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/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
54 struct btrfs_iget_args
{
56 struct btrfs_root
*root
;
59 static struct inode_operations btrfs_dir_inode_operations
;
60 static struct inode_operations btrfs_symlink_inode_operations
;
61 static struct inode_operations btrfs_dir_ro_inode_operations
;
62 static struct inode_operations btrfs_special_inode_operations
;
63 static struct inode_operations btrfs_file_inode_operations
;
64 static struct address_space_operations btrfs_aops
;
65 static struct address_space_operations btrfs_symlink_aops
;
66 static struct file_operations btrfs_dir_file_operations
;
67 static struct extent_io_ops btrfs_extent_io_ops
;
69 static struct kmem_cache
*btrfs_inode_cachep
;
70 struct kmem_cache
*btrfs_trans_handle_cachep
;
71 struct kmem_cache
*btrfs_transaction_cachep
;
72 struct kmem_cache
*btrfs_path_cachep
;
75 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
76 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
77 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
78 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
79 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
80 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
81 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
82 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
85 static void btrfs_truncate(struct inode
*inode
);
86 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
87 static noinline
int cow_file_range(struct inode
*inode
,
88 struct page
*locked_page
,
89 u64 start
, u64 end
, int *page_started
,
90 unsigned long *nr_written
, int unlock
);
92 static int btrfs_init_inode_security(struct inode
*inode
, struct inode
*dir
)
96 err
= btrfs_init_acl(inode
, dir
);
98 err
= btrfs_xattr_security_init(inode
, dir
);
103 * this does all the hard work for inserting an inline extent into
104 * the btree. The caller should have done a btrfs_drop_extents so that
105 * no overlapping inline items exist in the btree
107 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
108 struct btrfs_root
*root
, struct inode
*inode
,
109 u64 start
, size_t size
, size_t compressed_size
,
110 struct page
**compressed_pages
)
112 struct btrfs_key key
;
113 struct btrfs_path
*path
;
114 struct extent_buffer
*leaf
;
115 struct page
*page
= NULL
;
118 struct btrfs_file_extent_item
*ei
;
121 size_t cur_size
= size
;
123 unsigned long offset
;
124 int use_compress
= 0;
126 if (compressed_size
&& compressed_pages
) {
128 cur_size
= compressed_size
;
131 path
= btrfs_alloc_path();
135 path
->leave_spinning
= 1;
136 btrfs_set_trans_block_group(trans
, inode
);
138 key
.objectid
= inode
->i_ino
;
140 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
141 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
143 inode_add_bytes(inode
, size
);
144 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
151 leaf
= path
->nodes
[0];
152 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
153 struct btrfs_file_extent_item
);
154 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
155 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
156 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
157 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
158 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
159 ptr
= btrfs_file_extent_inline_start(ei
);
164 while (compressed_size
> 0) {
165 cpage
= compressed_pages
[i
];
166 cur_size
= min_t(unsigned long, compressed_size
,
169 kaddr
= kmap_atomic(cpage
, KM_USER0
);
170 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
171 kunmap_atomic(kaddr
, KM_USER0
);
175 compressed_size
-= cur_size
;
177 btrfs_set_file_extent_compression(leaf
, ei
,
178 BTRFS_COMPRESS_ZLIB
);
180 page
= find_get_page(inode
->i_mapping
,
181 start
>> PAGE_CACHE_SHIFT
);
182 btrfs_set_file_extent_compression(leaf
, ei
, 0);
183 kaddr
= kmap_atomic(page
, KM_USER0
);
184 offset
= start
& (PAGE_CACHE_SIZE
- 1);
185 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
186 kunmap_atomic(kaddr
, KM_USER0
);
187 page_cache_release(page
);
189 btrfs_mark_buffer_dirty(leaf
);
190 btrfs_free_path(path
);
192 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
193 btrfs_update_inode(trans
, root
, inode
);
196 btrfs_free_path(path
);
202 * conditionally insert an inline extent into the file. This
203 * does the checks required to make sure the data is small enough
204 * to fit as an inline extent.
206 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
207 struct btrfs_root
*root
,
208 struct inode
*inode
, u64 start
, u64 end
,
209 size_t compressed_size
,
210 struct page
**compressed_pages
)
212 u64 isize
= i_size_read(inode
);
213 u64 actual_end
= min(end
+ 1, isize
);
214 u64 inline_len
= actual_end
- start
;
215 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
216 ~((u64
)root
->sectorsize
- 1);
218 u64 data_len
= inline_len
;
222 data_len
= compressed_size
;
225 actual_end
>= PAGE_CACHE_SIZE
||
226 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
228 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
230 data_len
> root
->fs_info
->max_inline
) {
234 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
235 aligned_end
, aligned_end
, start
, &hint_byte
);
238 if (isize
> actual_end
)
239 inline_len
= min_t(u64
, isize
, actual_end
);
240 ret
= insert_inline_extent(trans
, root
, inode
, start
,
241 inline_len
, compressed_size
,
244 btrfs_drop_extent_cache(inode
, start
, aligned_end
, 0);
248 struct async_extent
{
253 unsigned long nr_pages
;
254 struct list_head list
;
259 struct btrfs_root
*root
;
260 struct page
*locked_page
;
263 struct list_head extents
;
264 struct btrfs_work work
;
267 static noinline
int add_async_extent(struct async_cow
*cow
,
268 u64 start
, u64 ram_size
,
271 unsigned long nr_pages
)
273 struct async_extent
*async_extent
;
275 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
276 async_extent
->start
= start
;
277 async_extent
->ram_size
= ram_size
;
278 async_extent
->compressed_size
= compressed_size
;
279 async_extent
->pages
= pages
;
280 async_extent
->nr_pages
= nr_pages
;
281 list_add_tail(&async_extent
->list
, &cow
->extents
);
286 * we create compressed extents in two phases. The first
287 * phase compresses a range of pages that have already been
288 * locked (both pages and state bits are locked).
290 * This is done inside an ordered work queue, and the compression
291 * is spread across many cpus. The actual IO submission is step
292 * two, and the ordered work queue takes care of making sure that
293 * happens in the same order things were put onto the queue by
294 * writepages and friends.
296 * If this code finds it can't get good compression, it puts an
297 * entry onto the work queue to write the uncompressed bytes. This
298 * makes sure that both compressed inodes and uncompressed inodes
299 * are written in the same order that pdflush sent them down.
301 static noinline
int compress_file_range(struct inode
*inode
,
302 struct page
*locked_page
,
304 struct async_cow
*async_cow
,
307 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
308 struct btrfs_trans_handle
*trans
;
312 u64 blocksize
= root
->sectorsize
;
314 u64 isize
= i_size_read(inode
);
316 struct page
**pages
= NULL
;
317 unsigned long nr_pages
;
318 unsigned long nr_pages_ret
= 0;
319 unsigned long total_compressed
= 0;
320 unsigned long total_in
= 0;
321 unsigned long max_compressed
= 128 * 1024;
322 unsigned long max_uncompressed
= 128 * 1024;
328 actual_end
= min_t(u64
, isize
, end
+ 1);
331 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
332 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
335 * we don't want to send crud past the end of i_size through
336 * compression, that's just a waste of CPU time. So, if the
337 * end of the file is before the start of our current
338 * requested range of bytes, we bail out to the uncompressed
339 * cleanup code that can deal with all of this.
341 * It isn't really the fastest way to fix things, but this is a
342 * very uncommon corner.
344 if (actual_end
<= start
)
345 goto cleanup_and_bail_uncompressed
;
347 total_compressed
= actual_end
- start
;
349 /* we want to make sure that amount of ram required to uncompress
350 * an extent is reasonable, so we limit the total size in ram
351 * of a compressed extent to 128k. This is a crucial number
352 * because it also controls how easily we can spread reads across
353 * cpus for decompression.
355 * We also want to make sure the amount of IO required to do
356 * a random read is reasonably small, so we limit the size of
357 * a compressed extent to 128k.
359 total_compressed
= min(total_compressed
, max_uncompressed
);
360 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
361 num_bytes
= max(blocksize
, num_bytes
);
362 disk_num_bytes
= num_bytes
;
367 * we do compression for mount -o compress and when the
368 * inode has not been flagged as nocompress. This flag can
369 * change at any time if we discover bad compression ratios.
371 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
372 btrfs_test_opt(root
, COMPRESS
)) {
374 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
376 ret
= btrfs_zlib_compress_pages(inode
->i_mapping
, start
,
377 total_compressed
, pages
,
378 nr_pages
, &nr_pages_ret
,
384 unsigned long offset
= total_compressed
&
385 (PAGE_CACHE_SIZE
- 1);
386 struct page
*page
= pages
[nr_pages_ret
- 1];
389 /* zero the tail end of the last page, we might be
390 * sending it down to disk
393 kaddr
= kmap_atomic(page
, KM_USER0
);
394 memset(kaddr
+ offset
, 0,
395 PAGE_CACHE_SIZE
- offset
);
396 kunmap_atomic(kaddr
, KM_USER0
);
402 trans
= btrfs_join_transaction(root
, 1);
404 btrfs_set_trans_block_group(trans
, inode
);
406 /* lets try to make an inline extent */
407 if (ret
|| total_in
< (actual_end
- start
)) {
408 /* we didn't compress the entire range, try
409 * to make an uncompressed inline extent.
411 ret
= cow_file_range_inline(trans
, root
, inode
,
412 start
, end
, 0, NULL
);
414 /* try making a compressed inline extent */
415 ret
= cow_file_range_inline(trans
, root
, inode
,
417 total_compressed
, pages
);
419 btrfs_end_transaction(trans
, root
);
422 * inline extent creation worked, we don't need
423 * to create any more async work items. Unlock
424 * and free up our temp pages.
426 extent_clear_unlock_delalloc(inode
,
427 &BTRFS_I(inode
)->io_tree
,
428 start
, end
, NULL
, 1, 0,
437 * we aren't doing an inline extent round the compressed size
438 * up to a block size boundary so the allocator does sane
441 total_compressed
= (total_compressed
+ blocksize
- 1) &
445 * one last check to make sure the compression is really a
446 * win, compare the page count read with the blocks on disk
448 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
449 ~(PAGE_CACHE_SIZE
- 1);
450 if (total_compressed
>= total_in
) {
453 disk_num_bytes
= total_compressed
;
454 num_bytes
= total_in
;
457 if (!will_compress
&& pages
) {
459 * the compression code ran but failed to make things smaller,
460 * free any pages it allocated and our page pointer array
462 for (i
= 0; i
< nr_pages_ret
; i
++) {
463 WARN_ON(pages
[i
]->mapping
);
464 page_cache_release(pages
[i
]);
468 total_compressed
= 0;
471 /* flag the file so we don't compress in the future */
472 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
477 /* the async work queues will take care of doing actual
478 * allocation on disk for these compressed pages,
479 * and will submit them to the elevator.
481 add_async_extent(async_cow
, start
, num_bytes
,
482 total_compressed
, pages
, nr_pages_ret
);
484 if (start
+ num_bytes
< end
&& start
+ num_bytes
< actual_end
) {
491 cleanup_and_bail_uncompressed
:
493 * No compression, but we still need to write the pages in
494 * the file we've been given so far. redirty the locked
495 * page if it corresponds to our extent and set things up
496 * for the async work queue to run cow_file_range to do
497 * the normal delalloc dance
499 if (page_offset(locked_page
) >= start
&&
500 page_offset(locked_page
) <= end
) {
501 __set_page_dirty_nobuffers(locked_page
);
502 /* unlocked later on in the async handlers */
504 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0);
512 for (i
= 0; i
< nr_pages_ret
; i
++) {
513 WARN_ON(pages
[i
]->mapping
);
514 page_cache_release(pages
[i
]);
522 * phase two of compressed writeback. This is the ordered portion
523 * of the code, which only gets called in the order the work was
524 * queued. We walk all the async extents created by compress_file_range
525 * and send them down to the disk.
527 static noinline
int submit_compressed_extents(struct inode
*inode
,
528 struct async_cow
*async_cow
)
530 struct async_extent
*async_extent
;
532 struct btrfs_trans_handle
*trans
;
533 struct btrfs_key ins
;
534 struct extent_map
*em
;
535 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
536 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
537 struct extent_io_tree
*io_tree
;
540 if (list_empty(&async_cow
->extents
))
543 trans
= btrfs_join_transaction(root
, 1);
545 while (!list_empty(&async_cow
->extents
)) {
546 async_extent
= list_entry(async_cow
->extents
.next
,
547 struct async_extent
, list
);
548 list_del(&async_extent
->list
);
550 io_tree
= &BTRFS_I(inode
)->io_tree
;
552 /* did the compression code fall back to uncompressed IO? */
553 if (!async_extent
->pages
) {
554 int page_started
= 0;
555 unsigned long nr_written
= 0;
557 lock_extent(io_tree
, async_extent
->start
,
558 async_extent
->start
+
559 async_extent
->ram_size
- 1, GFP_NOFS
);
561 /* allocate blocks */
562 cow_file_range(inode
, async_cow
->locked_page
,
564 async_extent
->start
+
565 async_extent
->ram_size
- 1,
566 &page_started
, &nr_written
, 0);
569 * if page_started, cow_file_range inserted an
570 * inline extent and took care of all the unlocking
571 * and IO for us. Otherwise, we need to submit
572 * all those pages down to the drive.
575 extent_write_locked_range(io_tree
,
576 inode
, async_extent
->start
,
577 async_extent
->start
+
578 async_extent
->ram_size
- 1,
586 lock_extent(io_tree
, async_extent
->start
,
587 async_extent
->start
+ async_extent
->ram_size
- 1,
590 * here we're doing allocation and writeback of the
593 btrfs_drop_extent_cache(inode
, async_extent
->start
,
594 async_extent
->start
+
595 async_extent
->ram_size
- 1, 0);
597 ret
= btrfs_reserve_extent(trans
, root
,
598 async_extent
->compressed_size
,
599 async_extent
->compressed_size
,
603 em
= alloc_extent_map(GFP_NOFS
);
604 em
->start
= async_extent
->start
;
605 em
->len
= async_extent
->ram_size
;
606 em
->orig_start
= em
->start
;
608 em
->block_start
= ins
.objectid
;
609 em
->block_len
= ins
.offset
;
610 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
611 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
612 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
615 spin_lock(&em_tree
->lock
);
616 ret
= add_extent_mapping(em_tree
, em
);
617 spin_unlock(&em_tree
->lock
);
618 if (ret
!= -EEXIST
) {
622 btrfs_drop_extent_cache(inode
, async_extent
->start
,
623 async_extent
->start
+
624 async_extent
->ram_size
- 1, 0);
627 ret
= btrfs_add_ordered_extent(inode
, async_extent
->start
,
629 async_extent
->ram_size
,
631 BTRFS_ORDERED_COMPRESSED
);
634 btrfs_end_transaction(trans
, root
);
637 * clear dirty, set writeback and unlock the pages.
639 extent_clear_unlock_delalloc(inode
,
640 &BTRFS_I(inode
)->io_tree
,
642 async_extent
->start
+
643 async_extent
->ram_size
- 1,
644 NULL
, 1, 1, 0, 1, 1, 0);
646 ret
= btrfs_submit_compressed_write(inode
,
648 async_extent
->ram_size
,
650 ins
.offset
, async_extent
->pages
,
651 async_extent
->nr_pages
);
654 trans
= btrfs_join_transaction(root
, 1);
655 alloc_hint
= ins
.objectid
+ ins
.offset
;
660 btrfs_end_transaction(trans
, root
);
665 * when extent_io.c finds a delayed allocation range in the file,
666 * the call backs end up in this code. The basic idea is to
667 * allocate extents on disk for the range, and create ordered data structs
668 * in ram to track those extents.
670 * locked_page is the page that writepage had locked already. We use
671 * it to make sure we don't do extra locks or unlocks.
673 * *page_started is set to one if we unlock locked_page and do everything
674 * required to start IO on it. It may be clean and already done with
677 static noinline
int cow_file_range(struct inode
*inode
,
678 struct page
*locked_page
,
679 u64 start
, u64 end
, int *page_started
,
680 unsigned long *nr_written
,
683 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
684 struct btrfs_trans_handle
*trans
;
687 unsigned long ram_size
;
690 u64 blocksize
= root
->sectorsize
;
692 u64 isize
= i_size_read(inode
);
693 struct btrfs_key ins
;
694 struct extent_map
*em
;
695 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
698 trans
= btrfs_join_transaction(root
, 1);
700 btrfs_set_trans_block_group(trans
, inode
);
702 actual_end
= min_t(u64
, isize
, end
+ 1);
704 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
705 num_bytes
= max(blocksize
, num_bytes
);
706 disk_num_bytes
= num_bytes
;
710 /* lets try to make an inline extent */
711 ret
= cow_file_range_inline(trans
, root
, inode
,
712 start
, end
, 0, NULL
);
714 extent_clear_unlock_delalloc(inode
,
715 &BTRFS_I(inode
)->io_tree
,
716 start
, end
, NULL
, 1, 1,
718 *nr_written
= *nr_written
+
719 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
726 BUG_ON(disk_num_bytes
>
727 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
729 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
731 while (disk_num_bytes
> 0) {
732 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
733 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
734 root
->sectorsize
, 0, alloc_hint
,
738 em
= alloc_extent_map(GFP_NOFS
);
740 em
->orig_start
= em
->start
;
742 ram_size
= ins
.offset
;
743 em
->len
= ins
.offset
;
745 em
->block_start
= ins
.objectid
;
746 em
->block_len
= ins
.offset
;
747 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
748 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
751 spin_lock(&em_tree
->lock
);
752 ret
= add_extent_mapping(em_tree
, em
);
753 spin_unlock(&em_tree
->lock
);
754 if (ret
!= -EEXIST
) {
758 btrfs_drop_extent_cache(inode
, start
,
759 start
+ ram_size
- 1, 0);
762 cur_alloc_size
= ins
.offset
;
763 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
764 ram_size
, cur_alloc_size
, 0);
767 if (root
->root_key
.objectid
==
768 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
769 ret
= btrfs_reloc_clone_csums(inode
, start
,
774 if (disk_num_bytes
< cur_alloc_size
)
777 /* we're not doing compressed IO, don't unlock the first
778 * page (which the caller expects to stay locked), don't
779 * clear any dirty bits and don't set any writeback bits
781 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
782 start
, start
+ ram_size
- 1,
783 locked_page
, unlock
, 1,
785 disk_num_bytes
-= cur_alloc_size
;
786 num_bytes
-= cur_alloc_size
;
787 alloc_hint
= ins
.objectid
+ ins
.offset
;
788 start
+= cur_alloc_size
;
792 btrfs_end_transaction(trans
, root
);
798 * work queue call back to started compression on a file and pages
800 static noinline
void async_cow_start(struct btrfs_work
*work
)
802 struct async_cow
*async_cow
;
804 async_cow
= container_of(work
, struct async_cow
, work
);
806 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
807 async_cow
->start
, async_cow
->end
, async_cow
,
810 async_cow
->inode
= NULL
;
814 * work queue call back to submit previously compressed pages
816 static noinline
void async_cow_submit(struct btrfs_work
*work
)
818 struct async_cow
*async_cow
;
819 struct btrfs_root
*root
;
820 unsigned long nr_pages
;
822 async_cow
= container_of(work
, struct async_cow
, work
);
824 root
= async_cow
->root
;
825 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
828 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
830 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
832 waitqueue_active(&root
->fs_info
->async_submit_wait
))
833 wake_up(&root
->fs_info
->async_submit_wait
);
835 if (async_cow
->inode
)
836 submit_compressed_extents(async_cow
->inode
, async_cow
);
839 static noinline
void async_cow_free(struct btrfs_work
*work
)
841 struct async_cow
*async_cow
;
842 async_cow
= container_of(work
, struct async_cow
, work
);
846 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
847 u64 start
, u64 end
, int *page_started
,
848 unsigned long *nr_written
)
850 struct async_cow
*async_cow
;
851 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
852 unsigned long nr_pages
;
854 int limit
= 10 * 1024 * 1042;
856 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
|
857 EXTENT_DELALLOC
, 1, 0, GFP_NOFS
);
858 while (start
< end
) {
859 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
860 async_cow
->inode
= inode
;
861 async_cow
->root
= root
;
862 async_cow
->locked_page
= locked_page
;
863 async_cow
->start
= start
;
865 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
868 cur_end
= min(end
, start
+ 512 * 1024 - 1);
870 async_cow
->end
= cur_end
;
871 INIT_LIST_HEAD(&async_cow
->extents
);
873 async_cow
->work
.func
= async_cow_start
;
874 async_cow
->work
.ordered_func
= async_cow_submit
;
875 async_cow
->work
.ordered_free
= async_cow_free
;
876 async_cow
->work
.flags
= 0;
878 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
880 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
882 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
885 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
886 wait_event(root
->fs_info
->async_submit_wait
,
887 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
891 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
892 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
893 wait_event(root
->fs_info
->async_submit_wait
,
894 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
898 *nr_written
+= nr_pages
;
905 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
906 u64 bytenr
, u64 num_bytes
)
909 struct btrfs_ordered_sum
*sums
;
912 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
913 bytenr
+ num_bytes
- 1, &list
);
914 if (ret
== 0 && list_empty(&list
))
917 while (!list_empty(&list
)) {
918 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
919 list_del(&sums
->list
);
926 * when nowcow writeback call back. This checks for snapshots or COW copies
927 * of the extents that exist in the file, and COWs the file as required.
929 * If no cow copies or snapshots exist, we write directly to the existing
932 static noinline
int run_delalloc_nocow(struct inode
*inode
,
933 struct page
*locked_page
,
934 u64 start
, u64 end
, int *page_started
, int force
,
935 unsigned long *nr_written
)
937 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
938 struct btrfs_trans_handle
*trans
;
939 struct extent_buffer
*leaf
;
940 struct btrfs_path
*path
;
941 struct btrfs_file_extent_item
*fi
;
942 struct btrfs_key found_key
;
955 path
= btrfs_alloc_path();
957 trans
= btrfs_join_transaction(root
, 1);
963 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
966 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
967 leaf
= path
->nodes
[0];
968 btrfs_item_key_to_cpu(leaf
, &found_key
,
970 if (found_key
.objectid
== inode
->i_ino
&&
971 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
976 leaf
= path
->nodes
[0];
977 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
978 ret
= btrfs_next_leaf(root
, path
);
983 leaf
= path
->nodes
[0];
989 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
991 if (found_key
.objectid
> inode
->i_ino
||
992 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
993 found_key
.offset
> end
)
996 if (found_key
.offset
> cur_offset
) {
997 extent_end
= found_key
.offset
;
1001 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1002 struct btrfs_file_extent_item
);
1003 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1005 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1006 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1007 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1008 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1009 extent_end
= found_key
.offset
+
1010 btrfs_file_extent_num_bytes(leaf
, fi
);
1011 if (extent_end
<= start
) {
1015 if (disk_bytenr
== 0)
1017 if (btrfs_file_extent_compression(leaf
, fi
) ||
1018 btrfs_file_extent_encryption(leaf
, fi
) ||
1019 btrfs_file_extent_other_encoding(leaf
, fi
))
1021 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1023 if (btrfs_extent_readonly(root
, disk_bytenr
))
1025 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1027 extent_offset
, disk_bytenr
))
1029 disk_bytenr
+= extent_offset
;
1030 disk_bytenr
+= cur_offset
- found_key
.offset
;
1031 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1033 * force cow if csum exists in the range.
1034 * this ensure that csum for a given extent are
1035 * either valid or do not exist.
1037 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1040 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1041 extent_end
= found_key
.offset
+
1042 btrfs_file_extent_inline_len(leaf
, fi
);
1043 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1048 if (extent_end
<= start
) {
1053 if (cow_start
== (u64
)-1)
1054 cow_start
= cur_offset
;
1055 cur_offset
= extent_end
;
1056 if (cur_offset
> end
)
1062 btrfs_release_path(root
, path
);
1063 if (cow_start
!= (u64
)-1) {
1064 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1065 found_key
.offset
- 1, page_started
,
1068 cow_start
= (u64
)-1;
1071 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1072 struct extent_map
*em
;
1073 struct extent_map_tree
*em_tree
;
1074 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1075 em
= alloc_extent_map(GFP_NOFS
);
1076 em
->start
= cur_offset
;
1077 em
->orig_start
= em
->start
;
1078 em
->len
= num_bytes
;
1079 em
->block_len
= num_bytes
;
1080 em
->block_start
= disk_bytenr
;
1081 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1082 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1084 spin_lock(&em_tree
->lock
);
1085 ret
= add_extent_mapping(em_tree
, em
);
1086 spin_unlock(&em_tree
->lock
);
1087 if (ret
!= -EEXIST
) {
1088 free_extent_map(em
);
1091 btrfs_drop_extent_cache(inode
, em
->start
,
1092 em
->start
+ em
->len
- 1, 0);
1094 type
= BTRFS_ORDERED_PREALLOC
;
1096 type
= BTRFS_ORDERED_NOCOW
;
1099 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1100 num_bytes
, num_bytes
, type
);
1103 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1104 cur_offset
, cur_offset
+ num_bytes
- 1,
1105 locked_page
, 1, 1, 1, 0, 0, 0);
1106 cur_offset
= extent_end
;
1107 if (cur_offset
> end
)
1110 btrfs_release_path(root
, path
);
1112 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1113 cow_start
= cur_offset
;
1114 if (cow_start
!= (u64
)-1) {
1115 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1116 page_started
, nr_written
, 1);
1120 ret
= btrfs_end_transaction(trans
, root
);
1122 btrfs_free_path(path
);
1127 * extent_io.c call back to do delayed allocation processing
1129 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1130 u64 start
, u64 end
, int *page_started
,
1131 unsigned long *nr_written
)
1134 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1136 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1137 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1138 page_started
, 1, nr_written
);
1139 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1140 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1141 page_started
, 0, nr_written
);
1142 else if (!btrfs_test_opt(root
, COMPRESS
))
1143 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1144 page_started
, nr_written
, 1);
1146 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1147 page_started
, nr_written
);
1152 * extent_io.c set_bit_hook, used to track delayed allocation
1153 * bytes in this file, and to maintain the list of inodes that
1154 * have pending delalloc work to be done.
1156 static int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1157 unsigned long old
, unsigned long bits
)
1160 * set_bit and clear bit hooks normally require _irqsave/restore
1161 * but in this case, we are only testeing for the DELALLOC
1162 * bit, which is only set or cleared with irqs on
1164 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1165 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1166 btrfs_delalloc_reserve_space(root
, inode
, end
- start
+ 1);
1167 spin_lock(&root
->fs_info
->delalloc_lock
);
1168 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
1169 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
1170 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1171 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1172 &root
->fs_info
->delalloc_inodes
);
1174 spin_unlock(&root
->fs_info
->delalloc_lock
);
1180 * extent_io.c clear_bit_hook, see set_bit_hook for why
1182 static int btrfs_clear_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1183 unsigned long old
, unsigned long bits
)
1186 * set_bit and clear bit hooks normally require _irqsave/restore
1187 * but in this case, we are only testeing for the DELALLOC
1188 * bit, which is only set or cleared with irqs on
1190 if ((old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1191 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1193 spin_lock(&root
->fs_info
->delalloc_lock
);
1194 if (end
- start
+ 1 > root
->fs_info
->delalloc_bytes
) {
1195 printk(KERN_INFO
"btrfs warning: delalloc account "
1197 (unsigned long long)end
- start
+ 1,
1198 (unsigned long long)
1199 root
->fs_info
->delalloc_bytes
);
1200 btrfs_delalloc_free_space(root
, inode
, (u64
)-1);
1201 root
->fs_info
->delalloc_bytes
= 0;
1202 BTRFS_I(inode
)->delalloc_bytes
= 0;
1204 btrfs_delalloc_free_space(root
, inode
,
1206 root
->fs_info
->delalloc_bytes
-= end
- start
+ 1;
1207 BTRFS_I(inode
)->delalloc_bytes
-= end
- start
+ 1;
1209 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1210 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1211 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1213 spin_unlock(&root
->fs_info
->delalloc_lock
);
1219 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1220 * we don't create bios that span stripes or chunks
1222 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1223 size_t size
, struct bio
*bio
,
1224 unsigned long bio_flags
)
1226 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1227 struct btrfs_mapping_tree
*map_tree
;
1228 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1233 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1236 length
= bio
->bi_size
;
1237 map_tree
= &root
->fs_info
->mapping_tree
;
1238 map_length
= length
;
1239 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1240 &map_length
, NULL
, 0);
1242 if (map_length
< length
+ size
)
1248 * in order to insert checksums into the metadata in large chunks,
1249 * we wait until bio submission time. All the pages in the bio are
1250 * checksummed and sums are attached onto the ordered extent record.
1252 * At IO completion time the cums attached on the ordered extent record
1253 * are inserted into the btree
1255 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1256 struct bio
*bio
, int mirror_num
,
1257 unsigned long bio_flags
)
1259 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1262 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1268 * in order to insert checksums into the metadata in large chunks,
1269 * we wait until bio submission time. All the pages in the bio are
1270 * checksummed and sums are attached onto the ordered extent record.
1272 * At IO completion time the cums attached on the ordered extent record
1273 * are inserted into the btree
1275 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1276 int mirror_num
, unsigned long bio_flags
)
1278 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1279 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1283 * extent_io.c submission hook. This does the right thing for csum calculation
1284 * on write, or reading the csums from the tree before a read
1286 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1287 int mirror_num
, unsigned long bio_flags
)
1289 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1293 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1295 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1298 if (!(rw
& (1 << BIO_RW
))) {
1299 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1300 return btrfs_submit_compressed_read(inode
, bio
,
1301 mirror_num
, bio_flags
);
1302 } else if (!skip_sum
)
1303 btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1305 } else if (!skip_sum
) {
1306 /* csum items have already been cloned */
1307 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1309 /* we're doing a write, do the async checksumming */
1310 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1311 inode
, rw
, bio
, mirror_num
,
1312 bio_flags
, __btrfs_submit_bio_start
,
1313 __btrfs_submit_bio_done
);
1317 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1321 * given a list of ordered sums record them in the inode. This happens
1322 * at IO completion time based on sums calculated at bio submission time.
1324 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1325 struct inode
*inode
, u64 file_offset
,
1326 struct list_head
*list
)
1328 struct btrfs_ordered_sum
*sum
;
1330 btrfs_set_trans_block_group(trans
, inode
);
1332 list_for_each_entry(sum
, list
, list
) {
1333 btrfs_csum_file_blocks(trans
,
1334 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1339 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
1341 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1343 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1347 /* see btrfs_writepage_start_hook for details on why this is required */
1348 struct btrfs_writepage_fixup
{
1350 struct btrfs_work work
;
1353 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1355 struct btrfs_writepage_fixup
*fixup
;
1356 struct btrfs_ordered_extent
*ordered
;
1358 struct inode
*inode
;
1362 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1366 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1367 ClearPageChecked(page
);
1371 inode
= page
->mapping
->host
;
1372 page_start
= page_offset(page
);
1373 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1375 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1377 /* already ordered? We're done */
1378 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1379 EXTENT_ORDERED
, 0)) {
1383 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1385 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
1386 page_end
, GFP_NOFS
);
1388 btrfs_start_ordered_extent(inode
, ordered
, 1);
1392 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
1393 ClearPageChecked(page
);
1395 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1398 page_cache_release(page
);
1402 * There are a few paths in the higher layers of the kernel that directly
1403 * set the page dirty bit without asking the filesystem if it is a
1404 * good idea. This causes problems because we want to make sure COW
1405 * properly happens and the data=ordered rules are followed.
1407 * In our case any range that doesn't have the ORDERED bit set
1408 * hasn't been properly setup for IO. We kick off an async process
1409 * to fix it up. The async helper will wait for ordered extents, set
1410 * the delalloc bit and make it safe to write the page.
1412 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1414 struct inode
*inode
= page
->mapping
->host
;
1415 struct btrfs_writepage_fixup
*fixup
;
1416 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1419 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1424 if (PageChecked(page
))
1427 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1431 SetPageChecked(page
);
1432 page_cache_get(page
);
1433 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1435 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1439 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1440 struct inode
*inode
, u64 file_pos
,
1441 u64 disk_bytenr
, u64 disk_num_bytes
,
1442 u64 num_bytes
, u64 ram_bytes
,
1444 u8 compression
, u8 encryption
,
1445 u16 other_encoding
, int extent_type
)
1447 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1448 struct btrfs_file_extent_item
*fi
;
1449 struct btrfs_path
*path
;
1450 struct extent_buffer
*leaf
;
1451 struct btrfs_key ins
;
1455 path
= btrfs_alloc_path();
1458 path
->leave_spinning
= 1;
1459 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1460 file_pos
+ num_bytes
, locked_end
,
1464 ins
.objectid
= inode
->i_ino
;
1465 ins
.offset
= file_pos
;
1466 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1467 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1469 leaf
= path
->nodes
[0];
1470 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1471 struct btrfs_file_extent_item
);
1472 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1473 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1474 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1475 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1476 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1477 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1478 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1479 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1480 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1481 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1483 btrfs_unlock_up_safe(path
, 1);
1484 btrfs_set_lock_blocking(leaf
);
1486 btrfs_mark_buffer_dirty(leaf
);
1488 inode_add_bytes(inode
, num_bytes
);
1489 btrfs_drop_extent_cache(inode
, file_pos
, file_pos
+ num_bytes
- 1, 0);
1491 ins
.objectid
= disk_bytenr
;
1492 ins
.offset
= disk_num_bytes
;
1493 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1494 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1495 root
->root_key
.objectid
,
1496 inode
->i_ino
, file_pos
, &ins
);
1498 btrfs_free_path(path
);
1504 * helper function for btrfs_finish_ordered_io, this
1505 * just reads in some of the csum leaves to prime them into ram
1506 * before we start the transaction. It limits the amount of btree
1507 * reads required while inside the transaction.
1509 static noinline
void reada_csum(struct btrfs_root
*root
,
1510 struct btrfs_path
*path
,
1511 struct btrfs_ordered_extent
*ordered_extent
)
1513 struct btrfs_ordered_sum
*sum
;
1516 sum
= list_entry(ordered_extent
->list
.next
, struct btrfs_ordered_sum
,
1518 bytenr
= sum
->sums
[0].bytenr
;
1521 * we don't care about the results, the point of this search is
1522 * just to get the btree leaves into ram
1524 btrfs_lookup_csum(NULL
, root
->fs_info
->csum_root
, path
, bytenr
, 0);
1527 /* as ordered data IO finishes, this gets called so we can finish
1528 * an ordered extent if the range of bytes in the file it covers are
1531 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1533 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1534 struct btrfs_trans_handle
*trans
;
1535 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1536 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1537 struct btrfs_path
*path
;
1541 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1546 * before we join the transaction, try to do some of our IO.
1547 * This will limit the amount of IO that we have to do with
1548 * the transaction running. We're unlikely to need to do any
1549 * IO if the file extents are new, the disk_i_size checks
1550 * covers the most common case.
1552 if (start
< BTRFS_I(inode
)->disk_i_size
) {
1553 path
= btrfs_alloc_path();
1555 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
1558 ordered_extent
= btrfs_lookup_ordered_extent(inode
,
1560 if (!list_empty(&ordered_extent
->list
)) {
1561 btrfs_release_path(root
, path
);
1562 reada_csum(root
, path
, ordered_extent
);
1564 btrfs_free_path(path
);
1568 trans
= btrfs_join_transaction(root
, 1);
1570 if (!ordered_extent
)
1571 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1572 BUG_ON(!ordered_extent
);
1573 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1576 lock_extent(io_tree
, ordered_extent
->file_offset
,
1577 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1580 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1582 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1584 ret
= btrfs_mark_extent_written(trans
, root
, inode
,
1585 ordered_extent
->file_offset
,
1586 ordered_extent
->file_offset
+
1587 ordered_extent
->len
);
1590 ret
= insert_reserved_file_extent(trans
, inode
,
1591 ordered_extent
->file_offset
,
1592 ordered_extent
->start
,
1593 ordered_extent
->disk_len
,
1594 ordered_extent
->len
,
1595 ordered_extent
->len
,
1596 ordered_extent
->file_offset
+
1597 ordered_extent
->len
,
1599 BTRFS_FILE_EXTENT_REG
);
1602 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1603 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1606 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1607 &ordered_extent
->list
);
1609 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1610 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1611 btrfs_update_inode(trans
, root
, inode
);
1612 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1613 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1616 btrfs_put_ordered_extent(ordered_extent
);
1617 /* once for the tree */
1618 btrfs_put_ordered_extent(ordered_extent
);
1620 btrfs_end_transaction(trans
, root
);
1624 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1625 struct extent_state
*state
, int uptodate
)
1627 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1631 * When IO fails, either with EIO or csum verification fails, we
1632 * try other mirrors that might have a good copy of the data. This
1633 * io_failure_record is used to record state as we go through all the
1634 * mirrors. If another mirror has good data, the page is set up to date
1635 * and things continue. If a good mirror can't be found, the original
1636 * bio end_io callback is called to indicate things have failed.
1638 struct io_failure_record
{
1643 unsigned long bio_flags
;
1647 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1648 struct page
*page
, u64 start
, u64 end
,
1649 struct extent_state
*state
)
1651 struct io_failure_record
*failrec
= NULL
;
1653 struct extent_map
*em
;
1654 struct inode
*inode
= page
->mapping
->host
;
1655 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1656 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1663 ret
= get_state_private(failure_tree
, start
, &private);
1665 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1668 failrec
->start
= start
;
1669 failrec
->len
= end
- start
+ 1;
1670 failrec
->last_mirror
= 0;
1671 failrec
->bio_flags
= 0;
1673 spin_lock(&em_tree
->lock
);
1674 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1675 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1676 free_extent_map(em
);
1679 spin_unlock(&em_tree
->lock
);
1681 if (!em
|| IS_ERR(em
)) {
1685 logical
= start
- em
->start
;
1686 logical
= em
->block_start
+ logical
;
1687 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1688 logical
= em
->block_start
;
1689 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1691 failrec
->logical
= logical
;
1692 free_extent_map(em
);
1693 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1694 EXTENT_DIRTY
, GFP_NOFS
);
1695 set_state_private(failure_tree
, start
,
1696 (u64
)(unsigned long)failrec
);
1698 failrec
= (struct io_failure_record
*)(unsigned long)private;
1700 num_copies
= btrfs_num_copies(
1701 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1702 failrec
->logical
, failrec
->len
);
1703 failrec
->last_mirror
++;
1705 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1706 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1709 if (state
&& state
->start
!= failrec
->start
)
1711 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1713 if (!state
|| failrec
->last_mirror
> num_copies
) {
1714 set_state_private(failure_tree
, failrec
->start
, 0);
1715 clear_extent_bits(failure_tree
, failrec
->start
,
1716 failrec
->start
+ failrec
->len
- 1,
1717 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1721 bio
= bio_alloc(GFP_NOFS
, 1);
1722 bio
->bi_private
= state
;
1723 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1724 bio
->bi_sector
= failrec
->logical
>> 9;
1725 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1728 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1729 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1734 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1735 failrec
->last_mirror
,
1736 failrec
->bio_flags
);
1741 * each time an IO finishes, we do a fast check in the IO failure tree
1742 * to see if we need to process or clean up an io_failure_record
1744 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1747 u64 private_failure
;
1748 struct io_failure_record
*failure
;
1752 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1753 (u64
)-1, 1, EXTENT_DIRTY
)) {
1754 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1755 start
, &private_failure
);
1757 failure
= (struct io_failure_record
*)(unsigned long)
1759 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1761 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1763 failure
->start
+ failure
->len
- 1,
1764 EXTENT_DIRTY
| EXTENT_LOCKED
,
1773 * when reads are done, we need to check csums to verify the data is correct
1774 * if there's a match, we allow the bio to finish. If not, we go through
1775 * the io_failure_record routines to find good copies
1777 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1778 struct extent_state
*state
)
1780 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1781 struct inode
*inode
= page
->mapping
->host
;
1782 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1784 u64
private = ~(u32
)0;
1786 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1789 if (PageChecked(page
)) {
1790 ClearPageChecked(page
);
1794 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1797 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1798 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1)) {
1799 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1804 if (state
&& state
->start
== start
) {
1805 private = state
->private;
1808 ret
= get_state_private(io_tree
, start
, &private);
1810 kaddr
= kmap_atomic(page
, KM_USER0
);
1814 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1815 btrfs_csum_final(csum
, (char *)&csum
);
1816 if (csum
!= private)
1819 kunmap_atomic(kaddr
, KM_USER0
);
1821 /* if the io failure tree for this inode is non-empty,
1822 * check to see if we've recovered from a failed IO
1824 btrfs_clean_io_failures(inode
, start
);
1828 if (printk_ratelimit()) {
1829 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
1830 "private %llu\n", page
->mapping
->host
->i_ino
,
1831 (unsigned long long)start
, csum
,
1832 (unsigned long long)private);
1834 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1835 flush_dcache_page(page
);
1836 kunmap_atomic(kaddr
, KM_USER0
);
1843 * This creates an orphan entry for the given inode in case something goes
1844 * wrong in the middle of an unlink/truncate.
1846 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1848 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1851 spin_lock(&root
->list_lock
);
1853 /* already on the orphan list, we're good */
1854 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1855 spin_unlock(&root
->list_lock
);
1859 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1861 spin_unlock(&root
->list_lock
);
1864 * insert an orphan item to track this unlinked/truncated file
1866 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1872 * We have done the truncate/delete so we can go ahead and remove the orphan
1873 * item for this particular inode.
1875 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1877 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1880 spin_lock(&root
->list_lock
);
1882 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1883 spin_unlock(&root
->list_lock
);
1887 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1889 spin_unlock(&root
->list_lock
);
1893 spin_unlock(&root
->list_lock
);
1895 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1901 * this cleans up any orphans that may be left on the list from the last use
1904 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1906 struct btrfs_path
*path
;
1907 struct extent_buffer
*leaf
;
1908 struct btrfs_item
*item
;
1909 struct btrfs_key key
, found_key
;
1910 struct btrfs_trans_handle
*trans
;
1911 struct inode
*inode
;
1912 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1914 path
= btrfs_alloc_path();
1919 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1920 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1921 key
.offset
= (u64
)-1;
1925 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1927 printk(KERN_ERR
"Error searching slot for orphan: %d"
1933 * if ret == 0 means we found what we were searching for, which
1934 * is weird, but possible, so only screw with path if we didnt
1935 * find the key and see if we have stuff that matches
1938 if (path
->slots
[0] == 0)
1943 /* pull out the item */
1944 leaf
= path
->nodes
[0];
1945 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1946 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1948 /* make sure the item matches what we want */
1949 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1951 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1954 /* release the path since we're done with it */
1955 btrfs_release_path(root
, path
);
1958 * this is where we are basically btrfs_lookup, without the
1959 * crossing root thing. we store the inode number in the
1960 * offset of the orphan item.
1962 found_key
.objectid
= found_key
.offset
;
1963 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
1964 found_key
.offset
= 0;
1965 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
);
1970 * add this inode to the orphan list so btrfs_orphan_del does
1971 * the proper thing when we hit it
1973 spin_lock(&root
->list_lock
);
1974 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1975 spin_unlock(&root
->list_lock
);
1978 * if this is a bad inode, means we actually succeeded in
1979 * removing the inode, but not the orphan record, which means
1980 * we need to manually delete the orphan since iput will just
1981 * do a destroy_inode
1983 if (is_bad_inode(inode
)) {
1984 trans
= btrfs_start_transaction(root
, 1);
1985 btrfs_orphan_del(trans
, inode
);
1986 btrfs_end_transaction(trans
, root
);
1991 /* if we have links, this was a truncate, lets do that */
1992 if (inode
->i_nlink
) {
1994 btrfs_truncate(inode
);
1999 /* this will do delete_inode and everything for us */
2004 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2006 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2008 btrfs_free_path(path
);
2012 * very simple check to peek ahead in the leaf looking for xattrs. If we
2013 * don't find any xattrs, we know there can't be any acls.
2015 * slot is the slot the inode is in, objectid is the objectid of the inode
2017 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2018 int slot
, u64 objectid
)
2020 u32 nritems
= btrfs_header_nritems(leaf
);
2021 struct btrfs_key found_key
;
2025 while (slot
< nritems
) {
2026 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2028 /* we found a different objectid, there must not be acls */
2029 if (found_key
.objectid
!= objectid
)
2032 /* we found an xattr, assume we've got an acl */
2033 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2037 * we found a key greater than an xattr key, there can't
2038 * be any acls later on
2040 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2047 * it goes inode, inode backrefs, xattrs, extents,
2048 * so if there are a ton of hard links to an inode there can
2049 * be a lot of backrefs. Don't waste time searching too hard,
2050 * this is just an optimization
2055 /* we hit the end of the leaf before we found an xattr or
2056 * something larger than an xattr. We have to assume the inode
2063 * read an inode from the btree into the in-memory inode
2065 static void btrfs_read_locked_inode(struct inode
*inode
)
2067 struct btrfs_path
*path
;
2068 struct extent_buffer
*leaf
;
2069 struct btrfs_inode_item
*inode_item
;
2070 struct btrfs_timespec
*tspec
;
2071 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2072 struct btrfs_key location
;
2074 u64 alloc_group_block
;
2078 path
= btrfs_alloc_path();
2080 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2082 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2086 leaf
= path
->nodes
[0];
2087 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2088 struct btrfs_inode_item
);
2090 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2091 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2092 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2093 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2094 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2096 tspec
= btrfs_inode_atime(inode_item
);
2097 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2098 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2100 tspec
= btrfs_inode_mtime(inode_item
);
2101 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2102 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2104 tspec
= btrfs_inode_ctime(inode_item
);
2105 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2106 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2108 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2109 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2110 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2111 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2113 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2115 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2116 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2118 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2121 * try to precache a NULL acl entry for files that don't have
2122 * any xattrs or acls
2124 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0], inode
->i_ino
);
2126 BTRFS_I(inode
)->i_acl
= NULL
;
2127 BTRFS_I(inode
)->i_default_acl
= NULL
;
2130 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2131 alloc_group_block
, 0);
2132 btrfs_free_path(path
);
2135 switch (inode
->i_mode
& S_IFMT
) {
2137 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2138 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2139 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2140 inode
->i_fop
= &btrfs_file_operations
;
2141 inode
->i_op
= &btrfs_file_inode_operations
;
2144 inode
->i_fop
= &btrfs_dir_file_operations
;
2145 if (root
== root
->fs_info
->tree_root
)
2146 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2148 inode
->i_op
= &btrfs_dir_inode_operations
;
2151 inode
->i_op
= &btrfs_symlink_inode_operations
;
2152 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2153 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2156 inode
->i_op
= &btrfs_special_inode_operations
;
2157 init_special_inode(inode
, inode
->i_mode
, rdev
);
2161 btrfs_update_iflags(inode
);
2165 btrfs_free_path(path
);
2166 make_bad_inode(inode
);
2170 * given a leaf and an inode, copy the inode fields into the leaf
2172 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2173 struct extent_buffer
*leaf
,
2174 struct btrfs_inode_item
*item
,
2175 struct inode
*inode
)
2177 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2178 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2179 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2180 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2181 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2183 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2184 inode
->i_atime
.tv_sec
);
2185 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2186 inode
->i_atime
.tv_nsec
);
2188 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2189 inode
->i_mtime
.tv_sec
);
2190 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2191 inode
->i_mtime
.tv_nsec
);
2193 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2194 inode
->i_ctime
.tv_sec
);
2195 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2196 inode
->i_ctime
.tv_nsec
);
2198 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2199 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2200 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2201 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2202 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2203 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2204 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2208 * copy everything in the in-memory inode into the btree.
2210 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2211 struct btrfs_root
*root
, struct inode
*inode
)
2213 struct btrfs_inode_item
*inode_item
;
2214 struct btrfs_path
*path
;
2215 struct extent_buffer
*leaf
;
2218 path
= btrfs_alloc_path();
2220 path
->leave_spinning
= 1;
2221 ret
= btrfs_lookup_inode(trans
, root
, path
,
2222 &BTRFS_I(inode
)->location
, 1);
2229 btrfs_unlock_up_safe(path
, 1);
2230 leaf
= path
->nodes
[0];
2231 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2232 struct btrfs_inode_item
);
2234 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2235 btrfs_mark_buffer_dirty(leaf
);
2236 btrfs_set_inode_last_trans(trans
, inode
);
2239 btrfs_free_path(path
);
2245 * unlink helper that gets used here in inode.c and in the tree logging
2246 * recovery code. It remove a link in a directory with a given name, and
2247 * also drops the back refs in the inode to the directory
2249 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2250 struct btrfs_root
*root
,
2251 struct inode
*dir
, struct inode
*inode
,
2252 const char *name
, int name_len
)
2254 struct btrfs_path
*path
;
2256 struct extent_buffer
*leaf
;
2257 struct btrfs_dir_item
*di
;
2258 struct btrfs_key key
;
2261 path
= btrfs_alloc_path();
2267 path
->leave_spinning
= 1;
2268 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2269 name
, name_len
, -1);
2278 leaf
= path
->nodes
[0];
2279 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2280 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2283 btrfs_release_path(root
, path
);
2285 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2287 dir
->i_ino
, &index
);
2289 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2290 "inode %lu parent %lu\n", name_len
, name
,
2291 inode
->i_ino
, dir
->i_ino
);
2295 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2296 index
, name
, name_len
, -1);
2305 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2306 btrfs_release_path(root
, path
);
2308 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2310 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2312 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2316 btrfs_free_path(path
);
2320 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2321 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2322 btrfs_update_inode(trans
, root
, dir
);
2323 btrfs_drop_nlink(inode
);
2324 ret
= btrfs_update_inode(trans
, root
, inode
);
2325 dir
->i_sb
->s_dirt
= 1;
2330 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2332 struct btrfs_root
*root
;
2333 struct btrfs_trans_handle
*trans
;
2334 struct inode
*inode
= dentry
->d_inode
;
2336 unsigned long nr
= 0;
2338 root
= BTRFS_I(dir
)->root
;
2340 trans
= btrfs_start_transaction(root
, 1);
2342 btrfs_set_trans_block_group(trans
, dir
);
2344 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2346 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2347 dentry
->d_name
.name
, dentry
->d_name
.len
);
2349 if (inode
->i_nlink
== 0)
2350 ret
= btrfs_orphan_add(trans
, inode
);
2352 nr
= trans
->blocks_used
;
2354 btrfs_end_transaction_throttle(trans
, root
);
2355 btrfs_btree_balance_dirty(root
, nr
);
2359 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2361 struct inode
*inode
= dentry
->d_inode
;
2364 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2365 struct btrfs_trans_handle
*trans
;
2366 unsigned long nr
= 0;
2369 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2370 * the root of a subvolume or snapshot
2372 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2373 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
2377 trans
= btrfs_start_transaction(root
, 1);
2378 btrfs_set_trans_block_group(trans
, dir
);
2380 err
= btrfs_orphan_add(trans
, inode
);
2384 /* now the directory is empty */
2385 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2386 dentry
->d_name
.name
, dentry
->d_name
.len
);
2388 btrfs_i_size_write(inode
, 0);
2391 nr
= trans
->blocks_used
;
2392 ret
= btrfs_end_transaction_throttle(trans
, root
);
2393 btrfs_btree_balance_dirty(root
, nr
);
2402 * when truncating bytes in a file, it is possible to avoid reading
2403 * the leaves that contain only checksum items. This can be the
2404 * majority of the IO required to delete a large file, but it must
2405 * be done carefully.
2407 * The keys in the level just above the leaves are checked to make sure
2408 * the lowest key in a given leaf is a csum key, and starts at an offset
2409 * after the new size.
2411 * Then the key for the next leaf is checked to make sure it also has
2412 * a checksum item for the same file. If it does, we know our target leaf
2413 * contains only checksum items, and it can be safely freed without reading
2416 * This is just an optimization targeted at large files. It may do
2417 * nothing. It will return 0 unless things went badly.
2419 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
2420 struct btrfs_root
*root
,
2421 struct btrfs_path
*path
,
2422 struct inode
*inode
, u64 new_size
)
2424 struct btrfs_key key
;
2427 struct btrfs_key found_key
;
2428 struct btrfs_key other_key
;
2429 struct btrfs_leaf_ref
*ref
;
2433 path
->lowest_level
= 1;
2434 key
.objectid
= inode
->i_ino
;
2435 key
.type
= BTRFS_CSUM_ITEM_KEY
;
2436 key
.offset
= new_size
;
2438 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2442 if (path
->nodes
[1] == NULL
) {
2447 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
2448 nritems
= btrfs_header_nritems(path
->nodes
[1]);
2453 if (path
->slots
[1] >= nritems
)
2456 /* did we find a key greater than anything we want to delete? */
2457 if (found_key
.objectid
> inode
->i_ino
||
2458 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
2461 /* we check the next key in the node to make sure the leave contains
2462 * only checksum items. This comparison doesn't work if our
2463 * leaf is the last one in the node
2465 if (path
->slots
[1] + 1 >= nritems
) {
2467 /* search forward from the last key in the node, this
2468 * will bring us into the next node in the tree
2470 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
2472 /* unlikely, but we inc below, so check to be safe */
2473 if (found_key
.offset
== (u64
)-1)
2476 /* search_forward needs a path with locks held, do the
2477 * search again for the original key. It is possible
2478 * this will race with a balance and return a path that
2479 * we could modify, but this drop is just an optimization
2480 * and is allowed to miss some leaves.
2482 btrfs_release_path(root
, path
);
2485 /* setup a max key for search_forward */
2486 other_key
.offset
= (u64
)-1;
2487 other_key
.type
= key
.type
;
2488 other_key
.objectid
= key
.objectid
;
2490 path
->keep_locks
= 1;
2491 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
2493 path
->keep_locks
= 0;
2494 if (ret
|| found_key
.objectid
!= key
.objectid
||
2495 found_key
.type
!= key
.type
) {
2500 key
.offset
= found_key
.offset
;
2501 btrfs_release_path(root
, path
);
2506 /* we know there's one more slot after us in the tree,
2507 * read that key so we can verify it is also a checksum item
2509 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
2511 if (found_key
.objectid
< inode
->i_ino
)
2514 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
2518 * if the key for the next leaf isn't a csum key from this objectid,
2519 * we can't be sure there aren't good items inside this leaf.
2522 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
2525 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
2526 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
2528 * it is safe to delete this leaf, it contains only
2529 * csum items from this inode at an offset >= new_size
2531 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
2534 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
2535 ref
= btrfs_alloc_leaf_ref(root
, 0);
2537 ref
->root_gen
= root
->root_key
.offset
;
2538 ref
->bytenr
= leaf_start
;
2540 ref
->generation
= leaf_gen
;
2543 btrfs_sort_leaf_ref(ref
);
2545 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
2547 btrfs_free_leaf_ref(root
, ref
);
2553 btrfs_release_path(root
, path
);
2555 if (other_key
.objectid
== inode
->i_ino
&&
2556 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
2557 key
.offset
= other_key
.offset
;
2563 /* fixup any changes we've made to the path */
2564 path
->lowest_level
= 0;
2565 path
->keep_locks
= 0;
2566 btrfs_release_path(root
, path
);
2573 * this can truncate away extent items, csum items and directory items.
2574 * It starts at a high offset and removes keys until it can't find
2575 * any higher than new_size
2577 * csum items that cross the new i_size are truncated to the new size
2580 * min_type is the minimum key type to truncate down to. If set to 0, this
2581 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2583 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2584 struct btrfs_root
*root
,
2585 struct inode
*inode
,
2586 u64 new_size
, u32 min_type
)
2589 struct btrfs_path
*path
;
2590 struct btrfs_key key
;
2591 struct btrfs_key found_key
;
2592 u32 found_type
= (u8
)-1;
2593 struct extent_buffer
*leaf
;
2594 struct btrfs_file_extent_item
*fi
;
2595 u64 extent_start
= 0;
2596 u64 extent_num_bytes
= 0;
2597 u64 extent_offset
= 0;
2601 int pending_del_nr
= 0;
2602 int pending_del_slot
= 0;
2603 int extent_type
= -1;
2605 u64 mask
= root
->sectorsize
- 1;
2608 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2609 path
= btrfs_alloc_path();
2613 /* FIXME, add redo link to tree so we don't leak on crash */
2614 key
.objectid
= inode
->i_ino
;
2615 key
.offset
= (u64
)-1;
2619 path
->leave_spinning
= 1;
2620 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2625 /* there are no items in the tree for us to truncate, we're
2628 if (path
->slots
[0] == 0) {
2637 leaf
= path
->nodes
[0];
2638 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2639 found_type
= btrfs_key_type(&found_key
);
2642 if (found_key
.objectid
!= inode
->i_ino
)
2645 if (found_type
< min_type
)
2648 item_end
= found_key
.offset
;
2649 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2650 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2651 struct btrfs_file_extent_item
);
2652 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2653 encoding
= btrfs_file_extent_compression(leaf
, fi
);
2654 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
2655 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
2657 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2659 btrfs_file_extent_num_bytes(leaf
, fi
);
2660 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2661 item_end
+= btrfs_file_extent_inline_len(leaf
,
2666 if (item_end
< new_size
) {
2667 if (found_type
== BTRFS_DIR_ITEM_KEY
)
2668 found_type
= BTRFS_INODE_ITEM_KEY
;
2669 else if (found_type
== BTRFS_EXTENT_ITEM_KEY
)
2670 found_type
= BTRFS_EXTENT_DATA_KEY
;
2671 else if (found_type
== BTRFS_EXTENT_DATA_KEY
)
2672 found_type
= BTRFS_XATTR_ITEM_KEY
;
2673 else if (found_type
== BTRFS_XATTR_ITEM_KEY
)
2674 found_type
= BTRFS_INODE_REF_KEY
;
2675 else if (found_type
)
2679 btrfs_set_key_type(&key
, found_type
);
2682 if (found_key
.offset
>= new_size
)
2688 /* FIXME, shrink the extent if the ref count is only 1 */
2689 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2692 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2694 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2695 if (!del_item
&& !encoding
) {
2696 u64 orig_num_bytes
=
2697 btrfs_file_extent_num_bytes(leaf
, fi
);
2698 extent_num_bytes
= new_size
-
2699 found_key
.offset
+ root
->sectorsize
- 1;
2700 extent_num_bytes
= extent_num_bytes
&
2701 ~((u64
)root
->sectorsize
- 1);
2702 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2704 num_dec
= (orig_num_bytes
-
2706 if (root
->ref_cows
&& extent_start
!= 0)
2707 inode_sub_bytes(inode
, num_dec
);
2708 btrfs_mark_buffer_dirty(leaf
);
2711 btrfs_file_extent_disk_num_bytes(leaf
,
2713 extent_offset
= found_key
.offset
-
2714 btrfs_file_extent_offset(leaf
, fi
);
2716 /* FIXME blocksize != 4096 */
2717 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2718 if (extent_start
!= 0) {
2721 inode_sub_bytes(inode
, num_dec
);
2724 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2726 * we can't truncate inline items that have had
2730 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2731 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2732 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2733 u32 size
= new_size
- found_key
.offset
;
2735 if (root
->ref_cows
) {
2736 inode_sub_bytes(inode
, item_end
+ 1 -
2740 btrfs_file_extent_calc_inline_size(size
);
2741 ret
= btrfs_truncate_item(trans
, root
, path
,
2744 } else if (root
->ref_cows
) {
2745 inode_sub_bytes(inode
, item_end
+ 1 -
2751 if (!pending_del_nr
) {
2752 /* no pending yet, add ourselves */
2753 pending_del_slot
= path
->slots
[0];
2755 } else if (pending_del_nr
&&
2756 path
->slots
[0] + 1 == pending_del_slot
) {
2757 /* hop on the pending chunk */
2759 pending_del_slot
= path
->slots
[0];
2766 if (found_extent
&& root
->ref_cows
) {
2767 btrfs_set_path_blocking(path
);
2768 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2769 extent_num_bytes
, 0,
2770 btrfs_header_owner(leaf
),
2771 inode
->i_ino
, extent_offset
);
2775 if (path
->slots
[0] == 0) {
2778 btrfs_release_path(root
, path
);
2779 if (found_type
== BTRFS_INODE_ITEM_KEY
)
2785 if (pending_del_nr
&&
2786 path
->slots
[0] + 1 != pending_del_slot
) {
2787 struct btrfs_key debug
;
2789 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2791 ret
= btrfs_del_items(trans
, root
, path
,
2796 btrfs_release_path(root
, path
);
2797 if (found_type
== BTRFS_INODE_ITEM_KEY
)
2804 if (pending_del_nr
) {
2805 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2808 btrfs_free_path(path
);
2809 inode
->i_sb
->s_dirt
= 1;
2814 * taken from block_truncate_page, but does cow as it zeros out
2815 * any bytes left in the last page in the file.
2817 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2819 struct inode
*inode
= mapping
->host
;
2820 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2821 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2822 struct btrfs_ordered_extent
*ordered
;
2824 u32 blocksize
= root
->sectorsize
;
2825 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2826 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2832 if ((offset
& (blocksize
- 1)) == 0)
2837 page
= grab_cache_page(mapping
, index
);
2841 page_start
= page_offset(page
);
2842 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2844 if (!PageUptodate(page
)) {
2845 ret
= btrfs_readpage(NULL
, page
);
2847 if (page
->mapping
!= mapping
) {
2849 page_cache_release(page
);
2852 if (!PageUptodate(page
)) {
2857 wait_on_page_writeback(page
);
2859 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2860 set_page_extent_mapped(page
);
2862 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2864 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2866 page_cache_release(page
);
2867 btrfs_start_ordered_extent(inode
, ordered
, 1);
2868 btrfs_put_ordered_extent(ordered
);
2872 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2874 if (offset
!= PAGE_CACHE_SIZE
) {
2876 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2877 flush_dcache_page(page
);
2880 ClearPageChecked(page
);
2881 set_page_dirty(page
);
2882 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2886 page_cache_release(page
);
2891 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
2893 struct btrfs_trans_handle
*trans
;
2894 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2895 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2896 struct extent_map
*em
;
2897 u64 mask
= root
->sectorsize
- 1;
2898 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2899 u64 block_end
= (size
+ mask
) & ~mask
;
2905 if (size
<= hole_start
)
2908 err
= btrfs_check_metadata_free_space(root
);
2912 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2915 struct btrfs_ordered_extent
*ordered
;
2916 btrfs_wait_ordered_range(inode
, hole_start
,
2917 block_end
- hole_start
);
2918 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2919 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2922 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2923 btrfs_put_ordered_extent(ordered
);
2926 trans
= btrfs_start_transaction(root
, 1);
2927 btrfs_set_trans_block_group(trans
, inode
);
2929 cur_offset
= hole_start
;
2931 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2932 block_end
- cur_offset
, 0);
2933 BUG_ON(IS_ERR(em
) || !em
);
2934 last_byte
= min(extent_map_end(em
), block_end
);
2935 last_byte
= (last_byte
+ mask
) & ~mask
;
2936 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2938 hole_size
= last_byte
- cur_offset
;
2939 err
= btrfs_drop_extents(trans
, root
, inode
,
2941 cur_offset
+ hole_size
,
2943 cur_offset
, &hint_byte
);
2946 err
= btrfs_insert_file_extent(trans
, root
,
2947 inode
->i_ino
, cur_offset
, 0,
2948 0, hole_size
, 0, hole_size
,
2950 btrfs_drop_extent_cache(inode
, hole_start
,
2953 free_extent_map(em
);
2954 cur_offset
= last_byte
;
2955 if (err
|| cur_offset
>= block_end
)
2959 btrfs_end_transaction(trans
, root
);
2960 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2964 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2966 struct inode
*inode
= dentry
->d_inode
;
2969 err
= inode_change_ok(inode
, attr
);
2973 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
2974 if (attr
->ia_size
> inode
->i_size
) {
2975 err
= btrfs_cont_expand(inode
, attr
->ia_size
);
2978 } else if (inode
->i_size
> 0 &&
2979 attr
->ia_size
== 0) {
2981 /* we're truncating a file that used to have good
2982 * data down to zero. Make sure it gets into
2983 * the ordered flush list so that any new writes
2984 * get down to disk quickly.
2986 BTRFS_I(inode
)->ordered_data_close
= 1;
2990 err
= inode_setattr(inode
, attr
);
2992 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2993 err
= btrfs_acl_chmod(inode
);
2997 void btrfs_delete_inode(struct inode
*inode
)
2999 struct btrfs_trans_handle
*trans
;
3000 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3004 truncate_inode_pages(&inode
->i_data
, 0);
3005 if (is_bad_inode(inode
)) {
3006 btrfs_orphan_del(NULL
, inode
);
3009 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3011 btrfs_i_size_write(inode
, 0);
3012 trans
= btrfs_join_transaction(root
, 1);
3014 btrfs_set_trans_block_group(trans
, inode
);
3015 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
3017 btrfs_orphan_del(NULL
, inode
);
3018 goto no_delete_lock
;
3021 btrfs_orphan_del(trans
, inode
);
3023 nr
= trans
->blocks_used
;
3026 btrfs_end_transaction(trans
, root
);
3027 btrfs_btree_balance_dirty(root
, nr
);
3031 nr
= trans
->blocks_used
;
3032 btrfs_end_transaction(trans
, root
);
3033 btrfs_btree_balance_dirty(root
, nr
);
3039 * this returns the key found in the dir entry in the location pointer.
3040 * If no dir entries were found, location->objectid is 0.
3042 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3043 struct btrfs_key
*location
)
3045 const char *name
= dentry
->d_name
.name
;
3046 int namelen
= dentry
->d_name
.len
;
3047 struct btrfs_dir_item
*di
;
3048 struct btrfs_path
*path
;
3049 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3052 path
= btrfs_alloc_path();
3055 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
3060 if (!di
|| IS_ERR(di
))
3063 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3065 btrfs_free_path(path
);
3068 location
->objectid
= 0;
3073 * when we hit a tree root in a directory, the btrfs part of the inode
3074 * needs to be changed to reflect the root directory of the tree root. This
3075 * is kind of like crossing a mount point.
3077 static int fixup_tree_root_location(struct btrfs_root
*root
,
3078 struct btrfs_key
*location
,
3079 struct btrfs_root
**sub_root
,
3080 struct dentry
*dentry
)
3082 struct btrfs_root_item
*ri
;
3084 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
3086 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
3089 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
3090 dentry
->d_name
.name
,
3091 dentry
->d_name
.len
);
3092 if (IS_ERR(*sub_root
))
3093 return PTR_ERR(*sub_root
);
3095 ri
= &(*sub_root
)->root_item
;
3096 location
->objectid
= btrfs_root_dirid(ri
);
3097 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3098 location
->offset
= 0;
3103 static void inode_tree_add(struct inode
*inode
)
3105 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3106 struct btrfs_inode
*entry
;
3107 struct rb_node
**p
= &root
->inode_tree
.rb_node
;
3108 struct rb_node
*parent
= NULL
;
3110 spin_lock(&root
->inode_lock
);
3113 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3115 if (inode
->i_ino
< entry
->vfs_inode
.i_ino
)
3117 else if (inode
->i_ino
> entry
->vfs_inode
.i_ino
)
3118 p
= &(*p
)->rb_right
;
3120 WARN_ON(!(entry
->vfs_inode
.i_state
&
3121 (I_WILL_FREE
| I_FREEING
| I_CLEAR
)));
3125 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3126 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3127 spin_unlock(&root
->inode_lock
);
3130 static void inode_tree_del(struct inode
*inode
)
3132 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3134 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3135 spin_lock(&root
->inode_lock
);
3136 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3137 spin_unlock(&root
->inode_lock
);
3138 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3142 static noinline
void init_btrfs_i(struct inode
*inode
)
3144 struct btrfs_inode
*bi
= BTRFS_I(inode
);
3146 bi
->i_acl
= BTRFS_ACL_NOT_CACHED
;
3147 bi
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
3152 bi
->logged_trans
= 0;
3153 bi
->delalloc_bytes
= 0;
3154 bi
->reserved_bytes
= 0;
3155 bi
->disk_i_size
= 0;
3157 bi
->index_cnt
= (u64
)-1;
3158 bi
->last_unlink_trans
= 0;
3159 bi
->ordered_data_close
= 0;
3160 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
3161 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
3162 inode
->i_mapping
, GFP_NOFS
);
3163 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
3164 inode
->i_mapping
, GFP_NOFS
);
3165 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
3166 INIT_LIST_HEAD(&BTRFS_I(inode
)->ordered_operations
);
3167 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3168 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
3169 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
3170 mutex_init(&BTRFS_I(inode
)->log_mutex
);
3173 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3175 struct btrfs_iget_args
*args
= p
;
3176 inode
->i_ino
= args
->ino
;
3177 init_btrfs_i(inode
);
3178 BTRFS_I(inode
)->root
= args
->root
;
3179 btrfs_set_inode_space_info(args
->root
, inode
);
3183 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3185 struct btrfs_iget_args
*args
= opaque
;
3186 return args
->ino
== inode
->i_ino
&&
3187 args
->root
== BTRFS_I(inode
)->root
;
3190 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
3192 struct btrfs_root
*root
)
3194 struct inode
*inode
;
3195 struct btrfs_iget_args args
;
3196 args
.ino
= objectid
;
3199 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3200 btrfs_init_locked_inode
,
3205 /* Get an inode object given its location and corresponding root.
3206 * Returns in *is_new if the inode was read from disk
3208 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3209 struct btrfs_root
*root
)
3211 struct inode
*inode
;
3213 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3215 return ERR_PTR(-ENOMEM
);
3217 if (inode
->i_state
& I_NEW
) {
3218 BTRFS_I(inode
)->root
= root
;
3219 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3220 btrfs_read_locked_inode(inode
);
3222 inode_tree_add(inode
);
3223 unlock_new_inode(inode
);
3229 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3231 struct inode
*inode
;
3232 struct btrfs_inode
*bi
= BTRFS_I(dir
);
3233 struct btrfs_root
*root
= bi
->root
;
3234 struct btrfs_root
*sub_root
= root
;
3235 struct btrfs_key location
;
3238 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3239 return ERR_PTR(-ENAMETOOLONG
);
3241 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3244 return ERR_PTR(ret
);
3247 if (location
.objectid
) {
3248 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
3251 return ERR_PTR(ret
);
3253 return ERR_PTR(-ENOENT
);
3254 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
);
3256 return ERR_CAST(inode
);
3261 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
3262 struct nameidata
*nd
)
3264 struct inode
*inode
;
3266 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3267 return ERR_PTR(-ENAMETOOLONG
);
3269 inode
= btrfs_lookup_dentry(dir
, dentry
);
3271 return ERR_CAST(inode
);
3273 return d_splice_alias(inode
, dentry
);
3276 static unsigned char btrfs_filetype_table
[] = {
3277 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
3280 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
3283 struct inode
*inode
= filp
->f_dentry
->d_inode
;
3284 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3285 struct btrfs_item
*item
;
3286 struct btrfs_dir_item
*di
;
3287 struct btrfs_key key
;
3288 struct btrfs_key found_key
;
3289 struct btrfs_path
*path
;
3292 struct extent_buffer
*leaf
;
3295 unsigned char d_type
;
3300 int key_type
= BTRFS_DIR_INDEX_KEY
;
3305 /* FIXME, use a real flag for deciding about the key type */
3306 if (root
->fs_info
->tree_root
== root
)
3307 key_type
= BTRFS_DIR_ITEM_KEY
;
3309 /* special case for "." */
3310 if (filp
->f_pos
== 0) {
3311 over
= filldir(dirent
, ".", 1,
3318 /* special case for .., just use the back ref */
3319 if (filp
->f_pos
== 1) {
3320 u64 pino
= parent_ino(filp
->f_path
.dentry
);
3321 over
= filldir(dirent
, "..", 2,
3327 path
= btrfs_alloc_path();
3330 btrfs_set_key_type(&key
, key_type
);
3331 key
.offset
= filp
->f_pos
;
3332 key
.objectid
= inode
->i_ino
;
3334 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3340 leaf
= path
->nodes
[0];
3341 nritems
= btrfs_header_nritems(leaf
);
3342 slot
= path
->slots
[0];
3343 if (advance
|| slot
>= nritems
) {
3344 if (slot
>= nritems
- 1) {
3345 ret
= btrfs_next_leaf(root
, path
);
3348 leaf
= path
->nodes
[0];
3349 nritems
= btrfs_header_nritems(leaf
);
3350 slot
= path
->slots
[0];
3358 item
= btrfs_item_nr(leaf
, slot
);
3359 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3361 if (found_key
.objectid
!= key
.objectid
)
3363 if (btrfs_key_type(&found_key
) != key_type
)
3365 if (found_key
.offset
< filp
->f_pos
)
3368 filp
->f_pos
= found_key
.offset
;
3370 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
3372 di_total
= btrfs_item_size(leaf
, item
);
3374 while (di_cur
< di_total
) {
3375 struct btrfs_key location
;
3377 name_len
= btrfs_dir_name_len(leaf
, di
);
3378 if (name_len
<= sizeof(tmp_name
)) {
3379 name_ptr
= tmp_name
;
3381 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
3387 read_extent_buffer(leaf
, name_ptr
,
3388 (unsigned long)(di
+ 1), name_len
);
3390 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
3391 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
3393 /* is this a reference to our own snapshot? If so
3396 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
3397 location
.objectid
== root
->root_key
.objectid
) {
3401 over
= filldir(dirent
, name_ptr
, name_len
,
3402 found_key
.offset
, location
.objectid
,
3406 if (name_ptr
!= tmp_name
)
3411 di_len
= btrfs_dir_name_len(leaf
, di
) +
3412 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
3414 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
3418 /* Reached end of directory/root. Bump pos past the last item. */
3419 if (key_type
== BTRFS_DIR_INDEX_KEY
)
3420 filp
->f_pos
= INT_LIMIT(off_t
);
3426 btrfs_free_path(path
);
3430 int btrfs_write_inode(struct inode
*inode
, int wait
)
3432 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3433 struct btrfs_trans_handle
*trans
;
3436 if (root
->fs_info
->btree_inode
== inode
)
3440 trans
= btrfs_join_transaction(root
, 1);
3441 btrfs_set_trans_block_group(trans
, inode
);
3442 ret
= btrfs_commit_transaction(trans
, root
);
3448 * This is somewhat expensive, updating the tree every time the
3449 * inode changes. But, it is most likely to find the inode in cache.
3450 * FIXME, needs more benchmarking...there are no reasons other than performance
3451 * to keep or drop this code.
3453 void btrfs_dirty_inode(struct inode
*inode
)
3455 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3456 struct btrfs_trans_handle
*trans
;
3458 trans
= btrfs_join_transaction(root
, 1);
3459 btrfs_set_trans_block_group(trans
, inode
);
3460 btrfs_update_inode(trans
, root
, inode
);
3461 btrfs_end_transaction(trans
, root
);
3465 * find the highest existing sequence number in a directory
3466 * and then set the in-memory index_cnt variable to reflect
3467 * free sequence numbers
3469 static int btrfs_set_inode_index_count(struct inode
*inode
)
3471 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3472 struct btrfs_key key
, found_key
;
3473 struct btrfs_path
*path
;
3474 struct extent_buffer
*leaf
;
3477 key
.objectid
= inode
->i_ino
;
3478 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
3479 key
.offset
= (u64
)-1;
3481 path
= btrfs_alloc_path();
3485 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3488 /* FIXME: we should be able to handle this */
3494 * MAGIC NUMBER EXPLANATION:
3495 * since we search a directory based on f_pos we have to start at 2
3496 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3497 * else has to start at 2
3499 if (path
->slots
[0] == 0) {
3500 BTRFS_I(inode
)->index_cnt
= 2;
3506 leaf
= path
->nodes
[0];
3507 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3509 if (found_key
.objectid
!= inode
->i_ino
||
3510 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
3511 BTRFS_I(inode
)->index_cnt
= 2;
3515 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
3517 btrfs_free_path(path
);
3522 * helper to find a free sequence number in a given directory. This current
3523 * code is very simple, later versions will do smarter things in the btree
3525 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
3529 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
3530 ret
= btrfs_set_inode_index_count(dir
);
3535 *index
= BTRFS_I(dir
)->index_cnt
;
3536 BTRFS_I(dir
)->index_cnt
++;
3541 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
3542 struct btrfs_root
*root
,
3544 const char *name
, int name_len
,
3545 u64 ref_objectid
, u64 objectid
,
3546 u64 alloc_hint
, int mode
, u64
*index
)
3548 struct inode
*inode
;
3549 struct btrfs_inode_item
*inode_item
;
3550 struct btrfs_key
*location
;
3551 struct btrfs_path
*path
;
3552 struct btrfs_inode_ref
*ref
;
3553 struct btrfs_key key
[2];
3559 path
= btrfs_alloc_path();
3562 inode
= new_inode(root
->fs_info
->sb
);
3564 return ERR_PTR(-ENOMEM
);
3567 ret
= btrfs_set_inode_index(dir
, index
);
3570 return ERR_PTR(ret
);
3574 * index_cnt is ignored for everything but a dir,
3575 * btrfs_get_inode_index_count has an explanation for the magic
3578 init_btrfs_i(inode
);
3579 BTRFS_I(inode
)->index_cnt
= 2;
3580 BTRFS_I(inode
)->root
= root
;
3581 BTRFS_I(inode
)->generation
= trans
->transid
;
3582 btrfs_set_inode_space_info(root
, inode
);
3588 BTRFS_I(inode
)->block_group
=
3589 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
3591 key
[0].objectid
= objectid
;
3592 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
3595 key
[1].objectid
= objectid
;
3596 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
3597 key
[1].offset
= ref_objectid
;
3599 sizes
[0] = sizeof(struct btrfs_inode_item
);
3600 sizes
[1] = name_len
+ sizeof(*ref
);
3602 path
->leave_spinning
= 1;
3603 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
3607 if (objectid
> root
->highest_inode
)
3608 root
->highest_inode
= objectid
;
3610 inode
->i_uid
= current_fsuid();
3612 if (dir
&& (dir
->i_mode
& S_ISGID
)) {
3613 inode
->i_gid
= dir
->i_gid
;
3617 inode
->i_gid
= current_fsgid();
3619 inode
->i_mode
= mode
;
3620 inode
->i_ino
= objectid
;
3621 inode_set_bytes(inode
, 0);
3622 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3623 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3624 struct btrfs_inode_item
);
3625 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
3627 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3628 struct btrfs_inode_ref
);
3629 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
3630 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
3631 ptr
= (unsigned long)(ref
+ 1);
3632 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
3634 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3635 btrfs_free_path(path
);
3637 location
= &BTRFS_I(inode
)->location
;
3638 location
->objectid
= objectid
;
3639 location
->offset
= 0;
3640 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3642 btrfs_inherit_iflags(inode
, dir
);
3644 if ((mode
& S_IFREG
)) {
3645 if (btrfs_test_opt(root
, NODATASUM
))
3646 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
3647 if (btrfs_test_opt(root
, NODATACOW
))
3648 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
3651 insert_inode_hash(inode
);
3652 inode_tree_add(inode
);
3656 BTRFS_I(dir
)->index_cnt
--;
3657 btrfs_free_path(path
);
3659 return ERR_PTR(ret
);
3662 static inline u8
btrfs_inode_type(struct inode
*inode
)
3664 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
3668 * utility function to add 'inode' into 'parent_inode' with
3669 * a give name and a given sequence number.
3670 * if 'add_backref' is true, also insert a backref from the
3671 * inode to the parent directory.
3673 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3674 struct inode
*parent_inode
, struct inode
*inode
,
3675 const char *name
, int name_len
, int add_backref
, u64 index
)
3678 struct btrfs_key key
;
3679 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3681 key
.objectid
= inode
->i_ino
;
3682 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3685 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3686 parent_inode
->i_ino
,
3687 &key
, btrfs_inode_type(inode
),
3691 ret
= btrfs_insert_inode_ref(trans
, root
,
3694 parent_inode
->i_ino
,
3697 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3699 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3700 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3705 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3706 struct dentry
*dentry
, struct inode
*inode
,
3707 int backref
, u64 index
)
3709 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3710 inode
, dentry
->d_name
.name
,
3711 dentry
->d_name
.len
, backref
, index
);
3713 d_instantiate(dentry
, inode
);
3721 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3722 int mode
, dev_t rdev
)
3724 struct btrfs_trans_handle
*trans
;
3725 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3726 struct inode
*inode
= NULL
;
3730 unsigned long nr
= 0;
3733 if (!new_valid_dev(rdev
))
3736 err
= btrfs_check_metadata_free_space(root
);
3740 trans
= btrfs_start_transaction(root
, 1);
3741 btrfs_set_trans_block_group(trans
, dir
);
3743 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3749 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3751 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3752 BTRFS_I(dir
)->block_group
, mode
, &index
);
3753 err
= PTR_ERR(inode
);
3757 err
= btrfs_init_inode_security(inode
, dir
);
3763 btrfs_set_trans_block_group(trans
, inode
);
3764 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3768 inode
->i_op
= &btrfs_special_inode_operations
;
3769 init_special_inode(inode
, inode
->i_mode
, rdev
);
3770 btrfs_update_inode(trans
, root
, inode
);
3772 dir
->i_sb
->s_dirt
= 1;
3773 btrfs_update_inode_block_group(trans
, inode
);
3774 btrfs_update_inode_block_group(trans
, dir
);
3776 nr
= trans
->blocks_used
;
3777 btrfs_end_transaction_throttle(trans
, root
);
3780 inode_dec_link_count(inode
);
3783 btrfs_btree_balance_dirty(root
, nr
);
3787 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3788 int mode
, struct nameidata
*nd
)
3790 struct btrfs_trans_handle
*trans
;
3791 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3792 struct inode
*inode
= NULL
;
3795 unsigned long nr
= 0;
3799 err
= btrfs_check_metadata_free_space(root
);
3802 trans
= btrfs_start_transaction(root
, 1);
3803 btrfs_set_trans_block_group(trans
, dir
);
3805 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3811 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3813 dentry
->d_parent
->d_inode
->i_ino
,
3814 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3816 err
= PTR_ERR(inode
);
3820 err
= btrfs_init_inode_security(inode
, dir
);
3826 btrfs_set_trans_block_group(trans
, inode
);
3827 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3831 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3832 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3833 inode
->i_fop
= &btrfs_file_operations
;
3834 inode
->i_op
= &btrfs_file_inode_operations
;
3835 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3837 dir
->i_sb
->s_dirt
= 1;
3838 btrfs_update_inode_block_group(trans
, inode
);
3839 btrfs_update_inode_block_group(trans
, dir
);
3841 nr
= trans
->blocks_used
;
3842 btrfs_end_transaction_throttle(trans
, root
);
3845 inode_dec_link_count(inode
);
3848 btrfs_btree_balance_dirty(root
, nr
);
3852 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3853 struct dentry
*dentry
)
3855 struct btrfs_trans_handle
*trans
;
3856 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3857 struct inode
*inode
= old_dentry
->d_inode
;
3859 unsigned long nr
= 0;
3863 if (inode
->i_nlink
== 0)
3866 btrfs_inc_nlink(inode
);
3867 err
= btrfs_check_metadata_free_space(root
);
3870 err
= btrfs_set_inode_index(dir
, &index
);
3874 trans
= btrfs_start_transaction(root
, 1);
3876 btrfs_set_trans_block_group(trans
, dir
);
3877 atomic_inc(&inode
->i_count
);
3879 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3884 dir
->i_sb
->s_dirt
= 1;
3885 btrfs_update_inode_block_group(trans
, dir
);
3886 err
= btrfs_update_inode(trans
, root
, inode
);
3891 nr
= trans
->blocks_used
;
3893 btrfs_log_new_name(trans
, inode
, NULL
, dentry
->d_parent
);
3894 btrfs_end_transaction_throttle(trans
, root
);
3897 inode_dec_link_count(inode
);
3900 btrfs_btree_balance_dirty(root
, nr
);
3904 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3906 struct inode
*inode
= NULL
;
3907 struct btrfs_trans_handle
*trans
;
3908 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3910 int drop_on_err
= 0;
3913 unsigned long nr
= 1;
3915 err
= btrfs_check_metadata_free_space(root
);
3919 trans
= btrfs_start_transaction(root
, 1);
3920 btrfs_set_trans_block_group(trans
, dir
);
3922 if (IS_ERR(trans
)) {
3923 err
= PTR_ERR(trans
);
3927 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3933 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3935 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3936 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3938 if (IS_ERR(inode
)) {
3939 err
= PTR_ERR(inode
);
3945 err
= btrfs_init_inode_security(inode
, dir
);
3949 inode
->i_op
= &btrfs_dir_inode_operations
;
3950 inode
->i_fop
= &btrfs_dir_file_operations
;
3951 btrfs_set_trans_block_group(trans
, inode
);
3953 btrfs_i_size_write(inode
, 0);
3954 err
= btrfs_update_inode(trans
, root
, inode
);
3958 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3959 inode
, dentry
->d_name
.name
,
3960 dentry
->d_name
.len
, 0, index
);
3964 d_instantiate(dentry
, inode
);
3966 dir
->i_sb
->s_dirt
= 1;
3967 btrfs_update_inode_block_group(trans
, inode
);
3968 btrfs_update_inode_block_group(trans
, dir
);
3971 nr
= trans
->blocks_used
;
3972 btrfs_end_transaction_throttle(trans
, root
);
3977 btrfs_btree_balance_dirty(root
, nr
);
3981 /* helper for btfs_get_extent. Given an existing extent in the tree,
3982 * and an extent that you want to insert, deal with overlap and insert
3983 * the new extent into the tree.
3985 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3986 struct extent_map
*existing
,
3987 struct extent_map
*em
,
3988 u64 map_start
, u64 map_len
)
3992 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3993 start_diff
= map_start
- em
->start
;
3994 em
->start
= map_start
;
3996 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3997 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3998 em
->block_start
+= start_diff
;
3999 em
->block_len
-= start_diff
;
4001 return add_extent_mapping(em_tree
, em
);
4004 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4005 struct inode
*inode
, struct page
*page
,
4006 size_t pg_offset
, u64 extent_offset
,
4007 struct btrfs_file_extent_item
*item
)
4010 struct extent_buffer
*leaf
= path
->nodes
[0];
4013 unsigned long inline_size
;
4016 WARN_ON(pg_offset
!= 0);
4017 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4018 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4019 btrfs_item_nr(leaf
, path
->slots
[0]));
4020 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4021 ptr
= btrfs_file_extent_inline_start(item
);
4023 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4025 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4026 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
4027 inline_size
, max_size
);
4029 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4030 unsigned long copy_size
= min_t(u64
,
4031 PAGE_CACHE_SIZE
- pg_offset
,
4032 max_size
- extent_offset
);
4033 memset(kaddr
+ pg_offset
, 0, copy_size
);
4034 kunmap_atomic(kaddr
, KM_USER0
);
4041 * a bit scary, this does extent mapping from logical file offset to the disk.
4042 * the ugly parts come from merging extents from the disk with the in-ram
4043 * representation. This gets more complex because of the data=ordered code,
4044 * where the in-ram extents might be locked pending data=ordered completion.
4046 * This also copies inline extents directly into the page.
4049 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
4050 size_t pg_offset
, u64 start
, u64 len
,
4056 u64 extent_start
= 0;
4058 u64 objectid
= inode
->i_ino
;
4060 struct btrfs_path
*path
= NULL
;
4061 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4062 struct btrfs_file_extent_item
*item
;
4063 struct extent_buffer
*leaf
;
4064 struct btrfs_key found_key
;
4065 struct extent_map
*em
= NULL
;
4066 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4067 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4068 struct btrfs_trans_handle
*trans
= NULL
;
4072 spin_lock(&em_tree
->lock
);
4073 em
= lookup_extent_mapping(em_tree
, start
, len
);
4075 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4076 spin_unlock(&em_tree
->lock
);
4079 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
4080 free_extent_map(em
);
4081 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
4082 free_extent_map(em
);
4086 em
= alloc_extent_map(GFP_NOFS
);
4091 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4092 em
->start
= EXTENT_MAP_HOLE
;
4093 em
->orig_start
= EXTENT_MAP_HOLE
;
4095 em
->block_len
= (u64
)-1;
4098 path
= btrfs_alloc_path();
4102 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
4103 objectid
, start
, trans
!= NULL
);
4110 if (path
->slots
[0] == 0)
4115 leaf
= path
->nodes
[0];
4116 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
4117 struct btrfs_file_extent_item
);
4118 /* are we inside the extent that was found? */
4119 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4120 found_type
= btrfs_key_type(&found_key
);
4121 if (found_key
.objectid
!= objectid
||
4122 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
4126 found_type
= btrfs_file_extent_type(leaf
, item
);
4127 extent_start
= found_key
.offset
;
4128 compressed
= btrfs_file_extent_compression(leaf
, item
);
4129 if (found_type
== BTRFS_FILE_EXTENT_REG
||
4130 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4131 extent_end
= extent_start
+
4132 btrfs_file_extent_num_bytes(leaf
, item
);
4133 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4135 size
= btrfs_file_extent_inline_len(leaf
, item
);
4136 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
4137 ~((u64
)root
->sectorsize
- 1);
4140 if (start
>= extent_end
) {
4142 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
4143 ret
= btrfs_next_leaf(root
, path
);
4150 leaf
= path
->nodes
[0];
4152 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4153 if (found_key
.objectid
!= objectid
||
4154 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4156 if (start
+ len
<= found_key
.offset
)
4159 em
->len
= found_key
.offset
- start
;
4163 if (found_type
== BTRFS_FILE_EXTENT_REG
||
4164 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4165 em
->start
= extent_start
;
4166 em
->len
= extent_end
- extent_start
;
4167 em
->orig_start
= extent_start
-
4168 btrfs_file_extent_offset(leaf
, item
);
4169 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
4171 em
->block_start
= EXTENT_MAP_HOLE
;
4175 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4176 em
->block_start
= bytenr
;
4177 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
4180 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
4181 em
->block_start
= bytenr
;
4182 em
->block_len
= em
->len
;
4183 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
4184 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
4187 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4191 size_t extent_offset
;
4194 em
->block_start
= EXTENT_MAP_INLINE
;
4195 if (!page
|| create
) {
4196 em
->start
= extent_start
;
4197 em
->len
= extent_end
- extent_start
;
4201 size
= btrfs_file_extent_inline_len(leaf
, item
);
4202 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
4203 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
4204 size
- extent_offset
);
4205 em
->start
= extent_start
+ extent_offset
;
4206 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
4207 ~((u64
)root
->sectorsize
- 1);
4208 em
->orig_start
= EXTENT_MAP_INLINE
;
4210 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4211 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
4212 if (create
== 0 && !PageUptodate(page
)) {
4213 if (btrfs_file_extent_compression(leaf
, item
) ==
4214 BTRFS_COMPRESS_ZLIB
) {
4215 ret
= uncompress_inline(path
, inode
, page
,
4217 extent_offset
, item
);
4221 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4225 flush_dcache_page(page
);
4226 } else if (create
&& PageUptodate(page
)) {
4229 free_extent_map(em
);
4231 btrfs_release_path(root
, path
);
4232 trans
= btrfs_join_transaction(root
, 1);
4236 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4239 btrfs_mark_buffer_dirty(leaf
);
4241 set_extent_uptodate(io_tree
, em
->start
,
4242 extent_map_end(em
) - 1, GFP_NOFS
);
4245 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
4252 em
->block_start
= EXTENT_MAP_HOLE
;
4253 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
4255 btrfs_release_path(root
, path
);
4256 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
4257 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
4258 "[%llu %llu]\n", (unsigned long long)em
->start
,
4259 (unsigned long long)em
->len
,
4260 (unsigned long long)start
,
4261 (unsigned long long)len
);
4267 spin_lock(&em_tree
->lock
);
4268 ret
= add_extent_mapping(em_tree
, em
);
4269 /* it is possible that someone inserted the extent into the tree
4270 * while we had the lock dropped. It is also possible that
4271 * an overlapping map exists in the tree
4273 if (ret
== -EEXIST
) {
4274 struct extent_map
*existing
;
4278 existing
= lookup_extent_mapping(em_tree
, start
, len
);
4279 if (existing
&& (existing
->start
> start
||
4280 existing
->start
+ existing
->len
<= start
)) {
4281 free_extent_map(existing
);
4285 existing
= lookup_extent_mapping(em_tree
, em
->start
,
4288 err
= merge_extent_mapping(em_tree
, existing
,
4291 free_extent_map(existing
);
4293 free_extent_map(em
);
4298 free_extent_map(em
);
4302 free_extent_map(em
);
4307 spin_unlock(&em_tree
->lock
);
4310 btrfs_free_path(path
);
4312 ret
= btrfs_end_transaction(trans
, root
);
4317 free_extent_map(em
);
4318 return ERR_PTR(err
);
4323 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
4324 const struct iovec
*iov
, loff_t offset
,
4325 unsigned long nr_segs
)
4330 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4331 __u64 start
, __u64 len
)
4333 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent
);
4336 int btrfs_readpage(struct file
*file
, struct page
*page
)
4338 struct extent_io_tree
*tree
;
4339 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4340 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
4343 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
4345 struct extent_io_tree
*tree
;
4348 if (current
->flags
& PF_MEMALLOC
) {
4349 redirty_page_for_writepage(wbc
, page
);
4353 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4354 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
4357 int btrfs_writepages(struct address_space
*mapping
,
4358 struct writeback_control
*wbc
)
4360 struct extent_io_tree
*tree
;
4362 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4363 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
4367 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
4368 struct list_head
*pages
, unsigned nr_pages
)
4370 struct extent_io_tree
*tree
;
4371 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4372 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
4375 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4377 struct extent_io_tree
*tree
;
4378 struct extent_map_tree
*map
;
4381 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4382 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
4383 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
4385 ClearPagePrivate(page
);
4386 set_page_private(page
, 0);
4387 page_cache_release(page
);
4392 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4394 if (PageWriteback(page
) || PageDirty(page
))
4396 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
4399 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
4401 struct extent_io_tree
*tree
;
4402 struct btrfs_ordered_extent
*ordered
;
4403 u64 page_start
= page_offset(page
);
4404 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4406 wait_on_page_writeback(page
);
4407 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4409 btrfs_releasepage(page
, GFP_NOFS
);
4413 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4414 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
4418 * IO on this page will never be started, so we need
4419 * to account for any ordered extents now
4421 clear_extent_bit(tree
, page_start
, page_end
,
4422 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4423 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
4424 btrfs_finish_ordered_io(page
->mapping
->host
,
4425 page_start
, page_end
);
4426 btrfs_put_ordered_extent(ordered
);
4427 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4429 clear_extent_bit(tree
, page_start
, page_end
,
4430 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4433 __btrfs_releasepage(page
, GFP_NOFS
);
4435 ClearPageChecked(page
);
4436 if (PagePrivate(page
)) {
4437 ClearPagePrivate(page
);
4438 set_page_private(page
, 0);
4439 page_cache_release(page
);
4444 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4445 * called from a page fault handler when a page is first dirtied. Hence we must
4446 * be careful to check for EOF conditions here. We set the page up correctly
4447 * for a written page which means we get ENOSPC checking when writing into
4448 * holes and correct delalloc and unwritten extent mapping on filesystems that
4449 * support these features.
4451 * We are not allowed to take the i_mutex here so we have to play games to
4452 * protect against truncate races as the page could now be beyond EOF. Because
4453 * vmtruncate() writes the inode size before removing pages, once we have the
4454 * page lock we can determine safely if the page is beyond EOF. If it is not
4455 * beyond EOF, then the page is guaranteed safe against truncation until we
4458 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4460 struct page
*page
= vmf
->page
;
4461 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
4462 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4463 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4464 struct btrfs_ordered_extent
*ordered
;
4466 unsigned long zero_start
;
4472 ret
= btrfs_check_data_free_space(root
, inode
, PAGE_CACHE_SIZE
);
4476 else /* -ENOSPC, -EIO, etc */
4477 ret
= VM_FAULT_SIGBUS
;
4481 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
4484 size
= i_size_read(inode
);
4485 page_start
= page_offset(page
);
4486 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4488 if ((page
->mapping
!= inode
->i_mapping
) ||
4489 (page_start
>= size
)) {
4490 btrfs_free_reserved_data_space(root
, inode
, PAGE_CACHE_SIZE
);
4491 /* page got truncated out from underneath us */
4494 wait_on_page_writeback(page
);
4496 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4497 set_page_extent_mapped(page
);
4500 * we can't set the delalloc bits if there are pending ordered
4501 * extents. Drop our locks and wait for them to finish
4503 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4505 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4507 btrfs_start_ordered_extent(inode
, ordered
, 1);
4508 btrfs_put_ordered_extent(ordered
);
4512 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
4515 /* page is wholly or partially inside EOF */
4516 if (page_start
+ PAGE_CACHE_SIZE
> size
)
4517 zero_start
= size
& ~PAGE_CACHE_MASK
;
4519 zero_start
= PAGE_CACHE_SIZE
;
4521 if (zero_start
!= PAGE_CACHE_SIZE
) {
4523 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
4524 flush_dcache_page(page
);
4527 ClearPageChecked(page
);
4528 set_page_dirty(page
);
4530 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
+ 1;
4531 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4539 static void btrfs_truncate(struct inode
*inode
)
4541 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4543 struct btrfs_trans_handle
*trans
;
4545 u64 mask
= root
->sectorsize
- 1;
4547 if (!S_ISREG(inode
->i_mode
))
4549 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4552 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
4553 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
4555 trans
= btrfs_start_transaction(root
, 1);
4558 * setattr is responsible for setting the ordered_data_close flag,
4559 * but that is only tested during the last file release. That
4560 * could happen well after the next commit, leaving a great big
4561 * window where new writes may get lost if someone chooses to write
4562 * to this file after truncating to zero
4564 * The inode doesn't have any dirty data here, and so if we commit
4565 * this is a noop. If someone immediately starts writing to the inode
4566 * it is very likely we'll catch some of their writes in this
4567 * transaction, and the commit will find this file on the ordered
4568 * data list with good things to send down.
4570 * This is a best effort solution, there is still a window where
4571 * using truncate to replace the contents of the file will
4572 * end up with a zero length file after a crash.
4574 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
4575 btrfs_add_ordered_operation(trans
, root
, inode
);
4577 btrfs_set_trans_block_group(trans
, inode
);
4578 btrfs_i_size_write(inode
, inode
->i_size
);
4580 ret
= btrfs_orphan_add(trans
, inode
);
4583 /* FIXME, add redo link to tree so we don't leak on crash */
4584 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
4585 BTRFS_EXTENT_DATA_KEY
);
4586 btrfs_update_inode(trans
, root
, inode
);
4588 ret
= btrfs_orphan_del(trans
, inode
);
4592 nr
= trans
->blocks_used
;
4593 ret
= btrfs_end_transaction_throttle(trans
, root
);
4595 btrfs_btree_balance_dirty(root
, nr
);
4599 * create a new subvolume directory/inode (helper for the ioctl).
4601 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
4602 struct btrfs_root
*new_root
, struct dentry
*dentry
,
4603 u64 new_dirid
, u64 alloc_hint
)
4605 struct inode
*inode
;
4609 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
4610 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
4612 return PTR_ERR(inode
);
4613 inode
->i_op
= &btrfs_dir_inode_operations
;
4614 inode
->i_fop
= &btrfs_dir_file_operations
;
4617 btrfs_i_size_write(inode
, 0);
4619 error
= btrfs_update_inode(trans
, new_root
, inode
);
4623 d_instantiate(dentry
, inode
);
4627 /* helper function for file defrag and space balancing. This
4628 * forces readahead on a given range of bytes in an inode
4630 unsigned long btrfs_force_ra(struct address_space
*mapping
,
4631 struct file_ra_state
*ra
, struct file
*file
,
4632 pgoff_t offset
, pgoff_t last_index
)
4634 pgoff_t req_size
= last_index
- offset
+ 1;
4636 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
4637 return offset
+ req_size
;
4640 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
4642 struct btrfs_inode
*ei
;
4644 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
4648 ei
->logged_trans
= 0;
4649 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
4650 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
4651 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
4652 INIT_LIST_HEAD(&ei
->i_orphan
);
4653 INIT_LIST_HEAD(&ei
->ordered_operations
);
4654 return &ei
->vfs_inode
;
4657 void btrfs_destroy_inode(struct inode
*inode
)
4659 struct btrfs_ordered_extent
*ordered
;
4660 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4662 WARN_ON(!list_empty(&inode
->i_dentry
));
4663 WARN_ON(inode
->i_data
.nrpages
);
4665 if (BTRFS_I(inode
)->i_acl
&&
4666 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
4667 posix_acl_release(BTRFS_I(inode
)->i_acl
);
4668 if (BTRFS_I(inode
)->i_default_acl
&&
4669 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
4670 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
4673 * Make sure we're properly removed from the ordered operation
4677 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
4678 spin_lock(&root
->fs_info
->ordered_extent_lock
);
4679 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
4680 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
4683 spin_lock(&root
->list_lock
);
4684 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
4685 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
4686 " list\n", inode
->i_ino
);
4689 spin_unlock(&root
->list_lock
);
4692 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
4696 printk(KERN_ERR
"btrfs found ordered "
4697 "extent %llu %llu on inode cleanup\n",
4698 (unsigned long long)ordered
->file_offset
,
4699 (unsigned long long)ordered
->len
);
4700 btrfs_remove_ordered_extent(inode
, ordered
);
4701 btrfs_put_ordered_extent(ordered
);
4702 btrfs_put_ordered_extent(ordered
);
4705 inode_tree_del(inode
);
4706 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4707 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4710 static void init_once(void *foo
)
4712 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4714 inode_init_once(&ei
->vfs_inode
);
4717 void btrfs_destroy_cachep(void)
4719 if (btrfs_inode_cachep
)
4720 kmem_cache_destroy(btrfs_inode_cachep
);
4721 if (btrfs_trans_handle_cachep
)
4722 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4723 if (btrfs_transaction_cachep
)
4724 kmem_cache_destroy(btrfs_transaction_cachep
);
4725 if (btrfs_path_cachep
)
4726 kmem_cache_destroy(btrfs_path_cachep
);
4729 int btrfs_init_cachep(void)
4731 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
4732 sizeof(struct btrfs_inode
), 0,
4733 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
4734 if (!btrfs_inode_cachep
)
4737 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
4738 sizeof(struct btrfs_trans_handle
), 0,
4739 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
4740 if (!btrfs_trans_handle_cachep
)
4743 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
4744 sizeof(struct btrfs_transaction
), 0,
4745 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
4746 if (!btrfs_transaction_cachep
)
4749 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
4750 sizeof(struct btrfs_path
), 0,
4751 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
4752 if (!btrfs_path_cachep
)
4757 btrfs_destroy_cachep();
4761 static int btrfs_getattr(struct vfsmount
*mnt
,
4762 struct dentry
*dentry
, struct kstat
*stat
)
4764 struct inode
*inode
= dentry
->d_inode
;
4765 generic_fillattr(inode
, stat
);
4766 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
4767 stat
->blksize
= PAGE_CACHE_SIZE
;
4768 stat
->blocks
= (inode_get_bytes(inode
) +
4769 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4773 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4774 struct inode
*new_dir
, struct dentry
*new_dentry
)
4776 struct btrfs_trans_handle
*trans
;
4777 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4778 struct inode
*new_inode
= new_dentry
->d_inode
;
4779 struct inode
*old_inode
= old_dentry
->d_inode
;
4780 struct timespec ctime
= CURRENT_TIME
;
4784 /* we're not allowed to rename between subvolumes */
4785 if (BTRFS_I(old_inode
)->root
->root_key
.objectid
!=
4786 BTRFS_I(new_dir
)->root
->root_key
.objectid
)
4789 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4790 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4794 /* to rename a snapshot or subvolume, we need to juggle the
4795 * backrefs. This isn't coded yet
4797 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
4800 ret
= btrfs_check_metadata_free_space(root
);
4805 * we're using rename to replace one file with another.
4806 * and the replacement file is large. Start IO on it now so
4807 * we don't add too much work to the end of the transaction
4809 if (new_inode
&& old_inode
&& S_ISREG(old_inode
->i_mode
) &&
4810 new_inode
->i_size
&&
4811 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
4812 filemap_flush(old_inode
->i_mapping
);
4814 trans
= btrfs_start_transaction(root
, 1);
4817 * make sure the inode gets flushed if it is replacing
4820 if (new_inode
&& new_inode
->i_size
&&
4821 old_inode
&& S_ISREG(old_inode
->i_mode
)) {
4822 btrfs_add_ordered_operation(trans
, root
, old_inode
);
4826 * this is an ugly little race, but the rename is required to make
4827 * sure that if we crash, the inode is either at the old name
4828 * or the new one. pinning the log transaction lets us make sure
4829 * we don't allow a log commit to come in after we unlink the
4830 * name but before we add the new name back in.
4832 btrfs_pin_log_trans(root
);
4834 btrfs_set_trans_block_group(trans
, new_dir
);
4836 btrfs_inc_nlink(old_dentry
->d_inode
);
4837 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4838 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4839 old_inode
->i_ctime
= ctime
;
4841 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
4842 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
4844 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4845 old_dentry
->d_name
.name
,
4846 old_dentry
->d_name
.len
);
4851 new_inode
->i_ctime
= CURRENT_TIME
;
4852 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4853 new_dentry
->d_inode
,
4854 new_dentry
->d_name
.name
,
4855 new_dentry
->d_name
.len
);
4858 if (new_inode
->i_nlink
== 0) {
4859 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4865 ret
= btrfs_set_inode_index(new_dir
, &index
);
4869 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4870 old_inode
, new_dentry
->d_name
.name
,
4871 new_dentry
->d_name
.len
, 1, index
);
4875 btrfs_log_new_name(trans
, old_inode
, old_dir
,
4876 new_dentry
->d_parent
);
4879 /* this btrfs_end_log_trans just allows the current
4880 * log-sub transaction to complete
4882 btrfs_end_log_trans(root
);
4883 btrfs_end_transaction_throttle(trans
, root
);
4889 * some fairly slow code that needs optimization. This walks the list
4890 * of all the inodes with pending delalloc and forces them to disk.
4892 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4894 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4895 struct btrfs_inode
*binode
;
4896 struct inode
*inode
;
4898 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
4901 spin_lock(&root
->fs_info
->delalloc_lock
);
4902 while (!list_empty(head
)) {
4903 binode
= list_entry(head
->next
, struct btrfs_inode
,
4905 inode
= igrab(&binode
->vfs_inode
);
4907 list_del_init(&binode
->delalloc_inodes
);
4908 spin_unlock(&root
->fs_info
->delalloc_lock
);
4910 filemap_flush(inode
->i_mapping
);
4914 spin_lock(&root
->fs_info
->delalloc_lock
);
4916 spin_unlock(&root
->fs_info
->delalloc_lock
);
4918 /* the filemap_flush will queue IO into the worker threads, but
4919 * we have to make sure the IO is actually started and that
4920 * ordered extents get created before we return
4922 atomic_inc(&root
->fs_info
->async_submit_draining
);
4923 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
4924 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
4925 wait_event(root
->fs_info
->async_submit_wait
,
4926 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
4927 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
4929 atomic_dec(&root
->fs_info
->async_submit_draining
);
4933 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4934 const char *symname
)
4936 struct btrfs_trans_handle
*trans
;
4937 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4938 struct btrfs_path
*path
;
4939 struct btrfs_key key
;
4940 struct inode
*inode
= NULL
;
4948 struct btrfs_file_extent_item
*ei
;
4949 struct extent_buffer
*leaf
;
4950 unsigned long nr
= 0;
4952 name_len
= strlen(symname
) + 1;
4953 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4954 return -ENAMETOOLONG
;
4956 err
= btrfs_check_metadata_free_space(root
);
4960 trans
= btrfs_start_transaction(root
, 1);
4961 btrfs_set_trans_block_group(trans
, dir
);
4963 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4969 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4971 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4972 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4974 err
= PTR_ERR(inode
);
4978 err
= btrfs_init_inode_security(inode
, dir
);
4984 btrfs_set_trans_block_group(trans
, inode
);
4985 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4989 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4990 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4991 inode
->i_fop
= &btrfs_file_operations
;
4992 inode
->i_op
= &btrfs_file_inode_operations
;
4993 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4995 dir
->i_sb
->s_dirt
= 1;
4996 btrfs_update_inode_block_group(trans
, inode
);
4997 btrfs_update_inode_block_group(trans
, dir
);
5001 path
= btrfs_alloc_path();
5003 key
.objectid
= inode
->i_ino
;
5005 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
5006 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
5007 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
5013 leaf
= path
->nodes
[0];
5014 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
5015 struct btrfs_file_extent_item
);
5016 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
5017 btrfs_set_file_extent_type(leaf
, ei
,
5018 BTRFS_FILE_EXTENT_INLINE
);
5019 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
5020 btrfs_set_file_extent_compression(leaf
, ei
, 0);
5021 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
5022 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
5024 ptr
= btrfs_file_extent_inline_start(ei
);
5025 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
5026 btrfs_mark_buffer_dirty(leaf
);
5027 btrfs_free_path(path
);
5029 inode
->i_op
= &btrfs_symlink_inode_operations
;
5030 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
5031 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
5032 inode_set_bytes(inode
, name_len
);
5033 btrfs_i_size_write(inode
, name_len
- 1);
5034 err
= btrfs_update_inode(trans
, root
, inode
);
5039 nr
= trans
->blocks_used
;
5040 btrfs_end_transaction_throttle(trans
, root
);
5043 inode_dec_link_count(inode
);
5046 btrfs_btree_balance_dirty(root
, nr
);
5050 static int prealloc_file_range(struct btrfs_trans_handle
*trans
,
5051 struct inode
*inode
, u64 start
, u64 end
,
5052 u64 locked_end
, u64 alloc_hint
, int mode
)
5054 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5055 struct btrfs_key ins
;
5057 u64 cur_offset
= start
;
5058 u64 num_bytes
= end
- start
;
5061 while (num_bytes
> 0) {
5062 alloc_size
= min(num_bytes
, root
->fs_info
->max_extent
);
5063 ret
= btrfs_reserve_extent(trans
, root
, alloc_size
,
5064 root
->sectorsize
, 0, alloc_hint
,
5070 ret
= insert_reserved_file_extent(trans
, inode
,
5071 cur_offset
, ins
.objectid
,
5072 ins
.offset
, ins
.offset
,
5073 ins
.offset
, locked_end
,
5075 BTRFS_FILE_EXTENT_PREALLOC
);
5077 num_bytes
-= ins
.offset
;
5078 cur_offset
+= ins
.offset
;
5079 alloc_hint
= ins
.objectid
+ ins
.offset
;
5082 if (cur_offset
> start
) {
5083 inode
->i_ctime
= CURRENT_TIME
;
5084 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
5085 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
5086 cur_offset
> i_size_read(inode
))
5087 btrfs_i_size_write(inode
, cur_offset
);
5088 ret
= btrfs_update_inode(trans
, root
, inode
);
5095 static long btrfs_fallocate(struct inode
*inode
, int mode
,
5096 loff_t offset
, loff_t len
)
5104 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
5105 struct extent_map
*em
;
5106 struct btrfs_trans_handle
*trans
;
5107 struct btrfs_root
*root
;
5110 alloc_start
= offset
& ~mask
;
5111 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
5114 * wait for ordered IO before we have any locks. We'll loop again
5115 * below with the locks held.
5117 btrfs_wait_ordered_range(inode
, alloc_start
, alloc_end
- alloc_start
);
5119 mutex_lock(&inode
->i_mutex
);
5120 if (alloc_start
> inode
->i_size
) {
5121 ret
= btrfs_cont_expand(inode
, alloc_start
);
5126 root
= BTRFS_I(inode
)->root
;
5128 ret
= btrfs_check_data_free_space(root
, inode
,
5129 alloc_end
- alloc_start
);
5133 locked_end
= alloc_end
- 1;
5135 struct btrfs_ordered_extent
*ordered
;
5137 trans
= btrfs_start_transaction(BTRFS_I(inode
)->root
, 1);
5143 /* the extent lock is ordered inside the running
5146 lock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
5148 ordered
= btrfs_lookup_first_ordered_extent(inode
,
5151 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
5152 ordered
->file_offset
< alloc_end
) {
5153 btrfs_put_ordered_extent(ordered
);
5154 unlock_extent(&BTRFS_I(inode
)->io_tree
,
5155 alloc_start
, locked_end
, GFP_NOFS
);
5156 btrfs_end_transaction(trans
, BTRFS_I(inode
)->root
);
5159 * we can't wait on the range with the transaction
5160 * running or with the extent lock held
5162 btrfs_wait_ordered_range(inode
, alloc_start
,
5163 alloc_end
- alloc_start
);
5166 btrfs_put_ordered_extent(ordered
);
5171 cur_offset
= alloc_start
;
5173 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
5174 alloc_end
- cur_offset
, 0);
5175 BUG_ON(IS_ERR(em
) || !em
);
5176 last_byte
= min(extent_map_end(em
), alloc_end
);
5177 last_byte
= (last_byte
+ mask
) & ~mask
;
5178 if (em
->block_start
== EXTENT_MAP_HOLE
) {
5179 ret
= prealloc_file_range(trans
, inode
, cur_offset
,
5180 last_byte
, locked_end
+ 1,
5183 free_extent_map(em
);
5187 if (em
->block_start
<= EXTENT_MAP_LAST_BYTE
)
5188 alloc_hint
= em
->block_start
;
5189 free_extent_map(em
);
5191 cur_offset
= last_byte
;
5192 if (cur_offset
>= alloc_end
) {
5197 unlock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, locked_end
,
5200 btrfs_end_transaction(trans
, BTRFS_I(inode
)->root
);
5202 btrfs_free_reserved_data_space(root
, inode
, alloc_end
- alloc_start
);
5204 mutex_unlock(&inode
->i_mutex
);
5208 static int btrfs_set_page_dirty(struct page
*page
)
5210 return __set_page_dirty_nobuffers(page
);
5213 static int btrfs_permission(struct inode
*inode
, int mask
)
5215 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
5217 return generic_permission(inode
, mask
, btrfs_check_acl
);
5220 static struct inode_operations btrfs_dir_inode_operations
= {
5221 .getattr
= btrfs_getattr
,
5222 .lookup
= btrfs_lookup
,
5223 .create
= btrfs_create
,
5224 .unlink
= btrfs_unlink
,
5226 .mkdir
= btrfs_mkdir
,
5227 .rmdir
= btrfs_rmdir
,
5228 .rename
= btrfs_rename
,
5229 .symlink
= btrfs_symlink
,
5230 .setattr
= btrfs_setattr
,
5231 .mknod
= btrfs_mknod
,
5232 .setxattr
= btrfs_setxattr
,
5233 .getxattr
= btrfs_getxattr
,
5234 .listxattr
= btrfs_listxattr
,
5235 .removexattr
= btrfs_removexattr
,
5236 .permission
= btrfs_permission
,
5238 static struct inode_operations btrfs_dir_ro_inode_operations
= {
5239 .lookup
= btrfs_lookup
,
5240 .permission
= btrfs_permission
,
5242 static struct file_operations btrfs_dir_file_operations
= {
5243 .llseek
= generic_file_llseek
,
5244 .read
= generic_read_dir
,
5245 .readdir
= btrfs_real_readdir
,
5246 .unlocked_ioctl
= btrfs_ioctl
,
5247 #ifdef CONFIG_COMPAT
5248 .compat_ioctl
= btrfs_ioctl
,
5250 .release
= btrfs_release_file
,
5251 .fsync
= btrfs_sync_file
,
5254 static struct extent_io_ops btrfs_extent_io_ops
= {
5255 .fill_delalloc
= run_delalloc_range
,
5256 .submit_bio_hook
= btrfs_submit_bio_hook
,
5257 .merge_bio_hook
= btrfs_merge_bio_hook
,
5258 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
5259 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
5260 .writepage_start_hook
= btrfs_writepage_start_hook
,
5261 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
5262 .set_bit_hook
= btrfs_set_bit_hook
,
5263 .clear_bit_hook
= btrfs_clear_bit_hook
,
5267 * btrfs doesn't support the bmap operation because swapfiles
5268 * use bmap to make a mapping of extents in the file. They assume
5269 * these extents won't change over the life of the file and they
5270 * use the bmap result to do IO directly to the drive.
5272 * the btrfs bmap call would return logical addresses that aren't
5273 * suitable for IO and they also will change frequently as COW
5274 * operations happen. So, swapfile + btrfs == corruption.
5276 * For now we're avoiding this by dropping bmap.
5278 static struct address_space_operations btrfs_aops
= {
5279 .readpage
= btrfs_readpage
,
5280 .writepage
= btrfs_writepage
,
5281 .writepages
= btrfs_writepages
,
5282 .readpages
= btrfs_readpages
,
5283 .sync_page
= block_sync_page
,
5284 .direct_IO
= btrfs_direct_IO
,
5285 .invalidatepage
= btrfs_invalidatepage
,
5286 .releasepage
= btrfs_releasepage
,
5287 .set_page_dirty
= btrfs_set_page_dirty
,
5290 static struct address_space_operations btrfs_symlink_aops
= {
5291 .readpage
= btrfs_readpage
,
5292 .writepage
= btrfs_writepage
,
5293 .invalidatepage
= btrfs_invalidatepage
,
5294 .releasepage
= btrfs_releasepage
,
5297 static struct inode_operations btrfs_file_inode_operations
= {
5298 .truncate
= btrfs_truncate
,
5299 .getattr
= btrfs_getattr
,
5300 .setattr
= btrfs_setattr
,
5301 .setxattr
= btrfs_setxattr
,
5302 .getxattr
= btrfs_getxattr
,
5303 .listxattr
= btrfs_listxattr
,
5304 .removexattr
= btrfs_removexattr
,
5305 .permission
= btrfs_permission
,
5306 .fallocate
= btrfs_fallocate
,
5307 .fiemap
= btrfs_fiemap
,
5309 static struct inode_operations btrfs_special_inode_operations
= {
5310 .getattr
= btrfs_getattr
,
5311 .setattr
= btrfs_setattr
,
5312 .permission
= btrfs_permission
,
5313 .setxattr
= btrfs_setxattr
,
5314 .getxattr
= btrfs_getxattr
,
5315 .listxattr
= btrfs_listxattr
,
5316 .removexattr
= btrfs_removexattr
,
5318 static struct inode_operations btrfs_symlink_inode_operations
= {
5319 .readlink
= generic_readlink
,
5320 .follow_link
= page_follow_link_light
,
5321 .put_link
= page_put_link
,
5322 .permission
= btrfs_permission
,
5323 .setxattr
= btrfs_setxattr
,
5324 .getxattr
= btrfs_getxattr
,
5325 .listxattr
= btrfs_listxattr
,
5326 .removexattr
= btrfs_removexattr
,