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/version.h>
38 #include <linux/xattr.h>
39 #include <linux/posix_acl.h>
42 #include "transaction.h"
43 #include "btrfs_inode.h"
45 #include "print-tree.h"
47 #include "ordered-data.h"
51 #include "ref-cache.h"
52 #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_bit_radix_cachep
;
73 struct kmem_cache
*btrfs_path_cachep
;
76 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
77 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
78 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
79 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
80 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
81 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
82 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
83 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
86 static void btrfs_truncate(struct inode
*inode
);
87 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
90 * a very lame attempt at stopping writes when the FS is 85% full. There
91 * are countless ways this is incorrect, but it is better than nothing.
93 int btrfs_check_free_space(struct btrfs_root
*root
, u64 num_required
,
102 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
103 total
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
104 used
= btrfs_super_bytes_used(&root
->fs_info
->super_copy
);
112 if (used
+ root
->fs_info
->delalloc_bytes
+ num_required
> thresh
)
114 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
119 * this does all the hard work for inserting an inline extent into
120 * the btree. The caller should have done a btrfs_drop_extents so that
121 * no overlapping inline items exist in the btree
123 static int noinline
insert_inline_extent(struct btrfs_trans_handle
*trans
,
124 struct btrfs_root
*root
, struct inode
*inode
,
125 u64 start
, size_t size
, size_t compressed_size
,
126 struct page
**compressed_pages
)
128 struct btrfs_key key
;
129 struct btrfs_path
*path
;
130 struct extent_buffer
*leaf
;
131 struct page
*page
= NULL
;
134 struct btrfs_file_extent_item
*ei
;
137 size_t cur_size
= size
;
139 unsigned long offset
;
140 int use_compress
= 0;
142 if (compressed_size
&& compressed_pages
) {
144 cur_size
= compressed_size
;
147 path
= btrfs_alloc_path(); if (!path
)
150 btrfs_set_trans_block_group(trans
, inode
);
152 key
.objectid
= inode
->i_ino
;
154 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
155 inode_add_bytes(inode
, size
);
156 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
158 inode_add_bytes(inode
, size
);
159 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
164 printk("got bad ret %d\n", ret
);
167 leaf
= path
->nodes
[0];
168 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
169 struct btrfs_file_extent_item
);
170 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
171 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
172 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
173 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
174 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
175 ptr
= btrfs_file_extent_inline_start(ei
);
180 while(compressed_size
> 0) {
181 cpage
= compressed_pages
[i
];
182 cur_size
= min(compressed_size
,
186 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
191 compressed_size
-= cur_size
;
193 btrfs_set_file_extent_compression(leaf
, ei
,
194 BTRFS_COMPRESS_ZLIB
);
196 page
= find_get_page(inode
->i_mapping
,
197 start
>> PAGE_CACHE_SHIFT
);
198 btrfs_set_file_extent_compression(leaf
, ei
, 0);
199 kaddr
= kmap_atomic(page
, KM_USER0
);
200 offset
= start
& (PAGE_CACHE_SIZE
- 1);
201 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
202 kunmap_atomic(kaddr
, KM_USER0
);
203 page_cache_release(page
);
205 btrfs_mark_buffer_dirty(leaf
);
206 btrfs_free_path(path
);
208 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
209 btrfs_update_inode(trans
, root
, inode
);
212 btrfs_free_path(path
);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
223 struct btrfs_root
*root
,
224 struct inode
*inode
, u64 start
, u64 end
,
225 size_t compressed_size
,
226 struct page
**compressed_pages
)
228 u64 isize
= i_size_read(inode
);
229 u64 actual_end
= min(end
+ 1, isize
);
230 u64 inline_len
= actual_end
- start
;
231 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
232 ~((u64
)root
->sectorsize
- 1);
234 u64 data_len
= inline_len
;
238 data_len
= compressed_size
;
241 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
243 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
245 data_len
> root
->fs_info
->max_inline
) {
249 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
250 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
251 aligned_end
, aligned_end
, &hint_byte
);
254 if (isize
> actual_end
)
255 inline_len
= min_t(u64
, isize
, actual_end
);
256 ret
= insert_inline_extent(trans
, root
, inode
, start
,
257 inline_len
, compressed_size
,
260 btrfs_drop_extent_cache(inode
, start
, aligned_end
, 0);
261 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
266 * when extent_io.c finds a delayed allocation range in the file,
267 * the call backs end up in this code. The basic idea is to
268 * allocate extents on disk for the range, and create ordered data structs
269 * in ram to track those extents.
271 * locked_page is the page that writepage had locked already. We use
272 * it to make sure we don't do extra locks or unlocks.
274 * *page_started is set to one if we unlock locked_page and do everything
275 * required to start IO on it. It may be clean and already done with
278 static int cow_file_range(struct inode
*inode
, struct page
*locked_page
,
279 u64 start
, u64 end
, int *page_started
)
281 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
282 struct btrfs_trans_handle
*trans
;
285 unsigned long ram_size
;
289 u64 blocksize
= root
->sectorsize
;
291 struct btrfs_key ins
;
292 struct extent_map
*em
;
293 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
295 struct page
**pages
= NULL
;
296 unsigned long nr_pages
;
297 unsigned long nr_pages_ret
= 0;
298 unsigned long total_compressed
= 0;
299 unsigned long total_in
= 0;
300 unsigned long max_compressed
= 128 * 1024;
301 unsigned long max_uncompressed
= 256 * 1024;
305 trans
= btrfs_join_transaction(root
, 1);
307 btrfs_set_trans_block_group(trans
, inode
);
311 * compression made this loop a bit ugly, but the basic idea is to
312 * compress some pages but keep the total size of the compressed
313 * extent relatively small. If compression is off, this goto target
318 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
319 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
321 actual_end
= min_t(u64
, i_size_read(inode
), end
+ 1);
322 total_compressed
= actual_end
- start
;
324 /* we want to make sure that amount of ram required to uncompress
325 * an extent is reasonable, so we limit the total size in ram
326 * of a compressed extent to 256k
328 total_compressed
= min(total_compressed
, max_uncompressed
);
329 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
330 num_bytes
= max(blocksize
, num_bytes
);
331 disk_num_bytes
= num_bytes
;
335 /* we do compression for mount -o compress and when the
336 * inode has not been flagged as nocompress
338 if (!btrfs_test_flag(inode
, NOCOMPRESS
) &&
339 btrfs_test_opt(root
, COMPRESS
)) {
341 pages
= kmalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
343 /* we want to make sure the amount of IO required to satisfy
344 * a random read is reasonably small, so we limit the size
345 * of a compressed extent to 128k
347 ret
= btrfs_zlib_compress_pages(inode
->i_mapping
, start
,
348 total_compressed
, pages
,
349 nr_pages
, &nr_pages_ret
,
355 unsigned long offset
= total_compressed
&
356 (PAGE_CACHE_SIZE
- 1);
357 struct page
*page
= pages
[nr_pages_ret
- 1];
360 /* zero the tail end of the last page, we might be
361 * sending it down to disk
364 kaddr
= kmap_atomic(page
, KM_USER0
);
365 memset(kaddr
+ offset
, 0,
366 PAGE_CACHE_SIZE
- offset
);
367 kunmap_atomic(kaddr
, KM_USER0
);
373 /* lets try to make an inline extent */
374 if (ret
|| total_in
< (end
- start
+ 1)) {
375 /* we didn't compress the entire range, try
376 * to make an uncompressed inline extent. This
377 * is almost sure to fail, but maybe inline sizes
378 * will get bigger later
380 ret
= cow_file_range_inline(trans
, root
, inode
,
381 start
, end
, 0, NULL
);
383 ret
= cow_file_range_inline(trans
, root
, inode
,
385 total_compressed
, pages
);
388 extent_clear_unlock_delalloc(inode
,
389 &BTRFS_I(inode
)->io_tree
,
400 * we aren't doing an inline extent round the compressed size
401 * up to a block size boundary so the allocator does sane
404 total_compressed
= (total_compressed
+ blocksize
- 1) &
408 * one last check to make sure the compression is really a
409 * win, compare the page count read with the blocks on disk
411 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
412 ~(PAGE_CACHE_SIZE
- 1);
413 if (total_compressed
>= total_in
) {
416 disk_num_bytes
= total_compressed
;
417 num_bytes
= total_in
;
420 if (!will_compress
&& pages
) {
422 * the compression code ran but failed to make things smaller,
423 * free any pages it allocated and our page pointer array
425 for (i
= 0; i
< nr_pages_ret
; i
++) {
426 page_cache_release(pages
[i
]);
430 total_compressed
= 0;
433 /* flag the file so we don't compress in the future */
434 btrfs_set_flag(inode
, NOCOMPRESS
);
437 BUG_ON(disk_num_bytes
>
438 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
440 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
441 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
442 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
444 while(disk_num_bytes
> 0) {
445 unsigned long min_bytes
;
448 * the max size of a compressed extent is pretty small,
449 * make the code a little less complex by forcing
450 * the allocator to find a whole compressed extent at once
453 min_bytes
= disk_num_bytes
;
455 min_bytes
= root
->sectorsize
;
457 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
458 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
459 min_bytes
, 0, alloc_hint
,
463 goto free_pages_out_fail
;
465 em
= alloc_extent_map(GFP_NOFS
);
469 ram_size
= num_bytes
;
472 /* ramsize == disk size */
473 ram_size
= ins
.offset
;
474 em
->len
= ins
.offset
;
477 em
->block_start
= ins
.objectid
;
478 em
->block_len
= ins
.offset
;
479 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
481 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
482 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
485 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
488 spin_lock(&em_tree
->lock
);
489 ret
= add_extent_mapping(em_tree
, em
);
490 spin_unlock(&em_tree
->lock
);
491 if (ret
!= -EEXIST
) {
495 btrfs_drop_extent_cache(inode
, start
,
496 start
+ ram_size
- 1, 0);
498 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
500 cur_alloc_size
= ins
.offset
;
501 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
502 ram_size
, cur_alloc_size
, 0,
506 if (disk_num_bytes
< cur_alloc_size
) {
507 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes
,
514 * we're doing compression, we and we need to
515 * submit the compressed extents down to the device.
517 * We lock down all the file pages, clearing their
518 * dirty bits and setting them writeback. Everyone
519 * that wants to modify the page will wait on the
520 * ordered extent above.
522 * The writeback bits on the file pages are
523 * cleared when the compressed pages are on disk
525 btrfs_end_transaction(trans
, root
);
527 if (start
<= page_offset(locked_page
) &&
528 page_offset(locked_page
) < start
+ ram_size
) {
532 extent_clear_unlock_delalloc(inode
,
533 &BTRFS_I(inode
)->io_tree
,
535 start
+ ram_size
- 1,
538 ret
= btrfs_submit_compressed_write(inode
, start
,
539 ram_size
, ins
.objectid
,
540 cur_alloc_size
, pages
,
544 trans
= btrfs_join_transaction(root
, 1);
545 if (start
+ ram_size
< end
) {
547 alloc_hint
= ins
.objectid
+ ins
.offset
;
548 /* pages will be freed at end_bio time */
552 /* we've written everything, time to go */
556 /* we're not doing compressed IO, don't unlock the first
557 * page (which the caller expects to stay locked), don't
558 * clear any dirty bits and don't set any writeback bits
560 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
561 start
, start
+ ram_size
- 1,
562 locked_page
, 0, 0, 0);
563 disk_num_bytes
-= cur_alloc_size
;
564 num_bytes
-= cur_alloc_size
;
565 alloc_hint
= ins
.objectid
+ ins
.offset
;
566 start
+= cur_alloc_size
;
571 btrfs_end_transaction(trans
, root
);
576 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
577 start
, end
, locked_page
, 0, 0, 0);
579 for (i
= 0; i
< nr_pages_ret
; i
++)
580 page_cache_release(pages
[i
]);
588 * when nowcow writeback call back. This checks for snapshots or COW copies
589 * of the extents that exist in the file, and COWs the file as required.
591 * If no cow copies or snapshots exist, we write directly to the existing
594 static int run_delalloc_nocow(struct inode
*inode
, struct page
*locked_page
,
595 u64 start
, u64 end
, int *page_started
)
602 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
603 struct btrfs_block_group_cache
*block_group
;
604 struct btrfs_trans_handle
*trans
;
605 struct extent_buffer
*leaf
;
607 struct btrfs_path
*path
;
608 struct btrfs_file_extent_item
*item
;
611 struct btrfs_key found_key
;
613 total_fs_bytes
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
614 path
= btrfs_alloc_path();
616 trans
= btrfs_join_transaction(root
, 1);
619 ret
= btrfs_lookup_file_extent(NULL
, root
, path
,
620 inode
->i_ino
, start
, 0);
627 if (path
->slots
[0] == 0)
632 leaf
= path
->nodes
[0];
633 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
634 struct btrfs_file_extent_item
);
636 /* are we inside the extent that was found? */
637 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
638 found_type
= btrfs_key_type(&found_key
);
639 if (found_key
.objectid
!= inode
->i_ino
||
640 found_type
!= BTRFS_EXTENT_DATA_KEY
)
643 found_type
= btrfs_file_extent_type(leaf
, item
);
644 extent_start
= found_key
.offset
;
645 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
646 u64 extent_num_bytes
;
648 extent_num_bytes
= btrfs_file_extent_num_bytes(leaf
, item
);
649 extent_end
= extent_start
+ extent_num_bytes
;
652 if (btrfs_file_extent_compression(leaf
, item
) ||
653 btrfs_file_extent_encryption(leaf
,item
) ||
654 btrfs_file_extent_other_encoding(leaf
, item
))
657 if (loops
&& start
!= extent_start
)
660 if (start
< extent_start
|| start
>= extent_end
)
663 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
667 if (btrfs_cross_ref_exists(trans
, root
, &found_key
, bytenr
))
670 * we may be called by the resizer, make sure we're inside
671 * the limits of the FS
673 block_group
= btrfs_lookup_block_group(root
->fs_info
,
675 if (!block_group
|| block_group
->ro
)
678 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
679 extent_num_bytes
= min(end
+ 1, extent_end
) - start
;
680 ret
= btrfs_add_ordered_extent(inode
, start
, bytenr
,
682 extent_num_bytes
, 1, 0);
688 btrfs_release_path(root
, path
);
696 btrfs_end_transaction(trans
, root
);
697 btrfs_free_path(path
);
698 return cow_file_range(inode
, locked_page
, start
, end
,
703 btrfs_end_transaction(trans
, root
);
704 btrfs_free_path(path
);
709 * extent_io.c call back to do delayed allocation processing
711 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
712 u64 start
, u64 end
, int *page_started
)
714 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
717 if (btrfs_test_opt(root
, NODATACOW
) ||
718 btrfs_test_flag(inode
, NODATACOW
))
719 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
722 ret
= cow_file_range(inode
, locked_page
, start
, end
,
729 * extent_io.c set_bit_hook, used to track delayed allocation
730 * bytes in this file, and to maintain the list of inodes that
731 * have pending delalloc work to be done.
733 int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
734 unsigned long old
, unsigned long bits
)
737 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
738 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
739 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
740 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
741 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
742 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
743 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
744 &root
->fs_info
->delalloc_inodes
);
746 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
752 * extent_io.c clear_bit_hook, see set_bit_hook for why
754 int btrfs_clear_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
755 unsigned long old
, unsigned long bits
)
757 if ((old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
758 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
761 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
762 if (end
- start
+ 1 > root
->fs_info
->delalloc_bytes
) {
763 printk("warning: delalloc account %Lu %Lu\n",
764 end
- start
+ 1, root
->fs_info
->delalloc_bytes
);
765 root
->fs_info
->delalloc_bytes
= 0;
766 BTRFS_I(inode
)->delalloc_bytes
= 0;
768 root
->fs_info
->delalloc_bytes
-= end
- start
+ 1;
769 BTRFS_I(inode
)->delalloc_bytes
-= end
- start
+ 1;
771 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
772 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
773 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
775 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
781 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
782 * we don't create bios that span stripes or chunks
784 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
785 size_t size
, struct bio
*bio
,
786 unsigned long bio_flags
)
788 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
789 struct btrfs_mapping_tree
*map_tree
;
790 u64 logical
= (u64
)bio
->bi_sector
<< 9;
795 length
= bio
->bi_size
;
796 map_tree
= &root
->fs_info
->mapping_tree
;
798 ret
= btrfs_map_block(map_tree
, READ
, logical
,
799 &map_length
, NULL
, 0);
801 if (map_length
< length
+ size
) {
808 * in order to insert checksums into the metadata in large chunks,
809 * we wait until bio submission time. All the pages in the bio are
810 * checksummed and sums are attached onto the ordered extent record.
812 * At IO completion time the cums attached on the ordered extent record
813 * are inserted into the btree
815 int __btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
816 int mirror_num
, unsigned long bio_flags
)
818 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
821 ret
= btrfs_csum_one_bio(root
, inode
, bio
);
824 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
828 * extent_io.c submission hook. This does the right thing for csum calculation on write,
829 * or reading the csums from the tree before a read
831 int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
832 int mirror_num
, unsigned long bio_flags
)
834 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
837 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
840 if (btrfs_test_opt(root
, NODATASUM
) ||
841 btrfs_test_flag(inode
, NODATASUM
)) {
845 if (!(rw
& (1 << BIO_RW
))) {
846 btrfs_lookup_bio_sums(root
, inode
, bio
);
848 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
849 return btrfs_submit_compressed_read(inode
, bio
,
850 mirror_num
, bio_flags
);
855 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
856 inode
, rw
, bio
, mirror_num
,
857 bio_flags
, __btrfs_submit_bio_hook
);
859 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
863 * given a list of ordered sums record them in the inode. This happens
864 * at IO completion time based on sums calculated at bio submission time.
866 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
867 struct inode
*inode
, u64 file_offset
,
868 struct list_head
*list
)
870 struct list_head
*cur
;
871 struct btrfs_ordered_sum
*sum
;
873 btrfs_set_trans_block_group(trans
, inode
);
874 list_for_each(cur
, list
) {
875 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
876 btrfs_csum_file_blocks(trans
, BTRFS_I(inode
)->root
,
882 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
884 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
888 /* see btrfs_writepage_start_hook for details on why this is required */
889 struct btrfs_writepage_fixup
{
891 struct btrfs_work work
;
894 void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
896 struct btrfs_writepage_fixup
*fixup
;
897 struct btrfs_ordered_extent
*ordered
;
903 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
907 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
908 ClearPageChecked(page
);
912 inode
= page
->mapping
->host
;
913 page_start
= page_offset(page
);
914 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
916 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
918 /* already ordered? We're done */
919 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
920 EXTENT_ORDERED
, 0)) {
924 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
926 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
929 btrfs_start_ordered_extent(inode
, ordered
, 1);
933 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
934 ClearPageChecked(page
);
936 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
939 page_cache_release(page
);
943 * There are a few paths in the higher layers of the kernel that directly
944 * set the page dirty bit without asking the filesystem if it is a
945 * good idea. This causes problems because we want to make sure COW
946 * properly happens and the data=ordered rules are followed.
948 * In our case any range that doesn't have the ORDERED bit set
949 * hasn't been properly setup for IO. We kick off an async process
950 * to fix it up. The async helper will wait for ordered extents, set
951 * the delalloc bit and make it safe to write the page.
953 int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
955 struct inode
*inode
= page
->mapping
->host
;
956 struct btrfs_writepage_fixup
*fixup
;
957 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
960 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
965 if (PageChecked(page
))
968 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
972 SetPageChecked(page
);
973 page_cache_get(page
);
974 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
976 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
980 /* as ordered data IO finishes, this gets called so we can finish
981 * an ordered extent if the range of bytes in the file it covers are
984 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
986 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
987 struct btrfs_trans_handle
*trans
;
988 struct btrfs_ordered_extent
*ordered_extent
;
989 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
990 struct btrfs_file_extent_item
*extent_item
;
991 struct btrfs_path
*path
= NULL
;
992 struct extent_buffer
*leaf
;
994 struct list_head list
;
995 struct btrfs_key ins
;
998 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1002 trans
= btrfs_join_transaction(root
, 1);
1004 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1005 BUG_ON(!ordered_extent
);
1006 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1009 path
= btrfs_alloc_path();
1012 lock_extent(io_tree
, ordered_extent
->file_offset
,
1013 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1016 INIT_LIST_HEAD(&list
);
1018 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1020 ret
= btrfs_drop_extents(trans
, root
, inode
,
1021 ordered_extent
->file_offset
,
1022 ordered_extent
->file_offset
+
1023 ordered_extent
->len
,
1024 ordered_extent
->file_offset
, &alloc_hint
);
1027 ins
.objectid
= inode
->i_ino
;
1028 ins
.offset
= ordered_extent
->file_offset
;
1029 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1030 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
1031 sizeof(*extent_item
));
1033 leaf
= path
->nodes
[0];
1034 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1035 struct btrfs_file_extent_item
);
1036 btrfs_set_file_extent_generation(leaf
, extent_item
, trans
->transid
);
1037 btrfs_set_file_extent_type(leaf
, extent_item
, BTRFS_FILE_EXTENT_REG
);
1038 btrfs_set_file_extent_disk_bytenr(leaf
, extent_item
,
1039 ordered_extent
->start
);
1040 btrfs_set_file_extent_disk_num_bytes(leaf
, extent_item
,
1041 ordered_extent
->disk_len
);
1042 btrfs_set_file_extent_offset(leaf
, extent_item
, 0);
1044 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1045 btrfs_set_file_extent_compression(leaf
, extent_item
, 1);
1047 btrfs_set_file_extent_compression(leaf
, extent_item
, 0);
1048 btrfs_set_file_extent_encryption(leaf
, extent_item
, 0);
1049 btrfs_set_file_extent_other_encoding(leaf
, extent_item
, 0);
1051 /* ram bytes = extent_num_bytes for now */
1052 btrfs_set_file_extent_num_bytes(leaf
, extent_item
,
1053 ordered_extent
->len
);
1054 btrfs_set_file_extent_ram_bytes(leaf
, extent_item
,
1055 ordered_extent
->len
);
1056 btrfs_mark_buffer_dirty(leaf
);
1058 btrfs_drop_extent_cache(inode
, ordered_extent
->file_offset
,
1059 ordered_extent
->file_offset
+
1060 ordered_extent
->len
- 1, 0);
1061 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1063 ins
.objectid
= ordered_extent
->start
;
1064 ins
.offset
= ordered_extent
->disk_len
;
1065 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1066 ret
= btrfs_alloc_reserved_extent(trans
, root
, leaf
->start
,
1067 root
->root_key
.objectid
,
1068 trans
->transid
, inode
->i_ino
, &ins
);
1070 btrfs_release_path(root
, path
);
1072 inode_add_bytes(inode
, ordered_extent
->len
);
1073 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1074 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1077 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1078 &ordered_extent
->list
);
1080 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1081 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1082 btrfs_update_inode(trans
, root
, inode
);
1083 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1084 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1087 btrfs_put_ordered_extent(ordered_extent
);
1088 /* once for the tree */
1089 btrfs_put_ordered_extent(ordered_extent
);
1091 btrfs_end_transaction(trans
, root
);
1093 btrfs_free_path(path
);
1097 int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1098 struct extent_state
*state
, int uptodate
)
1100 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1104 * When IO fails, either with EIO or csum verification fails, we
1105 * try other mirrors that might have a good copy of the data. This
1106 * io_failure_record is used to record state as we go through all the
1107 * mirrors. If another mirror has good data, the page is set up to date
1108 * and things continue. If a good mirror can't be found, the original
1109 * bio end_io callback is called to indicate things have failed.
1111 struct io_failure_record
{
1119 int btrfs_io_failed_hook(struct bio
*failed_bio
,
1120 struct page
*page
, u64 start
, u64 end
,
1121 struct extent_state
*state
)
1123 struct io_failure_record
*failrec
= NULL
;
1125 struct extent_map
*em
;
1126 struct inode
*inode
= page
->mapping
->host
;
1127 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1128 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1134 unsigned long bio_flags
= 0;
1136 ret
= get_state_private(failure_tree
, start
, &private);
1138 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1141 failrec
->start
= start
;
1142 failrec
->len
= end
- start
+ 1;
1143 failrec
->last_mirror
= 0;
1145 spin_lock(&em_tree
->lock
);
1146 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1147 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1148 free_extent_map(em
);
1151 spin_unlock(&em_tree
->lock
);
1153 if (!em
|| IS_ERR(em
)) {
1157 logical
= start
- em
->start
;
1158 logical
= em
->block_start
+ logical
;
1159 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
1160 bio_flags
= EXTENT_BIO_COMPRESSED
;
1161 failrec
->logical
= logical
;
1162 free_extent_map(em
);
1163 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1164 EXTENT_DIRTY
, GFP_NOFS
);
1165 set_state_private(failure_tree
, start
,
1166 (u64
)(unsigned long)failrec
);
1168 failrec
= (struct io_failure_record
*)(unsigned long)private;
1170 num_copies
= btrfs_num_copies(
1171 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1172 failrec
->logical
, failrec
->len
);
1173 failrec
->last_mirror
++;
1175 spin_lock_irq(&BTRFS_I(inode
)->io_tree
.lock
);
1176 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1179 if (state
&& state
->start
!= failrec
->start
)
1181 spin_unlock_irq(&BTRFS_I(inode
)->io_tree
.lock
);
1183 if (!state
|| failrec
->last_mirror
> num_copies
) {
1184 set_state_private(failure_tree
, failrec
->start
, 0);
1185 clear_extent_bits(failure_tree
, failrec
->start
,
1186 failrec
->start
+ failrec
->len
- 1,
1187 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1191 bio
= bio_alloc(GFP_NOFS
, 1);
1192 bio
->bi_private
= state
;
1193 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1194 bio
->bi_sector
= failrec
->logical
>> 9;
1195 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1197 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1198 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1203 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1204 failrec
->last_mirror
,
1210 * each time an IO finishes, we do a fast check in the IO failure tree
1211 * to see if we need to process or clean up an io_failure_record
1213 int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1216 u64 private_failure
;
1217 struct io_failure_record
*failure
;
1221 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1222 (u64
)-1, 1, EXTENT_DIRTY
)) {
1223 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1224 start
, &private_failure
);
1226 failure
= (struct io_failure_record
*)(unsigned long)
1228 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1230 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1232 failure
->start
+ failure
->len
- 1,
1233 EXTENT_DIRTY
| EXTENT_LOCKED
,
1242 * when reads are done, we need to check csums to verify the data is correct
1243 * if there's a match, we allow the bio to finish. If not, we go through
1244 * the io_failure_record routines to find good copies
1246 int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1247 struct extent_state
*state
)
1249 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1250 struct inode
*inode
= page
->mapping
->host
;
1251 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1253 u64
private = ~(u32
)0;
1255 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1257 unsigned long flags
;
1259 if (btrfs_test_opt(root
, NODATASUM
) ||
1260 btrfs_test_flag(inode
, NODATASUM
))
1262 if (state
&& state
->start
== start
) {
1263 private = state
->private;
1266 ret
= get_state_private(io_tree
, start
, &private);
1268 local_irq_save(flags
);
1269 kaddr
= kmap_atomic(page
, KM_IRQ0
);
1273 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1274 btrfs_csum_final(csum
, (char *)&csum
);
1275 if (csum
!= private) {
1278 kunmap_atomic(kaddr
, KM_IRQ0
);
1279 local_irq_restore(flags
);
1281 /* if the io failure tree for this inode is non-empty,
1282 * check to see if we've recovered from a failed IO
1284 btrfs_clean_io_failures(inode
, start
);
1288 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1289 page
->mapping
->host
->i_ino
, (unsigned long long)start
, csum
,
1291 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1292 flush_dcache_page(page
);
1293 kunmap_atomic(kaddr
, KM_IRQ0
);
1294 local_irq_restore(flags
);
1301 * This creates an orphan entry for the given inode in case something goes
1302 * wrong in the middle of an unlink/truncate.
1304 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1306 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1309 spin_lock(&root
->list_lock
);
1311 /* already on the orphan list, we're good */
1312 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1313 spin_unlock(&root
->list_lock
);
1317 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1319 spin_unlock(&root
->list_lock
);
1322 * insert an orphan item to track this unlinked/truncated file
1324 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1330 * We have done the truncate/delete so we can go ahead and remove the orphan
1331 * item for this particular inode.
1333 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1335 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1338 spin_lock(&root
->list_lock
);
1340 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1341 spin_unlock(&root
->list_lock
);
1345 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1347 spin_unlock(&root
->list_lock
);
1351 spin_unlock(&root
->list_lock
);
1353 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1359 * this cleans up any orphans that may be left on the list from the last use
1362 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1364 struct btrfs_path
*path
;
1365 struct extent_buffer
*leaf
;
1366 struct btrfs_item
*item
;
1367 struct btrfs_key key
, found_key
;
1368 struct btrfs_trans_handle
*trans
;
1369 struct inode
*inode
;
1370 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1372 /* don't do orphan cleanup if the fs is readonly. */
1373 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
1376 path
= btrfs_alloc_path();
1381 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1382 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1383 key
.offset
= (u64
)-1;
1387 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1389 printk(KERN_ERR
"Error searching slot for orphan: %d"
1395 * if ret == 0 means we found what we were searching for, which
1396 * is weird, but possible, so only screw with path if we didnt
1397 * find the key and see if we have stuff that matches
1400 if (path
->slots
[0] == 0)
1405 /* pull out the item */
1406 leaf
= path
->nodes
[0];
1407 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1408 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1410 /* make sure the item matches what we want */
1411 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1413 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1416 /* release the path since we're done with it */
1417 btrfs_release_path(root
, path
);
1420 * this is where we are basically btrfs_lookup, without the
1421 * crossing root thing. we store the inode number in the
1422 * offset of the orphan item.
1424 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
1425 found_key
.offset
, root
);
1429 if (inode
->i_state
& I_NEW
) {
1430 BTRFS_I(inode
)->root
= root
;
1432 /* have to set the location manually */
1433 BTRFS_I(inode
)->location
.objectid
= inode
->i_ino
;
1434 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
1435 BTRFS_I(inode
)->location
.offset
= 0;
1437 btrfs_read_locked_inode(inode
);
1438 unlock_new_inode(inode
);
1442 * add this inode to the orphan list so btrfs_orphan_del does
1443 * the proper thing when we hit it
1445 spin_lock(&root
->list_lock
);
1446 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1447 spin_unlock(&root
->list_lock
);
1450 * if this is a bad inode, means we actually succeeded in
1451 * removing the inode, but not the orphan record, which means
1452 * we need to manually delete the orphan since iput will just
1453 * do a destroy_inode
1455 if (is_bad_inode(inode
)) {
1456 trans
= btrfs_start_transaction(root
, 1);
1457 btrfs_orphan_del(trans
, inode
);
1458 btrfs_end_transaction(trans
, root
);
1463 /* if we have links, this was a truncate, lets do that */
1464 if (inode
->i_nlink
) {
1466 btrfs_truncate(inode
);
1471 /* this will do delete_inode and everything for us */
1476 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
1478 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
1480 btrfs_free_path(path
);
1484 * read an inode from the btree into the in-memory inode
1486 void btrfs_read_locked_inode(struct inode
*inode
)
1488 struct btrfs_path
*path
;
1489 struct extent_buffer
*leaf
;
1490 struct btrfs_inode_item
*inode_item
;
1491 struct btrfs_timespec
*tspec
;
1492 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1493 struct btrfs_key location
;
1494 u64 alloc_group_block
;
1498 path
= btrfs_alloc_path();
1500 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
1502 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
1506 leaf
= path
->nodes
[0];
1507 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1508 struct btrfs_inode_item
);
1510 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
1511 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
1512 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
1513 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
1514 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
1516 tspec
= btrfs_inode_atime(inode_item
);
1517 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1518 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1520 tspec
= btrfs_inode_mtime(inode_item
);
1521 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1522 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1524 tspec
= btrfs_inode_ctime(inode_item
);
1525 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1526 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1528 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
1529 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
1530 inode
->i_generation
= BTRFS_I(inode
)->generation
;
1532 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
1534 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1536 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
1537 BTRFS_I(inode
)->block_group
= btrfs_lookup_block_group(root
->fs_info
,
1539 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
1540 if (!BTRFS_I(inode
)->block_group
) {
1541 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
,
1543 BTRFS_BLOCK_GROUP_METADATA
, 0);
1545 btrfs_free_path(path
);
1548 switch (inode
->i_mode
& S_IFMT
) {
1550 inode
->i_mapping
->a_ops
= &btrfs_aops
;
1551 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
1552 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
1553 inode
->i_fop
= &btrfs_file_operations
;
1554 inode
->i_op
= &btrfs_file_inode_operations
;
1557 inode
->i_fop
= &btrfs_dir_file_operations
;
1558 if (root
== root
->fs_info
->tree_root
)
1559 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
1561 inode
->i_op
= &btrfs_dir_inode_operations
;
1564 inode
->i_op
= &btrfs_symlink_inode_operations
;
1565 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
1566 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
1569 init_special_inode(inode
, inode
->i_mode
, rdev
);
1575 btrfs_free_path(path
);
1576 make_bad_inode(inode
);
1580 * given a leaf and an inode, copy the inode fields into the leaf
1582 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
1583 struct extent_buffer
*leaf
,
1584 struct btrfs_inode_item
*item
,
1585 struct inode
*inode
)
1587 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
1588 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
1589 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
1590 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
1591 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
1593 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
1594 inode
->i_atime
.tv_sec
);
1595 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
1596 inode
->i_atime
.tv_nsec
);
1598 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
1599 inode
->i_mtime
.tv_sec
);
1600 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
1601 inode
->i_mtime
.tv_nsec
);
1603 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
1604 inode
->i_ctime
.tv_sec
);
1605 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
1606 inode
->i_ctime
.tv_nsec
);
1608 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
1609 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
1610 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
1611 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
1612 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
1613 btrfs_set_inode_block_group(leaf
, item
,
1614 BTRFS_I(inode
)->block_group
->key
.objectid
);
1618 * copy everything in the in-memory inode into the btree.
1620 int noinline
btrfs_update_inode(struct btrfs_trans_handle
*trans
,
1621 struct btrfs_root
*root
,
1622 struct inode
*inode
)
1624 struct btrfs_inode_item
*inode_item
;
1625 struct btrfs_path
*path
;
1626 struct extent_buffer
*leaf
;
1629 path
= btrfs_alloc_path();
1631 ret
= btrfs_lookup_inode(trans
, root
, path
,
1632 &BTRFS_I(inode
)->location
, 1);
1639 leaf
= path
->nodes
[0];
1640 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1641 struct btrfs_inode_item
);
1643 fill_inode_item(trans
, leaf
, inode_item
, inode
);
1644 btrfs_mark_buffer_dirty(leaf
);
1645 btrfs_set_inode_last_trans(trans
, inode
);
1648 btrfs_free_path(path
);
1654 * unlink helper that gets used here in inode.c and in the tree logging
1655 * recovery code. It remove a link in a directory with a given name, and
1656 * also drops the back refs in the inode to the directory
1658 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
1659 struct btrfs_root
*root
,
1660 struct inode
*dir
, struct inode
*inode
,
1661 const char *name
, int name_len
)
1663 struct btrfs_path
*path
;
1665 struct extent_buffer
*leaf
;
1666 struct btrfs_dir_item
*di
;
1667 struct btrfs_key key
;
1670 path
= btrfs_alloc_path();
1676 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
1677 name
, name_len
, -1);
1686 leaf
= path
->nodes
[0];
1687 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
1688 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
1691 btrfs_release_path(root
, path
);
1693 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
1695 dir
->i_ino
, &index
);
1697 printk("failed to delete reference to %.*s, "
1698 "inode %lu parent %lu\n", name_len
, name
,
1699 inode
->i_ino
, dir
->i_ino
);
1703 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
1704 index
, name
, name_len
, -1);
1713 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
1714 btrfs_release_path(root
, path
);
1716 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
1718 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
1720 BTRFS_I(dir
)->log_dirty_trans
= trans
->transid
;
1722 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
1726 btrfs_free_path(path
);
1730 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
1731 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
1732 btrfs_update_inode(trans
, root
, dir
);
1733 btrfs_drop_nlink(inode
);
1734 ret
= btrfs_update_inode(trans
, root
, inode
);
1735 dir
->i_sb
->s_dirt
= 1;
1740 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1742 struct btrfs_root
*root
;
1743 struct btrfs_trans_handle
*trans
;
1744 struct inode
*inode
= dentry
->d_inode
;
1746 unsigned long nr
= 0;
1748 root
= BTRFS_I(dir
)->root
;
1750 ret
= btrfs_check_free_space(root
, 1, 1);
1754 trans
= btrfs_start_transaction(root
, 1);
1756 btrfs_set_trans_block_group(trans
, dir
);
1757 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
1758 dentry
->d_name
.name
, dentry
->d_name
.len
);
1760 if (inode
->i_nlink
== 0)
1761 ret
= btrfs_orphan_add(trans
, inode
);
1763 nr
= trans
->blocks_used
;
1765 btrfs_end_transaction_throttle(trans
, root
);
1767 btrfs_btree_balance_dirty(root
, nr
);
1771 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1773 struct inode
*inode
= dentry
->d_inode
;
1776 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
1777 struct btrfs_trans_handle
*trans
;
1778 unsigned long nr
= 0;
1780 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
1784 ret
= btrfs_check_free_space(root
, 1, 1);
1788 trans
= btrfs_start_transaction(root
, 1);
1789 btrfs_set_trans_block_group(trans
, dir
);
1791 err
= btrfs_orphan_add(trans
, inode
);
1795 /* now the directory is empty */
1796 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
1797 dentry
->d_name
.name
, dentry
->d_name
.len
);
1799 btrfs_i_size_write(inode
, 0);
1803 nr
= trans
->blocks_used
;
1804 ret
= btrfs_end_transaction_throttle(trans
, root
);
1806 btrfs_btree_balance_dirty(root
, nr
);
1814 * when truncating bytes in a file, it is possible to avoid reading
1815 * the leaves that contain only checksum items. This can be the
1816 * majority of the IO required to delete a large file, but it must
1817 * be done carefully.
1819 * The keys in the level just above the leaves are checked to make sure
1820 * the lowest key in a given leaf is a csum key, and starts at an offset
1821 * after the new size.
1823 * Then the key for the next leaf is checked to make sure it also has
1824 * a checksum item for the same file. If it does, we know our target leaf
1825 * contains only checksum items, and it can be safely freed without reading
1828 * This is just an optimization targeted at large files. It may do
1829 * nothing. It will return 0 unless things went badly.
1831 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
1832 struct btrfs_root
*root
,
1833 struct btrfs_path
*path
,
1834 struct inode
*inode
, u64 new_size
)
1836 struct btrfs_key key
;
1839 struct btrfs_key found_key
;
1840 struct btrfs_key other_key
;
1841 struct btrfs_leaf_ref
*ref
;
1845 path
->lowest_level
= 1;
1846 key
.objectid
= inode
->i_ino
;
1847 key
.type
= BTRFS_CSUM_ITEM_KEY
;
1848 key
.offset
= new_size
;
1850 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1854 if (path
->nodes
[1] == NULL
) {
1859 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
1860 nritems
= btrfs_header_nritems(path
->nodes
[1]);
1865 if (path
->slots
[1] >= nritems
)
1868 /* did we find a key greater than anything we want to delete? */
1869 if (found_key
.objectid
> inode
->i_ino
||
1870 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
1873 /* we check the next key in the node to make sure the leave contains
1874 * only checksum items. This comparison doesn't work if our
1875 * leaf is the last one in the node
1877 if (path
->slots
[1] + 1 >= nritems
) {
1879 /* search forward from the last key in the node, this
1880 * will bring us into the next node in the tree
1882 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
1884 /* unlikely, but we inc below, so check to be safe */
1885 if (found_key
.offset
== (u64
)-1)
1888 /* search_forward needs a path with locks held, do the
1889 * search again for the original key. It is possible
1890 * this will race with a balance and return a path that
1891 * we could modify, but this drop is just an optimization
1892 * and is allowed to miss some leaves.
1894 btrfs_release_path(root
, path
);
1897 /* setup a max key for search_forward */
1898 other_key
.offset
= (u64
)-1;
1899 other_key
.type
= key
.type
;
1900 other_key
.objectid
= key
.objectid
;
1902 path
->keep_locks
= 1;
1903 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
1905 path
->keep_locks
= 0;
1906 if (ret
|| found_key
.objectid
!= key
.objectid
||
1907 found_key
.type
!= key
.type
) {
1912 key
.offset
= found_key
.offset
;
1913 btrfs_release_path(root
, path
);
1918 /* we know there's one more slot after us in the tree,
1919 * read that key so we can verify it is also a checksum item
1921 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
1923 if (found_key
.objectid
< inode
->i_ino
)
1926 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
1930 * if the key for the next leaf isn't a csum key from this objectid,
1931 * we can't be sure there aren't good items inside this leaf.
1934 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
1937 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
1938 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
1940 * it is safe to delete this leaf, it contains only
1941 * csum items from this inode at an offset >= new_size
1943 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
1946 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
1947 ref
= btrfs_alloc_leaf_ref(root
, 0);
1949 ref
->root_gen
= root
->root_key
.offset
;
1950 ref
->bytenr
= leaf_start
;
1952 ref
->generation
= leaf_gen
;
1955 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
1957 btrfs_free_leaf_ref(root
, ref
);
1963 btrfs_release_path(root
, path
);
1965 if (other_key
.objectid
== inode
->i_ino
&&
1966 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
1967 key
.offset
= other_key
.offset
;
1973 /* fixup any changes we've made to the path */
1974 path
->lowest_level
= 0;
1975 path
->keep_locks
= 0;
1976 btrfs_release_path(root
, path
);
1981 * this can truncate away extent items, csum items and directory items.
1982 * It starts at a high offset and removes keys until it can't find
1983 * any higher than new_size
1985 * csum items that cross the new i_size are truncated to the new size
1988 * min_type is the minimum key type to truncate down to. If set to 0, this
1989 * will kill all the items on this inode, including the INODE_ITEM_KEY.
1991 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
1992 struct btrfs_root
*root
,
1993 struct inode
*inode
,
1994 u64 new_size
, u32 min_type
)
1997 struct btrfs_path
*path
;
1998 struct btrfs_key key
;
1999 struct btrfs_key found_key
;
2001 struct extent_buffer
*leaf
;
2002 struct btrfs_file_extent_item
*fi
;
2003 u64 extent_start
= 0;
2004 u64 extent_num_bytes
= 0;
2010 int pending_del_nr
= 0;
2011 int pending_del_slot
= 0;
2012 int extent_type
= -1;
2013 u64 mask
= root
->sectorsize
- 1;
2016 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2017 path
= btrfs_alloc_path();
2021 /* FIXME, add redo link to tree so we don't leak on crash */
2022 key
.objectid
= inode
->i_ino
;
2023 key
.offset
= (u64
)-1;
2026 btrfs_init_path(path
);
2028 ret
= drop_csum_leaves(trans
, root
, path
, inode
, new_size
);
2032 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2037 /* there are no items in the tree for us to truncate, we're
2040 if (path
->slots
[0] == 0) {
2049 leaf
= path
->nodes
[0];
2050 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2051 found_type
= btrfs_key_type(&found_key
);
2053 if (found_key
.objectid
!= inode
->i_ino
)
2056 if (found_type
< min_type
)
2059 item_end
= found_key
.offset
;
2060 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2061 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2062 struct btrfs_file_extent_item
);
2063 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2064 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2066 btrfs_file_extent_num_bytes(leaf
, fi
);
2067 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2068 item_end
+= btrfs_file_extent_inline_len(leaf
,
2073 if (found_type
== BTRFS_CSUM_ITEM_KEY
) {
2074 ret
= btrfs_csum_truncate(trans
, root
, path
,
2078 if (item_end
< new_size
) {
2079 if (found_type
== BTRFS_DIR_ITEM_KEY
) {
2080 found_type
= BTRFS_INODE_ITEM_KEY
;
2081 } else if (found_type
== BTRFS_EXTENT_ITEM_KEY
) {
2082 found_type
= BTRFS_CSUM_ITEM_KEY
;
2083 } else if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2084 found_type
= BTRFS_XATTR_ITEM_KEY
;
2085 } else if (found_type
== BTRFS_XATTR_ITEM_KEY
) {
2086 found_type
= BTRFS_INODE_REF_KEY
;
2087 } else if (found_type
) {
2092 btrfs_set_key_type(&key
, found_type
);
2095 if (found_key
.offset
>= new_size
)
2101 /* FIXME, shrink the extent if the ref count is only 1 */
2102 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2105 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2107 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2109 u64 orig_num_bytes
=
2110 btrfs_file_extent_num_bytes(leaf
, fi
);
2111 extent_num_bytes
= new_size
-
2112 found_key
.offset
+ root
->sectorsize
- 1;
2113 extent_num_bytes
= extent_num_bytes
&
2114 ~((u64
)root
->sectorsize
- 1);
2115 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2117 num_dec
= (orig_num_bytes
-
2119 if (root
->ref_cows
&& extent_start
!= 0)
2120 inode_sub_bytes(inode
, num_dec
);
2121 btrfs_mark_buffer_dirty(leaf
);
2124 btrfs_file_extent_disk_num_bytes(leaf
,
2126 /* FIXME blocksize != 4096 */
2127 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2128 if (extent_start
!= 0) {
2131 inode_sub_bytes(inode
, num_dec
);
2133 root_gen
= btrfs_header_generation(leaf
);
2134 root_owner
= btrfs_header_owner(leaf
);
2136 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2138 * we can't truncate inline items that have had
2142 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2143 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2144 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2145 u32 size
= new_size
- found_key
.offset
;
2147 if (root
->ref_cows
) {
2148 inode_sub_bytes(inode
, item_end
+ 1 -
2152 btrfs_file_extent_calc_inline_size(size
);
2153 ret
= btrfs_truncate_item(trans
, root
, path
,
2156 } else if (root
->ref_cows
) {
2157 inode_sub_bytes(inode
, item_end
+ 1 -
2163 if (!pending_del_nr
) {
2164 /* no pending yet, add ourselves */
2165 pending_del_slot
= path
->slots
[0];
2167 } else if (pending_del_nr
&&
2168 path
->slots
[0] + 1 == pending_del_slot
) {
2169 /* hop on the pending chunk */
2171 pending_del_slot
= path
->slots
[0];
2173 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path
->slots
[0], pending_del_nr
, pending_del_slot
);
2179 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2181 leaf
->start
, root_owner
,
2182 root_gen
, inode
->i_ino
, 0);
2186 if (path
->slots
[0] == 0) {
2189 btrfs_release_path(root
, path
);
2194 if (pending_del_nr
&&
2195 path
->slots
[0] + 1 != pending_del_slot
) {
2196 struct btrfs_key debug
;
2198 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2200 ret
= btrfs_del_items(trans
, root
, path
,
2205 btrfs_release_path(root
, path
);
2211 if (pending_del_nr
) {
2212 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2215 btrfs_free_path(path
);
2216 inode
->i_sb
->s_dirt
= 1;
2221 * taken from block_truncate_page, but does cow as it zeros out
2222 * any bytes left in the last page in the file.
2224 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2226 struct inode
*inode
= mapping
->host
;
2227 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2228 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2229 struct btrfs_ordered_extent
*ordered
;
2231 u32 blocksize
= root
->sectorsize
;
2232 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2233 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2239 if ((offset
& (blocksize
- 1)) == 0)
2244 page
= grab_cache_page(mapping
, index
);
2248 page_start
= page_offset(page
);
2249 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2251 if (!PageUptodate(page
)) {
2252 ret
= btrfs_readpage(NULL
, page
);
2254 if (page
->mapping
!= mapping
) {
2256 page_cache_release(page
);
2259 if (!PageUptodate(page
)) {
2264 wait_on_page_writeback(page
);
2266 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2267 set_page_extent_mapped(page
);
2269 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2271 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2273 page_cache_release(page
);
2274 btrfs_start_ordered_extent(inode
, ordered
, 1);
2275 btrfs_put_ordered_extent(ordered
);
2279 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2281 if (offset
!= PAGE_CACHE_SIZE
) {
2283 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2284 flush_dcache_page(page
);
2287 ClearPageChecked(page
);
2288 set_page_dirty(page
);
2289 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2293 page_cache_release(page
);
2298 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2300 struct inode
*inode
= dentry
->d_inode
;
2303 err
= inode_change_ok(inode
, attr
);
2307 if (S_ISREG(inode
->i_mode
) &&
2308 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
> inode
->i_size
) {
2309 struct btrfs_trans_handle
*trans
;
2310 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2311 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2313 u64 mask
= root
->sectorsize
- 1;
2314 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2315 u64 block_end
= (attr
->ia_size
+ mask
) & ~mask
;
2319 if (attr
->ia_size
<= hole_start
)
2322 err
= btrfs_check_free_space(root
, 1, 0);
2326 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2328 hole_size
= block_end
- hole_start
;
2330 struct btrfs_ordered_extent
*ordered
;
2331 btrfs_wait_ordered_range(inode
, hole_start
, hole_size
);
2333 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2334 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2336 unlock_extent(io_tree
, hole_start
,
2337 block_end
- 1, GFP_NOFS
);
2338 btrfs_put_ordered_extent(ordered
);
2344 trans
= btrfs_start_transaction(root
, 1);
2345 btrfs_set_trans_block_group(trans
, inode
);
2346 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
2347 err
= btrfs_drop_extents(trans
, root
, inode
,
2348 hole_start
, block_end
, hole_start
,
2351 if (alloc_hint
!= EXTENT_MAP_INLINE
) {
2352 err
= btrfs_insert_file_extent(trans
, root
,
2355 hole_size
, 0, hole_size
,
2357 btrfs_drop_extent_cache(inode
, hole_start
,
2359 btrfs_check_file(root
, inode
);
2361 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
2362 btrfs_end_transaction(trans
, root
);
2363 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2368 err
= inode_setattr(inode
, attr
);
2370 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2371 err
= btrfs_acl_chmod(inode
);
2376 void btrfs_delete_inode(struct inode
*inode
)
2378 struct btrfs_trans_handle
*trans
;
2379 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2383 truncate_inode_pages(&inode
->i_data
, 0);
2384 if (is_bad_inode(inode
)) {
2385 btrfs_orphan_del(NULL
, inode
);
2388 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
2390 btrfs_i_size_write(inode
, 0);
2391 trans
= btrfs_start_transaction(root
, 1);
2393 btrfs_set_trans_block_group(trans
, inode
);
2394 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
2396 btrfs_orphan_del(NULL
, inode
);
2397 goto no_delete_lock
;
2400 btrfs_orphan_del(trans
, inode
);
2402 nr
= trans
->blocks_used
;
2405 btrfs_end_transaction(trans
, root
);
2406 btrfs_btree_balance_dirty(root
, nr
);
2410 nr
= trans
->blocks_used
;
2411 btrfs_end_transaction(trans
, root
);
2412 btrfs_btree_balance_dirty(root
, nr
);
2418 * this returns the key found in the dir entry in the location pointer.
2419 * If no dir entries were found, location->objectid is 0.
2421 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
2422 struct btrfs_key
*location
)
2424 const char *name
= dentry
->d_name
.name
;
2425 int namelen
= dentry
->d_name
.len
;
2426 struct btrfs_dir_item
*di
;
2427 struct btrfs_path
*path
;
2428 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2431 path
= btrfs_alloc_path();
2434 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
2438 if (!di
|| IS_ERR(di
)) {
2441 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
2443 btrfs_free_path(path
);
2446 location
->objectid
= 0;
2451 * when we hit a tree root in a directory, the btrfs part of the inode
2452 * needs to be changed to reflect the root directory of the tree root. This
2453 * is kind of like crossing a mount point.
2455 static int fixup_tree_root_location(struct btrfs_root
*root
,
2456 struct btrfs_key
*location
,
2457 struct btrfs_root
**sub_root
,
2458 struct dentry
*dentry
)
2460 struct btrfs_root_item
*ri
;
2462 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
2464 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
2467 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
2468 dentry
->d_name
.name
,
2469 dentry
->d_name
.len
);
2470 if (IS_ERR(*sub_root
))
2471 return PTR_ERR(*sub_root
);
2473 ri
= &(*sub_root
)->root_item
;
2474 location
->objectid
= btrfs_root_dirid(ri
);
2475 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2476 location
->offset
= 0;
2481 static noinline
void init_btrfs_i(struct inode
*inode
)
2483 struct btrfs_inode
*bi
= BTRFS_I(inode
);
2486 bi
->i_default_acl
= NULL
;
2490 bi
->logged_trans
= 0;
2491 bi
->delalloc_bytes
= 0;
2492 bi
->disk_i_size
= 0;
2494 bi
->index_cnt
= (u64
)-1;
2495 bi
->log_dirty_trans
= 0;
2496 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
2497 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
2498 inode
->i_mapping
, GFP_NOFS
);
2499 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
2500 inode
->i_mapping
, GFP_NOFS
);
2501 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
2502 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
2503 mutex_init(&BTRFS_I(inode
)->csum_mutex
);
2504 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
2505 mutex_init(&BTRFS_I(inode
)->log_mutex
);
2508 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
2510 struct btrfs_iget_args
*args
= p
;
2511 inode
->i_ino
= args
->ino
;
2512 init_btrfs_i(inode
);
2513 BTRFS_I(inode
)->root
= args
->root
;
2517 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
2519 struct btrfs_iget_args
*args
= opaque
;
2520 return (args
->ino
== inode
->i_ino
&&
2521 args
->root
== BTRFS_I(inode
)->root
);
2524 struct inode
*btrfs_ilookup(struct super_block
*s
, u64 objectid
,
2525 struct btrfs_root
*root
, int wait
)
2527 struct inode
*inode
;
2528 struct btrfs_iget_args args
;
2529 args
.ino
= objectid
;
2533 inode
= ilookup5(s
, objectid
, btrfs_find_actor
,
2536 inode
= ilookup5_nowait(s
, objectid
, btrfs_find_actor
,
2542 struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 objectid
,
2543 struct btrfs_root
*root
)
2545 struct inode
*inode
;
2546 struct btrfs_iget_args args
;
2547 args
.ino
= objectid
;
2550 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
2551 btrfs_init_locked_inode
,
2556 /* Get an inode object given its location and corresponding root.
2557 * Returns in *is_new if the inode was read from disk
2559 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
2560 struct btrfs_root
*root
, int *is_new
)
2562 struct inode
*inode
;
2564 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
2566 return ERR_PTR(-EACCES
);
2568 if (inode
->i_state
& I_NEW
) {
2569 BTRFS_I(inode
)->root
= root
;
2570 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
2571 btrfs_read_locked_inode(inode
);
2572 unlock_new_inode(inode
);
2583 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
2584 struct nameidata
*nd
)
2586 struct inode
* inode
;
2587 struct btrfs_inode
*bi
= BTRFS_I(dir
);
2588 struct btrfs_root
*root
= bi
->root
;
2589 struct btrfs_root
*sub_root
= root
;
2590 struct btrfs_key location
;
2591 int ret
, new, do_orphan
= 0;
2593 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
2594 return ERR_PTR(-ENAMETOOLONG
);
2596 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
2599 return ERR_PTR(ret
);
2602 if (location
.objectid
) {
2603 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
2606 return ERR_PTR(ret
);
2608 return ERR_PTR(-ENOENT
);
2609 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, &new);
2611 return ERR_CAST(inode
);
2613 /* the inode and parent dir are two different roots */
2614 if (new && root
!= sub_root
) {
2616 sub_root
->inode
= inode
;
2621 if (unlikely(do_orphan
))
2622 btrfs_orphan_cleanup(sub_root
);
2624 return d_splice_alias(inode
, dentry
);
2627 static unsigned char btrfs_filetype_table
[] = {
2628 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
2631 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
2634 struct inode
*inode
= filp
->f_dentry
->d_inode
;
2635 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2636 struct btrfs_item
*item
;
2637 struct btrfs_dir_item
*di
;
2638 struct btrfs_key key
;
2639 struct btrfs_key found_key
;
2640 struct btrfs_path
*path
;
2643 struct extent_buffer
*leaf
;
2646 unsigned char d_type
;
2651 int key_type
= BTRFS_DIR_INDEX_KEY
;
2656 /* FIXME, use a real flag for deciding about the key type */
2657 if (root
->fs_info
->tree_root
== root
)
2658 key_type
= BTRFS_DIR_ITEM_KEY
;
2660 /* special case for "." */
2661 if (filp
->f_pos
== 0) {
2662 over
= filldir(dirent
, ".", 1,
2669 /* special case for .., just use the back ref */
2670 if (filp
->f_pos
== 1) {
2671 u64 pino
= parent_ino(filp
->f_path
.dentry
);
2672 over
= filldir(dirent
, "..", 2,
2679 path
= btrfs_alloc_path();
2682 btrfs_set_key_type(&key
, key_type
);
2683 key
.offset
= filp
->f_pos
;
2684 key
.objectid
= inode
->i_ino
;
2686 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2692 leaf
= path
->nodes
[0];
2693 nritems
= btrfs_header_nritems(leaf
);
2694 slot
= path
->slots
[0];
2695 if (advance
|| slot
>= nritems
) {
2696 if (slot
>= nritems
- 1) {
2697 ret
= btrfs_next_leaf(root
, path
);
2700 leaf
= path
->nodes
[0];
2701 nritems
= btrfs_header_nritems(leaf
);
2702 slot
= path
->slots
[0];
2709 item
= btrfs_item_nr(leaf
, slot
);
2710 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2712 if (found_key
.objectid
!= key
.objectid
)
2714 if (btrfs_key_type(&found_key
) != key_type
)
2716 if (found_key
.offset
< filp
->f_pos
)
2719 filp
->f_pos
= found_key
.offset
;
2721 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
2723 di_total
= btrfs_item_size(leaf
, item
);
2725 while (di_cur
< di_total
) {
2726 struct btrfs_key location
;
2728 name_len
= btrfs_dir_name_len(leaf
, di
);
2729 if (name_len
<= sizeof(tmp_name
)) {
2730 name_ptr
= tmp_name
;
2732 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
2738 read_extent_buffer(leaf
, name_ptr
,
2739 (unsigned long)(di
+ 1), name_len
);
2741 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
2742 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
2743 over
= filldir(dirent
, name_ptr
, name_len
,
2744 found_key
.offset
, location
.objectid
,
2747 if (name_ptr
!= tmp_name
)
2753 di_len
= btrfs_dir_name_len(leaf
, di
) +
2754 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
2756 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
2760 /* Reached end of directory/root. Bump pos past the last item. */
2761 if (key_type
== BTRFS_DIR_INDEX_KEY
)
2762 filp
->f_pos
= INT_LIMIT(typeof(filp
->f_pos
));
2768 btrfs_free_path(path
);
2772 int btrfs_write_inode(struct inode
*inode
, int wait
)
2774 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2775 struct btrfs_trans_handle
*trans
;
2778 if (root
->fs_info
->closing
> 1)
2782 trans
= btrfs_join_transaction(root
, 1);
2783 btrfs_set_trans_block_group(trans
, inode
);
2784 ret
= btrfs_commit_transaction(trans
, root
);
2790 * This is somewhat expensive, updating the tree every time the
2791 * inode changes. But, it is most likely to find the inode in cache.
2792 * FIXME, needs more benchmarking...there are no reasons other than performance
2793 * to keep or drop this code.
2795 void btrfs_dirty_inode(struct inode
*inode
)
2797 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2798 struct btrfs_trans_handle
*trans
;
2800 trans
= btrfs_join_transaction(root
, 1);
2801 btrfs_set_trans_block_group(trans
, inode
);
2802 btrfs_update_inode(trans
, root
, inode
);
2803 btrfs_end_transaction(trans
, root
);
2807 * find the highest existing sequence number in a directory
2808 * and then set the in-memory index_cnt variable to reflect
2809 * free sequence numbers
2811 static int btrfs_set_inode_index_count(struct inode
*inode
)
2813 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2814 struct btrfs_key key
, found_key
;
2815 struct btrfs_path
*path
;
2816 struct extent_buffer
*leaf
;
2819 key
.objectid
= inode
->i_ino
;
2820 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
2821 key
.offset
= (u64
)-1;
2823 path
= btrfs_alloc_path();
2827 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2830 /* FIXME: we should be able to handle this */
2836 * MAGIC NUMBER EXPLANATION:
2837 * since we search a directory based on f_pos we have to start at 2
2838 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2839 * else has to start at 2
2841 if (path
->slots
[0] == 0) {
2842 BTRFS_I(inode
)->index_cnt
= 2;
2848 leaf
= path
->nodes
[0];
2849 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2851 if (found_key
.objectid
!= inode
->i_ino
||
2852 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
2853 BTRFS_I(inode
)->index_cnt
= 2;
2857 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
2859 btrfs_free_path(path
);
2864 * helper to find a free sequence number in a given directory. This current
2865 * code is very simple, later versions will do smarter things in the btree
2867 static int btrfs_set_inode_index(struct inode
*dir
, struct inode
*inode
,
2872 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
2873 ret
= btrfs_set_inode_index_count(dir
);
2879 *index
= BTRFS_I(dir
)->index_cnt
;
2880 BTRFS_I(dir
)->index_cnt
++;
2885 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
2886 struct btrfs_root
*root
,
2888 const char *name
, int name_len
,
2891 struct btrfs_block_group_cache
*group
,
2892 int mode
, u64
*index
)
2894 struct inode
*inode
;
2895 struct btrfs_inode_item
*inode_item
;
2896 struct btrfs_block_group_cache
*new_inode_group
;
2897 struct btrfs_key
*location
;
2898 struct btrfs_path
*path
;
2899 struct btrfs_inode_ref
*ref
;
2900 struct btrfs_key key
[2];
2906 path
= btrfs_alloc_path();
2909 inode
= new_inode(root
->fs_info
->sb
);
2911 return ERR_PTR(-ENOMEM
);
2914 ret
= btrfs_set_inode_index(dir
, inode
, index
);
2916 return ERR_PTR(ret
);
2919 * index_cnt is ignored for everything but a dir,
2920 * btrfs_get_inode_index_count has an explanation for the magic
2923 init_btrfs_i(inode
);
2924 BTRFS_I(inode
)->index_cnt
= 2;
2925 BTRFS_I(inode
)->root
= root
;
2926 BTRFS_I(inode
)->generation
= trans
->transid
;
2932 new_inode_group
= btrfs_find_block_group(root
, group
, 0,
2933 BTRFS_BLOCK_GROUP_METADATA
, owner
);
2934 if (!new_inode_group
) {
2935 printk("find_block group failed\n");
2936 new_inode_group
= group
;
2938 BTRFS_I(inode
)->block_group
= new_inode_group
;
2940 key
[0].objectid
= objectid
;
2941 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
2944 key
[1].objectid
= objectid
;
2945 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
2946 key
[1].offset
= ref_objectid
;
2948 sizes
[0] = sizeof(struct btrfs_inode_item
);
2949 sizes
[1] = name_len
+ sizeof(*ref
);
2951 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
2955 if (objectid
> root
->highest_inode
)
2956 root
->highest_inode
= objectid
;
2958 inode
->i_uid
= current
->fsuid
;
2959 inode
->i_gid
= current
->fsgid
;
2960 inode
->i_mode
= mode
;
2961 inode
->i_ino
= objectid
;
2962 inode_set_bytes(inode
, 0);
2963 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
2964 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2965 struct btrfs_inode_item
);
2966 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
2968 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
2969 struct btrfs_inode_ref
);
2970 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
2971 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
2972 ptr
= (unsigned long)(ref
+ 1);
2973 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
2975 btrfs_mark_buffer_dirty(path
->nodes
[0]);
2976 btrfs_free_path(path
);
2978 location
= &BTRFS_I(inode
)->location
;
2979 location
->objectid
= objectid
;
2980 location
->offset
= 0;
2981 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2983 insert_inode_hash(inode
);
2987 BTRFS_I(dir
)->index_cnt
--;
2988 btrfs_free_path(path
);
2989 return ERR_PTR(ret
);
2992 static inline u8
btrfs_inode_type(struct inode
*inode
)
2994 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
2998 * utility function to add 'inode' into 'parent_inode' with
2999 * a give name and a given sequence number.
3000 * if 'add_backref' is true, also insert a backref from the
3001 * inode to the parent directory.
3003 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3004 struct inode
*parent_inode
, struct inode
*inode
,
3005 const char *name
, int name_len
, int add_backref
, u64 index
)
3008 struct btrfs_key key
;
3009 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3011 key
.objectid
= inode
->i_ino
;
3012 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3015 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3016 parent_inode
->i_ino
,
3017 &key
, btrfs_inode_type(inode
),
3021 ret
= btrfs_insert_inode_ref(trans
, root
,
3024 parent_inode
->i_ino
,
3027 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3029 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3030 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3035 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3036 struct dentry
*dentry
, struct inode
*inode
,
3037 int backref
, u64 index
)
3039 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3040 inode
, dentry
->d_name
.name
,
3041 dentry
->d_name
.len
, backref
, index
);
3043 d_instantiate(dentry
, inode
);
3051 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3052 int mode
, dev_t rdev
)
3054 struct btrfs_trans_handle
*trans
;
3055 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3056 struct inode
*inode
= NULL
;
3060 unsigned long nr
= 0;
3063 if (!new_valid_dev(rdev
))
3066 err
= btrfs_check_free_space(root
, 1, 0);
3070 trans
= btrfs_start_transaction(root
, 1);
3071 btrfs_set_trans_block_group(trans
, dir
);
3073 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3079 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3081 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3082 BTRFS_I(dir
)->block_group
, mode
, &index
);
3083 err
= PTR_ERR(inode
);
3087 err
= btrfs_init_acl(inode
, dir
);
3093 btrfs_set_trans_block_group(trans
, inode
);
3094 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3098 inode
->i_op
= &btrfs_special_inode_operations
;
3099 init_special_inode(inode
, inode
->i_mode
, rdev
);
3100 btrfs_update_inode(trans
, root
, inode
);
3102 dir
->i_sb
->s_dirt
= 1;
3103 btrfs_update_inode_block_group(trans
, inode
);
3104 btrfs_update_inode_block_group(trans
, dir
);
3106 nr
= trans
->blocks_used
;
3107 btrfs_end_transaction_throttle(trans
, root
);
3110 inode_dec_link_count(inode
);
3113 btrfs_btree_balance_dirty(root
, nr
);
3117 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3118 int mode
, struct nameidata
*nd
)
3120 struct btrfs_trans_handle
*trans
;
3121 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3122 struct inode
*inode
= NULL
;
3125 unsigned long nr
= 0;
3129 err
= btrfs_check_free_space(root
, 1, 0);
3132 trans
= btrfs_start_transaction(root
, 1);
3133 btrfs_set_trans_block_group(trans
, dir
);
3135 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3141 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3143 dentry
->d_parent
->d_inode
->i_ino
,
3144 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3146 err
= PTR_ERR(inode
);
3150 err
= btrfs_init_acl(inode
, dir
);
3156 btrfs_set_trans_block_group(trans
, inode
);
3157 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3161 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3162 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3163 inode
->i_fop
= &btrfs_file_operations
;
3164 inode
->i_op
= &btrfs_file_inode_operations
;
3165 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3167 dir
->i_sb
->s_dirt
= 1;
3168 btrfs_update_inode_block_group(trans
, inode
);
3169 btrfs_update_inode_block_group(trans
, dir
);
3171 nr
= trans
->blocks_used
;
3172 btrfs_end_transaction_throttle(trans
, root
);
3175 inode_dec_link_count(inode
);
3178 btrfs_btree_balance_dirty(root
, nr
);
3182 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3183 struct dentry
*dentry
)
3185 struct btrfs_trans_handle
*trans
;
3186 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3187 struct inode
*inode
= old_dentry
->d_inode
;
3189 unsigned long nr
= 0;
3193 if (inode
->i_nlink
== 0)
3196 btrfs_inc_nlink(inode
);
3197 err
= btrfs_check_free_space(root
, 1, 0);
3200 err
= btrfs_set_inode_index(dir
, inode
, &index
);
3204 trans
= btrfs_start_transaction(root
, 1);
3206 btrfs_set_trans_block_group(trans
, dir
);
3207 atomic_inc(&inode
->i_count
);
3209 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3214 dir
->i_sb
->s_dirt
= 1;
3215 btrfs_update_inode_block_group(trans
, dir
);
3216 err
= btrfs_update_inode(trans
, root
, inode
);
3221 nr
= trans
->blocks_used
;
3222 btrfs_end_transaction_throttle(trans
, root
);
3225 inode_dec_link_count(inode
);
3228 btrfs_btree_balance_dirty(root
, nr
);
3232 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3234 struct inode
*inode
= NULL
;
3235 struct btrfs_trans_handle
*trans
;
3236 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3238 int drop_on_err
= 0;
3241 unsigned long nr
= 1;
3243 err
= btrfs_check_free_space(root
, 1, 0);
3247 trans
= btrfs_start_transaction(root
, 1);
3248 btrfs_set_trans_block_group(trans
, dir
);
3250 if (IS_ERR(trans
)) {
3251 err
= PTR_ERR(trans
);
3255 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3261 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3263 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3264 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3266 if (IS_ERR(inode
)) {
3267 err
= PTR_ERR(inode
);
3273 err
= btrfs_init_acl(inode
, dir
);
3277 inode
->i_op
= &btrfs_dir_inode_operations
;
3278 inode
->i_fop
= &btrfs_dir_file_operations
;
3279 btrfs_set_trans_block_group(trans
, inode
);
3281 btrfs_i_size_write(inode
, 0);
3282 err
= btrfs_update_inode(trans
, root
, inode
);
3286 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3287 inode
, dentry
->d_name
.name
,
3288 dentry
->d_name
.len
, 0, index
);
3292 d_instantiate(dentry
, inode
);
3294 dir
->i_sb
->s_dirt
= 1;
3295 btrfs_update_inode_block_group(trans
, inode
);
3296 btrfs_update_inode_block_group(trans
, dir
);
3299 nr
= trans
->blocks_used
;
3300 btrfs_end_transaction_throttle(trans
, root
);
3305 btrfs_btree_balance_dirty(root
, nr
);
3309 /* helper for btfs_get_extent. Given an existing extent in the tree,
3310 * and an extent that you want to insert, deal with overlap and insert
3311 * the new extent into the tree.
3313 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3314 struct extent_map
*existing
,
3315 struct extent_map
*em
,
3316 u64 map_start
, u64 map_len
)
3320 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3321 start_diff
= map_start
- em
->start
;
3322 em
->start
= map_start
;
3324 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3325 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3326 em
->block_start
+= start_diff
;
3327 em
->block_len
-= start_diff
;
3329 return add_extent_mapping(em_tree
, em
);
3332 static noinline
int uncompress_inline(struct btrfs_path
*path
,
3333 struct inode
*inode
, struct page
*page
,
3334 size_t pg_offset
, u64 extent_offset
,
3335 struct btrfs_file_extent_item
*item
)
3338 struct extent_buffer
*leaf
= path
->nodes
[0];
3341 unsigned long inline_size
;
3344 WARN_ON(pg_offset
!= 0);
3345 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
3346 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
3347 btrfs_item_nr(leaf
, path
->slots
[0]));
3348 tmp
= kmalloc(inline_size
, GFP_NOFS
);
3349 ptr
= btrfs_file_extent_inline_start(item
);
3351 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
3353 max_size
= min(PAGE_CACHE_SIZE
, max_size
);
3354 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
3355 inline_size
, max_size
);
3357 char *kaddr
= kmap_atomic(page
, KM_USER0
);
3358 unsigned long copy_size
= min_t(u64
,
3359 PAGE_CACHE_SIZE
- pg_offset
,
3360 max_size
- extent_offset
);
3361 memset(kaddr
+ pg_offset
, 0, copy_size
);
3362 kunmap_atomic(kaddr
, KM_USER0
);
3369 * a bit scary, this does extent mapping from logical file offset to the disk.
3370 * the ugly parts come from merging extents from the disk with the
3371 * in-ram representation. This gets more complex because of the data=ordered code,
3372 * where the in-ram extents might be locked pending data=ordered completion.
3374 * This also copies inline extents directly into the page.
3376 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
3377 size_t pg_offset
, u64 start
, u64 len
,
3383 u64 extent_start
= 0;
3385 u64 objectid
= inode
->i_ino
;
3387 struct btrfs_path
*path
= NULL
;
3388 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3389 struct btrfs_file_extent_item
*item
;
3390 struct extent_buffer
*leaf
;
3391 struct btrfs_key found_key
;
3392 struct extent_map
*em
= NULL
;
3393 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3394 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3395 struct btrfs_trans_handle
*trans
= NULL
;
3399 spin_lock(&em_tree
->lock
);
3400 em
= lookup_extent_mapping(em_tree
, start
, len
);
3402 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3403 spin_unlock(&em_tree
->lock
);
3406 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
3407 free_extent_map(em
);
3408 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
3409 free_extent_map(em
);
3413 em
= alloc_extent_map(GFP_NOFS
);
3418 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3419 em
->start
= EXTENT_MAP_HOLE
;
3421 em
->block_len
= (u64
)-1;
3424 path
= btrfs_alloc_path();
3428 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3429 objectid
, start
, trans
!= NULL
);
3436 if (path
->slots
[0] == 0)
3441 leaf
= path
->nodes
[0];
3442 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3443 struct btrfs_file_extent_item
);
3444 /* are we inside the extent that was found? */
3445 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3446 found_type
= btrfs_key_type(&found_key
);
3447 if (found_key
.objectid
!= objectid
||
3448 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3452 found_type
= btrfs_file_extent_type(leaf
, item
);
3453 extent_start
= found_key
.offset
;
3454 compressed
= btrfs_file_extent_compression(leaf
, item
);
3455 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
3456 extent_end
= extent_start
+
3457 btrfs_file_extent_num_bytes(leaf
, item
);
3459 if (start
< extent_start
|| start
>= extent_end
) {
3461 if (start
< extent_start
) {
3462 if (start
+ len
<= extent_start
)
3464 em
->len
= extent_end
- extent_start
;
3470 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
3472 em
->start
= extent_start
;
3473 em
->len
= extent_end
- extent_start
;
3474 em
->block_start
= EXTENT_MAP_HOLE
;
3477 em
->start
= extent_start
;
3478 em
->len
= extent_end
- extent_start
;
3480 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3481 em
->block_start
= bytenr
;
3482 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
3485 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
3486 em
->block_start
= bytenr
;
3487 em
->block_len
= em
->len
;
3490 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3495 size_t extent_offset
;
3498 size
= btrfs_file_extent_inline_len(leaf
, item
);
3499 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
3500 ~((u64
)root
->sectorsize
- 1);
3501 if (start
< extent_start
|| start
>= extent_end
) {
3503 if (start
< extent_start
) {
3504 if (start
+ len
<= extent_start
)
3506 em
->len
= extent_end
- extent_start
;
3512 em
->block_start
= EXTENT_MAP_INLINE
;
3514 if (!page
|| create
) {
3515 em
->start
= extent_start
;
3516 em
->len
= (size
+ root
->sectorsize
- 1) &
3517 ~((u64
)root
->sectorsize
- 1);
3521 page_start
= page_offset(page
) + pg_offset
;
3522 extent_offset
= page_start
- extent_start
;
3523 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
3524 size
- extent_offset
);
3525 em
->start
= extent_start
+ extent_offset
;
3526 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
3527 ~((u64
)root
->sectorsize
- 1);
3529 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3530 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
3531 if (create
== 0 && !PageUptodate(page
)) {
3532 if (btrfs_file_extent_compression(leaf
, item
) ==
3533 BTRFS_COMPRESS_ZLIB
) {
3534 ret
= uncompress_inline(path
, inode
, page
,
3536 extent_offset
, item
);
3540 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
3544 flush_dcache_page(page
);
3545 } else if (create
&& PageUptodate(page
)) {
3548 free_extent_map(em
);
3550 btrfs_release_path(root
, path
);
3551 trans
= btrfs_join_transaction(root
, 1);
3555 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
3558 btrfs_mark_buffer_dirty(leaf
);
3560 set_extent_uptodate(io_tree
, em
->start
,
3561 extent_map_end(em
) - 1, GFP_NOFS
);
3564 printk("unkknown found_type %d\n", found_type
);
3571 em
->block_start
= EXTENT_MAP_HOLE
;
3573 btrfs_release_path(root
, path
);
3574 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
3575 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em
->start
, em
->len
, start
, len
);
3581 spin_lock(&em_tree
->lock
);
3582 ret
= add_extent_mapping(em_tree
, em
);
3583 /* it is possible that someone inserted the extent into the tree
3584 * while we had the lock dropped. It is also possible that
3585 * an overlapping map exists in the tree
3587 if (ret
== -EEXIST
) {
3588 struct extent_map
*existing
;
3592 existing
= lookup_extent_mapping(em_tree
, start
, len
);
3593 if (existing
&& (existing
->start
> start
||
3594 existing
->start
+ existing
->len
<= start
)) {
3595 free_extent_map(existing
);
3599 existing
= lookup_extent_mapping(em_tree
, em
->start
,
3602 err
= merge_extent_mapping(em_tree
, existing
,
3605 free_extent_map(existing
);
3607 free_extent_map(em
);
3612 printk("failing to insert %Lu %Lu\n",
3614 free_extent_map(em
);
3618 free_extent_map(em
);
3623 spin_unlock(&em_tree
->lock
);
3626 btrfs_free_path(path
);
3628 ret
= btrfs_end_transaction(trans
, root
);
3634 free_extent_map(em
);
3636 return ERR_PTR(err
);
3641 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
3642 const struct iovec
*iov
, loff_t offset
,
3643 unsigned long nr_segs
)
3648 static sector_t
btrfs_bmap(struct address_space
*mapping
, sector_t iblock
)
3650 return extent_bmap(mapping
, iblock
, btrfs_get_extent
);
3653 int btrfs_readpage(struct file
*file
, struct page
*page
)
3655 struct extent_io_tree
*tree
;
3656 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
3657 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
3660 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
3662 struct extent_io_tree
*tree
;
3665 if (current
->flags
& PF_MEMALLOC
) {
3666 redirty_page_for_writepage(wbc
, page
);
3670 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
3671 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
3674 int btrfs_writepages(struct address_space
*mapping
,
3675 struct writeback_control
*wbc
)
3677 struct extent_io_tree
*tree
;
3678 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3679 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
3683 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
3684 struct list_head
*pages
, unsigned nr_pages
)
3686 struct extent_io_tree
*tree
;
3687 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3688 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
3691 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
3693 struct extent_io_tree
*tree
;
3694 struct extent_map_tree
*map
;
3697 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
3698 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
3699 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
3701 ClearPagePrivate(page
);
3702 set_page_private(page
, 0);
3703 page_cache_release(page
);
3708 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
3710 if (PageWriteback(page
) || PageDirty(page
))
3712 return __btrfs_releasepage(page
, gfp_flags
);
3715 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
3717 struct extent_io_tree
*tree
;
3718 struct btrfs_ordered_extent
*ordered
;
3719 u64 page_start
= page_offset(page
);
3720 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3722 wait_on_page_writeback(page
);
3723 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
3725 btrfs_releasepage(page
, GFP_NOFS
);
3729 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
3730 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
3734 * IO on this page will never be started, so we need
3735 * to account for any ordered extents now
3737 clear_extent_bit(tree
, page_start
, page_end
,
3738 EXTENT_DIRTY
| EXTENT_DELALLOC
|
3739 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
3740 btrfs_finish_ordered_io(page
->mapping
->host
,
3741 page_start
, page_end
);
3742 btrfs_put_ordered_extent(ordered
);
3743 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
3745 clear_extent_bit(tree
, page_start
, page_end
,
3746 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3749 __btrfs_releasepage(page
, GFP_NOFS
);
3751 ClearPageChecked(page
);
3752 if (PagePrivate(page
)) {
3753 ClearPagePrivate(page
);
3754 set_page_private(page
, 0);
3755 page_cache_release(page
);
3760 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3761 * called from a page fault handler when a page is first dirtied. Hence we must
3762 * be careful to check for EOF conditions here. We set the page up correctly
3763 * for a written page which means we get ENOSPC checking when writing into
3764 * holes and correct delalloc and unwritten extent mapping on filesystems that
3765 * support these features.
3767 * We are not allowed to take the i_mutex here so we have to play games to
3768 * protect against truncate races as the page could now be beyond EOF. Because
3769 * vmtruncate() writes the inode size before removing pages, once we have the
3770 * page lock we can determine safely if the page is beyond EOF. If it is not
3771 * beyond EOF, then the page is guaranteed safe against truncation until we
3774 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
3776 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
3777 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3778 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3779 struct btrfs_ordered_extent
*ordered
;
3781 unsigned long zero_start
;
3787 ret
= btrfs_check_free_space(root
, PAGE_CACHE_SIZE
, 0);
3794 size
= i_size_read(inode
);
3795 page_start
= page_offset(page
);
3796 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3798 if ((page
->mapping
!= inode
->i_mapping
) ||
3799 (page_start
>= size
)) {
3800 /* page got truncated out from underneath us */
3803 wait_on_page_writeback(page
);
3805 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3806 set_page_extent_mapped(page
);
3809 * we can't set the delalloc bits if there are pending ordered
3810 * extents. Drop our locks and wait for them to finish
3812 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3814 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3816 btrfs_start_ordered_extent(inode
, ordered
, 1);
3817 btrfs_put_ordered_extent(ordered
);
3821 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
3824 /* page is wholly or partially inside EOF */
3825 if (page_start
+ PAGE_CACHE_SIZE
> size
)
3826 zero_start
= size
& ~PAGE_CACHE_MASK
;
3828 zero_start
= PAGE_CACHE_SIZE
;
3830 if (zero_start
!= PAGE_CACHE_SIZE
) {
3832 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
3833 flush_dcache_page(page
);
3836 ClearPageChecked(page
);
3837 set_page_dirty(page
);
3838 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3846 static void btrfs_truncate(struct inode
*inode
)
3848 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3850 struct btrfs_trans_handle
*trans
;
3852 u64 mask
= root
->sectorsize
- 1;
3854 if (!S_ISREG(inode
->i_mode
))
3856 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3859 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
3860 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
3862 trans
= btrfs_start_transaction(root
, 1);
3863 btrfs_set_trans_block_group(trans
, inode
);
3864 btrfs_i_size_write(inode
, inode
->i_size
);
3866 ret
= btrfs_orphan_add(trans
, inode
);
3869 /* FIXME, add redo link to tree so we don't leak on crash */
3870 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
3871 BTRFS_EXTENT_DATA_KEY
);
3872 btrfs_update_inode(trans
, root
, inode
);
3874 ret
= btrfs_orphan_del(trans
, inode
);
3878 nr
= trans
->blocks_used
;
3879 ret
= btrfs_end_transaction_throttle(trans
, root
);
3881 btrfs_btree_balance_dirty(root
, nr
);
3885 * Invalidate a single dcache entry at the root of the filesystem.
3886 * Needed after creation of snapshot or subvolume.
3888 void btrfs_invalidate_dcache_root(struct btrfs_root
*root
, char *name
,
3891 struct dentry
*alias
, *entry
;
3894 alias
= d_find_alias(root
->fs_info
->sb
->s_root
->d_inode
);
3898 /* change me if btrfs ever gets a d_hash operation */
3899 qstr
.hash
= full_name_hash(qstr
.name
, qstr
.len
);
3900 entry
= d_lookup(alias
, &qstr
);
3903 d_invalidate(entry
);
3910 * create a new subvolume directory/inode (helper for the ioctl).
3912 int btrfs_create_subvol_root(struct btrfs_root
*new_root
, struct dentry
*dentry
,
3913 struct btrfs_trans_handle
*trans
, u64 new_dirid
,
3914 struct btrfs_block_group_cache
*block_group
)
3916 struct inode
*inode
;
3920 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
3921 new_dirid
, block_group
, S_IFDIR
| 0700, &index
);
3923 return PTR_ERR(inode
);
3924 inode
->i_op
= &btrfs_dir_inode_operations
;
3925 inode
->i_fop
= &btrfs_dir_file_operations
;
3926 new_root
->inode
= inode
;
3929 btrfs_i_size_write(inode
, 0);
3931 error
= btrfs_update_inode(trans
, new_root
, inode
);
3935 d_instantiate(dentry
, inode
);
3939 /* helper function for file defrag and space balancing. This
3940 * forces readahead on a given range of bytes in an inode
3942 unsigned long btrfs_force_ra(struct address_space
*mapping
,
3943 struct file_ra_state
*ra
, struct file
*file
,
3944 pgoff_t offset
, pgoff_t last_index
)
3946 pgoff_t req_size
= last_index
- offset
+ 1;
3948 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
3949 return offset
+ req_size
;
3952 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
3954 struct btrfs_inode
*ei
;
3956 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
3960 ei
->logged_trans
= 0;
3961 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
3962 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
3963 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
3964 INIT_LIST_HEAD(&ei
->i_orphan
);
3965 return &ei
->vfs_inode
;
3968 void btrfs_destroy_inode(struct inode
*inode
)
3970 struct btrfs_ordered_extent
*ordered
;
3971 WARN_ON(!list_empty(&inode
->i_dentry
));
3972 WARN_ON(inode
->i_data
.nrpages
);
3974 if (BTRFS_I(inode
)->i_acl
&&
3975 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
3976 posix_acl_release(BTRFS_I(inode
)->i_acl
);
3977 if (BTRFS_I(inode
)->i_default_acl
&&
3978 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
3979 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
3981 spin_lock(&BTRFS_I(inode
)->root
->list_lock
);
3982 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
3983 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
3984 " list\n", inode
->i_ino
);
3987 spin_unlock(&BTRFS_I(inode
)->root
->list_lock
);
3990 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
3994 printk("found ordered extent %Lu %Lu\n",
3995 ordered
->file_offset
, ordered
->len
);
3996 btrfs_remove_ordered_extent(inode
, ordered
);
3997 btrfs_put_ordered_extent(ordered
);
3998 btrfs_put_ordered_extent(ordered
);
4001 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4002 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4005 static void init_once(void *foo
)
4007 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4009 inode_init_once(&ei
->vfs_inode
);
4012 void btrfs_destroy_cachep(void)
4014 if (btrfs_inode_cachep
)
4015 kmem_cache_destroy(btrfs_inode_cachep
);
4016 if (btrfs_trans_handle_cachep
)
4017 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4018 if (btrfs_transaction_cachep
)
4019 kmem_cache_destroy(btrfs_transaction_cachep
);
4020 if (btrfs_bit_radix_cachep
)
4021 kmem_cache_destroy(btrfs_bit_radix_cachep
);
4022 if (btrfs_path_cachep
)
4023 kmem_cache_destroy(btrfs_path_cachep
);
4026 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
4027 unsigned long extra_flags
,
4028 void (*ctor
)(void *))
4030 return kmem_cache_create(name
, size
, 0, (SLAB_RECLAIM_ACCOUNT
|
4031 SLAB_MEM_SPREAD
| extra_flags
), ctor
);
4034 int btrfs_init_cachep(void)
4036 btrfs_inode_cachep
= btrfs_cache_create("btrfs_inode_cache",
4037 sizeof(struct btrfs_inode
),
4039 if (!btrfs_inode_cachep
)
4041 btrfs_trans_handle_cachep
=
4042 btrfs_cache_create("btrfs_trans_handle_cache",
4043 sizeof(struct btrfs_trans_handle
),
4045 if (!btrfs_trans_handle_cachep
)
4047 btrfs_transaction_cachep
= btrfs_cache_create("btrfs_transaction_cache",
4048 sizeof(struct btrfs_transaction
),
4050 if (!btrfs_transaction_cachep
)
4052 btrfs_path_cachep
= btrfs_cache_create("btrfs_path_cache",
4053 sizeof(struct btrfs_path
),
4055 if (!btrfs_path_cachep
)
4057 btrfs_bit_radix_cachep
= btrfs_cache_create("btrfs_radix", 256,
4058 SLAB_DESTROY_BY_RCU
, NULL
);
4059 if (!btrfs_bit_radix_cachep
)
4063 btrfs_destroy_cachep();
4067 static int btrfs_getattr(struct vfsmount
*mnt
,
4068 struct dentry
*dentry
, struct kstat
*stat
)
4070 struct inode
*inode
= dentry
->d_inode
;
4071 generic_fillattr(inode
, stat
);
4072 stat
->blksize
= PAGE_CACHE_SIZE
;
4073 stat
->blocks
= (inode_get_bytes(inode
) +
4074 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4078 static int btrfs_rename(struct inode
* old_dir
, struct dentry
*old_dentry
,
4079 struct inode
* new_dir
,struct dentry
*new_dentry
)
4081 struct btrfs_trans_handle
*trans
;
4082 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4083 struct inode
*new_inode
= new_dentry
->d_inode
;
4084 struct inode
*old_inode
= old_dentry
->d_inode
;
4085 struct timespec ctime
= CURRENT_TIME
;
4089 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4090 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4094 ret
= btrfs_check_free_space(root
, 1, 0);
4098 trans
= btrfs_start_transaction(root
, 1);
4100 btrfs_set_trans_block_group(trans
, new_dir
);
4102 btrfs_inc_nlink(old_dentry
->d_inode
);
4103 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4104 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4105 old_inode
->i_ctime
= ctime
;
4107 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4108 old_dentry
->d_name
.name
,
4109 old_dentry
->d_name
.len
);
4114 new_inode
->i_ctime
= CURRENT_TIME
;
4115 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4116 new_dentry
->d_inode
,
4117 new_dentry
->d_name
.name
,
4118 new_dentry
->d_name
.len
);
4121 if (new_inode
->i_nlink
== 0) {
4122 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4128 ret
= btrfs_set_inode_index(new_dir
, old_inode
, &index
);
4132 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4133 old_inode
, new_dentry
->d_name
.name
,
4134 new_dentry
->d_name
.len
, 1, index
);
4139 btrfs_end_transaction_throttle(trans
, root
);
4145 * some fairly slow code that needs optimization. This walks the list
4146 * of all the inodes with pending delalloc and forces them to disk.
4148 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4150 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4151 struct btrfs_inode
*binode
;
4152 struct inode
*inode
;
4153 unsigned long flags
;
4155 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
4156 while(!list_empty(head
)) {
4157 binode
= list_entry(head
->next
, struct btrfs_inode
,
4159 inode
= igrab(&binode
->vfs_inode
);
4161 list_del_init(&binode
->delalloc_inodes
);
4162 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
4164 filemap_flush(inode
->i_mapping
);
4168 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
4170 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
4172 /* the filemap_flush will queue IO into the worker threads, but
4173 * we have to make sure the IO is actually started and that
4174 * ordered extents get created before we return
4176 atomic_inc(&root
->fs_info
->async_submit_draining
);
4177 while(atomic_read(&root
->fs_info
->nr_async_submits
)) {
4178 wait_event(root
->fs_info
->async_submit_wait
,
4179 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0));
4181 atomic_dec(&root
->fs_info
->async_submit_draining
);
4185 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4186 const char *symname
)
4188 struct btrfs_trans_handle
*trans
;
4189 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4190 struct btrfs_path
*path
;
4191 struct btrfs_key key
;
4192 struct inode
*inode
= NULL
;
4200 struct btrfs_file_extent_item
*ei
;
4201 struct extent_buffer
*leaf
;
4202 unsigned long nr
= 0;
4204 name_len
= strlen(symname
) + 1;
4205 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4206 return -ENAMETOOLONG
;
4208 err
= btrfs_check_free_space(root
, 1, 0);
4212 trans
= btrfs_start_transaction(root
, 1);
4213 btrfs_set_trans_block_group(trans
, dir
);
4215 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4221 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4223 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4224 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4226 err
= PTR_ERR(inode
);
4230 err
= btrfs_init_acl(inode
, dir
);
4236 btrfs_set_trans_block_group(trans
, inode
);
4237 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4241 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4242 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4243 inode
->i_fop
= &btrfs_file_operations
;
4244 inode
->i_op
= &btrfs_file_inode_operations
;
4245 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4247 dir
->i_sb
->s_dirt
= 1;
4248 btrfs_update_inode_block_group(trans
, inode
);
4249 btrfs_update_inode_block_group(trans
, dir
);
4253 path
= btrfs_alloc_path();
4255 key
.objectid
= inode
->i_ino
;
4257 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
4258 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
4259 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
4265 leaf
= path
->nodes
[0];
4266 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
4267 struct btrfs_file_extent_item
);
4268 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
4269 btrfs_set_file_extent_type(leaf
, ei
,
4270 BTRFS_FILE_EXTENT_INLINE
);
4271 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
4272 btrfs_set_file_extent_compression(leaf
, ei
, 0);
4273 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
4274 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
4276 ptr
= btrfs_file_extent_inline_start(ei
);
4277 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
4278 btrfs_mark_buffer_dirty(leaf
);
4279 btrfs_free_path(path
);
4281 inode
->i_op
= &btrfs_symlink_inode_operations
;
4282 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
4283 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4284 btrfs_i_size_write(inode
, name_len
- 1);
4285 err
= btrfs_update_inode(trans
, root
, inode
);
4290 nr
= trans
->blocks_used
;
4291 btrfs_end_transaction_throttle(trans
, root
);
4294 inode_dec_link_count(inode
);
4297 btrfs_btree_balance_dirty(root
, nr
);
4301 static int btrfs_set_page_dirty(struct page
*page
)
4303 return __set_page_dirty_nobuffers(page
);
4306 static int btrfs_permission(struct inode
*inode
, int mask
)
4308 if (btrfs_test_flag(inode
, READONLY
) && (mask
& MAY_WRITE
))
4310 return generic_permission(inode
, mask
, btrfs_check_acl
);
4313 static struct inode_operations btrfs_dir_inode_operations
= {
4314 .lookup
= btrfs_lookup
,
4315 .create
= btrfs_create
,
4316 .unlink
= btrfs_unlink
,
4318 .mkdir
= btrfs_mkdir
,
4319 .rmdir
= btrfs_rmdir
,
4320 .rename
= btrfs_rename
,
4321 .symlink
= btrfs_symlink
,
4322 .setattr
= btrfs_setattr
,
4323 .mknod
= btrfs_mknod
,
4324 .setxattr
= btrfs_setxattr
,
4325 .getxattr
= btrfs_getxattr
,
4326 .listxattr
= btrfs_listxattr
,
4327 .removexattr
= btrfs_removexattr
,
4328 .permission
= btrfs_permission
,
4330 static struct inode_operations btrfs_dir_ro_inode_operations
= {
4331 .lookup
= btrfs_lookup
,
4332 .permission
= btrfs_permission
,
4334 static struct file_operations btrfs_dir_file_operations
= {
4335 .llseek
= generic_file_llseek
,
4336 .read
= generic_read_dir
,
4337 .readdir
= btrfs_real_readdir
,
4338 .unlocked_ioctl
= btrfs_ioctl
,
4339 #ifdef CONFIG_COMPAT
4340 .compat_ioctl
= btrfs_ioctl
,
4342 .release
= btrfs_release_file
,
4343 .fsync
= btrfs_sync_file
,
4346 static struct extent_io_ops btrfs_extent_io_ops
= {
4347 .fill_delalloc
= run_delalloc_range
,
4348 .submit_bio_hook
= btrfs_submit_bio_hook
,
4349 .merge_bio_hook
= btrfs_merge_bio_hook
,
4350 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
4351 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
4352 .writepage_start_hook
= btrfs_writepage_start_hook
,
4353 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
4354 .set_bit_hook
= btrfs_set_bit_hook
,
4355 .clear_bit_hook
= btrfs_clear_bit_hook
,
4358 static struct address_space_operations btrfs_aops
= {
4359 .readpage
= btrfs_readpage
,
4360 .writepage
= btrfs_writepage
,
4361 .writepages
= btrfs_writepages
,
4362 .readpages
= btrfs_readpages
,
4363 .sync_page
= block_sync_page
,
4365 .direct_IO
= btrfs_direct_IO
,
4366 .invalidatepage
= btrfs_invalidatepage
,
4367 .releasepage
= btrfs_releasepage
,
4368 .set_page_dirty
= btrfs_set_page_dirty
,
4371 static struct address_space_operations btrfs_symlink_aops
= {
4372 .readpage
= btrfs_readpage
,
4373 .writepage
= btrfs_writepage
,
4374 .invalidatepage
= btrfs_invalidatepage
,
4375 .releasepage
= btrfs_releasepage
,
4378 static struct inode_operations btrfs_file_inode_operations
= {
4379 .truncate
= btrfs_truncate
,
4380 .getattr
= btrfs_getattr
,
4381 .setattr
= btrfs_setattr
,
4382 .setxattr
= btrfs_setxattr
,
4383 .getxattr
= btrfs_getxattr
,
4384 .listxattr
= btrfs_listxattr
,
4385 .removexattr
= btrfs_removexattr
,
4386 .permission
= btrfs_permission
,
4388 static struct inode_operations btrfs_special_inode_operations
= {
4389 .getattr
= btrfs_getattr
,
4390 .setattr
= btrfs_setattr
,
4391 .permission
= btrfs_permission
,
4392 .setxattr
= btrfs_setxattr
,
4393 .getxattr
= btrfs_getxattr
,
4394 .listxattr
= btrfs_listxattr
,
4395 .removexattr
= btrfs_removexattr
,
4397 static struct inode_operations btrfs_symlink_inode_operations
= {
4398 .readlink
= generic_readlink
,
4399 .follow_link
= page_follow_link_light
,
4400 .put_link
= page_put_link
,
4401 .permission
= btrfs_permission
,