2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args
{
57 struct btrfs_root
*root
;
60 static const struct inode_operations btrfs_dir_inode_operations
;
61 static const struct inode_operations btrfs_symlink_inode_operations
;
62 static const struct inode_operations btrfs_dir_ro_inode_operations
;
63 static const struct inode_operations btrfs_special_inode_operations
;
64 static const struct inode_operations btrfs_file_inode_operations
;
65 static const struct address_space_operations btrfs_aops
;
66 static const struct address_space_operations btrfs_symlink_aops
;
67 static const struct file_operations btrfs_dir_file_operations
;
68 static struct extent_io_ops btrfs_extent_io_ops
;
70 static struct kmem_cache
*btrfs_inode_cachep
;
71 struct kmem_cache
*btrfs_trans_handle_cachep
;
72 struct kmem_cache
*btrfs_transaction_cachep
;
73 struct kmem_cache
*btrfs_path_cachep
;
74 struct kmem_cache
*btrfs_free_space_cachep
;
77 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
78 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
79 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
80 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
81 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
82 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
83 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
84 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
87 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
);
88 static int btrfs_truncate(struct inode
*inode
);
89 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
90 static noinline
int cow_file_range(struct inode
*inode
,
91 struct page
*locked_page
,
92 u64 start
, u64 end
, int *page_started
,
93 unsigned long *nr_written
, int unlock
);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
96 struct inode
*inode
, struct inode
*dir
)
100 err
= btrfs_init_acl(trans
, inode
, dir
);
102 err
= btrfs_xattr_security_init(trans
, inode
, dir
);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
112 struct btrfs_root
*root
, struct inode
*inode
,
113 u64 start
, size_t size
, size_t compressed_size
,
114 struct page
**compressed_pages
)
116 struct btrfs_key key
;
117 struct btrfs_path
*path
;
118 struct extent_buffer
*leaf
;
119 struct page
*page
= NULL
;
122 struct btrfs_file_extent_item
*ei
;
125 size_t cur_size
= size
;
127 unsigned long offset
;
128 int compress_type
= BTRFS_COMPRESS_NONE
;
130 if (compressed_size
&& compressed_pages
) {
131 compress_type
= root
->fs_info
->compress_type
;
132 cur_size
= compressed_size
;
135 path
= btrfs_alloc_path();
139 path
->leave_spinning
= 1;
140 btrfs_set_trans_block_group(trans
, inode
);
142 key
.objectid
= inode
->i_ino
;
144 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
145 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
147 inode_add_bytes(inode
, size
);
148 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
155 leaf
= path
->nodes
[0];
156 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
157 struct btrfs_file_extent_item
);
158 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
159 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
160 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
161 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
162 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
163 ptr
= btrfs_file_extent_inline_start(ei
);
165 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
168 while (compressed_size
> 0) {
169 cpage
= compressed_pages
[i
];
170 cur_size
= min_t(unsigned long, compressed_size
,
173 kaddr
= kmap_atomic(cpage
, KM_USER0
);
174 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
175 kunmap_atomic(kaddr
, KM_USER0
);
179 compressed_size
-= cur_size
;
181 btrfs_set_file_extent_compression(leaf
, ei
,
184 page
= find_get_page(inode
->i_mapping
,
185 start
>> PAGE_CACHE_SHIFT
);
186 btrfs_set_file_extent_compression(leaf
, ei
, 0);
187 kaddr
= kmap_atomic(page
, KM_USER0
);
188 offset
= start
& (PAGE_CACHE_SIZE
- 1);
189 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
190 kunmap_atomic(kaddr
, KM_USER0
);
191 page_cache_release(page
);
193 btrfs_mark_buffer_dirty(leaf
);
194 btrfs_free_path(path
);
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
205 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
206 btrfs_update_inode(trans
, root
, inode
);
210 btrfs_free_path(path
);
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
220 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
221 struct btrfs_root
*root
,
222 struct inode
*inode
, u64 start
, u64 end
,
223 size_t compressed_size
,
224 struct page
**compressed_pages
)
226 u64 isize
= i_size_read(inode
);
227 u64 actual_end
= min(end
+ 1, isize
);
228 u64 inline_len
= actual_end
- start
;
229 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
230 ~((u64
)root
->sectorsize
- 1);
232 u64 data_len
= inline_len
;
236 data_len
= compressed_size
;
239 actual_end
>= PAGE_CACHE_SIZE
||
240 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
242 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
244 data_len
> root
->fs_info
->max_inline
) {
248 ret
= btrfs_drop_extents(trans
, inode
, start
, aligned_end
,
252 if (isize
> actual_end
)
253 inline_len
= min_t(u64
, isize
, actual_end
);
254 ret
= insert_inline_extent(trans
, root
, inode
, start
,
255 inline_len
, compressed_size
,
258 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
259 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
263 struct async_extent
{
268 unsigned long nr_pages
;
270 struct list_head list
;
275 struct btrfs_root
*root
;
276 struct page
*locked_page
;
279 struct list_head extents
;
280 struct btrfs_work work
;
283 static noinline
int add_async_extent(struct async_cow
*cow
,
284 u64 start
, u64 ram_size
,
287 unsigned long nr_pages
,
290 struct async_extent
*async_extent
;
292 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
293 async_extent
->start
= start
;
294 async_extent
->ram_size
= ram_size
;
295 async_extent
->compressed_size
= compressed_size
;
296 async_extent
->pages
= pages
;
297 async_extent
->nr_pages
= nr_pages
;
298 async_extent
->compress_type
= compress_type
;
299 list_add_tail(&async_extent
->list
, &cow
->extents
);
304 * we create compressed extents in two phases. The first
305 * phase compresses a range of pages that have already been
306 * locked (both pages and state bits are locked).
308 * This is done inside an ordered work queue, and the compression
309 * is spread across many cpus. The actual IO submission is step
310 * two, and the ordered work queue takes care of making sure that
311 * happens in the same order things were put onto the queue by
312 * writepages and friends.
314 * If this code finds it can't get good compression, it puts an
315 * entry onto the work queue to write the uncompressed bytes. This
316 * makes sure that both compressed inodes and uncompressed inodes
317 * are written in the same order that pdflush sent them down.
319 static noinline
int compress_file_range(struct inode
*inode
,
320 struct page
*locked_page
,
322 struct async_cow
*async_cow
,
325 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
326 struct btrfs_trans_handle
*trans
;
328 u64 blocksize
= root
->sectorsize
;
330 u64 isize
= i_size_read(inode
);
332 struct page
**pages
= NULL
;
333 unsigned long nr_pages
;
334 unsigned long nr_pages_ret
= 0;
335 unsigned long total_compressed
= 0;
336 unsigned long total_in
= 0;
337 unsigned long max_compressed
= 128 * 1024;
338 unsigned long max_uncompressed
= 128 * 1024;
341 int compress_type
= root
->fs_info
->compress_type
;
343 actual_end
= min_t(u64
, isize
, end
+ 1);
346 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
347 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
350 * we don't want to send crud past the end of i_size through
351 * compression, that's just a waste of CPU time. So, if the
352 * end of the file is before the start of our current
353 * requested range of bytes, we bail out to the uncompressed
354 * cleanup code that can deal with all of this.
356 * It isn't really the fastest way to fix things, but this is a
357 * very uncommon corner.
359 if (actual_end
<= start
)
360 goto cleanup_and_bail_uncompressed
;
362 total_compressed
= actual_end
- start
;
364 /* we want to make sure that amount of ram required to uncompress
365 * an extent is reasonable, so we limit the total size in ram
366 * of a compressed extent to 128k. This is a crucial number
367 * because it also controls how easily we can spread reads across
368 * cpus for decompression.
370 * We also want to make sure the amount of IO required to do
371 * a random read is reasonably small, so we limit the size of
372 * a compressed extent to 128k.
374 total_compressed
= min(total_compressed
, max_uncompressed
);
375 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
376 num_bytes
= max(blocksize
, num_bytes
);
381 * we do compression for mount -o compress and when the
382 * inode has not been flagged as nocompress. This flag can
383 * change at any time if we discover bad compression ratios.
385 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
386 (btrfs_test_opt(root
, COMPRESS
) ||
387 (BTRFS_I(inode
)->force_compress
))) {
389 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
391 if (BTRFS_I(inode
)->force_compress
)
392 compress_type
= BTRFS_I(inode
)->force_compress
;
394 ret
= btrfs_compress_pages(compress_type
,
395 inode
->i_mapping
, start
,
396 total_compressed
, pages
,
397 nr_pages
, &nr_pages_ret
,
403 unsigned long offset
= total_compressed
&
404 (PAGE_CACHE_SIZE
- 1);
405 struct page
*page
= pages
[nr_pages_ret
- 1];
408 /* zero the tail end of the last page, we might be
409 * sending it down to disk
412 kaddr
= kmap_atomic(page
, KM_USER0
);
413 memset(kaddr
+ offset
, 0,
414 PAGE_CACHE_SIZE
- offset
);
415 kunmap_atomic(kaddr
, KM_USER0
);
421 trans
= btrfs_join_transaction(root
, 1);
422 BUG_ON(IS_ERR(trans
));
423 btrfs_set_trans_block_group(trans
, inode
);
424 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
426 /* lets try to make an inline extent */
427 if (ret
|| total_in
< (actual_end
- start
)) {
428 /* we didn't compress the entire range, try
429 * to make an uncompressed inline extent.
431 ret
= cow_file_range_inline(trans
, root
, inode
,
432 start
, end
, 0, NULL
);
434 /* try making a compressed inline extent */
435 ret
= cow_file_range_inline(trans
, root
, inode
,
437 total_compressed
, pages
);
441 * inline extent creation worked, we don't need
442 * to create any more async work items. Unlock
443 * and free up our temp pages.
445 extent_clear_unlock_delalloc(inode
,
446 &BTRFS_I(inode
)->io_tree
,
448 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
449 EXTENT_CLEAR_DELALLOC
|
450 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
452 btrfs_end_transaction(trans
, root
);
455 btrfs_end_transaction(trans
, root
);
460 * we aren't doing an inline extent round the compressed size
461 * up to a block size boundary so the allocator does sane
464 total_compressed
= (total_compressed
+ blocksize
- 1) &
468 * one last check to make sure the compression is really a
469 * win, compare the page count read with the blocks on disk
471 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
472 ~(PAGE_CACHE_SIZE
- 1);
473 if (total_compressed
>= total_in
) {
476 num_bytes
= total_in
;
479 if (!will_compress
&& pages
) {
481 * the compression code ran but failed to make things smaller,
482 * free any pages it allocated and our page pointer array
484 for (i
= 0; i
< nr_pages_ret
; i
++) {
485 WARN_ON(pages
[i
]->mapping
);
486 page_cache_release(pages
[i
]);
490 total_compressed
= 0;
493 /* flag the file so we don't compress in the future */
494 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
495 !(BTRFS_I(inode
)->force_compress
)) {
496 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
502 /* the async work queues will take care of doing actual
503 * allocation on disk for these compressed pages,
504 * and will submit them to the elevator.
506 add_async_extent(async_cow
, start
, num_bytes
,
507 total_compressed
, pages
, nr_pages_ret
,
510 if (start
+ num_bytes
< end
) {
517 cleanup_and_bail_uncompressed
:
519 * No compression, but we still need to write the pages in
520 * the file we've been given so far. redirty the locked
521 * page if it corresponds to our extent and set things up
522 * for the async work queue to run cow_file_range to do
523 * the normal delalloc dance
525 if (page_offset(locked_page
) >= start
&&
526 page_offset(locked_page
) <= end
) {
527 __set_page_dirty_nobuffers(locked_page
);
528 /* unlocked later on in the async handlers */
530 add_async_extent(async_cow
, start
, end
- start
+ 1,
531 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
539 for (i
= 0; i
< nr_pages_ret
; i
++) {
540 WARN_ON(pages
[i
]->mapping
);
541 page_cache_release(pages
[i
]);
549 * phase two of compressed writeback. This is the ordered portion
550 * of the code, which only gets called in the order the work was
551 * queued. We walk all the async extents created by compress_file_range
552 * and send them down to the disk.
554 static noinline
int submit_compressed_extents(struct inode
*inode
,
555 struct async_cow
*async_cow
)
557 struct async_extent
*async_extent
;
559 struct btrfs_trans_handle
*trans
;
560 struct btrfs_key ins
;
561 struct extent_map
*em
;
562 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
563 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
564 struct extent_io_tree
*io_tree
;
567 if (list_empty(&async_cow
->extents
))
571 while (!list_empty(&async_cow
->extents
)) {
572 async_extent
= list_entry(async_cow
->extents
.next
,
573 struct async_extent
, list
);
574 list_del(&async_extent
->list
);
576 io_tree
= &BTRFS_I(inode
)->io_tree
;
579 /* did the compression code fall back to uncompressed IO? */
580 if (!async_extent
->pages
) {
581 int page_started
= 0;
582 unsigned long nr_written
= 0;
584 lock_extent(io_tree
, async_extent
->start
,
585 async_extent
->start
+
586 async_extent
->ram_size
- 1, GFP_NOFS
);
588 /* allocate blocks */
589 ret
= cow_file_range(inode
, async_cow
->locked_page
,
591 async_extent
->start
+
592 async_extent
->ram_size
- 1,
593 &page_started
, &nr_written
, 0);
596 * if page_started, cow_file_range inserted an
597 * inline extent and took care of all the unlocking
598 * and IO for us. Otherwise, we need to submit
599 * all those pages down to the drive.
601 if (!page_started
&& !ret
)
602 extent_write_locked_range(io_tree
,
603 inode
, async_extent
->start
,
604 async_extent
->start
+
605 async_extent
->ram_size
- 1,
613 lock_extent(io_tree
, async_extent
->start
,
614 async_extent
->start
+ async_extent
->ram_size
- 1,
617 trans
= btrfs_join_transaction(root
, 1);
618 BUG_ON(IS_ERR(trans
));
619 ret
= btrfs_reserve_extent(trans
, root
,
620 async_extent
->compressed_size
,
621 async_extent
->compressed_size
,
624 btrfs_end_transaction(trans
, root
);
628 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
629 WARN_ON(async_extent
->pages
[i
]->mapping
);
630 page_cache_release(async_extent
->pages
[i
]);
632 kfree(async_extent
->pages
);
633 async_extent
->nr_pages
= 0;
634 async_extent
->pages
= NULL
;
635 unlock_extent(io_tree
, async_extent
->start
,
636 async_extent
->start
+
637 async_extent
->ram_size
- 1, GFP_NOFS
);
642 * here we're doing allocation and writeback of the
645 btrfs_drop_extent_cache(inode
, async_extent
->start
,
646 async_extent
->start
+
647 async_extent
->ram_size
- 1, 0);
649 em
= alloc_extent_map(GFP_NOFS
);
651 em
->start
= async_extent
->start
;
652 em
->len
= async_extent
->ram_size
;
653 em
->orig_start
= em
->start
;
655 em
->block_start
= ins
.objectid
;
656 em
->block_len
= ins
.offset
;
657 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
658 em
->compress_type
= async_extent
->compress_type
;
659 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
660 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
663 write_lock(&em_tree
->lock
);
664 ret
= add_extent_mapping(em_tree
, em
);
665 write_unlock(&em_tree
->lock
);
666 if (ret
!= -EEXIST
) {
670 btrfs_drop_extent_cache(inode
, async_extent
->start
,
671 async_extent
->start
+
672 async_extent
->ram_size
- 1, 0);
675 ret
= btrfs_add_ordered_extent_compress(inode
,
678 async_extent
->ram_size
,
680 BTRFS_ORDERED_COMPRESSED
,
681 async_extent
->compress_type
);
685 * clear dirty, set writeback and unlock the pages.
687 extent_clear_unlock_delalloc(inode
,
688 &BTRFS_I(inode
)->io_tree
,
690 async_extent
->start
+
691 async_extent
->ram_size
- 1,
692 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
693 EXTENT_CLEAR_UNLOCK
|
694 EXTENT_CLEAR_DELALLOC
|
695 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
697 ret
= btrfs_submit_compressed_write(inode
,
699 async_extent
->ram_size
,
701 ins
.offset
, async_extent
->pages
,
702 async_extent
->nr_pages
);
705 alloc_hint
= ins
.objectid
+ ins
.offset
;
713 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
716 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
717 struct extent_map
*em
;
720 read_lock(&em_tree
->lock
);
721 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
724 * if block start isn't an actual block number then find the
725 * first block in this inode and use that as a hint. If that
726 * block is also bogus then just don't worry about it.
728 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
730 em
= search_extent_mapping(em_tree
, 0, 0);
731 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
732 alloc_hint
= em
->block_start
;
736 alloc_hint
= em
->block_start
;
740 read_unlock(&em_tree
->lock
);
746 * when extent_io.c finds a delayed allocation range in the file,
747 * the call backs end up in this code. The basic idea is to
748 * allocate extents on disk for the range, and create ordered data structs
749 * in ram to track those extents.
751 * locked_page is the page that writepage had locked already. We use
752 * it to make sure we don't do extra locks or unlocks.
754 * *page_started is set to one if we unlock locked_page and do everything
755 * required to start IO on it. It may be clean and already done with
758 static noinline
int cow_file_range(struct inode
*inode
,
759 struct page
*locked_page
,
760 u64 start
, u64 end
, int *page_started
,
761 unsigned long *nr_written
,
764 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
765 struct btrfs_trans_handle
*trans
;
768 unsigned long ram_size
;
771 u64 blocksize
= root
->sectorsize
;
772 struct btrfs_key ins
;
773 struct extent_map
*em
;
774 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
777 BUG_ON(root
== root
->fs_info
->tree_root
);
778 trans
= btrfs_join_transaction(root
, 1);
779 BUG_ON(IS_ERR(trans
));
780 btrfs_set_trans_block_group(trans
, inode
);
781 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
783 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
784 num_bytes
= max(blocksize
, num_bytes
);
785 disk_num_bytes
= num_bytes
;
789 /* lets try to make an inline extent */
790 ret
= cow_file_range_inline(trans
, root
, inode
,
791 start
, end
, 0, NULL
);
793 extent_clear_unlock_delalloc(inode
,
794 &BTRFS_I(inode
)->io_tree
,
796 EXTENT_CLEAR_UNLOCK_PAGE
|
797 EXTENT_CLEAR_UNLOCK
|
798 EXTENT_CLEAR_DELALLOC
|
800 EXTENT_SET_WRITEBACK
|
801 EXTENT_END_WRITEBACK
);
803 *nr_written
= *nr_written
+
804 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
811 BUG_ON(disk_num_bytes
>
812 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
814 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
815 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
817 while (disk_num_bytes
> 0) {
820 cur_alloc_size
= disk_num_bytes
;
821 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
822 root
->sectorsize
, 0, alloc_hint
,
826 em
= alloc_extent_map(GFP_NOFS
);
829 em
->orig_start
= em
->start
;
830 ram_size
= ins
.offset
;
831 em
->len
= ins
.offset
;
833 em
->block_start
= ins
.objectid
;
834 em
->block_len
= ins
.offset
;
835 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
836 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
839 write_lock(&em_tree
->lock
);
840 ret
= add_extent_mapping(em_tree
, em
);
841 write_unlock(&em_tree
->lock
);
842 if (ret
!= -EEXIST
) {
846 btrfs_drop_extent_cache(inode
, start
,
847 start
+ ram_size
- 1, 0);
850 cur_alloc_size
= ins
.offset
;
851 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
852 ram_size
, cur_alloc_size
, 0);
855 if (root
->root_key
.objectid
==
856 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
857 ret
= btrfs_reloc_clone_csums(inode
, start
,
862 if (disk_num_bytes
< cur_alloc_size
)
865 /* we're not doing compressed IO, don't unlock the first
866 * page (which the caller expects to stay locked), don't
867 * clear any dirty bits and don't set any writeback bits
869 * Do set the Private2 bit so we know this page was properly
870 * setup for writepage
872 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
873 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
876 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
877 start
, start
+ ram_size
- 1,
879 disk_num_bytes
-= cur_alloc_size
;
880 num_bytes
-= cur_alloc_size
;
881 alloc_hint
= ins
.objectid
+ ins
.offset
;
882 start
+= cur_alloc_size
;
886 btrfs_end_transaction(trans
, root
);
892 * work queue call back to started compression on a file and pages
894 static noinline
void async_cow_start(struct btrfs_work
*work
)
896 struct async_cow
*async_cow
;
898 async_cow
= container_of(work
, struct async_cow
, work
);
900 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
901 async_cow
->start
, async_cow
->end
, async_cow
,
904 async_cow
->inode
= NULL
;
908 * work queue call back to submit previously compressed pages
910 static noinline
void async_cow_submit(struct btrfs_work
*work
)
912 struct async_cow
*async_cow
;
913 struct btrfs_root
*root
;
914 unsigned long nr_pages
;
916 async_cow
= container_of(work
, struct async_cow
, work
);
918 root
= async_cow
->root
;
919 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
922 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
924 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
926 waitqueue_active(&root
->fs_info
->async_submit_wait
))
927 wake_up(&root
->fs_info
->async_submit_wait
);
929 if (async_cow
->inode
)
930 submit_compressed_extents(async_cow
->inode
, async_cow
);
933 static noinline
void async_cow_free(struct btrfs_work
*work
)
935 struct async_cow
*async_cow
;
936 async_cow
= container_of(work
, struct async_cow
, work
);
940 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
941 u64 start
, u64 end
, int *page_started
,
942 unsigned long *nr_written
)
944 struct async_cow
*async_cow
;
945 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
946 unsigned long nr_pages
;
948 int limit
= 10 * 1024 * 1042;
950 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
951 1, 0, NULL
, GFP_NOFS
);
952 while (start
< end
) {
953 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
954 async_cow
->inode
= inode
;
955 async_cow
->root
= root
;
956 async_cow
->locked_page
= locked_page
;
957 async_cow
->start
= start
;
959 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
962 cur_end
= min(end
, start
+ 512 * 1024 - 1);
964 async_cow
->end
= cur_end
;
965 INIT_LIST_HEAD(&async_cow
->extents
);
967 async_cow
->work
.func
= async_cow_start
;
968 async_cow
->work
.ordered_func
= async_cow_submit
;
969 async_cow
->work
.ordered_free
= async_cow_free
;
970 async_cow
->work
.flags
= 0;
972 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
974 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
976 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
979 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
980 wait_event(root
->fs_info
->async_submit_wait
,
981 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
985 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
986 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
987 wait_event(root
->fs_info
->async_submit_wait
,
988 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
992 *nr_written
+= nr_pages
;
999 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1000 u64 bytenr
, u64 num_bytes
)
1003 struct btrfs_ordered_sum
*sums
;
1006 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1007 bytenr
+ num_bytes
- 1, &list
);
1008 if (ret
== 0 && list_empty(&list
))
1011 while (!list_empty(&list
)) {
1012 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1013 list_del(&sums
->list
);
1020 * when nowcow writeback call back. This checks for snapshots or COW copies
1021 * of the extents that exist in the file, and COWs the file as required.
1023 * If no cow copies or snapshots exist, we write directly to the existing
1026 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1027 struct page
*locked_page
,
1028 u64 start
, u64 end
, int *page_started
, int force
,
1029 unsigned long *nr_written
)
1031 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1032 struct btrfs_trans_handle
*trans
;
1033 struct extent_buffer
*leaf
;
1034 struct btrfs_path
*path
;
1035 struct btrfs_file_extent_item
*fi
;
1036 struct btrfs_key found_key
;
1048 bool nolock
= false;
1050 path
= btrfs_alloc_path();
1052 if (root
== root
->fs_info
->tree_root
) {
1054 trans
= btrfs_join_transaction_nolock(root
, 1);
1056 trans
= btrfs_join_transaction(root
, 1);
1058 BUG_ON(IS_ERR(trans
));
1060 cow_start
= (u64
)-1;
1063 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
1066 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1067 leaf
= path
->nodes
[0];
1068 btrfs_item_key_to_cpu(leaf
, &found_key
,
1069 path
->slots
[0] - 1);
1070 if (found_key
.objectid
== inode
->i_ino
&&
1071 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1076 leaf
= path
->nodes
[0];
1077 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1078 ret
= btrfs_next_leaf(root
, path
);
1083 leaf
= path
->nodes
[0];
1089 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1091 if (found_key
.objectid
> inode
->i_ino
||
1092 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1093 found_key
.offset
> end
)
1096 if (found_key
.offset
> cur_offset
) {
1097 extent_end
= found_key
.offset
;
1102 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1103 struct btrfs_file_extent_item
);
1104 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1106 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1107 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1108 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1109 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1110 extent_end
= found_key
.offset
+
1111 btrfs_file_extent_num_bytes(leaf
, fi
);
1112 if (extent_end
<= start
) {
1116 if (disk_bytenr
== 0)
1118 if (btrfs_file_extent_compression(leaf
, fi
) ||
1119 btrfs_file_extent_encryption(leaf
, fi
) ||
1120 btrfs_file_extent_other_encoding(leaf
, fi
))
1122 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1124 if (btrfs_extent_readonly(root
, disk_bytenr
))
1126 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1128 extent_offset
, disk_bytenr
))
1130 disk_bytenr
+= extent_offset
;
1131 disk_bytenr
+= cur_offset
- found_key
.offset
;
1132 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1134 * force cow if csum exists in the range.
1135 * this ensure that csum for a given extent are
1136 * either valid or do not exist.
1138 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1141 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1142 extent_end
= found_key
.offset
+
1143 btrfs_file_extent_inline_len(leaf
, fi
);
1144 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1149 if (extent_end
<= start
) {
1154 if (cow_start
== (u64
)-1)
1155 cow_start
= cur_offset
;
1156 cur_offset
= extent_end
;
1157 if (cur_offset
> end
)
1163 btrfs_release_path(root
, path
);
1164 if (cow_start
!= (u64
)-1) {
1165 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1166 found_key
.offset
- 1, page_started
,
1169 cow_start
= (u64
)-1;
1172 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1173 struct extent_map
*em
;
1174 struct extent_map_tree
*em_tree
;
1175 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1176 em
= alloc_extent_map(GFP_NOFS
);
1178 em
->start
= cur_offset
;
1179 em
->orig_start
= em
->start
;
1180 em
->len
= num_bytes
;
1181 em
->block_len
= num_bytes
;
1182 em
->block_start
= disk_bytenr
;
1183 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1184 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1186 write_lock(&em_tree
->lock
);
1187 ret
= add_extent_mapping(em_tree
, em
);
1188 write_unlock(&em_tree
->lock
);
1189 if (ret
!= -EEXIST
) {
1190 free_extent_map(em
);
1193 btrfs_drop_extent_cache(inode
, em
->start
,
1194 em
->start
+ em
->len
- 1, 0);
1196 type
= BTRFS_ORDERED_PREALLOC
;
1198 type
= BTRFS_ORDERED_NOCOW
;
1201 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1202 num_bytes
, num_bytes
, type
);
1205 if (root
->root_key
.objectid
==
1206 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1207 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1212 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1213 cur_offset
, cur_offset
+ num_bytes
- 1,
1214 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1215 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1216 EXTENT_SET_PRIVATE2
);
1217 cur_offset
= extent_end
;
1218 if (cur_offset
> end
)
1221 btrfs_release_path(root
, path
);
1223 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1224 cow_start
= cur_offset
;
1225 if (cow_start
!= (u64
)-1) {
1226 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1227 page_started
, nr_written
, 1);
1232 ret
= btrfs_end_transaction_nolock(trans
, root
);
1235 ret
= btrfs_end_transaction(trans
, root
);
1238 btrfs_free_path(path
);
1243 * extent_io.c call back to do delayed allocation processing
1245 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1246 u64 start
, u64 end
, int *page_started
,
1247 unsigned long *nr_written
)
1250 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1252 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1253 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1254 page_started
, 1, nr_written
);
1255 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1256 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1257 page_started
, 0, nr_written
);
1258 else if (!btrfs_test_opt(root
, COMPRESS
) &&
1259 !(BTRFS_I(inode
)->force_compress
))
1260 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1261 page_started
, nr_written
, 1);
1263 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1264 page_started
, nr_written
);
1268 static int btrfs_split_extent_hook(struct inode
*inode
,
1269 struct extent_state
*orig
, u64 split
)
1271 /* not delalloc, ignore it */
1272 if (!(orig
->state
& EXTENT_DELALLOC
))
1275 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1280 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1281 * extents so we can keep track of new extents that are just merged onto old
1282 * extents, such as when we are doing sequential writes, so we can properly
1283 * account for the metadata space we'll need.
1285 static int btrfs_merge_extent_hook(struct inode
*inode
,
1286 struct extent_state
*new,
1287 struct extent_state
*other
)
1289 /* not delalloc, ignore it */
1290 if (!(other
->state
& EXTENT_DELALLOC
))
1293 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1298 * extent_io.c set_bit_hook, used to track delayed allocation
1299 * bytes in this file, and to maintain the list of inodes that
1300 * have pending delalloc work to be done.
1302 static int btrfs_set_bit_hook(struct inode
*inode
,
1303 struct extent_state
*state
, int *bits
)
1307 * set_bit and clear bit hooks normally require _irqsave/restore
1308 * but in this case, we are only testeing for the DELALLOC
1309 * bit, which is only set or cleared with irqs on
1311 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1312 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1313 u64 len
= state
->end
+ 1 - state
->start
;
1314 int do_list
= (root
->root_key
.objectid
!=
1315 BTRFS_ROOT_TREE_OBJECTID
);
1317 if (*bits
& EXTENT_FIRST_DELALLOC
)
1318 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1320 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1322 spin_lock(&root
->fs_info
->delalloc_lock
);
1323 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1324 root
->fs_info
->delalloc_bytes
+= len
;
1325 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1326 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1327 &root
->fs_info
->delalloc_inodes
);
1329 spin_unlock(&root
->fs_info
->delalloc_lock
);
1335 * extent_io.c clear_bit_hook, see set_bit_hook for why
1337 static int btrfs_clear_bit_hook(struct inode
*inode
,
1338 struct extent_state
*state
, int *bits
)
1341 * set_bit and clear bit hooks normally require _irqsave/restore
1342 * but in this case, we are only testeing for the DELALLOC
1343 * bit, which is only set or cleared with irqs on
1345 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1346 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1347 u64 len
= state
->end
+ 1 - state
->start
;
1348 int do_list
= (root
->root_key
.objectid
!=
1349 BTRFS_ROOT_TREE_OBJECTID
);
1351 if (*bits
& EXTENT_FIRST_DELALLOC
)
1352 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1353 else if (!(*bits
& EXTENT_DO_ACCOUNTING
))
1354 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1356 if (*bits
& EXTENT_DO_ACCOUNTING
)
1357 btrfs_delalloc_release_metadata(inode
, len
);
1359 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1361 btrfs_free_reserved_data_space(inode
, len
);
1363 spin_lock(&root
->fs_info
->delalloc_lock
);
1364 root
->fs_info
->delalloc_bytes
-= len
;
1365 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1367 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1368 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1369 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1371 spin_unlock(&root
->fs_info
->delalloc_lock
);
1377 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1378 * we don't create bios that span stripes or chunks
1380 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1381 size_t size
, struct bio
*bio
,
1382 unsigned long bio_flags
)
1384 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1385 struct btrfs_mapping_tree
*map_tree
;
1386 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1391 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1394 length
= bio
->bi_size
;
1395 map_tree
= &root
->fs_info
->mapping_tree
;
1396 map_length
= length
;
1397 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1398 &map_length
, NULL
, 0);
1400 if (map_length
< length
+ size
)
1406 * in order to insert checksums into the metadata in large chunks,
1407 * we wait until bio submission time. All the pages in the bio are
1408 * checksummed and sums are attached onto the ordered extent record.
1410 * At IO completion time the cums attached on the ordered extent record
1411 * are inserted into the btree
1413 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1414 struct bio
*bio
, int mirror_num
,
1415 unsigned long bio_flags
,
1418 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1421 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1427 * in order to insert checksums into the metadata in large chunks,
1428 * we wait until bio submission time. All the pages in the bio are
1429 * checksummed and sums are attached onto the ordered extent record.
1431 * At IO completion time the cums attached on the ordered extent record
1432 * are inserted into the btree
1434 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1435 int mirror_num
, unsigned long bio_flags
,
1438 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1439 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1443 * extent_io.c submission hook. This does the right thing for csum calculation
1444 * on write, or reading the csums from the tree before a read
1446 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1447 int mirror_num
, unsigned long bio_flags
,
1450 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1454 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1456 if (root
== root
->fs_info
->tree_root
)
1457 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 2);
1459 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1462 if (!(rw
& REQ_WRITE
)) {
1463 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1464 return btrfs_submit_compressed_read(inode
, bio
,
1465 mirror_num
, bio_flags
);
1466 } else if (!skip_sum
)
1467 btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1469 } else if (!skip_sum
) {
1470 /* csum items have already been cloned */
1471 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1473 /* we're doing a write, do the async checksumming */
1474 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1475 inode
, rw
, bio
, mirror_num
,
1476 bio_flags
, bio_offset
,
1477 __btrfs_submit_bio_start
,
1478 __btrfs_submit_bio_done
);
1482 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1486 * given a list of ordered sums record them in the inode. This happens
1487 * at IO completion time based on sums calculated at bio submission time.
1489 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1490 struct inode
*inode
, u64 file_offset
,
1491 struct list_head
*list
)
1493 struct btrfs_ordered_sum
*sum
;
1495 btrfs_set_trans_block_group(trans
, inode
);
1497 list_for_each_entry(sum
, list
, list
) {
1498 btrfs_csum_file_blocks(trans
,
1499 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1504 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1505 struct extent_state
**cached_state
)
1507 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1509 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1510 cached_state
, GFP_NOFS
);
1513 /* see btrfs_writepage_start_hook for details on why this is required */
1514 struct btrfs_writepage_fixup
{
1516 struct btrfs_work work
;
1519 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1521 struct btrfs_writepage_fixup
*fixup
;
1522 struct btrfs_ordered_extent
*ordered
;
1523 struct extent_state
*cached_state
= NULL
;
1525 struct inode
*inode
;
1529 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1533 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1534 ClearPageChecked(page
);
1538 inode
= page
->mapping
->host
;
1539 page_start
= page_offset(page
);
1540 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1542 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1543 &cached_state
, GFP_NOFS
);
1545 /* already ordered? We're done */
1546 if (PagePrivate2(page
))
1549 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1551 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1552 page_end
, &cached_state
, GFP_NOFS
);
1554 btrfs_start_ordered_extent(inode
, ordered
, 1);
1559 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1560 ClearPageChecked(page
);
1562 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1563 &cached_state
, GFP_NOFS
);
1566 page_cache_release(page
);
1571 * There are a few paths in the higher layers of the kernel that directly
1572 * set the page dirty bit without asking the filesystem if it is a
1573 * good idea. This causes problems because we want to make sure COW
1574 * properly happens and the data=ordered rules are followed.
1576 * In our case any range that doesn't have the ORDERED bit set
1577 * hasn't been properly setup for IO. We kick off an async process
1578 * to fix it up. The async helper will wait for ordered extents, set
1579 * the delalloc bit and make it safe to write the page.
1581 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1583 struct inode
*inode
= page
->mapping
->host
;
1584 struct btrfs_writepage_fixup
*fixup
;
1585 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1587 /* this page is properly in the ordered list */
1588 if (TestClearPagePrivate2(page
))
1591 if (PageChecked(page
))
1594 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1598 SetPageChecked(page
);
1599 page_cache_get(page
);
1600 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1602 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1606 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1607 struct inode
*inode
, u64 file_pos
,
1608 u64 disk_bytenr
, u64 disk_num_bytes
,
1609 u64 num_bytes
, u64 ram_bytes
,
1610 u8 compression
, u8 encryption
,
1611 u16 other_encoding
, int extent_type
)
1613 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1614 struct btrfs_file_extent_item
*fi
;
1615 struct btrfs_path
*path
;
1616 struct extent_buffer
*leaf
;
1617 struct btrfs_key ins
;
1621 path
= btrfs_alloc_path();
1624 path
->leave_spinning
= 1;
1627 * we may be replacing one extent in the tree with another.
1628 * The new extent is pinned in the extent map, and we don't want
1629 * to drop it from the cache until it is completely in the btree.
1631 * So, tell btrfs_drop_extents to leave this extent in the cache.
1632 * the caller is expected to unpin it and allow it to be merged
1635 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1639 ins
.objectid
= inode
->i_ino
;
1640 ins
.offset
= file_pos
;
1641 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1642 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1644 leaf
= path
->nodes
[0];
1645 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1646 struct btrfs_file_extent_item
);
1647 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1648 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1649 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1650 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1651 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1652 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1653 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1654 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1655 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1656 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1658 btrfs_unlock_up_safe(path
, 1);
1659 btrfs_set_lock_blocking(leaf
);
1661 btrfs_mark_buffer_dirty(leaf
);
1663 inode_add_bytes(inode
, num_bytes
);
1665 ins
.objectid
= disk_bytenr
;
1666 ins
.offset
= disk_num_bytes
;
1667 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1668 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1669 root
->root_key
.objectid
,
1670 inode
->i_ino
, file_pos
, &ins
);
1672 btrfs_free_path(path
);
1678 * helper function for btrfs_finish_ordered_io, this
1679 * just reads in some of the csum leaves to prime them into ram
1680 * before we start the transaction. It limits the amount of btree
1681 * reads required while inside the transaction.
1683 /* as ordered data IO finishes, this gets called so we can finish
1684 * an ordered extent if the range of bytes in the file it covers are
1687 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1689 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1690 struct btrfs_trans_handle
*trans
= NULL
;
1691 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1692 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1693 struct extent_state
*cached_state
= NULL
;
1694 int compress_type
= 0;
1696 bool nolock
= false;
1698 ret
= btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1702 BUG_ON(!ordered_extent
);
1704 nolock
= (root
== root
->fs_info
->tree_root
);
1706 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1707 BUG_ON(!list_empty(&ordered_extent
->list
));
1708 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1711 trans
= btrfs_join_transaction_nolock(root
, 1);
1713 trans
= btrfs_join_transaction(root
, 1);
1714 BUG_ON(IS_ERR(trans
));
1715 btrfs_set_trans_block_group(trans
, inode
);
1716 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1717 ret
= btrfs_update_inode(trans
, root
, inode
);
1723 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1724 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1725 0, &cached_state
, GFP_NOFS
);
1728 trans
= btrfs_join_transaction_nolock(root
, 1);
1730 trans
= btrfs_join_transaction(root
, 1);
1731 BUG_ON(IS_ERR(trans
));
1732 btrfs_set_trans_block_group(trans
, inode
);
1733 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1735 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1736 compress_type
= ordered_extent
->compress_type
;
1737 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1738 BUG_ON(compress_type
);
1739 ret
= btrfs_mark_extent_written(trans
, inode
,
1740 ordered_extent
->file_offset
,
1741 ordered_extent
->file_offset
+
1742 ordered_extent
->len
);
1745 BUG_ON(root
== root
->fs_info
->tree_root
);
1746 ret
= insert_reserved_file_extent(trans
, inode
,
1747 ordered_extent
->file_offset
,
1748 ordered_extent
->start
,
1749 ordered_extent
->disk_len
,
1750 ordered_extent
->len
,
1751 ordered_extent
->len
,
1752 compress_type
, 0, 0,
1753 BTRFS_FILE_EXTENT_REG
);
1754 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1755 ordered_extent
->file_offset
,
1756 ordered_extent
->len
);
1759 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1760 ordered_extent
->file_offset
+
1761 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1763 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1764 &ordered_extent
->list
);
1766 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1767 ret
= btrfs_update_inode(trans
, root
, inode
);
1772 btrfs_end_transaction_nolock(trans
, root
);
1774 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1776 btrfs_end_transaction(trans
, root
);
1780 btrfs_put_ordered_extent(ordered_extent
);
1781 /* once for the tree */
1782 btrfs_put_ordered_extent(ordered_extent
);
1787 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1788 struct extent_state
*state
, int uptodate
)
1790 ClearPagePrivate2(page
);
1791 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1795 * When IO fails, either with EIO or csum verification fails, we
1796 * try other mirrors that might have a good copy of the data. This
1797 * io_failure_record is used to record state as we go through all the
1798 * mirrors. If another mirror has good data, the page is set up to date
1799 * and things continue. If a good mirror can't be found, the original
1800 * bio end_io callback is called to indicate things have failed.
1802 struct io_failure_record
{
1807 unsigned long bio_flags
;
1811 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1812 struct page
*page
, u64 start
, u64 end
,
1813 struct extent_state
*state
)
1815 struct io_failure_record
*failrec
= NULL
;
1817 struct extent_map
*em
;
1818 struct inode
*inode
= page
->mapping
->host
;
1819 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1820 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1827 ret
= get_state_private(failure_tree
, start
, &private);
1829 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1832 failrec
->start
= start
;
1833 failrec
->len
= end
- start
+ 1;
1834 failrec
->last_mirror
= 0;
1835 failrec
->bio_flags
= 0;
1837 read_lock(&em_tree
->lock
);
1838 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1839 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1840 free_extent_map(em
);
1843 read_unlock(&em_tree
->lock
);
1845 if (!em
|| IS_ERR(em
)) {
1849 logical
= start
- em
->start
;
1850 logical
= em
->block_start
+ logical
;
1851 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1852 logical
= em
->block_start
;
1853 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1854 extent_set_compress_type(&failrec
->bio_flags
,
1857 failrec
->logical
= logical
;
1858 free_extent_map(em
);
1859 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1860 EXTENT_DIRTY
, GFP_NOFS
);
1861 set_state_private(failure_tree
, start
,
1862 (u64
)(unsigned long)failrec
);
1864 failrec
= (struct io_failure_record
*)(unsigned long)private;
1866 num_copies
= btrfs_num_copies(
1867 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1868 failrec
->logical
, failrec
->len
);
1869 failrec
->last_mirror
++;
1871 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1872 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1875 if (state
&& state
->start
!= failrec
->start
)
1877 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1879 if (!state
|| failrec
->last_mirror
> num_copies
) {
1880 set_state_private(failure_tree
, failrec
->start
, 0);
1881 clear_extent_bits(failure_tree
, failrec
->start
,
1882 failrec
->start
+ failrec
->len
- 1,
1883 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1887 bio
= bio_alloc(GFP_NOFS
, 1);
1888 bio
->bi_private
= state
;
1889 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1890 bio
->bi_sector
= failrec
->logical
>> 9;
1891 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1894 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1895 if (failed_bio
->bi_rw
& REQ_WRITE
)
1900 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1901 failrec
->last_mirror
,
1902 failrec
->bio_flags
, 0);
1907 * each time an IO finishes, we do a fast check in the IO failure tree
1908 * to see if we need to process or clean up an io_failure_record
1910 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1913 u64 private_failure
;
1914 struct io_failure_record
*failure
;
1918 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1919 (u64
)-1, 1, EXTENT_DIRTY
, 0)) {
1920 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1921 start
, &private_failure
);
1923 failure
= (struct io_failure_record
*)(unsigned long)
1925 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1927 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1929 failure
->start
+ failure
->len
- 1,
1930 EXTENT_DIRTY
| EXTENT_LOCKED
,
1939 * when reads are done, we need to check csums to verify the data is correct
1940 * if there's a match, we allow the bio to finish. If not, we go through
1941 * the io_failure_record routines to find good copies
1943 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1944 struct extent_state
*state
)
1946 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1947 struct inode
*inode
= page
->mapping
->host
;
1948 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1950 u64
private = ~(u32
)0;
1952 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1955 if (PageChecked(page
)) {
1956 ClearPageChecked(page
);
1960 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1963 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1964 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1965 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1970 if (state
&& state
->start
== start
) {
1971 private = state
->private;
1974 ret
= get_state_private(io_tree
, start
, &private);
1976 kaddr
= kmap_atomic(page
, KM_USER0
);
1980 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1981 btrfs_csum_final(csum
, (char *)&csum
);
1982 if (csum
!= private)
1985 kunmap_atomic(kaddr
, KM_USER0
);
1987 /* if the io failure tree for this inode is non-empty,
1988 * check to see if we've recovered from a failed IO
1990 btrfs_clean_io_failures(inode
, start
);
1994 if (printk_ratelimit()) {
1995 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
1996 "private %llu\n", page
->mapping
->host
->i_ino
,
1997 (unsigned long long)start
, csum
,
1998 (unsigned long long)private);
2000 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2001 flush_dcache_page(page
);
2002 kunmap_atomic(kaddr
, KM_USER0
);
2008 struct delayed_iput
{
2009 struct list_head list
;
2010 struct inode
*inode
;
2013 void btrfs_add_delayed_iput(struct inode
*inode
)
2015 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2016 struct delayed_iput
*delayed
;
2018 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2021 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2022 delayed
->inode
= inode
;
2024 spin_lock(&fs_info
->delayed_iput_lock
);
2025 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2026 spin_unlock(&fs_info
->delayed_iput_lock
);
2029 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2032 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2033 struct delayed_iput
*delayed
;
2036 spin_lock(&fs_info
->delayed_iput_lock
);
2037 empty
= list_empty(&fs_info
->delayed_iputs
);
2038 spin_unlock(&fs_info
->delayed_iput_lock
);
2042 down_read(&root
->fs_info
->cleanup_work_sem
);
2043 spin_lock(&fs_info
->delayed_iput_lock
);
2044 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2045 spin_unlock(&fs_info
->delayed_iput_lock
);
2047 while (!list_empty(&list
)) {
2048 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2049 list_del(&delayed
->list
);
2050 iput(delayed
->inode
);
2053 up_read(&root
->fs_info
->cleanup_work_sem
);
2057 * calculate extra metadata reservation when snapshotting a subvolume
2058 * contains orphan files.
2060 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle
*trans
,
2061 struct btrfs_pending_snapshot
*pending
,
2062 u64
*bytes_to_reserve
)
2064 struct btrfs_root
*root
;
2065 struct btrfs_block_rsv
*block_rsv
;
2069 root
= pending
->root
;
2070 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2073 block_rsv
= root
->orphan_block_rsv
;
2075 /* orphan block reservation for the snapshot */
2076 num_bytes
= block_rsv
->size
;
2079 * after the snapshot is created, COWing tree blocks may use more
2080 * space than it frees. So we should make sure there is enough
2083 index
= trans
->transid
& 0x1;
2084 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2085 num_bytes
+= block_rsv
->size
-
2086 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2089 *bytes_to_reserve
+= num_bytes
;
2092 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle
*trans
,
2093 struct btrfs_pending_snapshot
*pending
)
2095 struct btrfs_root
*root
= pending
->root
;
2096 struct btrfs_root
*snap
= pending
->snap
;
2097 struct btrfs_block_rsv
*block_rsv
;
2102 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2105 /* refill source subvolume's orphan block reservation */
2106 block_rsv
= root
->orphan_block_rsv
;
2107 index
= trans
->transid
& 0x1;
2108 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2109 num_bytes
= block_rsv
->size
-
2110 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2111 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2112 root
->orphan_block_rsv
,
2117 /* setup orphan block reservation for the snapshot */
2118 block_rsv
= btrfs_alloc_block_rsv(snap
);
2121 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2122 snap
->orphan_block_rsv
= block_rsv
;
2124 num_bytes
= root
->orphan_block_rsv
->size
;
2125 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2126 block_rsv
, num_bytes
);
2130 /* insert orphan item for the snapshot */
2131 WARN_ON(!root
->orphan_item_inserted
);
2132 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2133 snap
->root_key
.objectid
);
2135 snap
->orphan_item_inserted
= 1;
2139 enum btrfs_orphan_cleanup_state
{
2140 ORPHAN_CLEANUP_STARTED
= 1,
2141 ORPHAN_CLEANUP_DONE
= 2,
2145 * This is called in transaction commmit time. If there are no orphan
2146 * files in the subvolume, it removes orphan item and frees block_rsv
2149 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2150 struct btrfs_root
*root
)
2154 if (!list_empty(&root
->orphan_list
) ||
2155 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2158 if (root
->orphan_item_inserted
&&
2159 btrfs_root_refs(&root
->root_item
) > 0) {
2160 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2161 root
->root_key
.objectid
);
2163 root
->orphan_item_inserted
= 0;
2166 if (root
->orphan_block_rsv
) {
2167 WARN_ON(root
->orphan_block_rsv
->size
> 0);
2168 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
2169 root
->orphan_block_rsv
= NULL
;
2174 * This creates an orphan entry for the given inode in case something goes
2175 * wrong in the middle of an unlink/truncate.
2177 * NOTE: caller of this function should reserve 5 units of metadata for
2180 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2182 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2183 struct btrfs_block_rsv
*block_rsv
= NULL
;
2188 if (!root
->orphan_block_rsv
) {
2189 block_rsv
= btrfs_alloc_block_rsv(root
);
2193 spin_lock(&root
->orphan_lock
);
2194 if (!root
->orphan_block_rsv
) {
2195 root
->orphan_block_rsv
= block_rsv
;
2196 } else if (block_rsv
) {
2197 btrfs_free_block_rsv(root
, block_rsv
);
2201 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2202 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2205 * For proper ENOSPC handling, we should do orphan
2206 * cleanup when mounting. But this introduces backward
2207 * compatibility issue.
2209 if (!xchg(&root
->orphan_item_inserted
, 1))
2216 WARN_ON(!BTRFS_I(inode
)->orphan_meta_reserved
);
2219 if (!BTRFS_I(inode
)->orphan_meta_reserved
) {
2220 BTRFS_I(inode
)->orphan_meta_reserved
= 1;
2223 spin_unlock(&root
->orphan_lock
);
2226 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2228 /* grab metadata reservation from transaction handle */
2230 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2234 /* insert an orphan item to track this unlinked/truncated file */
2236 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
2240 /* insert an orphan item to track subvolume contains orphan files */
2242 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2243 root
->root_key
.objectid
);
2250 * We have done the truncate/delete so we can go ahead and remove the orphan
2251 * item for this particular inode.
2253 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2255 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2256 int delete_item
= 0;
2257 int release_rsv
= 0;
2260 spin_lock(&root
->orphan_lock
);
2261 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2262 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2266 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2267 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2270 spin_unlock(&root
->orphan_lock
);
2272 if (trans
&& delete_item
) {
2273 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
2278 btrfs_orphan_release_metadata(inode
);
2284 * this cleans up any orphans that may be left on the list from the last use
2287 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2289 struct btrfs_path
*path
;
2290 struct extent_buffer
*leaf
;
2291 struct btrfs_key key
, found_key
;
2292 struct btrfs_trans_handle
*trans
;
2293 struct inode
*inode
;
2294 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2296 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2299 path
= btrfs_alloc_path();
2306 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2307 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2308 key
.offset
= (u64
)-1;
2311 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2316 * if ret == 0 means we found what we were searching for, which
2317 * is weird, but possible, so only screw with path if we didnt
2318 * find the key and see if we have stuff that matches
2322 if (path
->slots
[0] == 0)
2327 /* pull out the item */
2328 leaf
= path
->nodes
[0];
2329 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2331 /* make sure the item matches what we want */
2332 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2334 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2337 /* release the path since we're done with it */
2338 btrfs_release_path(root
, path
);
2341 * this is where we are basically btrfs_lookup, without the
2342 * crossing root thing. we store the inode number in the
2343 * offset of the orphan item.
2345 found_key
.objectid
= found_key
.offset
;
2346 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2347 found_key
.offset
= 0;
2348 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2349 if (IS_ERR(inode
)) {
2350 ret
= PTR_ERR(inode
);
2355 * add this inode to the orphan list so btrfs_orphan_del does
2356 * the proper thing when we hit it
2358 spin_lock(&root
->orphan_lock
);
2359 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2360 spin_unlock(&root
->orphan_lock
);
2363 * if this is a bad inode, means we actually succeeded in
2364 * removing the inode, but not the orphan record, which means
2365 * we need to manually delete the orphan since iput will just
2366 * do a destroy_inode
2368 if (is_bad_inode(inode
)) {
2369 trans
= btrfs_start_transaction(root
, 0);
2370 if (IS_ERR(trans
)) {
2371 ret
= PTR_ERR(trans
);
2374 btrfs_orphan_del(trans
, inode
);
2375 btrfs_end_transaction(trans
, root
);
2380 /* if we have links, this was a truncate, lets do that */
2381 if (inode
->i_nlink
) {
2382 if (!S_ISREG(inode
->i_mode
)) {
2388 ret
= btrfs_truncate(inode
);
2393 /* this will do delete_inode and everything for us */
2398 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2400 if (root
->orphan_block_rsv
)
2401 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2404 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2405 trans
= btrfs_join_transaction(root
, 1);
2407 btrfs_end_transaction(trans
, root
);
2411 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2413 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2417 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2418 btrfs_free_path(path
);
2423 * very simple check to peek ahead in the leaf looking for xattrs. If we
2424 * don't find any xattrs, we know there can't be any acls.
2426 * slot is the slot the inode is in, objectid is the objectid of the inode
2428 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2429 int slot
, u64 objectid
)
2431 u32 nritems
= btrfs_header_nritems(leaf
);
2432 struct btrfs_key found_key
;
2436 while (slot
< nritems
) {
2437 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2439 /* we found a different objectid, there must not be acls */
2440 if (found_key
.objectid
!= objectid
)
2443 /* we found an xattr, assume we've got an acl */
2444 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2448 * we found a key greater than an xattr key, there can't
2449 * be any acls later on
2451 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2458 * it goes inode, inode backrefs, xattrs, extents,
2459 * so if there are a ton of hard links to an inode there can
2460 * be a lot of backrefs. Don't waste time searching too hard,
2461 * this is just an optimization
2466 /* we hit the end of the leaf before we found an xattr or
2467 * something larger than an xattr. We have to assume the inode
2474 * read an inode from the btree into the in-memory inode
2476 static void btrfs_read_locked_inode(struct inode
*inode
)
2478 struct btrfs_path
*path
;
2479 struct extent_buffer
*leaf
;
2480 struct btrfs_inode_item
*inode_item
;
2481 struct btrfs_timespec
*tspec
;
2482 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2483 struct btrfs_key location
;
2485 u64 alloc_group_block
;
2489 path
= btrfs_alloc_path();
2491 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2493 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2497 leaf
= path
->nodes
[0];
2498 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2499 struct btrfs_inode_item
);
2501 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2502 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2503 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2504 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2505 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2507 tspec
= btrfs_inode_atime(inode_item
);
2508 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2509 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2511 tspec
= btrfs_inode_mtime(inode_item
);
2512 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2513 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2515 tspec
= btrfs_inode_ctime(inode_item
);
2516 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2517 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2519 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2520 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2521 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2522 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2524 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2526 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2527 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2529 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2532 * try to precache a NULL acl entry for files that don't have
2533 * any xattrs or acls
2535 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0], inode
->i_ino
);
2537 cache_no_acl(inode
);
2539 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2540 alloc_group_block
, 0);
2541 btrfs_free_path(path
);
2544 switch (inode
->i_mode
& S_IFMT
) {
2546 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2547 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2548 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2549 inode
->i_fop
= &btrfs_file_operations
;
2550 inode
->i_op
= &btrfs_file_inode_operations
;
2553 inode
->i_fop
= &btrfs_dir_file_operations
;
2554 if (root
== root
->fs_info
->tree_root
)
2555 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2557 inode
->i_op
= &btrfs_dir_inode_operations
;
2560 inode
->i_op
= &btrfs_symlink_inode_operations
;
2561 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2562 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2565 inode
->i_op
= &btrfs_special_inode_operations
;
2566 init_special_inode(inode
, inode
->i_mode
, rdev
);
2570 btrfs_update_iflags(inode
);
2574 btrfs_free_path(path
);
2575 make_bad_inode(inode
);
2579 * given a leaf and an inode, copy the inode fields into the leaf
2581 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2582 struct extent_buffer
*leaf
,
2583 struct btrfs_inode_item
*item
,
2584 struct inode
*inode
)
2586 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2587 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2588 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2589 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2590 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2592 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2593 inode
->i_atime
.tv_sec
);
2594 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2595 inode
->i_atime
.tv_nsec
);
2597 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2598 inode
->i_mtime
.tv_sec
);
2599 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2600 inode
->i_mtime
.tv_nsec
);
2602 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2603 inode
->i_ctime
.tv_sec
);
2604 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2605 inode
->i_ctime
.tv_nsec
);
2607 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2608 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2609 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2610 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2611 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2612 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2613 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2617 * copy everything in the in-memory inode into the btree.
2619 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2620 struct btrfs_root
*root
, struct inode
*inode
)
2622 struct btrfs_inode_item
*inode_item
;
2623 struct btrfs_path
*path
;
2624 struct extent_buffer
*leaf
;
2627 path
= btrfs_alloc_path();
2629 path
->leave_spinning
= 1;
2630 ret
= btrfs_lookup_inode(trans
, root
, path
,
2631 &BTRFS_I(inode
)->location
, 1);
2638 btrfs_unlock_up_safe(path
, 1);
2639 leaf
= path
->nodes
[0];
2640 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2641 struct btrfs_inode_item
);
2643 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2644 btrfs_mark_buffer_dirty(leaf
);
2645 btrfs_set_inode_last_trans(trans
, inode
);
2648 btrfs_free_path(path
);
2654 * unlink helper that gets used here in inode.c and in the tree logging
2655 * recovery code. It remove a link in a directory with a given name, and
2656 * also drops the back refs in the inode to the directory
2658 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2659 struct btrfs_root
*root
,
2660 struct inode
*dir
, struct inode
*inode
,
2661 const char *name
, int name_len
)
2663 struct btrfs_path
*path
;
2665 struct extent_buffer
*leaf
;
2666 struct btrfs_dir_item
*di
;
2667 struct btrfs_key key
;
2670 path
= btrfs_alloc_path();
2676 path
->leave_spinning
= 1;
2677 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2678 name
, name_len
, -1);
2687 leaf
= path
->nodes
[0];
2688 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2689 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2692 btrfs_release_path(root
, path
);
2694 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2696 dir
->i_ino
, &index
);
2698 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2699 "inode %lu parent %lu\n", name_len
, name
,
2700 inode
->i_ino
, dir
->i_ino
);
2704 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2705 index
, name
, name_len
, -1);
2714 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2715 btrfs_release_path(root
, path
);
2717 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2719 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2721 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2726 btrfs_free_path(path
);
2730 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2731 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2732 btrfs_update_inode(trans
, root
, dir
);
2733 btrfs_drop_nlink(inode
);
2734 ret
= btrfs_update_inode(trans
, root
, inode
);
2739 /* helper to check if there is any shared block in the path */
2740 static int check_path_shared(struct btrfs_root
*root
,
2741 struct btrfs_path
*path
)
2743 struct extent_buffer
*eb
;
2747 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2750 if (!path
->nodes
[level
])
2752 eb
= path
->nodes
[level
];
2753 if (!btrfs_block_can_be_shared(root
, eb
))
2755 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2764 * helper to start transaction for unlink and rmdir.
2766 * unlink and rmdir are special in btrfs, they do not always free space.
2767 * so in enospc case, we should make sure they will free space before
2768 * allowing them to use the global metadata reservation.
2770 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2771 struct dentry
*dentry
)
2773 struct btrfs_trans_handle
*trans
;
2774 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2775 struct btrfs_path
*path
;
2776 struct btrfs_inode_ref
*ref
;
2777 struct btrfs_dir_item
*di
;
2778 struct inode
*inode
= dentry
->d_inode
;
2784 trans
= btrfs_start_transaction(root
, 10);
2785 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2788 if (inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2789 return ERR_PTR(-ENOSPC
);
2791 /* check if there is someone else holds reference */
2792 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2793 return ERR_PTR(-ENOSPC
);
2795 if (atomic_read(&inode
->i_count
) > 2)
2796 return ERR_PTR(-ENOSPC
);
2798 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2799 return ERR_PTR(-ENOSPC
);
2801 path
= btrfs_alloc_path();
2803 root
->fs_info
->enospc_unlink
= 0;
2804 return ERR_PTR(-ENOMEM
);
2807 trans
= btrfs_start_transaction(root
, 0);
2808 if (IS_ERR(trans
)) {
2809 btrfs_free_path(path
);
2810 root
->fs_info
->enospc_unlink
= 0;
2814 path
->skip_locking
= 1;
2815 path
->search_commit_root
= 1;
2817 ret
= btrfs_lookup_inode(trans
, root
, path
,
2818 &BTRFS_I(dir
)->location
, 0);
2824 if (check_path_shared(root
, path
))
2829 btrfs_release_path(root
, path
);
2831 ret
= btrfs_lookup_inode(trans
, root
, path
,
2832 &BTRFS_I(inode
)->location
, 0);
2838 if (check_path_shared(root
, path
))
2843 btrfs_release_path(root
, path
);
2845 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2846 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2847 inode
->i_ino
, (u64
)-1, 0);
2853 if (check_path_shared(root
, path
))
2855 btrfs_release_path(root
, path
);
2863 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2864 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2870 if (check_path_shared(root
, path
))
2876 btrfs_release_path(root
, path
);
2878 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2879 dentry
->d_name
.name
, dentry
->d_name
.len
,
2880 inode
->i_ino
, dir
->i_ino
, 0);
2886 if (check_path_shared(root
, path
))
2888 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2889 btrfs_release_path(root
, path
);
2891 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
, index
,
2892 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2897 BUG_ON(ret
== -ENOENT
);
2898 if (check_path_shared(root
, path
))
2903 btrfs_free_path(path
);
2905 btrfs_end_transaction(trans
, root
);
2906 root
->fs_info
->enospc_unlink
= 0;
2907 return ERR_PTR(err
);
2910 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2914 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2915 struct btrfs_root
*root
)
2917 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2918 BUG_ON(!root
->fs_info
->enospc_unlink
);
2919 root
->fs_info
->enospc_unlink
= 0;
2921 btrfs_end_transaction_throttle(trans
, root
);
2924 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2926 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2927 struct btrfs_trans_handle
*trans
;
2928 struct inode
*inode
= dentry
->d_inode
;
2930 unsigned long nr
= 0;
2932 trans
= __unlink_start_trans(dir
, dentry
);
2934 return PTR_ERR(trans
);
2936 btrfs_set_trans_block_group(trans
, dir
);
2938 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2940 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2941 dentry
->d_name
.name
, dentry
->d_name
.len
);
2944 if (inode
->i_nlink
== 0) {
2945 ret
= btrfs_orphan_add(trans
, inode
);
2949 nr
= trans
->blocks_used
;
2950 __unlink_end_trans(trans
, root
);
2951 btrfs_btree_balance_dirty(root
, nr
);
2955 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2956 struct btrfs_root
*root
,
2957 struct inode
*dir
, u64 objectid
,
2958 const char *name
, int name_len
)
2960 struct btrfs_path
*path
;
2961 struct extent_buffer
*leaf
;
2962 struct btrfs_dir_item
*di
;
2963 struct btrfs_key key
;
2967 path
= btrfs_alloc_path();
2971 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2972 name
, name_len
, -1);
2973 BUG_ON(!di
|| IS_ERR(di
));
2975 leaf
= path
->nodes
[0];
2976 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2977 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
2978 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2980 btrfs_release_path(root
, path
);
2982 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
2983 objectid
, root
->root_key
.objectid
,
2984 dir
->i_ino
, &index
, name
, name_len
);
2986 BUG_ON(ret
!= -ENOENT
);
2987 di
= btrfs_search_dir_index_item(root
, path
, dir
->i_ino
,
2989 BUG_ON(!di
|| IS_ERR(di
));
2991 leaf
= path
->nodes
[0];
2992 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2993 btrfs_release_path(root
, path
);
2997 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2998 index
, name
, name_len
, -1);
2999 BUG_ON(!di
|| IS_ERR(di
));
3001 leaf
= path
->nodes
[0];
3002 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3003 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3004 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3006 btrfs_release_path(root
, path
);
3008 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3009 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3010 ret
= btrfs_update_inode(trans
, root
, dir
);
3013 btrfs_free_path(path
);
3017 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3019 struct inode
*inode
= dentry
->d_inode
;
3021 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3022 struct btrfs_trans_handle
*trans
;
3023 unsigned long nr
= 0;
3025 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3026 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
3029 trans
= __unlink_start_trans(dir
, dentry
);
3031 return PTR_ERR(trans
);
3033 btrfs_set_trans_block_group(trans
, dir
);
3035 if (unlikely(inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3036 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3037 BTRFS_I(inode
)->location
.objectid
,
3038 dentry
->d_name
.name
,
3039 dentry
->d_name
.len
);
3043 err
= btrfs_orphan_add(trans
, inode
);
3047 /* now the directory is empty */
3048 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3049 dentry
->d_name
.name
, dentry
->d_name
.len
);
3051 btrfs_i_size_write(inode
, 0);
3053 nr
= trans
->blocks_used
;
3054 __unlink_end_trans(trans
, root
);
3055 btrfs_btree_balance_dirty(root
, nr
);
3062 * when truncating bytes in a file, it is possible to avoid reading
3063 * the leaves that contain only checksum items. This can be the
3064 * majority of the IO required to delete a large file, but it must
3065 * be done carefully.
3067 * The keys in the level just above the leaves are checked to make sure
3068 * the lowest key in a given leaf is a csum key, and starts at an offset
3069 * after the new size.
3071 * Then the key for the next leaf is checked to make sure it also has
3072 * a checksum item for the same file. If it does, we know our target leaf
3073 * contains only checksum items, and it can be safely freed without reading
3076 * This is just an optimization targeted at large files. It may do
3077 * nothing. It will return 0 unless things went badly.
3079 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
3080 struct btrfs_root
*root
,
3081 struct btrfs_path
*path
,
3082 struct inode
*inode
, u64 new_size
)
3084 struct btrfs_key key
;
3087 struct btrfs_key found_key
;
3088 struct btrfs_key other_key
;
3089 struct btrfs_leaf_ref
*ref
;
3093 path
->lowest_level
= 1;
3094 key
.objectid
= inode
->i_ino
;
3095 key
.type
= BTRFS_CSUM_ITEM_KEY
;
3096 key
.offset
= new_size
;
3098 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3102 if (path
->nodes
[1] == NULL
) {
3107 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
3108 nritems
= btrfs_header_nritems(path
->nodes
[1]);
3113 if (path
->slots
[1] >= nritems
)
3116 /* did we find a key greater than anything we want to delete? */
3117 if (found_key
.objectid
> inode
->i_ino
||
3118 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
3121 /* we check the next key in the node to make sure the leave contains
3122 * only checksum items. This comparison doesn't work if our
3123 * leaf is the last one in the node
3125 if (path
->slots
[1] + 1 >= nritems
) {
3127 /* search forward from the last key in the node, this
3128 * will bring us into the next node in the tree
3130 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
3132 /* unlikely, but we inc below, so check to be safe */
3133 if (found_key
.offset
== (u64
)-1)
3136 /* search_forward needs a path with locks held, do the
3137 * search again for the original key. It is possible
3138 * this will race with a balance and return a path that
3139 * we could modify, but this drop is just an optimization
3140 * and is allowed to miss some leaves.
3142 btrfs_release_path(root
, path
);
3145 /* setup a max key for search_forward */
3146 other_key
.offset
= (u64
)-1;
3147 other_key
.type
= key
.type
;
3148 other_key
.objectid
= key
.objectid
;
3150 path
->keep_locks
= 1;
3151 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
3153 path
->keep_locks
= 0;
3154 if (ret
|| found_key
.objectid
!= key
.objectid
||
3155 found_key
.type
!= key
.type
) {
3160 key
.offset
= found_key
.offset
;
3161 btrfs_release_path(root
, path
);
3166 /* we know there's one more slot after us in the tree,
3167 * read that key so we can verify it is also a checksum item
3169 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
3171 if (found_key
.objectid
< inode
->i_ino
)
3174 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
3178 * if the key for the next leaf isn't a csum key from this objectid,
3179 * we can't be sure there aren't good items inside this leaf.
3182 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
3185 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
3186 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
3188 * it is safe to delete this leaf, it contains only
3189 * csum items from this inode at an offset >= new_size
3191 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
3194 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
3195 ref
= btrfs_alloc_leaf_ref(root
, 0);
3197 ref
->root_gen
= root
->root_key
.offset
;
3198 ref
->bytenr
= leaf_start
;
3200 ref
->generation
= leaf_gen
;
3203 btrfs_sort_leaf_ref(ref
);
3205 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
3207 btrfs_free_leaf_ref(root
, ref
);
3213 btrfs_release_path(root
, path
);
3215 if (other_key
.objectid
== inode
->i_ino
&&
3216 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
3217 key
.offset
= other_key
.offset
;
3223 /* fixup any changes we've made to the path */
3224 path
->lowest_level
= 0;
3225 path
->keep_locks
= 0;
3226 btrfs_release_path(root
, path
);
3233 * this can truncate away extent items, csum items and directory items.
3234 * It starts at a high offset and removes keys until it can't find
3235 * any higher than new_size
3237 * csum items that cross the new i_size are truncated to the new size
3240 * min_type is the minimum key type to truncate down to. If set to 0, this
3241 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3243 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3244 struct btrfs_root
*root
,
3245 struct inode
*inode
,
3246 u64 new_size
, u32 min_type
)
3248 struct btrfs_path
*path
;
3249 struct extent_buffer
*leaf
;
3250 struct btrfs_file_extent_item
*fi
;
3251 struct btrfs_key key
;
3252 struct btrfs_key found_key
;
3253 u64 extent_start
= 0;
3254 u64 extent_num_bytes
= 0;
3255 u64 extent_offset
= 0;
3257 u64 mask
= root
->sectorsize
- 1;
3258 u32 found_type
= (u8
)-1;
3261 int pending_del_nr
= 0;
3262 int pending_del_slot
= 0;
3263 int extent_type
= -1;
3268 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3270 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3271 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3273 path
= btrfs_alloc_path();
3277 key
.objectid
= inode
->i_ino
;
3278 key
.offset
= (u64
)-1;
3282 path
->leave_spinning
= 1;
3283 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3290 /* there are no items in the tree for us to truncate, we're
3293 if (path
->slots
[0] == 0)
3300 leaf
= path
->nodes
[0];
3301 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3302 found_type
= btrfs_key_type(&found_key
);
3305 if (found_key
.objectid
!= inode
->i_ino
)
3308 if (found_type
< min_type
)
3311 item_end
= found_key
.offset
;
3312 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3313 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3314 struct btrfs_file_extent_item
);
3315 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3316 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3317 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3318 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3320 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3322 btrfs_file_extent_num_bytes(leaf
, fi
);
3323 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3324 item_end
+= btrfs_file_extent_inline_len(leaf
,
3329 if (found_type
> min_type
) {
3332 if (item_end
< new_size
)
3334 if (found_key
.offset
>= new_size
)
3340 /* FIXME, shrink the extent if the ref count is only 1 */
3341 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3344 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3346 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3347 if (!del_item
&& !encoding
) {
3348 u64 orig_num_bytes
=
3349 btrfs_file_extent_num_bytes(leaf
, fi
);
3350 extent_num_bytes
= new_size
-
3351 found_key
.offset
+ root
->sectorsize
- 1;
3352 extent_num_bytes
= extent_num_bytes
&
3353 ~((u64
)root
->sectorsize
- 1);
3354 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3356 num_dec
= (orig_num_bytes
-
3358 if (root
->ref_cows
&& extent_start
!= 0)
3359 inode_sub_bytes(inode
, num_dec
);
3360 btrfs_mark_buffer_dirty(leaf
);
3363 btrfs_file_extent_disk_num_bytes(leaf
,
3365 extent_offset
= found_key
.offset
-
3366 btrfs_file_extent_offset(leaf
, fi
);
3368 /* FIXME blocksize != 4096 */
3369 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3370 if (extent_start
!= 0) {
3373 inode_sub_bytes(inode
, num_dec
);
3376 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3378 * we can't truncate inline items that have had
3382 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3383 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3384 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3385 u32 size
= new_size
- found_key
.offset
;
3387 if (root
->ref_cows
) {
3388 inode_sub_bytes(inode
, item_end
+ 1 -
3392 btrfs_file_extent_calc_inline_size(size
);
3393 ret
= btrfs_truncate_item(trans
, root
, path
,
3396 } else if (root
->ref_cows
) {
3397 inode_sub_bytes(inode
, item_end
+ 1 -
3403 if (!pending_del_nr
) {
3404 /* no pending yet, add ourselves */
3405 pending_del_slot
= path
->slots
[0];
3407 } else if (pending_del_nr
&&
3408 path
->slots
[0] + 1 == pending_del_slot
) {
3409 /* hop on the pending chunk */
3411 pending_del_slot
= path
->slots
[0];
3418 if (found_extent
&& (root
->ref_cows
||
3419 root
== root
->fs_info
->tree_root
)) {
3420 btrfs_set_path_blocking(path
);
3421 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3422 extent_num_bytes
, 0,
3423 btrfs_header_owner(leaf
),
3424 inode
->i_ino
, extent_offset
);
3428 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3431 if (path
->slots
[0] == 0 ||
3432 path
->slots
[0] != pending_del_slot
) {
3433 if (root
->ref_cows
) {
3437 if (pending_del_nr
) {
3438 ret
= btrfs_del_items(trans
, root
, path
,
3444 btrfs_release_path(root
, path
);
3451 if (pending_del_nr
) {
3452 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3456 btrfs_free_path(path
);
3461 * taken from block_truncate_page, but does cow as it zeros out
3462 * any bytes left in the last page in the file.
3464 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3466 struct inode
*inode
= mapping
->host
;
3467 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3468 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3469 struct btrfs_ordered_extent
*ordered
;
3470 struct extent_state
*cached_state
= NULL
;
3472 u32 blocksize
= root
->sectorsize
;
3473 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3474 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3480 if ((offset
& (blocksize
- 1)) == 0)
3482 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3488 page
= grab_cache_page(mapping
, index
);
3490 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3494 page_start
= page_offset(page
);
3495 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3497 if (!PageUptodate(page
)) {
3498 ret
= btrfs_readpage(NULL
, page
);
3500 if (page
->mapping
!= mapping
) {
3502 page_cache_release(page
);
3505 if (!PageUptodate(page
)) {
3510 wait_on_page_writeback(page
);
3512 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3514 set_page_extent_mapped(page
);
3516 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3518 unlock_extent_cached(io_tree
, page_start
, page_end
,
3519 &cached_state
, GFP_NOFS
);
3521 page_cache_release(page
);
3522 btrfs_start_ordered_extent(inode
, ordered
, 1);
3523 btrfs_put_ordered_extent(ordered
);
3527 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3528 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3529 0, 0, &cached_state
, GFP_NOFS
);
3531 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3534 unlock_extent_cached(io_tree
, page_start
, page_end
,
3535 &cached_state
, GFP_NOFS
);
3540 if (offset
!= PAGE_CACHE_SIZE
) {
3542 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3543 flush_dcache_page(page
);
3546 ClearPageChecked(page
);
3547 set_page_dirty(page
);
3548 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3553 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3555 page_cache_release(page
);
3561 * This function puts in dummy file extents for the area we're creating a hole
3562 * for. So if we are truncating this file to a larger size we need to insert
3563 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3564 * the range between oldsize and size
3566 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3568 struct btrfs_trans_handle
*trans
;
3569 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3570 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3571 struct extent_map
*em
= NULL
;
3572 struct extent_state
*cached_state
= NULL
;
3573 u64 mask
= root
->sectorsize
- 1;
3574 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3575 u64 block_end
= (size
+ mask
) & ~mask
;
3581 if (size
<= hole_start
)
3585 struct btrfs_ordered_extent
*ordered
;
3586 btrfs_wait_ordered_range(inode
, hole_start
,
3587 block_end
- hole_start
);
3588 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3589 &cached_state
, GFP_NOFS
);
3590 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3593 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3594 &cached_state
, GFP_NOFS
);
3595 btrfs_put_ordered_extent(ordered
);
3598 cur_offset
= hole_start
;
3600 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3601 block_end
- cur_offset
, 0);
3602 BUG_ON(IS_ERR(em
) || !em
);
3603 last_byte
= min(extent_map_end(em
), block_end
);
3604 last_byte
= (last_byte
+ mask
) & ~mask
;
3605 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3607 hole_size
= last_byte
- cur_offset
;
3609 trans
= btrfs_start_transaction(root
, 2);
3610 if (IS_ERR(trans
)) {
3611 err
= PTR_ERR(trans
);
3614 btrfs_set_trans_block_group(trans
, inode
);
3616 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3617 cur_offset
+ hole_size
,
3622 err
= btrfs_insert_file_extent(trans
, root
,
3623 inode
->i_ino
, cur_offset
, 0,
3624 0, hole_size
, 0, hole_size
,
3629 btrfs_drop_extent_cache(inode
, hole_start
,
3632 btrfs_end_transaction(trans
, root
);
3634 free_extent_map(em
);
3636 cur_offset
= last_byte
;
3637 if (cur_offset
>= block_end
)
3641 free_extent_map(em
);
3642 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3647 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3649 loff_t oldsize
= i_size_read(inode
);
3652 if (newsize
== oldsize
)
3655 if (newsize
> oldsize
) {
3656 i_size_write(inode
, newsize
);
3657 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3658 truncate_pagecache(inode
, oldsize
, newsize
);
3659 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3661 btrfs_setsize(inode
, oldsize
);
3665 mark_inode_dirty(inode
);
3669 * We're truncating a file that used to have good data down to
3670 * zero. Make sure it gets into the ordered flush list so that
3671 * any new writes get down to disk quickly.
3674 BTRFS_I(inode
)->ordered_data_close
= 1;
3676 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3677 truncate_setsize(inode
, newsize
);
3678 ret
= btrfs_truncate(inode
);
3684 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3686 struct inode
*inode
= dentry
->d_inode
;
3687 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3690 if (btrfs_root_readonly(root
))
3693 err
= inode_change_ok(inode
, attr
);
3697 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3698 err
= btrfs_setsize(inode
, attr
->ia_size
);
3703 if (attr
->ia_valid
) {
3704 setattr_copy(inode
, attr
);
3705 mark_inode_dirty(inode
);
3707 if (attr
->ia_valid
& ATTR_MODE
)
3708 err
= btrfs_acl_chmod(inode
);
3714 void btrfs_evict_inode(struct inode
*inode
)
3716 struct btrfs_trans_handle
*trans
;
3717 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3721 truncate_inode_pages(&inode
->i_data
, 0);
3722 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3723 root
== root
->fs_info
->tree_root
))
3726 if (is_bad_inode(inode
)) {
3727 btrfs_orphan_del(NULL
, inode
);
3730 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3731 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3733 if (root
->fs_info
->log_root_recovering
) {
3734 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3738 if (inode
->i_nlink
> 0) {
3739 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3743 btrfs_i_size_write(inode
, 0);
3746 trans
= btrfs_start_transaction(root
, 0);
3747 BUG_ON(IS_ERR(trans
));
3748 btrfs_set_trans_block_group(trans
, inode
);
3749 trans
->block_rsv
= root
->orphan_block_rsv
;
3751 ret
= btrfs_block_rsv_check(trans
, root
,
3752 root
->orphan_block_rsv
, 0, 5);
3754 BUG_ON(ret
!= -EAGAIN
);
3755 ret
= btrfs_commit_transaction(trans
, root
);
3760 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3764 nr
= trans
->blocks_used
;
3765 btrfs_end_transaction(trans
, root
);
3767 btrfs_btree_balance_dirty(root
, nr
);
3772 ret
= btrfs_orphan_del(trans
, inode
);
3776 nr
= trans
->blocks_used
;
3777 btrfs_end_transaction(trans
, root
);
3778 btrfs_btree_balance_dirty(root
, nr
);
3780 end_writeback(inode
);
3785 * this returns the key found in the dir entry in the location pointer.
3786 * If no dir entries were found, location->objectid is 0.
3788 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3789 struct btrfs_key
*location
)
3791 const char *name
= dentry
->d_name
.name
;
3792 int namelen
= dentry
->d_name
.len
;
3793 struct btrfs_dir_item
*di
;
3794 struct btrfs_path
*path
;
3795 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3798 path
= btrfs_alloc_path();
3801 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
3806 if (!di
|| IS_ERR(di
))
3809 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3811 btrfs_free_path(path
);
3814 location
->objectid
= 0;
3819 * when we hit a tree root in a directory, the btrfs part of the inode
3820 * needs to be changed to reflect the root directory of the tree root. This
3821 * is kind of like crossing a mount point.
3823 static int fixup_tree_root_location(struct btrfs_root
*root
,
3825 struct dentry
*dentry
,
3826 struct btrfs_key
*location
,
3827 struct btrfs_root
**sub_root
)
3829 struct btrfs_path
*path
;
3830 struct btrfs_root
*new_root
;
3831 struct btrfs_root_ref
*ref
;
3832 struct extent_buffer
*leaf
;
3836 path
= btrfs_alloc_path();
3843 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3844 BTRFS_I(dir
)->root
->root_key
.objectid
,
3845 location
->objectid
);
3852 leaf
= path
->nodes
[0];
3853 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3854 if (btrfs_root_ref_dirid(leaf
, ref
) != dir
->i_ino
||
3855 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3858 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3859 (unsigned long)(ref
+ 1),
3860 dentry
->d_name
.len
);
3864 btrfs_release_path(root
->fs_info
->tree_root
, path
);
3866 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3867 if (IS_ERR(new_root
)) {
3868 err
= PTR_ERR(new_root
);
3872 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3877 *sub_root
= new_root
;
3878 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3879 location
->type
= BTRFS_INODE_ITEM_KEY
;
3880 location
->offset
= 0;
3883 btrfs_free_path(path
);
3887 static void inode_tree_add(struct inode
*inode
)
3889 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3890 struct btrfs_inode
*entry
;
3892 struct rb_node
*parent
;
3894 p
= &root
->inode_tree
.rb_node
;
3897 if (inode_unhashed(inode
))
3900 spin_lock(&root
->inode_lock
);
3903 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3905 if (inode
->i_ino
< entry
->vfs_inode
.i_ino
)
3906 p
= &parent
->rb_left
;
3907 else if (inode
->i_ino
> entry
->vfs_inode
.i_ino
)
3908 p
= &parent
->rb_right
;
3910 WARN_ON(!(entry
->vfs_inode
.i_state
&
3911 (I_WILL_FREE
| I_FREEING
)));
3912 rb_erase(parent
, &root
->inode_tree
);
3913 RB_CLEAR_NODE(parent
);
3914 spin_unlock(&root
->inode_lock
);
3918 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3919 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3920 spin_unlock(&root
->inode_lock
);
3923 static void inode_tree_del(struct inode
*inode
)
3925 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3928 spin_lock(&root
->inode_lock
);
3929 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3930 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3931 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3932 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3934 spin_unlock(&root
->inode_lock
);
3937 * Free space cache has inodes in the tree root, but the tree root has a
3938 * root_refs of 0, so this could end up dropping the tree root as a
3939 * snapshot, so we need the extra !root->fs_info->tree_root check to
3940 * make sure we don't drop it.
3942 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3943 root
!= root
->fs_info
->tree_root
) {
3944 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3945 spin_lock(&root
->inode_lock
);
3946 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3947 spin_unlock(&root
->inode_lock
);
3949 btrfs_add_dead_root(root
);
3953 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3955 struct rb_node
*node
;
3956 struct rb_node
*prev
;
3957 struct btrfs_inode
*entry
;
3958 struct inode
*inode
;
3961 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3963 spin_lock(&root
->inode_lock
);
3965 node
= root
->inode_tree
.rb_node
;
3969 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3971 if (objectid
< entry
->vfs_inode
.i_ino
)
3972 node
= node
->rb_left
;
3973 else if (objectid
> entry
->vfs_inode
.i_ino
)
3974 node
= node
->rb_right
;
3980 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
3981 if (objectid
<= entry
->vfs_inode
.i_ino
) {
3985 prev
= rb_next(prev
);
3989 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3990 objectid
= entry
->vfs_inode
.i_ino
+ 1;
3991 inode
= igrab(&entry
->vfs_inode
);
3993 spin_unlock(&root
->inode_lock
);
3994 if (atomic_read(&inode
->i_count
) > 1)
3995 d_prune_aliases(inode
);
3997 * btrfs_drop_inode will have it removed from
3998 * the inode cache when its usage count
4003 spin_lock(&root
->inode_lock
);
4007 if (cond_resched_lock(&root
->inode_lock
))
4010 node
= rb_next(node
);
4012 spin_unlock(&root
->inode_lock
);
4016 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4018 struct btrfs_iget_args
*args
= p
;
4019 inode
->i_ino
= args
->ino
;
4020 BTRFS_I(inode
)->root
= args
->root
;
4021 btrfs_set_inode_space_info(args
->root
, inode
);
4025 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4027 struct btrfs_iget_args
*args
= opaque
;
4028 return args
->ino
== inode
->i_ino
&&
4029 args
->root
== BTRFS_I(inode
)->root
;
4032 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4034 struct btrfs_root
*root
)
4036 struct inode
*inode
;
4037 struct btrfs_iget_args args
;
4038 args
.ino
= objectid
;
4041 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4042 btrfs_init_locked_inode
,
4047 /* Get an inode object given its location and corresponding root.
4048 * Returns in *is_new if the inode was read from disk
4050 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4051 struct btrfs_root
*root
, int *new)
4053 struct inode
*inode
;
4055 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4057 return ERR_PTR(-ENOMEM
);
4059 if (inode
->i_state
& I_NEW
) {
4060 BTRFS_I(inode
)->root
= root
;
4061 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4062 btrfs_read_locked_inode(inode
);
4064 inode_tree_add(inode
);
4065 unlock_new_inode(inode
);
4073 static struct inode
*new_simple_dir(struct super_block
*s
,
4074 struct btrfs_key
*key
,
4075 struct btrfs_root
*root
)
4077 struct inode
*inode
= new_inode(s
);
4080 return ERR_PTR(-ENOMEM
);
4082 BTRFS_I(inode
)->root
= root
;
4083 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4084 BTRFS_I(inode
)->dummy_inode
= 1;
4086 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4087 inode
->i_op
= &simple_dir_inode_operations
;
4088 inode
->i_fop
= &simple_dir_operations
;
4089 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4090 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4095 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4097 struct inode
*inode
;
4098 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4099 struct btrfs_root
*sub_root
= root
;
4100 struct btrfs_key location
;
4104 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4105 return ERR_PTR(-ENAMETOOLONG
);
4107 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4110 return ERR_PTR(ret
);
4112 if (location
.objectid
== 0)
4115 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4116 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4120 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4122 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4123 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4124 &location
, &sub_root
);
4127 inode
= ERR_PTR(ret
);
4129 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4131 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4133 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4135 if (!IS_ERR(inode
) && root
!= sub_root
) {
4136 down_read(&root
->fs_info
->cleanup_work_sem
);
4137 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4138 ret
= btrfs_orphan_cleanup(sub_root
);
4139 up_read(&root
->fs_info
->cleanup_work_sem
);
4141 inode
= ERR_PTR(ret
);
4147 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4149 struct btrfs_root
*root
;
4151 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
4152 dentry
= dentry
->d_parent
;
4154 if (dentry
->d_inode
) {
4155 root
= BTRFS_I(dentry
->d_inode
)->root
;
4156 if (btrfs_root_refs(&root
->root_item
) == 0)
4162 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4163 struct nameidata
*nd
)
4165 struct inode
*inode
;
4167 inode
= btrfs_lookup_dentry(dir
, dentry
);
4169 return ERR_CAST(inode
);
4171 return d_splice_alias(inode
, dentry
);
4174 static unsigned char btrfs_filetype_table
[] = {
4175 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4178 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4181 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4182 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4183 struct btrfs_item
*item
;
4184 struct btrfs_dir_item
*di
;
4185 struct btrfs_key key
;
4186 struct btrfs_key found_key
;
4187 struct btrfs_path
*path
;
4190 struct extent_buffer
*leaf
;
4193 unsigned char d_type
;
4198 int key_type
= BTRFS_DIR_INDEX_KEY
;
4203 /* FIXME, use a real flag for deciding about the key type */
4204 if (root
->fs_info
->tree_root
== root
)
4205 key_type
= BTRFS_DIR_ITEM_KEY
;
4207 /* special case for "." */
4208 if (filp
->f_pos
== 0) {
4209 over
= filldir(dirent
, ".", 1,
4216 /* special case for .., just use the back ref */
4217 if (filp
->f_pos
== 1) {
4218 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4219 over
= filldir(dirent
, "..", 2,
4225 path
= btrfs_alloc_path();
4228 btrfs_set_key_type(&key
, key_type
);
4229 key
.offset
= filp
->f_pos
;
4230 key
.objectid
= inode
->i_ino
;
4232 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4238 leaf
= path
->nodes
[0];
4239 nritems
= btrfs_header_nritems(leaf
);
4240 slot
= path
->slots
[0];
4241 if (advance
|| slot
>= nritems
) {
4242 if (slot
>= nritems
- 1) {
4243 ret
= btrfs_next_leaf(root
, path
);
4246 leaf
= path
->nodes
[0];
4247 nritems
= btrfs_header_nritems(leaf
);
4248 slot
= path
->slots
[0];
4256 item
= btrfs_item_nr(leaf
, slot
);
4257 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4259 if (found_key
.objectid
!= key
.objectid
)
4261 if (btrfs_key_type(&found_key
) != key_type
)
4263 if (found_key
.offset
< filp
->f_pos
)
4266 filp
->f_pos
= found_key
.offset
;
4268 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4270 di_total
= btrfs_item_size(leaf
, item
);
4272 while (di_cur
< di_total
) {
4273 struct btrfs_key location
;
4275 if (verify_dir_item(root
, leaf
, di
))
4278 name_len
= btrfs_dir_name_len(leaf
, di
);
4279 if (name_len
<= sizeof(tmp_name
)) {
4280 name_ptr
= tmp_name
;
4282 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4288 read_extent_buffer(leaf
, name_ptr
,
4289 (unsigned long)(di
+ 1), name_len
);
4291 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4292 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4294 /* is this a reference to our own snapshot? If so
4297 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4298 location
.objectid
== root
->root_key
.objectid
) {
4302 over
= filldir(dirent
, name_ptr
, name_len
,
4303 found_key
.offset
, location
.objectid
,
4307 if (name_ptr
!= tmp_name
)
4312 di_len
= btrfs_dir_name_len(leaf
, di
) +
4313 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4315 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4319 /* Reached end of directory/root. Bump pos past the last item. */
4320 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4322 * 32-bit glibc will use getdents64, but then strtol -
4323 * so the last number we can serve is this.
4325 filp
->f_pos
= 0x7fffffff;
4331 btrfs_free_path(path
);
4335 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4337 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4338 struct btrfs_trans_handle
*trans
;
4340 bool nolock
= false;
4342 if (BTRFS_I(inode
)->dummy_inode
)
4346 nolock
= (root
->fs_info
->closing
&& root
== root
->fs_info
->tree_root
);
4348 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4350 trans
= btrfs_join_transaction_nolock(root
, 1);
4352 trans
= btrfs_join_transaction(root
, 1);
4354 return PTR_ERR(trans
);
4355 btrfs_set_trans_block_group(trans
, inode
);
4357 ret
= btrfs_end_transaction_nolock(trans
, root
);
4359 ret
= btrfs_commit_transaction(trans
, root
);
4365 * This is somewhat expensive, updating the tree every time the
4366 * inode changes. But, it is most likely to find the inode in cache.
4367 * FIXME, needs more benchmarking...there are no reasons other than performance
4368 * to keep or drop this code.
4370 void btrfs_dirty_inode(struct inode
*inode
)
4372 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4373 struct btrfs_trans_handle
*trans
;
4376 if (BTRFS_I(inode
)->dummy_inode
)
4379 trans
= btrfs_join_transaction(root
, 1);
4380 BUG_ON(IS_ERR(trans
));
4381 btrfs_set_trans_block_group(trans
, inode
);
4383 ret
= btrfs_update_inode(trans
, root
, inode
);
4384 if (ret
&& ret
== -ENOSPC
) {
4385 /* whoops, lets try again with the full transaction */
4386 btrfs_end_transaction(trans
, root
);
4387 trans
= btrfs_start_transaction(root
, 1);
4388 if (IS_ERR(trans
)) {
4389 if (printk_ratelimit()) {
4390 printk(KERN_ERR
"btrfs: fail to "
4391 "dirty inode %lu error %ld\n",
4392 inode
->i_ino
, PTR_ERR(trans
));
4396 btrfs_set_trans_block_group(trans
, inode
);
4398 ret
= btrfs_update_inode(trans
, root
, inode
);
4400 if (printk_ratelimit()) {
4401 printk(KERN_ERR
"btrfs: fail to "
4402 "dirty inode %lu error %d\n",
4407 btrfs_end_transaction(trans
, root
);
4411 * find the highest existing sequence number in a directory
4412 * and then set the in-memory index_cnt variable to reflect
4413 * free sequence numbers
4415 static int btrfs_set_inode_index_count(struct inode
*inode
)
4417 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4418 struct btrfs_key key
, found_key
;
4419 struct btrfs_path
*path
;
4420 struct extent_buffer
*leaf
;
4423 key
.objectid
= inode
->i_ino
;
4424 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4425 key
.offset
= (u64
)-1;
4427 path
= btrfs_alloc_path();
4431 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4434 /* FIXME: we should be able to handle this */
4440 * MAGIC NUMBER EXPLANATION:
4441 * since we search a directory based on f_pos we have to start at 2
4442 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4443 * else has to start at 2
4445 if (path
->slots
[0] == 0) {
4446 BTRFS_I(inode
)->index_cnt
= 2;
4452 leaf
= path
->nodes
[0];
4453 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4455 if (found_key
.objectid
!= inode
->i_ino
||
4456 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4457 BTRFS_I(inode
)->index_cnt
= 2;
4461 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4463 btrfs_free_path(path
);
4468 * helper to find a free sequence number in a given directory. This current
4469 * code is very simple, later versions will do smarter things in the btree
4471 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4475 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4476 ret
= btrfs_set_inode_index_count(dir
);
4481 *index
= BTRFS_I(dir
)->index_cnt
;
4482 BTRFS_I(dir
)->index_cnt
++;
4487 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4488 struct btrfs_root
*root
,
4490 const char *name
, int name_len
,
4491 u64 ref_objectid
, u64 objectid
,
4492 u64 alloc_hint
, int mode
, u64
*index
)
4494 struct inode
*inode
;
4495 struct btrfs_inode_item
*inode_item
;
4496 struct btrfs_key
*location
;
4497 struct btrfs_path
*path
;
4498 struct btrfs_inode_ref
*ref
;
4499 struct btrfs_key key
[2];
4505 path
= btrfs_alloc_path();
4508 inode
= new_inode(root
->fs_info
->sb
);
4510 return ERR_PTR(-ENOMEM
);
4513 ret
= btrfs_set_inode_index(dir
, index
);
4516 return ERR_PTR(ret
);
4520 * index_cnt is ignored for everything but a dir,
4521 * btrfs_get_inode_index_count has an explanation for the magic
4524 BTRFS_I(inode
)->index_cnt
= 2;
4525 BTRFS_I(inode
)->root
= root
;
4526 BTRFS_I(inode
)->generation
= trans
->transid
;
4527 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4528 btrfs_set_inode_space_info(root
, inode
);
4534 BTRFS_I(inode
)->block_group
=
4535 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
4537 key
[0].objectid
= objectid
;
4538 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4541 key
[1].objectid
= objectid
;
4542 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4543 key
[1].offset
= ref_objectid
;
4545 sizes
[0] = sizeof(struct btrfs_inode_item
);
4546 sizes
[1] = name_len
+ sizeof(*ref
);
4548 path
->leave_spinning
= 1;
4549 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4553 inode_init_owner(inode
, dir
, mode
);
4554 inode
->i_ino
= objectid
;
4555 inode_set_bytes(inode
, 0);
4556 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4557 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4558 struct btrfs_inode_item
);
4559 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4561 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4562 struct btrfs_inode_ref
);
4563 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4564 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4565 ptr
= (unsigned long)(ref
+ 1);
4566 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4568 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4569 btrfs_free_path(path
);
4571 location
= &BTRFS_I(inode
)->location
;
4572 location
->objectid
= objectid
;
4573 location
->offset
= 0;
4574 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4576 btrfs_inherit_iflags(inode
, dir
);
4578 if ((mode
& S_IFREG
)) {
4579 if (btrfs_test_opt(root
, NODATASUM
))
4580 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4581 if (btrfs_test_opt(root
, NODATACOW
))
4582 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4585 insert_inode_hash(inode
);
4586 inode_tree_add(inode
);
4590 BTRFS_I(dir
)->index_cnt
--;
4591 btrfs_free_path(path
);
4593 return ERR_PTR(ret
);
4596 static inline u8
btrfs_inode_type(struct inode
*inode
)
4598 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4602 * utility function to add 'inode' into 'parent_inode' with
4603 * a give name and a given sequence number.
4604 * if 'add_backref' is true, also insert a backref from the
4605 * inode to the parent directory.
4607 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4608 struct inode
*parent_inode
, struct inode
*inode
,
4609 const char *name
, int name_len
, int add_backref
, u64 index
)
4612 struct btrfs_key key
;
4613 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4615 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4616 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4618 key
.objectid
= inode
->i_ino
;
4619 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4623 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4624 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4625 key
.objectid
, root
->root_key
.objectid
,
4626 parent_inode
->i_ino
,
4627 index
, name
, name_len
);
4628 } else if (add_backref
) {
4629 ret
= btrfs_insert_inode_ref(trans
, root
,
4630 name
, name_len
, inode
->i_ino
,
4631 parent_inode
->i_ino
, index
);
4635 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4636 parent_inode
->i_ino
, &key
,
4637 btrfs_inode_type(inode
), index
);
4640 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4642 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4643 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4648 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4649 struct inode
*dir
, struct dentry
*dentry
,
4650 struct inode
*inode
, int backref
, u64 index
)
4652 int err
= btrfs_add_link(trans
, dir
, inode
,
4653 dentry
->d_name
.name
, dentry
->d_name
.len
,
4656 d_instantiate(dentry
, inode
);
4664 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4665 int mode
, dev_t rdev
)
4667 struct btrfs_trans_handle
*trans
;
4668 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4669 struct inode
*inode
= NULL
;
4673 unsigned long nr
= 0;
4676 if (!new_valid_dev(rdev
))
4679 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4684 * 2 for inode item and ref
4686 * 1 for xattr if selinux is on
4688 trans
= btrfs_start_transaction(root
, 5);
4690 return PTR_ERR(trans
);
4692 btrfs_set_trans_block_group(trans
, dir
);
4694 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4695 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4696 BTRFS_I(dir
)->block_group
, mode
, &index
);
4697 err
= PTR_ERR(inode
);
4701 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4707 btrfs_set_trans_block_group(trans
, inode
);
4708 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4712 inode
->i_op
= &btrfs_special_inode_operations
;
4713 init_special_inode(inode
, inode
->i_mode
, rdev
);
4714 btrfs_update_inode(trans
, root
, inode
);
4716 btrfs_update_inode_block_group(trans
, inode
);
4717 btrfs_update_inode_block_group(trans
, dir
);
4719 nr
= trans
->blocks_used
;
4720 btrfs_end_transaction_throttle(trans
, root
);
4721 btrfs_btree_balance_dirty(root
, nr
);
4723 inode_dec_link_count(inode
);
4729 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4730 int mode
, struct nameidata
*nd
)
4732 struct btrfs_trans_handle
*trans
;
4733 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4734 struct inode
*inode
= NULL
;
4737 unsigned long nr
= 0;
4741 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4745 * 2 for inode item and ref
4747 * 1 for xattr if selinux is on
4749 trans
= btrfs_start_transaction(root
, 5);
4751 return PTR_ERR(trans
);
4753 btrfs_set_trans_block_group(trans
, dir
);
4755 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4756 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4757 BTRFS_I(dir
)->block_group
, mode
, &index
);
4758 err
= PTR_ERR(inode
);
4762 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4768 btrfs_set_trans_block_group(trans
, inode
);
4769 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4773 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4774 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4775 inode
->i_fop
= &btrfs_file_operations
;
4776 inode
->i_op
= &btrfs_file_inode_operations
;
4777 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4779 btrfs_update_inode_block_group(trans
, inode
);
4780 btrfs_update_inode_block_group(trans
, dir
);
4782 nr
= trans
->blocks_used
;
4783 btrfs_end_transaction_throttle(trans
, root
);
4785 inode_dec_link_count(inode
);
4788 btrfs_btree_balance_dirty(root
, nr
);
4792 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4793 struct dentry
*dentry
)
4795 struct btrfs_trans_handle
*trans
;
4796 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4797 struct inode
*inode
= old_dentry
->d_inode
;
4799 unsigned long nr
= 0;
4803 if (inode
->i_nlink
== 0)
4806 /* do not allow sys_link's with other subvols of the same device */
4807 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4810 btrfs_inc_nlink(inode
);
4811 inode
->i_ctime
= CURRENT_TIME
;
4813 err
= btrfs_set_inode_index(dir
, &index
);
4818 * 2 items for inode and inode ref
4819 * 2 items for dir items
4820 * 1 item for parent inode
4822 trans
= btrfs_start_transaction(root
, 5);
4823 if (IS_ERR(trans
)) {
4824 err
= PTR_ERR(trans
);
4828 btrfs_set_trans_block_group(trans
, dir
);
4831 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4836 struct dentry
*parent
= dget_parent(dentry
);
4837 btrfs_update_inode_block_group(trans
, dir
);
4838 err
= btrfs_update_inode(trans
, root
, inode
);
4840 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4844 nr
= trans
->blocks_used
;
4845 btrfs_end_transaction_throttle(trans
, root
);
4848 inode_dec_link_count(inode
);
4851 btrfs_btree_balance_dirty(root
, nr
);
4855 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4857 struct inode
*inode
= NULL
;
4858 struct btrfs_trans_handle
*trans
;
4859 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4861 int drop_on_err
= 0;
4864 unsigned long nr
= 1;
4866 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4871 * 2 items for inode and ref
4872 * 2 items for dir items
4873 * 1 for xattr if selinux is on
4875 trans
= btrfs_start_transaction(root
, 5);
4877 return PTR_ERR(trans
);
4878 btrfs_set_trans_block_group(trans
, dir
);
4880 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4881 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4882 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
4884 if (IS_ERR(inode
)) {
4885 err
= PTR_ERR(inode
);
4891 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4895 inode
->i_op
= &btrfs_dir_inode_operations
;
4896 inode
->i_fop
= &btrfs_dir_file_operations
;
4897 btrfs_set_trans_block_group(trans
, inode
);
4899 btrfs_i_size_write(inode
, 0);
4900 err
= btrfs_update_inode(trans
, root
, inode
);
4904 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4905 dentry
->d_name
.len
, 0, index
);
4909 d_instantiate(dentry
, inode
);
4911 btrfs_update_inode_block_group(trans
, inode
);
4912 btrfs_update_inode_block_group(trans
, dir
);
4915 nr
= trans
->blocks_used
;
4916 btrfs_end_transaction_throttle(trans
, root
);
4919 btrfs_btree_balance_dirty(root
, nr
);
4923 /* helper for btfs_get_extent. Given an existing extent in the tree,
4924 * and an extent that you want to insert, deal with overlap and insert
4925 * the new extent into the tree.
4927 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4928 struct extent_map
*existing
,
4929 struct extent_map
*em
,
4930 u64 map_start
, u64 map_len
)
4934 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4935 start_diff
= map_start
- em
->start
;
4936 em
->start
= map_start
;
4938 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4939 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4940 em
->block_start
+= start_diff
;
4941 em
->block_len
-= start_diff
;
4943 return add_extent_mapping(em_tree
, em
);
4946 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4947 struct inode
*inode
, struct page
*page
,
4948 size_t pg_offset
, u64 extent_offset
,
4949 struct btrfs_file_extent_item
*item
)
4952 struct extent_buffer
*leaf
= path
->nodes
[0];
4955 unsigned long inline_size
;
4959 WARN_ON(pg_offset
!= 0);
4960 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4961 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4962 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4963 btrfs_item_nr(leaf
, path
->slots
[0]));
4964 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4965 ptr
= btrfs_file_extent_inline_start(item
);
4967 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4969 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4970 ret
= btrfs_decompress(compress_type
, tmp
, page
,
4971 extent_offset
, inline_size
, max_size
);
4973 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4974 unsigned long copy_size
= min_t(u64
,
4975 PAGE_CACHE_SIZE
- pg_offset
,
4976 max_size
- extent_offset
);
4977 memset(kaddr
+ pg_offset
, 0, copy_size
);
4978 kunmap_atomic(kaddr
, KM_USER0
);
4985 * a bit scary, this does extent mapping from logical file offset to the disk.
4986 * the ugly parts come from merging extents from the disk with the in-ram
4987 * representation. This gets more complex because of the data=ordered code,
4988 * where the in-ram extents might be locked pending data=ordered completion.
4990 * This also copies inline extents directly into the page.
4993 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
4994 size_t pg_offset
, u64 start
, u64 len
,
5000 u64 extent_start
= 0;
5002 u64 objectid
= inode
->i_ino
;
5004 struct btrfs_path
*path
= NULL
;
5005 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5006 struct btrfs_file_extent_item
*item
;
5007 struct extent_buffer
*leaf
;
5008 struct btrfs_key found_key
;
5009 struct extent_map
*em
= NULL
;
5010 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5011 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5012 struct btrfs_trans_handle
*trans
= NULL
;
5016 read_lock(&em_tree
->lock
);
5017 em
= lookup_extent_mapping(em_tree
, start
, len
);
5019 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5020 read_unlock(&em_tree
->lock
);
5023 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5024 free_extent_map(em
);
5025 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5026 free_extent_map(em
);
5030 em
= alloc_extent_map(GFP_NOFS
);
5035 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5036 em
->start
= EXTENT_MAP_HOLE
;
5037 em
->orig_start
= EXTENT_MAP_HOLE
;
5039 em
->block_len
= (u64
)-1;
5042 path
= btrfs_alloc_path();
5046 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5047 objectid
, start
, trans
!= NULL
);
5054 if (path
->slots
[0] == 0)
5059 leaf
= path
->nodes
[0];
5060 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5061 struct btrfs_file_extent_item
);
5062 /* are we inside the extent that was found? */
5063 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5064 found_type
= btrfs_key_type(&found_key
);
5065 if (found_key
.objectid
!= objectid
||
5066 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5070 found_type
= btrfs_file_extent_type(leaf
, item
);
5071 extent_start
= found_key
.offset
;
5072 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5073 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5074 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5075 extent_end
= extent_start
+
5076 btrfs_file_extent_num_bytes(leaf
, item
);
5077 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5079 size
= btrfs_file_extent_inline_len(leaf
, item
);
5080 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5081 ~((u64
)root
->sectorsize
- 1);
5084 if (start
>= extent_end
) {
5086 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5087 ret
= btrfs_next_leaf(root
, path
);
5094 leaf
= path
->nodes
[0];
5096 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5097 if (found_key
.objectid
!= objectid
||
5098 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5100 if (start
+ len
<= found_key
.offset
)
5103 em
->len
= found_key
.offset
- start
;
5107 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5108 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5109 em
->start
= extent_start
;
5110 em
->len
= extent_end
- extent_start
;
5111 em
->orig_start
= extent_start
-
5112 btrfs_file_extent_offset(leaf
, item
);
5113 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5115 em
->block_start
= EXTENT_MAP_HOLE
;
5118 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5119 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5120 em
->compress_type
= compress_type
;
5121 em
->block_start
= bytenr
;
5122 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5125 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5126 em
->block_start
= bytenr
;
5127 em
->block_len
= em
->len
;
5128 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5129 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5132 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5136 size_t extent_offset
;
5139 em
->block_start
= EXTENT_MAP_INLINE
;
5140 if (!page
|| create
) {
5141 em
->start
= extent_start
;
5142 em
->len
= extent_end
- extent_start
;
5146 size
= btrfs_file_extent_inline_len(leaf
, item
);
5147 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5148 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5149 size
- extent_offset
);
5150 em
->start
= extent_start
+ extent_offset
;
5151 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5152 ~((u64
)root
->sectorsize
- 1);
5153 em
->orig_start
= EXTENT_MAP_INLINE
;
5154 if (compress_type
) {
5155 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5156 em
->compress_type
= compress_type
;
5158 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5159 if (create
== 0 && !PageUptodate(page
)) {
5160 if (btrfs_file_extent_compression(leaf
, item
) !=
5161 BTRFS_COMPRESS_NONE
) {
5162 ret
= uncompress_inline(path
, inode
, page
,
5164 extent_offset
, item
);
5168 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5170 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5171 memset(map
+ pg_offset
+ copy_size
, 0,
5172 PAGE_CACHE_SIZE
- pg_offset
-
5177 flush_dcache_page(page
);
5178 } else if (create
&& PageUptodate(page
)) {
5182 free_extent_map(em
);
5184 btrfs_release_path(root
, path
);
5185 trans
= btrfs_join_transaction(root
, 1);
5187 return ERR_CAST(trans
);
5191 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5194 btrfs_mark_buffer_dirty(leaf
);
5196 set_extent_uptodate(io_tree
, em
->start
,
5197 extent_map_end(em
) - 1, GFP_NOFS
);
5200 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5207 em
->block_start
= EXTENT_MAP_HOLE
;
5208 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5210 btrfs_release_path(root
, path
);
5211 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5212 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5213 "[%llu %llu]\n", (unsigned long long)em
->start
,
5214 (unsigned long long)em
->len
,
5215 (unsigned long long)start
,
5216 (unsigned long long)len
);
5222 write_lock(&em_tree
->lock
);
5223 ret
= add_extent_mapping(em_tree
, em
);
5224 /* it is possible that someone inserted the extent into the tree
5225 * while we had the lock dropped. It is also possible that
5226 * an overlapping map exists in the tree
5228 if (ret
== -EEXIST
) {
5229 struct extent_map
*existing
;
5233 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5234 if (existing
&& (existing
->start
> start
||
5235 existing
->start
+ existing
->len
<= start
)) {
5236 free_extent_map(existing
);
5240 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5243 err
= merge_extent_mapping(em_tree
, existing
,
5246 free_extent_map(existing
);
5248 free_extent_map(em
);
5253 free_extent_map(em
);
5257 free_extent_map(em
);
5262 write_unlock(&em_tree
->lock
);
5265 btrfs_free_path(path
);
5267 ret
= btrfs_end_transaction(trans
, root
);
5272 free_extent_map(em
);
5273 return ERR_PTR(err
);
5278 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5279 size_t pg_offset
, u64 start
, u64 len
,
5282 struct extent_map
*em
;
5283 struct extent_map
*hole_em
= NULL
;
5284 u64 range_start
= start
;
5290 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5295 * if our em maps to a hole, there might
5296 * actually be delalloc bytes behind it
5298 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5304 /* check to see if we've wrapped (len == -1 or similar) */
5313 /* ok, we didn't find anything, lets look for delalloc */
5314 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5315 end
, len
, EXTENT_DELALLOC
, 1);
5316 found_end
= range_start
+ found
;
5317 if (found_end
< range_start
)
5318 found_end
= (u64
)-1;
5321 * we didn't find anything useful, return
5322 * the original results from get_extent()
5324 if (range_start
> end
|| found_end
<= start
) {
5330 /* adjust the range_start to make sure it doesn't
5331 * go backwards from the start they passed in
5333 range_start
= max(start
,range_start
);
5334 found
= found_end
- range_start
;
5337 u64 hole_start
= start
;
5340 em
= alloc_extent_map(GFP_NOFS
);
5346 * when btrfs_get_extent can't find anything it
5347 * returns one huge hole
5349 * make sure what it found really fits our range, and
5350 * adjust to make sure it is based on the start from
5354 u64 calc_end
= extent_map_end(hole_em
);
5356 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5357 free_extent_map(hole_em
);
5360 hole_start
= max(hole_em
->start
, start
);
5361 hole_len
= calc_end
- hole_start
;
5365 if (hole_em
&& range_start
> hole_start
) {
5366 /* our hole starts before our delalloc, so we
5367 * have to return just the parts of the hole
5368 * that go until the delalloc starts
5370 em
->len
= min(hole_len
,
5371 range_start
- hole_start
);
5372 em
->start
= hole_start
;
5373 em
->orig_start
= hole_start
;
5375 * don't adjust block start at all,
5376 * it is fixed at EXTENT_MAP_HOLE
5378 em
->block_start
= hole_em
->block_start
;
5379 em
->block_len
= hole_len
;
5381 em
->start
= range_start
;
5383 em
->orig_start
= range_start
;
5384 em
->block_start
= EXTENT_MAP_DELALLOC
;
5385 em
->block_len
= found
;
5387 } else if (hole_em
) {
5392 free_extent_map(hole_em
);
5394 free_extent_map(em
);
5395 return ERR_PTR(err
);
5400 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5403 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5404 struct btrfs_trans_handle
*trans
;
5405 struct extent_map
*em
;
5406 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5407 struct btrfs_key ins
;
5411 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5413 trans
= btrfs_join_transaction(root
, 0);
5415 return ERR_CAST(trans
);
5417 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5419 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5420 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5421 alloc_hint
, (u64
)-1, &ins
, 1);
5427 em
= alloc_extent_map(GFP_NOFS
);
5429 em
= ERR_PTR(-ENOMEM
);
5434 em
->orig_start
= em
->start
;
5435 em
->len
= ins
.offset
;
5437 em
->block_start
= ins
.objectid
;
5438 em
->block_len
= ins
.offset
;
5439 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5440 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5443 write_lock(&em_tree
->lock
);
5444 ret
= add_extent_mapping(em_tree
, em
);
5445 write_unlock(&em_tree
->lock
);
5448 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5451 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5452 ins
.offset
, ins
.offset
, 0);
5454 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5458 btrfs_end_transaction(trans
, root
);
5463 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5464 * block must be cow'd
5466 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5467 struct inode
*inode
, u64 offset
, u64 len
)
5469 struct btrfs_path
*path
;
5471 struct extent_buffer
*leaf
;
5472 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5473 struct btrfs_file_extent_item
*fi
;
5474 struct btrfs_key key
;
5482 path
= btrfs_alloc_path();
5486 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
5491 slot
= path
->slots
[0];
5494 /* can't find the item, must cow */
5501 leaf
= path
->nodes
[0];
5502 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5503 if (key
.objectid
!= inode
->i_ino
||
5504 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5505 /* not our file or wrong item type, must cow */
5509 if (key
.offset
> offset
) {
5510 /* Wrong offset, must cow */
5514 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5515 found_type
= btrfs_file_extent_type(leaf
, fi
);
5516 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5517 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5518 /* not a regular extent, must cow */
5521 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5522 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5524 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5525 if (extent_end
< offset
+ len
) {
5526 /* extent doesn't include our full range, must cow */
5530 if (btrfs_extent_readonly(root
, disk_bytenr
))
5534 * look for other files referencing this extent, if we
5535 * find any we must cow
5537 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
5538 key
.offset
- backref_offset
, disk_bytenr
))
5542 * adjust disk_bytenr and num_bytes to cover just the bytes
5543 * in this extent we are about to write. If there
5544 * are any csums in that range we have to cow in order
5545 * to keep the csums correct
5547 disk_bytenr
+= backref_offset
;
5548 disk_bytenr
+= offset
- key
.offset
;
5549 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5550 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5553 * all of the above have passed, it is safe to overwrite this extent
5558 btrfs_free_path(path
);
5562 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5563 struct buffer_head
*bh_result
, int create
)
5565 struct extent_map
*em
;
5566 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5567 u64 start
= iblock
<< inode
->i_blkbits
;
5568 u64 len
= bh_result
->b_size
;
5569 struct btrfs_trans_handle
*trans
;
5571 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5576 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5577 * io. INLINE is special, and we could probably kludge it in here, but
5578 * it's still buffered so for safety lets just fall back to the generic
5581 * For COMPRESSED we _have_ to read the entire extent in so we can
5582 * decompress it, so there will be buffering required no matter what we
5583 * do, so go ahead and fallback to buffered.
5585 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5586 * to buffered IO. Don't blame me, this is the price we pay for using
5589 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5590 em
->block_start
== EXTENT_MAP_INLINE
) {
5591 free_extent_map(em
);
5595 /* Just a good old fashioned hole, return */
5596 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5597 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5598 free_extent_map(em
);
5599 /* DIO will do one hole at a time, so just unlock a sector */
5600 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5601 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5606 * We don't allocate a new extent in the following cases
5608 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5610 * 2) The extent is marked as PREALLOC. We're good to go here and can
5611 * just use the extent.
5615 len
= em
->len
- (start
- em
->start
);
5619 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5620 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5621 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5626 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5627 type
= BTRFS_ORDERED_PREALLOC
;
5629 type
= BTRFS_ORDERED_NOCOW
;
5630 len
= min(len
, em
->len
- (start
- em
->start
));
5631 block_start
= em
->block_start
+ (start
- em
->start
);
5634 * we're not going to log anything, but we do need
5635 * to make sure the current transaction stays open
5636 * while we look for nocow cross refs
5638 trans
= btrfs_join_transaction(root
, 0);
5642 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5643 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5644 block_start
, len
, len
, type
);
5645 btrfs_end_transaction(trans
, root
);
5647 free_extent_map(em
);
5652 btrfs_end_transaction(trans
, root
);
5656 * this will cow the extent, reset the len in case we changed
5659 len
= bh_result
->b_size
;
5660 free_extent_map(em
);
5661 em
= btrfs_new_extent_direct(inode
, start
, len
);
5664 len
= min(len
, em
->len
- (start
- em
->start
));
5666 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5667 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5670 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5672 bh_result
->b_size
= len
;
5673 bh_result
->b_bdev
= em
->bdev
;
5674 set_buffer_mapped(bh_result
);
5675 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5676 set_buffer_new(bh_result
);
5678 free_extent_map(em
);
5683 struct btrfs_dio_private
{
5684 struct inode
*inode
;
5691 /* number of bios pending for this dio */
5692 atomic_t pending_bios
;
5697 struct bio
*orig_bio
;
5700 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5702 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5703 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5704 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5705 struct inode
*inode
= dip
->inode
;
5706 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5708 u32
*private = dip
->csums
;
5710 start
= dip
->logical_offset
;
5712 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5713 struct page
*page
= bvec
->bv_page
;
5716 unsigned long flags
;
5718 local_irq_save(flags
);
5719 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5720 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5721 csum
, bvec
->bv_len
);
5722 btrfs_csum_final(csum
, (char *)&csum
);
5723 kunmap_atomic(kaddr
, KM_IRQ0
);
5724 local_irq_restore(flags
);
5726 flush_dcache_page(bvec
->bv_page
);
5727 if (csum
!= *private) {
5728 printk(KERN_ERR
"btrfs csum failed ino %lu off"
5729 " %llu csum %u private %u\n",
5730 inode
->i_ino
, (unsigned long long)start
,
5736 start
+= bvec
->bv_len
;
5739 } while (bvec
<= bvec_end
);
5741 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5742 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5743 bio
->bi_private
= dip
->private;
5747 dio_end_io(bio
, err
);
5750 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5752 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5753 struct inode
*inode
= dip
->inode
;
5754 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5755 struct btrfs_trans_handle
*trans
;
5756 struct btrfs_ordered_extent
*ordered
= NULL
;
5757 struct extent_state
*cached_state
= NULL
;
5758 u64 ordered_offset
= dip
->logical_offset
;
5759 u64 ordered_bytes
= dip
->bytes
;
5765 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5773 trans
= btrfs_join_transaction(root
, 1);
5774 if (IS_ERR(trans
)) {
5778 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5780 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5781 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5783 ret
= btrfs_update_inode(trans
, root
, inode
);
5788 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5789 ordered
->file_offset
+ ordered
->len
- 1, 0,
5790 &cached_state
, GFP_NOFS
);
5792 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5793 ret
= btrfs_mark_extent_written(trans
, inode
,
5794 ordered
->file_offset
,
5795 ordered
->file_offset
+
5802 ret
= insert_reserved_file_extent(trans
, inode
,
5803 ordered
->file_offset
,
5809 BTRFS_FILE_EXTENT_REG
);
5810 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5811 ordered
->file_offset
, ordered
->len
);
5819 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5820 btrfs_ordered_update_i_size(inode
, 0, ordered
);
5821 btrfs_update_inode(trans
, root
, inode
);
5823 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5824 ordered
->file_offset
+ ordered
->len
- 1,
5825 &cached_state
, GFP_NOFS
);
5827 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5828 btrfs_end_transaction(trans
, root
);
5829 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5830 btrfs_put_ordered_extent(ordered
);
5831 btrfs_put_ordered_extent(ordered
);
5835 * our bio might span multiple ordered extents. If we haven't
5836 * completed the accounting for the whole dio, go back and try again
5838 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5839 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5844 bio
->bi_private
= dip
->private;
5848 dio_end_io(bio
, err
);
5851 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5852 struct bio
*bio
, int mirror_num
,
5853 unsigned long bio_flags
, u64 offset
)
5856 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5857 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5862 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5864 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5867 printk(KERN_ERR
"btrfs direct IO failed ino %lu rw %lu "
5868 "sector %#Lx len %u err no %d\n",
5869 dip
->inode
->i_ino
, bio
->bi_rw
,
5870 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5874 * before atomic variable goto zero, we must make sure
5875 * dip->errors is perceived to be set.
5877 smp_mb__before_atomic_dec();
5880 /* if there are more bios still pending for this dio, just exit */
5881 if (!atomic_dec_and_test(&dip
->pending_bios
))
5885 bio_io_error(dip
->orig_bio
);
5887 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5888 bio_endio(dip
->orig_bio
, 0);
5894 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5895 u64 first_sector
, gfp_t gfp_flags
)
5897 int nr_vecs
= bio_get_nr_vecs(bdev
);
5898 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5901 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5902 int rw
, u64 file_offset
, int skip_sum
,
5905 int write
= rw
& REQ_WRITE
;
5906 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5910 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5914 if (write
&& !skip_sum
) {
5915 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5916 inode
, rw
, bio
, 0, 0,
5918 __btrfs_submit_bio_start_direct_io
,
5919 __btrfs_submit_bio_done
);
5921 } else if (!skip_sum
)
5922 btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5923 file_offset
, csums
);
5925 ret
= btrfs_map_bio(root
, rw
, bio
, 0, 1);
5931 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5934 struct inode
*inode
= dip
->inode
;
5935 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5936 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5938 struct bio
*orig_bio
= dip
->orig_bio
;
5939 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5940 u64 start_sector
= orig_bio
->bi_sector
;
5941 u64 file_offset
= dip
->logical_offset
;
5945 u32
*csums
= dip
->csums
;
5947 int write
= rw
& REQ_WRITE
;
5949 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
5952 bio
->bi_private
= dip
;
5953 bio
->bi_end_io
= btrfs_end_dio_bio
;
5954 atomic_inc(&dip
->pending_bios
);
5956 map_length
= orig_bio
->bi_size
;
5957 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5958 &map_length
, NULL
, 0);
5964 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
5965 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
5966 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
5967 bvec
->bv_offset
) < bvec
->bv_len
)) {
5969 * inc the count before we submit the bio so
5970 * we know the end IO handler won't happen before
5971 * we inc the count. Otherwise, the dip might get freed
5972 * before we're done setting it up
5974 atomic_inc(&dip
->pending_bios
);
5975 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
5976 file_offset
, skip_sum
,
5980 atomic_dec(&dip
->pending_bios
);
5984 /* Write's use the ordered csums */
5985 if (!write
&& !skip_sum
)
5986 csums
= csums
+ nr_pages
;
5987 start_sector
+= submit_len
>> 9;
5988 file_offset
+= submit_len
;
5993 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
5994 start_sector
, GFP_NOFS
);
5997 bio
->bi_private
= dip
;
5998 bio
->bi_end_io
= btrfs_end_dio_bio
;
6000 map_length
= orig_bio
->bi_size
;
6001 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6002 &map_length
, NULL
, 0);
6008 submit_len
+= bvec
->bv_len
;
6014 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6023 * before atomic variable goto zero, we must
6024 * make sure dip->errors is perceived to be set.
6026 smp_mb__before_atomic_dec();
6027 if (atomic_dec_and_test(&dip
->pending_bios
))
6028 bio_io_error(dip
->orig_bio
);
6030 /* bio_end_io() will handle error, so we needn't return it */
6034 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6037 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6038 struct btrfs_dio_private
*dip
;
6039 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6041 int write
= rw
& REQ_WRITE
;
6044 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6046 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6053 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6054 if (!write
&& !skip_sum
) {
6055 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6063 dip
->private = bio
->bi_private
;
6065 dip
->logical_offset
= file_offset
;
6069 dip
->bytes
+= bvec
->bv_len
;
6071 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6073 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6074 bio
->bi_private
= dip
;
6076 dip
->orig_bio
= bio
;
6077 atomic_set(&dip
->pending_bios
, 0);
6080 bio
->bi_end_io
= btrfs_endio_direct_write
;
6082 bio
->bi_end_io
= btrfs_endio_direct_read
;
6084 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6089 * If this is a write, we need to clean up the reserved space and kill
6090 * the ordered extent.
6093 struct btrfs_ordered_extent
*ordered
;
6094 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6095 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6096 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6097 btrfs_free_reserved_extent(root
, ordered
->start
,
6099 btrfs_put_ordered_extent(ordered
);
6100 btrfs_put_ordered_extent(ordered
);
6102 bio_endio(bio
, ret
);
6105 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6106 const struct iovec
*iov
, loff_t offset
,
6107 unsigned long nr_segs
)
6112 unsigned blocksize_mask
= root
->sectorsize
- 1;
6113 ssize_t retval
= -EINVAL
;
6114 loff_t end
= offset
;
6116 if (offset
& blocksize_mask
)
6119 /* Check the memory alignment. Blocks cannot straddle pages */
6120 for (seg
= 0; seg
< nr_segs
; seg
++) {
6121 addr
= (unsigned long)iov
[seg
].iov_base
;
6122 size
= iov
[seg
].iov_len
;
6124 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6131 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6132 const struct iovec
*iov
, loff_t offset
,
6133 unsigned long nr_segs
)
6135 struct file
*file
= iocb
->ki_filp
;
6136 struct inode
*inode
= file
->f_mapping
->host
;
6137 struct btrfs_ordered_extent
*ordered
;
6138 struct extent_state
*cached_state
= NULL
;
6139 u64 lockstart
, lockend
;
6141 int writing
= rw
& WRITE
;
6143 size_t count
= iov_length(iov
, nr_segs
);
6145 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6151 lockend
= offset
+ count
- 1;
6154 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6160 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6161 0, &cached_state
, GFP_NOFS
);
6163 * We're concerned with the entire range that we're going to be
6164 * doing DIO to, so we need to make sure theres no ordered
6165 * extents in this range.
6167 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6168 lockend
- lockstart
+ 1);
6171 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6172 &cached_state
, GFP_NOFS
);
6173 btrfs_start_ordered_extent(inode
, ordered
, 1);
6174 btrfs_put_ordered_extent(ordered
);
6179 * we don't use btrfs_set_extent_delalloc because we don't want
6180 * the dirty or uptodate bits
6183 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6184 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6185 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6188 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6189 lockend
, EXTENT_LOCKED
| write_bits
,
6190 1, 0, &cached_state
, GFP_NOFS
);
6195 free_extent_state(cached_state
);
6196 cached_state
= NULL
;
6198 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6199 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6200 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6201 btrfs_submit_direct
, 0);
6203 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6204 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6205 offset
+ iov_length(iov
, nr_segs
) - 1,
6206 EXTENT_LOCKED
| write_bits
, 1, 0,
6207 &cached_state
, GFP_NOFS
);
6208 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6210 * We're falling back to buffered, unlock the section we didn't
6213 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6214 offset
+ iov_length(iov
, nr_segs
) - 1,
6215 EXTENT_LOCKED
| write_bits
, 1, 0,
6216 &cached_state
, GFP_NOFS
);
6219 free_extent_state(cached_state
);
6223 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6224 __u64 start
, __u64 len
)
6226 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6229 int btrfs_readpage(struct file
*file
, struct page
*page
)
6231 struct extent_io_tree
*tree
;
6232 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6233 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6236 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6238 struct extent_io_tree
*tree
;
6241 if (current
->flags
& PF_MEMALLOC
) {
6242 redirty_page_for_writepage(wbc
, page
);
6246 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6247 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6250 int btrfs_writepages(struct address_space
*mapping
,
6251 struct writeback_control
*wbc
)
6253 struct extent_io_tree
*tree
;
6255 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6256 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6260 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6261 struct list_head
*pages
, unsigned nr_pages
)
6263 struct extent_io_tree
*tree
;
6264 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6265 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6268 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6270 struct extent_io_tree
*tree
;
6271 struct extent_map_tree
*map
;
6274 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6275 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6276 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6278 ClearPagePrivate(page
);
6279 set_page_private(page
, 0);
6280 page_cache_release(page
);
6285 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6287 if (PageWriteback(page
) || PageDirty(page
))
6289 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6292 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6294 struct extent_io_tree
*tree
;
6295 struct btrfs_ordered_extent
*ordered
;
6296 struct extent_state
*cached_state
= NULL
;
6297 u64 page_start
= page_offset(page
);
6298 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6302 * we have the page locked, so new writeback can't start,
6303 * and the dirty bit won't be cleared while we are here.
6305 * Wait for IO on this page so that we can safely clear
6306 * the PagePrivate2 bit and do ordered accounting
6308 wait_on_page_writeback(page
);
6310 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6312 btrfs_releasepage(page
, GFP_NOFS
);
6315 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6317 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6321 * IO on this page will never be started, so we need
6322 * to account for any ordered extents now
6324 clear_extent_bit(tree
, page_start
, page_end
,
6325 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6326 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6327 &cached_state
, GFP_NOFS
);
6329 * whoever cleared the private bit is responsible
6330 * for the finish_ordered_io
6332 if (TestClearPagePrivate2(page
)) {
6333 btrfs_finish_ordered_io(page
->mapping
->host
,
6334 page_start
, page_end
);
6336 btrfs_put_ordered_extent(ordered
);
6337 cached_state
= NULL
;
6338 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6341 clear_extent_bit(tree
, page_start
, page_end
,
6342 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6343 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6344 __btrfs_releasepage(page
, GFP_NOFS
);
6346 ClearPageChecked(page
);
6347 if (PagePrivate(page
)) {
6348 ClearPagePrivate(page
);
6349 set_page_private(page
, 0);
6350 page_cache_release(page
);
6355 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6356 * called from a page fault handler when a page is first dirtied. Hence we must
6357 * be careful to check for EOF conditions here. We set the page up correctly
6358 * for a written page which means we get ENOSPC checking when writing into
6359 * holes and correct delalloc and unwritten extent mapping on filesystems that
6360 * support these features.
6362 * We are not allowed to take the i_mutex here so we have to play games to
6363 * protect against truncate races as the page could now be beyond EOF. Because
6364 * vmtruncate() writes the inode size before removing pages, once we have the
6365 * page lock we can determine safely if the page is beyond EOF. If it is not
6366 * beyond EOF, then the page is guaranteed safe against truncation until we
6369 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6371 struct page
*page
= vmf
->page
;
6372 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6373 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6374 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6375 struct btrfs_ordered_extent
*ordered
;
6376 struct extent_state
*cached_state
= NULL
;
6378 unsigned long zero_start
;
6384 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6388 else /* -ENOSPC, -EIO, etc */
6389 ret
= VM_FAULT_SIGBUS
;
6393 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6396 size
= i_size_read(inode
);
6397 page_start
= page_offset(page
);
6398 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6400 if ((page
->mapping
!= inode
->i_mapping
) ||
6401 (page_start
>= size
)) {
6402 /* page got truncated out from underneath us */
6405 wait_on_page_writeback(page
);
6407 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6409 set_page_extent_mapped(page
);
6412 * we can't set the delalloc bits if there are pending ordered
6413 * extents. Drop our locks and wait for them to finish
6415 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6417 unlock_extent_cached(io_tree
, page_start
, page_end
,
6418 &cached_state
, GFP_NOFS
);
6420 btrfs_start_ordered_extent(inode
, ordered
, 1);
6421 btrfs_put_ordered_extent(ordered
);
6426 * XXX - page_mkwrite gets called every time the page is dirtied, even
6427 * if it was already dirty, so for space accounting reasons we need to
6428 * clear any delalloc bits for the range we are fixing to save. There
6429 * is probably a better way to do this, but for now keep consistent with
6430 * prepare_pages in the normal write path.
6432 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6433 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6434 0, 0, &cached_state
, GFP_NOFS
);
6436 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6439 unlock_extent_cached(io_tree
, page_start
, page_end
,
6440 &cached_state
, GFP_NOFS
);
6441 ret
= VM_FAULT_SIGBUS
;
6446 /* page is wholly or partially inside EOF */
6447 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6448 zero_start
= size
& ~PAGE_CACHE_MASK
;
6450 zero_start
= PAGE_CACHE_SIZE
;
6452 if (zero_start
!= PAGE_CACHE_SIZE
) {
6454 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6455 flush_dcache_page(page
);
6458 ClearPageChecked(page
);
6459 set_page_dirty(page
);
6460 SetPageUptodate(page
);
6462 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6463 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6465 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6469 return VM_FAULT_LOCKED
;
6471 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6476 static int btrfs_truncate(struct inode
*inode
)
6478 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6481 struct btrfs_trans_handle
*trans
;
6483 u64 mask
= root
->sectorsize
- 1;
6485 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6489 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6490 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6492 trans
= btrfs_start_transaction(root
, 5);
6494 return PTR_ERR(trans
);
6496 btrfs_set_trans_block_group(trans
, inode
);
6498 ret
= btrfs_orphan_add(trans
, inode
);
6500 btrfs_end_transaction(trans
, root
);
6504 nr
= trans
->blocks_used
;
6505 btrfs_end_transaction(trans
, root
);
6506 btrfs_btree_balance_dirty(root
, nr
);
6508 /* Now start a transaction for the truncate */
6509 trans
= btrfs_start_transaction(root
, 0);
6511 return PTR_ERR(trans
);
6512 btrfs_set_trans_block_group(trans
, inode
);
6513 trans
->block_rsv
= root
->orphan_block_rsv
;
6516 * setattr is responsible for setting the ordered_data_close flag,
6517 * but that is only tested during the last file release. That
6518 * could happen well after the next commit, leaving a great big
6519 * window where new writes may get lost if someone chooses to write
6520 * to this file after truncating to zero
6522 * The inode doesn't have any dirty data here, and so if we commit
6523 * this is a noop. If someone immediately starts writing to the inode
6524 * it is very likely we'll catch some of their writes in this
6525 * transaction, and the commit will find this file on the ordered
6526 * data list with good things to send down.
6528 * This is a best effort solution, there is still a window where
6529 * using truncate to replace the contents of the file will
6530 * end up with a zero length file after a crash.
6532 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6533 btrfs_add_ordered_operation(trans
, root
, inode
);
6537 trans
= btrfs_start_transaction(root
, 0);
6539 return PTR_ERR(trans
);
6540 btrfs_set_trans_block_group(trans
, inode
);
6541 trans
->block_rsv
= root
->orphan_block_rsv
;
6544 ret
= btrfs_block_rsv_check(trans
, root
,
6545 root
->orphan_block_rsv
, 0, 5);
6546 if (ret
== -EAGAIN
) {
6547 ret
= btrfs_commit_transaction(trans
, root
);
6557 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6559 BTRFS_EXTENT_DATA_KEY
);
6560 if (ret
!= -EAGAIN
) {
6565 ret
= btrfs_update_inode(trans
, root
, inode
);
6571 nr
= trans
->blocks_used
;
6572 btrfs_end_transaction(trans
, root
);
6574 btrfs_btree_balance_dirty(root
, nr
);
6577 if (ret
== 0 && inode
->i_nlink
> 0) {
6578 ret
= btrfs_orphan_del(trans
, inode
);
6581 } else if (ret
&& inode
->i_nlink
> 0) {
6583 * Failed to do the truncate, remove us from the in memory
6586 ret
= btrfs_orphan_del(NULL
, inode
);
6589 ret
= btrfs_update_inode(trans
, root
, inode
);
6593 nr
= trans
->blocks_used
;
6594 ret
= btrfs_end_transaction_throttle(trans
, root
);
6597 btrfs_btree_balance_dirty(root
, nr
);
6603 * create a new subvolume directory/inode (helper for the ioctl).
6605 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6606 struct btrfs_root
*new_root
,
6607 u64 new_dirid
, u64 alloc_hint
)
6609 struct inode
*inode
;
6613 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6614 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
6616 return PTR_ERR(inode
);
6617 inode
->i_op
= &btrfs_dir_inode_operations
;
6618 inode
->i_fop
= &btrfs_dir_file_operations
;
6621 btrfs_i_size_write(inode
, 0);
6623 err
= btrfs_update_inode(trans
, new_root
, inode
);
6630 /* helper function for file defrag and space balancing. This
6631 * forces readahead on a given range of bytes in an inode
6633 unsigned long btrfs_force_ra(struct address_space
*mapping
,
6634 struct file_ra_state
*ra
, struct file
*file
,
6635 pgoff_t offset
, pgoff_t last_index
)
6637 pgoff_t req_size
= last_index
- offset
+ 1;
6639 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
6640 return offset
+ req_size
;
6643 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6645 struct btrfs_inode
*ei
;
6646 struct inode
*inode
;
6648 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6653 ei
->space_info
= NULL
;
6657 ei
->last_sub_trans
= 0;
6658 ei
->logged_trans
= 0;
6659 ei
->delalloc_bytes
= 0;
6660 ei
->reserved_bytes
= 0;
6661 ei
->disk_i_size
= 0;
6663 ei
->index_cnt
= (u64
)-1;
6664 ei
->last_unlink_trans
= 0;
6666 atomic_set(&ei
->outstanding_extents
, 0);
6667 atomic_set(&ei
->reserved_extents
, 0);
6669 ei
->ordered_data_close
= 0;
6670 ei
->orphan_meta_reserved
= 0;
6671 ei
->dummy_inode
= 0;
6672 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6674 inode
= &ei
->vfs_inode
;
6675 extent_map_tree_init(&ei
->extent_tree
, GFP_NOFS
);
6676 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
, GFP_NOFS
);
6677 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
, GFP_NOFS
);
6678 mutex_init(&ei
->log_mutex
);
6679 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6680 INIT_LIST_HEAD(&ei
->i_orphan
);
6681 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6682 INIT_LIST_HEAD(&ei
->ordered_operations
);
6683 RB_CLEAR_NODE(&ei
->rb_node
);
6688 static void btrfs_i_callback(struct rcu_head
*head
)
6690 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6691 INIT_LIST_HEAD(&inode
->i_dentry
);
6692 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6695 void btrfs_destroy_inode(struct inode
*inode
)
6697 struct btrfs_ordered_extent
*ordered
;
6698 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6700 WARN_ON(!list_empty(&inode
->i_dentry
));
6701 WARN_ON(inode
->i_data
.nrpages
);
6702 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
));
6703 WARN_ON(atomic_read(&BTRFS_I(inode
)->reserved_extents
));
6706 * This can happen where we create an inode, but somebody else also
6707 * created the same inode and we need to destroy the one we already
6714 * Make sure we're properly removed from the ordered operation
6718 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6719 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6720 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6721 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6724 if (root
== root
->fs_info
->tree_root
) {
6725 struct btrfs_block_group_cache
*block_group
;
6727 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6728 BTRFS_I(inode
)->block_group
);
6729 if (block_group
&& block_group
->inode
== inode
) {
6730 spin_lock(&block_group
->lock
);
6731 block_group
->inode
= NULL
;
6732 spin_unlock(&block_group
->lock
);
6733 btrfs_put_block_group(block_group
);
6734 } else if (block_group
) {
6735 btrfs_put_block_group(block_group
);
6739 spin_lock(&root
->orphan_lock
);
6740 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6741 printk(KERN_INFO
"BTRFS: inode %lu still on the orphan list\n",
6743 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6745 spin_unlock(&root
->orphan_lock
);
6748 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6752 printk(KERN_ERR
"btrfs found ordered "
6753 "extent %llu %llu on inode cleanup\n",
6754 (unsigned long long)ordered
->file_offset
,
6755 (unsigned long long)ordered
->len
);
6756 btrfs_remove_ordered_extent(inode
, ordered
);
6757 btrfs_put_ordered_extent(ordered
);
6758 btrfs_put_ordered_extent(ordered
);
6761 inode_tree_del(inode
);
6762 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6764 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6767 int btrfs_drop_inode(struct inode
*inode
)
6769 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6771 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6772 root
!= root
->fs_info
->tree_root
)
6775 return generic_drop_inode(inode
);
6778 static void init_once(void *foo
)
6780 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6782 inode_init_once(&ei
->vfs_inode
);
6785 void btrfs_destroy_cachep(void)
6787 if (btrfs_inode_cachep
)
6788 kmem_cache_destroy(btrfs_inode_cachep
);
6789 if (btrfs_trans_handle_cachep
)
6790 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6791 if (btrfs_transaction_cachep
)
6792 kmem_cache_destroy(btrfs_transaction_cachep
);
6793 if (btrfs_path_cachep
)
6794 kmem_cache_destroy(btrfs_path_cachep
);
6795 if (btrfs_free_space_cachep
)
6796 kmem_cache_destroy(btrfs_free_space_cachep
);
6799 int btrfs_init_cachep(void)
6801 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6802 sizeof(struct btrfs_inode
), 0,
6803 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6804 if (!btrfs_inode_cachep
)
6807 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6808 sizeof(struct btrfs_trans_handle
), 0,
6809 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6810 if (!btrfs_trans_handle_cachep
)
6813 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6814 sizeof(struct btrfs_transaction
), 0,
6815 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6816 if (!btrfs_transaction_cachep
)
6819 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6820 sizeof(struct btrfs_path
), 0,
6821 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6822 if (!btrfs_path_cachep
)
6825 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6826 sizeof(struct btrfs_free_space
), 0,
6827 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6828 if (!btrfs_free_space_cachep
)
6833 btrfs_destroy_cachep();
6837 static int btrfs_getattr(struct vfsmount
*mnt
,
6838 struct dentry
*dentry
, struct kstat
*stat
)
6840 struct inode
*inode
= dentry
->d_inode
;
6841 generic_fillattr(inode
, stat
);
6842 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
6843 stat
->blksize
= PAGE_CACHE_SIZE
;
6844 stat
->blocks
= (inode_get_bytes(inode
) +
6845 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
6849 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6850 struct inode
*new_dir
, struct dentry
*new_dentry
)
6852 struct btrfs_trans_handle
*trans
;
6853 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6854 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6855 struct inode
*new_inode
= new_dentry
->d_inode
;
6856 struct inode
*old_inode
= old_dentry
->d_inode
;
6857 struct timespec ctime
= CURRENT_TIME
;
6862 if (new_dir
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
6865 /* we only allow rename subvolume link between subvolumes */
6866 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
6869 if (old_inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
6870 (new_inode
&& new_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
))
6873 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
6874 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
6877 * we're using rename to replace one file with another.
6878 * and the replacement file is large. Start IO on it now so
6879 * we don't add too much work to the end of the transaction
6881 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
6882 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
6883 filemap_flush(old_inode
->i_mapping
);
6885 /* close the racy window with snapshot create/destroy ioctl */
6886 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
6887 down_read(&root
->fs_info
->subvol_sem
);
6889 * We want to reserve the absolute worst case amount of items. So if
6890 * both inodes are subvols and we need to unlink them then that would
6891 * require 4 item modifications, but if they are both normal inodes it
6892 * would require 5 item modifications, so we'll assume their normal
6893 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6894 * should cover the worst case number of items we'll modify.
6896 trans
= btrfs_start_transaction(root
, 20);
6898 return PTR_ERR(trans
);
6900 btrfs_set_trans_block_group(trans
, new_dir
);
6903 btrfs_record_root_in_trans(trans
, dest
);
6905 ret
= btrfs_set_inode_index(new_dir
, &index
);
6909 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6910 /* force full log commit if subvolume involved. */
6911 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
6913 ret
= btrfs_insert_inode_ref(trans
, dest
,
6914 new_dentry
->d_name
.name
,
6915 new_dentry
->d_name
.len
,
6917 new_dir
->i_ino
, index
);
6921 * this is an ugly little race, but the rename is required
6922 * to make sure that if we crash, the inode is either at the
6923 * old name or the new one. pinning the log transaction lets
6924 * us make sure we don't allow a log commit to come in after
6925 * we unlink the name but before we add the new name back in.
6927 btrfs_pin_log_trans(root
);
6930 * make sure the inode gets flushed if it is replacing
6933 if (new_inode
&& new_inode
->i_size
&&
6934 old_inode
&& S_ISREG(old_inode
->i_mode
)) {
6935 btrfs_add_ordered_operation(trans
, root
, old_inode
);
6938 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
6939 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
6940 old_inode
->i_ctime
= ctime
;
6942 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
6943 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
6945 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6946 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
6947 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
6948 old_dentry
->d_name
.name
,
6949 old_dentry
->d_name
.len
);
6951 btrfs_inc_nlink(old_dentry
->d_inode
);
6952 ret
= btrfs_unlink_inode(trans
, root
, old_dir
,
6953 old_dentry
->d_inode
,
6954 old_dentry
->d_name
.name
,
6955 old_dentry
->d_name
.len
);
6960 new_inode
->i_ctime
= CURRENT_TIME
;
6961 if (unlikely(new_inode
->i_ino
==
6962 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
6963 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
6964 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
6966 new_dentry
->d_name
.name
,
6967 new_dentry
->d_name
.len
);
6968 BUG_ON(new_inode
->i_nlink
== 0);
6970 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
6971 new_dentry
->d_inode
,
6972 new_dentry
->d_name
.name
,
6973 new_dentry
->d_name
.len
);
6976 if (new_inode
->i_nlink
== 0) {
6977 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
6982 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
6983 new_dentry
->d_name
.name
,
6984 new_dentry
->d_name
.len
, 0, index
);
6987 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
6988 struct dentry
*parent
= dget_parent(new_dentry
);
6989 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
6991 btrfs_end_log_trans(root
);
6994 btrfs_end_transaction_throttle(trans
, root
);
6996 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
6997 up_read(&root
->fs_info
->subvol_sem
);
7003 * some fairly slow code that needs optimization. This walks the list
7004 * of all the inodes with pending delalloc and forces them to disk.
7006 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7008 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7009 struct btrfs_inode
*binode
;
7010 struct inode
*inode
;
7012 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7015 spin_lock(&root
->fs_info
->delalloc_lock
);
7016 while (!list_empty(head
)) {
7017 binode
= list_entry(head
->next
, struct btrfs_inode
,
7019 inode
= igrab(&binode
->vfs_inode
);
7021 list_del_init(&binode
->delalloc_inodes
);
7022 spin_unlock(&root
->fs_info
->delalloc_lock
);
7024 filemap_flush(inode
->i_mapping
);
7026 btrfs_add_delayed_iput(inode
);
7031 spin_lock(&root
->fs_info
->delalloc_lock
);
7033 spin_unlock(&root
->fs_info
->delalloc_lock
);
7035 /* the filemap_flush will queue IO into the worker threads, but
7036 * we have to make sure the IO is actually started and that
7037 * ordered extents get created before we return
7039 atomic_inc(&root
->fs_info
->async_submit_draining
);
7040 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7041 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7042 wait_event(root
->fs_info
->async_submit_wait
,
7043 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7044 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7046 atomic_dec(&root
->fs_info
->async_submit_draining
);
7050 int btrfs_start_one_delalloc_inode(struct btrfs_root
*root
, int delay_iput
,
7053 struct btrfs_inode
*binode
;
7054 struct inode
*inode
= NULL
;
7056 spin_lock(&root
->fs_info
->delalloc_lock
);
7057 while (!list_empty(&root
->fs_info
->delalloc_inodes
)) {
7058 binode
= list_entry(root
->fs_info
->delalloc_inodes
.next
,
7059 struct btrfs_inode
, delalloc_inodes
);
7060 inode
= igrab(&binode
->vfs_inode
);
7062 list_move_tail(&binode
->delalloc_inodes
,
7063 &root
->fs_info
->delalloc_inodes
);
7067 list_del_init(&binode
->delalloc_inodes
);
7068 cond_resched_lock(&root
->fs_info
->delalloc_lock
);
7070 spin_unlock(&root
->fs_info
->delalloc_lock
);
7074 filemap_write_and_wait(inode
->i_mapping
);
7076 * We have to do this because compression doesn't
7077 * actually set PG_writeback until it submits the pages
7078 * for IO, which happens in an async thread, so we could
7079 * race and not actually wait for any writeback pages
7080 * because they've not been submitted yet. Technically
7081 * this could still be the case for the ordered stuff
7082 * since the async thread may not have started to do its
7083 * work yet. If this becomes the case then we need to
7084 * figure out a way to make sure that in writepage we
7085 * wait for any async pages to be submitted before
7086 * returning so that fdatawait does what its supposed to
7089 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
7091 filemap_flush(inode
->i_mapping
);
7094 btrfs_add_delayed_iput(inode
);
7102 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7103 const char *symname
)
7105 struct btrfs_trans_handle
*trans
;
7106 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7107 struct btrfs_path
*path
;
7108 struct btrfs_key key
;
7109 struct inode
*inode
= NULL
;
7117 struct btrfs_file_extent_item
*ei
;
7118 struct extent_buffer
*leaf
;
7119 unsigned long nr
= 0;
7121 name_len
= strlen(symname
) + 1;
7122 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7123 return -ENAMETOOLONG
;
7125 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
7129 * 2 items for inode item and ref
7130 * 2 items for dir items
7131 * 1 item for xattr if selinux is on
7133 trans
= btrfs_start_transaction(root
, 5);
7135 return PTR_ERR(trans
);
7137 btrfs_set_trans_block_group(trans
, dir
);
7139 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7140 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
7141 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
7143 err
= PTR_ERR(inode
);
7147 err
= btrfs_init_inode_security(trans
, inode
, dir
);
7153 btrfs_set_trans_block_group(trans
, inode
);
7154 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7158 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7159 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7160 inode
->i_fop
= &btrfs_file_operations
;
7161 inode
->i_op
= &btrfs_file_inode_operations
;
7162 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7164 btrfs_update_inode_block_group(trans
, inode
);
7165 btrfs_update_inode_block_group(trans
, dir
);
7169 path
= btrfs_alloc_path();
7171 key
.objectid
= inode
->i_ino
;
7173 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7174 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7175 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7181 leaf
= path
->nodes
[0];
7182 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7183 struct btrfs_file_extent_item
);
7184 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7185 btrfs_set_file_extent_type(leaf
, ei
,
7186 BTRFS_FILE_EXTENT_INLINE
);
7187 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7188 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7189 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7190 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7192 ptr
= btrfs_file_extent_inline_start(ei
);
7193 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7194 btrfs_mark_buffer_dirty(leaf
);
7195 btrfs_free_path(path
);
7197 inode
->i_op
= &btrfs_symlink_inode_operations
;
7198 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7199 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7200 inode_set_bytes(inode
, name_len
);
7201 btrfs_i_size_write(inode
, name_len
- 1);
7202 err
= btrfs_update_inode(trans
, root
, inode
);
7207 nr
= trans
->blocks_used
;
7208 btrfs_end_transaction_throttle(trans
, root
);
7210 inode_dec_link_count(inode
);
7213 btrfs_btree_balance_dirty(root
, nr
);
7217 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7218 u64 start
, u64 num_bytes
, u64 min_size
,
7219 loff_t actual_len
, u64
*alloc_hint
,
7220 struct btrfs_trans_handle
*trans
)
7222 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7223 struct btrfs_key ins
;
7224 u64 cur_offset
= start
;
7227 bool own_trans
= true;
7231 while (num_bytes
> 0) {
7233 trans
= btrfs_start_transaction(root
, 3);
7234 if (IS_ERR(trans
)) {
7235 ret
= PTR_ERR(trans
);
7240 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7241 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7244 btrfs_end_transaction(trans
, root
);
7248 ret
= insert_reserved_file_extent(trans
, inode
,
7249 cur_offset
, ins
.objectid
,
7250 ins
.offset
, ins
.offset
,
7251 ins
.offset
, 0, 0, 0,
7252 BTRFS_FILE_EXTENT_PREALLOC
);
7254 btrfs_drop_extent_cache(inode
, cur_offset
,
7255 cur_offset
+ ins
.offset
-1, 0);
7257 num_bytes
-= ins
.offset
;
7258 cur_offset
+= ins
.offset
;
7259 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7261 inode
->i_ctime
= CURRENT_TIME
;
7262 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7263 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7264 (actual_len
> inode
->i_size
) &&
7265 (cur_offset
> inode
->i_size
)) {
7266 if (cur_offset
> actual_len
)
7267 i_size
= actual_len
;
7269 i_size
= cur_offset
;
7270 i_size_write(inode
, i_size
);
7271 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7274 ret
= btrfs_update_inode(trans
, root
, inode
);
7278 btrfs_end_transaction(trans
, root
);
7283 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7284 u64 start
, u64 num_bytes
, u64 min_size
,
7285 loff_t actual_len
, u64
*alloc_hint
)
7287 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7288 min_size
, actual_len
, alloc_hint
,
7292 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7293 struct btrfs_trans_handle
*trans
, int mode
,
7294 u64 start
, u64 num_bytes
, u64 min_size
,
7295 loff_t actual_len
, u64
*alloc_hint
)
7297 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7298 min_size
, actual_len
, alloc_hint
, trans
);
7301 static int btrfs_set_page_dirty(struct page
*page
)
7303 return __set_page_dirty_nobuffers(page
);
7306 static int btrfs_permission(struct inode
*inode
, int mask
, unsigned int flags
)
7308 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7310 if (btrfs_root_readonly(root
) && (mask
& MAY_WRITE
))
7312 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
7314 return generic_permission(inode
, mask
, flags
, btrfs_check_acl
);
7317 static const struct inode_operations btrfs_dir_inode_operations
= {
7318 .getattr
= btrfs_getattr
,
7319 .lookup
= btrfs_lookup
,
7320 .create
= btrfs_create
,
7321 .unlink
= btrfs_unlink
,
7323 .mkdir
= btrfs_mkdir
,
7324 .rmdir
= btrfs_rmdir
,
7325 .rename
= btrfs_rename
,
7326 .symlink
= btrfs_symlink
,
7327 .setattr
= btrfs_setattr
,
7328 .mknod
= btrfs_mknod
,
7329 .setxattr
= btrfs_setxattr
,
7330 .getxattr
= btrfs_getxattr
,
7331 .listxattr
= btrfs_listxattr
,
7332 .removexattr
= btrfs_removexattr
,
7333 .permission
= btrfs_permission
,
7335 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7336 .lookup
= btrfs_lookup
,
7337 .permission
= btrfs_permission
,
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 .sync_page
= block_sync_page
,
7384 .direct_IO
= btrfs_direct_IO
,
7385 .invalidatepage
= btrfs_invalidatepage
,
7386 .releasepage
= btrfs_releasepage
,
7387 .set_page_dirty
= btrfs_set_page_dirty
,
7388 .error_remove_page
= generic_error_remove_page
,
7391 static const struct address_space_operations btrfs_symlink_aops
= {
7392 .readpage
= btrfs_readpage
,
7393 .writepage
= btrfs_writepage
,
7394 .invalidatepage
= btrfs_invalidatepage
,
7395 .releasepage
= btrfs_releasepage
,
7398 static const struct inode_operations btrfs_file_inode_operations
= {
7399 .getattr
= btrfs_getattr
,
7400 .setattr
= btrfs_setattr
,
7401 .setxattr
= btrfs_setxattr
,
7402 .getxattr
= btrfs_getxattr
,
7403 .listxattr
= btrfs_listxattr
,
7404 .removexattr
= btrfs_removexattr
,
7405 .permission
= btrfs_permission
,
7406 .fiemap
= btrfs_fiemap
,
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
,
7417 static const struct inode_operations btrfs_symlink_inode_operations
= {
7418 .readlink
= generic_readlink
,
7419 .follow_link
= page_follow_link_light
,
7420 .put_link
= page_put_link
,
7421 .getattr
= btrfs_getattr
,
7422 .permission
= btrfs_permission
,
7423 .setxattr
= btrfs_setxattr
,
7424 .getxattr
= btrfs_getxattr
,
7425 .listxattr
= btrfs_listxattr
,
7426 .removexattr
= btrfs_removexattr
,
7429 const struct dentry_operations btrfs_dentry_operations
= {
7430 .d_delete
= btrfs_dentry_delete
,