, extents,
2401 * so if there are a ton of hard links to an inode there can
2402 * be a lot of backrefs. Don't waste time searching too hard,
2403 * this is just an optimization
2408 /* we hit the end of the leaf before we found an xattr or
2409 * something larger than an xattr. We have to assume the inode
2416 * read an inode from the btree into the in-memory inode
2418 static void btrfs_read_locked_inode(struct inode
*inode
)
2420 struct btrfs_path
*path
;
2421 struct extent_buffer
*leaf
;
2422 struct btrfs_inode_item
*inode_item
;
2423 struct btrfs_timespec
*tspec
;
2424 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2425 struct btrfs_key location
;
2429 bool filled
= false;
2431 ret
= btrfs_fill_inode(inode
, &rdev
);
2435 path
= btrfs_alloc_path();
2439 path
->leave_spinning
= 1;
2440 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2442 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2446 leaf
= path
->nodes
[0];
2451 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2452 struct btrfs_inode_item
);
2453 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2454 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2455 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2456 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2457 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2459 tspec
= btrfs_inode_atime(inode_item
);
2460 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2461 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2463 tspec
= btrfs_inode_mtime(inode_item
);
2464 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2465 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2467 tspec
= btrfs_inode_ctime(inode_item
);
2468 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2469 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2471 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2472 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2473 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2474 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2476 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2478 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2479 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2482 * try to precache a NULL acl entry for files that don't have
2483 * any xattrs or acls
2485 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
2488 cache_no_acl(inode
);
2490 btrfs_free_path(path
);
2492 switch (inode
->i_mode
& S_IFMT
) {
2494 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2495 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2496 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2497 inode
->i_fop
= &btrfs_file_operations
;
2498 inode
->i_op
= &btrfs_file_inode_operations
;
2501 inode
->i_fop
= &btrfs_dir_file_operations
;
2502 if (root
== root
->fs_info
->tree_root
)
2503 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2505 inode
->i_op
= &btrfs_dir_inode_operations
;
2508 inode
->i_op
= &btrfs_symlink_inode_operations
;
2509 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2510 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2513 inode
->i_op
= &btrfs_special_inode_operations
;
2514 init_special_inode(inode
, inode
->i_mode
, rdev
);
2518 btrfs_update_iflags(inode
);
2522 btrfs_free_path(path
);
2523 make_bad_inode(inode
);
2527 * given a leaf and an inode, copy the inode fields into the leaf
2529 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2530 struct extent_buffer
*leaf
,
2531 struct btrfs_inode_item
*item
,
2532 struct inode
*inode
)
2534 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2535 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2536 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2537 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2538 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2540 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2541 inode
->i_atime
.tv_sec
);
2542 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2543 inode
->i_atime
.tv_nsec
);
2545 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2546 inode
->i_mtime
.tv_sec
);
2547 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2548 inode
->i_mtime
.tv_nsec
);
2550 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2551 inode
->i_ctime
.tv_sec
);
2552 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2553 inode
->i_ctime
.tv_nsec
);
2555 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2556 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2557 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2558 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2559 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2560 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2561 btrfs_set_inode_block_group(leaf
, item
, 0);
2565 * copy everything in the in-memory inode into the btree.
2567 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2568 struct btrfs_root
*root
, struct inode
*inode
)
2570 struct btrfs_inode_item
*inode_item
;
2571 struct btrfs_path
*path
;
2572 struct extent_buffer
*leaf
;
2576 * If the inode is a free space inode, we can deadlock during commit
2577 * if we put it into the delayed code.
2579 * The data relocation inode should also be directly updated
2582 if (!btrfs_is_free_space_inode(root
, inode
)
2583 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
2584 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
2586 btrfs_set_inode_last_trans(trans
, inode
);
2590 path
= btrfs_alloc_path();
2594 path
->leave_spinning
= 1;
2595 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
2603 btrfs_unlock_up_safe(path
, 1);
2604 leaf
= path
->nodes
[0];
2605 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2606 struct btrfs_inode_item
);
2608 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2609 btrfs_mark_buffer_dirty(leaf
);
2610 btrfs_set_inode_last_trans(trans
, inode
);
2613 btrfs_free_path(path
);
2618 * unlink helper that gets used here in inode.c and in the tree logging
2619 * recovery code. It remove a link in a directory with a given name, and
2620 * also drops the back refs in the inode to the directory
2622 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2623 struct btrfs_root
*root
,
2624 struct inode
*dir
, struct inode
*inode
,
2625 const char *name
, int name_len
)
2627 struct btrfs_path
*path
;
2629 struct extent_buffer
*leaf
;
2630 struct btrfs_dir_item
*di
;
2631 struct btrfs_key key
;
2633 u64 ino
= btrfs_ino(inode
);
2634 u64 dir_ino
= btrfs_ino(dir
);
2636 path
= btrfs_alloc_path();
2642 path
->leave_spinning
= 1;
2643 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2644 name
, name_len
, -1);
2653 leaf
= path
->nodes
[0];
2654 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2655 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2658 btrfs_release_path(path
);
2660 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
2663 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2664 "inode %llu parent %llu\n", name_len
, name
,
2665 (unsigned long long)ino
, (unsigned long long)dir_ino
);
2669 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2673 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2675 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2677 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2682 btrfs_free_path(path
);
2686 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2687 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2688 btrfs_update_inode(trans
, root
, dir
);
2693 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2694 struct btrfs_root
*root
,
2695 struct inode
*dir
, struct inode
*inode
,
2696 const char *name
, int name_len
)
2699 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2701 btrfs_drop_nlink(inode
);
2702 ret
= btrfs_update_inode(trans
, root
, inode
);
2708 /* helper to check if there is any shared block in the path */
2709 static int check_path_shared(struct btrfs_root
*root
,
2710 struct btrfs_path
*path
)
2712 struct extent_buffer
*eb
;
2716 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2719 if (!path
->nodes
[level
])
2721 eb
= path
->nodes
[level
];
2722 if (!btrfs_block_can_be_shared(root
, eb
))
2724 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2733 * helper to start transaction for unlink and rmdir.
2735 * unlink and rmdir are special in btrfs, they do not always free space.
2736 * so in enospc case, we should make sure they will free space before
2737 * allowing them to use the global metadata reservation.
2739 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2740 struct dentry
*dentry
)
2742 struct btrfs_trans_handle
*trans
;
2743 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2744 struct btrfs_path
*path
;
2745 struct btrfs_inode_ref
*ref
;
2746 struct btrfs_dir_item
*di
;
2747 struct inode
*inode
= dentry
->d_inode
;
2752 u64 ino
= btrfs_ino(inode
);
2753 u64 dir_ino
= btrfs_ino(dir
);
2755 trans
= btrfs_start_transaction(root
, 10);
2756 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2759 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2760 return ERR_PTR(-ENOSPC
);
2762 /* check if there is someone else holds reference */
2763 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2764 return ERR_PTR(-ENOSPC
);
2766 if (atomic_read(&inode
->i_count
) > 2)
2767 return ERR_PTR(-ENOSPC
);
2769 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2770 return ERR_PTR(-ENOSPC
);
2772 path
= btrfs_alloc_path();
2774 root
->fs_info
->enospc_unlink
= 0;
2775 return ERR_PTR(-ENOMEM
);
2778 trans
= btrfs_start_transaction(root
, 0);
2779 if (IS_ERR(trans
)) {
2780 btrfs_free_path(path
);
2781 root
->fs_info
->enospc_unlink
= 0;
2785 path
->skip_locking
= 1;
2786 path
->search_commit_root
= 1;
2788 ret
= btrfs_lookup_inode(trans
, root
, path
,
2789 &BTRFS_I(dir
)->location
, 0);
2795 if (check_path_shared(root
, path
))
2800 btrfs_release_path(path
);
2802 ret
= btrfs_lookup_inode(trans
, root
, path
,
2803 &BTRFS_I(inode
)->location
, 0);
2809 if (check_path_shared(root
, path
))
2814 btrfs_release_path(path
);
2816 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2817 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2824 if (check_path_shared(root
, path
))
2826 btrfs_release_path(path
);
2834 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2835 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2841 if (check_path_shared(root
, path
))
2847 btrfs_release_path(path
);
2849 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2850 dentry
->d_name
.name
, dentry
->d_name
.len
,
2857 if (check_path_shared(root
, path
))
2859 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2860 btrfs_release_path(path
);
2863 * This is a commit root search, if we can lookup inode item and other
2864 * relative items in the commit root, it means the transaction of
2865 * dir/file creation has been committed, and the dir index item that we
2866 * delay to insert has also been inserted into the commit root. So
2867 * we needn't worry about the delayed insertion of the dir index item
2870 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
2871 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2876 BUG_ON(ret
== -ENOENT
);
2877 if (check_path_shared(root
, path
))
2882 btrfs_free_path(path
);
2884 btrfs_end_transaction(trans
, root
);
2885 root
->fs_info
->enospc_unlink
= 0;
2886 return ERR_PTR(err
);
2889 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2893 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2894 struct btrfs_root
*root
)
2896 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2897 BUG_ON(!root
->fs_info
->enospc_unlink
);
2898 root
->fs_info
->enospc_unlink
= 0;
2900 btrfs_end_transaction_throttle(trans
, root
);
2903 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2905 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2906 struct btrfs_trans_handle
*trans
;
2907 struct inode
*inode
= dentry
->d_inode
;
2909 unsigned long nr
= 0;
2911 trans
= __unlink_start_trans(dir
, dentry
);
2913 return PTR_ERR(trans
);
2915 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2917 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2918 dentry
->d_name
.name
, dentry
->d_name
.len
);
2922 if (inode
->i_nlink
== 0) {
2923 ret
= btrfs_orphan_add(trans
, inode
);
2929 nr
= trans
->blocks_used
;
2930 __unlink_end_trans(trans
, root
);
2931 btrfs_btree_balance_dirty(root
, nr
);
2935 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2936 struct btrfs_root
*root
,
2937 struct inode
*dir
, u64 objectid
,
2938 const char *name
, int name_len
)
2940 struct btrfs_path
*path
;
2941 struct extent_buffer
*leaf
;
2942 struct btrfs_dir_item
*di
;
2943 struct btrfs_key key
;
2946 u64 dir_ino
= btrfs_ino(dir
);
2948 path
= btrfs_alloc_path();
2952 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2953 name
, name_len
, -1);
2954 BUG_ON(IS_ERR_OR_NULL(di
));
2956 leaf
= path
->nodes
[0];
2957 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2958 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
2959 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2961 btrfs_release_path(path
);
2963 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
2964 objectid
, root
->root_key
.objectid
,
2965 dir_ino
, &index
, name
, name_len
);
2967 BUG_ON(ret
!= -ENOENT
);
2968 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
2970 BUG_ON(IS_ERR_OR_NULL(di
));
2972 leaf
= path
->nodes
[0];
2973 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2974 btrfs_release_path(path
);
2977 btrfs_release_path(path
);
2979 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2982 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2983 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2984 ret
= btrfs_update_inode(trans
, root
, dir
);
2987 btrfs_free_path(path
);
2991 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2993 struct inode
*inode
= dentry
->d_inode
;
2995 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2996 struct btrfs_trans_handle
*trans
;
2997 unsigned long nr
= 0;
2999 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3000 btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3003 trans
= __unlink_start_trans(dir
, dentry
);
3005 return PTR_ERR(trans
);
3007 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3008 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3009 BTRFS_I(inode
)->location
.objectid
,
3010 dentry
->d_name
.name
,
3011 dentry
->d_name
.len
);
3015 err
= btrfs_orphan_add(trans
, inode
);
3019 /* now the directory is empty */
3020 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3021 dentry
->d_name
.name
, dentry
->d_name
.len
);
3023 btrfs_i_size_write(inode
, 0);
3025 nr
= trans
->blocks_used
;
3026 __unlink_end_trans(trans
, root
);
3027 btrfs_btree_balance_dirty(root
, nr
);
3033 * this can truncate away extent items, csum items and directory items.
3034 * It starts at a high offset and removes keys until it can't find
3035 * any higher than new_size
3037 * csum items that cross the new i_size are truncated to the new size
3040 * min_type is the minimum key type to truncate down to. If set to 0, this
3041 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3043 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3044 struct btrfs_root
*root
,
3045 struct inode
*inode
,
3046 u64 new_size
, u32 min_type
)
3048 struct btrfs_path
*path
;
3049 struct extent_buffer
*leaf
;
3050 struct btrfs_file_extent_item
*fi
;
3051 struct btrfs_key key
;
3052 struct btrfs_key found_key
;
3053 u64 extent_start
= 0;
3054 u64 extent_num_bytes
= 0;
3055 u64 extent_offset
= 0;
3057 u64 mask
= root
->sectorsize
- 1;
3058 u32 found_type
= (u8
)-1;
3061 int pending_del_nr
= 0;
3062 int pending_del_slot
= 0;
3063 int extent_type
= -1;
3067 u64 ino
= btrfs_ino(inode
);
3069 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3071 path
= btrfs_alloc_path();
3076 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3077 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3080 * This function is also used to drop the items in the log tree before
3081 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3082 * it is used to drop the loged items. So we shouldn't kill the delayed
3085 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3086 btrfs_kill_delayed_inode_items(inode
);
3089 key
.offset
= (u64
)-1;
3093 path
->leave_spinning
= 1;
3094 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3101 /* there are no items in the tree for us to truncate, we're
3104 if (path
->slots
[0] == 0)
3111 leaf
= path
->nodes
[0];
3112 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3113 found_type
= btrfs_key_type(&found_key
);
3116 if (found_key
.objectid
!= ino
)
3119 if (found_type
< min_type
)
3122 item_end
= found_key
.offset
;
3123 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3124 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3125 struct btrfs_file_extent_item
);
3126 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3127 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3128 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3129 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3131 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3133 btrfs_file_extent_num_bytes(leaf
, fi
);
3134 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3135 item_end
+= btrfs_file_extent_inline_len(leaf
,
3140 if (found_type
> min_type
) {
3143 if (item_end
< new_size
)
3145 if (found_key
.offset
>= new_size
)
3151 /* FIXME, shrink the extent if the ref count is only 1 */
3152 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3155 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3157 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3158 if (!del_item
&& !encoding
) {
3159 u64 orig_num_bytes
=
3160 btrfs_file_extent_num_bytes(leaf
, fi
);
3161 extent_num_bytes
= new_size
-
3162 found_key
.offset
+ root
->sectorsize
- 1;
3163 extent_num_bytes
= extent_num_bytes
&
3164 ~((u64
)root
->sectorsize
- 1);
3165 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3167 num_dec
= (orig_num_bytes
-
3169 if (root
->ref_cows
&& extent_start
!= 0)
3170 inode_sub_bytes(inode
, num_dec
);
3171 btrfs_mark_buffer_dirty(leaf
);
3174 btrfs_file_extent_disk_num_bytes(leaf
,
3176 extent_offset
= found_key
.offset
-
3177 btrfs_file_extent_offset(leaf
, fi
);
3179 /* FIXME blocksize != 4096 */
3180 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3181 if (extent_start
!= 0) {
3184 inode_sub_bytes(inode
, num_dec
);
3187 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3189 * we can't truncate inline items that have had
3193 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3194 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3195 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3196 u32 size
= new_size
- found_key
.offset
;
3198 if (root
->ref_cows
) {
3199 inode_sub_bytes(inode
, item_end
+ 1 -
3203 btrfs_file_extent_calc_inline_size(size
);
3204 ret
= btrfs_truncate_item(trans
, root
, path
,
3206 } else if (root
->ref_cows
) {
3207 inode_sub_bytes(inode
, item_end
+ 1 -
3213 if (!pending_del_nr
) {
3214 /* no pending yet, add ourselves */
3215 pending_del_slot
= path
->slots
[0];
3217 } else if (pending_del_nr
&&
3218 path
->slots
[0] + 1 == pending_del_slot
) {
3219 /* hop on the pending chunk */
3221 pending_del_slot
= path
->slots
[0];
3228 if (found_extent
&& (root
->ref_cows
||
3229 root
== root
->fs_info
->tree_root
)) {
3230 btrfs_set_path_blocking(path
);
3231 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3232 extent_num_bytes
, 0,
3233 btrfs_header_owner(leaf
),
3234 ino
, extent_offset
);
3238 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3241 if (path
->slots
[0] == 0 ||
3242 path
->slots
[0] != pending_del_slot
) {
3243 if (root
->ref_cows
&&
3244 BTRFS_I(inode
)->location
.objectid
!=
3245 BTRFS_FREE_INO_OBJECTID
) {
3249 if (pending_del_nr
) {
3250 ret
= btrfs_del_items(trans
, root
, path
,
3256 btrfs_release_path(path
);
3263 if (pending_del_nr
) {
3264 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3268 btrfs_free_path(path
);
3273 * taken from block_truncate_page, but does cow as it zeros out
3274 * any bytes left in the last page in the file.
3276 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3278 struct inode
*inode
= mapping
->host
;
3279 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3280 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3281 struct btrfs_ordered_extent
*ordered
;
3282 struct extent_state
*cached_state
= NULL
;
3284 u32 blocksize
= root
->sectorsize
;
3285 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3286 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3292 if ((offset
& (blocksize
- 1)) == 0)
3294 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3300 page
= find_or_create_page(mapping
, index
, GFP_NOFS
);
3302 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3306 page_start
= page_offset(page
);
3307 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3309 if (!PageUptodate(page
)) {
3310 ret
= btrfs_readpage(NULL
, page
);
3312 if (page
->mapping
!= mapping
) {
3314 page_cache_release(page
);
3317 if (!PageUptodate(page
)) {
3322 wait_on_page_writeback(page
);
3324 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3326 set_page_extent_mapped(page
);
3328 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3330 unlock_extent_cached(io_tree
, page_start
, page_end
,
3331 &cached_state
, GFP_NOFS
);
3333 page_cache_release(page
);
3334 btrfs_start_ordered_extent(inode
, ordered
, 1);
3335 btrfs_put_ordered_extent(ordered
);
3339 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3340 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3341 0, 0, &cached_state
, GFP_NOFS
);
3343 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3346 unlock_extent_cached(io_tree
, page_start
, page_end
,
3347 &cached_state
, GFP_NOFS
);
3352 if (offset
!= PAGE_CACHE_SIZE
) {
3354 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3355 flush_dcache_page(page
);
3358 ClearPageChecked(page
);
3359 set_page_dirty(page
);
3360 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3365 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3367 page_cache_release(page
);
3373 * This function puts in dummy file extents for the area we're creating a hole
3374 * for. So if we are truncating this file to a larger size we need to insert
3375 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3376 * the range between oldsize and size
3378 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3380 struct btrfs_trans_handle
*trans
;
3381 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3382 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3383 struct extent_map
*em
= NULL
;
3384 struct extent_state
*cached_state
= NULL
;
3385 u64 mask
= root
->sectorsize
- 1;
3386 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3387 u64 block_end
= (size
+ mask
) & ~mask
;
3393 if (size
<= hole_start
)
3397 struct btrfs_ordered_extent
*ordered
;
3398 btrfs_wait_ordered_range(inode
, hole_start
,
3399 block_end
- hole_start
);
3400 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3401 &cached_state
, GFP_NOFS
);
3402 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3405 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3406 &cached_state
, GFP_NOFS
);
3407 btrfs_put_ordered_extent(ordered
);
3410 cur_offset
= hole_start
;
3412 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3413 block_end
- cur_offset
, 0);
3414 BUG_ON(IS_ERR_OR_NULL(em
));
3415 last_byte
= min(extent_map_end(em
), block_end
);
3416 last_byte
= (last_byte
+ mask
) & ~mask
;
3417 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3419 hole_size
= last_byte
- cur_offset
;
3421 trans
= btrfs_start_transaction(root
, 2);
3422 if (IS_ERR(trans
)) {
3423 err
= PTR_ERR(trans
);
3427 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3428 cur_offset
+ hole_size
,
3431 btrfs_end_transaction(trans
, root
);
3435 err
= btrfs_insert_file_extent(trans
, root
,
3436 btrfs_ino(inode
), cur_offset
, 0,
3437 0, hole_size
, 0, hole_size
,
3440 btrfs_end_transaction(trans
, root
);
3444 btrfs_drop_extent_cache(inode
, hole_start
,
3447 btrfs_end_transaction(trans
, root
);
3449 free_extent_map(em
);
3451 cur_offset
= last_byte
;
3452 if (cur_offset
>= block_end
)
3456 free_extent_map(em
);
3457 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3462 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3464 loff_t oldsize
= i_size_read(inode
);
3467 if (newsize
== oldsize
)
3470 if (newsize
> oldsize
) {
3471 i_size_write(inode
, newsize
);
3472 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3473 truncate_pagecache(inode
, oldsize
, newsize
);
3474 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3476 btrfs_setsize(inode
, oldsize
);
3480 mark_inode_dirty(inode
);
3484 * We're truncating a file that used to have good data down to
3485 * zero. Make sure it gets into the ordered flush list so that
3486 * any new writes get down to disk quickly.
3489 BTRFS_I(inode
)->ordered_data_close
= 1;
3491 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3492 truncate_setsize(inode
, newsize
);
3493 ret
= btrfs_truncate(inode
);
3499 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3501 struct inode
*inode
= dentry
->d_inode
;
3502 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3505 if (btrfs_root_readonly(root
))
3508 err
= inode_change_ok(inode
, attr
);
3512 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3513 err
= btrfs_setsize(inode
, attr
->ia_size
);
3518 if (attr
->ia_valid
) {
3519 setattr_copy(inode
, attr
);
3520 mark_inode_dirty(inode
);
3522 if (attr
->ia_valid
& ATTR_MODE
)
3523 err
= btrfs_acl_chmod(inode
);
3529 void btrfs_evict_inode(struct inode
*inode
)
3531 struct btrfs_trans_handle
*trans
;
3532 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3536 trace_btrfs_inode_evict(inode
);
3538 truncate_inode_pages(&inode
->i_data
, 0);
3539 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3540 btrfs_is_free_space_inode(root
, inode
)))
3543 if (is_bad_inode(inode
)) {
3544 btrfs_orphan_del(NULL
, inode
);
3547 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3548 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3550 if (root
->fs_info
->log_root_recovering
) {
3551 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3555 if (inode
->i_nlink
> 0) {
3556 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3560 btrfs_i_size_write(inode
, 0);
3563 trans
= btrfs_join_transaction(root
);
3564 BUG_ON(IS_ERR(trans
));
3565 trans
->block_rsv
= root
->orphan_block_rsv
;
3567 ret
= btrfs_block_rsv_check(trans
, root
,
3568 root
->orphan_block_rsv
, 0, 5);
3570 BUG_ON(ret
!= -EAGAIN
);
3571 ret
= btrfs_commit_transaction(trans
, root
);
3576 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3580 nr
= trans
->blocks_used
;
3581 btrfs_end_transaction(trans
, root
);
3583 btrfs_btree_balance_dirty(root
, nr
);
3588 ret
= btrfs_orphan_del(trans
, inode
);
3592 if (!(root
== root
->fs_info
->tree_root
||
3593 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
3594 btrfs_return_ino(root
, btrfs_ino(inode
));
3596 nr
= trans
->blocks_used
;
3597 btrfs_end_transaction(trans
, root
);
3598 btrfs_btree_balance_dirty(root
, nr
);
3600 end_writeback(inode
);
3605 * this returns the key found in the dir entry in the location pointer.
3606 * If no dir entries were found, location->objectid is 0.
3608 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3609 struct btrfs_key
*location
)
3611 const char *name
= dentry
->d_name
.name
;
3612 int namelen
= dentry
->d_name
.len
;
3613 struct btrfs_dir_item
*di
;
3614 struct btrfs_path
*path
;
3615 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3618 path
= btrfs_alloc_path();
3622 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
3627 if (IS_ERR_OR_NULL(di
))
3630 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3632 btrfs_free_path(path
);
3635 location
->objectid
= 0;
3640 * when we hit a tree root in a directory, the btrfs part of the inode
3641 * needs to be changed to reflect the root directory of the tree root. This
3642 * is kind of like crossing a mount point.
3644 static int fixup_tree_root_location(struct btrfs_root
*root
,
3646 struct dentry
*dentry
,
3647 struct btrfs_key
*location
,
3648 struct btrfs_root
**sub_root
)
3650 struct btrfs_path
*path
;
3651 struct btrfs_root
*new_root
;
3652 struct btrfs_root_ref
*ref
;
3653 struct extent_buffer
*leaf
;
3657 path
= btrfs_alloc_path();
3664 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3665 BTRFS_I(dir
)->root
->root_key
.objectid
,
3666 location
->objectid
);
3673 leaf
= path
->nodes
[0];
3674 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3675 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
3676 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3679 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3680 (unsigned long)(ref
+ 1),
3681 dentry
->d_name
.len
);
3685 btrfs_release_path(path
);
3687 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3688 if (IS_ERR(new_root
)) {
3689 err
= PTR_ERR(new_root
);
3693 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3698 *sub_root
= new_root
;
3699 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3700 location
->type
= BTRFS_INODE_ITEM_KEY
;
3701 location
->offset
= 0;
3704 btrfs_free_path(path
);
3708 static void inode_tree_add(struct inode
*inode
)
3710 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3711 struct btrfs_inode
*entry
;
3713 struct rb_node
*parent
;
3714 u64 ino
= btrfs_ino(inode
);
3716 p
= &root
->inode_tree
.rb_node
;
3719 if (inode_unhashed(inode
))
3722 spin_lock(&root
->inode_lock
);
3725 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3727 if (ino
< btrfs_ino(&entry
->vfs_inode
))
3728 p
= &parent
->rb_left
;
3729 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
3730 p
= &parent
->rb_right
;
3732 WARN_ON(!(entry
->vfs_inode
.i_state
&
3733 (I_WILL_FREE
| I_FREEING
)));
3734 rb_erase(parent
, &root
->inode_tree
);
3735 RB_CLEAR_NODE(parent
);
3736 spin_unlock(&root
->inode_lock
);
3740 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3741 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3742 spin_unlock(&root
->inode_lock
);
3745 static void inode_tree_del(struct inode
*inode
)
3747 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3750 spin_lock(&root
->inode_lock
);
3751 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3752 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3753 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3754 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3756 spin_unlock(&root
->inode_lock
);
3759 * Free space cache has inodes in the tree root, but the tree root has a
3760 * root_refs of 0, so this could end up dropping the tree root as a
3761 * snapshot, so we need the extra !root->fs_info->tree_root check to
3762 * make sure we don't drop it.
3764 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3765 root
!= root
->fs_info
->tree_root
) {
3766 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3767 spin_lock(&root
->inode_lock
);
3768 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3769 spin_unlock(&root
->inode_lock
);
3771 btrfs_add_dead_root(root
);
3775 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3777 struct rb_node
*node
;
3778 struct rb_node
*prev
;
3779 struct btrfs_inode
*entry
;
3780 struct inode
*inode
;
3783 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3785 spin_lock(&root
->inode_lock
);
3787 node
= root
->inode_tree
.rb_node
;
3791 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3793 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
3794 node
= node
->rb_left
;
3795 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
3796 node
= node
->rb_right
;
3802 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
3803 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
3807 prev
= rb_next(prev
);
3811 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3812 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
3813 inode
= igrab(&entry
->vfs_inode
);
3815 spin_unlock(&root
->inode_lock
);
3816 if (atomic_read(&inode
->i_count
) > 1)
3817 d_prune_aliases(inode
);
3819 * btrfs_drop_inode will have it removed from
3820 * the inode cache when its usage count
3825 spin_lock(&root
->inode_lock
);
3829 if (cond_resched_lock(&root
->inode_lock
))
3832 node
= rb_next(node
);
3834 spin_unlock(&root
->inode_lock
);
3838 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3840 struct btrfs_iget_args
*args
= p
;
3841 inode
->i_ino
= args
->ino
;
3842 BTRFS_I(inode
)->root
= args
->root
;
3843 btrfs_set_inode_space_info(args
->root
, inode
);
3847 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3849 struct btrfs_iget_args
*args
= opaque
;
3850 return args
->ino
== btrfs_ino(inode
) &&
3851 args
->root
== BTRFS_I(inode
)->root
;
3854 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
3856 struct btrfs_root
*root
)
3858 struct inode
*inode
;
3859 struct btrfs_iget_args args
;
3860 args
.ino
= objectid
;
3863 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3864 btrfs_init_locked_inode
,
3869 /* Get an inode object given its location and corresponding root.
3870 * Returns in *is_new if the inode was read from disk
3872 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3873 struct btrfs_root
*root
, int *new)
3875 struct inode
*inode
;
3877 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3879 return ERR_PTR(-ENOMEM
);
3881 if (inode
->i_state
& I_NEW
) {
3882 BTRFS_I(inode
)->root
= root
;
3883 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3884 btrfs_read_locked_inode(inode
);
3885 if (!is_bad_inode(inode
)) {
3886 inode_tree_add(inode
);
3887 unlock_new_inode(inode
);
3891 unlock_new_inode(inode
);
3893 inode
= ERR_PTR(-ESTALE
);
3900 static struct inode
*new_simple_dir(struct super_block
*s
,
3901 struct btrfs_key
*key
,
3902 struct btrfs_root
*root
)
3904 struct inode
*inode
= new_inode(s
);
3907 return ERR_PTR(-ENOMEM
);
3909 BTRFS_I(inode
)->root
= root
;
3910 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
3911 BTRFS_I(inode
)->dummy_inode
= 1;
3913 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
3914 inode
->i_op
= &simple_dir_inode_operations
;
3915 inode
->i_fop
= &simple_dir_operations
;
3916 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
3917 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3922 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3924 struct inode
*inode
;
3925 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3926 struct btrfs_root
*sub_root
= root
;
3927 struct btrfs_key location
;
3931 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3932 return ERR_PTR(-ENAMETOOLONG
);
3934 if (unlikely(d_need_lookup(dentry
))) {
3935 memcpy(&location
, dentry
->d_fsdata
, sizeof(struct btrfs_key
));
3936 kfree(dentry
->d_fsdata
);
3937 dentry
->d_fsdata
= NULL
;
3938 /* This thing is hashed, drop it for now */
3941 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3945 return ERR_PTR(ret
);
3947 if (location
.objectid
== 0)
3950 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
3951 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
3955 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
3957 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
3958 ret
= fixup_tree_root_location(root
, dir
, dentry
,
3959 &location
, &sub_root
);
3962 inode
= ERR_PTR(ret
);
3964 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
3966 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
3968 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
3970 if (!IS_ERR(inode
) && root
!= sub_root
) {
3971 down_read(&root
->fs_info
->cleanup_work_sem
);
3972 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
3973 ret
= btrfs_orphan_cleanup(sub_root
);
3974 up_read(&root
->fs_info
->cleanup_work_sem
);
3976 inode
= ERR_PTR(ret
);
3982 static int btrfs_dentry_delete(const struct dentry
*dentry
)
3984 struct btrfs_root
*root
;
3986 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
3987 dentry
= dentry
->d_parent
;
3989 if (dentry
->d_inode
) {
3990 root
= BTRFS_I(dentry
->d_inode
)->root
;
3991 if (btrfs_root_refs(&root
->root_item
) == 0)
3997 static void btrfs_dentry_release(struct dentry
*dentry
)
3999 if (dentry
->d_fsdata
)
4000 kfree(dentry
->d_fsdata
);
4003 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4004 struct nameidata
*nd
)
4008 ret
= d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
4009 if (unlikely(d_need_lookup(dentry
))) {
4010 spin_lock(&dentry
->d_lock
);
4011 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
4012 spin_unlock(&dentry
->d_lock
);
4017 unsigned char btrfs_filetype_table
[] = {
4018 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4021 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4024 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4025 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4026 struct btrfs_item
*item
;
4027 struct btrfs_dir_item
*di
;
4028 struct btrfs_key key
;
4029 struct btrfs_key found_key
;
4030 struct btrfs_path
*path
;
4031 struct list_head ins_list
;
4032 struct list_head del_list
;
4035 struct extent_buffer
*leaf
;
4037 unsigned char d_type
;
4042 int key_type
= BTRFS_DIR_INDEX_KEY
;
4046 int is_curr
= 0; /* filp->f_pos points to the current index? */
4048 /* FIXME, use a real flag for deciding about the key type */
4049 if (root
->fs_info
->tree_root
== root
)
4050 key_type
= BTRFS_DIR_ITEM_KEY
;
4052 /* special case for "." */
4053 if (filp
->f_pos
== 0) {
4054 over
= filldir(dirent
, ".", 1,
4055 filp
->f_pos
, btrfs_ino(inode
), DT_DIR
);
4060 /* special case for .., just use the back ref */
4061 if (filp
->f_pos
== 1) {
4062 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4063 over
= filldir(dirent
, "..", 2,
4064 filp
->f_pos
, pino
, DT_DIR
);
4069 path
= btrfs_alloc_path();
4075 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4076 INIT_LIST_HEAD(&ins_list
);
4077 INIT_LIST_HEAD(&del_list
);
4078 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
4081 btrfs_set_key_type(&key
, key_type
);
4082 key
.offset
= filp
->f_pos
;
4083 key
.objectid
= btrfs_ino(inode
);
4085 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4090 leaf
= path
->nodes
[0];
4091 slot
= path
->slots
[0];
4092 if (slot
>= btrfs_header_nritems(leaf
)) {
4093 ret
= btrfs_next_leaf(root
, path
);
4101 item
= btrfs_item_nr(leaf
, slot
);
4102 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4104 if (found_key
.objectid
!= key
.objectid
)
4106 if (btrfs_key_type(&found_key
) != key_type
)
4108 if (found_key
.offset
< filp
->f_pos
)
4110 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
4111 btrfs_should_delete_dir_index(&del_list
,
4115 filp
->f_pos
= found_key
.offset
;
4118 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4120 di_total
= btrfs_item_size(leaf
, item
);
4122 while (di_cur
< di_total
) {
4123 struct btrfs_key location
;
4126 if (verify_dir_item(root
, leaf
, di
))
4129 name_len
= btrfs_dir_name_len(leaf
, di
);
4130 if (name_len
<= sizeof(tmp_name
)) {
4131 name_ptr
= tmp_name
;
4133 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4139 read_extent_buffer(leaf
, name_ptr
,
4140 (unsigned long)(di
+ 1), name_len
);
4142 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4143 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4147 q
.hash
= full_name_hash(q
.name
, q
.len
);
4148 tmp
= d_lookup(filp
->f_dentry
, &q
);
4150 struct btrfs_key
*newkey
;
4152 newkey
= kzalloc(sizeof(struct btrfs_key
),
4156 tmp
= d_alloc(filp
->f_dentry
, &q
);
4162 memcpy(newkey
, &location
,
4163 sizeof(struct btrfs_key
));
4164 tmp
->d_fsdata
= newkey
;
4165 tmp
->d_flags
|= DCACHE_NEED_LOOKUP
;
4172 /* is this a reference to our own snapshot? If so
4175 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4176 location
.objectid
== root
->root_key
.objectid
) {
4180 over
= filldir(dirent
, name_ptr
, name_len
,
4181 found_key
.offset
, location
.objectid
,
4185 if (name_ptr
!= tmp_name
)
4190 di_len
= btrfs_dir_name_len(leaf
, di
) +
4191 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4193 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4199 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4202 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
4208 /* Reached end of directory/root. Bump pos past the last item. */
4209 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4211 * 32-bit glibc will use getdents64, but then strtol -
4212 * so the last number we can serve is this.
4214 filp
->f_pos
= 0x7fffffff;
4220 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4221 btrfs_put_delayed_items(&ins_list
, &del_list
);
4222 btrfs_free_path(path
);
4226 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4228 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4229 struct btrfs_trans_handle
*trans
;
4231 bool nolock
= false;
4233 if (BTRFS_I(inode
)->dummy_inode
)
4236 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(root
, inode
))
4239 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4241 trans
= btrfs_join_transaction_nolock(root
);
4243 trans
= btrfs_join_transaction(root
);
4245 return PTR_ERR(trans
);
4247 ret
= btrfs_end_transaction_nolock(trans
, root
);
4249 ret
= btrfs_commit_transaction(trans
, root
);
4255 * This is somewhat expensive, updating the tree every time the
4256 * inode changes. But, it is most likely to find the inode in cache.
4257 * FIXME, needs more benchmarking...there are no reasons other than performance
4258 * to keep or drop this code.
4260 void btrfs_dirty_inode(struct inode
*inode
, int flags
)
4262 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4263 struct btrfs_trans_handle
*trans
;
4266 if (BTRFS_I(inode
)->dummy_inode
)
4269 trans
= btrfs_join_transaction(root
);
4270 BUG_ON(IS_ERR(trans
));
4272 ret
= btrfs_update_inode(trans
, root
, inode
);
4273 if (ret
&& ret
== -ENOSPC
) {
4274 /* whoops, lets try again with the full transaction */
4275 btrfs_end_transaction(trans
, root
);
4276 trans
= btrfs_start_transaction(root
, 1);
4277 if (IS_ERR(trans
)) {
4278 printk_ratelimited(KERN_ERR
"btrfs: fail to "
4279 "dirty inode %llu error %ld\n",
4280 (unsigned long long)btrfs_ino(inode
),
4285 ret
= btrfs_update_inode(trans
, root
, inode
);
4287 printk_ratelimited(KERN_ERR
"btrfs: fail to "
4288 "dirty inode %llu error %d\n",
4289 (unsigned long long)btrfs_ino(inode
),
4293 btrfs_end_transaction(trans
, root
);
4294 if (BTRFS_I(inode
)->delayed_node
)
4295 btrfs_balance_delayed_items(root
);
4299 * find the highest existing sequence number in a directory
4300 * and then set the in-memory index_cnt variable to reflect
4301 * free sequence numbers
4303 static int btrfs_set_inode_index_count(struct inode
*inode
)
4305 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4306 struct btrfs_key key
, found_key
;
4307 struct btrfs_path
*path
;
4308 struct extent_buffer
*leaf
;
4311 key
.objectid
= btrfs_ino(inode
);
4312 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4313 key
.offset
= (u64
)-1;
4315 path
= btrfs_alloc_path();
4319 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4322 /* FIXME: we should be able to handle this */
4328 * MAGIC NUMBER EXPLANATION:
4329 * since we search a directory based on f_pos we have to start at 2
4330 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4331 * else has to start at 2
4333 if (path
->slots
[0] == 0) {
4334 BTRFS_I(inode
)->index_cnt
= 2;
4340 leaf
= path
->nodes
[0];
4341 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4343 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4344 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4345 BTRFS_I(inode
)->index_cnt
= 2;
4349 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4351 btrfs_free_path(path
);
4356 * helper to find a free sequence number in a given directory. This current
4357 * code is very simple, later versions will do smarter things in the btree
4359 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4363 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4364 ret
= btrfs_inode_delayed_dir_index_count(dir
);
4366 ret
= btrfs_set_inode_index_count(dir
);
4372 *index
= BTRFS_I(dir
)->index_cnt
;
4373 BTRFS_I(dir
)->index_cnt
++;
4378 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4379 struct btrfs_root
*root
,
4381 const char *name
, int name_len
,
4382 u64 ref_objectid
, u64 objectid
, int mode
,
4385 struct inode
*inode
;
4386 struct btrfs_inode_item
*inode_item
;
4387 struct btrfs_key
*location
;
4388 struct btrfs_path
*path
;
4389 struct btrfs_inode_ref
*ref
;
4390 struct btrfs_key key
[2];
4396 path
= btrfs_alloc_path();
4398 return ERR_PTR(-ENOMEM
);
4400 inode
= new_inode(root
->fs_info
->sb
);
4402 btrfs_free_path(path
);
4403 return ERR_PTR(-ENOMEM
);
4407 * we have to initialize this early, so we can reclaim the inode
4408 * number if we fail afterwards in this function.
4410 inode
->i_ino
= objectid
;
4413 trace_btrfs_inode_request(dir
);
4415 ret
= btrfs_set_inode_index(dir
, index
);
4417 btrfs_free_path(path
);
4419 return ERR_PTR(ret
);
4423 * index_cnt is ignored for everything but a dir,
4424 * btrfs_get_inode_index_count has an explanation for the magic
4427 BTRFS_I(inode
)->index_cnt
= 2;
4428 BTRFS_I(inode
)->root
= root
;
4429 BTRFS_I(inode
)->generation
= trans
->transid
;
4430 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4431 btrfs_set_inode_space_info(root
, inode
);
4438 key
[0].objectid
= objectid
;
4439 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4442 key
[1].objectid
= objectid
;
4443 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4444 key
[1].offset
= ref_objectid
;
4446 sizes
[0] = sizeof(struct btrfs_inode_item
);
4447 sizes
[1] = name_len
+ sizeof(*ref
);
4449 path
->leave_spinning
= 1;
4450 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4454 inode_init_owner(inode
, dir
, mode
);
4455 inode_set_bytes(inode
, 0);
4456 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4457 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4458 struct btrfs_inode_item
);
4459 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4461 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4462 struct btrfs_inode_ref
);
4463 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4464 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4465 ptr
= (unsigned long)(ref
+ 1);
4466 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4468 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4469 btrfs_free_path(path
);
4471 location
= &BTRFS_I(inode
)->location
;
4472 location
->objectid
= objectid
;
4473 location
->offset
= 0;
4474 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4476 btrfs_inherit_iflags(inode
, dir
);
4478 if (S_ISREG(mode
)) {
4479 if (btrfs_test_opt(root
, NODATASUM
))
4480 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4481 if (btrfs_test_opt(root
, NODATACOW
) ||
4482 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4483 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4486 insert_inode_hash(inode
);
4487 inode_tree_add(inode
);
4489 trace_btrfs_inode_new(inode
);
4490 btrfs_set_inode_last_trans(trans
, inode
);
4495 BTRFS_I(dir
)->index_cnt
--;
4496 btrfs_free_path(path
);
4498 return ERR_PTR(ret
);
4501 static inline u8
btrfs_inode_type(struct inode
*inode
)
4503 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4507 * utility function to add 'inode' into 'parent_inode' with
4508 * a give name and a given sequence number.
4509 * if 'add_backref' is true, also insert a backref from the
4510 * inode to the parent directory.
4512 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4513 struct inode
*parent_inode
, struct inode
*inode
,
4514 const char *name
, int name_len
, int add_backref
, u64 index
)
4517 struct btrfs_key key
;
4518 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4519 u64 ino
= btrfs_ino(inode
);
4520 u64 parent_ino
= btrfs_ino(parent_inode
);
4522 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4523 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4526 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4530 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4531 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4532 key
.objectid
, root
->root_key
.objectid
,
4533 parent_ino
, index
, name
, name_len
);
4534 } else if (add_backref
) {
4535 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
4540 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4542 btrfs_inode_type(inode
), index
);
4545 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4547 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4548 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4553 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4554 struct inode
*dir
, struct dentry
*dentry
,
4555 struct inode
*inode
, int backref
, u64 index
)
4557 int err
= btrfs_add_link(trans
, dir
, inode
,
4558 dentry
->d_name
.name
, dentry
->d_name
.len
,
4561 d_instantiate(dentry
, inode
);
4569 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4570 int mode
, dev_t rdev
)
4572 struct btrfs_trans_handle
*trans
;
4573 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4574 struct inode
*inode
= NULL
;
4578 unsigned long nr
= 0;
4581 if (!new_valid_dev(rdev
))
4585 * 2 for inode item and ref
4587 * 1 for xattr if selinux is on
4589 trans
= btrfs_start_transaction(root
, 5);
4591 return PTR_ERR(trans
);
4593 err
= btrfs_find_free_ino(root
, &objectid
);
4597 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4598 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4600 if (IS_ERR(inode
)) {
4601 err
= PTR_ERR(inode
);
4605 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4611 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4615 inode
->i_op
= &btrfs_special_inode_operations
;
4616 init_special_inode(inode
, inode
->i_mode
, rdev
);
4617 btrfs_update_inode(trans
, root
, inode
);
4620 nr
= trans
->blocks_used
;
4621 btrfs_end_transaction_throttle(trans
, root
);
4622 btrfs_btree_balance_dirty(root
, nr
);
4624 inode_dec_link_count(inode
);
4630 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4631 int mode
, struct nameidata
*nd
)
4633 struct btrfs_trans_handle
*trans
;
4634 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4635 struct inode
*inode
= NULL
;
4638 unsigned long nr
= 0;
4643 * 2 for inode item and ref
4645 * 1 for xattr if selinux is on
4647 trans
= btrfs_start_transaction(root
, 5);
4649 return PTR_ERR(trans
);
4651 err
= btrfs_find_free_ino(root
, &objectid
);
4655 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4656 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4658 if (IS_ERR(inode
)) {
4659 err
= PTR_ERR(inode
);
4663 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4669 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4673 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4674 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4675 inode
->i_fop
= &btrfs_file_operations
;
4676 inode
->i_op
= &btrfs_file_inode_operations
;
4677 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4680 nr
= trans
->blocks_used
;
4681 btrfs_end_transaction_throttle(trans
, root
);
4683 inode_dec_link_count(inode
);
4686 btrfs_btree_balance_dirty(root
, nr
);
4690 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4691 struct dentry
*dentry
)
4693 struct btrfs_trans_handle
*trans
;
4694 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4695 struct inode
*inode
= old_dentry
->d_inode
;
4697 unsigned long nr
= 0;
4701 /* do not allow sys_link's with other subvols of the same device */
4702 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4705 if (inode
->i_nlink
== ~0U)
4708 err
= btrfs_set_inode_index(dir
, &index
);
4713 * 2 items for inode and inode ref
4714 * 2 items for dir items
4715 * 1 item for parent inode
4717 trans
= btrfs_start_transaction(root
, 5);
4718 if (IS_ERR(trans
)) {
4719 err
= PTR_ERR(trans
);
4723 btrfs_inc_nlink(inode
);
4724 inode
->i_ctime
= CURRENT_TIME
;
4727 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4732 struct dentry
*parent
= dentry
->d_parent
;
4733 err
= btrfs_update_inode(trans
, root
, inode
);
4735 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4738 nr
= trans
->blocks_used
;
4739 btrfs_end_transaction_throttle(trans
, root
);
4742 inode_dec_link_count(inode
);
4745 btrfs_btree_balance_dirty(root
, nr
);
4749 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4751 struct inode
*inode
= NULL
;
4752 struct btrfs_trans_handle
*trans
;
4753 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4755 int drop_on_err
= 0;
4758 unsigned long nr
= 1;
4761 * 2 items for inode and ref
4762 * 2 items for dir items
4763 * 1 for xattr if selinux is on
4765 trans
= btrfs_start_transaction(root
, 5);
4767 return PTR_ERR(trans
);
4769 err
= btrfs_find_free_ino(root
, &objectid
);
4773 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4774 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4775 S_IFDIR
| mode
, &index
);
4776 if (IS_ERR(inode
)) {
4777 err
= PTR_ERR(inode
);
4783 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4787 inode
->i_op
= &btrfs_dir_inode_operations
;
4788 inode
->i_fop
= &btrfs_dir_file_operations
;
4790 btrfs_i_size_write(inode
, 0);
4791 err
= btrfs_update_inode(trans
, root
, inode
);
4795 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4796 dentry
->d_name
.len
, 0, index
);
4800 d_instantiate(dentry
, inode
);
4804 nr
= trans
->blocks_used
;
4805 btrfs_end_transaction_throttle(trans
, root
);
4808 btrfs_btree_balance_dirty(root
, nr
);
4812 /* helper for btfs_get_extent. Given an existing extent in the tree,
4813 * and an extent that you want to insert, deal with overlap and insert
4814 * the new extent into the tree.
4816 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4817 struct extent_map
*existing
,
4818 struct extent_map
*em
,
4819 u64 map_start
, u64 map_len
)
4823 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4824 start_diff
= map_start
- em
->start
;
4825 em
->start
= map_start
;
4827 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4828 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4829 em
->block_start
+= start_diff
;
4830 em
->block_len
-= start_diff
;
4832 return add_extent_mapping(em_tree
, em
);
4835 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4836 struct inode
*inode
, struct page
*page
,
4837 size_t pg_offset
, u64 extent_offset
,
4838 struct btrfs_file_extent_item
*item
)
4841 struct extent_buffer
*leaf
= path
->nodes
[0];
4844 unsigned long inline_size
;
4848 WARN_ON(pg_offset
!= 0);
4849 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4850 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4851 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4852 btrfs_item_nr(leaf
, path
->slots
[0]));
4853 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4856 ptr
= btrfs_file_extent_inline_start(item
);
4858 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4860 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4861 ret
= btrfs_decompress(compress_type
, tmp
, page
,
4862 extent_offset
, inline_size
, max_size
);
4864 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4865 unsigned long copy_size
= min_t(u64
,
4866 PAGE_CACHE_SIZE
- pg_offset
,
4867 max_size
- extent_offset
);
4868 memset(kaddr
+ pg_offset
, 0, copy_size
);
4869 kunmap_atomic(kaddr
, KM_USER0
);
4876 * a bit scary, this does extent mapping from logical file offset to the disk.
4877 * the ugly parts come from merging extents from the disk with the in-ram
4878 * representation. This gets more complex because of the data=ordered code,
4879 * where the in-ram extents might be locked pending data=ordered completion.
4881 * This also copies inline extents directly into the page.
4884 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
4885 size_t pg_offset
, u64 start
, u64 len
,
4891 u64 extent_start
= 0;
4893 u64 objectid
= btrfs_ino(inode
);
4895 struct btrfs_path
*path
= NULL
;
4896 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4897 struct btrfs_file_extent_item
*item
;
4898 struct extent_buffer
*leaf
;
4899 struct btrfs_key found_key
;
4900 struct extent_map
*em
= NULL
;
4901 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4902 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4903 struct btrfs_trans_handle
*trans
= NULL
;
4907 read_lock(&em_tree
->lock
);
4908 em
= lookup_extent_mapping(em_tree
, start
, len
);
4910 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4911 read_unlock(&em_tree
->lock
);
4914 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
4915 free_extent_map(em
);
4916 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
4917 free_extent_map(em
);
4921 em
= alloc_extent_map();
4926 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4927 em
->start
= EXTENT_MAP_HOLE
;
4928 em
->orig_start
= EXTENT_MAP_HOLE
;
4930 em
->block_len
= (u64
)-1;
4933 path
= btrfs_alloc_path();
4939 * Chances are we'll be called again, so go ahead and do
4945 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
4946 objectid
, start
, trans
!= NULL
);
4953 if (path
->slots
[0] == 0)
4958 leaf
= path
->nodes
[0];
4959 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4960 struct btrfs_file_extent_item
);
4961 /* are we inside the extent that was found? */
4962 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4963 found_type
= btrfs_key_type(&found_key
);
4964 if (found_key
.objectid
!= objectid
||
4965 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4969 found_type
= btrfs_file_extent_type(leaf
, item
);
4970 extent_start
= found_key
.offset
;
4971 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4972 if (found_type
== BTRFS_FILE_EXTENT_REG
||
4973 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4974 extent_end
= extent_start
+
4975 btrfs_file_extent_num_bytes(leaf
, item
);
4976 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4978 size
= btrfs_file_extent_inline_len(leaf
, item
);
4979 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
4980 ~((u64
)root
->sectorsize
- 1);
4983 if (start
>= extent_end
) {
4985 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
4986 ret
= btrfs_next_leaf(root
, path
);
4993 leaf
= path
->nodes
[0];
4995 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4996 if (found_key
.objectid
!= objectid
||
4997 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4999 if (start
+ len
<= found_key
.offset
)
5002 em
->len
= found_key
.offset
- start
;
5006 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5007 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5008 em
->start
= extent_start
;
5009 em
->len
= extent_end
- extent_start
;
5010 em
->orig_start
= extent_start
-
5011 btrfs_file_extent_offset(leaf
, item
);
5012 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5014 em
->block_start
= EXTENT_MAP_HOLE
;
5017 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5018 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5019 em
->compress_type
= compress_type
;
5020 em
->block_start
= bytenr
;
5021 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5024 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5025 em
->block_start
= bytenr
;
5026 em
->block_len
= em
->len
;
5027 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5028 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5031 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5035 size_t extent_offset
;
5038 em
->block_start
= EXTENT_MAP_INLINE
;
5039 if (!page
|| create
) {
5040 em
->start
= extent_start
;
5041 em
->len
= extent_end
- extent_start
;
5045 size
= btrfs_file_extent_inline_len(leaf
, item
);
5046 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5047 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5048 size
- extent_offset
);
5049 em
->start
= extent_start
+ extent_offset
;
5050 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5051 ~((u64
)root
->sectorsize
- 1);
5052 em
->orig_start
= EXTENT_MAP_INLINE
;
5053 if (compress_type
) {
5054 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5055 em
->compress_type
= compress_type
;
5057 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5058 if (create
== 0 && !PageUptodate(page
)) {
5059 if (btrfs_file_extent_compression(leaf
, item
) !=
5060 BTRFS_COMPRESS_NONE
) {
5061 ret
= uncompress_inline(path
, inode
, page
,
5063 extent_offset
, item
);
5067 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5069 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5070 memset(map
+ pg_offset
+ copy_size
, 0,
5071 PAGE_CACHE_SIZE
- pg_offset
-
5076 flush_dcache_page(page
);
5077 } else if (create
&& PageUptodate(page
)) {
5081 free_extent_map(em
);
5084 btrfs_release_path(path
);
5085 trans
= btrfs_join_transaction(root
);
5088 return ERR_CAST(trans
);
5092 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5095 btrfs_mark_buffer_dirty(leaf
);
5097 set_extent_uptodate(io_tree
, em
->start
,
5098 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5101 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5108 em
->block_start
= EXTENT_MAP_HOLE
;
5109 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5111 btrfs_release_path(path
);
5112 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5113 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5114 "[%llu %llu]\n", (unsigned long long)em
->start
,
5115 (unsigned long long)em
->len
,
5116 (unsigned long long)start
,
5117 (unsigned long long)len
);
5123 write_lock(&em_tree
->lock
);
5124 ret
= add_extent_mapping(em_tree
, em
);
5125 /* it is possible that someone inserted the extent into the tree
5126 * while we had the lock dropped. It is also possible that
5127 * an overlapping map exists in the tree
5129 if (ret
== -EEXIST
) {
5130 struct extent_map
*existing
;
5134 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5135 if (existing
&& (existing
->start
> start
||
5136 existing
->start
+ existing
->len
<= start
)) {
5137 free_extent_map(existing
);
5141 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5144 err
= merge_extent_mapping(em_tree
, existing
,
5147 free_extent_map(existing
);
5149 free_extent_map(em
);
5154 free_extent_map(em
);
5158 free_extent_map(em
);
5163 write_unlock(&em_tree
->lock
);
5166 trace_btrfs_get_extent(root
, em
);
5169 btrfs_free_path(path
);
5171 ret
= btrfs_end_transaction(trans
, root
);
5176 free_extent_map(em
);
5177 return ERR_PTR(err
);
5182 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5183 size_t pg_offset
, u64 start
, u64 len
,
5186 struct extent_map
*em
;
5187 struct extent_map
*hole_em
= NULL
;
5188 u64 range_start
= start
;
5194 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5199 * if our em maps to a hole, there might
5200 * actually be delalloc bytes behind it
5202 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5208 /* check to see if we've wrapped (len == -1 or similar) */
5217 /* ok, we didn't find anything, lets look for delalloc */
5218 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5219 end
, len
, EXTENT_DELALLOC
, 1);
5220 found_end
= range_start
+ found
;
5221 if (found_end
< range_start
)
5222 found_end
= (u64
)-1;
5225 * we didn't find anything useful, return
5226 * the original results from get_extent()
5228 if (range_start
> end
|| found_end
<= start
) {
5234 /* adjust the range_start to make sure it doesn't
5235 * go backwards from the start they passed in
5237 range_start
= max(start
,range_start
);
5238 found
= found_end
- range_start
;
5241 u64 hole_start
= start
;
5244 em
= alloc_extent_map();
5250 * when btrfs_get_extent can't find anything it
5251 * returns one huge hole
5253 * make sure what it found really fits our range, and
5254 * adjust to make sure it is based on the start from
5258 u64 calc_end
= extent_map_end(hole_em
);
5260 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5261 free_extent_map(hole_em
);
5264 hole_start
= max(hole_em
->start
, start
);
5265 hole_len
= calc_end
- hole_start
;
5269 if (hole_em
&& range_start
> hole_start
) {
5270 /* our hole starts before our delalloc, so we
5271 * have to return just the parts of the hole
5272 * that go until the delalloc starts
5274 em
->len
= min(hole_len
,
5275 range_start
- hole_start
);
5276 em
->start
= hole_start
;
5277 em
->orig_start
= hole_start
;
5279 * don't adjust block start at all,
5280 * it is fixed at EXTENT_MAP_HOLE
5282 em
->block_start
= hole_em
->block_start
;
5283 em
->block_len
= hole_len
;
5285 em
->start
= range_start
;
5287 em
->orig_start
= range_start
;
5288 em
->block_start
= EXTENT_MAP_DELALLOC
;
5289 em
->block_len
= found
;
5291 } else if (hole_em
) {
5296 free_extent_map(hole_em
);
5298 free_extent_map(em
);
5299 return ERR_PTR(err
);
5304 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5305 struct extent_map
*em
,
5308 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5309 struct btrfs_trans_handle
*trans
;
5310 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5311 struct btrfs_key ins
;
5314 bool insert
= false;
5317 * Ok if the extent map we looked up is a hole and is for the exact
5318 * range we want, there is no reason to allocate a new one, however if
5319 * it is not right then we need to free this one and drop the cache for
5322 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5324 free_extent_map(em
);
5327 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5330 trans
= btrfs_join_transaction(root
);
5332 return ERR_CAST(trans
);
5334 if (start
<= BTRFS_I(inode
)->disk_i_size
&& len
< 64 * 1024)
5335 btrfs_add_inode_defrag(trans
, inode
);
5337 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5339 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5340 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5341 alloc_hint
, (u64
)-1, &ins
, 1);
5348 em
= alloc_extent_map();
5350 em
= ERR_PTR(-ENOMEM
);
5356 em
->orig_start
= em
->start
;
5357 em
->len
= ins
.offset
;
5359 em
->block_start
= ins
.objectid
;
5360 em
->block_len
= ins
.offset
;
5361 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5364 * We need to do this because if we're using the original em we searched
5365 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5368 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5371 write_lock(&em_tree
->lock
);
5372 ret
= add_extent_mapping(em_tree
, em
);
5373 write_unlock(&em_tree
->lock
);
5376 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5379 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5380 ins
.offset
, ins
.offset
, 0);
5382 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5386 btrfs_end_transaction(trans
, root
);
5391 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5392 * block must be cow'd
5394 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5395 struct inode
*inode
, u64 offset
, u64 len
)
5397 struct btrfs_path
*path
;
5399 struct extent_buffer
*leaf
;
5400 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5401 struct btrfs_file_extent_item
*fi
;
5402 struct btrfs_key key
;
5410 path
= btrfs_alloc_path();
5414 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
5419 slot
= path
->slots
[0];
5422 /* can't find the item, must cow */
5429 leaf
= path
->nodes
[0];
5430 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5431 if (key
.objectid
!= btrfs_ino(inode
) ||
5432 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5433 /* not our file or wrong item type, must cow */
5437 if (key
.offset
> offset
) {
5438 /* Wrong offset, must cow */
5442 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5443 found_type
= btrfs_file_extent_type(leaf
, fi
);
5444 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5445 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5446 /* not a regular extent, must cow */
5449 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5450 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5452 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5453 if (extent_end
< offset
+ len
) {
5454 /* extent doesn't include our full range, must cow */
5458 if (btrfs_extent_readonly(root
, disk_bytenr
))
5462 * look for other files referencing this extent, if we
5463 * find any we must cow
5465 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
5466 key
.offset
- backref_offset
, disk_bytenr
))
5470 * adjust disk_bytenr and num_bytes to cover just the bytes
5471 * in this extent we are about to write. If there
5472 * are any csums in that range we have to cow in order
5473 * to keep the csums correct
5475 disk_bytenr
+= backref_offset
;
5476 disk_bytenr
+= offset
- key
.offset
;
5477 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5478 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5481 * all of the above have passed, it is safe to overwrite this extent
5486 btrfs_free_path(path
);
5490 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5491 struct buffer_head
*bh_result
, int create
)
5493 struct extent_map
*em
;
5494 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5495 u64 start
= iblock
<< inode
->i_blkbits
;
5496 u64 len
= bh_result
->b_size
;
5497 struct btrfs_trans_handle
*trans
;
5499 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5504 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5505 * io. INLINE is special, and we could probably kludge it in here, but
5506 * it's still buffered so for safety lets just fall back to the generic
5509 * For COMPRESSED we _have_ to read the entire extent in so we can
5510 * decompress it, so there will be buffering required no matter what we
5511 * do, so go ahead and fallback to buffered.
5513 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5514 * to buffered IO. Don't blame me, this is the price we pay for using
5517 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5518 em
->block_start
== EXTENT_MAP_INLINE
) {
5519 free_extent_map(em
);
5523 /* Just a good old fashioned hole, return */
5524 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5525 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5526 free_extent_map(em
);
5527 /* DIO will do one hole at a time, so just unlock a sector */
5528 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5529 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5534 * We don't allocate a new extent in the following cases
5536 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5538 * 2) The extent is marked as PREALLOC. We're good to go here and can
5539 * just use the extent.
5543 len
= em
->len
- (start
- em
->start
);
5547 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5548 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5549 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5554 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5555 type
= BTRFS_ORDERED_PREALLOC
;
5557 type
= BTRFS_ORDERED_NOCOW
;
5558 len
= min(len
, em
->len
- (start
- em
->start
));
5559 block_start
= em
->block_start
+ (start
- em
->start
);
5562 * we're not going to log anything, but we do need
5563 * to make sure the current transaction stays open
5564 * while we look for nocow cross refs
5566 trans
= btrfs_join_transaction(root
);
5570 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5571 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5572 block_start
, len
, len
, type
);
5573 btrfs_end_transaction(trans
, root
);
5575 free_extent_map(em
);
5580 btrfs_end_transaction(trans
, root
);
5584 * this will cow the extent, reset the len in case we changed
5587 len
= bh_result
->b_size
;
5588 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
5591 len
= min(len
, em
->len
- (start
- em
->start
));
5593 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5594 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5597 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5599 bh_result
->b_size
= len
;
5600 bh_result
->b_bdev
= em
->bdev
;
5601 set_buffer_mapped(bh_result
);
5602 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5603 set_buffer_new(bh_result
);
5605 free_extent_map(em
);
5610 struct btrfs_dio_private
{
5611 struct inode
*inode
;
5618 /* number of bios pending for this dio */
5619 atomic_t pending_bios
;
5624 struct bio
*orig_bio
;
5627 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5629 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5630 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5631 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5632 struct inode
*inode
= dip
->inode
;
5633 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5635 u32
*private = dip
->csums
;
5637 start
= dip
->logical_offset
;
5639 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5640 struct page
*page
= bvec
->bv_page
;
5643 unsigned long flags
;
5645 local_irq_save(flags
);
5646 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5647 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5648 csum
, bvec
->bv_len
);
5649 btrfs_csum_final(csum
, (char *)&csum
);
5650 kunmap_atomic(kaddr
, KM_IRQ0
);
5651 local_irq_restore(flags
);
5653 flush_dcache_page(bvec
->bv_page
);
5654 if (csum
!= *private) {
5655 printk(KERN_ERR
"btrfs csum failed ino %llu off"
5656 " %llu csum %u private %u\n",
5657 (unsigned long long)btrfs_ino(inode
),
5658 (unsigned long long)start
,
5664 start
+= bvec
->bv_len
;
5667 } while (bvec
<= bvec_end
);
5669 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5670 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5671 bio
->bi_private
= dip
->private;
5676 /* If we had a csum failure make sure to clear the uptodate flag */
5678 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5679 dio_end_io(bio
, err
);
5682 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5684 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5685 struct inode
*inode
= dip
->inode
;
5686 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5687 struct btrfs_trans_handle
*trans
;
5688 struct btrfs_ordered_extent
*ordered
= NULL
;
5689 struct extent_state
*cached_state
= NULL
;
5690 u64 ordered_offset
= dip
->logical_offset
;
5691 u64 ordered_bytes
= dip
->bytes
;
5697 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5705 trans
= btrfs_join_transaction(root
);
5706 if (IS_ERR(trans
)) {
5710 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5712 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5713 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5715 ret
= btrfs_update_inode(trans
, root
, inode
);
5720 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5721 ordered
->file_offset
+ ordered
->len
- 1, 0,
5722 &cached_state
, GFP_NOFS
);
5724 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5725 ret
= btrfs_mark_extent_written(trans
, inode
,
5726 ordered
->file_offset
,
5727 ordered
->file_offset
+
5734 ret
= insert_reserved_file_extent(trans
, inode
,
5735 ordered
->file_offset
,
5741 BTRFS_FILE_EXTENT_REG
);
5742 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5743 ordered
->file_offset
, ordered
->len
);
5751 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5752 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5753 if (!ret
|| !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
))
5754 btrfs_update_inode(trans
, root
, inode
);
5757 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5758 ordered
->file_offset
+ ordered
->len
- 1,
5759 &cached_state
, GFP_NOFS
);
5761 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5762 btrfs_end_transaction(trans
, root
);
5763 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5764 btrfs_put_ordered_extent(ordered
);
5765 btrfs_put_ordered_extent(ordered
);
5769 * our bio might span multiple ordered extents. If we haven't
5770 * completed the accounting for the whole dio, go back and try again
5772 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5773 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5778 bio
->bi_private
= dip
->private;
5783 /* If we had an error make sure to clear the uptodate flag */
5785 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5786 dio_end_io(bio
, err
);
5789 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5790 struct bio
*bio
, int mirror_num
,
5791 unsigned long bio_flags
, u64 offset
)
5794 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5795 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5800 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5802 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5805 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
5806 "sector %#Lx len %u err no %d\n",
5807 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
5808 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5812 * before atomic variable goto zero, we must make sure
5813 * dip->errors is perceived to be set.
5815 smp_mb__before_atomic_dec();
5818 /* if there are more bios still pending for this dio, just exit */
5819 if (!atomic_dec_and_test(&dip
->pending_bios
))
5823 bio_io_error(dip
->orig_bio
);
5825 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5826 bio_endio(dip
->orig_bio
, 0);
5832 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5833 u64 first_sector
, gfp_t gfp_flags
)
5835 int nr_vecs
= bio_get_nr_vecs(bdev
);
5836 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5839 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5840 int rw
, u64 file_offset
, int skip_sum
,
5841 u32
*csums
, int async_submit
)
5843 int write
= rw
& REQ_WRITE
;
5844 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5848 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5855 if (write
&& async_submit
) {
5856 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5857 inode
, rw
, bio
, 0, 0,
5859 __btrfs_submit_bio_start_direct_io
,
5860 __btrfs_submit_bio_done
);
5864 * If we aren't doing async submit, calculate the csum of the
5867 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
5870 } else if (!skip_sum
) {
5871 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5872 file_offset
, csums
);
5878 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
5884 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5887 struct inode
*inode
= dip
->inode
;
5888 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5889 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5891 struct bio
*orig_bio
= dip
->orig_bio
;
5892 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5893 u64 start_sector
= orig_bio
->bi_sector
;
5894 u64 file_offset
= dip
->logical_offset
;
5898 u32
*csums
= dip
->csums
;
5900 int async_submit
= 0;
5901 int write
= rw
& REQ_WRITE
;
5903 map_length
= orig_bio
->bi_size
;
5904 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5905 &map_length
, NULL
, 0);
5911 if (map_length
>= orig_bio
->bi_size
) {
5917 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
5920 bio
->bi_private
= dip
;
5921 bio
->bi_end_io
= btrfs_end_dio_bio
;
5922 atomic_inc(&dip
->pending_bios
);
5924 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
5925 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
5926 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5927 bvec
->bv_offset
) < bvec
->bv_len
)) {
5929 * inc the count before we submit the bio so
5930 * we know the end IO handler won't happen before
5931 * we inc the count. Otherwise, the dip might get freed
5932 * before we're done setting it up
5934 atomic_inc(&dip
->pending_bios
);
5935 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
5936 file_offset
, skip_sum
,
5937 csums
, async_submit
);
5940 atomic_dec(&dip
->pending_bios
);
5944 /* Write's use the ordered csums */
5945 if (!write
&& !skip_sum
)
5946 csums
= csums
+ nr_pages
;
5947 start_sector
+= submit_len
>> 9;
5948 file_offset
+= submit_len
;
5953 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
5954 start_sector
, GFP_NOFS
);
5957 bio
->bi_private
= dip
;
5958 bio
->bi_end_io
= btrfs_end_dio_bio
;
5960 map_length
= orig_bio
->bi_size
;
5961 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5962 &map_length
, NULL
, 0);
5968 submit_len
+= bvec
->bv_len
;
5975 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
5976 csums
, async_submit
);
5984 * before atomic variable goto zero, we must
5985 * make sure dip->errors is perceived to be set.
5987 smp_mb__before_atomic_dec();
5988 if (atomic_dec_and_test(&dip
->pending_bios
))
5989 bio_io_error(dip
->orig_bio
);
5991 /* bio_end_io() will handle error, so we needn't return it */
5995 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
5998 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5999 struct btrfs_dio_private
*dip
;
6000 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6002 int write
= rw
& REQ_WRITE
;
6005 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6007 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6014 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6015 if (!write
&& !skip_sum
) {
6016 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6024 dip
->private = bio
->bi_private
;
6026 dip
->logical_offset
= file_offset
;
6030 dip
->bytes
+= bvec
->bv_len
;
6032 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6034 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6035 bio
->bi_private
= dip
;
6037 dip
->orig_bio
= bio
;
6038 atomic_set(&dip
->pending_bios
, 0);
6041 bio
->bi_end_io
= btrfs_endio_direct_write
;
6043 bio
->bi_end_io
= btrfs_endio_direct_read
;
6045 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6050 * If this is a write, we need to clean up the reserved space and kill
6051 * the ordered extent.
6054 struct btrfs_ordered_extent
*ordered
;
6055 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6056 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6057 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6058 btrfs_free_reserved_extent(root
, ordered
->start
,
6060 btrfs_put_ordered_extent(ordered
);
6061 btrfs_put_ordered_extent(ordered
);
6063 bio_endio(bio
, ret
);
6066 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6067 const struct iovec
*iov
, loff_t offset
,
6068 unsigned long nr_segs
)
6074 unsigned blocksize_mask
= root
->sectorsize
- 1;
6075 ssize_t retval
= -EINVAL
;
6076 loff_t end
= offset
;
6078 if (offset
& blocksize_mask
)
6081 /* Check the memory alignment. Blocks cannot straddle pages */
6082 for (seg
= 0; seg
< nr_segs
; seg
++) {
6083 addr
= (unsigned long)iov
[seg
].iov_base
;
6084 size
= iov
[seg
].iov_len
;
6086 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6089 /* If this is a write we don't need to check anymore */
6094 * Check to make sure we don't have duplicate iov_base's in this
6095 * iovec, if so return EINVAL, otherwise we'll get csum errors
6096 * when reading back.
6098 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6099 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6107 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6108 const struct iovec
*iov
, loff_t offset
,
6109 unsigned long nr_segs
)
6111 struct file
*file
= iocb
->ki_filp
;
6112 struct inode
*inode
= file
->f_mapping
->host
;
6113 struct btrfs_ordered_extent
*ordered
;
6114 struct extent_state
*cached_state
= NULL
;
6115 u64 lockstart
, lockend
;
6117 int writing
= rw
& WRITE
;
6119 size_t count
= iov_length(iov
, nr_segs
);
6121 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6127 lockend
= offset
+ count
- 1;
6130 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6136 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6137 0, &cached_state
, GFP_NOFS
);
6139 * We're concerned with the entire range that we're going to be
6140 * doing DIO to, so we need to make sure theres no ordered
6141 * extents in this range.
6143 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6144 lockend
- lockstart
+ 1);
6147 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6148 &cached_state
, GFP_NOFS
);
6149 btrfs_start_ordered_extent(inode
, ordered
, 1);
6150 btrfs_put_ordered_extent(ordered
);
6155 * we don't use btrfs_set_extent_delalloc because we don't want
6156 * the dirty or uptodate bits
6159 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6160 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6161 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6164 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6165 lockend
, EXTENT_LOCKED
| write_bits
,
6166 1, 0, &cached_state
, GFP_NOFS
);
6171 free_extent_state(cached_state
);
6172 cached_state
= NULL
;
6174 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6175 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6176 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6177 btrfs_submit_direct
, 0);
6179 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6180 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6181 offset
+ iov_length(iov
, nr_segs
) - 1,
6182 EXTENT_LOCKED
| write_bits
, 1, 0,
6183 &cached_state
, GFP_NOFS
);
6184 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6186 * We're falling back to buffered, unlock the section we didn't
6189 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6190 offset
+ iov_length(iov
, nr_segs
) - 1,
6191 EXTENT_LOCKED
| write_bits
, 1, 0,
6192 &cached_state
, GFP_NOFS
);
6195 free_extent_state(cached_state
);
6199 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6200 __u64 start
, __u64 len
)
6202 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6205 int btrfs_readpage(struct file
*file
, struct page
*page
)
6207 struct extent_io_tree
*tree
;
6208 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6209 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6212 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6214 struct extent_io_tree
*tree
;
6217 if (current
->flags
& PF_MEMALLOC
) {
6218 redirty_page_for_writepage(wbc
, page
);
6222 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6223 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6226 int btrfs_writepages(struct address_space
*mapping
,
6227 struct writeback_control
*wbc
)
6229 struct extent_io_tree
*tree
;
6231 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6232 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6236 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6237 struct list_head
*pages
, unsigned nr_pages
)
6239 struct extent_io_tree
*tree
;
6240 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6241 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6244 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6246 struct extent_io_tree
*tree
;
6247 struct extent_map_tree
*map
;
6250 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6251 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6252 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6254 ClearPagePrivate(page
);
6255 set_page_private(page
, 0);
6256 page_cache_release(page
);
6261 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6263 if (PageWriteback(page
) || PageDirty(page
))
6265 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6268 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6270 struct extent_io_tree
*tree
;
6271 struct btrfs_ordered_extent
*ordered
;
6272 struct extent_state
*cached_state
= NULL
;
6273 u64 page_start
= page_offset(page
);
6274 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6278 * we have the page locked, so new writeback can't start,
6279 * and the dirty bit won't be cleared while we are here.
6281 * Wait for IO on this page so that we can safely clear
6282 * the PagePrivate2 bit and do ordered accounting
6284 wait_on_page_writeback(page
);
6286 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6288 btrfs_releasepage(page
, GFP_NOFS
);
6291 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6293 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6297 * IO on this page will never be started, so we need
6298 * to account for any ordered extents now
6300 clear_extent_bit(tree
, page_start
, page_end
,
6301 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6302 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6303 &cached_state
, GFP_NOFS
);
6305 * whoever cleared the private bit is responsible
6306 * for the finish_ordered_io
6308 if (TestClearPagePrivate2(page
)) {
6309 btrfs_finish_ordered_io(page
->mapping
->host
,
6310 page_start
, page_end
);
6312 btrfs_put_ordered_extent(ordered
);
6313 cached_state
= NULL
;
6314 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6317 clear_extent_bit(tree
, page_start
, page_end
,
6318 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6319 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6320 __btrfs_releasepage(page
, GFP_NOFS
);
6322 ClearPageChecked(page
);
6323 if (PagePrivate(page
)) {
6324 ClearPagePrivate(page
);
6325 set_page_private(page
, 0);
6326 page_cache_release(page
);
6331 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6332 * called from a page fault handler when a page is first dirtied. Hence we must
6333 * be careful to check for EOF conditions here. We set the page up correctly
6334 * for a written page which means we get ENOSPC checking when writing into
6335 * holes and correct delalloc and unwritten extent mapping on filesystems that
6336 * support these features.
6338 * We are not allowed to take the i_mutex here so we have to play games to
6339 * protect against truncate races as the page could now be beyond EOF. Because
6340 * vmtruncate() writes the inode size before removing pages, once we have the
6341 * page lock we can determine safely if the page is beyond EOF. If it is not
6342 * beyond EOF, then the page is guaranteed safe against truncation until we
6345 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6347 struct page
*page
= vmf
->page
;
6348 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6349 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6350 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6351 struct btrfs_ordered_extent
*ordered
;
6352 struct extent_state
*cached_state
= NULL
;
6354 unsigned long zero_start
;
6360 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6364 else /* -ENOSPC, -EIO, etc */
6365 ret
= VM_FAULT_SIGBUS
;
6369 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6372 size
= i_size_read(inode
);
6373 page_start
= page_offset(page
);
6374 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6376 if ((page
->mapping
!= inode
->i_mapping
) ||
6377 (page_start
>= size
)) {
6378 /* page got truncated out from underneath us */
6381 wait_on_page_writeback(page
);
6383 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6385 set_page_extent_mapped(page
);
6388 * we can't set the delalloc bits if there are pending ordered
6389 * extents. Drop our locks and wait for them to finish
6391 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6393 unlock_extent_cached(io_tree
, page_start
, page_end
,
6394 &cached_state
, GFP_NOFS
);
6396 btrfs_start_ordered_extent(inode
, ordered
, 1);
6397 btrfs_put_ordered_extent(ordered
);
6402 * XXX - page_mkwrite gets called every time the page is dirtied, even
6403 * if it was already dirty, so for space accounting reasons we need to
6404 * clear any delalloc bits for the range we are fixing to save. There
6405 * is probably a better way to do this, but for now keep consistent with
6406 * prepare_pages in the normal write path.
6408 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6409 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6410 0, 0, &cached_state
, GFP_NOFS
);
6412 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6415 unlock_extent_cached(io_tree
, page_start
, page_end
,
6416 &cached_state
, GFP_NOFS
);
6417 ret
= VM_FAULT_SIGBUS
;
6422 /* page is wholly or partially inside EOF */
6423 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6424 zero_start
= size
& ~PAGE_CACHE_MASK
;
6426 zero_start
= PAGE_CACHE_SIZE
;
6428 if (zero_start
!= PAGE_CACHE_SIZE
) {
6430 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6431 flush_dcache_page(page
);
6434 ClearPageChecked(page
);
6435 set_page_dirty(page
);
6436 SetPageUptodate(page
);
6438 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6439 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6441 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6445 return VM_FAULT_LOCKED
;
6447 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6452 static int btrfs_truncate(struct inode
*inode
)
6454 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6455 struct btrfs_block_rsv
*rsv
;
6458 struct btrfs_trans_handle
*trans
;
6460 u64 mask
= root
->sectorsize
- 1;
6462 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6466 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6467 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6470 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6471 * 3 things going on here
6473 * 1) We need to reserve space for our orphan item and the space to
6474 * delete our orphan item. Lord knows we don't want to have a dangling
6475 * orphan item because we didn't reserve space to remove it.
6477 * 2) We need to reserve space to update our inode.
6479 * 3) We need to have something to cache all the space that is going to
6480 * be free'd up by the truncate operation, but also have some slack
6481 * space reserved in case it uses space during the truncate (thank you
6482 * very much snapshotting).
6484 * And we need these to all be seperate. The fact is we can use alot of
6485 * space doing the truncate, and we have no earthly idea how much space
6486 * we will use, so we need the truncate reservation to be seperate so it
6487 * doesn't end up using space reserved for updating the inode or
6488 * removing the orphan item. We also need to be able to stop the
6489 * transaction and start a new one, which means we need to be able to
6490 * update the inode several times, and we have no idea of knowing how
6491 * many times that will be, so we can't just reserve 1 item for the
6492 * entirety of the opration, so that has to be done seperately as well.
6493 * Then there is the orphan item, which does indeed need to be held on
6494 * to for the whole operation, and we need nobody to touch this reserved
6495 * space except the orphan code.
6497 * So that leaves us with
6499 * 1) root->orphan_block_rsv - for the orphan deletion.
6500 * 2) rsv - for the truncate reservation, which we will steal from the
6501 * transaction reservation.
6502 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6503 * updating the inode.
6505 rsv
= btrfs_alloc_block_rsv(root
);
6508 btrfs_add_durable_block_rsv(root
->fs_info
, rsv
);
6510 trans
= btrfs_start_transaction(root
, 4);
6511 if (IS_ERR(trans
)) {
6512 err
= PTR_ERR(trans
);
6517 * Reserve space for the truncate process. Truncate should be adding
6518 * space, but if there are snapshots it may end up using space.
6520 ret
= btrfs_truncate_reserve_metadata(trans
, root
, rsv
);
6523 ret
= btrfs_orphan_add(trans
, inode
);
6525 btrfs_end_transaction(trans
, root
);
6529 nr
= trans
->blocks_used
;
6530 btrfs_end_transaction(trans
, root
);
6531 btrfs_btree_balance_dirty(root
, nr
);
6534 * Ok so we've already migrated our bytes over for the truncate, so here
6535 * just reserve the one slot we need for updating the inode.
6537 trans
= btrfs_start_transaction(root
, 1);
6538 if (IS_ERR(trans
)) {
6539 err
= PTR_ERR(trans
);
6542 trans
->block_rsv
= rsv
;
6545 * setattr is responsible for setting the ordered_data_close flag,
6546 * but that is only tested during the last file release. That
6547 * could happen well after the next commit, leaving a great big
6548 * window where new writes may get lost if someone chooses to write
6549 * to this file after truncating to zero
6551 * The inode doesn't have any dirty data here, and so if we commit
6552 * this is a noop. If someone immediately starts writing to the inode
6553 * it is very likely we'll catch some of their writes in this
6554 * transaction, and the commit will find this file on the ordered
6555 * data list with good things to send down.
6557 * This is a best effort solution, there is still a window where
6558 * using truncate to replace the contents of the file will
6559 * end up with a zero length file after a crash.
6561 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6562 btrfs_add_ordered_operation(trans
, root
, inode
);
6566 trans
= btrfs_start_transaction(root
, 3);
6567 if (IS_ERR(trans
)) {
6568 err
= PTR_ERR(trans
);
6572 ret
= btrfs_truncate_reserve_metadata(trans
, root
,
6576 trans
->block_rsv
= rsv
;
6579 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6581 BTRFS_EXTENT_DATA_KEY
);
6582 if (ret
!= -EAGAIN
) {
6587 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6588 ret
= btrfs_update_inode(trans
, root
, inode
);
6594 nr
= trans
->blocks_used
;
6595 btrfs_end_transaction(trans
, root
);
6597 btrfs_btree_balance_dirty(root
, nr
);
6600 if (ret
== 0 && inode
->i_nlink
> 0) {
6601 trans
->block_rsv
= root
->orphan_block_rsv
;
6602 ret
= btrfs_orphan_del(trans
, inode
);
6605 } else if (ret
&& inode
->i_nlink
> 0) {
6607 * Failed to do the truncate, remove us from the in memory
6610 ret
= btrfs_orphan_del(NULL
, inode
);
6613 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6614 ret
= btrfs_update_inode(trans
, root
, inode
);
6618 nr
= trans
->blocks_used
;
6619 ret
= btrfs_end_transaction_throttle(trans
, root
);
6620 btrfs_btree_balance_dirty(root
, nr
);
6623 btrfs_free_block_rsv(root
, rsv
);
6632 * create a new subvolume directory/inode (helper for the ioctl).
6634 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6635 struct btrfs_root
*new_root
, u64 new_dirid
)
6637 struct inode
*inode
;
6641 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6642 new_dirid
, S_IFDIR
| 0700, &index
);
6644 return PTR_ERR(inode
);
6645 inode
->i_op
= &btrfs_dir_inode_operations
;
6646 inode
->i_fop
= &btrfs_dir_file_operations
;
6649 btrfs_i_size_write(inode
, 0);
6651 err
= btrfs_update_inode(trans
, new_root
, inode
);
6658 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6660 struct btrfs_inode
*ei
;
6661 struct inode
*inode
;
6663 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6668 ei
->space_info
= NULL
;
6672 ei
->last_sub_trans
= 0;
6673 ei
->logged_trans
= 0;
6674 ei
->delalloc_bytes
= 0;
6675 ei
->disk_i_size
= 0;
6678 ei
->index_cnt
= (u64
)-1;
6679 ei
->last_unlink_trans
= 0;
6681 spin_lock_init(&ei
->lock
);
6682 ei
->outstanding_extents
= 0;
6683 ei
->reserved_extents
= 0;
6685 ei
->ordered_data_close
= 0;
6686 ei
->orphan_meta_reserved
= 0;
6687 ei
->dummy_inode
= 0;
6689 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6691 ei
->delayed_node
= NULL
;
6693 inode
= &ei
->vfs_inode
;
6694 extent_map_tree_init(&ei
->extent_tree
);
6695 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
6696 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
6697 mutex_init(&ei
->log_mutex
);
6698 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6699 INIT_LIST_HEAD(&ei
->i_orphan
);
6700 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6701 INIT_LIST_HEAD(&ei
->ordered_operations
);
6702 RB_CLEAR_NODE(&ei
->rb_node
);
6707 static void btrfs_i_callback(struct rcu_head
*head
)
6709 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6710 INIT_LIST_HEAD(&inode
->i_dentry
);
6711 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6714 void btrfs_destroy_inode(struct inode
*inode
)
6716 struct btrfs_ordered_extent
*ordered
;
6717 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6719 WARN_ON(!list_empty(&inode
->i_dentry
));
6720 WARN_ON(inode
->i_data
.nrpages
);
6721 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
6722 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
6723 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
6724 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
6727 * This can happen where we create an inode, but somebody else also
6728 * created the same inode and we need to destroy the one we already
6735 * Make sure we're properly removed from the ordered operation
6739 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6740 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6741 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6742 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6745 spin_lock(&root
->orphan_lock
);
6746 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6747 printk(KERN_INFO
"BTRFS: inode %llu still on the orphan list\n",
6748 (unsigned long long)btrfs_ino(inode
));
6749 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6751 spin_unlock(&root
->orphan_lock
);
6754 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6758 printk(KERN_ERR
"btrfs found ordered "
6759 "extent %llu %llu on inode cleanup\n",
6760 (unsigned long long)ordered
->file_offset
,
6761 (unsigned long long)ordered
->len
);
6762 btrfs_remove_ordered_extent(inode
, ordered
);
6763 btrfs_put_ordered_extent(ordered
);
6764 btrfs_put_ordered_extent(ordered
);
6767 inode_tree_del(inode
);
6768 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6770 btrfs_remove_delayed_node(inode
);
6771 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6774 int btrfs_drop_inode(struct inode
*inode
)
6776 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6778 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6779 !btrfs_is_free_space_inode(root
, inode
))
6782 return generic_drop_inode(inode
);
6785 static void init_once(void *foo
)
6787 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6789 inode_init_once(&ei
->vfs_inode
);
6792 void btrfs_destroy_cachep(void)
6794 if (btrfs_inode_cachep
)
6795 kmem_cache_destroy(btrfs_inode_cachep
);
6796 if (btrfs_trans_handle_cachep
)
6797 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6798 if (btrfs_transaction_cachep
)
6799 kmem_cache_destroy(btrfs_transaction_cachep
);
6800 if (btrfs_path_cachep
)
6801 kmem_cache_destroy(btrfs_path_cachep
);
6802 if (btrfs_free_space_cachep
)
6803 kmem_cache_destroy(btrfs_free_space_cachep
);
6806 int btrfs_init_cachep(void)
6808 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6809 sizeof(struct btrfs_inode
), 0,
6810 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6811 if (!btrfs_inode_cachep
)
6814 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6815 sizeof(struct btrfs_trans_handle
), 0,
6816 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6817 if (!btrfs_trans_handle_cachep
)
6820 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6821 sizeof(struct btrfs_transaction
), 0,
6822 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6823 if (!btrfs_transaction_cachep
)
6826 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6827 sizeof(struct btrfs_path
), 0,
6828 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6829 if (!btrfs_path_cachep
)
6832 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6833 sizeof(struct btrfs_free_space
), 0,
6834 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6835 if (!btrfs_free_space_cachep
)
6840 btrfs_destroy_cachep();
6844 static int btrfs_getattr(struct vfsmount
*mnt
,
6845 struct dentry
*dentry
, struct kstat
*stat
)
6847 struct inode
*inode
= dentry
->d_inode
;
6848 generic_fillattr(inode
, stat
);
6849 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
6850 stat
->blksize
= PAGE_CACHE_SIZE
;
6851 stat
->blocks
= (inode_get_bytes(inode
) +
6852 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
6857 * If a file is moved, it will inherit the cow and compression flags of the new
6860 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
6862 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
6863 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
6865 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
6866 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
6868 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
6870 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
)
6871 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
6873 b_inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6876 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6877 struct inode
*new_dir
, struct dentry
*new_dentry
)
6879 struct btrfs_trans_handle
*trans
;
6880 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6881 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6882 struct inode
*new_inode
= new_dentry
->d_inode
;
6883 struct inode
*old_inode
= old_dentry
->d_inode
;
6884 struct timespec ctime
= CURRENT_TIME
;
6888 u64 old_ino
= btrfs_ino(old_inode
);
6890 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
6893 /* we only allow rename subvolume link between subvolumes */
6894 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
6897 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
6898 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
6901 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
6902 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
6905 * we're using rename to replace one file with another.
6906 * and the replacement file is large. Start IO on it now so
6907 * we don't add too much work to the end of the transaction
6909 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
6910 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
6911 filemap_flush(old_inode
->i_mapping
);
6913 /* close the racy window with snapshot create/destroy ioctl */
6914 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
6915 down_read(&root
->fs_info
->subvol_sem
);
6917 * We want to reserve the absolute worst case amount of items. So if
6918 * both inodes are subvols and we need to unlink them then that would
6919 * require 4 item modifications, but if they are both normal inodes it
6920 * would require 5 item modifications, so we'll assume their normal
6921 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6922 * should cover the worst case number of items we'll modify.
6924 trans
= btrfs_start_transaction(root
, 20);
6925 if (IS_ERR(trans
)) {
6926 ret
= PTR_ERR(trans
);
6931 btrfs_record_root_in_trans(trans
, dest
);
6933 ret
= btrfs_set_inode_index(new_dir
, &index
);
6937 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6938 /* force full log commit if subvolume involved. */
6939 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
6941 ret
= btrfs_insert_inode_ref(trans
, dest
,
6942 new_dentry
->d_name
.name
,
6943 new_dentry
->d_name
.len
,
6945 btrfs_ino(new_dir
), index
);
6949 * this is an ugly little race, but the rename is required
6950 * to make sure that if we crash, the inode is either at the
6951 * old name or the new one. pinning the log transaction lets
6952 * us make sure we don't allow a log commit to come in after
6953 * we unlink the name but before we add the new name back in.
6955 btrfs_pin_log_trans(root
);
6958 * make sure the inode gets flushed if it is replacing
6961 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
6962 btrfs_add_ordered_operation(trans
, root
, old_inode
);
6964 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
6965 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
6966 old_inode
->i_ctime
= ctime
;
6968 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
6969 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
6971 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6972 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
6973 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
6974 old_dentry
->d_name
.name
,
6975 old_dentry
->d_name
.len
);
6977 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
6978 old_dentry
->d_inode
,
6979 old_dentry
->d_name
.name
,
6980 old_dentry
->d_name
.len
);
6982 ret
= btrfs_update_inode(trans
, root
, old_inode
);
6987 new_inode
->i_ctime
= CURRENT_TIME
;
6988 if (unlikely(btrfs_ino(new_inode
) ==
6989 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
6990 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
6991 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
6993 new_dentry
->d_name
.name
,
6994 new_dentry
->d_name
.len
);
6995 BUG_ON(new_inode
->i_nlink
== 0);
6997 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
6998 new_dentry
->d_inode
,
6999 new_dentry
->d_name
.name
,
7000 new_dentry
->d_name
.len
);
7003 if (new_inode
->i_nlink
== 0) {
7004 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7009 fixup_inode_flags(new_dir
, old_inode
);
7011 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7012 new_dentry
->d_name
.name
,
7013 new_dentry
->d_name
.len
, 0, index
);
7016 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7017 struct dentry
*parent
= new_dentry
->d_parent
;
7018 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7019 btrfs_end_log_trans(root
);
7022 btrfs_end_transaction_throttle(trans
, root
);
7024 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7025 up_read(&root
->fs_info
->subvol_sem
);
7031 * some fairly slow code that needs optimization. This walks the list
7032 * of all the inodes with pending delalloc and forces them to disk.
7034 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7036 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7037 struct btrfs_inode
*binode
;
7038 struct inode
*inode
;
7040 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7043 spin_lock(&root
->fs_info
->delalloc_lock
);
7044 while (!list_empty(head
)) {
7045 binode
= list_entry(head
->next
, struct btrfs_inode
,
7047 inode
= igrab(&binode
->vfs_inode
);
7049 list_del_init(&binode
->delalloc_inodes
);
7050 spin_unlock(&root
->fs_info
->delalloc_lock
);
7052 filemap_flush(inode
->i_mapping
);
7054 btrfs_add_delayed_iput(inode
);
7059 spin_lock(&root
->fs_info
->delalloc_lock
);
7061 spin_unlock(&root
->fs_info
->delalloc_lock
);
7063 /* the filemap_flush will queue IO into the worker threads, but
7064 * we have to make sure the IO is actually started and that
7065 * ordered extents get created before we return
7067 atomic_inc(&root
->fs_info
->async_submit_draining
);
7068 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7069 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7070 wait_event(root
->fs_info
->async_submit_wait
,
7071 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7072 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7074 atomic_dec(&root
->fs_info
->async_submit_draining
);
7078 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7079 const char *symname
)
7081 struct btrfs_trans_handle
*trans
;
7082 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7083 struct btrfs_path
*path
;
7084 struct btrfs_key key
;
7085 struct inode
*inode
= NULL
;
7093 struct btrfs_file_extent_item
*ei
;
7094 struct extent_buffer
*leaf
;
7095 unsigned long nr
= 0;
7097 name_len
= strlen(symname
) + 1;
7098 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7099 return -ENAMETOOLONG
;
7102 * 2 items for inode item and ref
7103 * 2 items for dir items
7104 * 1 item for xattr if selinux is on
7106 trans
= btrfs_start_transaction(root
, 5);
7108 return PTR_ERR(trans
);
7110 err
= btrfs_find_free_ino(root
, &objectid
);
7114 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7115 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
7116 S_IFLNK
|S_IRWXUGO
, &index
);
7117 if (IS_ERR(inode
)) {
7118 err
= PTR_ERR(inode
);
7122 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7128 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7132 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7133 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7134 inode
->i_fop
= &btrfs_file_operations
;
7135 inode
->i_op
= &btrfs_file_inode_operations
;
7136 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7141 path
= btrfs_alloc_path();
7147 key
.objectid
= btrfs_ino(inode
);
7149 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7150 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7151 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7155 btrfs_free_path(path
);
7158 leaf
= path
->nodes
[0];
7159 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7160 struct btrfs_file_extent_item
);
7161 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7162 btrfs_set_file_extent_type(leaf
, ei
,
7163 BTRFS_FILE_EXTENT_INLINE
);
7164 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7165 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7166 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7167 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7169 ptr
= btrfs_file_extent_inline_start(ei
);
7170 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7171 btrfs_mark_buffer_dirty(leaf
);
7172 btrfs_free_path(path
);
7174 inode
->i_op
= &btrfs_symlink_inode_operations
;
7175 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7176 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7177 inode_set_bytes(inode
, name_len
);
7178 btrfs_i_size_write(inode
, name_len
- 1);
7179 err
= btrfs_update_inode(trans
, root
, inode
);
7184 nr
= trans
->blocks_used
;
7185 btrfs_end_transaction_throttle(trans
, root
);
7187 inode_dec_link_count(inode
);
7190 btrfs_btree_balance_dirty(root
, nr
);
7194 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7195 u64 start
, u64 num_bytes
, u64 min_size
,
7196 loff_t actual_len
, u64
*alloc_hint
,
7197 struct btrfs_trans_handle
*trans
)
7199 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7200 struct btrfs_key ins
;
7201 u64 cur_offset
= start
;
7204 bool own_trans
= true;
7208 while (num_bytes
> 0) {
7210 trans
= btrfs_start_transaction(root
, 3);
7211 if (IS_ERR(trans
)) {
7212 ret
= PTR_ERR(trans
);
7217 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7218 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7221 btrfs_end_transaction(trans
, root
);
7225 ret
= insert_reserved_file_extent(trans
, inode
,
7226 cur_offset
, ins
.objectid
,
7227 ins
.offset
, ins
.offset
,
7228 ins
.offset
, 0, 0, 0,
7229 BTRFS_FILE_EXTENT_PREALLOC
);
7231 btrfs_drop_extent_cache(inode
, cur_offset
,
7232 cur_offset
+ ins
.offset
-1, 0);
7234 num_bytes
-= ins
.offset
;
7235 cur_offset
+= ins
.offset
;
7236 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7238 inode
->i_ctime
= CURRENT_TIME
;
7239 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7240 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7241 (actual_len
> inode
->i_size
) &&
7242 (cur_offset
> inode
->i_size
)) {
7243 if (cur_offset
> actual_len
)
7244 i_size
= actual_len
;
7246 i_size
= cur_offset
;
7247 i_size_write(inode
, i_size
);
7248 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7251 ret
= btrfs_update_inode(trans
, root
, inode
);
7255 btrfs_end_transaction(trans
, root
);
7260 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7261 u64 start
, u64 num_bytes
, u64 min_size
,
7262 loff_t actual_len
, u64
*alloc_hint
)
7264 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7265 min_size
, actual_len
, alloc_hint
,
7269 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7270 struct btrfs_trans_handle
*trans
, int mode
,
7271 u64 start
, u64 num_bytes
, u64 min_size
,
7272 loff_t actual_len
, u64
*alloc_hint
)
7274 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7275 min_size
, actual_len
, alloc_hint
, trans
);
7278 static int btrfs_set_page_dirty(struct page
*page
)
7280 return __set_page_dirty_nobuffers(page
);
7283 static int btrfs_permission(struct inode
*inode
, int mask
)
7285 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7286 umode_t mode
= inode
->i_mode
;
7288 if (mask
& MAY_WRITE
&&
7289 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
7290 if (btrfs_root_readonly(root
))
7292 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
7295 return generic_permission(inode
, mask
);
7298 static const struct inode_operations btrfs_dir_inode_operations
= {
7299 .getattr
= btrfs_getattr
,
7300 .lookup
= btrfs_lookup
,
7301 .create
= btrfs_create
,
7302 .unlink
= btrfs_unlink
,
7304 .mkdir
= btrfs_mkdir
,
7305 .rmdir
= btrfs_rmdir
,
7306 .rename
= btrfs_rename
,
7307 .symlink
= btrfs_symlink
,
7308 .setattr
= btrfs_setattr
,
7309 .mknod
= btrfs_mknod
,
7310 .setxattr
= btrfs_setxattr
,
7311 .getxattr
= btrfs_getxattr
,
7312 .listxattr
= btrfs_listxattr
,
7313 .removexattr
= btrfs_removexattr
,
7314 .permission
= btrfs_permission
,
7315 .get_acl
= btrfs_get_acl
,
7317 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7318 .lookup
= btrfs_lookup
,
7319 .permission
= btrfs_permission
,
7320 .get_acl
= btrfs_get_acl
,
7323 static const struct file_operations btrfs_dir_file_operations
= {
7324 .llseek
= generic_file_llseek
,
7325 .read
= generic_read_dir
,
7326 .readdir
= btrfs_real_readdir
,
7327 .unlocked_ioctl
= btrfs_ioctl
,
7328 #ifdef CONFIG_COMPAT
7329 .compat_ioctl
= btrfs_ioctl
,
7331 .release
= btrfs_release_file
,
7332 .fsync
= btrfs_sync_file
,
7335 static struct extent_io_ops btrfs_extent_io_ops
= {
7336 .fill_delalloc
= run_delalloc_range
,
7337 .submit_bio_hook
= btrfs_submit_bio_hook
,
7338 .merge_bio_hook
= btrfs_merge_bio_hook
,
7339 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7340 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7341 .writepage_start_hook
= btrfs_writepage_start_hook
,
7342 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
7343 .set_bit_hook
= btrfs_set_bit_hook
,
7344 .clear_bit_hook
= btrfs_clear_bit_hook
,
7345 .merge_extent_hook
= btrfs_merge_extent_hook
,
7346 .split_extent_hook
= btrfs_split_extent_hook
,
7350 * btrfs doesn't support the bmap operation because swapfiles
7351 * use bmap to make a mapping of extents in the file. They assume
7352 * these extents won't change over the life of the file and they
7353 * use the bmap result to do IO directly to the drive.
7355 * the btrfs bmap call would return logical addresses that aren't
7356 * suitable for IO and they also will change frequently as COW
7357 * operations happen. So, swapfile + btrfs == corruption.
7359 * For now we're avoiding this by dropping bmap.
7361 static const struct address_space_operations btrfs_aops
= {
7362 .readpage
= btrfs_readpage
,
7363 .writepage
= btrfs_writepage
,
7364 .writepages
= btrfs_writepages
,
7365 .readpages
= btrfs_readpages
,
7366 .direct_IO
= btrfs_direct_IO
,
7367 .invalidatepage
= btrfs_invalidatepage
,
7368 .releasepage
= btrfs_releasepage
,
7369 .set_page_dirty
= btrfs_set_page_dirty
,
7370 .error_remove_page
= generic_error_remove_page
,
7373 static const struct address_space_operations btrfs_symlink_aops
= {
7374 .readpage
= btrfs_readpage
,
7375 .writepage
= btrfs_writepage
,
7376 .invalidatepage
= btrfs_invalidatepage
,
7377 .releasepage
= btrfs_releasepage
,
7380 static const struct inode_operations btrfs_file_inode_operations
= {
7381 .getattr
= btrfs_getattr
,
7382 .setattr
= btrfs_setattr
,
7383 .setxattr
= btrfs_setxattr
,
7384 .getxattr
= btrfs_getxattr
,
7385 .listxattr
= btrfs_listxattr
,
7386 .removexattr
= btrfs_removexattr
,
7387 .permission
= btrfs_permission
,
7388 .fiemap
= btrfs_fiemap
,
7389 .get_acl
= btrfs_get_acl
,
7391 static const struct inode_operations btrfs_special_inode_operations
= {
7392 .getattr
= btrfs_getattr
,
7393 .setattr
= btrfs_setattr
,
7394 .permission
= btrfs_permission
,
7395 .setxattr
= btrfs_setxattr
,
7396 .getxattr
= btrfs_getxattr
,
7397 .listxattr
= btrfs_listxattr
,
7398 .removexattr
= btrfs_removexattr
,
7399 .get_acl
= btrfs_get_acl
,
7401 static const struct inode_operations btrfs_symlink_inode_operations
= {
7402 .readlink
= generic_readlink
,
7403 .follow_link
= page_follow_link_light
,
7404 .put_link
= page_put_link
,
7405 .getattr
= btrfs_getattr
,
7406 .permission
= btrfs_permission
,
7407 .setxattr
= btrfs_setxattr
,
7408 .getxattr
= btrfs_getxattr
,
7409 .listxattr
= btrfs_listxattr
,
7410 .removexattr
= btrfs_removexattr
,
7411 .get_acl
= btrfs_get_acl
,
7414 const struct dentry_operations btrfs_dentry_operations
= {
7415 .d_delete
= btrfs_dentry_delete
,
7416 .d_release
= btrfs_dentry_release
,