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();
1066 nolock
= btrfs_is_free_space_inode(root
, inode
);
1069 trans
= btrfs_join_transaction_nolock(root
);
1071 trans
= btrfs_join_transaction(root
);
1073 BUG_ON(IS_ERR(trans
));
1074 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1076 cow_start
= (u64
)-1;
1079 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1082 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1083 leaf
= path
->nodes
[0];
1084 btrfs_item_key_to_cpu(leaf
, &found_key
,
1085 path
->slots
[0] - 1);
1086 if (found_key
.objectid
== ino
&&
1087 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1092 leaf
= path
->nodes
[0];
1093 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1094 ret
= btrfs_next_leaf(root
, path
);
1099 leaf
= path
->nodes
[0];
1105 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1107 if (found_key
.objectid
> ino
||
1108 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1109 found_key
.offset
> end
)
1112 if (found_key
.offset
> cur_offset
) {
1113 extent_end
= found_key
.offset
;
1118 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1119 struct btrfs_file_extent_item
);
1120 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1122 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1123 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1124 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1125 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1126 extent_end
= found_key
.offset
+
1127 btrfs_file_extent_num_bytes(leaf
, fi
);
1128 if (extent_end
<= start
) {
1132 if (disk_bytenr
== 0)
1134 if (btrfs_file_extent_compression(leaf
, fi
) ||
1135 btrfs_file_extent_encryption(leaf
, fi
) ||
1136 btrfs_file_extent_other_encoding(leaf
, fi
))
1138 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1140 if (btrfs_extent_readonly(root
, disk_bytenr
))
1142 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1144 extent_offset
, disk_bytenr
))
1146 disk_bytenr
+= extent_offset
;
1147 disk_bytenr
+= cur_offset
- found_key
.offset
;
1148 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1150 * force cow if csum exists in the range.
1151 * this ensure that csum for a given extent are
1152 * either valid or do not exist.
1154 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1157 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1158 extent_end
= found_key
.offset
+
1159 btrfs_file_extent_inline_len(leaf
, fi
);
1160 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1165 if (extent_end
<= start
) {
1170 if (cow_start
== (u64
)-1)
1171 cow_start
= cur_offset
;
1172 cur_offset
= extent_end
;
1173 if (cur_offset
> end
)
1179 btrfs_release_path(path
);
1180 if (cow_start
!= (u64
)-1) {
1181 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1182 found_key
.offset
- 1, page_started
,
1185 cow_start
= (u64
)-1;
1188 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1189 struct extent_map
*em
;
1190 struct extent_map_tree
*em_tree
;
1191 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1192 em
= alloc_extent_map();
1194 em
->start
= cur_offset
;
1195 em
->orig_start
= em
->start
;
1196 em
->len
= num_bytes
;
1197 em
->block_len
= num_bytes
;
1198 em
->block_start
= disk_bytenr
;
1199 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1200 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1202 write_lock(&em_tree
->lock
);
1203 ret
= add_extent_mapping(em_tree
, em
);
1204 write_unlock(&em_tree
->lock
);
1205 if (ret
!= -EEXIST
) {
1206 free_extent_map(em
);
1209 btrfs_drop_extent_cache(inode
, em
->start
,
1210 em
->start
+ em
->len
- 1, 0);
1212 type
= BTRFS_ORDERED_PREALLOC
;
1214 type
= BTRFS_ORDERED_NOCOW
;
1217 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1218 num_bytes
, num_bytes
, type
);
1221 if (root
->root_key
.objectid
==
1222 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1223 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1228 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1229 cur_offset
, cur_offset
+ num_bytes
- 1,
1230 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1231 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1232 EXTENT_SET_PRIVATE2
);
1233 cur_offset
= extent_end
;
1234 if (cur_offset
> end
)
1237 btrfs_release_path(path
);
1239 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1240 cow_start
= cur_offset
;
1241 if (cow_start
!= (u64
)-1) {
1242 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1243 page_started
, nr_written
, 1);
1248 ret
= btrfs_end_transaction_nolock(trans
, root
);
1251 ret
= btrfs_end_transaction(trans
, root
);
1254 btrfs_free_path(path
);
1259 * extent_io.c call back to do delayed allocation processing
1261 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1262 u64 start
, u64 end
, int *page_started
,
1263 unsigned long *nr_written
)
1266 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1268 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1269 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1270 page_started
, 1, nr_written
);
1271 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1272 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1273 page_started
, 0, nr_written
);
1274 else if (!btrfs_test_opt(root
, COMPRESS
) &&
1275 !(BTRFS_I(inode
)->force_compress
) &&
1276 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))
1277 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1278 page_started
, nr_written
, 1);
1280 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1281 page_started
, nr_written
);
1285 static int btrfs_split_extent_hook(struct inode
*inode
,
1286 struct extent_state
*orig
, u64 split
)
1288 /* not delalloc, ignore it */
1289 if (!(orig
->state
& EXTENT_DELALLOC
))
1292 spin_lock(&BTRFS_I(inode
)->lock
);
1293 BTRFS_I(inode
)->outstanding_extents
++;
1294 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 int 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
);
1319 * extent_io.c set_bit_hook, used to track delayed allocation
1320 * bytes in this file, and to maintain the list of inodes that
1321 * have pending delalloc work to be done.
1323 static int btrfs_set_bit_hook(struct inode
*inode
,
1324 struct extent_state
*state
, int *bits
)
1328 * set_bit and clear bit hooks normally require _irqsave/restore
1329 * but in this case, we are only testing for the DELALLOC
1330 * bit, which is only set or cleared with irqs on
1332 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1333 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1334 u64 len
= state
->end
+ 1 - state
->start
;
1335 bool do_list
= !btrfs_is_free_space_inode(root
, inode
);
1337 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1338 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1340 spin_lock(&BTRFS_I(inode
)->lock
);
1341 BTRFS_I(inode
)->outstanding_extents
++;
1342 spin_unlock(&BTRFS_I(inode
)->lock
);
1345 spin_lock(&root
->fs_info
->delalloc_lock
);
1346 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1347 root
->fs_info
->delalloc_bytes
+= len
;
1348 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1349 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1350 &root
->fs_info
->delalloc_inodes
);
1352 spin_unlock(&root
->fs_info
->delalloc_lock
);
1358 * extent_io.c clear_bit_hook, see set_bit_hook for why
1360 static int btrfs_clear_bit_hook(struct inode
*inode
,
1361 struct extent_state
*state
, int *bits
)
1364 * set_bit and clear bit hooks normally require _irqsave/restore
1365 * but in this case, we are only testing for the DELALLOC
1366 * bit, which is only set or cleared with irqs on
1368 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1369 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1370 u64 len
= state
->end
+ 1 - state
->start
;
1371 bool do_list
= !btrfs_is_free_space_inode(root
, inode
);
1373 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1374 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1375 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1376 spin_lock(&BTRFS_I(inode
)->lock
);
1377 BTRFS_I(inode
)->outstanding_extents
--;
1378 spin_unlock(&BTRFS_I(inode
)->lock
);
1381 if (*bits
& EXTENT_DO_ACCOUNTING
)
1382 btrfs_delalloc_release_metadata(inode
, len
);
1384 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1386 btrfs_free_reserved_data_space(inode
, len
);
1388 spin_lock(&root
->fs_info
->delalloc_lock
);
1389 root
->fs_info
->delalloc_bytes
-= len
;
1390 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1392 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1393 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1394 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1396 spin_unlock(&root
->fs_info
->delalloc_lock
);
1402 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1403 * we don't create bios that span stripes or chunks
1405 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1406 size_t size
, struct bio
*bio
,
1407 unsigned long bio_flags
)
1409 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1410 struct btrfs_mapping_tree
*map_tree
;
1411 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1416 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1419 length
= bio
->bi_size
;
1420 map_tree
= &root
->fs_info
->mapping_tree
;
1421 map_length
= length
;
1422 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1423 &map_length
, NULL
, 0);
1425 if (map_length
< length
+ size
)
1431 * in order to insert checksums into the metadata in large chunks,
1432 * we wait until bio submission time. All the pages in the bio are
1433 * checksummed and sums are attached onto the ordered extent record.
1435 * At IO completion time the cums attached on the ordered extent record
1436 * are inserted into the btree
1438 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1439 struct bio
*bio
, int mirror_num
,
1440 unsigned long bio_flags
,
1443 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1446 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1452 * in order to insert checksums into the metadata in large chunks,
1453 * we wait until bio submission time. All the pages in the bio are
1454 * checksummed and sums are attached onto the ordered extent record.
1456 * At IO completion time the cums attached on the ordered extent record
1457 * are inserted into the btree
1459 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1460 int mirror_num
, unsigned long bio_flags
,
1463 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1464 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1468 * extent_io.c submission hook. This does the right thing for csum calculation
1469 * on write, or reading the csums from the tree before a read
1471 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1472 int mirror_num
, unsigned long bio_flags
,
1475 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1479 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1481 if (btrfs_is_free_space_inode(root
, inode
))
1482 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 2);
1484 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1487 if (!(rw
& REQ_WRITE
)) {
1488 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1489 return btrfs_submit_compressed_read(inode
, bio
,
1490 mirror_num
, bio_flags
);
1491 } else if (!skip_sum
) {
1492 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1497 } else if (!skip_sum
) {
1498 /* csum items have already been cloned */
1499 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1501 /* we're doing a write, do the async checksumming */
1502 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1503 inode
, rw
, bio
, mirror_num
,
1504 bio_flags
, bio_offset
,
1505 __btrfs_submit_bio_start
,
1506 __btrfs_submit_bio_done
);
1510 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1514 * given a list of ordered sums record them in the inode. This happens
1515 * at IO completion time based on sums calculated at bio submission time.
1517 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1518 struct inode
*inode
, u64 file_offset
,
1519 struct list_head
*list
)
1521 struct btrfs_ordered_sum
*sum
;
1523 list_for_each_entry(sum
, list
, list
) {
1524 btrfs_csum_file_blocks(trans
,
1525 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1530 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1531 struct extent_state
**cached_state
)
1533 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1535 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1536 cached_state
, GFP_NOFS
);
1539 /* see btrfs_writepage_start_hook for details on why this is required */
1540 struct btrfs_writepage_fixup
{
1542 struct btrfs_work work
;
1545 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1547 struct btrfs_writepage_fixup
*fixup
;
1548 struct btrfs_ordered_extent
*ordered
;
1549 struct extent_state
*cached_state
= NULL
;
1551 struct inode
*inode
;
1555 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1559 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1560 ClearPageChecked(page
);
1564 inode
= page
->mapping
->host
;
1565 page_start
= page_offset(page
);
1566 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1568 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1569 &cached_state
, GFP_NOFS
);
1571 /* already ordered? We're done */
1572 if (PagePrivate2(page
))
1575 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1577 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1578 page_end
, &cached_state
, GFP_NOFS
);
1580 btrfs_start_ordered_extent(inode
, ordered
, 1);
1585 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1586 ClearPageChecked(page
);
1588 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1589 &cached_state
, GFP_NOFS
);
1592 page_cache_release(page
);
1597 * There are a few paths in the higher layers of the kernel that directly
1598 * set the page dirty bit without asking the filesystem if it is a
1599 * good idea. This causes problems because we want to make sure COW
1600 * properly happens and the data=ordered rules are followed.
1602 * In our case any range that doesn't have the ORDERED bit set
1603 * hasn't been properly setup for IO. We kick off an async process
1604 * to fix it up. The async helper will wait for ordered extents, set
1605 * the delalloc bit and make it safe to write the page.
1607 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1609 struct inode
*inode
= page
->mapping
->host
;
1610 struct btrfs_writepage_fixup
*fixup
;
1611 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1613 /* this page is properly in the ordered list */
1614 if (TestClearPagePrivate2(page
))
1617 if (PageChecked(page
))
1620 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1624 SetPageChecked(page
);
1625 page_cache_get(page
);
1626 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1628 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1632 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1633 struct inode
*inode
, u64 file_pos
,
1634 u64 disk_bytenr
, u64 disk_num_bytes
,
1635 u64 num_bytes
, u64 ram_bytes
,
1636 u8 compression
, u8 encryption
,
1637 u16 other_encoding
, int extent_type
)
1639 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1640 struct btrfs_file_extent_item
*fi
;
1641 struct btrfs_path
*path
;
1642 struct extent_buffer
*leaf
;
1643 struct btrfs_key ins
;
1647 path
= btrfs_alloc_path();
1650 path
->leave_spinning
= 1;
1653 * we may be replacing one extent in the tree with another.
1654 * The new extent is pinned in the extent map, and we don't want
1655 * to drop it from the cache until it is completely in the btree.
1657 * So, tell btrfs_drop_extents to leave this extent in the cache.
1658 * the caller is expected to unpin it and allow it to be merged
1661 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1665 ins
.objectid
= btrfs_ino(inode
);
1666 ins
.offset
= file_pos
;
1667 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1668 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1670 leaf
= path
->nodes
[0];
1671 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1672 struct btrfs_file_extent_item
);
1673 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1674 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1675 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1676 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1677 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1678 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1679 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1680 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1681 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1682 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1684 btrfs_unlock_up_safe(path
, 1);
1685 btrfs_set_lock_blocking(leaf
);
1687 btrfs_mark_buffer_dirty(leaf
);
1689 inode_add_bytes(inode
, num_bytes
);
1691 ins
.objectid
= disk_bytenr
;
1692 ins
.offset
= disk_num_bytes
;
1693 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1694 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1695 root
->root_key
.objectid
,
1696 btrfs_ino(inode
), file_pos
, &ins
);
1698 btrfs_free_path(path
);
1704 * helper function for btrfs_finish_ordered_io, this
1705 * just reads in some of the csum leaves to prime them into ram
1706 * before we start the transaction. It limits the amount of btree
1707 * reads required while inside the transaction.
1709 /* as ordered data IO finishes, this gets called so we can finish
1710 * an ordered extent if the range of bytes in the file it covers are
1713 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1715 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1716 struct btrfs_trans_handle
*trans
= NULL
;
1717 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1718 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1719 struct extent_state
*cached_state
= NULL
;
1720 int compress_type
= 0;
1724 ret
= btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1728 BUG_ON(!ordered_extent
);
1730 nolock
= btrfs_is_free_space_inode(root
, inode
);
1732 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1733 BUG_ON(!list_empty(&ordered_extent
->list
));
1734 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1737 trans
= btrfs_join_transaction_nolock(root
);
1739 trans
= btrfs_join_transaction(root
);
1740 BUG_ON(IS_ERR(trans
));
1741 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1742 ret
= btrfs_update_inode(trans
, root
, inode
);
1748 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1749 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1750 0, &cached_state
, GFP_NOFS
);
1753 trans
= btrfs_join_transaction_nolock(root
);
1755 trans
= btrfs_join_transaction(root
);
1756 BUG_ON(IS_ERR(trans
));
1757 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1759 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1760 compress_type
= ordered_extent
->compress_type
;
1761 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1762 BUG_ON(compress_type
);
1763 ret
= btrfs_mark_extent_written(trans
, inode
,
1764 ordered_extent
->file_offset
,
1765 ordered_extent
->file_offset
+
1766 ordered_extent
->len
);
1769 BUG_ON(root
== root
->fs_info
->tree_root
);
1770 ret
= insert_reserved_file_extent(trans
, inode
,
1771 ordered_extent
->file_offset
,
1772 ordered_extent
->start
,
1773 ordered_extent
->disk_len
,
1774 ordered_extent
->len
,
1775 ordered_extent
->len
,
1776 compress_type
, 0, 0,
1777 BTRFS_FILE_EXTENT_REG
);
1778 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1779 ordered_extent
->file_offset
,
1780 ordered_extent
->len
);
1783 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1784 ordered_extent
->file_offset
+
1785 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1787 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1788 &ordered_extent
->list
);
1790 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1792 ret
= btrfs_update_inode(trans
, root
, inode
);
1799 btrfs_end_transaction_nolock(trans
, root
);
1801 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1803 btrfs_end_transaction(trans
, root
);
1807 btrfs_put_ordered_extent(ordered_extent
);
1808 /* once for the tree */
1809 btrfs_put_ordered_extent(ordered_extent
);
1814 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1815 struct extent_state
*state
, int uptodate
)
1817 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1819 ClearPagePrivate2(page
);
1820 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1824 * When IO fails, either with EIO or csum verification fails, we
1825 * try other mirrors that might have a good copy of the data. This
1826 * io_failure_record is used to record state as we go through all the
1827 * mirrors. If another mirror has good data, the page is set up to date
1828 * and things continue. If a good mirror can't be found, the original
1829 * bio end_io callback is called to indicate things have failed.
1831 struct io_failure_record
{
1836 unsigned long bio_flags
;
1840 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1841 struct page
*page
, u64 start
, u64 end
,
1842 struct extent_state
*state
)
1844 struct io_failure_record
*failrec
= NULL
;
1846 struct extent_map
*em
;
1847 struct inode
*inode
= page
->mapping
->host
;
1848 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1849 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1856 ret
= get_state_private(failure_tree
, start
, &private);
1858 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1861 failrec
->start
= start
;
1862 failrec
->len
= end
- start
+ 1;
1863 failrec
->last_mirror
= 0;
1864 failrec
->bio_flags
= 0;
1866 read_lock(&em_tree
->lock
);
1867 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1868 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1869 free_extent_map(em
);
1872 read_unlock(&em_tree
->lock
);
1874 if (IS_ERR_OR_NULL(em
)) {
1878 logical
= start
- em
->start
;
1879 logical
= em
->block_start
+ logical
;
1880 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1881 logical
= em
->block_start
;
1882 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1883 extent_set_compress_type(&failrec
->bio_flags
,
1886 failrec
->logical
= logical
;
1887 free_extent_map(em
);
1888 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1889 EXTENT_DIRTY
, GFP_NOFS
);
1890 set_state_private(failure_tree
, start
,
1891 (u64
)(unsigned long)failrec
);
1893 failrec
= (struct io_failure_record
*)(unsigned long)private;
1895 num_copies
= btrfs_num_copies(
1896 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1897 failrec
->logical
, failrec
->len
);
1898 failrec
->last_mirror
++;
1900 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1901 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1904 if (state
&& state
->start
!= failrec
->start
)
1906 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1908 if (!state
|| failrec
->last_mirror
> num_copies
) {
1909 set_state_private(failure_tree
, failrec
->start
, 0);
1910 clear_extent_bits(failure_tree
, failrec
->start
,
1911 failrec
->start
+ failrec
->len
- 1,
1912 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1916 bio
= bio_alloc(GFP_NOFS
, 1);
1917 bio
->bi_private
= state
;
1918 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1919 bio
->bi_sector
= failrec
->logical
>> 9;
1920 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1923 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1924 if (failed_bio
->bi_rw
& REQ_WRITE
)
1929 ret
= BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1930 failrec
->last_mirror
,
1931 failrec
->bio_flags
, 0);
1936 * each time an IO finishes, we do a fast check in the IO failure tree
1937 * to see if we need to process or clean up an io_failure_record
1939 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1942 u64 private_failure
;
1943 struct io_failure_record
*failure
;
1947 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1948 (u64
)-1, 1, EXTENT_DIRTY
, 0)) {
1949 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1950 start
, &private_failure
);
1952 failure
= (struct io_failure_record
*)(unsigned long)
1954 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1956 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1958 failure
->start
+ failure
->len
- 1,
1959 EXTENT_DIRTY
| EXTENT_LOCKED
,
1968 * when reads are done, we need to check csums to verify the data is correct
1969 * if there's a match, we allow the bio to finish. If not, we go through
1970 * the io_failure_record routines to find good copies
1972 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1973 struct extent_state
*state
)
1975 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1976 struct inode
*inode
= page
->mapping
->host
;
1977 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1979 u64
private = ~(u32
)0;
1981 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1984 if (PageChecked(page
)) {
1985 ClearPageChecked(page
);
1989 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1992 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1993 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1994 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1999 if (state
&& state
->start
== start
) {
2000 private = state
->private;
2003 ret
= get_state_private(io_tree
, start
, &private);
2005 kaddr
= kmap_atomic(page
, KM_USER0
);
2009 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
2010 btrfs_csum_final(csum
, (char *)&csum
);
2011 if (csum
!= private)
2014 kunmap_atomic(kaddr
, KM_USER0
);
2016 /* if the io failure tree for this inode is non-empty,
2017 * check to see if we've recovered from a failed IO
2019 btrfs_clean_io_failures(inode
, start
);
2023 printk_ratelimited(KERN_INFO
"btrfs csum failed ino %llu off %llu csum %u "
2025 (unsigned long long)btrfs_ino(page
->mapping
->host
),
2026 (unsigned long long)start
, csum
,
2027 (unsigned long long)private);
2028 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2029 flush_dcache_page(page
);
2030 kunmap_atomic(kaddr
, KM_USER0
);
2036 struct delayed_iput
{
2037 struct list_head list
;
2038 struct inode
*inode
;
2041 void btrfs_add_delayed_iput(struct inode
*inode
)
2043 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2044 struct delayed_iput
*delayed
;
2046 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2049 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2050 delayed
->inode
= inode
;
2052 spin_lock(&fs_info
->delayed_iput_lock
);
2053 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2054 spin_unlock(&fs_info
->delayed_iput_lock
);
2057 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2060 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2061 struct delayed_iput
*delayed
;
2064 spin_lock(&fs_info
->delayed_iput_lock
);
2065 empty
= list_empty(&fs_info
->delayed_iputs
);
2066 spin_unlock(&fs_info
->delayed_iput_lock
);
2070 down_read(&root
->fs_info
->cleanup_work_sem
);
2071 spin_lock(&fs_info
->delayed_iput_lock
);
2072 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2073 spin_unlock(&fs_info
->delayed_iput_lock
);
2075 while (!list_empty(&list
)) {
2076 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2077 list_del(&delayed
->list
);
2078 iput(delayed
->inode
);
2081 up_read(&root
->fs_info
->cleanup_work_sem
);
2085 * calculate extra metadata reservation when snapshotting a subvolume
2086 * contains orphan files.
2088 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle
*trans
,
2089 struct btrfs_pending_snapshot
*pending
,
2090 u64
*bytes_to_reserve
)
2092 struct btrfs_root
*root
;
2093 struct btrfs_block_rsv
*block_rsv
;
2097 root
= pending
->root
;
2098 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2101 block_rsv
= root
->orphan_block_rsv
;
2103 /* orphan block reservation for the snapshot */
2104 num_bytes
= block_rsv
->size
;
2107 * after the snapshot is created, COWing tree blocks may use more
2108 * space than it frees. So we should make sure there is enough
2111 index
= trans
->transid
& 0x1;
2112 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2113 num_bytes
+= block_rsv
->size
-
2114 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2117 *bytes_to_reserve
+= num_bytes
;
2120 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle
*trans
,
2121 struct btrfs_pending_snapshot
*pending
)
2123 struct btrfs_root
*root
= pending
->root
;
2124 struct btrfs_root
*snap
= pending
->snap
;
2125 struct btrfs_block_rsv
*block_rsv
;
2130 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2133 /* refill source subvolume's orphan block reservation */
2134 block_rsv
= root
->orphan_block_rsv
;
2135 index
= trans
->transid
& 0x1;
2136 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2137 num_bytes
= block_rsv
->size
-
2138 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2139 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2140 root
->orphan_block_rsv
,
2145 /* setup orphan block reservation for the snapshot */
2146 block_rsv
= btrfs_alloc_block_rsv(snap
);
2149 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2150 snap
->orphan_block_rsv
= block_rsv
;
2152 num_bytes
= root
->orphan_block_rsv
->size
;
2153 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2154 block_rsv
, num_bytes
);
2158 /* insert orphan item for the snapshot */
2159 WARN_ON(!root
->orphan_item_inserted
);
2160 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2161 snap
->root_key
.objectid
);
2163 snap
->orphan_item_inserted
= 1;
2167 enum btrfs_orphan_cleanup_state
{
2168 ORPHAN_CLEANUP_STARTED
= 1,
2169 ORPHAN_CLEANUP_DONE
= 2,
2173 * This is called in transaction commmit time. If there are no orphan
2174 * files in the subvolume, it removes orphan item and frees block_rsv
2177 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2178 struct btrfs_root
*root
)
2182 if (!list_empty(&root
->orphan_list
) ||
2183 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2186 if (root
->orphan_item_inserted
&&
2187 btrfs_root_refs(&root
->root_item
) > 0) {
2188 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2189 root
->root_key
.objectid
);
2191 root
->orphan_item_inserted
= 0;
2194 if (root
->orphan_block_rsv
) {
2195 WARN_ON(root
->orphan_block_rsv
->size
> 0);
2196 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
2197 root
->orphan_block_rsv
= NULL
;
2202 * This creates an orphan entry for the given inode in case something goes
2203 * wrong in the middle of an unlink/truncate.
2205 * NOTE: caller of this function should reserve 5 units of metadata for
2208 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2210 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2211 struct btrfs_block_rsv
*block_rsv
= NULL
;
2216 if (!root
->orphan_block_rsv
) {
2217 block_rsv
= btrfs_alloc_block_rsv(root
);
2221 spin_lock(&root
->orphan_lock
);
2222 if (!root
->orphan_block_rsv
) {
2223 root
->orphan_block_rsv
= block_rsv
;
2224 } else if (block_rsv
) {
2225 btrfs_free_block_rsv(root
, block_rsv
);
2229 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2230 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2233 * For proper ENOSPC handling, we should do orphan
2234 * cleanup when mounting. But this introduces backward
2235 * compatibility issue.
2237 if (!xchg(&root
->orphan_item_inserted
, 1))
2245 if (!BTRFS_I(inode
)->orphan_meta_reserved
) {
2246 BTRFS_I(inode
)->orphan_meta_reserved
= 1;
2249 spin_unlock(&root
->orphan_lock
);
2252 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2254 /* grab metadata reservation from transaction handle */
2256 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2260 /* insert an orphan item to track this unlinked/truncated file */
2262 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
2266 /* insert an orphan item to track subvolume contains orphan files */
2268 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2269 root
->root_key
.objectid
);
2276 * We have done the truncate/delete so we can go ahead and remove the orphan
2277 * item for this particular inode.
2279 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2281 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2282 int delete_item
= 0;
2283 int release_rsv
= 0;
2286 spin_lock(&root
->orphan_lock
);
2287 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2288 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2292 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2293 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2296 spin_unlock(&root
->orphan_lock
);
2298 if (trans
&& delete_item
) {
2299 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
2304 btrfs_orphan_release_metadata(inode
);
2310 * this cleans up any orphans that may be left on the list from the last use
2313 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2315 struct btrfs_path
*path
;
2316 struct extent_buffer
*leaf
;
2317 struct btrfs_key key
, found_key
;
2318 struct btrfs_trans_handle
*trans
;
2319 struct inode
*inode
;
2320 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2322 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2325 path
= btrfs_alloc_path();
2332 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2333 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2334 key
.offset
= (u64
)-1;
2337 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2342 * if ret == 0 means we found what we were searching for, which
2343 * is weird, but possible, so only screw with path if we didn't
2344 * find the key and see if we have stuff that matches
2348 if (path
->slots
[0] == 0)
2353 /* pull out the item */
2354 leaf
= path
->nodes
[0];
2355 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2357 /* make sure the item matches what we want */
2358 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2360 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2363 /* release the path since we're done with it */
2364 btrfs_release_path(path
);
2367 * this is where we are basically btrfs_lookup, without the
2368 * crossing root thing. we store the inode number in the
2369 * offset of the orphan item.
2371 found_key
.objectid
= found_key
.offset
;
2372 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2373 found_key
.offset
= 0;
2374 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2375 if (IS_ERR(inode
)) {
2376 ret
= PTR_ERR(inode
);
2381 * add this inode to the orphan list so btrfs_orphan_del does
2382 * the proper thing when we hit it
2384 spin_lock(&root
->orphan_lock
);
2385 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2386 spin_unlock(&root
->orphan_lock
);
2389 * if this is a bad inode, means we actually succeeded in
2390 * removing the inode, but not the orphan record, which means
2391 * we need to manually delete the orphan since iput will just
2392 * do a destroy_inode
2394 if (is_bad_inode(inode
)) {
2395 trans
= btrfs_start_transaction(root
, 0);
2396 if (IS_ERR(trans
)) {
2397 ret
= PTR_ERR(trans
);
2400 btrfs_orphan_del(trans
, inode
);
2401 btrfs_end_transaction(trans
, root
);
2406 /* if we have links, this was a truncate, lets do that */
2407 if (inode
->i_nlink
) {
2408 if (!S_ISREG(inode
->i_mode
)) {
2414 ret
= btrfs_truncate(inode
);
2419 /* this will do delete_inode and everything for us */
2424 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2426 if (root
->orphan_block_rsv
)
2427 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2430 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2431 trans
= btrfs_join_transaction(root
);
2433 btrfs_end_transaction(trans
, root
);
2437 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2439 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2443 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2444 btrfs_free_path(path
);
2449 * very simple check to peek ahead in the leaf looking for xattrs. If we
2450 * don't find any xattrs, we know there can't be any acls.
2452 * slot is the slot the inode is in, objectid is the objectid of the inode
2454 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2455 int slot
, u64 objectid
)
2457 u32 nritems
= btrfs_header_nritems(leaf
);
2458 struct btrfs_key found_key
;
2462 while (slot
< nritems
) {
2463 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2465 /* we found a different objectid, there must not be acls */
2466 if (found_key
.objectid
!= objectid
)
2469 /* we found an xattr, assume we've got an acl */
2470 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2474 * we found a key greater than an xattr key, there can't
2475 * be any acls later on
2477 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2484 * it goes inode, inode backrefs, xattrs, extents,
2485 * so if there are a ton of hard links to an inode there can
2486 * be a lot of backrefs. Don't waste time searching too hard,
2487 * this is just an optimization
2492 /* we hit the end of the leaf before we found an xattr or
2493 * something larger than an xattr. We have to assume the inode
2500 * read an inode from the btree into the in-memory inode
2502 static void btrfs_read_locked_inode(struct inode
*inode
)
2504 struct btrfs_path
*path
;
2505 struct extent_buffer
*leaf
;
2506 struct btrfs_inode_item
*inode_item
;
2507 struct btrfs_timespec
*tspec
;
2508 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2509 struct btrfs_key location
;
2513 bool filled
= false;
2515 ret
= btrfs_fill_inode(inode
, &rdev
);
2519 path
= btrfs_alloc_path();
2521 path
->leave_spinning
= 1;
2522 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2524 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2528 leaf
= path
->nodes
[0];
2533 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2534 struct btrfs_inode_item
);
2535 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2536 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2537 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2538 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2539 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2541 tspec
= btrfs_inode_atime(inode_item
);
2542 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2543 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2545 tspec
= btrfs_inode_mtime(inode_item
);
2546 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2547 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2549 tspec
= btrfs_inode_ctime(inode_item
);
2550 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2551 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2553 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2554 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2555 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2556 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2558 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2560 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2561 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2564 * try to precache a NULL acl entry for files that don't have
2565 * any xattrs or acls
2567 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
2570 cache_no_acl(inode
);
2572 btrfs_free_path(path
);
2574 switch (inode
->i_mode
& S_IFMT
) {
2576 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2577 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2578 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2579 inode
->i_fop
= &btrfs_file_operations
;
2580 inode
->i_op
= &btrfs_file_inode_operations
;
2583 inode
->i_fop
= &btrfs_dir_file_operations
;
2584 if (root
== root
->fs_info
->tree_root
)
2585 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2587 inode
->i_op
= &btrfs_dir_inode_operations
;
2590 inode
->i_op
= &btrfs_symlink_inode_operations
;
2591 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2592 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2595 inode
->i_op
= &btrfs_special_inode_operations
;
2596 init_special_inode(inode
, inode
->i_mode
, rdev
);
2600 btrfs_update_iflags(inode
);
2604 btrfs_free_path(path
);
2605 make_bad_inode(inode
);
2609 * given a leaf and an inode, copy the inode fields into the leaf
2611 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2612 struct extent_buffer
*leaf
,
2613 struct btrfs_inode_item
*item
,
2614 struct inode
*inode
)
2616 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2617 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2618 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2619 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2620 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2622 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2623 inode
->i_atime
.tv_sec
);
2624 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2625 inode
->i_atime
.tv_nsec
);
2627 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2628 inode
->i_mtime
.tv_sec
);
2629 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2630 inode
->i_mtime
.tv_nsec
);
2632 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2633 inode
->i_ctime
.tv_sec
);
2634 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2635 inode
->i_ctime
.tv_nsec
);
2637 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2638 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2639 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2640 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2641 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2642 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2643 btrfs_set_inode_block_group(leaf
, item
, 0);
2647 * copy everything in the in-memory inode into the btree.
2649 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2650 struct btrfs_root
*root
, struct inode
*inode
)
2652 struct btrfs_inode_item
*inode_item
;
2653 struct btrfs_path
*path
;
2654 struct extent_buffer
*leaf
;
2658 * If the inode is a free space inode, we can deadlock during commit
2659 * if we put it into the delayed code.
2661 * The data relocation inode should also be directly updated
2664 if (!btrfs_is_free_space_inode(root
, inode
)
2665 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
2666 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
2668 btrfs_set_inode_last_trans(trans
, inode
);
2672 path
= btrfs_alloc_path();
2676 path
->leave_spinning
= 1;
2677 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
2685 btrfs_unlock_up_safe(path
, 1);
2686 leaf
= path
->nodes
[0];
2687 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2688 struct btrfs_inode_item
);
2690 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2691 btrfs_mark_buffer_dirty(leaf
);
2692 btrfs_set_inode_last_trans(trans
, inode
);
2695 btrfs_free_path(path
);
2700 * unlink helper that gets used here in inode.c and in the tree logging
2701 * recovery code. It remove a link in a directory with a given name, and
2702 * also drops the back refs in the inode to the directory
2704 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2705 struct btrfs_root
*root
,
2706 struct inode
*dir
, struct inode
*inode
,
2707 const char *name
, int name_len
)
2709 struct btrfs_path
*path
;
2711 struct extent_buffer
*leaf
;
2712 struct btrfs_dir_item
*di
;
2713 struct btrfs_key key
;
2715 u64 ino
= btrfs_ino(inode
);
2716 u64 dir_ino
= btrfs_ino(dir
);
2718 path
= btrfs_alloc_path();
2724 path
->leave_spinning
= 1;
2725 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2726 name
, name_len
, -1);
2735 leaf
= path
->nodes
[0];
2736 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2737 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2740 btrfs_release_path(path
);
2742 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
2745 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2746 "inode %llu parent %llu\n", name_len
, name
,
2747 (unsigned long long)ino
, (unsigned long long)dir_ino
);
2751 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2755 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2757 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2759 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2764 btrfs_free_path(path
);
2768 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2769 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2770 btrfs_update_inode(trans
, root
, dir
);
2775 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2776 struct btrfs_root
*root
,
2777 struct inode
*dir
, struct inode
*inode
,
2778 const char *name
, int name_len
)
2781 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2783 btrfs_drop_nlink(inode
);
2784 ret
= btrfs_update_inode(trans
, root
, inode
);
2790 /* helper to check if there is any shared block in the path */
2791 static int check_path_shared(struct btrfs_root
*root
,
2792 struct btrfs_path
*path
)
2794 struct extent_buffer
*eb
;
2798 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2801 if (!path
->nodes
[level
])
2803 eb
= path
->nodes
[level
];
2804 if (!btrfs_block_can_be_shared(root
, eb
))
2806 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2815 * helper to start transaction for unlink and rmdir.
2817 * unlink and rmdir are special in btrfs, they do not always free space.
2818 * so in enospc case, we should make sure they will free space before
2819 * allowing them to use the global metadata reservation.
2821 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2822 struct dentry
*dentry
)
2824 struct btrfs_trans_handle
*trans
;
2825 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2826 struct btrfs_path
*path
;
2827 struct btrfs_inode_ref
*ref
;
2828 struct btrfs_dir_item
*di
;
2829 struct inode
*inode
= dentry
->d_inode
;
2834 u64 ino
= btrfs_ino(inode
);
2835 u64 dir_ino
= btrfs_ino(dir
);
2837 trans
= btrfs_start_transaction(root
, 10);
2838 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2841 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2842 return ERR_PTR(-ENOSPC
);
2844 /* check if there is someone else holds reference */
2845 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2846 return ERR_PTR(-ENOSPC
);
2848 if (atomic_read(&inode
->i_count
) > 2)
2849 return ERR_PTR(-ENOSPC
);
2851 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2852 return ERR_PTR(-ENOSPC
);
2854 path
= btrfs_alloc_path();
2856 root
->fs_info
->enospc_unlink
= 0;
2857 return ERR_PTR(-ENOMEM
);
2860 trans
= btrfs_start_transaction(root
, 0);
2861 if (IS_ERR(trans
)) {
2862 btrfs_free_path(path
);
2863 root
->fs_info
->enospc_unlink
= 0;
2867 path
->skip_locking
= 1;
2868 path
->search_commit_root
= 1;
2870 ret
= btrfs_lookup_inode(trans
, root
, path
,
2871 &BTRFS_I(dir
)->location
, 0);
2877 if (check_path_shared(root
, path
))
2882 btrfs_release_path(path
);
2884 ret
= btrfs_lookup_inode(trans
, root
, path
,
2885 &BTRFS_I(inode
)->location
, 0);
2891 if (check_path_shared(root
, path
))
2896 btrfs_release_path(path
);
2898 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2899 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2906 if (check_path_shared(root
, path
))
2908 btrfs_release_path(path
);
2916 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2917 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2923 if (check_path_shared(root
, path
))
2929 btrfs_release_path(path
);
2931 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2932 dentry
->d_name
.name
, dentry
->d_name
.len
,
2939 if (check_path_shared(root
, path
))
2941 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2942 btrfs_release_path(path
);
2945 * This is a commit root search, if we can lookup inode item and other
2946 * relative items in the commit root, it means the transaction of
2947 * dir/file creation has been committed, and the dir index item that we
2948 * delay to insert has also been inserted into the commit root. So
2949 * we needn't worry about the delayed insertion of the dir index item
2952 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
2953 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2958 BUG_ON(ret
== -ENOENT
);
2959 if (check_path_shared(root
, path
))
2964 btrfs_free_path(path
);
2966 btrfs_end_transaction(trans
, root
);
2967 root
->fs_info
->enospc_unlink
= 0;
2968 return ERR_PTR(err
);
2971 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2975 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2976 struct btrfs_root
*root
)
2978 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2979 BUG_ON(!root
->fs_info
->enospc_unlink
);
2980 root
->fs_info
->enospc_unlink
= 0;
2982 btrfs_end_transaction_throttle(trans
, root
);
2985 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2987 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2988 struct btrfs_trans_handle
*trans
;
2989 struct inode
*inode
= dentry
->d_inode
;
2991 unsigned long nr
= 0;
2993 trans
= __unlink_start_trans(dir
, dentry
);
2995 return PTR_ERR(trans
);
2997 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2999 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3000 dentry
->d_name
.name
, dentry
->d_name
.len
);
3003 if (inode
->i_nlink
== 0) {
3004 ret
= btrfs_orphan_add(trans
, inode
);
3008 nr
= trans
->blocks_used
;
3009 __unlink_end_trans(trans
, root
);
3010 btrfs_btree_balance_dirty(root
, nr
);
3014 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
3015 struct btrfs_root
*root
,
3016 struct inode
*dir
, u64 objectid
,
3017 const char *name
, int name_len
)
3019 struct btrfs_path
*path
;
3020 struct extent_buffer
*leaf
;
3021 struct btrfs_dir_item
*di
;
3022 struct btrfs_key key
;
3025 u64 dir_ino
= btrfs_ino(dir
);
3027 path
= btrfs_alloc_path();
3031 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3032 name
, name_len
, -1);
3033 BUG_ON(IS_ERR_OR_NULL(di
));
3035 leaf
= path
->nodes
[0];
3036 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3037 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3038 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3040 btrfs_release_path(path
);
3042 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3043 objectid
, root
->root_key
.objectid
,
3044 dir_ino
, &index
, name
, name_len
);
3046 BUG_ON(ret
!= -ENOENT
);
3047 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
3049 BUG_ON(IS_ERR_OR_NULL(di
));
3051 leaf
= path
->nodes
[0];
3052 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3053 btrfs_release_path(path
);
3056 btrfs_release_path(path
);
3058 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3061 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3062 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3063 ret
= btrfs_update_inode(trans
, root
, dir
);
3066 btrfs_free_path(path
);
3070 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3072 struct inode
*inode
= dentry
->d_inode
;
3074 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3075 struct btrfs_trans_handle
*trans
;
3076 unsigned long nr
= 0;
3078 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3079 btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3082 trans
= __unlink_start_trans(dir
, dentry
);
3084 return PTR_ERR(trans
);
3086 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3087 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3088 BTRFS_I(inode
)->location
.objectid
,
3089 dentry
->d_name
.name
,
3090 dentry
->d_name
.len
);
3094 err
= btrfs_orphan_add(trans
, inode
);
3098 /* now the directory is empty */
3099 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3100 dentry
->d_name
.name
, dentry
->d_name
.len
);
3102 btrfs_i_size_write(inode
, 0);
3104 nr
= trans
->blocks_used
;
3105 __unlink_end_trans(trans
, root
);
3106 btrfs_btree_balance_dirty(root
, nr
);
3112 * this can truncate away extent items, csum items and directory items.
3113 * It starts at a high offset and removes keys until it can't find
3114 * any higher than new_size
3116 * csum items that cross the new i_size are truncated to the new size
3119 * min_type is the minimum key type to truncate down to. If set to 0, this
3120 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3122 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3123 struct btrfs_root
*root
,
3124 struct inode
*inode
,
3125 u64 new_size
, u32 min_type
)
3127 struct btrfs_path
*path
;
3128 struct extent_buffer
*leaf
;
3129 struct btrfs_file_extent_item
*fi
;
3130 struct btrfs_key key
;
3131 struct btrfs_key found_key
;
3132 u64 extent_start
= 0;
3133 u64 extent_num_bytes
= 0;
3134 u64 extent_offset
= 0;
3136 u64 mask
= root
->sectorsize
- 1;
3137 u32 found_type
= (u8
)-1;
3140 int pending_del_nr
= 0;
3141 int pending_del_slot
= 0;
3142 int extent_type
= -1;
3146 u64 ino
= btrfs_ino(inode
);
3148 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3150 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3151 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3154 * This function is also used to drop the items in the log tree before
3155 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3156 * it is used to drop the loged items. So we shouldn't kill the delayed
3159 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3160 btrfs_kill_delayed_inode_items(inode
);
3162 path
= btrfs_alloc_path();
3167 key
.offset
= (u64
)-1;
3171 path
->leave_spinning
= 1;
3172 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3179 /* there are no items in the tree for us to truncate, we're
3182 if (path
->slots
[0] == 0)
3189 leaf
= path
->nodes
[0];
3190 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3191 found_type
= btrfs_key_type(&found_key
);
3194 if (found_key
.objectid
!= ino
)
3197 if (found_type
< min_type
)
3200 item_end
= found_key
.offset
;
3201 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3202 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3203 struct btrfs_file_extent_item
);
3204 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3205 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3206 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3207 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3209 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3211 btrfs_file_extent_num_bytes(leaf
, fi
);
3212 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3213 item_end
+= btrfs_file_extent_inline_len(leaf
,
3218 if (found_type
> min_type
) {
3221 if (item_end
< new_size
)
3223 if (found_key
.offset
>= new_size
)
3229 /* FIXME, shrink the extent if the ref count is only 1 */
3230 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3233 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3235 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3236 if (!del_item
&& !encoding
) {
3237 u64 orig_num_bytes
=
3238 btrfs_file_extent_num_bytes(leaf
, fi
);
3239 extent_num_bytes
= new_size
-
3240 found_key
.offset
+ root
->sectorsize
- 1;
3241 extent_num_bytes
= extent_num_bytes
&
3242 ~((u64
)root
->sectorsize
- 1);
3243 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3245 num_dec
= (orig_num_bytes
-
3247 if (root
->ref_cows
&& extent_start
!= 0)
3248 inode_sub_bytes(inode
, num_dec
);
3249 btrfs_mark_buffer_dirty(leaf
);
3252 btrfs_file_extent_disk_num_bytes(leaf
,
3254 extent_offset
= found_key
.offset
-
3255 btrfs_file_extent_offset(leaf
, fi
);
3257 /* FIXME blocksize != 4096 */
3258 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3259 if (extent_start
!= 0) {
3262 inode_sub_bytes(inode
, num_dec
);
3265 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3267 * we can't truncate inline items that have had
3271 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3272 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3273 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3274 u32 size
= new_size
- found_key
.offset
;
3276 if (root
->ref_cows
) {
3277 inode_sub_bytes(inode
, item_end
+ 1 -
3281 btrfs_file_extent_calc_inline_size(size
);
3282 ret
= btrfs_truncate_item(trans
, root
, path
,
3284 } else if (root
->ref_cows
) {
3285 inode_sub_bytes(inode
, item_end
+ 1 -
3291 if (!pending_del_nr
) {
3292 /* no pending yet, add ourselves */
3293 pending_del_slot
= path
->slots
[0];
3295 } else if (pending_del_nr
&&
3296 path
->slots
[0] + 1 == pending_del_slot
) {
3297 /* hop on the pending chunk */
3299 pending_del_slot
= path
->slots
[0];
3306 if (found_extent
&& (root
->ref_cows
||
3307 root
== root
->fs_info
->tree_root
)) {
3308 btrfs_set_path_blocking(path
);
3309 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3310 extent_num_bytes
, 0,
3311 btrfs_header_owner(leaf
),
3312 ino
, extent_offset
);
3316 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3319 if (path
->slots
[0] == 0 ||
3320 path
->slots
[0] != pending_del_slot
) {
3321 if (root
->ref_cows
&&
3322 BTRFS_I(inode
)->location
.objectid
!=
3323 BTRFS_FREE_INO_OBJECTID
) {
3327 if (pending_del_nr
) {
3328 ret
= btrfs_del_items(trans
, root
, path
,
3334 btrfs_release_path(path
);
3341 if (pending_del_nr
) {
3342 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3346 btrfs_free_path(path
);
3351 * taken from block_truncate_page, but does cow as it zeros out
3352 * any bytes left in the last page in the file.
3354 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3356 struct inode
*inode
= mapping
->host
;
3357 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3358 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3359 struct btrfs_ordered_extent
*ordered
;
3360 struct extent_state
*cached_state
= NULL
;
3362 u32 blocksize
= root
->sectorsize
;
3363 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3364 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3370 if ((offset
& (blocksize
- 1)) == 0)
3372 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3378 page
= find_or_create_page(mapping
, index
, GFP_NOFS
);
3380 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3384 page_start
= page_offset(page
);
3385 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3387 if (!PageUptodate(page
)) {
3388 ret
= btrfs_readpage(NULL
, page
);
3390 if (page
->mapping
!= mapping
) {
3392 page_cache_release(page
);
3395 if (!PageUptodate(page
)) {
3400 wait_on_page_writeback(page
);
3402 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3404 set_page_extent_mapped(page
);
3406 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3408 unlock_extent_cached(io_tree
, page_start
, page_end
,
3409 &cached_state
, GFP_NOFS
);
3411 page_cache_release(page
);
3412 btrfs_start_ordered_extent(inode
, ordered
, 1);
3413 btrfs_put_ordered_extent(ordered
);
3417 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3418 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3419 0, 0, &cached_state
, GFP_NOFS
);
3421 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3424 unlock_extent_cached(io_tree
, page_start
, page_end
,
3425 &cached_state
, GFP_NOFS
);
3430 if (offset
!= PAGE_CACHE_SIZE
) {
3432 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3433 flush_dcache_page(page
);
3436 ClearPageChecked(page
);
3437 set_page_dirty(page
);
3438 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3443 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3445 page_cache_release(page
);
3451 * This function puts in dummy file extents for the area we're creating a hole
3452 * for. So if we are truncating this file to a larger size we need to insert
3453 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3454 * the range between oldsize and size
3456 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3458 struct btrfs_trans_handle
*trans
;
3459 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3460 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3461 struct extent_map
*em
= NULL
;
3462 struct extent_state
*cached_state
= NULL
;
3463 u64 mask
= root
->sectorsize
- 1;
3464 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3465 u64 block_end
= (size
+ mask
) & ~mask
;
3471 if (size
<= hole_start
)
3475 struct btrfs_ordered_extent
*ordered
;
3476 btrfs_wait_ordered_range(inode
, hole_start
,
3477 block_end
- hole_start
);
3478 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3479 &cached_state
, GFP_NOFS
);
3480 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3483 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3484 &cached_state
, GFP_NOFS
);
3485 btrfs_put_ordered_extent(ordered
);
3488 cur_offset
= hole_start
;
3490 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3491 block_end
- cur_offset
, 0);
3492 BUG_ON(IS_ERR_OR_NULL(em
));
3493 last_byte
= min(extent_map_end(em
), block_end
);
3494 last_byte
= (last_byte
+ mask
) & ~mask
;
3495 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3497 hole_size
= last_byte
- cur_offset
;
3499 trans
= btrfs_start_transaction(root
, 2);
3500 if (IS_ERR(trans
)) {
3501 err
= PTR_ERR(trans
);
3505 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3506 cur_offset
+ hole_size
,
3511 err
= btrfs_insert_file_extent(trans
, root
,
3512 btrfs_ino(inode
), cur_offset
, 0,
3513 0, hole_size
, 0, hole_size
,
3518 btrfs_drop_extent_cache(inode
, hole_start
,
3521 btrfs_end_transaction(trans
, root
);
3523 free_extent_map(em
);
3525 cur_offset
= last_byte
;
3526 if (cur_offset
>= block_end
)
3530 free_extent_map(em
);
3531 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3536 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3538 loff_t oldsize
= i_size_read(inode
);
3541 if (newsize
== oldsize
)
3544 if (newsize
> oldsize
) {
3545 i_size_write(inode
, newsize
);
3546 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3547 truncate_pagecache(inode
, oldsize
, newsize
);
3548 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3550 btrfs_setsize(inode
, oldsize
);
3554 mark_inode_dirty(inode
);
3558 * We're truncating a file that used to have good data down to
3559 * zero. Make sure it gets into the ordered flush list so that
3560 * any new writes get down to disk quickly.
3563 BTRFS_I(inode
)->ordered_data_close
= 1;
3565 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3566 truncate_setsize(inode
, newsize
);
3567 ret
= btrfs_truncate(inode
);
3573 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3575 struct inode
*inode
= dentry
->d_inode
;
3576 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3579 if (btrfs_root_readonly(root
))
3582 err
= inode_change_ok(inode
, attr
);
3586 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3587 err
= btrfs_setsize(inode
, attr
->ia_size
);
3592 if (attr
->ia_valid
) {
3593 setattr_copy(inode
, attr
);
3594 mark_inode_dirty(inode
);
3596 if (attr
->ia_valid
& ATTR_MODE
)
3597 err
= btrfs_acl_chmod(inode
);
3603 void btrfs_evict_inode(struct inode
*inode
)
3605 struct btrfs_trans_handle
*trans
;
3606 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3610 trace_btrfs_inode_evict(inode
);
3612 truncate_inode_pages(&inode
->i_data
, 0);
3613 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3614 btrfs_is_free_space_inode(root
, inode
)))
3617 if (is_bad_inode(inode
)) {
3618 btrfs_orphan_del(NULL
, inode
);
3621 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3622 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3624 if (root
->fs_info
->log_root_recovering
) {
3625 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3629 if (inode
->i_nlink
> 0) {
3630 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3634 btrfs_i_size_write(inode
, 0);
3637 trans
= btrfs_join_transaction(root
);
3638 BUG_ON(IS_ERR(trans
));
3639 trans
->block_rsv
= root
->orphan_block_rsv
;
3641 ret
= btrfs_block_rsv_check(trans
, root
,
3642 root
->orphan_block_rsv
, 0, 5);
3644 BUG_ON(ret
!= -EAGAIN
);
3645 ret
= btrfs_commit_transaction(trans
, root
);
3650 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3654 nr
= trans
->blocks_used
;
3655 btrfs_end_transaction(trans
, root
);
3657 btrfs_btree_balance_dirty(root
, nr
);
3662 ret
= btrfs_orphan_del(trans
, inode
);
3666 if (!(root
== root
->fs_info
->tree_root
||
3667 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
3668 btrfs_return_ino(root
, btrfs_ino(inode
));
3670 nr
= trans
->blocks_used
;
3671 btrfs_end_transaction(trans
, root
);
3672 btrfs_btree_balance_dirty(root
, nr
);
3674 end_writeback(inode
);
3679 * this returns the key found in the dir entry in the location pointer.
3680 * If no dir entries were found, location->objectid is 0.
3682 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3683 struct btrfs_key
*location
)
3685 const char *name
= dentry
->d_name
.name
;
3686 int namelen
= dentry
->d_name
.len
;
3687 struct btrfs_dir_item
*di
;
3688 struct btrfs_path
*path
;
3689 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3692 path
= btrfs_alloc_path();
3695 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
3700 if (IS_ERR_OR_NULL(di
))
3703 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3705 btrfs_free_path(path
);
3708 location
->objectid
= 0;
3713 * when we hit a tree root in a directory, the btrfs part of the inode
3714 * needs to be changed to reflect the root directory of the tree root. This
3715 * is kind of like crossing a mount point.
3717 static int fixup_tree_root_location(struct btrfs_root
*root
,
3719 struct dentry
*dentry
,
3720 struct btrfs_key
*location
,
3721 struct btrfs_root
**sub_root
)
3723 struct btrfs_path
*path
;
3724 struct btrfs_root
*new_root
;
3725 struct btrfs_root_ref
*ref
;
3726 struct extent_buffer
*leaf
;
3730 path
= btrfs_alloc_path();
3737 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3738 BTRFS_I(dir
)->root
->root_key
.objectid
,
3739 location
->objectid
);
3746 leaf
= path
->nodes
[0];
3747 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3748 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
3749 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3752 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3753 (unsigned long)(ref
+ 1),
3754 dentry
->d_name
.len
);
3758 btrfs_release_path(path
);
3760 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3761 if (IS_ERR(new_root
)) {
3762 err
= PTR_ERR(new_root
);
3766 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3771 *sub_root
= new_root
;
3772 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3773 location
->type
= BTRFS_INODE_ITEM_KEY
;
3774 location
->offset
= 0;
3777 btrfs_free_path(path
);
3781 static void inode_tree_add(struct inode
*inode
)
3783 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3784 struct btrfs_inode
*entry
;
3786 struct rb_node
*parent
;
3787 u64 ino
= btrfs_ino(inode
);
3789 p
= &root
->inode_tree
.rb_node
;
3792 if (inode_unhashed(inode
))
3795 spin_lock(&root
->inode_lock
);
3798 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3800 if (ino
< btrfs_ino(&entry
->vfs_inode
))
3801 p
= &parent
->rb_left
;
3802 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
3803 p
= &parent
->rb_right
;
3805 WARN_ON(!(entry
->vfs_inode
.i_state
&
3806 (I_WILL_FREE
| I_FREEING
)));
3807 rb_erase(parent
, &root
->inode_tree
);
3808 RB_CLEAR_NODE(parent
);
3809 spin_unlock(&root
->inode_lock
);
3813 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3814 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3815 spin_unlock(&root
->inode_lock
);
3818 static void inode_tree_del(struct inode
*inode
)
3820 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3823 spin_lock(&root
->inode_lock
);
3824 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3825 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3826 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3827 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3829 spin_unlock(&root
->inode_lock
);
3832 * Free space cache has inodes in the tree root, but the tree root has a
3833 * root_refs of 0, so this could end up dropping the tree root as a
3834 * snapshot, so we need the extra !root->fs_info->tree_root check to
3835 * make sure we don't drop it.
3837 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3838 root
!= root
->fs_info
->tree_root
) {
3839 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3840 spin_lock(&root
->inode_lock
);
3841 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3842 spin_unlock(&root
->inode_lock
);
3844 btrfs_add_dead_root(root
);
3848 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3850 struct rb_node
*node
;
3851 struct rb_node
*prev
;
3852 struct btrfs_inode
*entry
;
3853 struct inode
*inode
;
3856 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3858 spin_lock(&root
->inode_lock
);
3860 node
= root
->inode_tree
.rb_node
;
3864 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3866 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
3867 node
= node
->rb_left
;
3868 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
3869 node
= node
->rb_right
;
3875 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
3876 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
3880 prev
= rb_next(prev
);
3884 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3885 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
3886 inode
= igrab(&entry
->vfs_inode
);
3888 spin_unlock(&root
->inode_lock
);
3889 if (atomic_read(&inode
->i_count
) > 1)
3890 d_prune_aliases(inode
);
3892 * btrfs_drop_inode will have it removed from
3893 * the inode cache when its usage count
3898 spin_lock(&root
->inode_lock
);
3902 if (cond_resched_lock(&root
->inode_lock
))
3905 node
= rb_next(node
);
3907 spin_unlock(&root
->inode_lock
);
3911 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3913 struct btrfs_iget_args
*args
= p
;
3914 inode
->i_ino
= args
->ino
;
3915 BTRFS_I(inode
)->root
= args
->root
;
3916 btrfs_set_inode_space_info(args
->root
, inode
);
3920 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3922 struct btrfs_iget_args
*args
= opaque
;
3923 return args
->ino
== btrfs_ino(inode
) &&
3924 args
->root
== BTRFS_I(inode
)->root
;
3927 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
3929 struct btrfs_root
*root
)
3931 struct inode
*inode
;
3932 struct btrfs_iget_args args
;
3933 args
.ino
= objectid
;
3936 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3937 btrfs_init_locked_inode
,
3942 /* Get an inode object given its location and corresponding root.
3943 * Returns in *is_new if the inode was read from disk
3945 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3946 struct btrfs_root
*root
, int *new)
3948 struct inode
*inode
;
3950 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3952 return ERR_PTR(-ENOMEM
);
3954 if (inode
->i_state
& I_NEW
) {
3955 BTRFS_I(inode
)->root
= root
;
3956 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3957 btrfs_read_locked_inode(inode
);
3958 inode_tree_add(inode
);
3959 unlock_new_inode(inode
);
3967 static struct inode
*new_simple_dir(struct super_block
*s
,
3968 struct btrfs_key
*key
,
3969 struct btrfs_root
*root
)
3971 struct inode
*inode
= new_inode(s
);
3974 return ERR_PTR(-ENOMEM
);
3976 BTRFS_I(inode
)->root
= root
;
3977 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
3978 BTRFS_I(inode
)->dummy_inode
= 1;
3980 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
3981 inode
->i_op
= &simple_dir_inode_operations
;
3982 inode
->i_fop
= &simple_dir_operations
;
3983 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
3984 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3989 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3991 struct inode
*inode
;
3992 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3993 struct btrfs_root
*sub_root
= root
;
3994 struct btrfs_key location
;
3998 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3999 return ERR_PTR(-ENAMETOOLONG
);
4001 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4004 return ERR_PTR(ret
);
4006 if (location
.objectid
== 0)
4009 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4010 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4014 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4016 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4017 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4018 &location
, &sub_root
);
4021 inode
= ERR_PTR(ret
);
4023 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4025 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4027 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4029 if (!IS_ERR(inode
) && root
!= sub_root
) {
4030 down_read(&root
->fs_info
->cleanup_work_sem
);
4031 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4032 ret
= btrfs_orphan_cleanup(sub_root
);
4033 up_read(&root
->fs_info
->cleanup_work_sem
);
4035 inode
= ERR_PTR(ret
);
4041 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4043 struct btrfs_root
*root
;
4045 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
4046 dentry
= dentry
->d_parent
;
4048 if (dentry
->d_inode
) {
4049 root
= BTRFS_I(dentry
->d_inode
)->root
;
4050 if (btrfs_root_refs(&root
->root_item
) == 0)
4056 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4057 struct nameidata
*nd
)
4059 return d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
4062 unsigned char btrfs_filetype_table
[] = {
4063 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4066 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4069 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4070 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4071 struct btrfs_item
*item
;
4072 struct btrfs_dir_item
*di
;
4073 struct btrfs_key key
;
4074 struct btrfs_key found_key
;
4075 struct btrfs_path
*path
;
4076 struct list_head ins_list
;
4077 struct list_head del_list
;
4079 struct extent_buffer
*leaf
;
4081 unsigned char d_type
;
4086 int key_type
= BTRFS_DIR_INDEX_KEY
;
4090 int is_curr
= 0; /* filp->f_pos points to the current index? */
4092 /* FIXME, use a real flag for deciding about the key type */
4093 if (root
->fs_info
->tree_root
== root
)
4094 key_type
= BTRFS_DIR_ITEM_KEY
;
4096 /* special case for "." */
4097 if (filp
->f_pos
== 0) {
4098 over
= filldir(dirent
, ".", 1, 1, btrfs_ino(inode
), DT_DIR
);
4103 /* special case for .., just use the back ref */
4104 if (filp
->f_pos
== 1) {
4105 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4106 over
= filldir(dirent
, "..", 2,
4112 path
= btrfs_alloc_path();
4118 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4119 INIT_LIST_HEAD(&ins_list
);
4120 INIT_LIST_HEAD(&del_list
);
4121 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
4124 btrfs_set_key_type(&key
, key_type
);
4125 key
.offset
= filp
->f_pos
;
4126 key
.objectid
= btrfs_ino(inode
);
4128 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4133 leaf
= path
->nodes
[0];
4134 slot
= path
->slots
[0];
4135 if (slot
>= btrfs_header_nritems(leaf
)) {
4136 ret
= btrfs_next_leaf(root
, path
);
4144 item
= btrfs_item_nr(leaf
, slot
);
4145 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4147 if (found_key
.objectid
!= key
.objectid
)
4149 if (btrfs_key_type(&found_key
) != key_type
)
4151 if (found_key
.offset
< filp
->f_pos
)
4153 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
4154 btrfs_should_delete_dir_index(&del_list
,
4158 filp
->f_pos
= found_key
.offset
;
4161 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4163 di_total
= btrfs_item_size(leaf
, item
);
4165 while (di_cur
< di_total
) {
4166 struct btrfs_key location
;
4168 if (verify_dir_item(root
, leaf
, di
))
4171 name_len
= btrfs_dir_name_len(leaf
, di
);
4172 if (name_len
<= sizeof(tmp_name
)) {
4173 name_ptr
= tmp_name
;
4175 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4181 read_extent_buffer(leaf
, name_ptr
,
4182 (unsigned long)(di
+ 1), name_len
);
4184 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4185 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4187 /* is this a reference to our own snapshot? If so
4190 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4191 location
.objectid
== root
->root_key
.objectid
) {
4195 over
= filldir(dirent
, name_ptr
, name_len
,
4196 found_key
.offset
, location
.objectid
,
4200 if (name_ptr
!= tmp_name
)
4205 di_len
= btrfs_dir_name_len(leaf
, di
) +
4206 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4208 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4214 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4217 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
4223 /* Reached end of directory/root. Bump pos past the last item. */
4224 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4226 * 32-bit glibc will use getdents64, but then strtol -
4227 * so the last number we can serve is this.
4229 filp
->f_pos
= 0x7fffffff;
4235 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4236 btrfs_put_delayed_items(&ins_list
, &del_list
);
4237 btrfs_free_path(path
);
4241 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4243 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4244 struct btrfs_trans_handle
*trans
;
4246 bool nolock
= false;
4248 if (BTRFS_I(inode
)->dummy_inode
)
4251 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(root
, inode
))
4254 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4256 trans
= btrfs_join_transaction_nolock(root
);
4258 trans
= btrfs_join_transaction(root
);
4260 return PTR_ERR(trans
);
4262 ret
= btrfs_end_transaction_nolock(trans
, root
);
4264 ret
= btrfs_commit_transaction(trans
, root
);
4270 * This is somewhat expensive, updating the tree every time the
4271 * inode changes. But, it is most likely to find the inode in cache.
4272 * FIXME, needs more benchmarking...there are no reasons other than performance
4273 * to keep or drop this code.
4275 void btrfs_dirty_inode(struct inode
*inode
, int flags
)
4277 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4278 struct btrfs_trans_handle
*trans
;
4281 if (BTRFS_I(inode
)->dummy_inode
)
4284 trans
= btrfs_join_transaction(root
);
4285 BUG_ON(IS_ERR(trans
));
4287 ret
= btrfs_update_inode(trans
, root
, inode
);
4288 if (ret
&& ret
== -ENOSPC
) {
4289 /* whoops, lets try again with the full transaction */
4290 btrfs_end_transaction(trans
, root
);
4291 trans
= btrfs_start_transaction(root
, 1);
4292 if (IS_ERR(trans
)) {
4293 printk_ratelimited(KERN_ERR
"btrfs: fail to "
4294 "dirty inode %llu error %ld\n",
4295 (unsigned long long)btrfs_ino(inode
),
4300 ret
= btrfs_update_inode(trans
, root
, inode
);
4302 printk_ratelimited(KERN_ERR
"btrfs: fail to "
4303 "dirty inode %llu error %d\n",
4304 (unsigned long long)btrfs_ino(inode
),
4308 btrfs_end_transaction(trans
, root
);
4309 if (BTRFS_I(inode
)->delayed_node
)
4310 btrfs_balance_delayed_items(root
);
4314 * find the highest existing sequence number in a directory
4315 * and then set the in-memory index_cnt variable to reflect
4316 * free sequence numbers
4318 static int btrfs_set_inode_index_count(struct inode
*inode
)
4320 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4321 struct btrfs_key key
, found_key
;
4322 struct btrfs_path
*path
;
4323 struct extent_buffer
*leaf
;
4326 key
.objectid
= btrfs_ino(inode
);
4327 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4328 key
.offset
= (u64
)-1;
4330 path
= btrfs_alloc_path();
4334 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4337 /* FIXME: we should be able to handle this */
4343 * MAGIC NUMBER EXPLANATION:
4344 * since we search a directory based on f_pos we have to start at 2
4345 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4346 * else has to start at 2
4348 if (path
->slots
[0] == 0) {
4349 BTRFS_I(inode
)->index_cnt
= 2;
4355 leaf
= path
->nodes
[0];
4356 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4358 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4359 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4360 BTRFS_I(inode
)->index_cnt
= 2;
4364 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4366 btrfs_free_path(path
);
4371 * helper to find a free sequence number in a given directory. This current
4372 * code is very simple, later versions will do smarter things in the btree
4374 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4378 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4379 ret
= btrfs_inode_delayed_dir_index_count(dir
);
4381 ret
= btrfs_set_inode_index_count(dir
);
4387 *index
= BTRFS_I(dir
)->index_cnt
;
4388 BTRFS_I(dir
)->index_cnt
++;
4393 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4394 struct btrfs_root
*root
,
4396 const char *name
, int name_len
,
4397 u64 ref_objectid
, u64 objectid
, int mode
,
4400 struct inode
*inode
;
4401 struct btrfs_inode_item
*inode_item
;
4402 struct btrfs_key
*location
;
4403 struct btrfs_path
*path
;
4404 struct btrfs_inode_ref
*ref
;
4405 struct btrfs_key key
[2];
4411 path
= btrfs_alloc_path();
4414 inode
= new_inode(root
->fs_info
->sb
);
4416 btrfs_free_path(path
);
4417 return ERR_PTR(-ENOMEM
);
4421 * we have to initialize this early, so we can reclaim the inode
4422 * number if we fail afterwards in this function.
4424 inode
->i_ino
= objectid
;
4427 trace_btrfs_inode_request(dir
);
4429 ret
= btrfs_set_inode_index(dir
, index
);
4431 btrfs_free_path(path
);
4433 return ERR_PTR(ret
);
4437 * index_cnt is ignored for everything but a dir,
4438 * btrfs_get_inode_index_count has an explanation for the magic
4441 BTRFS_I(inode
)->index_cnt
= 2;
4442 BTRFS_I(inode
)->root
= root
;
4443 BTRFS_I(inode
)->generation
= trans
->transid
;
4444 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4445 btrfs_set_inode_space_info(root
, inode
);
4452 key
[0].objectid
= objectid
;
4453 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4456 key
[1].objectid
= objectid
;
4457 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4458 key
[1].offset
= ref_objectid
;
4460 sizes
[0] = sizeof(struct btrfs_inode_item
);
4461 sizes
[1] = name_len
+ sizeof(*ref
);
4463 path
->leave_spinning
= 1;
4464 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4468 inode_init_owner(inode
, dir
, mode
);
4469 inode_set_bytes(inode
, 0);
4470 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4471 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4472 struct btrfs_inode_item
);
4473 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4475 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4476 struct btrfs_inode_ref
);
4477 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4478 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4479 ptr
= (unsigned long)(ref
+ 1);
4480 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4482 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4483 btrfs_free_path(path
);
4485 location
= &BTRFS_I(inode
)->location
;
4486 location
->objectid
= objectid
;
4487 location
->offset
= 0;
4488 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4490 btrfs_inherit_iflags(inode
, dir
);
4492 if (S_ISREG(mode
)) {
4493 if (btrfs_test_opt(root
, NODATASUM
))
4494 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4495 if (btrfs_test_opt(root
, NODATACOW
) ||
4496 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4497 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4500 insert_inode_hash(inode
);
4501 inode_tree_add(inode
);
4503 trace_btrfs_inode_new(inode
);
4504 btrfs_set_inode_last_trans(trans
, inode
);
4509 BTRFS_I(dir
)->index_cnt
--;
4510 btrfs_free_path(path
);
4512 return ERR_PTR(ret
);
4515 static inline u8
btrfs_inode_type(struct inode
*inode
)
4517 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4521 * utility function to add 'inode' into 'parent_inode' with
4522 * a give name and a given sequence number.
4523 * if 'add_backref' is true, also insert a backref from the
4524 * inode to the parent directory.
4526 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4527 struct inode
*parent_inode
, struct inode
*inode
,
4528 const char *name
, int name_len
, int add_backref
, u64 index
)
4531 struct btrfs_key key
;
4532 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4533 u64 ino
= btrfs_ino(inode
);
4534 u64 parent_ino
= btrfs_ino(parent_inode
);
4536 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4537 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4540 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4544 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4545 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4546 key
.objectid
, root
->root_key
.objectid
,
4547 parent_ino
, index
, name
, name_len
);
4548 } else if (add_backref
) {
4549 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
4554 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4556 btrfs_inode_type(inode
), index
);
4559 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4561 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4562 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4567 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4568 struct inode
*dir
, struct dentry
*dentry
,
4569 struct inode
*inode
, int backref
, u64 index
)
4571 int err
= btrfs_add_link(trans
, dir
, inode
,
4572 dentry
->d_name
.name
, dentry
->d_name
.len
,
4575 d_instantiate(dentry
, inode
);
4583 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4584 int mode
, dev_t rdev
)
4586 struct btrfs_trans_handle
*trans
;
4587 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4588 struct inode
*inode
= NULL
;
4592 unsigned long nr
= 0;
4595 if (!new_valid_dev(rdev
))
4599 * 2 for inode item and ref
4601 * 1 for xattr if selinux is on
4603 trans
= btrfs_start_transaction(root
, 5);
4605 return PTR_ERR(trans
);
4607 err
= btrfs_find_free_ino(root
, &objectid
);
4611 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4612 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4614 if (IS_ERR(inode
)) {
4615 err
= PTR_ERR(inode
);
4619 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4625 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4629 inode
->i_op
= &btrfs_special_inode_operations
;
4630 init_special_inode(inode
, inode
->i_mode
, rdev
);
4631 btrfs_update_inode(trans
, root
, inode
);
4634 nr
= trans
->blocks_used
;
4635 btrfs_end_transaction_throttle(trans
, root
);
4636 btrfs_btree_balance_dirty(root
, nr
);
4638 inode_dec_link_count(inode
);
4644 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4645 int mode
, struct nameidata
*nd
)
4647 struct btrfs_trans_handle
*trans
;
4648 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4649 struct inode
*inode
= NULL
;
4652 unsigned long nr
= 0;
4657 * 2 for inode item and ref
4659 * 1 for xattr if selinux is on
4661 trans
= btrfs_start_transaction(root
, 5);
4663 return PTR_ERR(trans
);
4665 err
= btrfs_find_free_ino(root
, &objectid
);
4669 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4670 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4672 if (IS_ERR(inode
)) {
4673 err
= PTR_ERR(inode
);
4677 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4683 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4687 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4688 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4689 inode
->i_fop
= &btrfs_file_operations
;
4690 inode
->i_op
= &btrfs_file_inode_operations
;
4691 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4694 nr
= trans
->blocks_used
;
4695 btrfs_end_transaction_throttle(trans
, root
);
4697 inode_dec_link_count(inode
);
4700 btrfs_btree_balance_dirty(root
, nr
);
4704 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4705 struct dentry
*dentry
)
4707 struct btrfs_trans_handle
*trans
;
4708 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4709 struct inode
*inode
= old_dentry
->d_inode
;
4711 unsigned long nr
= 0;
4715 /* do not allow sys_link's with other subvols of the same device */
4716 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4719 if (inode
->i_nlink
== ~0U)
4722 err
= btrfs_set_inode_index(dir
, &index
);
4727 * 2 items for inode and inode ref
4728 * 2 items for dir items
4729 * 1 item for parent inode
4731 trans
= btrfs_start_transaction(root
, 5);
4732 if (IS_ERR(trans
)) {
4733 err
= PTR_ERR(trans
);
4737 btrfs_inc_nlink(inode
);
4738 inode
->i_ctime
= CURRENT_TIME
;
4741 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4746 struct dentry
*parent
= dentry
->d_parent
;
4747 err
= btrfs_update_inode(trans
, root
, inode
);
4749 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4752 nr
= trans
->blocks_used
;
4753 btrfs_end_transaction_throttle(trans
, root
);
4756 inode_dec_link_count(inode
);
4759 btrfs_btree_balance_dirty(root
, nr
);
4763 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4765 struct inode
*inode
= NULL
;
4766 struct btrfs_trans_handle
*trans
;
4767 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4769 int drop_on_err
= 0;
4772 unsigned long nr
= 1;
4775 * 2 items for inode and ref
4776 * 2 items for dir items
4777 * 1 for xattr if selinux is on
4779 trans
= btrfs_start_transaction(root
, 5);
4781 return PTR_ERR(trans
);
4783 err
= btrfs_find_free_ino(root
, &objectid
);
4787 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4788 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4789 S_IFDIR
| mode
, &index
);
4790 if (IS_ERR(inode
)) {
4791 err
= PTR_ERR(inode
);
4797 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4801 inode
->i_op
= &btrfs_dir_inode_operations
;
4802 inode
->i_fop
= &btrfs_dir_file_operations
;
4804 btrfs_i_size_write(inode
, 0);
4805 err
= btrfs_update_inode(trans
, root
, inode
);
4809 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4810 dentry
->d_name
.len
, 0, index
);
4814 d_instantiate(dentry
, inode
);
4818 nr
= trans
->blocks_used
;
4819 btrfs_end_transaction_throttle(trans
, root
);
4822 btrfs_btree_balance_dirty(root
, nr
);
4826 /* helper for btfs_get_extent. Given an existing extent in the tree,
4827 * and an extent that you want to insert, deal with overlap and insert
4828 * the new extent into the tree.
4830 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4831 struct extent_map
*existing
,
4832 struct extent_map
*em
,
4833 u64 map_start
, u64 map_len
)
4837 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4838 start_diff
= map_start
- em
->start
;
4839 em
->start
= map_start
;
4841 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4842 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4843 em
->block_start
+= start_diff
;
4844 em
->block_len
-= start_diff
;
4846 return add_extent_mapping(em_tree
, em
);
4849 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4850 struct inode
*inode
, struct page
*page
,
4851 size_t pg_offset
, u64 extent_offset
,
4852 struct btrfs_file_extent_item
*item
)
4855 struct extent_buffer
*leaf
= path
->nodes
[0];
4858 unsigned long inline_size
;
4862 WARN_ON(pg_offset
!= 0);
4863 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4864 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4865 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4866 btrfs_item_nr(leaf
, path
->slots
[0]));
4867 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4870 ptr
= btrfs_file_extent_inline_start(item
);
4872 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4874 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4875 ret
= btrfs_decompress(compress_type
, tmp
, page
,
4876 extent_offset
, inline_size
, max_size
);
4878 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4879 unsigned long copy_size
= min_t(u64
,
4880 PAGE_CACHE_SIZE
- pg_offset
,
4881 max_size
- extent_offset
);
4882 memset(kaddr
+ pg_offset
, 0, copy_size
);
4883 kunmap_atomic(kaddr
, KM_USER0
);
4890 * a bit scary, this does extent mapping from logical file offset to the disk.
4891 * the ugly parts come from merging extents from the disk with the in-ram
4892 * representation. This gets more complex because of the data=ordered code,
4893 * where the in-ram extents might be locked pending data=ordered completion.
4895 * This also copies inline extents directly into the page.
4898 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
4899 size_t pg_offset
, u64 start
, u64 len
,
4905 u64 extent_start
= 0;
4907 u64 objectid
= btrfs_ino(inode
);
4909 struct btrfs_path
*path
= NULL
;
4910 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4911 struct btrfs_file_extent_item
*item
;
4912 struct extent_buffer
*leaf
;
4913 struct btrfs_key found_key
;
4914 struct extent_map
*em
= NULL
;
4915 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4916 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4917 struct btrfs_trans_handle
*trans
= NULL
;
4921 read_lock(&em_tree
->lock
);
4922 em
= lookup_extent_mapping(em_tree
, start
, len
);
4924 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4925 read_unlock(&em_tree
->lock
);
4928 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
4929 free_extent_map(em
);
4930 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
4931 free_extent_map(em
);
4935 em
= alloc_extent_map();
4940 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4941 em
->start
= EXTENT_MAP_HOLE
;
4942 em
->orig_start
= EXTENT_MAP_HOLE
;
4944 em
->block_len
= (u64
)-1;
4947 path
= btrfs_alloc_path();
4953 * Chances are we'll be called again, so go ahead and do
4959 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
4960 objectid
, start
, trans
!= NULL
);
4967 if (path
->slots
[0] == 0)
4972 leaf
= path
->nodes
[0];
4973 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4974 struct btrfs_file_extent_item
);
4975 /* are we inside the extent that was found? */
4976 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4977 found_type
= btrfs_key_type(&found_key
);
4978 if (found_key
.objectid
!= objectid
||
4979 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4983 found_type
= btrfs_file_extent_type(leaf
, item
);
4984 extent_start
= found_key
.offset
;
4985 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4986 if (found_type
== BTRFS_FILE_EXTENT_REG
||
4987 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4988 extent_end
= extent_start
+
4989 btrfs_file_extent_num_bytes(leaf
, item
);
4990 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4992 size
= btrfs_file_extent_inline_len(leaf
, item
);
4993 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
4994 ~((u64
)root
->sectorsize
- 1);
4997 if (start
>= extent_end
) {
4999 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5000 ret
= btrfs_next_leaf(root
, path
);
5007 leaf
= path
->nodes
[0];
5009 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5010 if (found_key
.objectid
!= objectid
||
5011 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5013 if (start
+ len
<= found_key
.offset
)
5016 em
->len
= found_key
.offset
- start
;
5020 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5021 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5022 em
->start
= extent_start
;
5023 em
->len
= extent_end
- extent_start
;
5024 em
->orig_start
= extent_start
-
5025 btrfs_file_extent_offset(leaf
, item
);
5026 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5028 em
->block_start
= EXTENT_MAP_HOLE
;
5031 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5032 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5033 em
->compress_type
= compress_type
;
5034 em
->block_start
= bytenr
;
5035 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5038 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5039 em
->block_start
= bytenr
;
5040 em
->block_len
= em
->len
;
5041 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5042 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5045 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5049 size_t extent_offset
;
5052 em
->block_start
= EXTENT_MAP_INLINE
;
5053 if (!page
|| create
) {
5054 em
->start
= extent_start
;
5055 em
->len
= extent_end
- extent_start
;
5059 size
= btrfs_file_extent_inline_len(leaf
, item
);
5060 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5061 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5062 size
- extent_offset
);
5063 em
->start
= extent_start
+ extent_offset
;
5064 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5065 ~((u64
)root
->sectorsize
- 1);
5066 em
->orig_start
= EXTENT_MAP_INLINE
;
5067 if (compress_type
) {
5068 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5069 em
->compress_type
= compress_type
;
5071 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5072 if (create
== 0 && !PageUptodate(page
)) {
5073 if (btrfs_file_extent_compression(leaf
, item
) !=
5074 BTRFS_COMPRESS_NONE
) {
5075 ret
= uncompress_inline(path
, inode
, page
,
5077 extent_offset
, item
);
5081 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5083 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5084 memset(map
+ pg_offset
+ copy_size
, 0,
5085 PAGE_CACHE_SIZE
- pg_offset
-
5090 flush_dcache_page(page
);
5091 } else if (create
&& PageUptodate(page
)) {
5095 free_extent_map(em
);
5098 btrfs_release_path(path
);
5099 trans
= btrfs_join_transaction(root
);
5102 return ERR_CAST(trans
);
5106 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5109 btrfs_mark_buffer_dirty(leaf
);
5111 set_extent_uptodate(io_tree
, em
->start
,
5112 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5115 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5122 em
->block_start
= EXTENT_MAP_HOLE
;
5123 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5125 btrfs_release_path(path
);
5126 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5127 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5128 "[%llu %llu]\n", (unsigned long long)em
->start
,
5129 (unsigned long long)em
->len
,
5130 (unsigned long long)start
,
5131 (unsigned long long)len
);
5137 write_lock(&em_tree
->lock
);
5138 ret
= add_extent_mapping(em_tree
, em
);
5139 /* it is possible that someone inserted the extent into the tree
5140 * while we had the lock dropped. It is also possible that
5141 * an overlapping map exists in the tree
5143 if (ret
== -EEXIST
) {
5144 struct extent_map
*existing
;
5148 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5149 if (existing
&& (existing
->start
> start
||
5150 existing
->start
+ existing
->len
<= start
)) {
5151 free_extent_map(existing
);
5155 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5158 err
= merge_extent_mapping(em_tree
, existing
,
5161 free_extent_map(existing
);
5163 free_extent_map(em
);
5168 free_extent_map(em
);
5172 free_extent_map(em
);
5177 write_unlock(&em_tree
->lock
);
5180 trace_btrfs_get_extent(root
, em
);
5183 btrfs_free_path(path
);
5185 ret
= btrfs_end_transaction(trans
, root
);
5190 free_extent_map(em
);
5191 return ERR_PTR(err
);
5196 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5197 size_t pg_offset
, u64 start
, u64 len
,
5200 struct extent_map
*em
;
5201 struct extent_map
*hole_em
= NULL
;
5202 u64 range_start
= start
;
5208 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5213 * if our em maps to a hole, there might
5214 * actually be delalloc bytes behind it
5216 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5222 /* check to see if we've wrapped (len == -1 or similar) */
5231 /* ok, we didn't find anything, lets look for delalloc */
5232 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5233 end
, len
, EXTENT_DELALLOC
, 1);
5234 found_end
= range_start
+ found
;
5235 if (found_end
< range_start
)
5236 found_end
= (u64
)-1;
5239 * we didn't find anything useful, return
5240 * the original results from get_extent()
5242 if (range_start
> end
|| found_end
<= start
) {
5248 /* adjust the range_start to make sure it doesn't
5249 * go backwards from the start they passed in
5251 range_start
= max(start
,range_start
);
5252 found
= found_end
- range_start
;
5255 u64 hole_start
= start
;
5258 em
= alloc_extent_map();
5264 * when btrfs_get_extent can't find anything it
5265 * returns one huge hole
5267 * make sure what it found really fits our range, and
5268 * adjust to make sure it is based on the start from
5272 u64 calc_end
= extent_map_end(hole_em
);
5274 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5275 free_extent_map(hole_em
);
5278 hole_start
= max(hole_em
->start
, start
);
5279 hole_len
= calc_end
- hole_start
;
5283 if (hole_em
&& range_start
> hole_start
) {
5284 /* our hole starts before our delalloc, so we
5285 * have to return just the parts of the hole
5286 * that go until the delalloc starts
5288 em
->len
= min(hole_len
,
5289 range_start
- hole_start
);
5290 em
->start
= hole_start
;
5291 em
->orig_start
= hole_start
;
5293 * don't adjust block start at all,
5294 * it is fixed at EXTENT_MAP_HOLE
5296 em
->block_start
= hole_em
->block_start
;
5297 em
->block_len
= hole_len
;
5299 em
->start
= range_start
;
5301 em
->orig_start
= range_start
;
5302 em
->block_start
= EXTENT_MAP_DELALLOC
;
5303 em
->block_len
= found
;
5305 } else if (hole_em
) {
5310 free_extent_map(hole_em
);
5312 free_extent_map(em
);
5313 return ERR_PTR(err
);
5318 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5319 struct extent_map
*em
,
5322 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5323 struct btrfs_trans_handle
*trans
;
5324 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5325 struct btrfs_key ins
;
5328 bool insert
= false;
5331 * Ok if the extent map we looked up is a hole and is for the exact
5332 * range we want, there is no reason to allocate a new one, however if
5333 * it is not right then we need to free this one and drop the cache for
5336 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5338 free_extent_map(em
);
5341 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5344 trans
= btrfs_join_transaction(root
);
5346 return ERR_CAST(trans
);
5348 if (start
<= BTRFS_I(inode
)->disk_i_size
&& len
< 64 * 1024)
5349 btrfs_add_inode_defrag(trans
, inode
);
5351 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5353 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5354 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5355 alloc_hint
, (u64
)-1, &ins
, 1);
5362 em
= alloc_extent_map();
5364 em
= ERR_PTR(-ENOMEM
);
5370 em
->orig_start
= em
->start
;
5371 em
->len
= ins
.offset
;
5373 em
->block_start
= ins
.objectid
;
5374 em
->block_len
= ins
.offset
;
5375 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5378 * We need to do this because if we're using the original em we searched
5379 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5382 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5385 write_lock(&em_tree
->lock
);
5386 ret
= add_extent_mapping(em_tree
, em
);
5387 write_unlock(&em_tree
->lock
);
5390 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5393 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5394 ins
.offset
, ins
.offset
, 0);
5396 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5400 btrfs_end_transaction(trans
, root
);
5405 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5406 * block must be cow'd
5408 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5409 struct inode
*inode
, u64 offset
, u64 len
)
5411 struct btrfs_path
*path
;
5413 struct extent_buffer
*leaf
;
5414 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5415 struct btrfs_file_extent_item
*fi
;
5416 struct btrfs_key key
;
5424 path
= btrfs_alloc_path();
5428 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
5433 slot
= path
->slots
[0];
5436 /* can't find the item, must cow */
5443 leaf
= path
->nodes
[0];
5444 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5445 if (key
.objectid
!= btrfs_ino(inode
) ||
5446 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5447 /* not our file or wrong item type, must cow */
5451 if (key
.offset
> offset
) {
5452 /* Wrong offset, must cow */
5456 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5457 found_type
= btrfs_file_extent_type(leaf
, fi
);
5458 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5459 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5460 /* not a regular extent, must cow */
5463 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5464 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5466 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5467 if (extent_end
< offset
+ len
) {
5468 /* extent doesn't include our full range, must cow */
5472 if (btrfs_extent_readonly(root
, disk_bytenr
))
5476 * look for other files referencing this extent, if we
5477 * find any we must cow
5479 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
5480 key
.offset
- backref_offset
, disk_bytenr
))
5484 * adjust disk_bytenr and num_bytes to cover just the bytes
5485 * in this extent we are about to write. If there
5486 * are any csums in that range we have to cow in order
5487 * to keep the csums correct
5489 disk_bytenr
+= backref_offset
;
5490 disk_bytenr
+= offset
- key
.offset
;
5491 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5492 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5495 * all of the above have passed, it is safe to overwrite this extent
5500 btrfs_free_path(path
);
5504 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5505 struct buffer_head
*bh_result
, int create
)
5507 struct extent_map
*em
;
5508 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5509 u64 start
= iblock
<< inode
->i_blkbits
;
5510 u64 len
= bh_result
->b_size
;
5511 struct btrfs_trans_handle
*trans
;
5513 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5518 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5519 * io. INLINE is special, and we could probably kludge it in here, but
5520 * it's still buffered so for safety lets just fall back to the generic
5523 * For COMPRESSED we _have_ to read the entire extent in so we can
5524 * decompress it, so there will be buffering required no matter what we
5525 * do, so go ahead and fallback to buffered.
5527 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5528 * to buffered IO. Don't blame me, this is the price we pay for using
5531 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5532 em
->block_start
== EXTENT_MAP_INLINE
) {
5533 free_extent_map(em
);
5537 /* Just a good old fashioned hole, return */
5538 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5539 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5540 free_extent_map(em
);
5541 /* DIO will do one hole at a time, so just unlock a sector */
5542 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5543 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5548 * We don't allocate a new extent in the following cases
5550 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5552 * 2) The extent is marked as PREALLOC. We're good to go here and can
5553 * just use the extent.
5557 len
= em
->len
- (start
- em
->start
);
5561 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5562 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5563 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5568 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5569 type
= BTRFS_ORDERED_PREALLOC
;
5571 type
= BTRFS_ORDERED_NOCOW
;
5572 len
= min(len
, em
->len
- (start
- em
->start
));
5573 block_start
= em
->block_start
+ (start
- em
->start
);
5576 * we're not going to log anything, but we do need
5577 * to make sure the current transaction stays open
5578 * while we look for nocow cross refs
5580 trans
= btrfs_join_transaction(root
);
5584 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5585 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5586 block_start
, len
, len
, type
);
5587 btrfs_end_transaction(trans
, root
);
5589 free_extent_map(em
);
5594 btrfs_end_transaction(trans
, root
);
5598 * this will cow the extent, reset the len in case we changed
5601 len
= bh_result
->b_size
;
5602 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
5605 len
= min(len
, em
->len
- (start
- em
->start
));
5607 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5608 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5611 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5613 bh_result
->b_size
= len
;
5614 bh_result
->b_bdev
= em
->bdev
;
5615 set_buffer_mapped(bh_result
);
5616 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5617 set_buffer_new(bh_result
);
5619 free_extent_map(em
);
5624 struct btrfs_dio_private
{
5625 struct inode
*inode
;
5632 /* number of bios pending for this dio */
5633 atomic_t pending_bios
;
5638 struct bio
*orig_bio
;
5641 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5643 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5644 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5645 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5646 struct inode
*inode
= dip
->inode
;
5647 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5649 u32
*private = dip
->csums
;
5651 start
= dip
->logical_offset
;
5653 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5654 struct page
*page
= bvec
->bv_page
;
5657 unsigned long flags
;
5659 local_irq_save(flags
);
5660 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5661 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5662 csum
, bvec
->bv_len
);
5663 btrfs_csum_final(csum
, (char *)&csum
);
5664 kunmap_atomic(kaddr
, KM_IRQ0
);
5665 local_irq_restore(flags
);
5667 flush_dcache_page(bvec
->bv_page
);
5668 if (csum
!= *private) {
5669 printk(KERN_ERR
"btrfs csum failed ino %llu off"
5670 " %llu csum %u private %u\n",
5671 (unsigned long long)btrfs_ino(inode
),
5672 (unsigned long long)start
,
5678 start
+= bvec
->bv_len
;
5681 } while (bvec
<= bvec_end
);
5683 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5684 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5685 bio
->bi_private
= dip
->private;
5690 /* If we had a csum failure make sure to clear the uptodate flag */
5692 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5693 dio_end_io(bio
, err
);
5696 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5698 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5699 struct inode
*inode
= dip
->inode
;
5700 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5701 struct btrfs_trans_handle
*trans
;
5702 struct btrfs_ordered_extent
*ordered
= NULL
;
5703 struct extent_state
*cached_state
= NULL
;
5704 u64 ordered_offset
= dip
->logical_offset
;
5705 u64 ordered_bytes
= dip
->bytes
;
5711 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5719 trans
= btrfs_join_transaction(root
);
5720 if (IS_ERR(trans
)) {
5724 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5726 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5727 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5729 ret
= btrfs_update_inode(trans
, root
, inode
);
5734 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5735 ordered
->file_offset
+ ordered
->len
- 1, 0,
5736 &cached_state
, GFP_NOFS
);
5738 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5739 ret
= btrfs_mark_extent_written(trans
, inode
,
5740 ordered
->file_offset
,
5741 ordered
->file_offset
+
5748 ret
= insert_reserved_file_extent(trans
, inode
,
5749 ordered
->file_offset
,
5755 BTRFS_FILE_EXTENT_REG
);
5756 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5757 ordered
->file_offset
, ordered
->len
);
5765 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5766 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5768 btrfs_update_inode(trans
, root
, inode
);
5771 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5772 ordered
->file_offset
+ ordered
->len
- 1,
5773 &cached_state
, GFP_NOFS
);
5775 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5776 btrfs_end_transaction(trans
, root
);
5777 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5778 btrfs_put_ordered_extent(ordered
);
5779 btrfs_put_ordered_extent(ordered
);
5783 * our bio might span multiple ordered extents. If we haven't
5784 * completed the accounting for the whole dio, go back and try again
5786 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5787 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5792 bio
->bi_private
= dip
->private;
5797 /* If we had an error make sure to clear the uptodate flag */
5799 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5800 dio_end_io(bio
, err
);
5803 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5804 struct bio
*bio
, int mirror_num
,
5805 unsigned long bio_flags
, u64 offset
)
5808 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5809 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5814 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5816 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5819 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
5820 "sector %#Lx len %u err no %d\n",
5821 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
5822 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5826 * before atomic variable goto zero, we must make sure
5827 * dip->errors is perceived to be set.
5829 smp_mb__before_atomic_dec();
5832 /* if there are more bios still pending for this dio, just exit */
5833 if (!atomic_dec_and_test(&dip
->pending_bios
))
5837 bio_io_error(dip
->orig_bio
);
5839 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5840 bio_endio(dip
->orig_bio
, 0);
5846 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5847 u64 first_sector
, gfp_t gfp_flags
)
5849 int nr_vecs
= bio_get_nr_vecs(bdev
);
5850 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5853 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5854 int rw
, u64 file_offset
, int skip_sum
,
5855 u32
*csums
, int async_submit
)
5857 int write
= rw
& REQ_WRITE
;
5858 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5862 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5869 if (write
&& async_submit
) {
5870 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5871 inode
, rw
, bio
, 0, 0,
5873 __btrfs_submit_bio_start_direct_io
,
5874 __btrfs_submit_bio_done
);
5878 * If we aren't doing async submit, calculate the csum of the
5881 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
5884 } else if (!skip_sum
) {
5885 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5886 file_offset
, csums
);
5892 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
5898 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5901 struct inode
*inode
= dip
->inode
;
5902 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5903 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5905 struct bio
*orig_bio
= dip
->orig_bio
;
5906 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5907 u64 start_sector
= orig_bio
->bi_sector
;
5908 u64 file_offset
= dip
->logical_offset
;
5912 u32
*csums
= dip
->csums
;
5914 int async_submit
= 0;
5915 int write
= rw
& REQ_WRITE
;
5917 map_length
= orig_bio
->bi_size
;
5918 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5919 &map_length
, NULL
, 0);
5925 if (map_length
>= orig_bio
->bi_size
) {
5931 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
5934 bio
->bi_private
= dip
;
5935 bio
->bi_end_io
= btrfs_end_dio_bio
;
5936 atomic_inc(&dip
->pending_bios
);
5938 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
5939 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
5940 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5941 bvec
->bv_offset
) < bvec
->bv_len
)) {
5943 * inc the count before we submit the bio so
5944 * we know the end IO handler won't happen before
5945 * we inc the count. Otherwise, the dip might get freed
5946 * before we're done setting it up
5948 atomic_inc(&dip
->pending_bios
);
5949 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
5950 file_offset
, skip_sum
,
5951 csums
, async_submit
);
5954 atomic_dec(&dip
->pending_bios
);
5958 /* Write's use the ordered csums */
5959 if (!write
&& !skip_sum
)
5960 csums
= csums
+ nr_pages
;
5961 start_sector
+= submit_len
>> 9;
5962 file_offset
+= submit_len
;
5967 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
5968 start_sector
, GFP_NOFS
);
5971 bio
->bi_private
= dip
;
5972 bio
->bi_end_io
= btrfs_end_dio_bio
;
5974 map_length
= orig_bio
->bi_size
;
5975 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5976 &map_length
, NULL
, 0);
5982 submit_len
+= bvec
->bv_len
;
5989 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
5990 csums
, async_submit
);
5998 * before atomic variable goto zero, we must
5999 * make sure dip->errors is perceived to be set.
6001 smp_mb__before_atomic_dec();
6002 if (atomic_dec_and_test(&dip
->pending_bios
))
6003 bio_io_error(dip
->orig_bio
);
6005 /* bio_end_io() will handle error, so we needn't return it */
6009 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6012 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6013 struct btrfs_dio_private
*dip
;
6014 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6016 int write
= rw
& REQ_WRITE
;
6019 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6021 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6028 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6029 if (!write
&& !skip_sum
) {
6030 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6038 dip
->private = bio
->bi_private
;
6040 dip
->logical_offset
= file_offset
;
6044 dip
->bytes
+= bvec
->bv_len
;
6046 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6048 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6049 bio
->bi_private
= dip
;
6051 dip
->orig_bio
= bio
;
6052 atomic_set(&dip
->pending_bios
, 0);
6055 bio
->bi_end_io
= btrfs_endio_direct_write
;
6057 bio
->bi_end_io
= btrfs_endio_direct_read
;
6059 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6064 * If this is a write, we need to clean up the reserved space and kill
6065 * the ordered extent.
6068 struct btrfs_ordered_extent
*ordered
;
6069 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6070 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6071 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6072 btrfs_free_reserved_extent(root
, ordered
->start
,
6074 btrfs_put_ordered_extent(ordered
);
6075 btrfs_put_ordered_extent(ordered
);
6077 bio_endio(bio
, ret
);
6080 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6081 const struct iovec
*iov
, loff_t offset
,
6082 unsigned long nr_segs
)
6088 unsigned blocksize_mask
= root
->sectorsize
- 1;
6089 ssize_t retval
= -EINVAL
;
6090 loff_t end
= offset
;
6092 if (offset
& blocksize_mask
)
6095 /* Check the memory alignment. Blocks cannot straddle pages */
6096 for (seg
= 0; seg
< nr_segs
; seg
++) {
6097 addr
= (unsigned long)iov
[seg
].iov_base
;
6098 size
= iov
[seg
].iov_len
;
6100 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6103 /* If this is a write we don't need to check anymore */
6108 * Check to make sure we don't have duplicate iov_base's in this
6109 * iovec, if so return EINVAL, otherwise we'll get csum errors
6110 * when reading back.
6112 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6113 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6121 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6122 const struct iovec
*iov
, loff_t offset
,
6123 unsigned long nr_segs
)
6125 struct file
*file
= iocb
->ki_filp
;
6126 struct inode
*inode
= file
->f_mapping
->host
;
6127 struct btrfs_ordered_extent
*ordered
;
6128 struct extent_state
*cached_state
= NULL
;
6129 u64 lockstart
, lockend
;
6131 int writing
= rw
& WRITE
;
6133 size_t count
= iov_length(iov
, nr_segs
);
6135 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6141 lockend
= offset
+ count
- 1;
6144 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6150 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6151 0, &cached_state
, GFP_NOFS
);
6153 * We're concerned with the entire range that we're going to be
6154 * doing DIO to, so we need to make sure theres no ordered
6155 * extents in this range.
6157 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6158 lockend
- lockstart
+ 1);
6161 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6162 &cached_state
, GFP_NOFS
);
6163 btrfs_start_ordered_extent(inode
, ordered
, 1);
6164 btrfs_put_ordered_extent(ordered
);
6169 * we don't use btrfs_set_extent_delalloc because we don't want
6170 * the dirty or uptodate bits
6173 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6174 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6175 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6178 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6179 lockend
, EXTENT_LOCKED
| write_bits
,
6180 1, 0, &cached_state
, GFP_NOFS
);
6185 free_extent_state(cached_state
);
6186 cached_state
= NULL
;
6188 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6189 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6190 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6191 btrfs_submit_direct
, 0);
6193 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6194 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6195 offset
+ iov_length(iov
, nr_segs
) - 1,
6196 EXTENT_LOCKED
| write_bits
, 1, 0,
6197 &cached_state
, GFP_NOFS
);
6198 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6200 * We're falling back to buffered, unlock the section we didn't
6203 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6204 offset
+ iov_length(iov
, nr_segs
) - 1,
6205 EXTENT_LOCKED
| write_bits
, 1, 0,
6206 &cached_state
, GFP_NOFS
);
6209 free_extent_state(cached_state
);
6213 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6214 __u64 start
, __u64 len
)
6216 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6219 int btrfs_readpage(struct file
*file
, struct page
*page
)
6221 struct extent_io_tree
*tree
;
6222 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6223 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6226 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6228 struct extent_io_tree
*tree
;
6231 if (current
->flags
& PF_MEMALLOC
) {
6232 redirty_page_for_writepage(wbc
, page
);
6236 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6237 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6240 int btrfs_writepages(struct address_space
*mapping
,
6241 struct writeback_control
*wbc
)
6243 struct extent_io_tree
*tree
;
6245 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6246 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6250 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6251 struct list_head
*pages
, unsigned nr_pages
)
6253 struct extent_io_tree
*tree
;
6254 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6255 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6258 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6260 struct extent_io_tree
*tree
;
6261 struct extent_map_tree
*map
;
6264 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6265 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6266 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6268 ClearPagePrivate(page
);
6269 set_page_private(page
, 0);
6270 page_cache_release(page
);
6275 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6277 if (PageWriteback(page
) || PageDirty(page
))
6279 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6282 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6284 struct extent_io_tree
*tree
;
6285 struct btrfs_ordered_extent
*ordered
;
6286 struct extent_state
*cached_state
= NULL
;
6287 u64 page_start
= page_offset(page
);
6288 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6292 * we have the page locked, so new writeback can't start,
6293 * and the dirty bit won't be cleared while we are here.
6295 * Wait for IO on this page so that we can safely clear
6296 * the PagePrivate2 bit and do ordered accounting
6298 wait_on_page_writeback(page
);
6300 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6302 btrfs_releasepage(page
, GFP_NOFS
);
6305 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6307 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6311 * IO on this page will never be started, so we need
6312 * to account for any ordered extents now
6314 clear_extent_bit(tree
, page_start
, page_end
,
6315 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6316 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6317 &cached_state
, GFP_NOFS
);
6319 * whoever cleared the private bit is responsible
6320 * for the finish_ordered_io
6322 if (TestClearPagePrivate2(page
)) {
6323 btrfs_finish_ordered_io(page
->mapping
->host
,
6324 page_start
, page_end
);
6326 btrfs_put_ordered_extent(ordered
);
6327 cached_state
= NULL
;
6328 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6331 clear_extent_bit(tree
, page_start
, page_end
,
6332 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6333 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6334 __btrfs_releasepage(page
, GFP_NOFS
);
6336 ClearPageChecked(page
);
6337 if (PagePrivate(page
)) {
6338 ClearPagePrivate(page
);
6339 set_page_private(page
, 0);
6340 page_cache_release(page
);
6345 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6346 * called from a page fault handler when a page is first dirtied. Hence we must
6347 * be careful to check for EOF conditions here. We set the page up correctly
6348 * for a written page which means we get ENOSPC checking when writing into
6349 * holes and correct delalloc and unwritten extent mapping on filesystems that
6350 * support these features.
6352 * We are not allowed to take the i_mutex here so we have to play games to
6353 * protect against truncate races as the page could now be beyond EOF. Because
6354 * vmtruncate() writes the inode size before removing pages, once we have the
6355 * page lock we can determine safely if the page is beyond EOF. If it is not
6356 * beyond EOF, then the page is guaranteed safe against truncation until we
6359 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6361 struct page
*page
= vmf
->page
;
6362 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6363 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6364 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6365 struct btrfs_ordered_extent
*ordered
;
6366 struct extent_state
*cached_state
= NULL
;
6368 unsigned long zero_start
;
6374 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6378 else /* -ENOSPC, -EIO, etc */
6379 ret
= VM_FAULT_SIGBUS
;
6383 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6386 size
= i_size_read(inode
);
6387 page_start
= page_offset(page
);
6388 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6390 if ((page
->mapping
!= inode
->i_mapping
) ||
6391 (page_start
>= size
)) {
6392 /* page got truncated out from underneath us */
6395 wait_on_page_writeback(page
);
6397 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6399 set_page_extent_mapped(page
);
6402 * we can't set the delalloc bits if there are pending ordered
6403 * extents. Drop our locks and wait for them to finish
6405 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6407 unlock_extent_cached(io_tree
, page_start
, page_end
,
6408 &cached_state
, GFP_NOFS
);
6410 btrfs_start_ordered_extent(inode
, ordered
, 1);
6411 btrfs_put_ordered_extent(ordered
);
6416 * XXX - page_mkwrite gets called every time the page is dirtied, even
6417 * if it was already dirty, so for space accounting reasons we need to
6418 * clear any delalloc bits for the range we are fixing to save. There
6419 * is probably a better way to do this, but for now keep consistent with
6420 * prepare_pages in the normal write path.
6422 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6423 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6424 0, 0, &cached_state
, GFP_NOFS
);
6426 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6429 unlock_extent_cached(io_tree
, page_start
, page_end
,
6430 &cached_state
, GFP_NOFS
);
6431 ret
= VM_FAULT_SIGBUS
;
6436 /* page is wholly or partially inside EOF */
6437 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6438 zero_start
= size
& ~PAGE_CACHE_MASK
;
6440 zero_start
= PAGE_CACHE_SIZE
;
6442 if (zero_start
!= PAGE_CACHE_SIZE
) {
6444 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6445 flush_dcache_page(page
);
6448 ClearPageChecked(page
);
6449 set_page_dirty(page
);
6450 SetPageUptodate(page
);
6452 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6453 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6455 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6459 return VM_FAULT_LOCKED
;
6461 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6466 static int btrfs_truncate(struct inode
*inode
)
6468 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6469 struct btrfs_block_rsv
*rsv
;
6472 struct btrfs_trans_handle
*trans
;
6474 u64 mask
= root
->sectorsize
- 1;
6476 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6480 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6481 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6484 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6485 * 3 things going on here
6487 * 1) We need to reserve space for our orphan item and the space to
6488 * delete our orphan item. Lord knows we don't want to have a dangling
6489 * orphan item because we didn't reserve space to remove it.
6491 * 2) We need to reserve space to update our inode.
6493 * 3) We need to have something to cache all the space that is going to
6494 * be free'd up by the truncate operation, but also have some slack
6495 * space reserved in case it uses space during the truncate (thank you
6496 * very much snapshotting).
6498 * And we need these to all be seperate. The fact is we can use alot of
6499 * space doing the truncate, and we have no earthly idea how much space
6500 * we will use, so we need the truncate reservation to be seperate so it
6501 * doesn't end up using space reserved for updating the inode or
6502 * removing the orphan item. We also need to be able to stop the
6503 * transaction and start a new one, which means we need to be able to
6504 * update the inode several times, and we have no idea of knowing how
6505 * many times that will be, so we can't just reserve 1 item for the
6506 * entirety of the opration, so that has to be done seperately as well.
6507 * Then there is the orphan item, which does indeed need to be held on
6508 * to for the whole operation, and we need nobody to touch this reserved
6509 * space except the orphan code.
6511 * So that leaves us with
6513 * 1) root->orphan_block_rsv - for the orphan deletion.
6514 * 2) rsv - for the truncate reservation, which we will steal from the
6515 * transaction reservation.
6516 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6517 * updating the inode.
6519 rsv
= btrfs_alloc_block_rsv(root
);
6522 btrfs_add_durable_block_rsv(root
->fs_info
, rsv
);
6524 trans
= btrfs_start_transaction(root
, 4);
6525 if (IS_ERR(trans
)) {
6526 err
= PTR_ERR(trans
);
6531 * Reserve space for the truncate process. Truncate should be adding
6532 * space, but if there are snapshots it may end up using space.
6534 ret
= btrfs_truncate_reserve_metadata(trans
, root
, rsv
);
6537 ret
= btrfs_orphan_add(trans
, inode
);
6539 btrfs_end_transaction(trans
, root
);
6543 nr
= trans
->blocks_used
;
6544 btrfs_end_transaction(trans
, root
);
6545 btrfs_btree_balance_dirty(root
, nr
);
6548 * Ok so we've already migrated our bytes over for the truncate, so here
6549 * just reserve the one slot we need for updating the inode.
6551 trans
= btrfs_start_transaction(root
, 1);
6552 if (IS_ERR(trans
)) {
6553 err
= PTR_ERR(trans
);
6556 trans
->block_rsv
= rsv
;
6559 * setattr is responsible for setting the ordered_data_close flag,
6560 * but that is only tested during the last file release. That
6561 * could happen well after the next commit, leaving a great big
6562 * window where new writes may get lost if someone chooses to write
6563 * to this file after truncating to zero
6565 * The inode doesn't have any dirty data here, and so if we commit
6566 * this is a noop. If someone immediately starts writing to the inode
6567 * it is very likely we'll catch some of their writes in this
6568 * transaction, and the commit will find this file on the ordered
6569 * data list with good things to send down.
6571 * This is a best effort solution, there is still a window where
6572 * using truncate to replace the contents of the file will
6573 * end up with a zero length file after a crash.
6575 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6576 btrfs_add_ordered_operation(trans
, root
, inode
);
6580 trans
= btrfs_start_transaction(root
, 3);
6581 if (IS_ERR(trans
)) {
6582 err
= PTR_ERR(trans
);
6586 ret
= btrfs_truncate_reserve_metadata(trans
, root
,
6590 trans
->block_rsv
= rsv
;
6593 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6595 BTRFS_EXTENT_DATA_KEY
);
6596 if (ret
!= -EAGAIN
) {
6601 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6602 ret
= btrfs_update_inode(trans
, root
, inode
);
6608 nr
= trans
->blocks_used
;
6609 btrfs_end_transaction(trans
, root
);
6611 btrfs_btree_balance_dirty(root
, nr
);
6614 if (ret
== 0 && inode
->i_nlink
> 0) {
6615 trans
->block_rsv
= root
->orphan_block_rsv
;
6616 ret
= btrfs_orphan_del(trans
, inode
);
6619 } else if (ret
&& inode
->i_nlink
> 0) {
6621 * Failed to do the truncate, remove us from the in memory
6624 ret
= btrfs_orphan_del(NULL
, inode
);
6627 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6628 ret
= btrfs_update_inode(trans
, root
, inode
);
6632 nr
= trans
->blocks_used
;
6633 ret
= btrfs_end_transaction_throttle(trans
, root
);
6634 btrfs_btree_balance_dirty(root
, nr
);
6637 btrfs_free_block_rsv(root
, rsv
);
6646 * create a new subvolume directory/inode (helper for the ioctl).
6648 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6649 struct btrfs_root
*new_root
, u64 new_dirid
)
6651 struct inode
*inode
;
6655 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6656 new_dirid
, S_IFDIR
| 0700, &index
);
6658 return PTR_ERR(inode
);
6659 inode
->i_op
= &btrfs_dir_inode_operations
;
6660 inode
->i_fop
= &btrfs_dir_file_operations
;
6663 btrfs_i_size_write(inode
, 0);
6665 err
= btrfs_update_inode(trans
, new_root
, inode
);
6672 /* helper function for file defrag and space balancing. This
6673 * forces readahead on a given range of bytes in an inode
6675 unsigned long btrfs_force_ra(struct address_space
*mapping
,
6676 struct file_ra_state
*ra
, struct file
*file
,
6677 pgoff_t offset
, pgoff_t last_index
)
6679 pgoff_t req_size
= last_index
- offset
+ 1;
6681 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
6682 return offset
+ req_size
;
6685 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6687 struct btrfs_inode
*ei
;
6688 struct inode
*inode
;
6690 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6695 ei
->space_info
= NULL
;
6699 ei
->last_sub_trans
= 0;
6700 ei
->logged_trans
= 0;
6701 ei
->delalloc_bytes
= 0;
6702 ei
->reserved_bytes
= 0;
6703 ei
->disk_i_size
= 0;
6705 ei
->index_cnt
= (u64
)-1;
6706 ei
->last_unlink_trans
= 0;
6708 spin_lock_init(&ei
->lock
);
6709 ei
->outstanding_extents
= 0;
6710 ei
->reserved_extents
= 0;
6712 ei
->ordered_data_close
= 0;
6713 ei
->orphan_meta_reserved
= 0;
6714 ei
->dummy_inode
= 0;
6716 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6718 ei
->delayed_node
= NULL
;
6720 inode
= &ei
->vfs_inode
;
6721 extent_map_tree_init(&ei
->extent_tree
);
6722 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
6723 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
6724 mutex_init(&ei
->log_mutex
);
6725 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6726 INIT_LIST_HEAD(&ei
->i_orphan
);
6727 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6728 INIT_LIST_HEAD(&ei
->ordered_operations
);
6729 RB_CLEAR_NODE(&ei
->rb_node
);
6734 static void btrfs_i_callback(struct rcu_head
*head
)
6736 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6737 INIT_LIST_HEAD(&inode
->i_dentry
);
6738 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6741 void btrfs_destroy_inode(struct inode
*inode
)
6743 struct btrfs_ordered_extent
*ordered
;
6744 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6746 WARN_ON(!list_empty(&inode
->i_dentry
));
6747 WARN_ON(inode
->i_data
.nrpages
);
6748 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
6749 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
6752 * This can happen where we create an inode, but somebody else also
6753 * created the same inode and we need to destroy the one we already
6760 * Make sure we're properly removed from the ordered operation
6764 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6765 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6766 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6767 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6770 spin_lock(&root
->orphan_lock
);
6771 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6772 printk(KERN_INFO
"BTRFS: inode %llu still on the orphan list\n",
6773 (unsigned long long)btrfs_ino(inode
));
6774 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6776 spin_unlock(&root
->orphan_lock
);
6779 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6783 printk(KERN_ERR
"btrfs found ordered "
6784 "extent %llu %llu on inode cleanup\n",
6785 (unsigned long long)ordered
->file_offset
,
6786 (unsigned long long)ordered
->len
);
6787 btrfs_remove_ordered_extent(inode
, ordered
);
6788 btrfs_put_ordered_extent(ordered
);
6789 btrfs_put_ordered_extent(ordered
);
6792 inode_tree_del(inode
);
6793 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6795 btrfs_remove_delayed_node(inode
);
6796 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6799 int btrfs_drop_inode(struct inode
*inode
)
6801 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6803 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6804 !btrfs_is_free_space_inode(root
, inode
))
6807 return generic_drop_inode(inode
);
6810 static void init_once(void *foo
)
6812 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6814 inode_init_once(&ei
->vfs_inode
);
6817 void btrfs_destroy_cachep(void)
6819 if (btrfs_inode_cachep
)
6820 kmem_cache_destroy(btrfs_inode_cachep
);
6821 if (btrfs_trans_handle_cachep
)
6822 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6823 if (btrfs_transaction_cachep
)
6824 kmem_cache_destroy(btrfs_transaction_cachep
);
6825 if (btrfs_path_cachep
)
6826 kmem_cache_destroy(btrfs_path_cachep
);
6827 if (btrfs_free_space_cachep
)
6828 kmem_cache_destroy(btrfs_free_space_cachep
);
6831 int btrfs_init_cachep(void)
6833 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6834 sizeof(struct btrfs_inode
), 0,
6835 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6836 if (!btrfs_inode_cachep
)
6839 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6840 sizeof(struct btrfs_trans_handle
), 0,
6841 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6842 if (!btrfs_trans_handle_cachep
)
6845 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6846 sizeof(struct btrfs_transaction
), 0,
6847 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6848 if (!btrfs_transaction_cachep
)
6851 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6852 sizeof(struct btrfs_path
), 0,
6853 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6854 if (!btrfs_path_cachep
)
6857 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6858 sizeof(struct btrfs_free_space
), 0,
6859 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6860 if (!btrfs_free_space_cachep
)
6865 btrfs_destroy_cachep();
6869 static int btrfs_getattr(struct vfsmount
*mnt
,
6870 struct dentry
*dentry
, struct kstat
*stat
)
6872 struct inode
*inode
= dentry
->d_inode
;
6873 generic_fillattr(inode
, stat
);
6874 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
6875 stat
->blksize
= PAGE_CACHE_SIZE
;
6876 stat
->blocks
= (inode_get_bytes(inode
) +
6877 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
6882 * If a file is moved, it will inherit the cow and compression flags of the new
6885 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
6887 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
6888 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
6890 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
6891 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
6893 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
6895 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
)
6896 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
6898 b_inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6901 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6902 struct inode
*new_dir
, struct dentry
*new_dentry
)
6904 struct btrfs_trans_handle
*trans
;
6905 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6906 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6907 struct inode
*new_inode
= new_dentry
->d_inode
;
6908 struct inode
*old_inode
= old_dentry
->d_inode
;
6909 struct timespec ctime
= CURRENT_TIME
;
6913 u64 old_ino
= btrfs_ino(old_inode
);
6915 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
6918 /* we only allow rename subvolume link between subvolumes */
6919 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
6922 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
6923 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
6926 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
6927 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
6930 * we're using rename to replace one file with another.
6931 * and the replacement file is large. Start IO on it now so
6932 * we don't add too much work to the end of the transaction
6934 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
6935 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
6936 filemap_flush(old_inode
->i_mapping
);
6938 /* close the racy window with snapshot create/destroy ioctl */
6939 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
6940 down_read(&root
->fs_info
->subvol_sem
);
6942 * We want to reserve the absolute worst case amount of items. So if
6943 * both inodes are subvols and we need to unlink them then that would
6944 * require 4 item modifications, but if they are both normal inodes it
6945 * would require 5 item modifications, so we'll assume their normal
6946 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6947 * should cover the worst case number of items we'll modify.
6949 trans
= btrfs_start_transaction(root
, 20);
6950 if (IS_ERR(trans
)) {
6951 ret
= PTR_ERR(trans
);
6956 btrfs_record_root_in_trans(trans
, dest
);
6958 ret
= btrfs_set_inode_index(new_dir
, &index
);
6962 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6963 /* force full log commit if subvolume involved. */
6964 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
6966 ret
= btrfs_insert_inode_ref(trans
, dest
,
6967 new_dentry
->d_name
.name
,
6968 new_dentry
->d_name
.len
,
6970 btrfs_ino(new_dir
), index
);
6974 * this is an ugly little race, but the rename is required
6975 * to make sure that if we crash, the inode is either at the
6976 * old name or the new one. pinning the log transaction lets
6977 * us make sure we don't allow a log commit to come in after
6978 * we unlink the name but before we add the new name back in.
6980 btrfs_pin_log_trans(root
);
6983 * make sure the inode gets flushed if it is replacing
6986 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
6987 btrfs_add_ordered_operation(trans
, root
, old_inode
);
6989 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
6990 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
6991 old_inode
->i_ctime
= ctime
;
6993 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
6994 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
6996 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6997 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
6998 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
6999 old_dentry
->d_name
.name
,
7000 old_dentry
->d_name
.len
);
7002 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7003 old_dentry
->d_inode
,
7004 old_dentry
->d_name
.name
,
7005 old_dentry
->d_name
.len
);
7007 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7012 new_inode
->i_ctime
= CURRENT_TIME
;
7013 if (unlikely(btrfs_ino(new_inode
) ==
7014 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7015 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7016 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7018 new_dentry
->d_name
.name
,
7019 new_dentry
->d_name
.len
);
7020 BUG_ON(new_inode
->i_nlink
== 0);
7022 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7023 new_dentry
->d_inode
,
7024 new_dentry
->d_name
.name
,
7025 new_dentry
->d_name
.len
);
7028 if (new_inode
->i_nlink
== 0) {
7029 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7034 fixup_inode_flags(new_dir
, old_inode
);
7036 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7037 new_dentry
->d_name
.name
,
7038 new_dentry
->d_name
.len
, 0, index
);
7041 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7042 struct dentry
*parent
= new_dentry
->d_parent
;
7043 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7044 btrfs_end_log_trans(root
);
7047 btrfs_end_transaction_throttle(trans
, root
);
7049 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7050 up_read(&root
->fs_info
->subvol_sem
);
7056 * some fairly slow code that needs optimization. This walks the list
7057 * of all the inodes with pending delalloc and forces them to disk.
7059 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7061 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7062 struct btrfs_inode
*binode
;
7063 struct inode
*inode
;
7065 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7068 spin_lock(&root
->fs_info
->delalloc_lock
);
7069 while (!list_empty(head
)) {
7070 binode
= list_entry(head
->next
, struct btrfs_inode
,
7072 inode
= igrab(&binode
->vfs_inode
);
7074 list_del_init(&binode
->delalloc_inodes
);
7075 spin_unlock(&root
->fs_info
->delalloc_lock
);
7077 filemap_flush(inode
->i_mapping
);
7079 btrfs_add_delayed_iput(inode
);
7084 spin_lock(&root
->fs_info
->delalloc_lock
);
7086 spin_unlock(&root
->fs_info
->delalloc_lock
);
7088 /* the filemap_flush will queue IO into the worker threads, but
7089 * we have to make sure the IO is actually started and that
7090 * ordered extents get created before we return
7092 atomic_inc(&root
->fs_info
->async_submit_draining
);
7093 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7094 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7095 wait_event(root
->fs_info
->async_submit_wait
,
7096 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7097 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7099 atomic_dec(&root
->fs_info
->async_submit_draining
);
7103 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7104 const char *symname
)
7106 struct btrfs_trans_handle
*trans
;
7107 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7108 struct btrfs_path
*path
;
7109 struct btrfs_key key
;
7110 struct inode
*inode
= NULL
;
7118 struct btrfs_file_extent_item
*ei
;
7119 struct extent_buffer
*leaf
;
7120 unsigned long nr
= 0;
7122 name_len
= strlen(symname
) + 1;
7123 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7124 return -ENAMETOOLONG
;
7127 * 2 items for inode item and ref
7128 * 2 items for dir items
7129 * 1 item for xattr if selinux is on
7131 trans
= btrfs_start_transaction(root
, 5);
7133 return PTR_ERR(trans
);
7135 err
= btrfs_find_free_ino(root
, &objectid
);
7139 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7140 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
7141 S_IFLNK
|S_IRWXUGO
, &index
);
7142 if (IS_ERR(inode
)) {
7143 err
= PTR_ERR(inode
);
7147 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7153 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7157 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7158 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7159 inode
->i_fop
= &btrfs_file_operations
;
7160 inode
->i_op
= &btrfs_file_inode_operations
;
7161 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7166 path
= btrfs_alloc_path();
7168 key
.objectid
= btrfs_ino(inode
);
7170 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7171 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7172 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7176 btrfs_free_path(path
);
7179 leaf
= path
->nodes
[0];
7180 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7181 struct btrfs_file_extent_item
);
7182 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7183 btrfs_set_file_extent_type(leaf
, ei
,
7184 BTRFS_FILE_EXTENT_INLINE
);
7185 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7186 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7187 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7188 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7190 ptr
= btrfs_file_extent_inline_start(ei
);
7191 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7192 btrfs_mark_buffer_dirty(leaf
);
7193 btrfs_free_path(path
);
7195 inode
->i_op
= &btrfs_symlink_inode_operations
;
7196 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7197 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7198 inode_set_bytes(inode
, name_len
);
7199 btrfs_i_size_write(inode
, name_len
- 1);
7200 err
= btrfs_update_inode(trans
, root
, inode
);
7205 nr
= trans
->blocks_used
;
7206 btrfs_end_transaction_throttle(trans
, root
);
7208 inode_dec_link_count(inode
);
7211 btrfs_btree_balance_dirty(root
, nr
);
7215 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7216 u64 start
, u64 num_bytes
, u64 min_size
,
7217 loff_t actual_len
, u64
*alloc_hint
,
7218 struct btrfs_trans_handle
*trans
)
7220 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7221 struct btrfs_key ins
;
7222 u64 cur_offset
= start
;
7225 bool own_trans
= true;
7229 while (num_bytes
> 0) {
7231 trans
= btrfs_start_transaction(root
, 3);
7232 if (IS_ERR(trans
)) {
7233 ret
= PTR_ERR(trans
);
7238 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7239 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7242 btrfs_end_transaction(trans
, root
);
7246 ret
= insert_reserved_file_extent(trans
, inode
,
7247 cur_offset
, ins
.objectid
,
7248 ins
.offset
, ins
.offset
,
7249 ins
.offset
, 0, 0, 0,
7250 BTRFS_FILE_EXTENT_PREALLOC
);
7252 btrfs_drop_extent_cache(inode
, cur_offset
,
7253 cur_offset
+ ins
.offset
-1, 0);
7255 num_bytes
-= ins
.offset
;
7256 cur_offset
+= ins
.offset
;
7257 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7259 inode
->i_ctime
= CURRENT_TIME
;
7260 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7261 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7262 (actual_len
> inode
->i_size
) &&
7263 (cur_offset
> inode
->i_size
)) {
7264 if (cur_offset
> actual_len
)
7265 i_size
= actual_len
;
7267 i_size
= cur_offset
;
7268 i_size_write(inode
, i_size
);
7269 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7272 ret
= btrfs_update_inode(trans
, root
, inode
);
7276 btrfs_end_transaction(trans
, root
);
7281 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7282 u64 start
, u64 num_bytes
, u64 min_size
,
7283 loff_t actual_len
, u64
*alloc_hint
)
7285 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7286 min_size
, actual_len
, alloc_hint
,
7290 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7291 struct btrfs_trans_handle
*trans
, int mode
,
7292 u64 start
, u64 num_bytes
, u64 min_size
,
7293 loff_t actual_len
, u64
*alloc_hint
)
7295 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7296 min_size
, actual_len
, alloc_hint
, trans
);
7299 static int btrfs_set_page_dirty(struct page
*page
)
7301 return __set_page_dirty_nobuffers(page
);
7304 static int btrfs_permission(struct inode
*inode
, int mask
)
7306 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7308 if (btrfs_root_readonly(root
) && (mask
& MAY_WRITE
))
7310 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
7312 return generic_permission(inode
, mask
);
7315 static const struct inode_operations btrfs_dir_inode_operations
= {
7316 .getattr
= btrfs_getattr
,
7317 .lookup
= btrfs_lookup
,
7318 .create
= btrfs_create
,
7319 .unlink
= btrfs_unlink
,
7321 .mkdir
= btrfs_mkdir
,
7322 .rmdir
= btrfs_rmdir
,
7323 .rename
= btrfs_rename
,
7324 .symlink
= btrfs_symlink
,
7325 .setattr
= btrfs_setattr
,
7326 .mknod
= btrfs_mknod
,
7327 .setxattr
= btrfs_setxattr
,
7328 .getxattr
= btrfs_getxattr
,
7329 .listxattr
= btrfs_listxattr
,
7330 .removexattr
= btrfs_removexattr
,
7331 .permission
= btrfs_permission
,
7332 .get_acl
= btrfs_get_acl
,
7334 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7335 .lookup
= btrfs_lookup
,
7336 .permission
= btrfs_permission
,
7337 .get_acl
= btrfs_get_acl
,
7340 static const struct file_operations btrfs_dir_file_operations
= {
7341 .llseek
= generic_file_llseek
,
7342 .read
= generic_read_dir
,
7343 .readdir
= btrfs_real_readdir
,
7344 .unlocked_ioctl
= btrfs_ioctl
,
7345 #ifdef CONFIG_COMPAT
7346 .compat_ioctl
= btrfs_ioctl
,
7348 .release
= btrfs_release_file
,
7349 .fsync
= btrfs_sync_file
,
7352 static struct extent_io_ops btrfs_extent_io_ops
= {
7353 .fill_delalloc
= run_delalloc_range
,
7354 .submit_bio_hook
= btrfs_submit_bio_hook
,
7355 .merge_bio_hook
= btrfs_merge_bio_hook
,
7356 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7357 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7358 .writepage_start_hook
= btrfs_writepage_start_hook
,
7359 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
7360 .set_bit_hook
= btrfs_set_bit_hook
,
7361 .clear_bit_hook
= btrfs_clear_bit_hook
,
7362 .merge_extent_hook
= btrfs_merge_extent_hook
,
7363 .split_extent_hook
= btrfs_split_extent_hook
,
7367 * btrfs doesn't support the bmap operation because swapfiles
7368 * use bmap to make a mapping of extents in the file. They assume
7369 * these extents won't change over the life of the file and they
7370 * use the bmap result to do IO directly to the drive.
7372 * the btrfs bmap call would return logical addresses that aren't
7373 * suitable for IO and they also will change frequently as COW
7374 * operations happen. So, swapfile + btrfs == corruption.
7376 * For now we're avoiding this by dropping bmap.
7378 static const struct address_space_operations btrfs_aops
= {
7379 .readpage
= btrfs_readpage
,
7380 .writepage
= btrfs_writepage
,
7381 .writepages
= btrfs_writepages
,
7382 .readpages
= btrfs_readpages
,
7383 .direct_IO
= btrfs_direct_IO
,
7384 .invalidatepage
= btrfs_invalidatepage
,
7385 .releasepage
= btrfs_releasepage
,
7386 .set_page_dirty
= btrfs_set_page_dirty
,
7387 .error_remove_page
= generic_error_remove_page
,
7390 static const struct address_space_operations btrfs_symlink_aops
= {
7391 .readpage
= btrfs_readpage
,
7392 .writepage
= btrfs_writepage
,
7393 .invalidatepage
= btrfs_invalidatepage
,
7394 .releasepage
= btrfs_releasepage
,
7397 static const struct inode_operations btrfs_file_inode_operations
= {
7398 .getattr
= btrfs_getattr
,
7399 .setattr
= btrfs_setattr
,
7400 .setxattr
= btrfs_setxattr
,
7401 .getxattr
= btrfs_getxattr
,
7402 .listxattr
= btrfs_listxattr
,
7403 .removexattr
= btrfs_removexattr
,
7404 .permission
= btrfs_permission
,
7405 .fiemap
= btrfs_fiemap
,
7406 .get_acl
= btrfs_get_acl
,
7408 static const struct inode_operations btrfs_special_inode_operations
= {
7409 .getattr
= btrfs_getattr
,
7410 .setattr
= btrfs_setattr
,
7411 .permission
= btrfs_permission
,
7412 .setxattr
= btrfs_setxattr
,
7413 .getxattr
= btrfs_getxattr
,
7414 .listxattr
= btrfs_listxattr
,
7415 .removexattr
= btrfs_removexattr
,
7416 .get_acl
= btrfs_get_acl
,
7418 static const struct inode_operations btrfs_symlink_inode_operations
= {
7419 .readlink
= generic_readlink
,
7420 .follow_link
= page_follow_link_light
,
7421 .put_link
= page_put_link
,
7422 .getattr
= btrfs_getattr
,
7423 .permission
= btrfs_permission
,
7424 .setxattr
= btrfs_setxattr
,
7425 .getxattr
= btrfs_getxattr
,
7426 .listxattr
= btrfs_listxattr
,
7427 .removexattr
= btrfs_removexattr
,
7428 .get_acl
= btrfs_get_acl
,
7431 const struct dentry_operations btrfs_dentry_operations
= {
7432 .d_delete
= btrfs_dentry_delete
,