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 btrfs_ordered_sum
*sum
;
1328 btrfs_set_trans_block_group(trans
, inode
);
1330 list_for_each_entry(sum
, list
, list
) {
1331 btrfs_csum_file_blocks(trans
,
1332 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1337 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
1339 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1341 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1345 /* see btrfs_writepage_start_hook for details on why this is required */
1346 struct btrfs_writepage_fixup
{
1348 struct btrfs_work work
;
1351 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1353 struct btrfs_writepage_fixup
*fixup
;
1354 struct btrfs_ordered_extent
*ordered
;
1356 struct inode
*inode
;
1360 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1364 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1365 ClearPageChecked(page
);
1369 inode
= page
->mapping
->host
;
1370 page_start
= page_offset(page
);
1371 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1373 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1375 /* already ordered? We're done */
1376 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1377 EXTENT_ORDERED
, 0)) {
1381 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1383 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
1384 page_end
, GFP_NOFS
);
1386 btrfs_start_ordered_extent(inode
, ordered
, 1);
1390 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
1391 ClearPageChecked(page
);
1393 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1396 page_cache_release(page
);
1400 * There are a few paths in the higher layers of the kernel that directly
1401 * set the page dirty bit without asking the filesystem if it is a
1402 * good idea. This causes problems because we want to make sure COW
1403 * properly happens and the data=ordered rules are followed.
1405 * In our case any range that doesn't have the ORDERED bit set
1406 * hasn't been properly setup for IO. We kick off an async process
1407 * to fix it up. The async helper will wait for ordered extents, set
1408 * the delalloc bit and make it safe to write the page.
1410 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1412 struct inode
*inode
= page
->mapping
->host
;
1413 struct btrfs_writepage_fixup
*fixup
;
1414 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1417 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1422 if (PageChecked(page
))
1425 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1429 SetPageChecked(page
);
1430 page_cache_get(page
);
1431 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1433 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1437 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1438 struct inode
*inode
, u64 file_pos
,
1439 u64 disk_bytenr
, u64 disk_num_bytes
,
1440 u64 num_bytes
, u64 ram_bytes
,
1441 u8 compression
, u8 encryption
,
1442 u16 other_encoding
, int extent_type
)
1444 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1445 struct btrfs_file_extent_item
*fi
;
1446 struct btrfs_path
*path
;
1447 struct extent_buffer
*leaf
;
1448 struct btrfs_key ins
;
1452 path
= btrfs_alloc_path();
1455 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1456 file_pos
+ num_bytes
, file_pos
, &hint
);
1459 ins
.objectid
= inode
->i_ino
;
1460 ins
.offset
= file_pos
;
1461 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1462 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1464 leaf
= path
->nodes
[0];
1465 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1466 struct btrfs_file_extent_item
);
1467 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1468 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1469 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1470 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1471 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1472 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1473 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1474 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1475 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1476 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1477 btrfs_mark_buffer_dirty(leaf
);
1479 inode_add_bytes(inode
, num_bytes
);
1480 btrfs_drop_extent_cache(inode
, file_pos
, file_pos
+ num_bytes
- 1, 0);
1482 ins
.objectid
= disk_bytenr
;
1483 ins
.offset
= disk_num_bytes
;
1484 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1485 ret
= btrfs_alloc_reserved_extent(trans
, root
, leaf
->start
,
1486 root
->root_key
.objectid
,
1487 trans
->transid
, inode
->i_ino
, &ins
);
1490 btrfs_free_path(path
);
1494 /* as ordered data IO finishes, this gets called so we can finish
1495 * an ordered extent if the range of bytes in the file it covers are
1498 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1500 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1501 struct btrfs_trans_handle
*trans
;
1502 struct btrfs_ordered_extent
*ordered_extent
;
1503 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1507 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1511 trans
= btrfs_join_transaction(root
, 1);
1513 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1514 BUG_ON(!ordered_extent
);
1515 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1518 lock_extent(io_tree
, ordered_extent
->file_offset
,
1519 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1522 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1524 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1526 ret
= btrfs_mark_extent_written(trans
, root
, inode
,
1527 ordered_extent
->file_offset
,
1528 ordered_extent
->file_offset
+
1529 ordered_extent
->len
);
1532 ret
= insert_reserved_file_extent(trans
, inode
,
1533 ordered_extent
->file_offset
,
1534 ordered_extent
->start
,
1535 ordered_extent
->disk_len
,
1536 ordered_extent
->len
,
1537 ordered_extent
->len
,
1539 BTRFS_FILE_EXTENT_REG
);
1542 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1543 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1546 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1547 &ordered_extent
->list
);
1549 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1550 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1551 btrfs_update_inode(trans
, root
, inode
);
1552 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1553 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1556 btrfs_put_ordered_extent(ordered_extent
);
1557 /* once for the tree */
1558 btrfs_put_ordered_extent(ordered_extent
);
1560 btrfs_end_transaction(trans
, root
);
1564 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1565 struct extent_state
*state
, int uptodate
)
1567 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1571 * When IO fails, either with EIO or csum verification fails, we
1572 * try other mirrors that might have a good copy of the data. This
1573 * io_failure_record is used to record state as we go through all the
1574 * mirrors. If another mirror has good data, the page is set up to date
1575 * and things continue. If a good mirror can't be found, the original
1576 * bio end_io callback is called to indicate things have failed.
1578 struct io_failure_record
{
1583 unsigned long bio_flags
;
1587 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1588 struct page
*page
, u64 start
, u64 end
,
1589 struct extent_state
*state
)
1591 struct io_failure_record
*failrec
= NULL
;
1593 struct extent_map
*em
;
1594 struct inode
*inode
= page
->mapping
->host
;
1595 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1596 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1603 ret
= get_state_private(failure_tree
, start
, &private);
1605 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1608 failrec
->start
= start
;
1609 failrec
->len
= end
- start
+ 1;
1610 failrec
->last_mirror
= 0;
1611 failrec
->bio_flags
= 0;
1613 spin_lock(&em_tree
->lock
);
1614 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1615 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1616 free_extent_map(em
);
1619 spin_unlock(&em_tree
->lock
);
1621 if (!em
|| IS_ERR(em
)) {
1625 logical
= start
- em
->start
;
1626 logical
= em
->block_start
+ logical
;
1627 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1628 logical
= em
->block_start
;
1629 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1631 failrec
->logical
= logical
;
1632 free_extent_map(em
);
1633 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1634 EXTENT_DIRTY
, GFP_NOFS
);
1635 set_state_private(failure_tree
, start
,
1636 (u64
)(unsigned long)failrec
);
1638 failrec
= (struct io_failure_record
*)(unsigned long)private;
1640 num_copies
= btrfs_num_copies(
1641 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1642 failrec
->logical
, failrec
->len
);
1643 failrec
->last_mirror
++;
1645 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1646 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1649 if (state
&& state
->start
!= failrec
->start
)
1651 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1653 if (!state
|| failrec
->last_mirror
> num_copies
) {
1654 set_state_private(failure_tree
, failrec
->start
, 0);
1655 clear_extent_bits(failure_tree
, failrec
->start
,
1656 failrec
->start
+ failrec
->len
- 1,
1657 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1661 bio
= bio_alloc(GFP_NOFS
, 1);
1662 bio
->bi_private
= state
;
1663 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1664 bio
->bi_sector
= failrec
->logical
>> 9;
1665 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1668 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1669 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1674 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1675 failrec
->last_mirror
,
1676 failrec
->bio_flags
);
1681 * each time an IO finishes, we do a fast check in the IO failure tree
1682 * to see if we need to process or clean up an io_failure_record
1684 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1687 u64 private_failure
;
1688 struct io_failure_record
*failure
;
1692 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1693 (u64
)-1, 1, EXTENT_DIRTY
)) {
1694 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1695 start
, &private_failure
);
1697 failure
= (struct io_failure_record
*)(unsigned long)
1699 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1701 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1703 failure
->start
+ failure
->len
- 1,
1704 EXTENT_DIRTY
| EXTENT_LOCKED
,
1713 * when reads are done, we need to check csums to verify the data is correct
1714 * if there's a match, we allow the bio to finish. If not, we go through
1715 * the io_failure_record routines to find good copies
1717 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1718 struct extent_state
*state
)
1720 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1721 struct inode
*inode
= page
->mapping
->host
;
1722 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1724 u64
private = ~(u32
)0;
1726 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1729 if (PageChecked(page
)) {
1730 ClearPageChecked(page
);
1733 if (btrfs_test_flag(inode
, NODATASUM
))
1736 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1737 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1)) {
1738 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1743 if (state
&& state
->start
== start
) {
1744 private = state
->private;
1747 ret
= get_state_private(io_tree
, start
, &private);
1749 kaddr
= kmap_atomic(page
, KM_USER0
);
1753 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1754 btrfs_csum_final(csum
, (char *)&csum
);
1755 if (csum
!= private)
1758 kunmap_atomic(kaddr
, KM_USER0
);
1760 /* if the io failure tree for this inode is non-empty,
1761 * check to see if we've recovered from a failed IO
1763 btrfs_clean_io_failures(inode
, start
);
1767 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
1768 "private %llu\n", page
->mapping
->host
->i_ino
,
1769 (unsigned long long)start
, csum
,
1770 (unsigned long long)private);
1771 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1772 flush_dcache_page(page
);
1773 kunmap_atomic(kaddr
, KM_USER0
);
1780 * This creates an orphan entry for the given inode in case something goes
1781 * wrong in the middle of an unlink/truncate.
1783 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1785 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1788 spin_lock(&root
->list_lock
);
1790 /* already on the orphan list, we're good */
1791 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1792 spin_unlock(&root
->list_lock
);
1796 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1798 spin_unlock(&root
->list_lock
);
1801 * insert an orphan item to track this unlinked/truncated file
1803 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1809 * We have done the truncate/delete so we can go ahead and remove the orphan
1810 * item for this particular inode.
1812 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1814 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1817 spin_lock(&root
->list_lock
);
1819 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1820 spin_unlock(&root
->list_lock
);
1824 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1826 spin_unlock(&root
->list_lock
);
1830 spin_unlock(&root
->list_lock
);
1832 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1838 * this cleans up any orphans that may be left on the list from the last use
1841 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1843 struct btrfs_path
*path
;
1844 struct extent_buffer
*leaf
;
1845 struct btrfs_item
*item
;
1846 struct btrfs_key key
, found_key
;
1847 struct btrfs_trans_handle
*trans
;
1848 struct inode
*inode
;
1849 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1851 path
= btrfs_alloc_path();
1856 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1857 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1858 key
.offset
= (u64
)-1;
1862 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1864 printk(KERN_ERR
"Error searching slot for orphan: %d"
1870 * if ret == 0 means we found what we were searching for, which
1871 * is weird, but possible, so only screw with path if we didnt
1872 * find the key and see if we have stuff that matches
1875 if (path
->slots
[0] == 0)
1880 /* pull out the item */
1881 leaf
= path
->nodes
[0];
1882 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1883 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1885 /* make sure the item matches what we want */
1886 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1888 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1891 /* release the path since we're done with it */
1892 btrfs_release_path(root
, path
);
1895 * this is where we are basically btrfs_lookup, without the
1896 * crossing root thing. we store the inode number in the
1897 * offset of the orphan item.
1899 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
1900 found_key
.offset
, root
);
1904 if (inode
->i_state
& I_NEW
) {
1905 BTRFS_I(inode
)->root
= root
;
1907 /* have to set the location manually */
1908 BTRFS_I(inode
)->location
.objectid
= inode
->i_ino
;
1909 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
1910 BTRFS_I(inode
)->location
.offset
= 0;
1912 btrfs_read_locked_inode(inode
);
1913 unlock_new_inode(inode
);
1917 * add this inode to the orphan list so btrfs_orphan_del does
1918 * the proper thing when we hit it
1920 spin_lock(&root
->list_lock
);
1921 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1922 spin_unlock(&root
->list_lock
);
1925 * if this is a bad inode, means we actually succeeded in
1926 * removing the inode, but not the orphan record, which means
1927 * we need to manually delete the orphan since iput will just
1928 * do a destroy_inode
1930 if (is_bad_inode(inode
)) {
1931 trans
= btrfs_start_transaction(root
, 1);
1932 btrfs_orphan_del(trans
, inode
);
1933 btrfs_end_transaction(trans
, root
);
1938 /* if we have links, this was a truncate, lets do that */
1939 if (inode
->i_nlink
) {
1941 btrfs_truncate(inode
);
1946 /* this will do delete_inode and everything for us */
1951 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
1953 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
1955 btrfs_free_path(path
);
1959 * read an inode from the btree into the in-memory inode
1961 void btrfs_read_locked_inode(struct inode
*inode
)
1963 struct btrfs_path
*path
;
1964 struct extent_buffer
*leaf
;
1965 struct btrfs_inode_item
*inode_item
;
1966 struct btrfs_timespec
*tspec
;
1967 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1968 struct btrfs_key location
;
1969 u64 alloc_group_block
;
1973 path
= btrfs_alloc_path();
1975 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
1977 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
1981 leaf
= path
->nodes
[0];
1982 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1983 struct btrfs_inode_item
);
1985 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
1986 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
1987 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
1988 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
1989 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
1991 tspec
= btrfs_inode_atime(inode_item
);
1992 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1993 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1995 tspec
= btrfs_inode_mtime(inode_item
);
1996 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1997 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1999 tspec
= btrfs_inode_ctime(inode_item
);
2000 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2001 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2003 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2004 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2005 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2006 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2008 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2010 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2011 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2013 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2014 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2015 alloc_group_block
, 0);
2016 btrfs_free_path(path
);
2019 switch (inode
->i_mode
& S_IFMT
) {
2021 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2022 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2023 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2024 inode
->i_fop
= &btrfs_file_operations
;
2025 inode
->i_op
= &btrfs_file_inode_operations
;
2028 inode
->i_fop
= &btrfs_dir_file_operations
;
2029 if (root
== root
->fs_info
->tree_root
)
2030 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2032 inode
->i_op
= &btrfs_dir_inode_operations
;
2035 inode
->i_op
= &btrfs_symlink_inode_operations
;
2036 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2037 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2040 init_special_inode(inode
, inode
->i_mode
, rdev
);
2046 btrfs_free_path(path
);
2047 make_bad_inode(inode
);
2051 * given a leaf and an inode, copy the inode fields into the leaf
2053 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2054 struct extent_buffer
*leaf
,
2055 struct btrfs_inode_item
*item
,
2056 struct inode
*inode
)
2058 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2059 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2060 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2061 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2062 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2064 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2065 inode
->i_atime
.tv_sec
);
2066 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2067 inode
->i_atime
.tv_nsec
);
2069 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2070 inode
->i_mtime
.tv_sec
);
2071 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2072 inode
->i_mtime
.tv_nsec
);
2074 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2075 inode
->i_ctime
.tv_sec
);
2076 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2077 inode
->i_ctime
.tv_nsec
);
2079 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2080 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2081 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2082 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2083 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2084 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2085 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2089 * copy everything in the in-memory inode into the btree.
2091 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2092 struct btrfs_root
*root
, struct inode
*inode
)
2094 struct btrfs_inode_item
*inode_item
;
2095 struct btrfs_path
*path
;
2096 struct extent_buffer
*leaf
;
2099 path
= btrfs_alloc_path();
2101 ret
= btrfs_lookup_inode(trans
, root
, path
,
2102 &BTRFS_I(inode
)->location
, 1);
2109 leaf
= path
->nodes
[0];
2110 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2111 struct btrfs_inode_item
);
2113 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2114 btrfs_mark_buffer_dirty(leaf
);
2115 btrfs_set_inode_last_trans(trans
, inode
);
2118 btrfs_free_path(path
);
2124 * unlink helper that gets used here in inode.c and in the tree logging
2125 * recovery code. It remove a link in a directory with a given name, and
2126 * also drops the back refs in the inode to the directory
2128 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2129 struct btrfs_root
*root
,
2130 struct inode
*dir
, struct inode
*inode
,
2131 const char *name
, int name_len
)
2133 struct btrfs_path
*path
;
2135 struct extent_buffer
*leaf
;
2136 struct btrfs_dir_item
*di
;
2137 struct btrfs_key key
;
2140 path
= btrfs_alloc_path();
2146 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2147 name
, name_len
, -1);
2156 leaf
= path
->nodes
[0];
2157 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2158 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2161 btrfs_release_path(root
, path
);
2163 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2165 dir
->i_ino
, &index
);
2167 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2168 "inode %lu parent %lu\n", name_len
, name
,
2169 inode
->i_ino
, dir
->i_ino
);
2173 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2174 index
, name
, name_len
, -1);
2183 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2184 btrfs_release_path(root
, path
);
2186 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2188 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2190 BTRFS_I(dir
)->log_dirty_trans
= trans
->transid
;
2192 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2196 btrfs_free_path(path
);
2200 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2201 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2202 btrfs_update_inode(trans
, root
, dir
);
2203 btrfs_drop_nlink(inode
);
2204 ret
= btrfs_update_inode(trans
, root
, inode
);
2205 dir
->i_sb
->s_dirt
= 1;
2210 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2212 struct btrfs_root
*root
;
2213 struct btrfs_trans_handle
*trans
;
2214 struct inode
*inode
= dentry
->d_inode
;
2216 unsigned long nr
= 0;
2218 root
= BTRFS_I(dir
)->root
;
2220 ret
= btrfs_check_free_space(root
, 1, 1);
2224 trans
= btrfs_start_transaction(root
, 1);
2226 btrfs_set_trans_block_group(trans
, dir
);
2227 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2228 dentry
->d_name
.name
, dentry
->d_name
.len
);
2230 if (inode
->i_nlink
== 0)
2231 ret
= btrfs_orphan_add(trans
, inode
);
2233 nr
= trans
->blocks_used
;
2235 btrfs_end_transaction_throttle(trans
, root
);
2237 btrfs_btree_balance_dirty(root
, nr
);
2241 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2243 struct inode
*inode
= dentry
->d_inode
;
2246 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2247 struct btrfs_trans_handle
*trans
;
2248 unsigned long nr
= 0;
2251 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2252 * the root of a subvolume or snapshot
2254 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2255 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
2259 ret
= btrfs_check_free_space(root
, 1, 1);
2263 trans
= btrfs_start_transaction(root
, 1);
2264 btrfs_set_trans_block_group(trans
, dir
);
2266 err
= btrfs_orphan_add(trans
, inode
);
2270 /* now the directory is empty */
2271 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2272 dentry
->d_name
.name
, dentry
->d_name
.len
);
2274 btrfs_i_size_write(inode
, 0);
2277 nr
= trans
->blocks_used
;
2278 ret
= btrfs_end_transaction_throttle(trans
, root
);
2280 btrfs_btree_balance_dirty(root
, nr
);
2289 * when truncating bytes in a file, it is possible to avoid reading
2290 * the leaves that contain only checksum items. This can be the
2291 * majority of the IO required to delete a large file, but it must
2292 * be done carefully.
2294 * The keys in the level just above the leaves are checked to make sure
2295 * the lowest key in a given leaf is a csum key, and starts at an offset
2296 * after the new size.
2298 * Then the key for the next leaf is checked to make sure it also has
2299 * a checksum item for the same file. If it does, we know our target leaf
2300 * contains only checksum items, and it can be safely freed without reading
2303 * This is just an optimization targeted at large files. It may do
2304 * nothing. It will return 0 unless things went badly.
2306 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
2307 struct btrfs_root
*root
,
2308 struct btrfs_path
*path
,
2309 struct inode
*inode
, u64 new_size
)
2311 struct btrfs_key key
;
2314 struct btrfs_key found_key
;
2315 struct btrfs_key other_key
;
2316 struct btrfs_leaf_ref
*ref
;
2320 path
->lowest_level
= 1;
2321 key
.objectid
= inode
->i_ino
;
2322 key
.type
= BTRFS_CSUM_ITEM_KEY
;
2323 key
.offset
= new_size
;
2325 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2329 if (path
->nodes
[1] == NULL
) {
2334 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
2335 nritems
= btrfs_header_nritems(path
->nodes
[1]);
2340 if (path
->slots
[1] >= nritems
)
2343 /* did we find a key greater than anything we want to delete? */
2344 if (found_key
.objectid
> inode
->i_ino
||
2345 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
2348 /* we check the next key in the node to make sure the leave contains
2349 * only checksum items. This comparison doesn't work if our
2350 * leaf is the last one in the node
2352 if (path
->slots
[1] + 1 >= nritems
) {
2354 /* search forward from the last key in the node, this
2355 * will bring us into the next node in the tree
2357 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
2359 /* unlikely, but we inc below, so check to be safe */
2360 if (found_key
.offset
== (u64
)-1)
2363 /* search_forward needs a path with locks held, do the
2364 * search again for the original key. It is possible
2365 * this will race with a balance and return a path that
2366 * we could modify, but this drop is just an optimization
2367 * and is allowed to miss some leaves.
2369 btrfs_release_path(root
, path
);
2372 /* setup a max key for search_forward */
2373 other_key
.offset
= (u64
)-1;
2374 other_key
.type
= key
.type
;
2375 other_key
.objectid
= key
.objectid
;
2377 path
->keep_locks
= 1;
2378 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
2380 path
->keep_locks
= 0;
2381 if (ret
|| found_key
.objectid
!= key
.objectid
||
2382 found_key
.type
!= key
.type
) {
2387 key
.offset
= found_key
.offset
;
2388 btrfs_release_path(root
, path
);
2393 /* we know there's one more slot after us in the tree,
2394 * read that key so we can verify it is also a checksum item
2396 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
2398 if (found_key
.objectid
< inode
->i_ino
)
2401 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
2405 * if the key for the next leaf isn't a csum key from this objectid,
2406 * we can't be sure there aren't good items inside this leaf.
2409 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
2412 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
2413 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
2415 * it is safe to delete this leaf, it contains only
2416 * csum items from this inode at an offset >= new_size
2418 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
2421 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
2422 ref
= btrfs_alloc_leaf_ref(root
, 0);
2424 ref
->root_gen
= root
->root_key
.offset
;
2425 ref
->bytenr
= leaf_start
;
2427 ref
->generation
= leaf_gen
;
2430 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
2432 btrfs_free_leaf_ref(root
, ref
);
2438 btrfs_release_path(root
, path
);
2440 if (other_key
.objectid
== inode
->i_ino
&&
2441 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
2442 key
.offset
= other_key
.offset
;
2448 /* fixup any changes we've made to the path */
2449 path
->lowest_level
= 0;
2450 path
->keep_locks
= 0;
2451 btrfs_release_path(root
, path
);
2458 * this can truncate away extent items, csum items and directory items.
2459 * It starts at a high offset and removes keys until it can't find
2460 * any higher than new_size
2462 * csum items that cross the new i_size are truncated to the new size
2465 * min_type is the minimum key type to truncate down to. If set to 0, this
2466 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2468 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2469 struct btrfs_root
*root
,
2470 struct inode
*inode
,
2471 u64 new_size
, u32 min_type
)
2474 struct btrfs_path
*path
;
2475 struct btrfs_key key
;
2476 struct btrfs_key found_key
;
2478 struct extent_buffer
*leaf
;
2479 struct btrfs_file_extent_item
*fi
;
2480 u64 extent_start
= 0;
2481 u64 extent_num_bytes
= 0;
2487 int pending_del_nr
= 0;
2488 int pending_del_slot
= 0;
2489 int extent_type
= -1;
2491 u64 mask
= root
->sectorsize
- 1;
2494 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2495 path
= btrfs_alloc_path();
2499 /* FIXME, add redo link to tree so we don't leak on crash */
2500 key
.objectid
= inode
->i_ino
;
2501 key
.offset
= (u64
)-1;
2504 btrfs_init_path(path
);
2507 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2512 /* there are no items in the tree for us to truncate, we're
2515 if (path
->slots
[0] == 0) {
2524 leaf
= path
->nodes
[0];
2525 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2526 found_type
= btrfs_key_type(&found_key
);
2529 if (found_key
.objectid
!= inode
->i_ino
)
2532 if (found_type
< min_type
)
2535 item_end
= found_key
.offset
;
2536 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2537 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2538 struct btrfs_file_extent_item
);
2539 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2540 encoding
= btrfs_file_extent_compression(leaf
, fi
);
2541 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
2542 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
2544 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2546 btrfs_file_extent_num_bytes(leaf
, fi
);
2547 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2548 item_end
+= btrfs_file_extent_inline_len(leaf
,
2553 if (item_end
< new_size
) {
2554 if (found_type
== BTRFS_DIR_ITEM_KEY
)
2555 found_type
= BTRFS_INODE_ITEM_KEY
;
2556 else if (found_type
== BTRFS_EXTENT_ITEM_KEY
)
2557 found_type
= BTRFS_EXTENT_DATA_KEY
;
2558 else if (found_type
== BTRFS_EXTENT_DATA_KEY
)
2559 found_type
= BTRFS_XATTR_ITEM_KEY
;
2560 else if (found_type
== BTRFS_XATTR_ITEM_KEY
)
2561 found_type
= BTRFS_INODE_REF_KEY
;
2562 else if (found_type
)
2566 btrfs_set_key_type(&key
, found_type
);
2569 if (found_key
.offset
>= new_size
)
2575 /* FIXME, shrink the extent if the ref count is only 1 */
2576 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2579 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2581 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2582 if (!del_item
&& !encoding
) {
2583 u64 orig_num_bytes
=
2584 btrfs_file_extent_num_bytes(leaf
, fi
);
2585 extent_num_bytes
= new_size
-
2586 found_key
.offset
+ root
->sectorsize
- 1;
2587 extent_num_bytes
= extent_num_bytes
&
2588 ~((u64
)root
->sectorsize
- 1);
2589 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2591 num_dec
= (orig_num_bytes
-
2593 if (root
->ref_cows
&& extent_start
!= 0)
2594 inode_sub_bytes(inode
, num_dec
);
2595 btrfs_mark_buffer_dirty(leaf
);
2598 btrfs_file_extent_disk_num_bytes(leaf
,
2600 /* FIXME blocksize != 4096 */
2601 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2602 if (extent_start
!= 0) {
2605 inode_sub_bytes(inode
, num_dec
);
2607 root_gen
= btrfs_header_generation(leaf
);
2608 root_owner
= btrfs_header_owner(leaf
);
2610 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2612 * we can't truncate inline items that have had
2616 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2617 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2618 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2619 u32 size
= new_size
- found_key
.offset
;
2621 if (root
->ref_cows
) {
2622 inode_sub_bytes(inode
, item_end
+ 1 -
2626 btrfs_file_extent_calc_inline_size(size
);
2627 ret
= btrfs_truncate_item(trans
, root
, path
,
2630 } else if (root
->ref_cows
) {
2631 inode_sub_bytes(inode
, item_end
+ 1 -
2637 if (!pending_del_nr
) {
2638 /* no pending yet, add ourselves */
2639 pending_del_slot
= path
->slots
[0];
2641 } else if (pending_del_nr
&&
2642 path
->slots
[0] + 1 == pending_del_slot
) {
2643 /* hop on the pending chunk */
2645 pending_del_slot
= path
->slots
[0];
2653 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2655 leaf
->start
, root_owner
,
2656 root_gen
, inode
->i_ino
, 0);
2660 if (path
->slots
[0] == 0) {
2663 btrfs_release_path(root
, path
);
2668 if (pending_del_nr
&&
2669 path
->slots
[0] + 1 != pending_del_slot
) {
2670 struct btrfs_key debug
;
2672 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2674 ret
= btrfs_del_items(trans
, root
, path
,
2679 btrfs_release_path(root
, path
);
2685 if (pending_del_nr
) {
2686 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2689 btrfs_free_path(path
);
2690 inode
->i_sb
->s_dirt
= 1;
2695 * taken from block_truncate_page, but does cow as it zeros out
2696 * any bytes left in the last page in the file.
2698 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2700 struct inode
*inode
= mapping
->host
;
2701 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2702 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2703 struct btrfs_ordered_extent
*ordered
;
2705 u32 blocksize
= root
->sectorsize
;
2706 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2707 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2713 if ((offset
& (blocksize
- 1)) == 0)
2718 page
= grab_cache_page(mapping
, index
);
2722 page_start
= page_offset(page
);
2723 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2725 if (!PageUptodate(page
)) {
2726 ret
= btrfs_readpage(NULL
, page
);
2728 if (page
->mapping
!= mapping
) {
2730 page_cache_release(page
);
2733 if (!PageUptodate(page
)) {
2738 wait_on_page_writeback(page
);
2740 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2741 set_page_extent_mapped(page
);
2743 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2745 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2747 page_cache_release(page
);
2748 btrfs_start_ordered_extent(inode
, ordered
, 1);
2749 btrfs_put_ordered_extent(ordered
);
2753 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2755 if (offset
!= PAGE_CACHE_SIZE
) {
2757 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2758 flush_dcache_page(page
);
2761 ClearPageChecked(page
);
2762 set_page_dirty(page
);
2763 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2767 page_cache_release(page
);
2772 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
2774 struct btrfs_trans_handle
*trans
;
2775 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2776 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2777 struct extent_map
*em
;
2778 u64 mask
= root
->sectorsize
- 1;
2779 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2780 u64 block_end
= (size
+ mask
) & ~mask
;
2786 if (size
<= hole_start
)
2789 err
= btrfs_check_free_space(root
, 1, 0);
2793 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2796 struct btrfs_ordered_extent
*ordered
;
2797 btrfs_wait_ordered_range(inode
, hole_start
,
2798 block_end
- hole_start
);
2799 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2800 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2803 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2804 btrfs_put_ordered_extent(ordered
);
2807 trans
= btrfs_start_transaction(root
, 1);
2808 btrfs_set_trans_block_group(trans
, inode
);
2810 cur_offset
= hole_start
;
2812 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2813 block_end
- cur_offset
, 0);
2814 BUG_ON(IS_ERR(em
) || !em
);
2815 last_byte
= min(extent_map_end(em
), block_end
);
2816 last_byte
= (last_byte
+ mask
) & ~mask
;
2817 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2819 hole_size
= last_byte
- cur_offset
;
2820 err
= btrfs_drop_extents(trans
, root
, inode
,
2822 cur_offset
+ hole_size
,
2823 cur_offset
, &hint_byte
);
2826 err
= btrfs_insert_file_extent(trans
, root
,
2827 inode
->i_ino
, cur_offset
, 0,
2828 0, hole_size
, 0, hole_size
,
2830 btrfs_drop_extent_cache(inode
, hole_start
,
2833 free_extent_map(em
);
2834 cur_offset
= last_byte
;
2835 if (err
|| cur_offset
>= block_end
)
2839 btrfs_end_transaction(trans
, root
);
2840 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2844 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2846 struct inode
*inode
= dentry
->d_inode
;
2849 err
= inode_change_ok(inode
, attr
);
2853 if (S_ISREG(inode
->i_mode
) &&
2854 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
> inode
->i_size
) {
2855 err
= btrfs_cont_expand(inode
, attr
->ia_size
);
2860 err
= inode_setattr(inode
, attr
);
2862 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2863 err
= btrfs_acl_chmod(inode
);
2867 void btrfs_delete_inode(struct inode
*inode
)
2869 struct btrfs_trans_handle
*trans
;
2870 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2874 truncate_inode_pages(&inode
->i_data
, 0);
2875 if (is_bad_inode(inode
)) {
2876 btrfs_orphan_del(NULL
, inode
);
2879 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
2881 btrfs_i_size_write(inode
, 0);
2882 trans
= btrfs_join_transaction(root
, 1);
2884 btrfs_set_trans_block_group(trans
, inode
);
2885 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
2887 btrfs_orphan_del(NULL
, inode
);
2888 goto no_delete_lock
;
2891 btrfs_orphan_del(trans
, inode
);
2893 nr
= trans
->blocks_used
;
2896 btrfs_end_transaction(trans
, root
);
2897 btrfs_btree_balance_dirty(root
, nr
);
2901 nr
= trans
->blocks_used
;
2902 btrfs_end_transaction(trans
, root
);
2903 btrfs_btree_balance_dirty(root
, nr
);
2909 * this returns the key found in the dir entry in the location pointer.
2910 * If no dir entries were found, location->objectid is 0.
2912 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
2913 struct btrfs_key
*location
)
2915 const char *name
= dentry
->d_name
.name
;
2916 int namelen
= dentry
->d_name
.len
;
2917 struct btrfs_dir_item
*di
;
2918 struct btrfs_path
*path
;
2919 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2922 path
= btrfs_alloc_path();
2925 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
2930 if (!di
|| IS_ERR(di
))
2933 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
2935 btrfs_free_path(path
);
2938 location
->objectid
= 0;
2943 * when we hit a tree root in a directory, the btrfs part of the inode
2944 * needs to be changed to reflect the root directory of the tree root. This
2945 * is kind of like crossing a mount point.
2947 static int fixup_tree_root_location(struct btrfs_root
*root
,
2948 struct btrfs_key
*location
,
2949 struct btrfs_root
**sub_root
,
2950 struct dentry
*dentry
)
2952 struct btrfs_root_item
*ri
;
2954 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
2956 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
2959 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
2960 dentry
->d_name
.name
,
2961 dentry
->d_name
.len
);
2962 if (IS_ERR(*sub_root
))
2963 return PTR_ERR(*sub_root
);
2965 ri
= &(*sub_root
)->root_item
;
2966 location
->objectid
= btrfs_root_dirid(ri
);
2967 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2968 location
->offset
= 0;
2973 static noinline
void init_btrfs_i(struct inode
*inode
)
2975 struct btrfs_inode
*bi
= BTRFS_I(inode
);
2978 bi
->i_default_acl
= NULL
;
2983 bi
->logged_trans
= 0;
2984 bi
->delalloc_bytes
= 0;
2985 bi
->disk_i_size
= 0;
2987 bi
->index_cnt
= (u64
)-1;
2988 bi
->log_dirty_trans
= 0;
2989 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
2990 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
2991 inode
->i_mapping
, GFP_NOFS
);
2992 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
2993 inode
->i_mapping
, GFP_NOFS
);
2994 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
2995 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
2996 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
2997 mutex_init(&BTRFS_I(inode
)->log_mutex
);
3000 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3002 struct btrfs_iget_args
*args
= p
;
3003 inode
->i_ino
= args
->ino
;
3004 init_btrfs_i(inode
);
3005 BTRFS_I(inode
)->root
= args
->root
;
3009 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3011 struct btrfs_iget_args
*args
= opaque
;
3012 return args
->ino
== inode
->i_ino
&&
3013 args
->root
== BTRFS_I(inode
)->root
;
3016 struct inode
*btrfs_ilookup(struct super_block
*s
, u64 objectid
,
3017 struct btrfs_root
*root
, int wait
)
3019 struct inode
*inode
;
3020 struct btrfs_iget_args args
;
3021 args
.ino
= objectid
;
3025 inode
= ilookup5(s
, objectid
, btrfs_find_actor
,
3028 inode
= ilookup5_nowait(s
, objectid
, btrfs_find_actor
,
3034 struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 objectid
,
3035 struct btrfs_root
*root
)
3037 struct inode
*inode
;
3038 struct btrfs_iget_args args
;
3039 args
.ino
= objectid
;
3042 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3043 btrfs_init_locked_inode
,
3048 /* Get an inode object given its location and corresponding root.
3049 * Returns in *is_new if the inode was read from disk
3051 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3052 struct btrfs_root
*root
, int *is_new
)
3054 struct inode
*inode
;
3056 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3058 return ERR_PTR(-EACCES
);
3060 if (inode
->i_state
& I_NEW
) {
3061 BTRFS_I(inode
)->root
= root
;
3062 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3063 btrfs_read_locked_inode(inode
);
3064 unlock_new_inode(inode
);
3075 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3077 struct inode
*inode
;
3078 struct btrfs_inode
*bi
= BTRFS_I(dir
);
3079 struct btrfs_root
*root
= bi
->root
;
3080 struct btrfs_root
*sub_root
= root
;
3081 struct btrfs_key location
;
3084 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3085 return ERR_PTR(-ENAMETOOLONG
);
3087 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3090 return ERR_PTR(ret
);
3093 if (location
.objectid
) {
3094 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
3097 return ERR_PTR(ret
);
3099 return ERR_PTR(-ENOENT
);
3100 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, &new);
3102 return ERR_CAST(inode
);
3107 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
3108 struct nameidata
*nd
)
3110 struct inode
*inode
;
3112 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3113 return ERR_PTR(-ENAMETOOLONG
);
3115 inode
= btrfs_lookup_dentry(dir
, dentry
);
3117 return ERR_CAST(inode
);
3119 return d_splice_alias(inode
, dentry
);
3122 static unsigned char btrfs_filetype_table
[] = {
3123 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
3126 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
3129 struct inode
*inode
= filp
->f_dentry
->d_inode
;
3130 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3131 struct btrfs_item
*item
;
3132 struct btrfs_dir_item
*di
;
3133 struct btrfs_key key
;
3134 struct btrfs_key found_key
;
3135 struct btrfs_path
*path
;
3138 struct extent_buffer
*leaf
;
3141 unsigned char d_type
;
3146 int key_type
= BTRFS_DIR_INDEX_KEY
;
3151 /* FIXME, use a real flag for deciding about the key type */
3152 if (root
->fs_info
->tree_root
== root
)
3153 key_type
= BTRFS_DIR_ITEM_KEY
;
3155 /* special case for "." */
3156 if (filp
->f_pos
== 0) {
3157 over
= filldir(dirent
, ".", 1,
3164 /* special case for .., just use the back ref */
3165 if (filp
->f_pos
== 1) {
3166 u64 pino
= parent_ino(filp
->f_path
.dentry
);
3167 over
= filldir(dirent
, "..", 2,
3173 path
= btrfs_alloc_path();
3176 btrfs_set_key_type(&key
, key_type
);
3177 key
.offset
= filp
->f_pos
;
3178 key
.objectid
= inode
->i_ino
;
3180 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3186 leaf
= path
->nodes
[0];
3187 nritems
= btrfs_header_nritems(leaf
);
3188 slot
= path
->slots
[0];
3189 if (advance
|| slot
>= nritems
) {
3190 if (slot
>= nritems
- 1) {
3191 ret
= btrfs_next_leaf(root
, path
);
3194 leaf
= path
->nodes
[0];
3195 nritems
= btrfs_header_nritems(leaf
);
3196 slot
= path
->slots
[0];
3204 item
= btrfs_item_nr(leaf
, slot
);
3205 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3207 if (found_key
.objectid
!= key
.objectid
)
3209 if (btrfs_key_type(&found_key
) != key_type
)
3211 if (found_key
.offset
< filp
->f_pos
)
3214 filp
->f_pos
= found_key
.offset
;
3216 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
3218 di_total
= btrfs_item_size(leaf
, item
);
3220 while (di_cur
< di_total
) {
3221 struct btrfs_key location
;
3223 name_len
= btrfs_dir_name_len(leaf
, di
);
3224 if (name_len
<= sizeof(tmp_name
)) {
3225 name_ptr
= tmp_name
;
3227 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
3233 read_extent_buffer(leaf
, name_ptr
,
3234 (unsigned long)(di
+ 1), name_len
);
3236 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
3237 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
3239 /* is this a reference to our own snapshot? If so
3242 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
3243 location
.objectid
== root
->root_key
.objectid
) {
3247 over
= filldir(dirent
, name_ptr
, name_len
,
3248 found_key
.offset
, location
.objectid
,
3252 if (name_ptr
!= tmp_name
)
3257 di_len
= btrfs_dir_name_len(leaf
, di
) +
3258 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
3260 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
3264 /* Reached end of directory/root. Bump pos past the last item. */
3265 if (key_type
== BTRFS_DIR_INDEX_KEY
)
3266 filp
->f_pos
= INT_LIMIT(typeof(filp
->f_pos
));
3272 btrfs_free_path(path
);
3276 int btrfs_write_inode(struct inode
*inode
, int wait
)
3278 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3279 struct btrfs_trans_handle
*trans
;
3282 if (root
->fs_info
->btree_inode
== inode
)
3286 trans
= btrfs_join_transaction(root
, 1);
3287 btrfs_set_trans_block_group(trans
, inode
);
3288 ret
= btrfs_commit_transaction(trans
, root
);
3294 * This is somewhat expensive, updating the tree every time the
3295 * inode changes. But, it is most likely to find the inode in cache.
3296 * FIXME, needs more benchmarking...there are no reasons other than performance
3297 * to keep or drop this code.
3299 void btrfs_dirty_inode(struct inode
*inode
)
3301 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3302 struct btrfs_trans_handle
*trans
;
3304 trans
= btrfs_join_transaction(root
, 1);
3305 btrfs_set_trans_block_group(trans
, inode
);
3306 btrfs_update_inode(trans
, root
, inode
);
3307 btrfs_end_transaction(trans
, root
);
3311 * find the highest existing sequence number in a directory
3312 * and then set the in-memory index_cnt variable to reflect
3313 * free sequence numbers
3315 static int btrfs_set_inode_index_count(struct inode
*inode
)
3317 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3318 struct btrfs_key key
, found_key
;
3319 struct btrfs_path
*path
;
3320 struct extent_buffer
*leaf
;
3323 key
.objectid
= inode
->i_ino
;
3324 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
3325 key
.offset
= (u64
)-1;
3327 path
= btrfs_alloc_path();
3331 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3334 /* FIXME: we should be able to handle this */
3340 * MAGIC NUMBER EXPLANATION:
3341 * since we search a directory based on f_pos we have to start at 2
3342 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3343 * else has to start at 2
3345 if (path
->slots
[0] == 0) {
3346 BTRFS_I(inode
)->index_cnt
= 2;
3352 leaf
= path
->nodes
[0];
3353 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3355 if (found_key
.objectid
!= inode
->i_ino
||
3356 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
3357 BTRFS_I(inode
)->index_cnt
= 2;
3361 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
3363 btrfs_free_path(path
);
3368 * helper to find a free sequence number in a given directory. This current
3369 * code is very simple, later versions will do smarter things in the btree
3371 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
3375 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
3376 ret
= btrfs_set_inode_index_count(dir
);
3381 *index
= BTRFS_I(dir
)->index_cnt
;
3382 BTRFS_I(dir
)->index_cnt
++;
3387 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
3388 struct btrfs_root
*root
,
3390 const char *name
, int name_len
,
3391 u64 ref_objectid
, u64 objectid
,
3392 u64 alloc_hint
, int mode
, u64
*index
)
3394 struct inode
*inode
;
3395 struct btrfs_inode_item
*inode_item
;
3396 struct btrfs_key
*location
;
3397 struct btrfs_path
*path
;
3398 struct btrfs_inode_ref
*ref
;
3399 struct btrfs_key key
[2];
3405 path
= btrfs_alloc_path();
3408 inode
= new_inode(root
->fs_info
->sb
);
3410 return ERR_PTR(-ENOMEM
);
3413 ret
= btrfs_set_inode_index(dir
, index
);
3415 return ERR_PTR(ret
);
3418 * index_cnt is ignored for everything but a dir,
3419 * btrfs_get_inode_index_count has an explanation for the magic
3422 init_btrfs_i(inode
);
3423 BTRFS_I(inode
)->index_cnt
= 2;
3424 BTRFS_I(inode
)->root
= root
;
3425 BTRFS_I(inode
)->generation
= trans
->transid
;
3431 BTRFS_I(inode
)->block_group
=
3432 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
3433 if ((mode
& S_IFREG
)) {
3434 if (btrfs_test_opt(root
, NODATASUM
))
3435 btrfs_set_flag(inode
, NODATASUM
);
3436 if (btrfs_test_opt(root
, NODATACOW
))
3437 btrfs_set_flag(inode
, NODATACOW
);
3440 key
[0].objectid
= objectid
;
3441 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
3444 key
[1].objectid
= objectid
;
3445 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
3446 key
[1].offset
= ref_objectid
;
3448 sizes
[0] = sizeof(struct btrfs_inode_item
);
3449 sizes
[1] = name_len
+ sizeof(*ref
);
3451 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
3455 if (objectid
> root
->highest_inode
)
3456 root
->highest_inode
= objectid
;
3458 inode
->i_uid
= current_fsuid();
3459 inode
->i_gid
= current_fsgid();
3460 inode
->i_mode
= mode
;
3461 inode
->i_ino
= objectid
;
3462 inode_set_bytes(inode
, 0);
3463 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3464 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3465 struct btrfs_inode_item
);
3466 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
3468 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3469 struct btrfs_inode_ref
);
3470 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
3471 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
3472 ptr
= (unsigned long)(ref
+ 1);
3473 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
3475 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3476 btrfs_free_path(path
);
3478 location
= &BTRFS_I(inode
)->location
;
3479 location
->objectid
= objectid
;
3480 location
->offset
= 0;
3481 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3483 insert_inode_hash(inode
);
3487 BTRFS_I(dir
)->index_cnt
--;
3488 btrfs_free_path(path
);
3489 return ERR_PTR(ret
);
3492 static inline u8
btrfs_inode_type(struct inode
*inode
)
3494 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
3498 * utility function to add 'inode' into 'parent_inode' with
3499 * a give name and a given sequence number.
3500 * if 'add_backref' is true, also insert a backref from the
3501 * inode to the parent directory.
3503 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3504 struct inode
*parent_inode
, struct inode
*inode
,
3505 const char *name
, int name_len
, int add_backref
, u64 index
)
3508 struct btrfs_key key
;
3509 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3511 key
.objectid
= inode
->i_ino
;
3512 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3515 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3516 parent_inode
->i_ino
,
3517 &key
, btrfs_inode_type(inode
),
3521 ret
= btrfs_insert_inode_ref(trans
, root
,
3524 parent_inode
->i_ino
,
3527 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3529 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3530 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3535 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3536 struct dentry
*dentry
, struct inode
*inode
,
3537 int backref
, u64 index
)
3539 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3540 inode
, dentry
->d_name
.name
,
3541 dentry
->d_name
.len
, backref
, index
);
3543 d_instantiate(dentry
, inode
);
3551 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3552 int mode
, dev_t rdev
)
3554 struct btrfs_trans_handle
*trans
;
3555 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3556 struct inode
*inode
= NULL
;
3560 unsigned long nr
= 0;
3563 if (!new_valid_dev(rdev
))
3566 err
= btrfs_check_free_space(root
, 1, 0);
3570 trans
= btrfs_start_transaction(root
, 1);
3571 btrfs_set_trans_block_group(trans
, dir
);
3573 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3579 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3581 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3582 BTRFS_I(dir
)->block_group
, mode
, &index
);
3583 err
= PTR_ERR(inode
);
3587 err
= btrfs_init_acl(inode
, dir
);
3593 btrfs_set_trans_block_group(trans
, inode
);
3594 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3598 inode
->i_op
= &btrfs_special_inode_operations
;
3599 init_special_inode(inode
, inode
->i_mode
, rdev
);
3600 btrfs_update_inode(trans
, root
, inode
);
3602 dir
->i_sb
->s_dirt
= 1;
3603 btrfs_update_inode_block_group(trans
, inode
);
3604 btrfs_update_inode_block_group(trans
, dir
);
3606 nr
= trans
->blocks_used
;
3607 btrfs_end_transaction_throttle(trans
, root
);
3610 inode_dec_link_count(inode
);
3613 btrfs_btree_balance_dirty(root
, nr
);
3617 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3618 int mode
, struct nameidata
*nd
)
3620 struct btrfs_trans_handle
*trans
;
3621 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3622 struct inode
*inode
= NULL
;
3625 unsigned long nr
= 0;
3629 err
= btrfs_check_free_space(root
, 1, 0);
3632 trans
= btrfs_start_transaction(root
, 1);
3633 btrfs_set_trans_block_group(trans
, dir
);
3635 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3641 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3643 dentry
->d_parent
->d_inode
->i_ino
,
3644 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3646 err
= PTR_ERR(inode
);
3650 err
= btrfs_init_acl(inode
, dir
);
3656 btrfs_set_trans_block_group(trans
, inode
);
3657 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3661 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3662 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3663 inode
->i_fop
= &btrfs_file_operations
;
3664 inode
->i_op
= &btrfs_file_inode_operations
;
3665 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3667 dir
->i_sb
->s_dirt
= 1;
3668 btrfs_update_inode_block_group(trans
, inode
);
3669 btrfs_update_inode_block_group(trans
, dir
);
3671 nr
= trans
->blocks_used
;
3672 btrfs_end_transaction_throttle(trans
, root
);
3675 inode_dec_link_count(inode
);
3678 btrfs_btree_balance_dirty(root
, nr
);
3682 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3683 struct dentry
*dentry
)
3685 struct btrfs_trans_handle
*trans
;
3686 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3687 struct inode
*inode
= old_dentry
->d_inode
;
3689 unsigned long nr
= 0;
3693 if (inode
->i_nlink
== 0)
3696 btrfs_inc_nlink(inode
);
3697 err
= btrfs_check_free_space(root
, 1, 0);
3700 err
= btrfs_set_inode_index(dir
, &index
);
3704 trans
= btrfs_start_transaction(root
, 1);
3706 btrfs_set_trans_block_group(trans
, dir
);
3707 atomic_inc(&inode
->i_count
);
3709 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3714 dir
->i_sb
->s_dirt
= 1;
3715 btrfs_update_inode_block_group(trans
, dir
);
3716 err
= btrfs_update_inode(trans
, root
, inode
);
3721 nr
= trans
->blocks_used
;
3722 btrfs_end_transaction_throttle(trans
, root
);
3725 inode_dec_link_count(inode
);
3728 btrfs_btree_balance_dirty(root
, nr
);
3732 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3734 struct inode
*inode
= NULL
;
3735 struct btrfs_trans_handle
*trans
;
3736 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3738 int drop_on_err
= 0;
3741 unsigned long nr
= 1;
3743 err
= btrfs_check_free_space(root
, 1, 0);
3747 trans
= btrfs_start_transaction(root
, 1);
3748 btrfs_set_trans_block_group(trans
, dir
);
3750 if (IS_ERR(trans
)) {
3751 err
= PTR_ERR(trans
);
3755 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3761 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3763 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3764 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3766 if (IS_ERR(inode
)) {
3767 err
= PTR_ERR(inode
);
3773 err
= btrfs_init_acl(inode
, dir
);
3777 inode
->i_op
= &btrfs_dir_inode_operations
;
3778 inode
->i_fop
= &btrfs_dir_file_operations
;
3779 btrfs_set_trans_block_group(trans
, inode
);
3781 btrfs_i_size_write(inode
, 0);
3782 err
= btrfs_update_inode(trans
, root
, inode
);
3786 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3787 inode
, dentry
->d_name
.name
,
3788 dentry
->d_name
.len
, 0, index
);
3792 d_instantiate(dentry
, inode
);
3794 dir
->i_sb
->s_dirt
= 1;
3795 btrfs_update_inode_block_group(trans
, inode
);
3796 btrfs_update_inode_block_group(trans
, dir
);
3799 nr
= trans
->blocks_used
;
3800 btrfs_end_transaction_throttle(trans
, root
);
3805 btrfs_btree_balance_dirty(root
, nr
);
3809 /* helper for btfs_get_extent. Given an existing extent in the tree,
3810 * and an extent that you want to insert, deal with overlap and insert
3811 * the new extent into the tree.
3813 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3814 struct extent_map
*existing
,
3815 struct extent_map
*em
,
3816 u64 map_start
, u64 map_len
)
3820 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3821 start_diff
= map_start
- em
->start
;
3822 em
->start
= map_start
;
3824 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3825 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3826 em
->block_start
+= start_diff
;
3827 em
->block_len
-= start_diff
;
3829 return add_extent_mapping(em_tree
, em
);
3832 static noinline
int uncompress_inline(struct btrfs_path
*path
,
3833 struct inode
*inode
, struct page
*page
,
3834 size_t pg_offset
, u64 extent_offset
,
3835 struct btrfs_file_extent_item
*item
)
3838 struct extent_buffer
*leaf
= path
->nodes
[0];
3841 unsigned long inline_size
;
3844 WARN_ON(pg_offset
!= 0);
3845 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
3846 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
3847 btrfs_item_nr(leaf
, path
->slots
[0]));
3848 tmp
= kmalloc(inline_size
, GFP_NOFS
);
3849 ptr
= btrfs_file_extent_inline_start(item
);
3851 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
3853 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
3854 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
3855 inline_size
, max_size
);
3857 char *kaddr
= kmap_atomic(page
, KM_USER0
);
3858 unsigned long copy_size
= min_t(u64
,
3859 PAGE_CACHE_SIZE
- pg_offset
,
3860 max_size
- extent_offset
);
3861 memset(kaddr
+ pg_offset
, 0, copy_size
);
3862 kunmap_atomic(kaddr
, KM_USER0
);
3869 * a bit scary, this does extent mapping from logical file offset to the disk.
3870 * the ugly parts come from merging extents from the disk with the in-ram
3871 * representation. This gets more complex because of the data=ordered code,
3872 * where the in-ram extents might be locked pending data=ordered completion.
3874 * This also copies inline extents directly into the page.
3877 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
3878 size_t pg_offset
, u64 start
, u64 len
,
3884 u64 extent_start
= 0;
3886 u64 objectid
= inode
->i_ino
;
3888 struct btrfs_path
*path
= NULL
;
3889 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3890 struct btrfs_file_extent_item
*item
;
3891 struct extent_buffer
*leaf
;
3892 struct btrfs_key found_key
;
3893 struct extent_map
*em
= NULL
;
3894 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3895 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3896 struct btrfs_trans_handle
*trans
= NULL
;
3900 spin_lock(&em_tree
->lock
);
3901 em
= lookup_extent_mapping(em_tree
, start
, len
);
3903 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3904 spin_unlock(&em_tree
->lock
);
3907 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
3908 free_extent_map(em
);
3909 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
3910 free_extent_map(em
);
3914 em
= alloc_extent_map(GFP_NOFS
);
3919 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3920 em
->start
= EXTENT_MAP_HOLE
;
3921 em
->orig_start
= EXTENT_MAP_HOLE
;
3923 em
->block_len
= (u64
)-1;
3926 path
= btrfs_alloc_path();
3930 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3931 objectid
, start
, trans
!= NULL
);
3938 if (path
->slots
[0] == 0)
3943 leaf
= path
->nodes
[0];
3944 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3945 struct btrfs_file_extent_item
);
3946 /* are we inside the extent that was found? */
3947 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3948 found_type
= btrfs_key_type(&found_key
);
3949 if (found_key
.objectid
!= objectid
||
3950 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3954 found_type
= btrfs_file_extent_type(leaf
, item
);
3955 extent_start
= found_key
.offset
;
3956 compressed
= btrfs_file_extent_compression(leaf
, item
);
3957 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3958 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3959 extent_end
= extent_start
+
3960 btrfs_file_extent_num_bytes(leaf
, item
);
3961 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3963 size
= btrfs_file_extent_inline_len(leaf
, item
);
3964 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
3965 ~((u64
)root
->sectorsize
- 1);
3968 if (start
>= extent_end
) {
3970 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3971 ret
= btrfs_next_leaf(root
, path
);
3978 leaf
= path
->nodes
[0];
3980 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3981 if (found_key
.objectid
!= objectid
||
3982 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3984 if (start
+ len
<= found_key
.offset
)
3987 em
->len
= found_key
.offset
- start
;
3991 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3992 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3993 em
->start
= extent_start
;
3994 em
->len
= extent_end
- extent_start
;
3995 em
->orig_start
= extent_start
-
3996 btrfs_file_extent_offset(leaf
, item
);
3997 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
3999 em
->block_start
= EXTENT_MAP_HOLE
;
4003 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4004 em
->block_start
= bytenr
;
4005 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
4008 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
4009 em
->block_start
= bytenr
;
4010 em
->block_len
= em
->len
;
4011 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
4012 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
4015 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4019 size_t extent_offset
;
4022 em
->block_start
= EXTENT_MAP_INLINE
;
4023 if (!page
|| create
) {
4024 em
->start
= extent_start
;
4025 em
->len
= extent_end
- extent_start
;
4029 size
= btrfs_file_extent_inline_len(leaf
, item
);
4030 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
4031 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
4032 size
- extent_offset
);
4033 em
->start
= extent_start
+ extent_offset
;
4034 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
4035 ~((u64
)root
->sectorsize
- 1);
4036 em
->orig_start
= EXTENT_MAP_INLINE
;
4038 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4039 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
4040 if (create
== 0 && !PageUptodate(page
)) {
4041 if (btrfs_file_extent_compression(leaf
, item
) ==
4042 BTRFS_COMPRESS_ZLIB
) {
4043 ret
= uncompress_inline(path
, inode
, page
,
4045 extent_offset
, item
);
4049 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4053 flush_dcache_page(page
);
4054 } else if (create
&& PageUptodate(page
)) {
4057 free_extent_map(em
);
4059 btrfs_release_path(root
, path
);
4060 trans
= btrfs_join_transaction(root
, 1);
4064 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4067 btrfs_mark_buffer_dirty(leaf
);
4069 set_extent_uptodate(io_tree
, em
->start
,
4070 extent_map_end(em
) - 1, GFP_NOFS
);
4073 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
4080 em
->block_start
= EXTENT_MAP_HOLE
;
4081 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
4083 btrfs_release_path(root
, path
);
4084 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
4085 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
4086 "[%llu %llu]\n", (unsigned long long)em
->start
,
4087 (unsigned long long)em
->len
,
4088 (unsigned long long)start
,
4089 (unsigned long long)len
);
4095 spin_lock(&em_tree
->lock
);
4096 ret
= add_extent_mapping(em_tree
, em
);
4097 /* it is possible that someone inserted the extent into the tree
4098 * while we had the lock dropped. It is also possible that
4099 * an overlapping map exists in the tree
4101 if (ret
== -EEXIST
) {
4102 struct extent_map
*existing
;
4106 existing
= lookup_extent_mapping(em_tree
, start
, len
);
4107 if (existing
&& (existing
->start
> start
||
4108 existing
->start
+ existing
->len
<= start
)) {
4109 free_extent_map(existing
);
4113 existing
= lookup_extent_mapping(em_tree
, em
->start
,
4116 err
= merge_extent_mapping(em_tree
, existing
,
4119 free_extent_map(existing
);
4121 free_extent_map(em
);
4126 free_extent_map(em
);
4130 free_extent_map(em
);
4135 spin_unlock(&em_tree
->lock
);
4138 btrfs_free_path(path
);
4140 ret
= btrfs_end_transaction(trans
, root
);
4145 free_extent_map(em
);
4147 return ERR_PTR(err
);
4152 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
4153 const struct iovec
*iov
, loff_t offset
,
4154 unsigned long nr_segs
)
4159 int btrfs_readpage(struct file
*file
, struct page
*page
)
4161 struct extent_io_tree
*tree
;
4162 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4163 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
4166 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
4168 struct extent_io_tree
*tree
;
4171 if (current
->flags
& PF_MEMALLOC
) {
4172 redirty_page_for_writepage(wbc
, page
);
4176 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4177 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
4180 int btrfs_writepages(struct address_space
*mapping
,
4181 struct writeback_control
*wbc
)
4183 struct extent_io_tree
*tree
;
4185 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4186 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
4190 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
4191 struct list_head
*pages
, unsigned nr_pages
)
4193 struct extent_io_tree
*tree
;
4194 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4195 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
4198 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4200 struct extent_io_tree
*tree
;
4201 struct extent_map_tree
*map
;
4204 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4205 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
4206 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
4208 ClearPagePrivate(page
);
4209 set_page_private(page
, 0);
4210 page_cache_release(page
);
4215 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4217 if (PageWriteback(page
) || PageDirty(page
))
4219 return __btrfs_releasepage(page
, gfp_flags
);
4222 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
4224 struct extent_io_tree
*tree
;
4225 struct btrfs_ordered_extent
*ordered
;
4226 u64 page_start
= page_offset(page
);
4227 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4229 wait_on_page_writeback(page
);
4230 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4232 btrfs_releasepage(page
, GFP_NOFS
);
4236 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4237 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
4241 * IO on this page will never be started, so we need
4242 * to account for any ordered extents now
4244 clear_extent_bit(tree
, page_start
, page_end
,
4245 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4246 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
4247 btrfs_finish_ordered_io(page
->mapping
->host
,
4248 page_start
, page_end
);
4249 btrfs_put_ordered_extent(ordered
);
4250 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4252 clear_extent_bit(tree
, page_start
, page_end
,
4253 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4256 __btrfs_releasepage(page
, GFP_NOFS
);
4258 ClearPageChecked(page
);
4259 if (PagePrivate(page
)) {
4260 ClearPagePrivate(page
);
4261 set_page_private(page
, 0);
4262 page_cache_release(page
);
4267 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4268 * called from a page fault handler when a page is first dirtied. Hence we must
4269 * be careful to check for EOF conditions here. We set the page up correctly
4270 * for a written page which means we get ENOSPC checking when writing into
4271 * holes and correct delalloc and unwritten extent mapping on filesystems that
4272 * support these features.
4274 * We are not allowed to take the i_mutex here so we have to play games to
4275 * protect against truncate races as the page could now be beyond EOF. Because
4276 * vmtruncate() writes the inode size before removing pages, once we have the
4277 * page lock we can determine safely if the page is beyond EOF. If it is not
4278 * beyond EOF, then the page is guaranteed safe against truncation until we
4281 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4283 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
4284 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4285 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4286 struct btrfs_ordered_extent
*ordered
;
4288 unsigned long zero_start
;
4294 ret
= btrfs_check_free_space(root
, PAGE_CACHE_SIZE
, 0);
4301 size
= i_size_read(inode
);
4302 page_start
= page_offset(page
);
4303 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4305 if ((page
->mapping
!= inode
->i_mapping
) ||
4306 (page_start
>= size
)) {
4307 /* page got truncated out from underneath us */
4310 wait_on_page_writeback(page
);
4312 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4313 set_page_extent_mapped(page
);
4316 * we can't set the delalloc bits if there are pending ordered
4317 * extents. Drop our locks and wait for them to finish
4319 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4321 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4323 btrfs_start_ordered_extent(inode
, ordered
, 1);
4324 btrfs_put_ordered_extent(ordered
);
4328 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
4331 /* page is wholly or partially inside EOF */
4332 if (page_start
+ PAGE_CACHE_SIZE
> size
)
4333 zero_start
= size
& ~PAGE_CACHE_MASK
;
4335 zero_start
= PAGE_CACHE_SIZE
;
4337 if (zero_start
!= PAGE_CACHE_SIZE
) {
4339 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
4340 flush_dcache_page(page
);
4343 ClearPageChecked(page
);
4344 set_page_dirty(page
);
4345 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4353 static void btrfs_truncate(struct inode
*inode
)
4355 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4357 struct btrfs_trans_handle
*trans
;
4359 u64 mask
= root
->sectorsize
- 1;
4361 if (!S_ISREG(inode
->i_mode
))
4363 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4366 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
4367 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
4369 trans
= btrfs_start_transaction(root
, 1);
4370 btrfs_set_trans_block_group(trans
, inode
);
4371 btrfs_i_size_write(inode
, inode
->i_size
);
4373 ret
= btrfs_orphan_add(trans
, inode
);
4376 /* FIXME, add redo link to tree so we don't leak on crash */
4377 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
4378 BTRFS_EXTENT_DATA_KEY
);
4379 btrfs_update_inode(trans
, root
, inode
);
4381 ret
= btrfs_orphan_del(trans
, inode
);
4385 nr
= trans
->blocks_used
;
4386 ret
= btrfs_end_transaction_throttle(trans
, root
);
4388 btrfs_btree_balance_dirty(root
, nr
);
4392 * create a new subvolume directory/inode (helper for the ioctl).
4394 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
4395 struct btrfs_root
*new_root
, struct dentry
*dentry
,
4396 u64 new_dirid
, u64 alloc_hint
)
4398 struct inode
*inode
;
4402 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
4403 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
4405 return PTR_ERR(inode
);
4406 inode
->i_op
= &btrfs_dir_inode_operations
;
4407 inode
->i_fop
= &btrfs_dir_file_operations
;
4410 btrfs_i_size_write(inode
, 0);
4412 error
= btrfs_update_inode(trans
, new_root
, inode
);
4416 d_instantiate(dentry
, inode
);
4420 /* helper function for file defrag and space balancing. This
4421 * forces readahead on a given range of bytes in an inode
4423 unsigned long btrfs_force_ra(struct address_space
*mapping
,
4424 struct file_ra_state
*ra
, struct file
*file
,
4425 pgoff_t offset
, pgoff_t last_index
)
4427 pgoff_t req_size
= last_index
- offset
+ 1;
4429 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
4430 return offset
+ req_size
;
4433 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
4435 struct btrfs_inode
*ei
;
4437 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
4441 ei
->logged_trans
= 0;
4442 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
4443 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
4444 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
4445 INIT_LIST_HEAD(&ei
->i_orphan
);
4446 return &ei
->vfs_inode
;
4449 void btrfs_destroy_inode(struct inode
*inode
)
4451 struct btrfs_ordered_extent
*ordered
;
4452 WARN_ON(!list_empty(&inode
->i_dentry
));
4453 WARN_ON(inode
->i_data
.nrpages
);
4455 if (BTRFS_I(inode
)->i_acl
&&
4456 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
4457 posix_acl_release(BTRFS_I(inode
)->i_acl
);
4458 if (BTRFS_I(inode
)->i_default_acl
&&
4459 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
4460 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
4462 spin_lock(&BTRFS_I(inode
)->root
->list_lock
);
4463 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
4464 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
4465 " list\n", inode
->i_ino
);
4468 spin_unlock(&BTRFS_I(inode
)->root
->list_lock
);
4471 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
4475 printk(KERN_ERR
"btrfs found ordered "
4476 "extent %llu %llu on inode cleanup\n",
4477 (unsigned long long)ordered
->file_offset
,
4478 (unsigned long long)ordered
->len
);
4479 btrfs_remove_ordered_extent(inode
, ordered
);
4480 btrfs_put_ordered_extent(ordered
);
4481 btrfs_put_ordered_extent(ordered
);
4484 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4485 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4488 static void init_once(void *foo
)
4490 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4492 inode_init_once(&ei
->vfs_inode
);
4495 void btrfs_destroy_cachep(void)
4497 if (btrfs_inode_cachep
)
4498 kmem_cache_destroy(btrfs_inode_cachep
);
4499 if (btrfs_trans_handle_cachep
)
4500 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4501 if (btrfs_transaction_cachep
)
4502 kmem_cache_destroy(btrfs_transaction_cachep
);
4503 if (btrfs_bit_radix_cachep
)
4504 kmem_cache_destroy(btrfs_bit_radix_cachep
);
4505 if (btrfs_path_cachep
)
4506 kmem_cache_destroy(btrfs_path_cachep
);
4509 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
4510 unsigned long extra_flags
,
4511 void (*ctor
)(void *))
4513 return kmem_cache_create(name
, size
, 0, (SLAB_RECLAIM_ACCOUNT
|
4514 SLAB_MEM_SPREAD
| extra_flags
), ctor
);
4517 int btrfs_init_cachep(void)
4519 btrfs_inode_cachep
= btrfs_cache_create("btrfs_inode_cache",
4520 sizeof(struct btrfs_inode
),
4522 if (!btrfs_inode_cachep
)
4524 btrfs_trans_handle_cachep
=
4525 btrfs_cache_create("btrfs_trans_handle_cache",
4526 sizeof(struct btrfs_trans_handle
),
4528 if (!btrfs_trans_handle_cachep
)
4530 btrfs_transaction_cachep
= btrfs_cache_create("btrfs_transaction_cache",
4531 sizeof(struct btrfs_transaction
),
4533 if (!btrfs_transaction_cachep
)
4535 btrfs_path_cachep
= btrfs_cache_create("btrfs_path_cache",
4536 sizeof(struct btrfs_path
),
4538 if (!btrfs_path_cachep
)
4540 btrfs_bit_radix_cachep
= btrfs_cache_create("btrfs_radix", 256,
4541 SLAB_DESTROY_BY_RCU
, NULL
);
4542 if (!btrfs_bit_radix_cachep
)
4546 btrfs_destroy_cachep();
4550 static int btrfs_getattr(struct vfsmount
*mnt
,
4551 struct dentry
*dentry
, struct kstat
*stat
)
4553 struct inode
*inode
= dentry
->d_inode
;
4554 generic_fillattr(inode
, stat
);
4555 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
4556 stat
->blksize
= PAGE_CACHE_SIZE
;
4557 stat
->blocks
= (inode_get_bytes(inode
) +
4558 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4562 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4563 struct inode
*new_dir
, struct dentry
*new_dentry
)
4565 struct btrfs_trans_handle
*trans
;
4566 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4567 struct inode
*new_inode
= new_dentry
->d_inode
;
4568 struct inode
*old_inode
= old_dentry
->d_inode
;
4569 struct timespec ctime
= CURRENT_TIME
;
4573 /* we're not allowed to rename between subvolumes */
4574 if (BTRFS_I(old_inode
)->root
->root_key
.objectid
!=
4575 BTRFS_I(new_dir
)->root
->root_key
.objectid
)
4578 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4579 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4583 /* to rename a snapshot or subvolume, we need to juggle the
4584 * backrefs. This isn't coded yet
4586 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
4589 ret
= btrfs_check_free_space(root
, 1, 0);
4593 trans
= btrfs_start_transaction(root
, 1);
4595 btrfs_set_trans_block_group(trans
, new_dir
);
4597 btrfs_inc_nlink(old_dentry
->d_inode
);
4598 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4599 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4600 old_inode
->i_ctime
= ctime
;
4602 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4603 old_dentry
->d_name
.name
,
4604 old_dentry
->d_name
.len
);
4609 new_inode
->i_ctime
= CURRENT_TIME
;
4610 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4611 new_dentry
->d_inode
,
4612 new_dentry
->d_name
.name
,
4613 new_dentry
->d_name
.len
);
4616 if (new_inode
->i_nlink
== 0) {
4617 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4623 ret
= btrfs_set_inode_index(new_dir
, &index
);
4627 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4628 old_inode
, new_dentry
->d_name
.name
,
4629 new_dentry
->d_name
.len
, 1, index
);
4634 btrfs_end_transaction_throttle(trans
, root
);
4640 * some fairly slow code that needs optimization. This walks the list
4641 * of all the inodes with pending delalloc and forces them to disk.
4643 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4645 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4646 struct btrfs_inode
*binode
;
4647 struct inode
*inode
;
4649 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
4652 spin_lock(&root
->fs_info
->delalloc_lock
);
4653 while (!list_empty(head
)) {
4654 binode
= list_entry(head
->next
, struct btrfs_inode
,
4656 inode
= igrab(&binode
->vfs_inode
);
4658 list_del_init(&binode
->delalloc_inodes
);
4659 spin_unlock(&root
->fs_info
->delalloc_lock
);
4661 filemap_flush(inode
->i_mapping
);
4665 spin_lock(&root
->fs_info
->delalloc_lock
);
4667 spin_unlock(&root
->fs_info
->delalloc_lock
);
4669 /* the filemap_flush will queue IO into the worker threads, but
4670 * we have to make sure the IO is actually started and that
4671 * ordered extents get created before we return
4673 atomic_inc(&root
->fs_info
->async_submit_draining
);
4674 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
4675 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
4676 wait_event(root
->fs_info
->async_submit_wait
,
4677 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
4678 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
4680 atomic_dec(&root
->fs_info
->async_submit_draining
);
4684 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4685 const char *symname
)
4687 struct btrfs_trans_handle
*trans
;
4688 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4689 struct btrfs_path
*path
;
4690 struct btrfs_key key
;
4691 struct inode
*inode
= NULL
;
4699 struct btrfs_file_extent_item
*ei
;
4700 struct extent_buffer
*leaf
;
4701 unsigned long nr
= 0;
4703 name_len
= strlen(symname
) + 1;
4704 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4705 return -ENAMETOOLONG
;
4707 err
= btrfs_check_free_space(root
, 1, 0);
4711 trans
= btrfs_start_transaction(root
, 1);
4712 btrfs_set_trans_block_group(trans
, dir
);
4714 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4720 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4722 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4723 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4725 err
= PTR_ERR(inode
);
4729 err
= btrfs_init_acl(inode
, dir
);
4735 btrfs_set_trans_block_group(trans
, inode
);
4736 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4740 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4741 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4742 inode
->i_fop
= &btrfs_file_operations
;
4743 inode
->i_op
= &btrfs_file_inode_operations
;
4744 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4746 dir
->i_sb
->s_dirt
= 1;
4747 btrfs_update_inode_block_group(trans
, inode
);
4748 btrfs_update_inode_block_group(trans
, dir
);
4752 path
= btrfs_alloc_path();
4754 key
.objectid
= inode
->i_ino
;
4756 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
4757 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
4758 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
4764 leaf
= path
->nodes
[0];
4765 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
4766 struct btrfs_file_extent_item
);
4767 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
4768 btrfs_set_file_extent_type(leaf
, ei
,
4769 BTRFS_FILE_EXTENT_INLINE
);
4770 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
4771 btrfs_set_file_extent_compression(leaf
, ei
, 0);
4772 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
4773 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
4775 ptr
= btrfs_file_extent_inline_start(ei
);
4776 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
4777 btrfs_mark_buffer_dirty(leaf
);
4778 btrfs_free_path(path
);
4780 inode
->i_op
= &btrfs_symlink_inode_operations
;
4781 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
4782 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4783 inode_set_bytes(inode
, name_len
);
4784 btrfs_i_size_write(inode
, name_len
- 1);
4785 err
= btrfs_update_inode(trans
, root
, inode
);
4790 nr
= trans
->blocks_used
;
4791 btrfs_end_transaction_throttle(trans
, root
);
4794 inode_dec_link_count(inode
);
4797 btrfs_btree_balance_dirty(root
, nr
);
4801 static int prealloc_file_range(struct inode
*inode
, u64 start
, u64 end
,
4802 u64 alloc_hint
, int mode
)
4804 struct btrfs_trans_handle
*trans
;
4805 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4806 struct btrfs_key ins
;
4808 u64 cur_offset
= start
;
4809 u64 num_bytes
= end
- start
;
4812 trans
= btrfs_join_transaction(root
, 1);
4814 btrfs_set_trans_block_group(trans
, inode
);
4816 while (num_bytes
> 0) {
4817 alloc_size
= min(num_bytes
, root
->fs_info
->max_extent
);
4818 ret
= btrfs_reserve_extent(trans
, root
, alloc_size
,
4819 root
->sectorsize
, 0, alloc_hint
,
4825 ret
= insert_reserved_file_extent(trans
, inode
,
4826 cur_offset
, ins
.objectid
,
4827 ins
.offset
, ins
.offset
,
4828 ins
.offset
, 0, 0, 0,
4829 BTRFS_FILE_EXTENT_PREALLOC
);
4831 num_bytes
-= ins
.offset
;
4832 cur_offset
+= ins
.offset
;
4833 alloc_hint
= ins
.objectid
+ ins
.offset
;
4836 if (cur_offset
> start
) {
4837 inode
->i_ctime
= CURRENT_TIME
;
4838 btrfs_set_flag(inode
, PREALLOC
);
4839 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4840 cur_offset
> i_size_read(inode
))
4841 btrfs_i_size_write(inode
, cur_offset
);
4842 ret
= btrfs_update_inode(trans
, root
, inode
);
4846 btrfs_end_transaction(trans
, root
);
4850 static long btrfs_fallocate(struct inode
*inode
, int mode
,
4851 loff_t offset
, loff_t len
)
4858 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
4859 struct extent_map
*em
;
4862 alloc_start
= offset
& ~mask
;
4863 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
4865 mutex_lock(&inode
->i_mutex
);
4866 if (alloc_start
> inode
->i_size
) {
4867 ret
= btrfs_cont_expand(inode
, alloc_start
);
4873 struct btrfs_ordered_extent
*ordered
;
4874 lock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
,
4875 alloc_end
- 1, GFP_NOFS
);
4876 ordered
= btrfs_lookup_first_ordered_extent(inode
,
4879 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
4880 ordered
->file_offset
< alloc_end
) {
4881 btrfs_put_ordered_extent(ordered
);
4882 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4883 alloc_start
, alloc_end
- 1, GFP_NOFS
);
4884 btrfs_wait_ordered_range(inode
, alloc_start
,
4885 alloc_end
- alloc_start
);
4888 btrfs_put_ordered_extent(ordered
);
4893 cur_offset
= alloc_start
;
4895 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4896 alloc_end
- cur_offset
, 0);
4897 BUG_ON(IS_ERR(em
) || !em
);
4898 last_byte
= min(extent_map_end(em
), alloc_end
);
4899 last_byte
= (last_byte
+ mask
) & ~mask
;
4900 if (em
->block_start
== EXTENT_MAP_HOLE
) {
4901 ret
= prealloc_file_range(inode
, cur_offset
,
4902 last_byte
, alloc_hint
, mode
);
4904 free_extent_map(em
);
4908 if (em
->block_start
<= EXTENT_MAP_LAST_BYTE
)
4909 alloc_hint
= em
->block_start
;
4910 free_extent_map(em
);
4912 cur_offset
= last_byte
;
4913 if (cur_offset
>= alloc_end
) {
4918 unlock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, alloc_end
- 1,
4921 mutex_unlock(&inode
->i_mutex
);
4925 static int btrfs_set_page_dirty(struct page
*page
)
4927 return __set_page_dirty_nobuffers(page
);
4930 static int btrfs_permission(struct inode
*inode
, int mask
)
4932 if (btrfs_test_flag(inode
, READONLY
) && (mask
& MAY_WRITE
))
4934 return generic_permission(inode
, mask
, btrfs_check_acl
);
4937 static struct inode_operations btrfs_dir_inode_operations
= {
4938 .getattr
= btrfs_getattr
,
4939 .lookup
= btrfs_lookup
,
4940 .create
= btrfs_create
,
4941 .unlink
= btrfs_unlink
,
4943 .mkdir
= btrfs_mkdir
,
4944 .rmdir
= btrfs_rmdir
,
4945 .rename
= btrfs_rename
,
4946 .symlink
= btrfs_symlink
,
4947 .setattr
= btrfs_setattr
,
4948 .mknod
= btrfs_mknod
,
4949 .setxattr
= btrfs_setxattr
,
4950 .getxattr
= btrfs_getxattr
,
4951 .listxattr
= btrfs_listxattr
,
4952 .removexattr
= btrfs_removexattr
,
4953 .permission
= btrfs_permission
,
4955 static struct inode_operations btrfs_dir_ro_inode_operations
= {
4956 .lookup
= btrfs_lookup
,
4957 .permission
= btrfs_permission
,
4959 static struct file_operations btrfs_dir_file_operations
= {
4960 .llseek
= generic_file_llseek
,
4961 .read
= generic_read_dir
,
4962 .readdir
= btrfs_real_readdir
,
4963 .unlocked_ioctl
= btrfs_ioctl
,
4964 #ifdef CONFIG_COMPAT
4965 .compat_ioctl
= btrfs_ioctl
,
4967 .release
= btrfs_release_file
,
4968 .fsync
= btrfs_sync_file
,
4971 static struct extent_io_ops btrfs_extent_io_ops
= {
4972 .fill_delalloc
= run_delalloc_range
,
4973 .submit_bio_hook
= btrfs_submit_bio_hook
,
4974 .merge_bio_hook
= btrfs_merge_bio_hook
,
4975 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
4976 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
4977 .writepage_start_hook
= btrfs_writepage_start_hook
,
4978 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
4979 .set_bit_hook
= btrfs_set_bit_hook
,
4980 .clear_bit_hook
= btrfs_clear_bit_hook
,
4984 * btrfs doesn't support the bmap operation because swapfiles
4985 * use bmap to make a mapping of extents in the file. They assume
4986 * these extents won't change over the life of the file and they
4987 * use the bmap result to do IO directly to the drive.
4989 * the btrfs bmap call would return logical addresses that aren't
4990 * suitable for IO and they also will change frequently as COW
4991 * operations happen. So, swapfile + btrfs == corruption.
4993 * For now we're avoiding this by dropping bmap.
4995 static struct address_space_operations btrfs_aops
= {
4996 .readpage
= btrfs_readpage
,
4997 .writepage
= btrfs_writepage
,
4998 .writepages
= btrfs_writepages
,
4999 .readpages
= btrfs_readpages
,
5000 .sync_page
= block_sync_page
,
5001 .direct_IO
= btrfs_direct_IO
,
5002 .invalidatepage
= btrfs_invalidatepage
,
5003 .releasepage
= btrfs_releasepage
,
5004 .set_page_dirty
= btrfs_set_page_dirty
,
5007 static struct address_space_operations btrfs_symlink_aops
= {
5008 .readpage
= btrfs_readpage
,
5009 .writepage
= btrfs_writepage
,
5010 .invalidatepage
= btrfs_invalidatepage
,
5011 .releasepage
= btrfs_releasepage
,
5014 static struct inode_operations btrfs_file_inode_operations
= {
5015 .truncate
= btrfs_truncate
,
5016 .getattr
= btrfs_getattr
,
5017 .setattr
= btrfs_setattr
,
5018 .setxattr
= btrfs_setxattr
,
5019 .getxattr
= btrfs_getxattr
,
5020 .listxattr
= btrfs_listxattr
,
5021 .removexattr
= btrfs_removexattr
,
5022 .permission
= btrfs_permission
,
5023 .fallocate
= btrfs_fallocate
,
5025 static struct inode_operations btrfs_special_inode_operations
= {
5026 .getattr
= btrfs_getattr
,
5027 .setattr
= btrfs_setattr
,
5028 .permission
= btrfs_permission
,
5029 .setxattr
= btrfs_setxattr
,
5030 .getxattr
= btrfs_getxattr
,
5031 .listxattr
= btrfs_listxattr
,
5032 .removexattr
= btrfs_removexattr
,
5034 static struct inode_operations btrfs_symlink_inode_operations
= {
5035 .readlink
= generic_readlink
,
5036 .follow_link
= page_follow_link_light
,
5037 .put_link
= page_put_link
,
5038 .permission
= btrfs_permission
,