2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "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
);
88 static noinline
int cow_file_range(struct inode
*inode
,
89 struct page
*locked_page
,
90 u64 start
, u64 end
, int *page_started
,
91 unsigned long *nr_written
, int unlock
);
94 * a very lame attempt at stopping writes when the FS is 85% full. There
95 * are countless ways this is incorrect, but it is better than nothing.
97 int btrfs_check_free_space(struct btrfs_root
*root
, u64 num_required
,
105 spin_lock(&root
->fs_info
->delalloc_lock
);
106 total
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
107 used
= btrfs_super_bytes_used(&root
->fs_info
->super_copy
);
115 if (used
+ root
->fs_info
->delalloc_bytes
+ num_required
> thresh
)
117 spin_unlock(&root
->fs_info
->delalloc_lock
);
122 * this does all the hard work for inserting an inline extent into
123 * the btree. The caller should have done a btrfs_drop_extents so that
124 * no overlapping inline items exist in the btree
126 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
127 struct btrfs_root
*root
, struct inode
*inode
,
128 u64 start
, size_t size
, size_t compressed_size
,
129 struct page
**compressed_pages
)
131 struct btrfs_key key
;
132 struct btrfs_path
*path
;
133 struct extent_buffer
*leaf
;
134 struct page
*page
= NULL
;
137 struct btrfs_file_extent_item
*ei
;
140 size_t cur_size
= size
;
142 unsigned long offset
;
143 int use_compress
= 0;
145 if (compressed_size
&& compressed_pages
) {
147 cur_size
= compressed_size
;
150 path
= btrfs_alloc_path();
154 btrfs_set_trans_block_group(trans
, inode
);
156 key
.objectid
= inode
->i_ino
;
158 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
159 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
161 inode_add_bytes(inode
, size
);
162 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
169 leaf
= path
->nodes
[0];
170 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
171 struct btrfs_file_extent_item
);
172 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
173 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
174 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
175 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
176 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
177 ptr
= btrfs_file_extent_inline_start(ei
);
182 while (compressed_size
> 0) {
183 cpage
= compressed_pages
[i
];
184 cur_size
= min_t(unsigned long, compressed_size
,
188 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
193 compressed_size
-= cur_size
;
195 btrfs_set_file_extent_compression(leaf
, ei
,
196 BTRFS_COMPRESS_ZLIB
);
198 page
= find_get_page(inode
->i_mapping
,
199 start
>> PAGE_CACHE_SHIFT
);
200 btrfs_set_file_extent_compression(leaf
, ei
, 0);
201 kaddr
= kmap_atomic(page
, KM_USER0
);
202 offset
= start
& (PAGE_CACHE_SIZE
- 1);
203 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
204 kunmap_atomic(kaddr
, KM_USER0
);
205 page_cache_release(page
);
207 btrfs_mark_buffer_dirty(leaf
);
208 btrfs_free_path(path
);
210 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
211 btrfs_update_inode(trans
, root
, inode
);
214 btrfs_free_path(path
);
220 * conditionally insert an inline extent into the file. This
221 * does the checks required to make sure the data is small enough
222 * to fit as an inline extent.
224 static int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
225 struct btrfs_root
*root
,
226 struct inode
*inode
, u64 start
, u64 end
,
227 size_t compressed_size
,
228 struct page
**compressed_pages
)
230 u64 isize
= i_size_read(inode
);
231 u64 actual_end
= min(end
+ 1, isize
);
232 u64 inline_len
= actual_end
- start
;
233 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
234 ~((u64
)root
->sectorsize
- 1);
236 u64 data_len
= inline_len
;
240 data_len
= compressed_size
;
243 actual_end
>= PAGE_CACHE_SIZE
||
244 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
246 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
248 data_len
> root
->fs_info
->max_inline
) {
252 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
253 aligned_end
, start
, &hint_byte
);
256 if (isize
> actual_end
)
257 inline_len
= min_t(u64
, isize
, actual_end
);
258 ret
= insert_inline_extent(trans
, root
, inode
, start
,
259 inline_len
, compressed_size
,
262 btrfs_drop_extent_cache(inode
, start
, aligned_end
, 0);
266 struct async_extent
{
271 unsigned long nr_pages
;
272 struct list_head list
;
277 struct btrfs_root
*root
;
278 struct page
*locked_page
;
281 struct list_head extents
;
282 struct btrfs_work work
;
285 static noinline
int add_async_extent(struct async_cow
*cow
,
286 u64 start
, u64 ram_size
,
289 unsigned long nr_pages
)
291 struct async_extent
*async_extent
;
293 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
294 async_extent
->start
= start
;
295 async_extent
->ram_size
= ram_size
;
296 async_extent
->compressed_size
= compressed_size
;
297 async_extent
->pages
= pages
;
298 async_extent
->nr_pages
= nr_pages
;
299 list_add_tail(&async_extent
->list
, &cow
->extents
);
304 * we create compressed extents in two phases. The first
305 * phase compresses a range of pages that have already been
306 * locked (both pages and state bits are locked).
308 * This is done inside an ordered work queue, and the compression
309 * is spread across many cpus. The actual IO submission is step
310 * two, and the ordered work queue takes care of making sure that
311 * happens in the same order things were put onto the queue by
312 * writepages and friends.
314 * If this code finds it can't get good compression, it puts an
315 * entry onto the work queue to write the uncompressed bytes. This
316 * makes sure that both compressed inodes and uncompressed inodes
317 * are written in the same order that pdflush sent them down.
319 static noinline
int compress_file_range(struct inode
*inode
,
320 struct page
*locked_page
,
322 struct async_cow
*async_cow
,
325 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
326 struct btrfs_trans_handle
*trans
;
330 u64 blocksize
= root
->sectorsize
;
332 u64 isize
= i_size_read(inode
);
334 struct page
**pages
= NULL
;
335 unsigned long nr_pages
;
336 unsigned long nr_pages_ret
= 0;
337 unsigned long total_compressed
= 0;
338 unsigned long total_in
= 0;
339 unsigned long max_compressed
= 128 * 1024;
340 unsigned long max_uncompressed
= 128 * 1024;
346 actual_end
= min_t(u64
, isize
, end
+ 1);
349 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
350 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
352 total_compressed
= actual_end
- start
;
354 /* we want to make sure that amount of ram required to uncompress
355 * an extent is reasonable, so we limit the total size in ram
356 * of a compressed extent to 128k. This is a crucial number
357 * because it also controls how easily we can spread reads across
358 * cpus for decompression.
360 * We also want to make sure the amount of IO required to do
361 * a random read is reasonably small, so we limit the size of
362 * a compressed extent to 128k.
364 total_compressed
= min(total_compressed
, max_uncompressed
);
365 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
366 num_bytes
= max(blocksize
, num_bytes
);
367 disk_num_bytes
= num_bytes
;
372 * we do compression for mount -o compress and when the
373 * inode has not been flagged as nocompress. This flag can
374 * change at any time if we discover bad compression ratios.
376 if (!btrfs_test_flag(inode
, NOCOMPRESS
) &&
377 btrfs_test_opt(root
, COMPRESS
)) {
379 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
381 ret
= btrfs_zlib_compress_pages(inode
->i_mapping
, start
,
382 total_compressed
, pages
,
383 nr_pages
, &nr_pages_ret
,
389 unsigned long offset
= total_compressed
&
390 (PAGE_CACHE_SIZE
- 1);
391 struct page
*page
= pages
[nr_pages_ret
- 1];
394 /* zero the tail end of the last page, we might be
395 * sending it down to disk
398 kaddr
= kmap_atomic(page
, KM_USER0
);
399 memset(kaddr
+ offset
, 0,
400 PAGE_CACHE_SIZE
- offset
);
401 kunmap_atomic(kaddr
, KM_USER0
);
407 trans
= btrfs_join_transaction(root
, 1);
409 btrfs_set_trans_block_group(trans
, inode
);
411 /* lets try to make an inline extent */
412 if (ret
|| total_in
< (actual_end
- start
)) {
413 /* we didn't compress the entire range, try
414 * to make an uncompressed inline extent.
416 ret
= cow_file_range_inline(trans
, root
, inode
,
417 start
, end
, 0, NULL
);
419 /* try making a compressed inline extent */
420 ret
= cow_file_range_inline(trans
, root
, inode
,
422 total_compressed
, pages
);
424 btrfs_end_transaction(trans
, root
);
427 * inline extent creation worked, we don't need
428 * to create any more async work items. Unlock
429 * and free up our temp pages.
431 extent_clear_unlock_delalloc(inode
,
432 &BTRFS_I(inode
)->io_tree
,
433 start
, end
, NULL
, 1, 0,
442 * we aren't doing an inline extent round the compressed size
443 * up to a block size boundary so the allocator does sane
446 total_compressed
= (total_compressed
+ blocksize
- 1) &
450 * one last check to make sure the compression is really a
451 * win, compare the page count read with the blocks on disk
453 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
454 ~(PAGE_CACHE_SIZE
- 1);
455 if (total_compressed
>= total_in
) {
458 disk_num_bytes
= total_compressed
;
459 num_bytes
= total_in
;
462 if (!will_compress
&& pages
) {
464 * the compression code ran but failed to make things smaller,
465 * free any pages it allocated and our page pointer array
467 for (i
= 0; i
< nr_pages_ret
; i
++) {
468 WARN_ON(pages
[i
]->mapping
);
469 page_cache_release(pages
[i
]);
473 total_compressed
= 0;
476 /* flag the file so we don't compress in the future */
477 btrfs_set_flag(inode
, NOCOMPRESS
);
482 /* the async work queues will take care of doing actual
483 * allocation on disk for these compressed pages,
484 * and will submit them to the elevator.
486 add_async_extent(async_cow
, start
, num_bytes
,
487 total_compressed
, pages
, nr_pages_ret
);
489 if (start
+ num_bytes
< end
&& start
+ num_bytes
< actual_end
) {
497 * No compression, but we still need to write the pages in
498 * the file we've been given so far. redirty the locked
499 * page if it corresponds to our extent and set things up
500 * for the async work queue to run cow_file_range to do
501 * the normal delalloc dance
503 if (page_offset(locked_page
) >= start
&&
504 page_offset(locked_page
) <= end
) {
505 __set_page_dirty_nobuffers(locked_page
);
506 /* unlocked later on in the async handlers */
508 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0);
516 for (i
= 0; i
< nr_pages_ret
; i
++) {
517 WARN_ON(pages
[i
]->mapping
);
518 page_cache_release(pages
[i
]);
526 * phase two of compressed writeback. This is the ordered portion
527 * of the code, which only gets called in the order the work was
528 * queued. We walk all the async extents created by compress_file_range
529 * and send them down to the disk.
531 static noinline
int submit_compressed_extents(struct inode
*inode
,
532 struct async_cow
*async_cow
)
534 struct async_extent
*async_extent
;
536 struct btrfs_trans_handle
*trans
;
537 struct btrfs_key ins
;
538 struct extent_map
*em
;
539 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
540 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
541 struct extent_io_tree
*io_tree
;
544 if (list_empty(&async_cow
->extents
))
547 trans
= btrfs_join_transaction(root
, 1);
549 while (!list_empty(&async_cow
->extents
)) {
550 async_extent
= list_entry(async_cow
->extents
.next
,
551 struct async_extent
, list
);
552 list_del(&async_extent
->list
);
554 io_tree
= &BTRFS_I(inode
)->io_tree
;
556 /* did the compression code fall back to uncompressed IO? */
557 if (!async_extent
->pages
) {
558 int page_started
= 0;
559 unsigned long nr_written
= 0;
561 lock_extent(io_tree
, async_extent
->start
,
562 async_extent
->start
+
563 async_extent
->ram_size
- 1, GFP_NOFS
);
565 /* allocate blocks */
566 cow_file_range(inode
, async_cow
->locked_page
,
568 async_extent
->start
+
569 async_extent
->ram_size
- 1,
570 &page_started
, &nr_written
, 0);
573 * if page_started, cow_file_range inserted an
574 * inline extent and took care of all the unlocking
575 * and IO for us. Otherwise, we need to submit
576 * all those pages down to the drive.
579 extent_write_locked_range(io_tree
,
580 inode
, async_extent
->start
,
581 async_extent
->start
+
582 async_extent
->ram_size
- 1,
590 lock_extent(io_tree
, async_extent
->start
,
591 async_extent
->start
+ async_extent
->ram_size
- 1,
594 * here we're doing allocation and writeback of the
597 btrfs_drop_extent_cache(inode
, async_extent
->start
,
598 async_extent
->start
+
599 async_extent
->ram_size
- 1, 0);
601 ret
= btrfs_reserve_extent(trans
, root
,
602 async_extent
->compressed_size
,
603 async_extent
->compressed_size
,
607 em
= alloc_extent_map(GFP_NOFS
);
608 em
->start
= async_extent
->start
;
609 em
->len
= async_extent
->ram_size
;
610 em
->orig_start
= em
->start
;
612 em
->block_start
= ins
.objectid
;
613 em
->block_len
= ins
.offset
;
614 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
615 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
616 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
619 spin_lock(&em_tree
->lock
);
620 ret
= add_extent_mapping(em_tree
, em
);
621 spin_unlock(&em_tree
->lock
);
622 if (ret
!= -EEXIST
) {
626 btrfs_drop_extent_cache(inode
, async_extent
->start
,
627 async_extent
->start
+
628 async_extent
->ram_size
- 1, 0);
631 ret
= btrfs_add_ordered_extent(inode
, async_extent
->start
,
633 async_extent
->ram_size
,
635 BTRFS_ORDERED_COMPRESSED
);
638 btrfs_end_transaction(trans
, root
);
641 * clear dirty, set writeback and unlock the pages.
643 extent_clear_unlock_delalloc(inode
,
644 &BTRFS_I(inode
)->io_tree
,
646 async_extent
->start
+
647 async_extent
->ram_size
- 1,
648 NULL
, 1, 1, 0, 1, 1, 0);
650 ret
= btrfs_submit_compressed_write(inode
,
652 async_extent
->ram_size
,
654 ins
.offset
, async_extent
->pages
,
655 async_extent
->nr_pages
);
658 trans
= btrfs_join_transaction(root
, 1);
659 alloc_hint
= ins
.objectid
+ ins
.offset
;
664 btrfs_end_transaction(trans
, root
);
669 * when extent_io.c finds a delayed allocation range in the file,
670 * the call backs end up in this code. The basic idea is to
671 * allocate extents on disk for the range, and create ordered data structs
672 * in ram to track those extents.
674 * locked_page is the page that writepage had locked already. We use
675 * it to make sure we don't do extra locks or unlocks.
677 * *page_started is set to one if we unlock locked_page and do everything
678 * required to start IO on it. It may be clean and already done with
681 static noinline
int cow_file_range(struct inode
*inode
,
682 struct page
*locked_page
,
683 u64 start
, u64 end
, int *page_started
,
684 unsigned long *nr_written
,
687 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
688 struct btrfs_trans_handle
*trans
;
691 unsigned long ram_size
;
694 u64 blocksize
= root
->sectorsize
;
696 u64 isize
= i_size_read(inode
);
697 struct btrfs_key ins
;
698 struct extent_map
*em
;
699 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
702 trans
= btrfs_join_transaction(root
, 1);
704 btrfs_set_trans_block_group(trans
, inode
);
706 actual_end
= min_t(u64
, isize
, end
+ 1);
708 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
709 num_bytes
= max(blocksize
, num_bytes
);
710 disk_num_bytes
= num_bytes
;
714 /* lets try to make an inline extent */
715 ret
= cow_file_range_inline(trans
, root
, inode
,
716 start
, end
, 0, NULL
);
718 extent_clear_unlock_delalloc(inode
,
719 &BTRFS_I(inode
)->io_tree
,
720 start
, end
, NULL
, 1, 1,
722 *nr_written
= *nr_written
+
723 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
730 BUG_ON(disk_num_bytes
>
731 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
733 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
735 while (disk_num_bytes
> 0) {
736 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
737 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
738 root
->sectorsize
, 0, alloc_hint
,
742 em
= alloc_extent_map(GFP_NOFS
);
744 em
->orig_start
= em
->start
;
746 ram_size
= ins
.offset
;
747 em
->len
= ins
.offset
;
749 em
->block_start
= ins
.objectid
;
750 em
->block_len
= ins
.offset
;
751 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
752 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
755 spin_lock(&em_tree
->lock
);
756 ret
= add_extent_mapping(em_tree
, em
);
757 spin_unlock(&em_tree
->lock
);
758 if (ret
!= -EEXIST
) {
762 btrfs_drop_extent_cache(inode
, start
,
763 start
+ ram_size
- 1, 0);
766 cur_alloc_size
= ins
.offset
;
767 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
768 ram_size
, cur_alloc_size
, 0);
771 if (root
->root_key
.objectid
==
772 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
773 ret
= btrfs_reloc_clone_csums(inode
, start
,
778 if (disk_num_bytes
< cur_alloc_size
)
781 /* we're not doing compressed IO, don't unlock the first
782 * page (which the caller expects to stay locked), don't
783 * clear any dirty bits and don't set any writeback bits
785 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
786 start
, start
+ ram_size
- 1,
787 locked_page
, unlock
, 1,
789 disk_num_bytes
-= cur_alloc_size
;
790 num_bytes
-= cur_alloc_size
;
791 alloc_hint
= ins
.objectid
+ ins
.offset
;
792 start
+= cur_alloc_size
;
796 btrfs_end_transaction(trans
, root
);
802 * work queue call back to started compression on a file and pages
804 static noinline
void async_cow_start(struct btrfs_work
*work
)
806 struct async_cow
*async_cow
;
808 async_cow
= container_of(work
, struct async_cow
, work
);
810 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
811 async_cow
->start
, async_cow
->end
, async_cow
,
814 async_cow
->inode
= NULL
;
818 * work queue call back to submit previously compressed pages
820 static noinline
void async_cow_submit(struct btrfs_work
*work
)
822 struct async_cow
*async_cow
;
823 struct btrfs_root
*root
;
824 unsigned long nr_pages
;
826 async_cow
= container_of(work
, struct async_cow
, work
);
828 root
= async_cow
->root
;
829 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
832 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
834 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
836 waitqueue_active(&root
->fs_info
->async_submit_wait
))
837 wake_up(&root
->fs_info
->async_submit_wait
);
839 if (async_cow
->inode
)
840 submit_compressed_extents(async_cow
->inode
, async_cow
);
843 static noinline
void async_cow_free(struct btrfs_work
*work
)
845 struct async_cow
*async_cow
;
846 async_cow
= container_of(work
, struct async_cow
, work
);
850 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
851 u64 start
, u64 end
, int *page_started
,
852 unsigned long *nr_written
)
854 struct async_cow
*async_cow
;
855 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
856 unsigned long nr_pages
;
858 int limit
= 10 * 1024 * 1042;
860 if (!btrfs_test_opt(root
, COMPRESS
)) {
861 return cow_file_range(inode
, locked_page
, start
, end
,
862 page_started
, nr_written
, 1);
865 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
|
866 EXTENT_DELALLOC
, 1, 0, GFP_NOFS
);
867 while (start
< end
) {
868 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
869 async_cow
->inode
= inode
;
870 async_cow
->root
= root
;
871 async_cow
->locked_page
= locked_page
;
872 async_cow
->start
= start
;
874 if (btrfs_test_flag(inode
, NOCOMPRESS
))
877 cur_end
= min(end
, start
+ 512 * 1024 - 1);
879 async_cow
->end
= cur_end
;
880 INIT_LIST_HEAD(&async_cow
->extents
);
882 async_cow
->work
.func
= async_cow_start
;
883 async_cow
->work
.ordered_func
= async_cow_submit
;
884 async_cow
->work
.ordered_free
= async_cow_free
;
885 async_cow
->work
.flags
= 0;
887 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
889 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
891 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
894 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
895 wait_event(root
->fs_info
->async_submit_wait
,
896 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
900 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
901 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
902 wait_event(root
->fs_info
->async_submit_wait
,
903 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
907 *nr_written
+= nr_pages
;
914 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
915 u64 bytenr
, u64 num_bytes
)
918 struct btrfs_ordered_sum
*sums
;
921 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
922 bytenr
+ num_bytes
- 1, &list
);
923 if (ret
== 0 && list_empty(&list
))
926 while (!list_empty(&list
)) {
927 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
928 list_del(&sums
->list
);
935 * when nowcow writeback call back. This checks for snapshots or COW copies
936 * of the extents that exist in the file, and COWs the file as required.
938 * If no cow copies or snapshots exist, we write directly to the existing
941 static int run_delalloc_nocow(struct inode
*inode
, struct page
*locked_page
,
942 u64 start
, u64 end
, int *page_started
, int force
,
943 unsigned long *nr_written
)
945 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
946 struct btrfs_trans_handle
*trans
;
947 struct extent_buffer
*leaf
;
948 struct btrfs_path
*path
;
949 struct btrfs_file_extent_item
*fi
;
950 struct btrfs_key found_key
;
962 path
= btrfs_alloc_path();
964 trans
= btrfs_join_transaction(root
, 1);
970 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
973 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
974 leaf
= path
->nodes
[0];
975 btrfs_item_key_to_cpu(leaf
, &found_key
,
977 if (found_key
.objectid
== inode
->i_ino
&&
978 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
983 leaf
= path
->nodes
[0];
984 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
985 ret
= btrfs_next_leaf(root
, path
);
990 leaf
= path
->nodes
[0];
996 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
998 if (found_key
.objectid
> inode
->i_ino
||
999 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1000 found_key
.offset
> end
)
1003 if (found_key
.offset
> cur_offset
) {
1004 extent_end
= found_key
.offset
;
1008 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1009 struct btrfs_file_extent_item
);
1010 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1012 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1013 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1014 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1015 extent_end
= found_key
.offset
+
1016 btrfs_file_extent_num_bytes(leaf
, fi
);
1017 if (extent_end
<= start
) {
1021 if (disk_bytenr
== 0)
1023 if (btrfs_file_extent_compression(leaf
, fi
) ||
1024 btrfs_file_extent_encryption(leaf
, fi
) ||
1025 btrfs_file_extent_other_encoding(leaf
, fi
))
1027 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1029 if (btrfs_extent_readonly(root
, disk_bytenr
))
1031 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1034 disk_bytenr
+= btrfs_file_extent_offset(leaf
, fi
);
1035 disk_bytenr
+= cur_offset
- found_key
.offset
;
1036 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1038 * force cow if csum exists in the range.
1039 * this ensure that csum for a given extent are
1040 * either valid or do not exist.
1042 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1045 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1046 extent_end
= found_key
.offset
+
1047 btrfs_file_extent_inline_len(leaf
, fi
);
1048 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1053 if (extent_end
<= start
) {
1058 if (cow_start
== (u64
)-1)
1059 cow_start
= cur_offset
;
1060 cur_offset
= extent_end
;
1061 if (cur_offset
> end
)
1067 btrfs_release_path(root
, path
);
1068 if (cow_start
!= (u64
)-1) {
1069 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1070 found_key
.offset
- 1, page_started
,
1073 cow_start
= (u64
)-1;
1076 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1077 struct extent_map
*em
;
1078 struct extent_map_tree
*em_tree
;
1079 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1080 em
= alloc_extent_map(GFP_NOFS
);
1081 em
->start
= cur_offset
;
1082 em
->orig_start
= em
->start
;
1083 em
->len
= num_bytes
;
1084 em
->block_len
= num_bytes
;
1085 em
->block_start
= disk_bytenr
;
1086 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1087 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1089 spin_lock(&em_tree
->lock
);
1090 ret
= add_extent_mapping(em_tree
, em
);
1091 spin_unlock(&em_tree
->lock
);
1092 if (ret
!= -EEXIST
) {
1093 free_extent_map(em
);
1096 btrfs_drop_extent_cache(inode
, em
->start
,
1097 em
->start
+ em
->len
- 1, 0);
1099 type
= BTRFS_ORDERED_PREALLOC
;
1101 type
= BTRFS_ORDERED_NOCOW
;
1104 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1105 num_bytes
, num_bytes
, type
);
1108 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1109 cur_offset
, cur_offset
+ num_bytes
- 1,
1110 locked_page
, 1, 1, 1, 0, 0, 0);
1111 cur_offset
= extent_end
;
1112 if (cur_offset
> end
)
1115 btrfs_release_path(root
, path
);
1117 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1118 cow_start
= cur_offset
;
1119 if (cow_start
!= (u64
)-1) {
1120 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1121 page_started
, nr_written
, 1);
1125 ret
= btrfs_end_transaction(trans
, root
);
1127 btrfs_free_path(path
);
1132 * extent_io.c call back to do delayed allocation processing
1134 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1135 u64 start
, u64 end
, int *page_started
,
1136 unsigned long *nr_written
)
1140 if (btrfs_test_flag(inode
, NODATACOW
))
1141 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1142 page_started
, 1, nr_written
);
1143 else if (btrfs_test_flag(inode
, PREALLOC
))
1144 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1145 page_started
, 0, nr_written
);
1147 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1148 page_started
, nr_written
);
1154 * extent_io.c set_bit_hook, used to track delayed allocation
1155 * bytes in this file, and to maintain the list of inodes that
1156 * have pending delalloc work to be done.
1158 static int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1159 unsigned long old
, unsigned long bits
)
1162 * set_bit and clear bit hooks normally require _irqsave/restore
1163 * but in this case, we are only testeing for the DELALLOC
1164 * bit, which is only set or cleared with irqs on
1166 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1167 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1168 spin_lock(&root
->fs_info
->delalloc_lock
);
1169 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
1170 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
1171 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1172 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1173 &root
->fs_info
->delalloc_inodes
);
1175 spin_unlock(&root
->fs_info
->delalloc_lock
);
1181 * extent_io.c clear_bit_hook, see set_bit_hook for why
1183 static int btrfs_clear_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1184 unsigned long old
, unsigned long bits
)
1187 * set_bit and clear bit hooks normally require _irqsave/restore
1188 * but in this case, we are only testeing for the DELALLOC
1189 * bit, which is only set or cleared with irqs on
1191 if ((old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1192 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1194 spin_lock(&root
->fs_info
->delalloc_lock
);
1195 if (end
- start
+ 1 > root
->fs_info
->delalloc_bytes
) {
1196 printk(KERN_INFO
"btrfs warning: delalloc account "
1198 (unsigned long long)end
- start
+ 1,
1199 (unsigned long long)
1200 root
->fs_info
->delalloc_bytes
);
1201 root
->fs_info
->delalloc_bytes
= 0;
1202 BTRFS_I(inode
)->delalloc_bytes
= 0;
1204 root
->fs_info
->delalloc_bytes
-= end
- start
+ 1;
1205 BTRFS_I(inode
)->delalloc_bytes
-= end
- start
+ 1;
1207 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1208 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1209 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1211 spin_unlock(&root
->fs_info
->delalloc_lock
);
1217 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1218 * we don't create bios that span stripes or chunks
1220 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1221 size_t size
, struct bio
*bio
,
1222 unsigned long bio_flags
)
1224 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1225 struct btrfs_mapping_tree
*map_tree
;
1226 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1231 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1234 length
= bio
->bi_size
;
1235 map_tree
= &root
->fs_info
->mapping_tree
;
1236 map_length
= length
;
1237 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1238 &map_length
, NULL
, 0);
1240 if (map_length
< length
+ size
)
1246 * in order to insert checksums into the metadata in large chunks,
1247 * we wait until bio submission time. All the pages in the bio are
1248 * checksummed and sums are attached onto the ordered extent record.
1250 * At IO completion time the cums attached on the ordered extent record
1251 * are inserted into the btree
1253 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1254 struct bio
*bio
, int mirror_num
,
1255 unsigned long bio_flags
)
1257 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1260 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1266 * in order to insert checksums into the metadata in large chunks,
1267 * we wait until bio submission time. All the pages in the bio are
1268 * checksummed and sums are attached onto the ordered extent record.
1270 * At IO completion time the cums attached on the ordered extent record
1271 * are inserted into the btree
1273 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1274 int mirror_num
, unsigned long bio_flags
)
1276 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1277 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1281 * extent_io.c submission hook. This does the right thing for csum calculation
1282 * on write, or reading the csums from the tree before a read
1284 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1285 int mirror_num
, unsigned long bio_flags
)
1287 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1291 skip_sum
= btrfs_test_flag(inode
, NODATASUM
);
1293 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1296 if (!(rw
& (1 << BIO_RW
))) {
1297 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1298 return btrfs_submit_compressed_read(inode
, bio
,
1299 mirror_num
, bio_flags
);
1300 } else if (!skip_sum
)
1301 btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1303 } else if (!skip_sum
) {
1304 /* csum items have already been cloned */
1305 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1307 /* we're doing a write, do the async checksumming */
1308 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1309 inode
, rw
, bio
, mirror_num
,
1310 bio_flags
, __btrfs_submit_bio_start
,
1311 __btrfs_submit_bio_done
);
1315 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1319 * given a list of ordered sums record them in the inode. This happens
1320 * at IO completion time based on sums calculated at bio submission time.
1322 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1323 struct inode
*inode
, u64 file_offset
,
1324 struct list_head
*list
)
1326 struct list_head
*cur
;
1327 struct btrfs_ordered_sum
*sum
;
1329 btrfs_set_trans_block_group(trans
, inode
);
1330 list_for_each(cur
, list
) {
1331 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
1332 btrfs_csum_file_blocks(trans
,
1333 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1338 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
1340 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1342 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1346 /* see btrfs_writepage_start_hook for details on why this is required */
1347 struct btrfs_writepage_fixup
{
1349 struct btrfs_work work
;
1352 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1354 struct btrfs_writepage_fixup
*fixup
;
1355 struct btrfs_ordered_extent
*ordered
;
1357 struct inode
*inode
;
1361 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1365 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1366 ClearPageChecked(page
);
1370 inode
= page
->mapping
->host
;
1371 page_start
= page_offset(page
);
1372 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1374 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1376 /* already ordered? We're done */
1377 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1378 EXTENT_ORDERED
, 0)) {
1382 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1384 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
1385 page_end
, GFP_NOFS
);
1387 btrfs_start_ordered_extent(inode
, ordered
, 1);
1391 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
1392 ClearPageChecked(page
);
1394 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1397 page_cache_release(page
);
1401 * There are a few paths in the higher layers of the kernel that directly
1402 * set the page dirty bit without asking the filesystem if it is a
1403 * good idea. This causes problems because we want to make sure COW
1404 * properly happens and the data=ordered rules are followed.
1406 * In our case any range that doesn't have the ORDERED bit set
1407 * hasn't been properly setup for IO. We kick off an async process
1408 * to fix it up. The async helper will wait for ordered extents, set
1409 * the delalloc bit and make it safe to write the page.
1411 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1413 struct inode
*inode
= page
->mapping
->host
;
1414 struct btrfs_writepage_fixup
*fixup
;
1415 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1418 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1423 if (PageChecked(page
))
1426 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1430 SetPageChecked(page
);
1431 page_cache_get(page
);
1432 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1434 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1438 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1439 struct inode
*inode
, u64 file_pos
,
1440 u64 disk_bytenr
, u64 disk_num_bytes
,
1441 u64 num_bytes
, u64 ram_bytes
,
1442 u8 compression
, u8 encryption
,
1443 u16 other_encoding
, int extent_type
)
1445 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1446 struct btrfs_file_extent_item
*fi
;
1447 struct btrfs_path
*path
;
1448 struct extent_buffer
*leaf
;
1449 struct btrfs_key ins
;
1453 path
= btrfs_alloc_path();
1456 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1457 file_pos
+ num_bytes
, file_pos
, &hint
);
1460 ins
.objectid
= inode
->i_ino
;
1461 ins
.offset
= file_pos
;
1462 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1463 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1465 leaf
= path
->nodes
[0];
1466 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1467 struct btrfs_file_extent_item
);
1468 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1469 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1470 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1471 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1472 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1473 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1474 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1475 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1476 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1477 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1478 btrfs_mark_buffer_dirty(leaf
);
1480 inode_add_bytes(inode
, num_bytes
);
1481 btrfs_drop_extent_cache(inode
, file_pos
, file_pos
+ num_bytes
- 1, 0);
1483 ins
.objectid
= disk_bytenr
;
1484 ins
.offset
= disk_num_bytes
;
1485 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1486 ret
= btrfs_alloc_reserved_extent(trans
, root
, leaf
->start
,
1487 root
->root_key
.objectid
,
1488 trans
->transid
, inode
->i_ino
, &ins
);
1491 btrfs_free_path(path
);
1495 /* as ordered data IO finishes, this gets called so we can finish
1496 * an ordered extent if the range of bytes in the file it covers are
1499 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1501 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1502 struct btrfs_trans_handle
*trans
;
1503 struct btrfs_ordered_extent
*ordered_extent
;
1504 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1508 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1512 trans
= btrfs_join_transaction(root
, 1);
1514 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1515 BUG_ON(!ordered_extent
);
1516 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1519 lock_extent(io_tree
, ordered_extent
->file_offset
,
1520 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1523 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1525 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1527 ret
= btrfs_mark_extent_written(trans
, root
, inode
,
1528 ordered_extent
->file_offset
,
1529 ordered_extent
->file_offset
+
1530 ordered_extent
->len
);
1533 ret
= insert_reserved_file_extent(trans
, inode
,
1534 ordered_extent
->file_offset
,
1535 ordered_extent
->start
,
1536 ordered_extent
->disk_len
,
1537 ordered_extent
->len
,
1538 ordered_extent
->len
,
1540 BTRFS_FILE_EXTENT_REG
);
1543 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1544 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1547 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1548 &ordered_extent
->list
);
1550 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1551 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1552 btrfs_update_inode(trans
, root
, inode
);
1553 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1554 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1557 btrfs_put_ordered_extent(ordered_extent
);
1558 /* once for the tree */
1559 btrfs_put_ordered_extent(ordered_extent
);
1561 btrfs_end_transaction(trans
, root
);
1565 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1566 struct extent_state
*state
, int uptodate
)
1568 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1572 * When IO fails, either with EIO or csum verification fails, we
1573 * try other mirrors that might have a good copy of the data. This
1574 * io_failure_record is used to record state as we go through all the
1575 * mirrors. If another mirror has good data, the page is set up to date
1576 * and things continue. If a good mirror can't be found, the original
1577 * bio end_io callback is called to indicate things have failed.
1579 struct io_failure_record
{
1584 unsigned long bio_flags
;
1588 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1589 struct page
*page
, u64 start
, u64 end
,
1590 struct extent_state
*state
)
1592 struct io_failure_record
*failrec
= NULL
;
1594 struct extent_map
*em
;
1595 struct inode
*inode
= page
->mapping
->host
;
1596 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1597 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1604 ret
= get_state_private(failure_tree
, start
, &private);
1606 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1609 failrec
->start
= start
;
1610 failrec
->len
= end
- start
+ 1;
1611 failrec
->last_mirror
= 0;
1612 failrec
->bio_flags
= 0;
1614 spin_lock(&em_tree
->lock
);
1615 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1616 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1617 free_extent_map(em
);
1620 spin_unlock(&em_tree
->lock
);
1622 if (!em
|| IS_ERR(em
)) {
1626 logical
= start
- em
->start
;
1627 logical
= em
->block_start
+ logical
;
1628 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1629 logical
= em
->block_start
;
1630 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1632 failrec
->logical
= logical
;
1633 free_extent_map(em
);
1634 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1635 EXTENT_DIRTY
, GFP_NOFS
);
1636 set_state_private(failure_tree
, start
,
1637 (u64
)(unsigned long)failrec
);
1639 failrec
= (struct io_failure_record
*)(unsigned long)private;
1641 num_copies
= btrfs_num_copies(
1642 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1643 failrec
->logical
, failrec
->len
);
1644 failrec
->last_mirror
++;
1646 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1647 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1650 if (state
&& state
->start
!= failrec
->start
)
1652 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1654 if (!state
|| failrec
->last_mirror
> num_copies
) {
1655 set_state_private(failure_tree
, failrec
->start
, 0);
1656 clear_extent_bits(failure_tree
, failrec
->start
,
1657 failrec
->start
+ failrec
->len
- 1,
1658 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1662 bio
= bio_alloc(GFP_NOFS
, 1);
1663 bio
->bi_private
= state
;
1664 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1665 bio
->bi_sector
= failrec
->logical
>> 9;
1666 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1669 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1670 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1675 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1676 failrec
->last_mirror
,
1677 failrec
->bio_flags
);
1682 * each time an IO finishes, we do a fast check in the IO failure tree
1683 * to see if we need to process or clean up an io_failure_record
1685 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1688 u64 private_failure
;
1689 struct io_failure_record
*failure
;
1693 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1694 (u64
)-1, 1, EXTENT_DIRTY
)) {
1695 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1696 start
, &private_failure
);
1698 failure
= (struct io_failure_record
*)(unsigned long)
1700 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1702 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1704 failure
->start
+ failure
->len
- 1,
1705 EXTENT_DIRTY
| EXTENT_LOCKED
,
1714 * when reads are done, we need to check csums to verify the data is correct
1715 * if there's a match, we allow the bio to finish. If not, we go through
1716 * the io_failure_record routines to find good copies
1718 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1719 struct extent_state
*state
)
1721 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1722 struct inode
*inode
= page
->mapping
->host
;
1723 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1725 u64
private = ~(u32
)0;
1727 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1730 if (PageChecked(page
)) {
1731 ClearPageChecked(page
);
1734 if (btrfs_test_flag(inode
, NODATASUM
))
1737 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1738 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1)) {
1739 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1744 if (state
&& state
->start
== start
) {
1745 private = state
->private;
1748 ret
= get_state_private(io_tree
, start
, &private);
1750 kaddr
= kmap_atomic(page
, KM_USER0
);
1754 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1755 btrfs_csum_final(csum
, (char *)&csum
);
1756 if (csum
!= private)
1759 kunmap_atomic(kaddr
, KM_USER0
);
1761 /* if the io failure tree for this inode is non-empty,
1762 * check to see if we've recovered from a failed IO
1764 btrfs_clean_io_failures(inode
, start
);
1768 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
1769 "private %llu\n", page
->mapping
->host
->i_ino
,
1770 (unsigned long long)start
, csum
,
1771 (unsigned long long)private);
1772 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1773 flush_dcache_page(page
);
1774 kunmap_atomic(kaddr
, KM_USER0
);
1781 * This creates an orphan entry for the given inode in case something goes
1782 * wrong in the middle of an unlink/truncate.
1784 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1786 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1789 spin_lock(&root
->list_lock
);
1791 /* already on the orphan list, we're good */
1792 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1793 spin_unlock(&root
->list_lock
);
1797 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1799 spin_unlock(&root
->list_lock
);
1802 * insert an orphan item to track this unlinked/truncated file
1804 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1810 * We have done the truncate/delete so we can go ahead and remove the orphan
1811 * item for this particular inode.
1813 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1815 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1818 spin_lock(&root
->list_lock
);
1820 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1821 spin_unlock(&root
->list_lock
);
1825 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1827 spin_unlock(&root
->list_lock
);
1831 spin_unlock(&root
->list_lock
);
1833 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1839 * this cleans up any orphans that may be left on the list from the last use
1842 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1844 struct btrfs_path
*path
;
1845 struct extent_buffer
*leaf
;
1846 struct btrfs_item
*item
;
1847 struct btrfs_key key
, found_key
;
1848 struct btrfs_trans_handle
*trans
;
1849 struct inode
*inode
;
1850 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1852 path
= btrfs_alloc_path();
1857 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1858 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1859 key
.offset
= (u64
)-1;
1863 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1865 printk(KERN_ERR
"Error searching slot for orphan: %d"
1871 * if ret == 0 means we found what we were searching for, which
1872 * is weird, but possible, so only screw with path if we didnt
1873 * find the key and see if we have stuff that matches
1876 if (path
->slots
[0] == 0)
1881 /* pull out the item */
1882 leaf
= path
->nodes
[0];
1883 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1884 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1886 /* make sure the item matches what we want */
1887 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1889 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1892 /* release the path since we're done with it */
1893 btrfs_release_path(root
, path
);
1896 * this is where we are basically btrfs_lookup, without the
1897 * crossing root thing. we store the inode number in the
1898 * offset of the orphan item.
1900 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
1901 found_key
.offset
, root
);
1905 if (inode
->i_state
& I_NEW
) {
1906 BTRFS_I(inode
)->root
= root
;
1908 /* have to set the location manually */
1909 BTRFS_I(inode
)->location
.objectid
= inode
->i_ino
;
1910 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
1911 BTRFS_I(inode
)->location
.offset
= 0;
1913 btrfs_read_locked_inode(inode
);
1914 unlock_new_inode(inode
);
1918 * add this inode to the orphan list so btrfs_orphan_del does
1919 * the proper thing when we hit it
1921 spin_lock(&root
->list_lock
);
1922 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1923 spin_unlock(&root
->list_lock
);
1926 * if this is a bad inode, means we actually succeeded in
1927 * removing the inode, but not the orphan record, which means
1928 * we need to manually delete the orphan since iput will just
1929 * do a destroy_inode
1931 if (is_bad_inode(inode
)) {
1932 trans
= btrfs_start_transaction(root
, 1);
1933 btrfs_orphan_del(trans
, inode
);
1934 btrfs_end_transaction(trans
, root
);
1939 /* if we have links, this was a truncate, lets do that */
1940 if (inode
->i_nlink
) {
1942 btrfs_truncate(inode
);
1947 /* this will do delete_inode and everything for us */
1952 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
1954 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
1956 btrfs_free_path(path
);
1960 * read an inode from the btree into the in-memory inode
1962 void btrfs_read_locked_inode(struct inode
*inode
)
1964 struct btrfs_path
*path
;
1965 struct extent_buffer
*leaf
;
1966 struct btrfs_inode_item
*inode_item
;
1967 struct btrfs_timespec
*tspec
;
1968 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1969 struct btrfs_key location
;
1970 u64 alloc_group_block
;
1974 path
= btrfs_alloc_path();
1976 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
1978 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
1982 leaf
= path
->nodes
[0];
1983 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1984 struct btrfs_inode_item
);
1986 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
1987 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
1988 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
1989 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
1990 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
1992 tspec
= btrfs_inode_atime(inode_item
);
1993 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1994 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1996 tspec
= btrfs_inode_mtime(inode_item
);
1997 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1998 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2000 tspec
= btrfs_inode_ctime(inode_item
);
2001 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2002 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2004 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2005 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2006 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2007 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2009 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2011 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2012 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2014 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2015 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2016 alloc_group_block
, 0);
2017 btrfs_free_path(path
);
2020 switch (inode
->i_mode
& S_IFMT
) {
2022 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2023 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2024 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2025 inode
->i_fop
= &btrfs_file_operations
;
2026 inode
->i_op
= &btrfs_file_inode_operations
;
2029 inode
->i_fop
= &btrfs_dir_file_operations
;
2030 if (root
== root
->fs_info
->tree_root
)
2031 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2033 inode
->i_op
= &btrfs_dir_inode_operations
;
2036 inode
->i_op
= &btrfs_symlink_inode_operations
;
2037 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2038 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2041 init_special_inode(inode
, inode
->i_mode
, rdev
);
2047 btrfs_free_path(path
);
2048 make_bad_inode(inode
);
2052 * given a leaf and an inode, copy the inode fields into the leaf
2054 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2055 struct extent_buffer
*leaf
,
2056 struct btrfs_inode_item
*item
,
2057 struct inode
*inode
)
2059 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2060 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2061 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2062 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2063 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2065 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2066 inode
->i_atime
.tv_sec
);
2067 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2068 inode
->i_atime
.tv_nsec
);
2070 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2071 inode
->i_mtime
.tv_sec
);
2072 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2073 inode
->i_mtime
.tv_nsec
);
2075 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2076 inode
->i_ctime
.tv_sec
);
2077 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2078 inode
->i_ctime
.tv_nsec
);
2080 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2081 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2082 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2083 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2084 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2085 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2086 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2090 * copy everything in the in-memory inode into the btree.
2092 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2093 struct btrfs_root
*root
, struct inode
*inode
)
2095 struct btrfs_inode_item
*inode_item
;
2096 struct btrfs_path
*path
;
2097 struct extent_buffer
*leaf
;
2100 path
= btrfs_alloc_path();
2102 ret
= btrfs_lookup_inode(trans
, root
, path
,
2103 &BTRFS_I(inode
)->location
, 1);
2110 leaf
= path
->nodes
[0];
2111 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2112 struct btrfs_inode_item
);
2114 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2115 btrfs_mark_buffer_dirty(leaf
);
2116 btrfs_set_inode_last_trans(trans
, inode
);
2119 btrfs_free_path(path
);
2125 * unlink helper that gets used here in inode.c and in the tree logging
2126 * recovery code. It remove a link in a directory with a given name, and
2127 * also drops the back refs in the inode to the directory
2129 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2130 struct btrfs_root
*root
,
2131 struct inode
*dir
, struct inode
*inode
,
2132 const char *name
, int name_len
)
2134 struct btrfs_path
*path
;
2136 struct extent_buffer
*leaf
;
2137 struct btrfs_dir_item
*di
;
2138 struct btrfs_key key
;
2141 path
= btrfs_alloc_path();
2147 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2148 name
, name_len
, -1);
2157 leaf
= path
->nodes
[0];
2158 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2159 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2162 btrfs_release_path(root
, path
);
2164 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2166 dir
->i_ino
, &index
);
2168 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2169 "inode %lu parent %lu\n", name_len
, name
,
2170 inode
->i_ino
, dir
->i_ino
);
2174 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2175 index
, name
, name_len
, -1);
2184 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2185 btrfs_release_path(root
, path
);
2187 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2189 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2191 BTRFS_I(dir
)->log_dirty_trans
= trans
->transid
;
2193 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2197 btrfs_free_path(path
);
2201 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2202 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2203 btrfs_update_inode(trans
, root
, dir
);
2204 btrfs_drop_nlink(inode
);
2205 ret
= btrfs_update_inode(trans
, root
, inode
);
2206 dir
->i_sb
->s_dirt
= 1;
2211 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2213 struct btrfs_root
*root
;
2214 struct btrfs_trans_handle
*trans
;
2215 struct inode
*inode
= dentry
->d_inode
;
2217 unsigned long nr
= 0;
2219 root
= BTRFS_I(dir
)->root
;
2221 ret
= btrfs_check_free_space(root
, 1, 1);
2225 trans
= btrfs_start_transaction(root
, 1);
2227 btrfs_set_trans_block_group(trans
, dir
);
2228 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2229 dentry
->d_name
.name
, dentry
->d_name
.len
);
2231 if (inode
->i_nlink
== 0)
2232 ret
= btrfs_orphan_add(trans
, inode
);
2234 nr
= trans
->blocks_used
;
2236 btrfs_end_transaction_throttle(trans
, root
);
2238 btrfs_btree_balance_dirty(root
, nr
);
2242 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2244 struct inode
*inode
= dentry
->d_inode
;
2247 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2248 struct btrfs_trans_handle
*trans
;
2249 unsigned long nr
= 0;
2252 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2253 * the root of a subvolume or snapshot
2255 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2256 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
2260 ret
= btrfs_check_free_space(root
, 1, 1);
2264 trans
= btrfs_start_transaction(root
, 1);
2265 btrfs_set_trans_block_group(trans
, dir
);
2267 err
= btrfs_orphan_add(trans
, inode
);
2271 /* now the directory is empty */
2272 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2273 dentry
->d_name
.name
, dentry
->d_name
.len
);
2275 btrfs_i_size_write(inode
, 0);
2278 nr
= trans
->blocks_used
;
2279 ret
= btrfs_end_transaction_throttle(trans
, root
);
2281 btrfs_btree_balance_dirty(root
, nr
);
2290 * when truncating bytes in a file, it is possible to avoid reading
2291 * the leaves that contain only checksum items. This can be the
2292 * majority of the IO required to delete a large file, but it must
2293 * be done carefully.
2295 * The keys in the level just above the leaves are checked to make sure
2296 * the lowest key in a given leaf is a csum key, and starts at an offset
2297 * after the new size.
2299 * Then the key for the next leaf is checked to make sure it also has
2300 * a checksum item for the same file. If it does, we know our target leaf
2301 * contains only checksum items, and it can be safely freed without reading
2304 * This is just an optimization targeted at large files. It may do
2305 * nothing. It will return 0 unless things went badly.
2307 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
2308 struct btrfs_root
*root
,
2309 struct btrfs_path
*path
,
2310 struct inode
*inode
, u64 new_size
)
2312 struct btrfs_key key
;
2315 struct btrfs_key found_key
;
2316 struct btrfs_key other_key
;
2317 struct btrfs_leaf_ref
*ref
;
2321 path
->lowest_level
= 1;
2322 key
.objectid
= inode
->i_ino
;
2323 key
.type
= BTRFS_CSUM_ITEM_KEY
;
2324 key
.offset
= new_size
;
2326 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2330 if (path
->nodes
[1] == NULL
) {
2335 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
2336 nritems
= btrfs_header_nritems(path
->nodes
[1]);
2341 if (path
->slots
[1] >= nritems
)
2344 /* did we find a key greater than anything we want to delete? */
2345 if (found_key
.objectid
> inode
->i_ino
||
2346 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
2349 /* we check the next key in the node to make sure the leave contains
2350 * only checksum items. This comparison doesn't work if our
2351 * leaf is the last one in the node
2353 if (path
->slots
[1] + 1 >= nritems
) {
2355 /* search forward from the last key in the node, this
2356 * will bring us into the next node in the tree
2358 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
2360 /* unlikely, but we inc below, so check to be safe */
2361 if (found_key
.offset
== (u64
)-1)
2364 /* search_forward needs a path with locks held, do the
2365 * search again for the original key. It is possible
2366 * this will race with a balance and return a path that
2367 * we could modify, but this drop is just an optimization
2368 * and is allowed to miss some leaves.
2370 btrfs_release_path(root
, path
);
2373 /* setup a max key for search_forward */
2374 other_key
.offset
= (u64
)-1;
2375 other_key
.type
= key
.type
;
2376 other_key
.objectid
= key
.objectid
;
2378 path
->keep_locks
= 1;
2379 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
2381 path
->keep_locks
= 0;
2382 if (ret
|| found_key
.objectid
!= key
.objectid
||
2383 found_key
.type
!= key
.type
) {
2388 key
.offset
= found_key
.offset
;
2389 btrfs_release_path(root
, path
);
2394 /* we know there's one more slot after us in the tree,
2395 * read that key so we can verify it is also a checksum item
2397 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
2399 if (found_key
.objectid
< inode
->i_ino
)
2402 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
2406 * if the key for the next leaf isn't a csum key from this objectid,
2407 * we can't be sure there aren't good items inside this leaf.
2410 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
2413 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
2414 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
2416 * it is safe to delete this leaf, it contains only
2417 * csum items from this inode at an offset >= new_size
2419 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
2422 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
2423 ref
= btrfs_alloc_leaf_ref(root
, 0);
2425 ref
->root_gen
= root
->root_key
.offset
;
2426 ref
->bytenr
= leaf_start
;
2428 ref
->generation
= leaf_gen
;
2431 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
2433 btrfs_free_leaf_ref(root
, ref
);
2439 btrfs_release_path(root
, path
);
2441 if (other_key
.objectid
== inode
->i_ino
&&
2442 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
2443 key
.offset
= other_key
.offset
;
2449 /* fixup any changes we've made to the path */
2450 path
->lowest_level
= 0;
2451 path
->keep_locks
= 0;
2452 btrfs_release_path(root
, path
);
2459 * this can truncate away extent items, csum items and directory items.
2460 * It starts at a high offset and removes keys until it can't find
2461 * any higher than new_size
2463 * csum items that cross the new i_size are truncated to the new size
2466 * min_type is the minimum key type to truncate down to. If set to 0, this
2467 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2469 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2470 struct btrfs_root
*root
,
2471 struct inode
*inode
,
2472 u64 new_size
, u32 min_type
)
2475 struct btrfs_path
*path
;
2476 struct btrfs_key key
;
2477 struct btrfs_key found_key
;
2479 struct extent_buffer
*leaf
;
2480 struct btrfs_file_extent_item
*fi
;
2481 u64 extent_start
= 0;
2482 u64 extent_num_bytes
= 0;
2488 int pending_del_nr
= 0;
2489 int pending_del_slot
= 0;
2490 int extent_type
= -1;
2492 u64 mask
= root
->sectorsize
- 1;
2495 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2496 path
= btrfs_alloc_path();
2500 /* FIXME, add redo link to tree so we don't leak on crash */
2501 key
.objectid
= inode
->i_ino
;
2502 key
.offset
= (u64
)-1;
2505 btrfs_init_path(path
);
2508 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2513 /* there are no items in the tree for us to truncate, we're
2516 if (path
->slots
[0] == 0) {
2525 leaf
= path
->nodes
[0];
2526 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2527 found_type
= btrfs_key_type(&found_key
);
2530 if (found_key
.objectid
!= inode
->i_ino
)
2533 if (found_type
< min_type
)
2536 item_end
= found_key
.offset
;
2537 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2538 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2539 struct btrfs_file_extent_item
);
2540 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2541 encoding
= btrfs_file_extent_compression(leaf
, fi
);
2542 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
2543 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
2545 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2547 btrfs_file_extent_num_bytes(leaf
, fi
);
2548 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2549 item_end
+= btrfs_file_extent_inline_len(leaf
,
2554 if (item_end
< new_size
) {
2555 if (found_type
== BTRFS_DIR_ITEM_KEY
)
2556 found_type
= BTRFS_INODE_ITEM_KEY
;
2557 else if (found_type
== BTRFS_EXTENT_ITEM_KEY
)
2558 found_type
= BTRFS_EXTENT_DATA_KEY
;
2559 else if (found_type
== BTRFS_EXTENT_DATA_KEY
)
2560 found_type
= BTRFS_XATTR_ITEM_KEY
;
2561 else if (found_type
== BTRFS_XATTR_ITEM_KEY
)
2562 found_type
= BTRFS_INODE_REF_KEY
;
2563 else if (found_type
)
2567 btrfs_set_key_type(&key
, found_type
);
2570 if (found_key
.offset
>= new_size
)
2576 /* FIXME, shrink the extent if the ref count is only 1 */
2577 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2580 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2582 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2583 if (!del_item
&& !encoding
) {
2584 u64 orig_num_bytes
=
2585 btrfs_file_extent_num_bytes(leaf
, fi
);
2586 extent_num_bytes
= new_size
-
2587 found_key
.offset
+ root
->sectorsize
- 1;
2588 extent_num_bytes
= extent_num_bytes
&
2589 ~((u64
)root
->sectorsize
- 1);
2590 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2592 num_dec
= (orig_num_bytes
-
2594 if (root
->ref_cows
&& extent_start
!= 0)
2595 inode_sub_bytes(inode
, num_dec
);
2596 btrfs_mark_buffer_dirty(leaf
);
2599 btrfs_file_extent_disk_num_bytes(leaf
,
2601 /* FIXME blocksize != 4096 */
2602 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2603 if (extent_start
!= 0) {
2606 inode_sub_bytes(inode
, num_dec
);
2608 root_gen
= btrfs_header_generation(leaf
);
2609 root_owner
= btrfs_header_owner(leaf
);
2611 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2613 * we can't truncate inline items that have had
2617 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2618 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2619 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2620 u32 size
= new_size
- found_key
.offset
;
2622 if (root
->ref_cows
) {
2623 inode_sub_bytes(inode
, item_end
+ 1 -
2627 btrfs_file_extent_calc_inline_size(size
);
2628 ret
= btrfs_truncate_item(trans
, root
, path
,
2631 } else if (root
->ref_cows
) {
2632 inode_sub_bytes(inode
, item_end
+ 1 -
2638 if (!pending_del_nr
) {
2639 /* no pending yet, add ourselves */
2640 pending_del_slot
= path
->slots
[0];
2642 } else if (pending_del_nr
&&
2643 path
->slots
[0] + 1 == pending_del_slot
) {
2644 /* hop on the pending chunk */
2646 pending_del_slot
= path
->slots
[0];
2654 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2656 leaf
->start
, root_owner
,
2657 root_gen
, inode
->i_ino
, 0);
2661 if (path
->slots
[0] == 0) {
2664 btrfs_release_path(root
, path
);
2669 if (pending_del_nr
&&
2670 path
->slots
[0] + 1 != pending_del_slot
) {
2671 struct btrfs_key debug
;
2673 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2675 ret
= btrfs_del_items(trans
, root
, path
,
2680 btrfs_release_path(root
, path
);
2686 if (pending_del_nr
) {
2687 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2690 btrfs_free_path(path
);
2691 inode
->i_sb
->s_dirt
= 1;
2696 * taken from block_truncate_page, but does cow as it zeros out
2697 * any bytes left in the last page in the file.
2699 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2701 struct inode
*inode
= mapping
->host
;
2702 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2703 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2704 struct btrfs_ordered_extent
*ordered
;
2706 u32 blocksize
= root
->sectorsize
;
2707 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2708 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2714 if ((offset
& (blocksize
- 1)) == 0)
2719 page
= grab_cache_page(mapping
, index
);
2723 page_start
= page_offset(page
);
2724 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2726 if (!PageUptodate(page
)) {
2727 ret
= btrfs_readpage(NULL
, page
);
2729 if (page
->mapping
!= mapping
) {
2731 page_cache_release(page
);
2734 if (!PageUptodate(page
)) {
2739 wait_on_page_writeback(page
);
2741 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2742 set_page_extent_mapped(page
);
2744 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2746 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2748 page_cache_release(page
);
2749 btrfs_start_ordered_extent(inode
, ordered
, 1);
2750 btrfs_put_ordered_extent(ordered
);
2754 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2756 if (offset
!= PAGE_CACHE_SIZE
) {
2758 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2759 flush_dcache_page(page
);
2762 ClearPageChecked(page
);
2763 set_page_dirty(page
);
2764 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2768 page_cache_release(page
);
2773 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
2775 struct btrfs_trans_handle
*trans
;
2776 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2777 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2778 struct extent_map
*em
;
2779 u64 mask
= root
->sectorsize
- 1;
2780 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2781 u64 block_end
= (size
+ mask
) & ~mask
;
2787 if (size
<= hole_start
)
2790 err
= btrfs_check_free_space(root
, 1, 0);
2794 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2797 struct btrfs_ordered_extent
*ordered
;
2798 btrfs_wait_ordered_range(inode
, hole_start
,
2799 block_end
- hole_start
);
2800 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2801 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2804 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2805 btrfs_put_ordered_extent(ordered
);
2808 trans
= btrfs_start_transaction(root
, 1);
2809 btrfs_set_trans_block_group(trans
, inode
);
2811 cur_offset
= hole_start
;
2813 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2814 block_end
- cur_offset
, 0);
2815 BUG_ON(IS_ERR(em
) || !em
);
2816 last_byte
= min(extent_map_end(em
), block_end
);
2817 last_byte
= (last_byte
+ mask
) & ~mask
;
2818 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2820 hole_size
= last_byte
- cur_offset
;
2821 err
= btrfs_drop_extents(trans
, root
, inode
,
2823 cur_offset
+ hole_size
,
2824 cur_offset
, &hint_byte
);
2827 err
= btrfs_insert_file_extent(trans
, root
,
2828 inode
->i_ino
, cur_offset
, 0,
2829 0, hole_size
, 0, hole_size
,
2831 btrfs_drop_extent_cache(inode
, hole_start
,
2834 free_extent_map(em
);
2835 cur_offset
= last_byte
;
2836 if (err
|| cur_offset
>= block_end
)
2840 btrfs_end_transaction(trans
, root
);
2841 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2845 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2847 struct inode
*inode
= dentry
->d_inode
;
2850 err
= inode_change_ok(inode
, attr
);
2854 if (S_ISREG(inode
->i_mode
) &&
2855 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
> inode
->i_size
) {
2856 err
= btrfs_cont_expand(inode
, attr
->ia_size
);
2861 err
= inode_setattr(inode
, attr
);
2863 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2864 err
= btrfs_acl_chmod(inode
);
2868 void btrfs_delete_inode(struct inode
*inode
)
2870 struct btrfs_trans_handle
*trans
;
2871 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2875 truncate_inode_pages(&inode
->i_data
, 0);
2876 if (is_bad_inode(inode
)) {
2877 btrfs_orphan_del(NULL
, inode
);
2880 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
2882 btrfs_i_size_write(inode
, 0);
2883 trans
= btrfs_join_transaction(root
, 1);
2885 btrfs_set_trans_block_group(trans
, inode
);
2886 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
2888 btrfs_orphan_del(NULL
, inode
);
2889 goto no_delete_lock
;
2892 btrfs_orphan_del(trans
, inode
);
2894 nr
= trans
->blocks_used
;
2897 btrfs_end_transaction(trans
, root
);
2898 btrfs_btree_balance_dirty(root
, nr
);
2902 nr
= trans
->blocks_used
;
2903 btrfs_end_transaction(trans
, root
);
2904 btrfs_btree_balance_dirty(root
, nr
);
2910 * this returns the key found in the dir entry in the location pointer.
2911 * If no dir entries were found, location->objectid is 0.
2913 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
2914 struct btrfs_key
*location
)
2916 const char *name
= dentry
->d_name
.name
;
2917 int namelen
= dentry
->d_name
.len
;
2918 struct btrfs_dir_item
*di
;
2919 struct btrfs_path
*path
;
2920 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2923 path
= btrfs_alloc_path();
2926 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
2931 if (!di
|| IS_ERR(di
))
2934 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
2936 btrfs_free_path(path
);
2939 location
->objectid
= 0;
2944 * when we hit a tree root in a directory, the btrfs part of the inode
2945 * needs to be changed to reflect the root directory of the tree root. This
2946 * is kind of like crossing a mount point.
2948 static int fixup_tree_root_location(struct btrfs_root
*root
,
2949 struct btrfs_key
*location
,
2950 struct btrfs_root
**sub_root
,
2951 struct dentry
*dentry
)
2953 struct btrfs_root_item
*ri
;
2955 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
2957 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
2960 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
2961 dentry
->d_name
.name
,
2962 dentry
->d_name
.len
);
2963 if (IS_ERR(*sub_root
))
2964 return PTR_ERR(*sub_root
);
2966 ri
= &(*sub_root
)->root_item
;
2967 location
->objectid
= btrfs_root_dirid(ri
);
2968 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2969 location
->offset
= 0;
2974 static noinline
void init_btrfs_i(struct inode
*inode
)
2976 struct btrfs_inode
*bi
= BTRFS_I(inode
);
2979 bi
->i_default_acl
= NULL
;
2984 bi
->logged_trans
= 0;
2985 bi
->delalloc_bytes
= 0;
2986 bi
->disk_i_size
= 0;
2988 bi
->index_cnt
= (u64
)-1;
2989 bi
->log_dirty_trans
= 0;
2990 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
2991 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
2992 inode
->i_mapping
, GFP_NOFS
);
2993 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
2994 inode
->i_mapping
, GFP_NOFS
);
2995 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
2996 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
2997 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
2998 mutex_init(&BTRFS_I(inode
)->log_mutex
);
3001 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3003 struct btrfs_iget_args
*args
= p
;
3004 inode
->i_ino
= args
->ino
;
3005 init_btrfs_i(inode
);
3006 BTRFS_I(inode
)->root
= args
->root
;
3010 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3012 struct btrfs_iget_args
*args
= opaque
;
3013 return args
->ino
== inode
->i_ino
&&
3014 args
->root
== BTRFS_I(inode
)->root
;
3017 struct inode
*btrfs_ilookup(struct super_block
*s
, u64 objectid
,
3018 struct btrfs_root
*root
, int wait
)
3020 struct inode
*inode
;
3021 struct btrfs_iget_args args
;
3022 args
.ino
= objectid
;
3026 inode
= ilookup5(s
, objectid
, btrfs_find_actor
,
3029 inode
= ilookup5_nowait(s
, objectid
, btrfs_find_actor
,
3035 struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 objectid
,
3036 struct btrfs_root
*root
)
3038 struct inode
*inode
;
3039 struct btrfs_iget_args args
;
3040 args
.ino
= objectid
;
3043 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3044 btrfs_init_locked_inode
,
3049 /* Get an inode object given its location and corresponding root.
3050 * Returns in *is_new if the inode was read from disk
3052 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3053 struct btrfs_root
*root
, int *is_new
)
3055 struct inode
*inode
;
3057 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3059 return ERR_PTR(-EACCES
);
3061 if (inode
->i_state
& I_NEW
) {
3062 BTRFS_I(inode
)->root
= root
;
3063 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3064 btrfs_read_locked_inode(inode
);
3065 unlock_new_inode(inode
);
3076 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3078 struct inode
*inode
;
3079 struct btrfs_inode
*bi
= BTRFS_I(dir
);
3080 struct btrfs_root
*root
= bi
->root
;
3081 struct btrfs_root
*sub_root
= root
;
3082 struct btrfs_key location
;
3085 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3086 return ERR_PTR(-ENAMETOOLONG
);
3088 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3091 return ERR_PTR(ret
);
3094 if (location
.objectid
) {
3095 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
3098 return ERR_PTR(ret
);
3100 return ERR_PTR(-ENOENT
);
3101 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, &new);
3103 return ERR_CAST(inode
);
3108 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
3109 struct nameidata
*nd
)
3111 struct inode
*inode
;
3113 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3114 return ERR_PTR(-ENAMETOOLONG
);
3116 inode
= btrfs_lookup_dentry(dir
, dentry
);
3118 return ERR_CAST(inode
);
3120 return d_splice_alias(inode
, dentry
);
3123 static unsigned char btrfs_filetype_table
[] = {
3124 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
3127 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
3130 struct inode
*inode
= filp
->f_dentry
->d_inode
;
3131 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3132 struct btrfs_item
*item
;
3133 struct btrfs_dir_item
*di
;
3134 struct btrfs_key key
;
3135 struct btrfs_key found_key
;
3136 struct btrfs_path
*path
;
3139 struct extent_buffer
*leaf
;
3142 unsigned char d_type
;
3147 int key_type
= BTRFS_DIR_INDEX_KEY
;
3152 /* FIXME, use a real flag for deciding about the key type */
3153 if (root
->fs_info
->tree_root
== root
)
3154 key_type
= BTRFS_DIR_ITEM_KEY
;
3156 /* special case for "." */
3157 if (filp
->f_pos
== 0) {
3158 over
= filldir(dirent
, ".", 1,
3165 /* special case for .., just use the back ref */
3166 if (filp
->f_pos
== 1) {
3167 u64 pino
= parent_ino(filp
->f_path
.dentry
);
3168 over
= filldir(dirent
, "..", 2,
3174 path
= btrfs_alloc_path();
3177 btrfs_set_key_type(&key
, key_type
);
3178 key
.offset
= filp
->f_pos
;
3179 key
.objectid
= inode
->i_ino
;
3181 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3187 leaf
= path
->nodes
[0];
3188 nritems
= btrfs_header_nritems(leaf
);
3189 slot
= path
->slots
[0];
3190 if (advance
|| slot
>= nritems
) {
3191 if (slot
>= nritems
- 1) {
3192 ret
= btrfs_next_leaf(root
, path
);
3195 leaf
= path
->nodes
[0];
3196 nritems
= btrfs_header_nritems(leaf
);
3197 slot
= path
->slots
[0];
3205 item
= btrfs_item_nr(leaf
, slot
);
3206 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3208 if (found_key
.objectid
!= key
.objectid
)
3210 if (btrfs_key_type(&found_key
) != key_type
)
3212 if (found_key
.offset
< filp
->f_pos
)
3215 filp
->f_pos
= found_key
.offset
;
3217 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
3219 di_total
= btrfs_item_size(leaf
, item
);
3221 while (di_cur
< di_total
) {
3222 struct btrfs_key location
;
3224 name_len
= btrfs_dir_name_len(leaf
, di
);
3225 if (name_len
<= sizeof(tmp_name
)) {
3226 name_ptr
= tmp_name
;
3228 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
3234 read_extent_buffer(leaf
, name_ptr
,
3235 (unsigned long)(di
+ 1), name_len
);
3237 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
3238 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
3240 /* is this a reference to our own snapshot? If so
3243 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
3244 location
.objectid
== root
->root_key
.objectid
) {
3248 over
= filldir(dirent
, name_ptr
, name_len
,
3249 found_key
.offset
, location
.objectid
,
3253 if (name_ptr
!= tmp_name
)
3258 di_len
= btrfs_dir_name_len(leaf
, di
) +
3259 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
3261 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
3265 /* Reached end of directory/root. Bump pos past the last item. */
3266 if (key_type
== BTRFS_DIR_INDEX_KEY
)
3267 filp
->f_pos
= INT_LIMIT(typeof(filp
->f_pos
));
3273 btrfs_free_path(path
);
3277 int btrfs_write_inode(struct inode
*inode
, int wait
)
3279 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3280 struct btrfs_trans_handle
*trans
;
3283 if (root
->fs_info
->btree_inode
== inode
)
3287 trans
= btrfs_join_transaction(root
, 1);
3288 btrfs_set_trans_block_group(trans
, inode
);
3289 ret
= btrfs_commit_transaction(trans
, root
);
3295 * This is somewhat expensive, updating the tree every time the
3296 * inode changes. But, it is most likely to find the inode in cache.
3297 * FIXME, needs more benchmarking...there are no reasons other than performance
3298 * to keep or drop this code.
3300 void btrfs_dirty_inode(struct inode
*inode
)
3302 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3303 struct btrfs_trans_handle
*trans
;
3305 trans
= btrfs_join_transaction(root
, 1);
3306 btrfs_set_trans_block_group(trans
, inode
);
3307 btrfs_update_inode(trans
, root
, inode
);
3308 btrfs_end_transaction(trans
, root
);
3312 * find the highest existing sequence number in a directory
3313 * and then set the in-memory index_cnt variable to reflect
3314 * free sequence numbers
3316 static int btrfs_set_inode_index_count(struct inode
*inode
)
3318 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3319 struct btrfs_key key
, found_key
;
3320 struct btrfs_path
*path
;
3321 struct extent_buffer
*leaf
;
3324 key
.objectid
= inode
->i_ino
;
3325 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
3326 key
.offset
= (u64
)-1;
3328 path
= btrfs_alloc_path();
3332 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3335 /* FIXME: we should be able to handle this */
3341 * MAGIC NUMBER EXPLANATION:
3342 * since we search a directory based on f_pos we have to start at 2
3343 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3344 * else has to start at 2
3346 if (path
->slots
[0] == 0) {
3347 BTRFS_I(inode
)->index_cnt
= 2;
3353 leaf
= path
->nodes
[0];
3354 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3356 if (found_key
.objectid
!= inode
->i_ino
||
3357 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
3358 BTRFS_I(inode
)->index_cnt
= 2;
3362 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
3364 btrfs_free_path(path
);
3369 * helper to find a free sequence number in a given directory. This current
3370 * code is very simple, later versions will do smarter things in the btree
3372 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
3376 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
3377 ret
= btrfs_set_inode_index_count(dir
);
3382 *index
= BTRFS_I(dir
)->index_cnt
;
3383 BTRFS_I(dir
)->index_cnt
++;
3388 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
3389 struct btrfs_root
*root
,
3391 const char *name
, int name_len
,
3392 u64 ref_objectid
, u64 objectid
,
3393 u64 alloc_hint
, int mode
, u64
*index
)
3395 struct inode
*inode
;
3396 struct btrfs_inode_item
*inode_item
;
3397 struct btrfs_key
*location
;
3398 struct btrfs_path
*path
;
3399 struct btrfs_inode_ref
*ref
;
3400 struct btrfs_key key
[2];
3406 path
= btrfs_alloc_path();
3409 inode
= new_inode(root
->fs_info
->sb
);
3411 return ERR_PTR(-ENOMEM
);
3414 ret
= btrfs_set_inode_index(dir
, index
);
3416 return ERR_PTR(ret
);
3419 * index_cnt is ignored for everything but a dir,
3420 * btrfs_get_inode_index_count has an explanation for the magic
3423 init_btrfs_i(inode
);
3424 BTRFS_I(inode
)->index_cnt
= 2;
3425 BTRFS_I(inode
)->root
= root
;
3426 BTRFS_I(inode
)->generation
= trans
->transid
;
3432 BTRFS_I(inode
)->block_group
=
3433 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
3434 if ((mode
& S_IFREG
)) {
3435 if (btrfs_test_opt(root
, NODATASUM
))
3436 btrfs_set_flag(inode
, NODATASUM
);
3437 if (btrfs_test_opt(root
, NODATACOW
))
3438 btrfs_set_flag(inode
, NODATACOW
);
3441 key
[0].objectid
= objectid
;
3442 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
3445 key
[1].objectid
= objectid
;
3446 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
3447 key
[1].offset
= ref_objectid
;
3449 sizes
[0] = sizeof(struct btrfs_inode_item
);
3450 sizes
[1] = name_len
+ sizeof(*ref
);
3452 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
3456 if (objectid
> root
->highest_inode
)
3457 root
->highest_inode
= objectid
;
3459 inode
->i_uid
= current_fsuid();
3460 inode
->i_gid
= current_fsgid();
3461 inode
->i_mode
= mode
;
3462 inode
->i_ino
= objectid
;
3463 inode_set_bytes(inode
, 0);
3464 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3465 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3466 struct btrfs_inode_item
);
3467 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
3469 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3470 struct btrfs_inode_ref
);
3471 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
3472 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
3473 ptr
= (unsigned long)(ref
+ 1);
3474 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
3476 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3477 btrfs_free_path(path
);
3479 location
= &BTRFS_I(inode
)->location
;
3480 location
->objectid
= objectid
;
3481 location
->offset
= 0;
3482 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3484 insert_inode_hash(inode
);
3488 BTRFS_I(dir
)->index_cnt
--;
3489 btrfs_free_path(path
);
3490 return ERR_PTR(ret
);
3493 static inline u8
btrfs_inode_type(struct inode
*inode
)
3495 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
3499 * utility function to add 'inode' into 'parent_inode' with
3500 * a give name and a given sequence number.
3501 * if 'add_backref' is true, also insert a backref from the
3502 * inode to the parent directory.
3504 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3505 struct inode
*parent_inode
, struct inode
*inode
,
3506 const char *name
, int name_len
, int add_backref
, u64 index
)
3509 struct btrfs_key key
;
3510 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3512 key
.objectid
= inode
->i_ino
;
3513 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3516 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3517 parent_inode
->i_ino
,
3518 &key
, btrfs_inode_type(inode
),
3522 ret
= btrfs_insert_inode_ref(trans
, root
,
3525 parent_inode
->i_ino
,
3528 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3530 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3531 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3536 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3537 struct dentry
*dentry
, struct inode
*inode
,
3538 int backref
, u64 index
)
3540 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3541 inode
, dentry
->d_name
.name
,
3542 dentry
->d_name
.len
, backref
, index
);
3544 d_instantiate(dentry
, inode
);
3552 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3553 int mode
, dev_t rdev
)
3555 struct btrfs_trans_handle
*trans
;
3556 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3557 struct inode
*inode
= NULL
;
3561 unsigned long nr
= 0;
3564 if (!new_valid_dev(rdev
))
3567 err
= btrfs_check_free_space(root
, 1, 0);
3571 trans
= btrfs_start_transaction(root
, 1);
3572 btrfs_set_trans_block_group(trans
, dir
);
3574 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3580 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3582 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3583 BTRFS_I(dir
)->block_group
, mode
, &index
);
3584 err
= PTR_ERR(inode
);
3588 err
= btrfs_init_acl(inode
, dir
);
3594 btrfs_set_trans_block_group(trans
, inode
);
3595 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3599 inode
->i_op
= &btrfs_special_inode_operations
;
3600 init_special_inode(inode
, inode
->i_mode
, rdev
);
3601 btrfs_update_inode(trans
, root
, inode
);
3603 dir
->i_sb
->s_dirt
= 1;
3604 btrfs_update_inode_block_group(trans
, inode
);
3605 btrfs_update_inode_block_group(trans
, dir
);
3607 nr
= trans
->blocks_used
;
3608 btrfs_end_transaction_throttle(trans
, root
);
3611 inode_dec_link_count(inode
);
3614 btrfs_btree_balance_dirty(root
, nr
);
3618 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3619 int mode
, struct nameidata
*nd
)
3621 struct btrfs_trans_handle
*trans
;
3622 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3623 struct inode
*inode
= NULL
;
3626 unsigned long nr
= 0;
3630 err
= btrfs_check_free_space(root
, 1, 0);
3633 trans
= btrfs_start_transaction(root
, 1);
3634 btrfs_set_trans_block_group(trans
, dir
);
3636 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3642 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3644 dentry
->d_parent
->d_inode
->i_ino
,
3645 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3647 err
= PTR_ERR(inode
);
3651 err
= btrfs_init_acl(inode
, dir
);
3657 btrfs_set_trans_block_group(trans
, inode
);
3658 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3662 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3663 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3664 inode
->i_fop
= &btrfs_file_operations
;
3665 inode
->i_op
= &btrfs_file_inode_operations
;
3666 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3668 dir
->i_sb
->s_dirt
= 1;
3669 btrfs_update_inode_block_group(trans
, inode
);
3670 btrfs_update_inode_block_group(trans
, dir
);
3672 nr
= trans
->blocks_used
;
3673 btrfs_end_transaction_throttle(trans
, root
);
3676 inode_dec_link_count(inode
);
3679 btrfs_btree_balance_dirty(root
, nr
);
3683 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3684 struct dentry
*dentry
)
3686 struct btrfs_trans_handle
*trans
;
3687 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3688 struct inode
*inode
= old_dentry
->d_inode
;
3690 unsigned long nr
= 0;
3694 if (inode
->i_nlink
== 0)
3697 btrfs_inc_nlink(inode
);
3698 err
= btrfs_check_free_space(root
, 1, 0);
3701 err
= btrfs_set_inode_index(dir
, &index
);
3705 trans
= btrfs_start_transaction(root
, 1);
3707 btrfs_set_trans_block_group(trans
, dir
);
3708 atomic_inc(&inode
->i_count
);
3710 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3715 dir
->i_sb
->s_dirt
= 1;
3716 btrfs_update_inode_block_group(trans
, dir
);
3717 err
= btrfs_update_inode(trans
, root
, inode
);
3722 nr
= trans
->blocks_used
;
3723 btrfs_end_transaction_throttle(trans
, root
);
3726 inode_dec_link_count(inode
);
3729 btrfs_btree_balance_dirty(root
, nr
);
3733 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3735 struct inode
*inode
= NULL
;
3736 struct btrfs_trans_handle
*trans
;
3737 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3739 int drop_on_err
= 0;
3742 unsigned long nr
= 1;
3744 err
= btrfs_check_free_space(root
, 1, 0);
3748 trans
= btrfs_start_transaction(root
, 1);
3749 btrfs_set_trans_block_group(trans
, dir
);
3751 if (IS_ERR(trans
)) {
3752 err
= PTR_ERR(trans
);
3756 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3762 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3764 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3765 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3767 if (IS_ERR(inode
)) {
3768 err
= PTR_ERR(inode
);
3774 err
= btrfs_init_acl(inode
, dir
);
3778 inode
->i_op
= &btrfs_dir_inode_operations
;
3779 inode
->i_fop
= &btrfs_dir_file_operations
;
3780 btrfs_set_trans_block_group(trans
, inode
);
3782 btrfs_i_size_write(inode
, 0);
3783 err
= btrfs_update_inode(trans
, root
, inode
);
3787 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3788 inode
, dentry
->d_name
.name
,
3789 dentry
->d_name
.len
, 0, index
);
3793 d_instantiate(dentry
, inode
);
3795 dir
->i_sb
->s_dirt
= 1;
3796 btrfs_update_inode_block_group(trans
, inode
);
3797 btrfs_update_inode_block_group(trans
, dir
);
3800 nr
= trans
->blocks_used
;
3801 btrfs_end_transaction_throttle(trans
, root
);
3806 btrfs_btree_balance_dirty(root
, nr
);
3810 /* helper for btfs_get_extent. Given an existing extent in the tree,
3811 * and an extent that you want to insert, deal with overlap and insert
3812 * the new extent into the tree.
3814 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3815 struct extent_map
*existing
,
3816 struct extent_map
*em
,
3817 u64 map_start
, u64 map_len
)
3821 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3822 start_diff
= map_start
- em
->start
;
3823 em
->start
= map_start
;
3825 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3826 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3827 em
->block_start
+= start_diff
;
3828 em
->block_len
-= start_diff
;
3830 return add_extent_mapping(em_tree
, em
);
3833 static noinline
int uncompress_inline(struct btrfs_path
*path
,
3834 struct inode
*inode
, struct page
*page
,
3835 size_t pg_offset
, u64 extent_offset
,
3836 struct btrfs_file_extent_item
*item
)
3839 struct extent_buffer
*leaf
= path
->nodes
[0];
3842 unsigned long inline_size
;
3845 WARN_ON(pg_offset
!= 0);
3846 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
3847 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
3848 btrfs_item_nr(leaf
, path
->slots
[0]));
3849 tmp
= kmalloc(inline_size
, GFP_NOFS
);
3850 ptr
= btrfs_file_extent_inline_start(item
);
3852 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
3854 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
3855 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
3856 inline_size
, max_size
);
3858 char *kaddr
= kmap_atomic(page
, KM_USER0
);
3859 unsigned long copy_size
= min_t(u64
,
3860 PAGE_CACHE_SIZE
- pg_offset
,
3861 max_size
- extent_offset
);
3862 memset(kaddr
+ pg_offset
, 0, copy_size
);
3863 kunmap_atomic(kaddr
, KM_USER0
);
3870 * a bit scary, this does extent mapping from logical file offset to the disk.
3871 * the ugly parts come from merging extents from the disk with the in-ram
3872 * representation. This gets more complex because of the data=ordered code,
3873 * where the in-ram extents might be locked pending data=ordered completion.
3875 * This also copies inline extents directly into the page.
3878 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
3879 size_t pg_offset
, u64 start
, u64 len
,
3885 u64 extent_start
= 0;
3887 u64 objectid
= inode
->i_ino
;
3889 struct btrfs_path
*path
= NULL
;
3890 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3891 struct btrfs_file_extent_item
*item
;
3892 struct extent_buffer
*leaf
;
3893 struct btrfs_key found_key
;
3894 struct extent_map
*em
= NULL
;
3895 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3896 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3897 struct btrfs_trans_handle
*trans
= NULL
;
3901 spin_lock(&em_tree
->lock
);
3902 em
= lookup_extent_mapping(em_tree
, start
, len
);
3904 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3905 spin_unlock(&em_tree
->lock
);
3908 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
3909 free_extent_map(em
);
3910 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
3911 free_extent_map(em
);
3915 em
= alloc_extent_map(GFP_NOFS
);
3920 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3921 em
->start
= EXTENT_MAP_HOLE
;
3922 em
->orig_start
= EXTENT_MAP_HOLE
;
3924 em
->block_len
= (u64
)-1;
3927 path
= btrfs_alloc_path();
3931 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3932 objectid
, start
, trans
!= NULL
);
3939 if (path
->slots
[0] == 0)
3944 leaf
= path
->nodes
[0];
3945 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3946 struct btrfs_file_extent_item
);
3947 /* are we inside the extent that was found? */
3948 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3949 found_type
= btrfs_key_type(&found_key
);
3950 if (found_key
.objectid
!= objectid
||
3951 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3955 found_type
= btrfs_file_extent_type(leaf
, item
);
3956 extent_start
= found_key
.offset
;
3957 compressed
= btrfs_file_extent_compression(leaf
, item
);
3958 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3959 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3960 extent_end
= extent_start
+
3961 btrfs_file_extent_num_bytes(leaf
, item
);
3962 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3964 size
= btrfs_file_extent_inline_len(leaf
, item
);
3965 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
3966 ~((u64
)root
->sectorsize
- 1);
3969 if (start
>= extent_end
) {
3971 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3972 ret
= btrfs_next_leaf(root
, path
);
3979 leaf
= path
->nodes
[0];
3981 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3982 if (found_key
.objectid
!= objectid
||
3983 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3985 if (start
+ len
<= found_key
.offset
)
3988 em
->len
= found_key
.offset
- start
;
3992 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3993 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3994 em
->start
= extent_start
;
3995 em
->len
= extent_end
- extent_start
;
3996 em
->orig_start
= extent_start
-
3997 btrfs_file_extent_offset(leaf
, item
);
3998 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
4000 em
->block_start
= EXTENT_MAP_HOLE
;
4004 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4005 em
->block_start
= bytenr
;
4006 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
4009 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
4010 em
->block_start
= bytenr
;
4011 em
->block_len
= em
->len
;
4012 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
4013 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
4016 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4020 size_t extent_offset
;
4023 em
->block_start
= EXTENT_MAP_INLINE
;
4024 if (!page
|| create
) {
4025 em
->start
= extent_start
;
4026 em
->len
= extent_end
- extent_start
;
4030 size
= btrfs_file_extent_inline_len(leaf
, item
);
4031 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
4032 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
4033 size
- extent_offset
);
4034 em
->start
= extent_start
+ extent_offset
;
4035 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
4036 ~((u64
)root
->sectorsize
- 1);
4037 em
->orig_start
= EXTENT_MAP_INLINE
;
4039 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4040 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
4041 if (create
== 0 && !PageUptodate(page
)) {
4042 if (btrfs_file_extent_compression(leaf
, item
) ==
4043 BTRFS_COMPRESS_ZLIB
) {
4044 ret
= uncompress_inline(path
, inode
, page
,
4046 extent_offset
, item
);
4050 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4054 flush_dcache_page(page
);
4055 } else if (create
&& PageUptodate(page
)) {
4058 free_extent_map(em
);
4060 btrfs_release_path(root
, path
);
4061 trans
= btrfs_join_transaction(root
, 1);
4065 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4068 btrfs_mark_buffer_dirty(leaf
);
4070 set_extent_uptodate(io_tree
, em
->start
,
4071 extent_map_end(em
) - 1, GFP_NOFS
);
4074 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
4081 em
->block_start
= EXTENT_MAP_HOLE
;
4082 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
4084 btrfs_release_path(root
, path
);
4085 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
4086 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
4087 "[%llu %llu]\n", (unsigned long long)em
->start
,
4088 (unsigned long long)em
->len
,
4089 (unsigned long long)start
,
4090 (unsigned long long)len
);
4096 spin_lock(&em_tree
->lock
);
4097 ret
= add_extent_mapping(em_tree
, em
);
4098 /* it is possible that someone inserted the extent into the tree
4099 * while we had the lock dropped. It is also possible that
4100 * an overlapping map exists in the tree
4102 if (ret
== -EEXIST
) {
4103 struct extent_map
*existing
;
4107 existing
= lookup_extent_mapping(em_tree
, start
, len
);
4108 if (existing
&& (existing
->start
> start
||
4109 existing
->start
+ existing
->len
<= start
)) {
4110 free_extent_map(existing
);
4114 existing
= lookup_extent_mapping(em_tree
, em
->start
,
4117 err
= merge_extent_mapping(em_tree
, existing
,
4120 free_extent_map(existing
);
4122 free_extent_map(em
);
4127 free_extent_map(em
);
4131 free_extent_map(em
);
4136 spin_unlock(&em_tree
->lock
);
4139 btrfs_free_path(path
);
4141 ret
= btrfs_end_transaction(trans
, root
);
4146 free_extent_map(em
);
4148 return ERR_PTR(err
);
4153 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
4154 const struct iovec
*iov
, loff_t offset
,
4155 unsigned long nr_segs
)
4160 static sector_t
btrfs_bmap(struct address_space
*mapping
, sector_t iblock
)
4162 return extent_bmap(mapping
, iblock
, btrfs_get_extent
);
4165 int btrfs_readpage(struct file
*file
, struct page
*page
)
4167 struct extent_io_tree
*tree
;
4168 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4169 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
4172 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
4174 struct extent_io_tree
*tree
;
4177 if (current
->flags
& PF_MEMALLOC
) {
4178 redirty_page_for_writepage(wbc
, page
);
4182 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4183 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
4186 int btrfs_writepages(struct address_space
*mapping
,
4187 struct writeback_control
*wbc
)
4189 struct extent_io_tree
*tree
;
4191 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4192 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
4196 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
4197 struct list_head
*pages
, unsigned nr_pages
)
4199 struct extent_io_tree
*tree
;
4200 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4201 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
4204 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4206 struct extent_io_tree
*tree
;
4207 struct extent_map_tree
*map
;
4210 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4211 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
4212 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
4214 ClearPagePrivate(page
);
4215 set_page_private(page
, 0);
4216 page_cache_release(page
);
4221 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4223 if (PageWriteback(page
) || PageDirty(page
))
4225 return __btrfs_releasepage(page
, gfp_flags
);
4228 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
4230 struct extent_io_tree
*tree
;
4231 struct btrfs_ordered_extent
*ordered
;
4232 u64 page_start
= page_offset(page
);
4233 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4235 wait_on_page_writeback(page
);
4236 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4238 btrfs_releasepage(page
, GFP_NOFS
);
4242 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4243 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
4247 * IO on this page will never be started, so we need
4248 * to account for any ordered extents now
4250 clear_extent_bit(tree
, page_start
, page_end
,
4251 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4252 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
4253 btrfs_finish_ordered_io(page
->mapping
->host
,
4254 page_start
, page_end
);
4255 btrfs_put_ordered_extent(ordered
);
4256 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4258 clear_extent_bit(tree
, page_start
, page_end
,
4259 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4262 __btrfs_releasepage(page
, GFP_NOFS
);
4264 ClearPageChecked(page
);
4265 if (PagePrivate(page
)) {
4266 ClearPagePrivate(page
);
4267 set_page_private(page
, 0);
4268 page_cache_release(page
);
4273 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4274 * called from a page fault handler when a page is first dirtied. Hence we must
4275 * be careful to check for EOF conditions here. We set the page up correctly
4276 * for a written page which means we get ENOSPC checking when writing into
4277 * holes and correct delalloc and unwritten extent mapping on filesystems that
4278 * support these features.
4280 * We are not allowed to take the i_mutex here so we have to play games to
4281 * protect against truncate races as the page could now be beyond EOF. Because
4282 * vmtruncate() writes the inode size before removing pages, once we have the
4283 * page lock we can determine safely if the page is beyond EOF. If it is not
4284 * beyond EOF, then the page is guaranteed safe against truncation until we
4287 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4289 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
4290 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4291 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4292 struct btrfs_ordered_extent
*ordered
;
4294 unsigned long zero_start
;
4300 ret
= btrfs_check_free_space(root
, PAGE_CACHE_SIZE
, 0);
4307 size
= i_size_read(inode
);
4308 page_start
= page_offset(page
);
4309 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4311 if ((page
->mapping
!= inode
->i_mapping
) ||
4312 (page_start
>= size
)) {
4313 /* page got truncated out from underneath us */
4316 wait_on_page_writeback(page
);
4318 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4319 set_page_extent_mapped(page
);
4322 * we can't set the delalloc bits if there are pending ordered
4323 * extents. Drop our locks and wait for them to finish
4325 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4327 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4329 btrfs_start_ordered_extent(inode
, ordered
, 1);
4330 btrfs_put_ordered_extent(ordered
);
4334 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
4337 /* page is wholly or partially inside EOF */
4338 if (page_start
+ PAGE_CACHE_SIZE
> size
)
4339 zero_start
= size
& ~PAGE_CACHE_MASK
;
4341 zero_start
= PAGE_CACHE_SIZE
;
4343 if (zero_start
!= PAGE_CACHE_SIZE
) {
4345 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
4346 flush_dcache_page(page
);
4349 ClearPageChecked(page
);
4350 set_page_dirty(page
);
4351 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4359 static void btrfs_truncate(struct inode
*inode
)
4361 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4363 struct btrfs_trans_handle
*trans
;
4365 u64 mask
= root
->sectorsize
- 1;
4367 if (!S_ISREG(inode
->i_mode
))
4369 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4372 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
4373 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
4375 trans
= btrfs_start_transaction(root
, 1);
4376 btrfs_set_trans_block_group(trans
, inode
);
4377 btrfs_i_size_write(inode
, inode
->i_size
);
4379 ret
= btrfs_orphan_add(trans
, inode
);
4382 /* FIXME, add redo link to tree so we don't leak on crash */
4383 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
4384 BTRFS_EXTENT_DATA_KEY
);
4385 btrfs_update_inode(trans
, root
, inode
);
4387 ret
= btrfs_orphan_del(trans
, inode
);
4391 nr
= trans
->blocks_used
;
4392 ret
= btrfs_end_transaction_throttle(trans
, root
);
4394 btrfs_btree_balance_dirty(root
, nr
);
4398 * create a new subvolume directory/inode (helper for the ioctl).
4400 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
4401 struct btrfs_root
*new_root
, struct dentry
*dentry
,
4402 u64 new_dirid
, u64 alloc_hint
)
4404 struct inode
*inode
;
4408 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
4409 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
4411 return PTR_ERR(inode
);
4412 inode
->i_op
= &btrfs_dir_inode_operations
;
4413 inode
->i_fop
= &btrfs_dir_file_operations
;
4416 btrfs_i_size_write(inode
, 0);
4418 error
= btrfs_update_inode(trans
, new_root
, inode
);
4422 d_instantiate(dentry
, inode
);
4426 /* helper function for file defrag and space balancing. This
4427 * forces readahead on a given range of bytes in an inode
4429 unsigned long btrfs_force_ra(struct address_space
*mapping
,
4430 struct file_ra_state
*ra
, struct file
*file
,
4431 pgoff_t offset
, pgoff_t last_index
)
4433 pgoff_t req_size
= last_index
- offset
+ 1;
4435 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
4436 return offset
+ req_size
;
4439 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
4441 struct btrfs_inode
*ei
;
4443 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
4447 ei
->logged_trans
= 0;
4448 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
4449 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
4450 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
4451 INIT_LIST_HEAD(&ei
->i_orphan
);
4452 return &ei
->vfs_inode
;
4455 void btrfs_destroy_inode(struct inode
*inode
)
4457 struct btrfs_ordered_extent
*ordered
;
4458 WARN_ON(!list_empty(&inode
->i_dentry
));
4459 WARN_ON(inode
->i_data
.nrpages
);
4461 if (BTRFS_I(inode
)->i_acl
&&
4462 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
4463 posix_acl_release(BTRFS_I(inode
)->i_acl
);
4464 if (BTRFS_I(inode
)->i_default_acl
&&
4465 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
4466 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
4468 spin_lock(&BTRFS_I(inode
)->root
->list_lock
);
4469 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
4470 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
4471 " list\n", inode
->i_ino
);
4474 spin_unlock(&BTRFS_I(inode
)->root
->list_lock
);
4477 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
4481 printk(KERN_ERR
"btrfs found ordered "
4482 "extent %llu %llu on inode cleanup\n",
4483 (unsigned long long)ordered
->file_offset
,
4484 (unsigned long long)ordered
->len
);
4485 btrfs_remove_ordered_extent(inode
, ordered
);
4486 btrfs_put_ordered_extent(ordered
);
4487 btrfs_put_ordered_extent(ordered
);
4490 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4491 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4494 static void init_once(void *foo
)
4496 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4498 inode_init_once(&ei
->vfs_inode
);
4501 void btrfs_destroy_cachep(void)
4503 if (btrfs_inode_cachep
)
4504 kmem_cache_destroy(btrfs_inode_cachep
);
4505 if (btrfs_trans_handle_cachep
)
4506 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4507 if (btrfs_transaction_cachep
)
4508 kmem_cache_destroy(btrfs_transaction_cachep
);
4509 if (btrfs_bit_radix_cachep
)
4510 kmem_cache_destroy(btrfs_bit_radix_cachep
);
4511 if (btrfs_path_cachep
)
4512 kmem_cache_destroy(btrfs_path_cachep
);
4515 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
4516 unsigned long extra_flags
,
4517 void (*ctor
)(void *))
4519 return kmem_cache_create(name
, size
, 0, (SLAB_RECLAIM_ACCOUNT
|
4520 SLAB_MEM_SPREAD
| extra_flags
), ctor
);
4523 int btrfs_init_cachep(void)
4525 btrfs_inode_cachep
= btrfs_cache_create("btrfs_inode_cache",
4526 sizeof(struct btrfs_inode
),
4528 if (!btrfs_inode_cachep
)
4530 btrfs_trans_handle_cachep
=
4531 btrfs_cache_create("btrfs_trans_handle_cache",
4532 sizeof(struct btrfs_trans_handle
),
4534 if (!btrfs_trans_handle_cachep
)
4536 btrfs_transaction_cachep
= btrfs_cache_create("btrfs_transaction_cache",
4537 sizeof(struct btrfs_transaction
),
4539 if (!btrfs_transaction_cachep
)
4541 btrfs_path_cachep
= btrfs_cache_create("btrfs_path_cache",
4542 sizeof(struct btrfs_path
),
4544 if (!btrfs_path_cachep
)
4546 btrfs_bit_radix_cachep
= btrfs_cache_create("btrfs_radix", 256,
4547 SLAB_DESTROY_BY_RCU
, NULL
);
4548 if (!btrfs_bit_radix_cachep
)
4552 btrfs_destroy_cachep();
4556 static int btrfs_getattr(struct vfsmount
*mnt
,
4557 struct dentry
*dentry
, struct kstat
*stat
)
4559 struct inode
*inode
= dentry
->d_inode
;
4560 generic_fillattr(inode
, stat
);
4561 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
4562 stat
->blksize
= PAGE_CACHE_SIZE
;
4563 stat
->blocks
= (inode_get_bytes(inode
) +
4564 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4568 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4569 struct inode
*new_dir
, struct dentry
*new_dentry
)
4571 struct btrfs_trans_handle
*trans
;
4572 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4573 struct inode
*new_inode
= new_dentry
->d_inode
;
4574 struct inode
*old_inode
= old_dentry
->d_inode
;
4575 struct timespec ctime
= CURRENT_TIME
;
4579 /* we're not allowed to rename between subvolumes */
4580 if (BTRFS_I(old_inode
)->root
->root_key
.objectid
!=
4581 BTRFS_I(new_dir
)->root
->root_key
.objectid
)
4584 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4585 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4589 /* to rename a snapshot or subvolume, we need to juggle the
4590 * backrefs. This isn't coded yet
4592 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
4595 ret
= btrfs_check_free_space(root
, 1, 0);
4599 trans
= btrfs_start_transaction(root
, 1);
4601 btrfs_set_trans_block_group(trans
, new_dir
);
4603 btrfs_inc_nlink(old_dentry
->d_inode
);
4604 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4605 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4606 old_inode
->i_ctime
= ctime
;
4608 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4609 old_dentry
->d_name
.name
,
4610 old_dentry
->d_name
.len
);
4615 new_inode
->i_ctime
= CURRENT_TIME
;
4616 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4617 new_dentry
->d_inode
,
4618 new_dentry
->d_name
.name
,
4619 new_dentry
->d_name
.len
);
4622 if (new_inode
->i_nlink
== 0) {
4623 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4629 ret
= btrfs_set_inode_index(new_dir
, &index
);
4633 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4634 old_inode
, new_dentry
->d_name
.name
,
4635 new_dentry
->d_name
.len
, 1, index
);
4640 btrfs_end_transaction_throttle(trans
, root
);
4646 * some fairly slow code that needs optimization. This walks the list
4647 * of all the inodes with pending delalloc and forces them to disk.
4649 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4651 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4652 struct btrfs_inode
*binode
;
4653 struct inode
*inode
;
4655 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
4658 spin_lock(&root
->fs_info
->delalloc_lock
);
4659 while (!list_empty(head
)) {
4660 binode
= list_entry(head
->next
, struct btrfs_inode
,
4662 inode
= igrab(&binode
->vfs_inode
);
4664 list_del_init(&binode
->delalloc_inodes
);
4665 spin_unlock(&root
->fs_info
->delalloc_lock
);
4667 filemap_flush(inode
->i_mapping
);
4671 spin_lock(&root
->fs_info
->delalloc_lock
);
4673 spin_unlock(&root
->fs_info
->delalloc_lock
);
4675 /* the filemap_flush will queue IO into the worker threads, but
4676 * we have to make sure the IO is actually started and that
4677 * ordered extents get created before we return
4679 atomic_inc(&root
->fs_info
->async_submit_draining
);
4680 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
4681 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
4682 wait_event(root
->fs_info
->async_submit_wait
,
4683 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
4684 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
4686 atomic_dec(&root
->fs_info
->async_submit_draining
);
4690 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4691 const char *symname
)
4693 struct btrfs_trans_handle
*trans
;
4694 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4695 struct btrfs_path
*path
;
4696 struct btrfs_key key
;
4697 struct inode
*inode
= NULL
;
4705 struct btrfs_file_extent_item
*ei
;
4706 struct extent_buffer
*leaf
;
4707 unsigned long nr
= 0;
4709 name_len
= strlen(symname
) + 1;
4710 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4711 return -ENAMETOOLONG
;
4713 err
= btrfs_check_free_space(root
, 1, 0);
4717 trans
= btrfs_start_transaction(root
, 1);
4718 btrfs_set_trans_block_group(trans
, dir
);
4720 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4726 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4728 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4729 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4731 err
= PTR_ERR(inode
);
4735 err
= btrfs_init_acl(inode
, dir
);
4741 btrfs_set_trans_block_group(trans
, inode
);
4742 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4746 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4747 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4748 inode
->i_fop
= &btrfs_file_operations
;
4749 inode
->i_op
= &btrfs_file_inode_operations
;
4750 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4752 dir
->i_sb
->s_dirt
= 1;
4753 btrfs_update_inode_block_group(trans
, inode
);
4754 btrfs_update_inode_block_group(trans
, dir
);
4758 path
= btrfs_alloc_path();
4760 key
.objectid
= inode
->i_ino
;
4762 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
4763 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
4764 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
4770 leaf
= path
->nodes
[0];
4771 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
4772 struct btrfs_file_extent_item
);
4773 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
4774 btrfs_set_file_extent_type(leaf
, ei
,
4775 BTRFS_FILE_EXTENT_INLINE
);
4776 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
4777 btrfs_set_file_extent_compression(leaf
, ei
, 0);
4778 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
4779 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
4781 ptr
= btrfs_file_extent_inline_start(ei
);
4782 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
4783 btrfs_mark_buffer_dirty(leaf
);
4784 btrfs_free_path(path
);
4786 inode
->i_op
= &btrfs_symlink_inode_operations
;
4787 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
4788 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4789 inode_set_bytes(inode
, name_len
);
4790 btrfs_i_size_write(inode
, name_len
- 1);
4791 err
= btrfs_update_inode(trans
, root
, inode
);
4796 nr
= trans
->blocks_used
;
4797 btrfs_end_transaction_throttle(trans
, root
);
4800 inode_dec_link_count(inode
);
4803 btrfs_btree_balance_dirty(root
, nr
);
4807 static int prealloc_file_range(struct inode
*inode
, u64 start
, u64 end
,
4808 u64 alloc_hint
, int mode
)
4810 struct btrfs_trans_handle
*trans
;
4811 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4812 struct btrfs_key ins
;
4814 u64 cur_offset
= start
;
4815 u64 num_bytes
= end
- start
;
4818 trans
= btrfs_join_transaction(root
, 1);
4820 btrfs_set_trans_block_group(trans
, inode
);
4822 while (num_bytes
> 0) {
4823 alloc_size
= min(num_bytes
, root
->fs_info
->max_extent
);
4824 ret
= btrfs_reserve_extent(trans
, root
, alloc_size
,
4825 root
->sectorsize
, 0, alloc_hint
,
4831 ret
= insert_reserved_file_extent(trans
, inode
,
4832 cur_offset
, ins
.objectid
,
4833 ins
.offset
, ins
.offset
,
4834 ins
.offset
, 0, 0, 0,
4835 BTRFS_FILE_EXTENT_PREALLOC
);
4837 num_bytes
-= ins
.offset
;
4838 cur_offset
+= ins
.offset
;
4839 alloc_hint
= ins
.objectid
+ ins
.offset
;
4842 if (cur_offset
> start
) {
4843 inode
->i_ctime
= CURRENT_TIME
;
4844 btrfs_set_flag(inode
, PREALLOC
);
4845 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4846 cur_offset
> i_size_read(inode
))
4847 btrfs_i_size_write(inode
, cur_offset
);
4848 ret
= btrfs_update_inode(trans
, root
, inode
);
4852 btrfs_end_transaction(trans
, root
);
4856 static long btrfs_fallocate(struct inode
*inode
, int mode
,
4857 loff_t offset
, loff_t len
)
4864 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
4865 struct extent_map
*em
;
4868 alloc_start
= offset
& ~mask
;
4869 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
4871 mutex_lock(&inode
->i_mutex
);
4872 if (alloc_start
> inode
->i_size
) {
4873 ret
= btrfs_cont_expand(inode
, alloc_start
);
4879 struct btrfs_ordered_extent
*ordered
;
4880 lock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
,
4881 alloc_end
- 1, GFP_NOFS
);
4882 ordered
= btrfs_lookup_first_ordered_extent(inode
,
4885 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
4886 ordered
->file_offset
< alloc_end
) {
4887 btrfs_put_ordered_extent(ordered
);
4888 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4889 alloc_start
, alloc_end
- 1, GFP_NOFS
);
4890 btrfs_wait_ordered_range(inode
, alloc_start
,
4891 alloc_end
- alloc_start
);
4894 btrfs_put_ordered_extent(ordered
);
4899 cur_offset
= alloc_start
;
4901 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4902 alloc_end
- cur_offset
, 0);
4903 BUG_ON(IS_ERR(em
) || !em
);
4904 last_byte
= min(extent_map_end(em
), alloc_end
);
4905 last_byte
= (last_byte
+ mask
) & ~mask
;
4906 if (em
->block_start
== EXTENT_MAP_HOLE
) {
4907 ret
= prealloc_file_range(inode
, cur_offset
,
4908 last_byte
, alloc_hint
, mode
);
4910 free_extent_map(em
);
4914 if (em
->block_start
<= EXTENT_MAP_LAST_BYTE
)
4915 alloc_hint
= em
->block_start
;
4916 free_extent_map(em
);
4918 cur_offset
= last_byte
;
4919 if (cur_offset
>= alloc_end
) {
4924 unlock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, alloc_end
- 1,
4927 mutex_unlock(&inode
->i_mutex
);
4931 static int btrfs_set_page_dirty(struct page
*page
)
4933 return __set_page_dirty_nobuffers(page
);
4936 static int btrfs_permission(struct inode
*inode
, int mask
)
4938 if (btrfs_test_flag(inode
, READONLY
) && (mask
& MAY_WRITE
))
4940 return generic_permission(inode
, mask
, btrfs_check_acl
);
4943 static struct inode_operations btrfs_dir_inode_operations
= {
4944 .getattr
= btrfs_getattr
,
4945 .lookup
= btrfs_lookup
,
4946 .create
= btrfs_create
,
4947 .unlink
= btrfs_unlink
,
4949 .mkdir
= btrfs_mkdir
,
4950 .rmdir
= btrfs_rmdir
,
4951 .rename
= btrfs_rename
,
4952 .symlink
= btrfs_symlink
,
4953 .setattr
= btrfs_setattr
,
4954 .mknod
= btrfs_mknod
,
4955 .setxattr
= btrfs_setxattr
,
4956 .getxattr
= btrfs_getxattr
,
4957 .listxattr
= btrfs_listxattr
,
4958 .removexattr
= btrfs_removexattr
,
4959 .permission
= btrfs_permission
,
4961 static struct inode_operations btrfs_dir_ro_inode_operations
= {
4962 .lookup
= btrfs_lookup
,
4963 .permission
= btrfs_permission
,
4965 static struct file_operations btrfs_dir_file_operations
= {
4966 .llseek
= generic_file_llseek
,
4967 .read
= generic_read_dir
,
4968 .readdir
= btrfs_real_readdir
,
4969 .unlocked_ioctl
= btrfs_ioctl
,
4970 #ifdef CONFIG_COMPAT
4971 .compat_ioctl
= btrfs_ioctl
,
4973 .release
= btrfs_release_file
,
4974 .fsync
= btrfs_sync_file
,
4977 static struct extent_io_ops btrfs_extent_io_ops
= {
4978 .fill_delalloc
= run_delalloc_range
,
4979 .submit_bio_hook
= btrfs_submit_bio_hook
,
4980 .merge_bio_hook
= btrfs_merge_bio_hook
,
4981 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
4982 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
4983 .writepage_start_hook
= btrfs_writepage_start_hook
,
4984 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
4985 .set_bit_hook
= btrfs_set_bit_hook
,
4986 .clear_bit_hook
= btrfs_clear_bit_hook
,
4989 static struct address_space_operations btrfs_aops
= {
4990 .readpage
= btrfs_readpage
,
4991 .writepage
= btrfs_writepage
,
4992 .writepages
= btrfs_writepages
,
4993 .readpages
= btrfs_readpages
,
4994 .sync_page
= block_sync_page
,
4996 .direct_IO
= btrfs_direct_IO
,
4997 .invalidatepage
= btrfs_invalidatepage
,
4998 .releasepage
= btrfs_releasepage
,
4999 .set_page_dirty
= btrfs_set_page_dirty
,
5002 static struct address_space_operations btrfs_symlink_aops
= {
5003 .readpage
= btrfs_readpage
,
5004 .writepage
= btrfs_writepage
,
5005 .invalidatepage
= btrfs_invalidatepage
,
5006 .releasepage
= btrfs_releasepage
,
5009 static struct inode_operations btrfs_file_inode_operations
= {
5010 .truncate
= btrfs_truncate
,
5011 .getattr
= btrfs_getattr
,
5012 .setattr
= btrfs_setattr
,
5013 .setxattr
= btrfs_setxattr
,
5014 .getxattr
= btrfs_getxattr
,
5015 .listxattr
= btrfs_listxattr
,
5016 .removexattr
= btrfs_removexattr
,
5017 .permission
= btrfs_permission
,
5018 .fallocate
= btrfs_fallocate
,
5020 static struct inode_operations btrfs_special_inode_operations
= {
5021 .getattr
= btrfs_getattr
,
5022 .setattr
= btrfs_setattr
,
5023 .permission
= btrfs_permission
,
5024 .setxattr
= btrfs_setxattr
,
5025 .getxattr
= btrfs_getxattr
,
5026 .listxattr
= btrfs_listxattr
,
5027 .removexattr
= btrfs_removexattr
,
5029 static struct inode_operations btrfs_symlink_inode_operations
= {
5030 .readlink
= generic_readlink
,
5031 .follow_link
= page_follow_link_light
,
5032 .put_link
= page_put_link
,
5033 .permission
= btrfs_permission
,