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"
55 struct btrfs_iget_args
{
57 struct btrfs_root
*root
;
60 static struct inode_operations btrfs_dir_inode_operations
;
61 static struct inode_operations btrfs_symlink_inode_operations
;
62 static struct inode_operations btrfs_dir_ro_inode_operations
;
63 static struct inode_operations btrfs_special_inode_operations
;
64 static struct inode_operations btrfs_file_inode_operations
;
65 static struct address_space_operations btrfs_aops
;
66 static struct address_space_operations btrfs_symlink_aops
;
67 static struct file_operations btrfs_dir_file_operations
;
68 static struct extent_io_ops btrfs_extent_io_ops
;
70 static struct kmem_cache
*btrfs_inode_cachep
;
71 struct kmem_cache
*btrfs_trans_handle_cachep
;
72 struct kmem_cache
*btrfs_transaction_cachep
;
73 struct kmem_cache
*btrfs_bit_radix_cachep
;
74 struct kmem_cache
*btrfs_path_cachep
;
77 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
78 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
79 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
80 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
81 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
82 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
83 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
84 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
87 static void btrfs_truncate(struct inode
*inode
);
88 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
89 static noinline
int cow_file_range(struct inode
*inode
,
90 struct page
*locked_page
,
91 u64 start
, u64 end
, int *page_started
,
92 unsigned long *nr_written
, int unlock
);
94 static int btrfs_init_inode_security(struct inode
*inode
, struct inode
*dir
)
98 err
= btrfs_init_acl(inode
, dir
);
100 err
= btrfs_xattr_security_init(inode
, dir
);
105 * a very lame attempt at stopping writes when the FS is 85% full. There
106 * are countless ways this is incorrect, but it is better than nothing.
108 int btrfs_check_free_space(struct btrfs_root
*root
, u64 num_required
,
116 spin_lock(&root
->fs_info
->delalloc_lock
);
117 total
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
118 used
= btrfs_super_bytes_used(&root
->fs_info
->super_copy
);
126 if (used
+ root
->fs_info
->delalloc_bytes
+ num_required
> thresh
)
128 spin_unlock(&root
->fs_info
->delalloc_lock
);
133 * this does all the hard work for inserting an inline extent into
134 * the btree. The caller should have done a btrfs_drop_extents so that
135 * no overlapping inline items exist in the btree
137 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
138 struct btrfs_root
*root
, struct inode
*inode
,
139 u64 start
, size_t size
, size_t compressed_size
,
140 struct page
**compressed_pages
)
142 struct btrfs_key key
;
143 struct btrfs_path
*path
;
144 struct extent_buffer
*leaf
;
145 struct page
*page
= NULL
;
148 struct btrfs_file_extent_item
*ei
;
151 size_t cur_size
= size
;
153 unsigned long offset
;
154 int use_compress
= 0;
156 if (compressed_size
&& compressed_pages
) {
158 cur_size
= compressed_size
;
161 path
= btrfs_alloc_path();
165 btrfs_set_trans_block_group(trans
, inode
);
167 key
.objectid
= inode
->i_ino
;
169 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
170 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
172 inode_add_bytes(inode
, size
);
173 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
180 leaf
= path
->nodes
[0];
181 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
182 struct btrfs_file_extent_item
);
183 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
184 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
185 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
186 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
187 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
188 ptr
= btrfs_file_extent_inline_start(ei
);
193 while (compressed_size
> 0) {
194 cpage
= compressed_pages
[i
];
195 cur_size
= min_t(unsigned long, compressed_size
,
199 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
204 compressed_size
-= cur_size
;
206 btrfs_set_file_extent_compression(leaf
, ei
,
207 BTRFS_COMPRESS_ZLIB
);
209 page
= find_get_page(inode
->i_mapping
,
210 start
>> PAGE_CACHE_SHIFT
);
211 btrfs_set_file_extent_compression(leaf
, ei
, 0);
212 kaddr
= kmap_atomic(page
, KM_USER0
);
213 offset
= start
& (PAGE_CACHE_SIZE
- 1);
214 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
215 kunmap_atomic(kaddr
, KM_USER0
);
216 page_cache_release(page
);
218 btrfs_mark_buffer_dirty(leaf
);
219 btrfs_free_path(path
);
221 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
222 btrfs_update_inode(trans
, root
, inode
);
225 btrfs_free_path(path
);
231 * conditionally insert an inline extent into the file. This
232 * does the checks required to make sure the data is small enough
233 * to fit as an inline extent.
235 static int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
236 struct btrfs_root
*root
,
237 struct inode
*inode
, u64 start
, u64 end
,
238 size_t compressed_size
,
239 struct page
**compressed_pages
)
241 u64 isize
= i_size_read(inode
);
242 u64 actual_end
= min(end
+ 1, isize
);
243 u64 inline_len
= actual_end
- start
;
244 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
245 ~((u64
)root
->sectorsize
- 1);
247 u64 data_len
= inline_len
;
251 data_len
= compressed_size
;
254 actual_end
>= PAGE_CACHE_SIZE
||
255 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
257 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
259 data_len
> root
->fs_info
->max_inline
) {
263 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
264 aligned_end
, start
, &hint_byte
);
267 if (isize
> actual_end
)
268 inline_len
= min_t(u64
, isize
, actual_end
);
269 ret
= insert_inline_extent(trans
, root
, inode
, start
,
270 inline_len
, compressed_size
,
273 btrfs_drop_extent_cache(inode
, start
, aligned_end
, 0);
277 struct async_extent
{
282 unsigned long nr_pages
;
283 struct list_head list
;
288 struct btrfs_root
*root
;
289 struct page
*locked_page
;
292 struct list_head extents
;
293 struct btrfs_work work
;
296 static noinline
int add_async_extent(struct async_cow
*cow
,
297 u64 start
, u64 ram_size
,
300 unsigned long nr_pages
)
302 struct async_extent
*async_extent
;
304 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
305 async_extent
->start
= start
;
306 async_extent
->ram_size
= ram_size
;
307 async_extent
->compressed_size
= compressed_size
;
308 async_extent
->pages
= pages
;
309 async_extent
->nr_pages
= nr_pages
;
310 list_add_tail(&async_extent
->list
, &cow
->extents
);
315 * we create compressed extents in two phases. The first
316 * phase compresses a range of pages that have already been
317 * locked (both pages and state bits are locked).
319 * This is done inside an ordered work queue, and the compression
320 * is spread across many cpus. The actual IO submission is step
321 * two, and the ordered work queue takes care of making sure that
322 * happens in the same order things were put onto the queue by
323 * writepages and friends.
325 * If this code finds it can't get good compression, it puts an
326 * entry onto the work queue to write the uncompressed bytes. This
327 * makes sure that both compressed inodes and uncompressed inodes
328 * are written in the same order that pdflush sent them down.
330 static noinline
int compress_file_range(struct inode
*inode
,
331 struct page
*locked_page
,
333 struct async_cow
*async_cow
,
336 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
337 struct btrfs_trans_handle
*trans
;
341 u64 blocksize
= root
->sectorsize
;
343 u64 isize
= i_size_read(inode
);
345 struct page
**pages
= NULL
;
346 unsigned long nr_pages
;
347 unsigned long nr_pages_ret
= 0;
348 unsigned long total_compressed
= 0;
349 unsigned long total_in
= 0;
350 unsigned long max_compressed
= 128 * 1024;
351 unsigned long max_uncompressed
= 128 * 1024;
357 actual_end
= min_t(u64
, isize
, end
+ 1);
360 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
361 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
363 total_compressed
= actual_end
- start
;
365 /* we want to make sure that amount of ram required to uncompress
366 * an extent is reasonable, so we limit the total size in ram
367 * of a compressed extent to 128k. This is a crucial number
368 * because it also controls how easily we can spread reads across
369 * cpus for decompression.
371 * We also want to make sure the amount of IO required to do
372 * a random read is reasonably small, so we limit the size of
373 * a compressed extent to 128k.
375 total_compressed
= min(total_compressed
, max_uncompressed
);
376 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
377 num_bytes
= max(blocksize
, num_bytes
);
378 disk_num_bytes
= num_bytes
;
383 * we do compression for mount -o compress and when the
384 * inode has not been flagged as nocompress. This flag can
385 * change at any time if we discover bad compression ratios.
387 if (!btrfs_test_flag(inode
, NOCOMPRESS
) &&
388 btrfs_test_opt(root
, COMPRESS
)) {
390 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
392 ret
= btrfs_zlib_compress_pages(inode
->i_mapping
, start
,
393 total_compressed
, pages
,
394 nr_pages
, &nr_pages_ret
,
400 unsigned long offset
= total_compressed
&
401 (PAGE_CACHE_SIZE
- 1);
402 struct page
*page
= pages
[nr_pages_ret
- 1];
405 /* zero the tail end of the last page, we might be
406 * sending it down to disk
409 kaddr
= kmap_atomic(page
, KM_USER0
);
410 memset(kaddr
+ offset
, 0,
411 PAGE_CACHE_SIZE
- offset
);
412 kunmap_atomic(kaddr
, KM_USER0
);
418 trans
= btrfs_join_transaction(root
, 1);
420 btrfs_set_trans_block_group(trans
, inode
);
422 /* lets try to make an inline extent */
423 if (ret
|| total_in
< (actual_end
- start
)) {
424 /* we didn't compress the entire range, try
425 * to make an uncompressed inline extent.
427 ret
= cow_file_range_inline(trans
, root
, inode
,
428 start
, end
, 0, NULL
);
430 /* try making a compressed inline extent */
431 ret
= cow_file_range_inline(trans
, root
, inode
,
433 total_compressed
, pages
);
435 btrfs_end_transaction(trans
, root
);
438 * inline extent creation worked, we don't need
439 * to create any more async work items. Unlock
440 * and free up our temp pages.
442 extent_clear_unlock_delalloc(inode
,
443 &BTRFS_I(inode
)->io_tree
,
444 start
, end
, NULL
, 1, 0,
453 * we aren't doing an inline extent round the compressed size
454 * up to a block size boundary so the allocator does sane
457 total_compressed
= (total_compressed
+ blocksize
- 1) &
461 * one last check to make sure the compression is really a
462 * win, compare the page count read with the blocks on disk
464 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
465 ~(PAGE_CACHE_SIZE
- 1);
466 if (total_compressed
>= total_in
) {
469 disk_num_bytes
= total_compressed
;
470 num_bytes
= total_in
;
473 if (!will_compress
&& pages
) {
475 * the compression code ran but failed to make things smaller,
476 * free any pages it allocated and our page pointer array
478 for (i
= 0; i
< nr_pages_ret
; i
++) {
479 WARN_ON(pages
[i
]->mapping
);
480 page_cache_release(pages
[i
]);
484 total_compressed
= 0;
487 /* flag the file so we don't compress in the future */
488 btrfs_set_flag(inode
, NOCOMPRESS
);
493 /* the async work queues will take care of doing actual
494 * allocation on disk for these compressed pages,
495 * and will submit them to the elevator.
497 add_async_extent(async_cow
, start
, num_bytes
,
498 total_compressed
, pages
, nr_pages_ret
);
500 if (start
+ num_bytes
< end
&& start
+ num_bytes
< actual_end
) {
508 * No compression, but we still need to write the pages in
509 * the file we've been given so far. redirty the locked
510 * page if it corresponds to our extent and set things up
511 * for the async work queue to run cow_file_range to do
512 * the normal delalloc dance
514 if (page_offset(locked_page
) >= start
&&
515 page_offset(locked_page
) <= end
) {
516 __set_page_dirty_nobuffers(locked_page
);
517 /* unlocked later on in the async handlers */
519 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0);
527 for (i
= 0; i
< nr_pages_ret
; i
++) {
528 WARN_ON(pages
[i
]->mapping
);
529 page_cache_release(pages
[i
]);
537 * phase two of compressed writeback. This is the ordered portion
538 * of the code, which only gets called in the order the work was
539 * queued. We walk all the async extents created by compress_file_range
540 * and send them down to the disk.
542 static noinline
int submit_compressed_extents(struct inode
*inode
,
543 struct async_cow
*async_cow
)
545 struct async_extent
*async_extent
;
547 struct btrfs_trans_handle
*trans
;
548 struct btrfs_key ins
;
549 struct extent_map
*em
;
550 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
551 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
552 struct extent_io_tree
*io_tree
;
555 if (list_empty(&async_cow
->extents
))
558 trans
= btrfs_join_transaction(root
, 1);
560 while (!list_empty(&async_cow
->extents
)) {
561 async_extent
= list_entry(async_cow
->extents
.next
,
562 struct async_extent
, list
);
563 list_del(&async_extent
->list
);
565 io_tree
= &BTRFS_I(inode
)->io_tree
;
567 /* did the compression code fall back to uncompressed IO? */
568 if (!async_extent
->pages
) {
569 int page_started
= 0;
570 unsigned long nr_written
= 0;
572 lock_extent(io_tree
, async_extent
->start
,
573 async_extent
->start
+
574 async_extent
->ram_size
- 1, GFP_NOFS
);
576 /* allocate blocks */
577 cow_file_range(inode
, async_cow
->locked_page
,
579 async_extent
->start
+
580 async_extent
->ram_size
- 1,
581 &page_started
, &nr_written
, 0);
584 * if page_started, cow_file_range inserted an
585 * inline extent and took care of all the unlocking
586 * and IO for us. Otherwise, we need to submit
587 * all those pages down to the drive.
590 extent_write_locked_range(io_tree
,
591 inode
, async_extent
->start
,
592 async_extent
->start
+
593 async_extent
->ram_size
- 1,
601 lock_extent(io_tree
, async_extent
->start
,
602 async_extent
->start
+ async_extent
->ram_size
- 1,
605 * here we're doing allocation and writeback of the
608 btrfs_drop_extent_cache(inode
, async_extent
->start
,
609 async_extent
->start
+
610 async_extent
->ram_size
- 1, 0);
612 ret
= btrfs_reserve_extent(trans
, root
,
613 async_extent
->compressed_size
,
614 async_extent
->compressed_size
,
618 em
= alloc_extent_map(GFP_NOFS
);
619 em
->start
= async_extent
->start
;
620 em
->len
= async_extent
->ram_size
;
621 em
->orig_start
= em
->start
;
623 em
->block_start
= ins
.objectid
;
624 em
->block_len
= ins
.offset
;
625 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
626 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
627 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
630 spin_lock(&em_tree
->lock
);
631 ret
= add_extent_mapping(em_tree
, em
);
632 spin_unlock(&em_tree
->lock
);
633 if (ret
!= -EEXIST
) {
637 btrfs_drop_extent_cache(inode
, async_extent
->start
,
638 async_extent
->start
+
639 async_extent
->ram_size
- 1, 0);
642 ret
= btrfs_add_ordered_extent(inode
, async_extent
->start
,
644 async_extent
->ram_size
,
646 BTRFS_ORDERED_COMPRESSED
);
649 btrfs_end_transaction(trans
, root
);
652 * clear dirty, set writeback and unlock the pages.
654 extent_clear_unlock_delalloc(inode
,
655 &BTRFS_I(inode
)->io_tree
,
657 async_extent
->start
+
658 async_extent
->ram_size
- 1,
659 NULL
, 1, 1, 0, 1, 1, 0);
661 ret
= btrfs_submit_compressed_write(inode
,
663 async_extent
->ram_size
,
665 ins
.offset
, async_extent
->pages
,
666 async_extent
->nr_pages
);
669 trans
= btrfs_join_transaction(root
, 1);
670 alloc_hint
= ins
.objectid
+ ins
.offset
;
675 btrfs_end_transaction(trans
, root
);
680 * when extent_io.c finds a delayed allocation range in the file,
681 * the call backs end up in this code. The basic idea is to
682 * allocate extents on disk for the range, and create ordered data structs
683 * in ram to track those extents.
685 * locked_page is the page that writepage had locked already. We use
686 * it to make sure we don't do extra locks or unlocks.
688 * *page_started is set to one if we unlock locked_page and do everything
689 * required to start IO on it. It may be clean and already done with
692 static noinline
int cow_file_range(struct inode
*inode
,
693 struct page
*locked_page
,
694 u64 start
, u64 end
, int *page_started
,
695 unsigned long *nr_written
,
698 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
699 struct btrfs_trans_handle
*trans
;
702 unsigned long ram_size
;
705 u64 blocksize
= root
->sectorsize
;
707 u64 isize
= i_size_read(inode
);
708 struct btrfs_key ins
;
709 struct extent_map
*em
;
710 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
713 trans
= btrfs_join_transaction(root
, 1);
715 btrfs_set_trans_block_group(trans
, inode
);
717 actual_end
= min_t(u64
, isize
, end
+ 1);
719 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
720 num_bytes
= max(blocksize
, num_bytes
);
721 disk_num_bytes
= num_bytes
;
725 /* lets try to make an inline extent */
726 ret
= cow_file_range_inline(trans
, root
, inode
,
727 start
, end
, 0, NULL
);
729 extent_clear_unlock_delalloc(inode
,
730 &BTRFS_I(inode
)->io_tree
,
731 start
, end
, NULL
, 1, 1,
733 *nr_written
= *nr_written
+
734 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
741 BUG_ON(disk_num_bytes
>
742 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
744 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
746 while (disk_num_bytes
> 0) {
747 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
748 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
749 root
->sectorsize
, 0, alloc_hint
,
753 em
= alloc_extent_map(GFP_NOFS
);
755 em
->orig_start
= em
->start
;
757 ram_size
= ins
.offset
;
758 em
->len
= ins
.offset
;
760 em
->block_start
= ins
.objectid
;
761 em
->block_len
= ins
.offset
;
762 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
763 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
766 spin_lock(&em_tree
->lock
);
767 ret
= add_extent_mapping(em_tree
, em
);
768 spin_unlock(&em_tree
->lock
);
769 if (ret
!= -EEXIST
) {
773 btrfs_drop_extent_cache(inode
, start
,
774 start
+ ram_size
- 1, 0);
777 cur_alloc_size
= ins
.offset
;
778 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
779 ram_size
, cur_alloc_size
, 0);
782 if (root
->root_key
.objectid
==
783 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
784 ret
= btrfs_reloc_clone_csums(inode
, start
,
789 if (disk_num_bytes
< cur_alloc_size
)
792 /* we're not doing compressed IO, don't unlock the first
793 * page (which the caller expects to stay locked), don't
794 * clear any dirty bits and don't set any writeback bits
796 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
797 start
, start
+ ram_size
- 1,
798 locked_page
, unlock
, 1,
800 disk_num_bytes
-= cur_alloc_size
;
801 num_bytes
-= cur_alloc_size
;
802 alloc_hint
= ins
.objectid
+ ins
.offset
;
803 start
+= cur_alloc_size
;
807 btrfs_end_transaction(trans
, root
);
813 * work queue call back to started compression on a file and pages
815 static noinline
void async_cow_start(struct btrfs_work
*work
)
817 struct async_cow
*async_cow
;
819 async_cow
= container_of(work
, struct async_cow
, work
);
821 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
822 async_cow
->start
, async_cow
->end
, async_cow
,
825 async_cow
->inode
= NULL
;
829 * work queue call back to submit previously compressed pages
831 static noinline
void async_cow_submit(struct btrfs_work
*work
)
833 struct async_cow
*async_cow
;
834 struct btrfs_root
*root
;
835 unsigned long nr_pages
;
837 async_cow
= container_of(work
, struct async_cow
, work
);
839 root
= async_cow
->root
;
840 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
843 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
845 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
847 waitqueue_active(&root
->fs_info
->async_submit_wait
))
848 wake_up(&root
->fs_info
->async_submit_wait
);
850 if (async_cow
->inode
)
851 submit_compressed_extents(async_cow
->inode
, async_cow
);
854 static noinline
void async_cow_free(struct btrfs_work
*work
)
856 struct async_cow
*async_cow
;
857 async_cow
= container_of(work
, struct async_cow
, work
);
861 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
862 u64 start
, u64 end
, int *page_started
,
863 unsigned long *nr_written
)
865 struct async_cow
*async_cow
;
866 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
867 unsigned long nr_pages
;
869 int limit
= 10 * 1024 * 1042;
871 if (!btrfs_test_opt(root
, COMPRESS
)) {
872 return cow_file_range(inode
, locked_page
, start
, end
,
873 page_started
, nr_written
, 1);
876 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
|
877 EXTENT_DELALLOC
, 1, 0, GFP_NOFS
);
878 while (start
< end
) {
879 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
880 async_cow
->inode
= inode
;
881 async_cow
->root
= root
;
882 async_cow
->locked_page
= locked_page
;
883 async_cow
->start
= start
;
885 if (btrfs_test_flag(inode
, NOCOMPRESS
))
888 cur_end
= min(end
, start
+ 512 * 1024 - 1);
890 async_cow
->end
= cur_end
;
891 INIT_LIST_HEAD(&async_cow
->extents
);
893 async_cow
->work
.func
= async_cow_start
;
894 async_cow
->work
.ordered_func
= async_cow_submit
;
895 async_cow
->work
.ordered_free
= async_cow_free
;
896 async_cow
->work
.flags
= 0;
898 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
900 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
902 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
905 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
906 wait_event(root
->fs_info
->async_submit_wait
,
907 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
911 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
912 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
913 wait_event(root
->fs_info
->async_submit_wait
,
914 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
918 *nr_written
+= nr_pages
;
925 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
926 u64 bytenr
, u64 num_bytes
)
929 struct btrfs_ordered_sum
*sums
;
932 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
933 bytenr
+ num_bytes
- 1, &list
);
934 if (ret
== 0 && list_empty(&list
))
937 while (!list_empty(&list
)) {
938 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
939 list_del(&sums
->list
);
946 * when nowcow writeback call back. This checks for snapshots or COW copies
947 * of the extents that exist in the file, and COWs the file as required.
949 * If no cow copies or snapshots exist, we write directly to the existing
952 static int run_delalloc_nocow(struct inode
*inode
, struct page
*locked_page
,
953 u64 start
, u64 end
, int *page_started
, int force
,
954 unsigned long *nr_written
)
956 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
957 struct btrfs_trans_handle
*trans
;
958 struct extent_buffer
*leaf
;
959 struct btrfs_path
*path
;
960 struct btrfs_file_extent_item
*fi
;
961 struct btrfs_key found_key
;
973 path
= btrfs_alloc_path();
975 trans
= btrfs_join_transaction(root
, 1);
981 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
984 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
985 leaf
= path
->nodes
[0];
986 btrfs_item_key_to_cpu(leaf
, &found_key
,
988 if (found_key
.objectid
== inode
->i_ino
&&
989 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
994 leaf
= path
->nodes
[0];
995 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
996 ret
= btrfs_next_leaf(root
, path
);
1001 leaf
= path
->nodes
[0];
1007 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1009 if (found_key
.objectid
> inode
->i_ino
||
1010 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1011 found_key
.offset
> end
)
1014 if (found_key
.offset
> cur_offset
) {
1015 extent_end
= found_key
.offset
;
1019 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1020 struct btrfs_file_extent_item
);
1021 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1023 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1024 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1025 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1026 extent_end
= found_key
.offset
+
1027 btrfs_file_extent_num_bytes(leaf
, fi
);
1028 if (extent_end
<= start
) {
1032 if (disk_bytenr
== 0)
1034 if (btrfs_file_extent_compression(leaf
, fi
) ||
1035 btrfs_file_extent_encryption(leaf
, fi
) ||
1036 btrfs_file_extent_other_encoding(leaf
, fi
))
1038 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1040 if (btrfs_extent_readonly(root
, disk_bytenr
))
1042 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1045 disk_bytenr
+= btrfs_file_extent_offset(leaf
, fi
);
1046 disk_bytenr
+= cur_offset
- found_key
.offset
;
1047 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1049 * force cow if csum exists in the range.
1050 * this ensure that csum for a given extent are
1051 * either valid or do not exist.
1053 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1056 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1057 extent_end
= found_key
.offset
+
1058 btrfs_file_extent_inline_len(leaf
, fi
);
1059 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1064 if (extent_end
<= start
) {
1069 if (cow_start
== (u64
)-1)
1070 cow_start
= cur_offset
;
1071 cur_offset
= extent_end
;
1072 if (cur_offset
> end
)
1078 btrfs_release_path(root
, path
);
1079 if (cow_start
!= (u64
)-1) {
1080 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1081 found_key
.offset
- 1, page_started
,
1084 cow_start
= (u64
)-1;
1087 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1088 struct extent_map
*em
;
1089 struct extent_map_tree
*em_tree
;
1090 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1091 em
= alloc_extent_map(GFP_NOFS
);
1092 em
->start
= cur_offset
;
1093 em
->orig_start
= em
->start
;
1094 em
->len
= num_bytes
;
1095 em
->block_len
= num_bytes
;
1096 em
->block_start
= disk_bytenr
;
1097 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1098 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1100 spin_lock(&em_tree
->lock
);
1101 ret
= add_extent_mapping(em_tree
, em
);
1102 spin_unlock(&em_tree
->lock
);
1103 if (ret
!= -EEXIST
) {
1104 free_extent_map(em
);
1107 btrfs_drop_extent_cache(inode
, em
->start
,
1108 em
->start
+ em
->len
- 1, 0);
1110 type
= BTRFS_ORDERED_PREALLOC
;
1112 type
= BTRFS_ORDERED_NOCOW
;
1115 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1116 num_bytes
, num_bytes
, type
);
1119 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1120 cur_offset
, cur_offset
+ num_bytes
- 1,
1121 locked_page
, 1, 1, 1, 0, 0, 0);
1122 cur_offset
= extent_end
;
1123 if (cur_offset
> end
)
1126 btrfs_release_path(root
, path
);
1128 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1129 cow_start
= cur_offset
;
1130 if (cow_start
!= (u64
)-1) {
1131 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1132 page_started
, nr_written
, 1);
1136 ret
= btrfs_end_transaction(trans
, root
);
1138 btrfs_free_path(path
);
1143 * extent_io.c call back to do delayed allocation processing
1145 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1146 u64 start
, u64 end
, int *page_started
,
1147 unsigned long *nr_written
)
1151 if (btrfs_test_flag(inode
, NODATACOW
))
1152 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1153 page_started
, 1, nr_written
);
1154 else if (btrfs_test_flag(inode
, PREALLOC
))
1155 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1156 page_started
, 0, nr_written
);
1158 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1159 page_started
, nr_written
);
1165 * extent_io.c set_bit_hook, used to track delayed allocation
1166 * bytes in this file, and to maintain the list of inodes that
1167 * have pending delalloc work to be done.
1169 static int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1170 unsigned long old
, unsigned long bits
)
1173 * set_bit and clear bit hooks normally require _irqsave/restore
1174 * but in this case, we are only testeing for the DELALLOC
1175 * bit, which is only set or cleared with irqs on
1177 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1178 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1179 spin_lock(&root
->fs_info
->delalloc_lock
);
1180 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
1181 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
1182 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1183 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1184 &root
->fs_info
->delalloc_inodes
);
1186 spin_unlock(&root
->fs_info
->delalloc_lock
);
1192 * extent_io.c clear_bit_hook, see set_bit_hook for why
1194 static int btrfs_clear_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1195 unsigned long old
, unsigned long bits
)
1198 * set_bit and clear bit hooks normally require _irqsave/restore
1199 * but in this case, we are only testeing for the DELALLOC
1200 * bit, which is only set or cleared with irqs on
1202 if ((old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1203 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1205 spin_lock(&root
->fs_info
->delalloc_lock
);
1206 if (end
- start
+ 1 > root
->fs_info
->delalloc_bytes
) {
1207 printk(KERN_INFO
"btrfs warning: delalloc account "
1209 (unsigned long long)end
- start
+ 1,
1210 (unsigned long long)
1211 root
->fs_info
->delalloc_bytes
);
1212 root
->fs_info
->delalloc_bytes
= 0;
1213 BTRFS_I(inode
)->delalloc_bytes
= 0;
1215 root
->fs_info
->delalloc_bytes
-= end
- start
+ 1;
1216 BTRFS_I(inode
)->delalloc_bytes
-= end
- start
+ 1;
1218 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1219 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1220 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1222 spin_unlock(&root
->fs_info
->delalloc_lock
);
1228 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1229 * we don't create bios that span stripes or chunks
1231 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1232 size_t size
, struct bio
*bio
,
1233 unsigned long bio_flags
)
1235 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1236 struct btrfs_mapping_tree
*map_tree
;
1237 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1242 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1245 length
= bio
->bi_size
;
1246 map_tree
= &root
->fs_info
->mapping_tree
;
1247 map_length
= length
;
1248 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1249 &map_length
, NULL
, 0);
1251 if (map_length
< length
+ size
)
1257 * in order to insert checksums into the metadata in large chunks,
1258 * we wait until bio submission time. All the pages in the bio are
1259 * checksummed and sums are attached onto the ordered extent record.
1261 * At IO completion time the cums attached on the ordered extent record
1262 * are inserted into the btree
1264 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1265 struct bio
*bio
, int mirror_num
,
1266 unsigned long bio_flags
)
1268 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1271 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1277 * in order to insert checksums into the metadata in large chunks,
1278 * we wait until bio submission time. All the pages in the bio are
1279 * checksummed and sums are attached onto the ordered extent record.
1281 * At IO completion time the cums attached on the ordered extent record
1282 * are inserted into the btree
1284 static int __btrfs_submit_bio_done(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
;
1288 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1292 * extent_io.c submission hook. This does the right thing for csum calculation
1293 * on write, or reading the csums from the tree before a read
1295 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1296 int mirror_num
, unsigned long bio_flags
)
1298 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1302 skip_sum
= btrfs_test_flag(inode
, NODATASUM
);
1304 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1307 if (!(rw
& (1 << BIO_RW
))) {
1308 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1309 return btrfs_submit_compressed_read(inode
, bio
,
1310 mirror_num
, bio_flags
);
1311 } else if (!skip_sum
)
1312 btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1314 } else if (!skip_sum
) {
1315 /* csum items have already been cloned */
1316 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1318 /* we're doing a write, do the async checksumming */
1319 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1320 inode
, rw
, bio
, mirror_num
,
1321 bio_flags
, __btrfs_submit_bio_start
,
1322 __btrfs_submit_bio_done
);
1326 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1330 * given a list of ordered sums record them in the inode. This happens
1331 * at IO completion time based on sums calculated at bio submission time.
1333 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1334 struct inode
*inode
, u64 file_offset
,
1335 struct list_head
*list
)
1337 struct btrfs_ordered_sum
*sum
;
1339 btrfs_set_trans_block_group(trans
, inode
);
1341 list_for_each_entry(sum
, list
, list
) {
1342 btrfs_csum_file_blocks(trans
,
1343 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1348 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
1350 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1352 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1356 /* see btrfs_writepage_start_hook for details on why this is required */
1357 struct btrfs_writepage_fixup
{
1359 struct btrfs_work work
;
1362 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1364 struct btrfs_writepage_fixup
*fixup
;
1365 struct btrfs_ordered_extent
*ordered
;
1367 struct inode
*inode
;
1371 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1375 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1376 ClearPageChecked(page
);
1380 inode
= page
->mapping
->host
;
1381 page_start
= page_offset(page
);
1382 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1384 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1386 /* already ordered? We're done */
1387 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1388 EXTENT_ORDERED
, 0)) {
1392 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1394 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
1395 page_end
, GFP_NOFS
);
1397 btrfs_start_ordered_extent(inode
, ordered
, 1);
1401 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
1402 ClearPageChecked(page
);
1404 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1407 page_cache_release(page
);
1411 * There are a few paths in the higher layers of the kernel that directly
1412 * set the page dirty bit without asking the filesystem if it is a
1413 * good idea. This causes problems because we want to make sure COW
1414 * properly happens and the data=ordered rules are followed.
1416 * In our case any range that doesn't have the ORDERED bit set
1417 * hasn't been properly setup for IO. We kick off an async process
1418 * to fix it up. The async helper will wait for ordered extents, set
1419 * the delalloc bit and make it safe to write the page.
1421 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1423 struct inode
*inode
= page
->mapping
->host
;
1424 struct btrfs_writepage_fixup
*fixup
;
1425 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1428 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1433 if (PageChecked(page
))
1436 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1440 SetPageChecked(page
);
1441 page_cache_get(page
);
1442 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1444 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1448 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1449 struct inode
*inode
, u64 file_pos
,
1450 u64 disk_bytenr
, u64 disk_num_bytes
,
1451 u64 num_bytes
, u64 ram_bytes
,
1452 u8 compression
, u8 encryption
,
1453 u16 other_encoding
, int extent_type
)
1455 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1456 struct btrfs_file_extent_item
*fi
;
1457 struct btrfs_path
*path
;
1458 struct extent_buffer
*leaf
;
1459 struct btrfs_key ins
;
1463 path
= btrfs_alloc_path();
1466 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1467 file_pos
+ num_bytes
, file_pos
, &hint
);
1470 ins
.objectid
= inode
->i_ino
;
1471 ins
.offset
= file_pos
;
1472 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1473 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1475 leaf
= path
->nodes
[0];
1476 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1477 struct btrfs_file_extent_item
);
1478 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1479 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1480 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1481 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1482 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1483 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1484 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1485 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1486 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1487 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1488 btrfs_mark_buffer_dirty(leaf
);
1490 inode_add_bytes(inode
, num_bytes
);
1491 btrfs_drop_extent_cache(inode
, file_pos
, file_pos
+ num_bytes
- 1, 0);
1493 ins
.objectid
= disk_bytenr
;
1494 ins
.offset
= disk_num_bytes
;
1495 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1496 ret
= btrfs_alloc_reserved_extent(trans
, root
, leaf
->start
,
1497 root
->root_key
.objectid
,
1498 trans
->transid
, inode
->i_ino
, &ins
);
1501 btrfs_free_path(path
);
1505 /* as ordered data IO finishes, this gets called so we can finish
1506 * an ordered extent if the range of bytes in the file it covers are
1509 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1511 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1512 struct btrfs_trans_handle
*trans
;
1513 struct btrfs_ordered_extent
*ordered_extent
;
1514 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1518 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1522 trans
= btrfs_join_transaction(root
, 1);
1524 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1525 BUG_ON(!ordered_extent
);
1526 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1529 lock_extent(io_tree
, ordered_extent
->file_offset
,
1530 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1533 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1535 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1537 ret
= btrfs_mark_extent_written(trans
, root
, inode
,
1538 ordered_extent
->file_offset
,
1539 ordered_extent
->file_offset
+
1540 ordered_extent
->len
);
1543 ret
= insert_reserved_file_extent(trans
, inode
,
1544 ordered_extent
->file_offset
,
1545 ordered_extent
->start
,
1546 ordered_extent
->disk_len
,
1547 ordered_extent
->len
,
1548 ordered_extent
->len
,
1550 BTRFS_FILE_EXTENT_REG
);
1553 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1554 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1557 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1558 &ordered_extent
->list
);
1560 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1561 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1562 btrfs_update_inode(trans
, root
, inode
);
1563 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1564 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1567 btrfs_put_ordered_extent(ordered_extent
);
1568 /* once for the tree */
1569 btrfs_put_ordered_extent(ordered_extent
);
1571 btrfs_end_transaction(trans
, root
);
1575 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1576 struct extent_state
*state
, int uptodate
)
1578 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1582 * When IO fails, either with EIO or csum verification fails, we
1583 * try other mirrors that might have a good copy of the data. This
1584 * io_failure_record is used to record state as we go through all the
1585 * mirrors. If another mirror has good data, the page is set up to date
1586 * and things continue. If a good mirror can't be found, the original
1587 * bio end_io callback is called to indicate things have failed.
1589 struct io_failure_record
{
1594 unsigned long bio_flags
;
1598 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1599 struct page
*page
, u64 start
, u64 end
,
1600 struct extent_state
*state
)
1602 struct io_failure_record
*failrec
= NULL
;
1604 struct extent_map
*em
;
1605 struct inode
*inode
= page
->mapping
->host
;
1606 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1607 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1614 ret
= get_state_private(failure_tree
, start
, &private);
1616 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1619 failrec
->start
= start
;
1620 failrec
->len
= end
- start
+ 1;
1621 failrec
->last_mirror
= 0;
1622 failrec
->bio_flags
= 0;
1624 spin_lock(&em_tree
->lock
);
1625 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1626 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1627 free_extent_map(em
);
1630 spin_unlock(&em_tree
->lock
);
1632 if (!em
|| IS_ERR(em
)) {
1636 logical
= start
- em
->start
;
1637 logical
= em
->block_start
+ logical
;
1638 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1639 logical
= em
->block_start
;
1640 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1642 failrec
->logical
= logical
;
1643 free_extent_map(em
);
1644 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1645 EXTENT_DIRTY
, GFP_NOFS
);
1646 set_state_private(failure_tree
, start
,
1647 (u64
)(unsigned long)failrec
);
1649 failrec
= (struct io_failure_record
*)(unsigned long)private;
1651 num_copies
= btrfs_num_copies(
1652 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1653 failrec
->logical
, failrec
->len
);
1654 failrec
->last_mirror
++;
1656 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1657 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1660 if (state
&& state
->start
!= failrec
->start
)
1662 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1664 if (!state
|| failrec
->last_mirror
> num_copies
) {
1665 set_state_private(failure_tree
, failrec
->start
, 0);
1666 clear_extent_bits(failure_tree
, failrec
->start
,
1667 failrec
->start
+ failrec
->len
- 1,
1668 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1672 bio
= bio_alloc(GFP_NOFS
, 1);
1673 bio
->bi_private
= state
;
1674 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1675 bio
->bi_sector
= failrec
->logical
>> 9;
1676 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1679 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1680 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1685 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1686 failrec
->last_mirror
,
1687 failrec
->bio_flags
);
1692 * each time an IO finishes, we do a fast check in the IO failure tree
1693 * to see if we need to process or clean up an io_failure_record
1695 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1698 u64 private_failure
;
1699 struct io_failure_record
*failure
;
1703 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1704 (u64
)-1, 1, EXTENT_DIRTY
)) {
1705 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1706 start
, &private_failure
);
1708 failure
= (struct io_failure_record
*)(unsigned long)
1710 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1712 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1714 failure
->start
+ failure
->len
- 1,
1715 EXTENT_DIRTY
| EXTENT_LOCKED
,
1724 * when reads are done, we need to check csums to verify the data is correct
1725 * if there's a match, we allow the bio to finish. If not, we go through
1726 * the io_failure_record routines to find good copies
1728 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1729 struct extent_state
*state
)
1731 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1732 struct inode
*inode
= page
->mapping
->host
;
1733 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1735 u64
private = ~(u32
)0;
1737 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1740 if (PageChecked(page
)) {
1741 ClearPageChecked(page
);
1744 if (btrfs_test_flag(inode
, NODATASUM
))
1747 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1748 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1)) {
1749 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1754 if (state
&& state
->start
== start
) {
1755 private = state
->private;
1758 ret
= get_state_private(io_tree
, start
, &private);
1760 kaddr
= kmap_atomic(page
, KM_USER0
);
1764 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1765 btrfs_csum_final(csum
, (char *)&csum
);
1766 if (csum
!= private)
1769 kunmap_atomic(kaddr
, KM_USER0
);
1771 /* if the io failure tree for this inode is non-empty,
1772 * check to see if we've recovered from a failed IO
1774 btrfs_clean_io_failures(inode
, start
);
1778 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
1779 "private %llu\n", page
->mapping
->host
->i_ino
,
1780 (unsigned long long)start
, csum
,
1781 (unsigned long long)private);
1782 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1783 flush_dcache_page(page
);
1784 kunmap_atomic(kaddr
, KM_USER0
);
1791 * This creates an orphan entry for the given inode in case something goes
1792 * wrong in the middle of an unlink/truncate.
1794 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1796 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1799 spin_lock(&root
->list_lock
);
1801 /* already on the orphan list, we're good */
1802 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1803 spin_unlock(&root
->list_lock
);
1807 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1809 spin_unlock(&root
->list_lock
);
1812 * insert an orphan item to track this unlinked/truncated file
1814 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1820 * We have done the truncate/delete so we can go ahead and remove the orphan
1821 * item for this particular inode.
1823 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1825 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1828 spin_lock(&root
->list_lock
);
1830 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1831 spin_unlock(&root
->list_lock
);
1835 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1837 spin_unlock(&root
->list_lock
);
1841 spin_unlock(&root
->list_lock
);
1843 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1849 * this cleans up any orphans that may be left on the list from the last use
1852 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1854 struct btrfs_path
*path
;
1855 struct extent_buffer
*leaf
;
1856 struct btrfs_item
*item
;
1857 struct btrfs_key key
, found_key
;
1858 struct btrfs_trans_handle
*trans
;
1859 struct inode
*inode
;
1860 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1862 path
= btrfs_alloc_path();
1867 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1868 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1869 key
.offset
= (u64
)-1;
1873 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1875 printk(KERN_ERR
"Error searching slot for orphan: %d"
1881 * if ret == 0 means we found what we were searching for, which
1882 * is weird, but possible, so only screw with path if we didnt
1883 * find the key and see if we have stuff that matches
1886 if (path
->slots
[0] == 0)
1891 /* pull out the item */
1892 leaf
= path
->nodes
[0];
1893 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1894 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1896 /* make sure the item matches what we want */
1897 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1899 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1902 /* release the path since we're done with it */
1903 btrfs_release_path(root
, path
);
1906 * this is where we are basically btrfs_lookup, without the
1907 * crossing root thing. we store the inode number in the
1908 * offset of the orphan item.
1910 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
1911 found_key
.offset
, root
);
1915 if (inode
->i_state
& I_NEW
) {
1916 BTRFS_I(inode
)->root
= root
;
1918 /* have to set the location manually */
1919 BTRFS_I(inode
)->location
.objectid
= inode
->i_ino
;
1920 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
1921 BTRFS_I(inode
)->location
.offset
= 0;
1923 btrfs_read_locked_inode(inode
);
1924 unlock_new_inode(inode
);
1928 * add this inode to the orphan list so btrfs_orphan_del does
1929 * the proper thing when we hit it
1931 spin_lock(&root
->list_lock
);
1932 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1933 spin_unlock(&root
->list_lock
);
1936 * if this is a bad inode, means we actually succeeded in
1937 * removing the inode, but not the orphan record, which means
1938 * we need to manually delete the orphan since iput will just
1939 * do a destroy_inode
1941 if (is_bad_inode(inode
)) {
1942 trans
= btrfs_start_transaction(root
, 1);
1943 btrfs_orphan_del(trans
, inode
);
1944 btrfs_end_transaction(trans
, root
);
1949 /* if we have links, this was a truncate, lets do that */
1950 if (inode
->i_nlink
) {
1952 btrfs_truncate(inode
);
1957 /* this will do delete_inode and everything for us */
1962 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
1964 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
1966 btrfs_free_path(path
);
1970 * read an inode from the btree into the in-memory inode
1972 void btrfs_read_locked_inode(struct inode
*inode
)
1974 struct btrfs_path
*path
;
1975 struct extent_buffer
*leaf
;
1976 struct btrfs_inode_item
*inode_item
;
1977 struct btrfs_timespec
*tspec
;
1978 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1979 struct btrfs_key location
;
1980 u64 alloc_group_block
;
1984 path
= btrfs_alloc_path();
1986 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
1988 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
1992 leaf
= path
->nodes
[0];
1993 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1994 struct btrfs_inode_item
);
1996 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
1997 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
1998 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
1999 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2000 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2002 tspec
= btrfs_inode_atime(inode_item
);
2003 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2004 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2006 tspec
= btrfs_inode_mtime(inode_item
);
2007 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2008 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2010 tspec
= btrfs_inode_ctime(inode_item
);
2011 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2012 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2014 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2015 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2016 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2017 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2019 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2021 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2022 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2024 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2026 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2027 alloc_group_block
, 0);
2028 btrfs_free_path(path
);
2031 switch (inode
->i_mode
& S_IFMT
) {
2033 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2034 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2035 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2036 inode
->i_fop
= &btrfs_file_operations
;
2037 inode
->i_op
= &btrfs_file_inode_operations
;
2040 inode
->i_fop
= &btrfs_dir_file_operations
;
2041 if (root
== root
->fs_info
->tree_root
)
2042 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2044 inode
->i_op
= &btrfs_dir_inode_operations
;
2047 inode
->i_op
= &btrfs_symlink_inode_operations
;
2048 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2049 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2052 inode
->i_op
= &btrfs_special_inode_operations
;
2053 init_special_inode(inode
, inode
->i_mode
, rdev
);
2059 btrfs_free_path(path
);
2060 make_bad_inode(inode
);
2064 * given a leaf and an inode, copy the inode fields into the leaf
2066 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2067 struct extent_buffer
*leaf
,
2068 struct btrfs_inode_item
*item
,
2069 struct inode
*inode
)
2071 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2072 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2073 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2074 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2075 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2077 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2078 inode
->i_atime
.tv_sec
);
2079 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2080 inode
->i_atime
.tv_nsec
);
2082 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2083 inode
->i_mtime
.tv_sec
);
2084 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2085 inode
->i_mtime
.tv_nsec
);
2087 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2088 inode
->i_ctime
.tv_sec
);
2089 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2090 inode
->i_ctime
.tv_nsec
);
2092 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2093 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2094 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2095 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2096 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2097 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2098 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2102 * copy everything in the in-memory inode into the btree.
2104 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2105 struct btrfs_root
*root
, struct inode
*inode
)
2107 struct btrfs_inode_item
*inode_item
;
2108 struct btrfs_path
*path
;
2109 struct extent_buffer
*leaf
;
2112 path
= btrfs_alloc_path();
2114 ret
= btrfs_lookup_inode(trans
, root
, path
,
2115 &BTRFS_I(inode
)->location
, 1);
2122 btrfs_unlock_up_safe(path
, 1);
2123 leaf
= path
->nodes
[0];
2124 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2125 struct btrfs_inode_item
);
2127 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2128 btrfs_mark_buffer_dirty(leaf
);
2129 btrfs_set_inode_last_trans(trans
, inode
);
2132 btrfs_free_path(path
);
2138 * unlink helper that gets used here in inode.c and in the tree logging
2139 * recovery code. It remove a link in a directory with a given name, and
2140 * also drops the back refs in the inode to the directory
2142 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2143 struct btrfs_root
*root
,
2144 struct inode
*dir
, struct inode
*inode
,
2145 const char *name
, int name_len
)
2147 struct btrfs_path
*path
;
2149 struct extent_buffer
*leaf
;
2150 struct btrfs_dir_item
*di
;
2151 struct btrfs_key key
;
2154 path
= btrfs_alloc_path();
2160 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2161 name
, name_len
, -1);
2170 leaf
= path
->nodes
[0];
2171 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2172 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2175 btrfs_release_path(root
, path
);
2177 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2179 dir
->i_ino
, &index
);
2181 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2182 "inode %lu parent %lu\n", name_len
, name
,
2183 inode
->i_ino
, dir
->i_ino
);
2187 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2188 index
, name
, name_len
, -1);
2197 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2198 btrfs_release_path(root
, path
);
2200 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2202 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2204 BTRFS_I(dir
)->log_dirty_trans
= trans
->transid
;
2206 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2210 btrfs_free_path(path
);
2214 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2215 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2216 btrfs_update_inode(trans
, root
, dir
);
2217 btrfs_drop_nlink(inode
);
2218 ret
= btrfs_update_inode(trans
, root
, inode
);
2219 dir
->i_sb
->s_dirt
= 1;
2224 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2226 struct btrfs_root
*root
;
2227 struct btrfs_trans_handle
*trans
;
2228 struct inode
*inode
= dentry
->d_inode
;
2230 unsigned long nr
= 0;
2232 root
= BTRFS_I(dir
)->root
;
2234 ret
= btrfs_check_free_space(root
, 1, 1);
2238 trans
= btrfs_start_transaction(root
, 1);
2240 btrfs_set_trans_block_group(trans
, dir
);
2241 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2242 dentry
->d_name
.name
, dentry
->d_name
.len
);
2244 if (inode
->i_nlink
== 0)
2245 ret
= btrfs_orphan_add(trans
, inode
);
2247 nr
= trans
->blocks_used
;
2249 btrfs_end_transaction_throttle(trans
, root
);
2251 btrfs_btree_balance_dirty(root
, nr
);
2255 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2257 struct inode
*inode
= dentry
->d_inode
;
2260 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2261 struct btrfs_trans_handle
*trans
;
2262 unsigned long nr
= 0;
2265 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2266 * the root of a subvolume or snapshot
2268 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2269 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
2273 ret
= btrfs_check_free_space(root
, 1, 1);
2277 trans
= btrfs_start_transaction(root
, 1);
2278 btrfs_set_trans_block_group(trans
, dir
);
2280 err
= btrfs_orphan_add(trans
, inode
);
2284 /* now the directory is empty */
2285 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2286 dentry
->d_name
.name
, dentry
->d_name
.len
);
2288 btrfs_i_size_write(inode
, 0);
2291 nr
= trans
->blocks_used
;
2292 ret
= btrfs_end_transaction_throttle(trans
, root
);
2294 btrfs_btree_balance_dirty(root
, nr
);
2303 * when truncating bytes in a file, it is possible to avoid reading
2304 * the leaves that contain only checksum items. This can be the
2305 * majority of the IO required to delete a large file, but it must
2306 * be done carefully.
2308 * The keys in the level just above the leaves are checked to make sure
2309 * the lowest key in a given leaf is a csum key, and starts at an offset
2310 * after the new size.
2312 * Then the key for the next leaf is checked to make sure it also has
2313 * a checksum item for the same file. If it does, we know our target leaf
2314 * contains only checksum items, and it can be safely freed without reading
2317 * This is just an optimization targeted at large files. It may do
2318 * nothing. It will return 0 unless things went badly.
2320 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
2321 struct btrfs_root
*root
,
2322 struct btrfs_path
*path
,
2323 struct inode
*inode
, u64 new_size
)
2325 struct btrfs_key key
;
2328 struct btrfs_key found_key
;
2329 struct btrfs_key other_key
;
2330 struct btrfs_leaf_ref
*ref
;
2334 path
->lowest_level
= 1;
2335 key
.objectid
= inode
->i_ino
;
2336 key
.type
= BTRFS_CSUM_ITEM_KEY
;
2337 key
.offset
= new_size
;
2339 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2343 if (path
->nodes
[1] == NULL
) {
2348 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
2349 nritems
= btrfs_header_nritems(path
->nodes
[1]);
2354 if (path
->slots
[1] >= nritems
)
2357 /* did we find a key greater than anything we want to delete? */
2358 if (found_key
.objectid
> inode
->i_ino
||
2359 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
2362 /* we check the next key in the node to make sure the leave contains
2363 * only checksum items. This comparison doesn't work if our
2364 * leaf is the last one in the node
2366 if (path
->slots
[1] + 1 >= nritems
) {
2368 /* search forward from the last key in the node, this
2369 * will bring us into the next node in the tree
2371 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
2373 /* unlikely, but we inc below, so check to be safe */
2374 if (found_key
.offset
== (u64
)-1)
2377 /* search_forward needs a path with locks held, do the
2378 * search again for the original key. It is possible
2379 * this will race with a balance and return a path that
2380 * we could modify, but this drop is just an optimization
2381 * and is allowed to miss some leaves.
2383 btrfs_release_path(root
, path
);
2386 /* setup a max key for search_forward */
2387 other_key
.offset
= (u64
)-1;
2388 other_key
.type
= key
.type
;
2389 other_key
.objectid
= key
.objectid
;
2391 path
->keep_locks
= 1;
2392 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
2394 path
->keep_locks
= 0;
2395 if (ret
|| found_key
.objectid
!= key
.objectid
||
2396 found_key
.type
!= key
.type
) {
2401 key
.offset
= found_key
.offset
;
2402 btrfs_release_path(root
, path
);
2407 /* we know there's one more slot after us in the tree,
2408 * read that key so we can verify it is also a checksum item
2410 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
2412 if (found_key
.objectid
< inode
->i_ino
)
2415 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
2419 * if the key for the next leaf isn't a csum key from this objectid,
2420 * we can't be sure there aren't good items inside this leaf.
2423 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
2426 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
2427 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
2429 * it is safe to delete this leaf, it contains only
2430 * csum items from this inode at an offset >= new_size
2432 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
2435 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
2436 ref
= btrfs_alloc_leaf_ref(root
, 0);
2438 ref
->root_gen
= root
->root_key
.offset
;
2439 ref
->bytenr
= leaf_start
;
2441 ref
->generation
= leaf_gen
;
2444 btrfs_sort_leaf_ref(ref
);
2446 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
2448 btrfs_free_leaf_ref(root
, ref
);
2454 btrfs_release_path(root
, path
);
2456 if (other_key
.objectid
== inode
->i_ino
&&
2457 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
2458 key
.offset
= other_key
.offset
;
2464 /* fixup any changes we've made to the path */
2465 path
->lowest_level
= 0;
2466 path
->keep_locks
= 0;
2467 btrfs_release_path(root
, path
);
2474 * this can truncate away extent items, csum items and directory items.
2475 * It starts at a high offset and removes keys until it can't find
2476 * any higher than new_size
2478 * csum items that cross the new i_size are truncated to the new size
2481 * min_type is the minimum key type to truncate down to. If set to 0, this
2482 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2484 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2485 struct btrfs_root
*root
,
2486 struct inode
*inode
,
2487 u64 new_size
, u32 min_type
)
2490 struct btrfs_path
*path
;
2491 struct btrfs_key key
;
2492 struct btrfs_key found_key
;
2494 struct extent_buffer
*leaf
;
2495 struct btrfs_file_extent_item
*fi
;
2496 u64 extent_start
= 0;
2497 u64 extent_num_bytes
= 0;
2503 int pending_del_nr
= 0;
2504 int pending_del_slot
= 0;
2505 int extent_type
= -1;
2507 u64 mask
= root
->sectorsize
- 1;
2510 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2511 path
= btrfs_alloc_path();
2515 /* FIXME, add redo link to tree so we don't leak on crash */
2516 key
.objectid
= inode
->i_ino
;
2517 key
.offset
= (u64
)-1;
2520 btrfs_init_path(path
);
2523 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2528 /* there are no items in the tree for us to truncate, we're
2531 if (path
->slots
[0] == 0) {
2540 leaf
= path
->nodes
[0];
2541 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2542 found_type
= btrfs_key_type(&found_key
);
2545 if (found_key
.objectid
!= inode
->i_ino
)
2548 if (found_type
< min_type
)
2551 item_end
= found_key
.offset
;
2552 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2553 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2554 struct btrfs_file_extent_item
);
2555 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2556 encoding
= btrfs_file_extent_compression(leaf
, fi
);
2557 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
2558 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
2560 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2562 btrfs_file_extent_num_bytes(leaf
, fi
);
2563 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2564 item_end
+= btrfs_file_extent_inline_len(leaf
,
2569 if (item_end
< new_size
) {
2570 if (found_type
== BTRFS_DIR_ITEM_KEY
)
2571 found_type
= BTRFS_INODE_ITEM_KEY
;
2572 else if (found_type
== BTRFS_EXTENT_ITEM_KEY
)
2573 found_type
= BTRFS_EXTENT_DATA_KEY
;
2574 else if (found_type
== BTRFS_EXTENT_DATA_KEY
)
2575 found_type
= BTRFS_XATTR_ITEM_KEY
;
2576 else if (found_type
== BTRFS_XATTR_ITEM_KEY
)
2577 found_type
= BTRFS_INODE_REF_KEY
;
2578 else if (found_type
)
2582 btrfs_set_key_type(&key
, found_type
);
2585 if (found_key
.offset
>= new_size
)
2591 /* FIXME, shrink the extent if the ref count is only 1 */
2592 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2595 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2597 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2598 if (!del_item
&& !encoding
) {
2599 u64 orig_num_bytes
=
2600 btrfs_file_extent_num_bytes(leaf
, fi
);
2601 extent_num_bytes
= new_size
-
2602 found_key
.offset
+ root
->sectorsize
- 1;
2603 extent_num_bytes
= extent_num_bytes
&
2604 ~((u64
)root
->sectorsize
- 1);
2605 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2607 num_dec
= (orig_num_bytes
-
2609 if (root
->ref_cows
&& extent_start
!= 0)
2610 inode_sub_bytes(inode
, num_dec
);
2611 btrfs_mark_buffer_dirty(leaf
);
2614 btrfs_file_extent_disk_num_bytes(leaf
,
2616 /* FIXME blocksize != 4096 */
2617 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2618 if (extent_start
!= 0) {
2621 inode_sub_bytes(inode
, num_dec
);
2623 root_gen
= btrfs_header_generation(leaf
);
2624 root_owner
= btrfs_header_owner(leaf
);
2626 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2628 * we can't truncate inline items that have had
2632 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2633 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2634 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2635 u32 size
= new_size
- found_key
.offset
;
2637 if (root
->ref_cows
) {
2638 inode_sub_bytes(inode
, item_end
+ 1 -
2642 btrfs_file_extent_calc_inline_size(size
);
2643 ret
= btrfs_truncate_item(trans
, root
, path
,
2646 } else if (root
->ref_cows
) {
2647 inode_sub_bytes(inode
, item_end
+ 1 -
2653 if (!pending_del_nr
) {
2654 /* no pending yet, add ourselves */
2655 pending_del_slot
= path
->slots
[0];
2657 } else if (pending_del_nr
&&
2658 path
->slots
[0] + 1 == pending_del_slot
) {
2659 /* hop on the pending chunk */
2661 pending_del_slot
= path
->slots
[0];
2669 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2671 leaf
->start
, root_owner
,
2672 root_gen
, inode
->i_ino
, 0);
2676 if (path
->slots
[0] == 0) {
2679 btrfs_release_path(root
, path
);
2684 if (pending_del_nr
&&
2685 path
->slots
[0] + 1 != pending_del_slot
) {
2686 struct btrfs_key debug
;
2688 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2690 ret
= btrfs_del_items(trans
, root
, path
,
2695 btrfs_release_path(root
, path
);
2701 if (pending_del_nr
) {
2702 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2705 btrfs_free_path(path
);
2706 inode
->i_sb
->s_dirt
= 1;
2711 * taken from block_truncate_page, but does cow as it zeros out
2712 * any bytes left in the last page in the file.
2714 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2716 struct inode
*inode
= mapping
->host
;
2717 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2718 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2719 struct btrfs_ordered_extent
*ordered
;
2721 u32 blocksize
= root
->sectorsize
;
2722 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2723 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2729 if ((offset
& (blocksize
- 1)) == 0)
2734 page
= grab_cache_page(mapping
, index
);
2738 page_start
= page_offset(page
);
2739 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2741 if (!PageUptodate(page
)) {
2742 ret
= btrfs_readpage(NULL
, page
);
2744 if (page
->mapping
!= mapping
) {
2746 page_cache_release(page
);
2749 if (!PageUptodate(page
)) {
2754 wait_on_page_writeback(page
);
2756 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2757 set_page_extent_mapped(page
);
2759 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2761 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2763 page_cache_release(page
);
2764 btrfs_start_ordered_extent(inode
, ordered
, 1);
2765 btrfs_put_ordered_extent(ordered
);
2769 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2771 if (offset
!= PAGE_CACHE_SIZE
) {
2773 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2774 flush_dcache_page(page
);
2777 ClearPageChecked(page
);
2778 set_page_dirty(page
);
2779 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2783 page_cache_release(page
);
2788 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
2790 struct btrfs_trans_handle
*trans
;
2791 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2792 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2793 struct extent_map
*em
;
2794 u64 mask
= root
->sectorsize
- 1;
2795 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2796 u64 block_end
= (size
+ mask
) & ~mask
;
2802 if (size
<= hole_start
)
2805 err
= btrfs_check_free_space(root
, 1, 0);
2809 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2812 struct btrfs_ordered_extent
*ordered
;
2813 btrfs_wait_ordered_range(inode
, hole_start
,
2814 block_end
- hole_start
);
2815 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2816 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2819 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2820 btrfs_put_ordered_extent(ordered
);
2823 trans
= btrfs_start_transaction(root
, 1);
2824 btrfs_set_trans_block_group(trans
, inode
);
2826 cur_offset
= hole_start
;
2828 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2829 block_end
- cur_offset
, 0);
2830 BUG_ON(IS_ERR(em
) || !em
);
2831 last_byte
= min(extent_map_end(em
), block_end
);
2832 last_byte
= (last_byte
+ mask
) & ~mask
;
2833 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2835 hole_size
= last_byte
- cur_offset
;
2836 err
= btrfs_drop_extents(trans
, root
, inode
,
2838 cur_offset
+ hole_size
,
2839 cur_offset
, &hint_byte
);
2842 err
= btrfs_insert_file_extent(trans
, root
,
2843 inode
->i_ino
, cur_offset
, 0,
2844 0, hole_size
, 0, hole_size
,
2846 btrfs_drop_extent_cache(inode
, hole_start
,
2849 free_extent_map(em
);
2850 cur_offset
= last_byte
;
2851 if (err
|| cur_offset
>= block_end
)
2855 btrfs_end_transaction(trans
, root
);
2856 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2860 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2862 struct inode
*inode
= dentry
->d_inode
;
2865 err
= inode_change_ok(inode
, attr
);
2869 if (S_ISREG(inode
->i_mode
) &&
2870 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
> inode
->i_size
) {
2871 err
= btrfs_cont_expand(inode
, attr
->ia_size
);
2876 err
= inode_setattr(inode
, attr
);
2878 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2879 err
= btrfs_acl_chmod(inode
);
2883 void btrfs_delete_inode(struct inode
*inode
)
2885 struct btrfs_trans_handle
*trans
;
2886 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2890 truncate_inode_pages(&inode
->i_data
, 0);
2891 if (is_bad_inode(inode
)) {
2892 btrfs_orphan_del(NULL
, inode
);
2895 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
2897 btrfs_i_size_write(inode
, 0);
2898 trans
= btrfs_join_transaction(root
, 1);
2900 btrfs_set_trans_block_group(trans
, inode
);
2901 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
2903 btrfs_orphan_del(NULL
, inode
);
2904 goto no_delete_lock
;
2907 btrfs_orphan_del(trans
, inode
);
2909 nr
= trans
->blocks_used
;
2912 btrfs_end_transaction(trans
, root
);
2913 btrfs_btree_balance_dirty(root
, nr
);
2917 nr
= trans
->blocks_used
;
2918 btrfs_end_transaction(trans
, root
);
2919 btrfs_btree_balance_dirty(root
, nr
);
2925 * this returns the key found in the dir entry in the location pointer.
2926 * If no dir entries were found, location->objectid is 0.
2928 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
2929 struct btrfs_key
*location
)
2931 const char *name
= dentry
->d_name
.name
;
2932 int namelen
= dentry
->d_name
.len
;
2933 struct btrfs_dir_item
*di
;
2934 struct btrfs_path
*path
;
2935 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2938 path
= btrfs_alloc_path();
2941 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
2946 if (!di
|| IS_ERR(di
))
2949 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
2951 btrfs_free_path(path
);
2954 location
->objectid
= 0;
2959 * when we hit a tree root in a directory, the btrfs part of the inode
2960 * needs to be changed to reflect the root directory of the tree root. This
2961 * is kind of like crossing a mount point.
2963 static int fixup_tree_root_location(struct btrfs_root
*root
,
2964 struct btrfs_key
*location
,
2965 struct btrfs_root
**sub_root
,
2966 struct dentry
*dentry
)
2968 struct btrfs_root_item
*ri
;
2970 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
2972 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
2975 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
2976 dentry
->d_name
.name
,
2977 dentry
->d_name
.len
);
2978 if (IS_ERR(*sub_root
))
2979 return PTR_ERR(*sub_root
);
2981 ri
= &(*sub_root
)->root_item
;
2982 location
->objectid
= btrfs_root_dirid(ri
);
2983 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2984 location
->offset
= 0;
2989 static noinline
void init_btrfs_i(struct inode
*inode
)
2991 struct btrfs_inode
*bi
= BTRFS_I(inode
);
2994 bi
->i_default_acl
= NULL
;
2999 bi
->logged_trans
= 0;
3000 bi
->delalloc_bytes
= 0;
3001 bi
->disk_i_size
= 0;
3003 bi
->index_cnt
= (u64
)-1;
3004 bi
->log_dirty_trans
= 0;
3005 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
3006 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
3007 inode
->i_mapping
, GFP_NOFS
);
3008 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
3009 inode
->i_mapping
, GFP_NOFS
);
3010 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
3011 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
3012 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
3013 mutex_init(&BTRFS_I(inode
)->log_mutex
);
3016 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3018 struct btrfs_iget_args
*args
= p
;
3019 inode
->i_ino
= args
->ino
;
3020 init_btrfs_i(inode
);
3021 BTRFS_I(inode
)->root
= args
->root
;
3025 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3027 struct btrfs_iget_args
*args
= opaque
;
3028 return args
->ino
== inode
->i_ino
&&
3029 args
->root
== BTRFS_I(inode
)->root
;
3032 struct inode
*btrfs_ilookup(struct super_block
*s
, u64 objectid
,
3033 struct btrfs_root
*root
, int wait
)
3035 struct inode
*inode
;
3036 struct btrfs_iget_args args
;
3037 args
.ino
= objectid
;
3041 inode
= ilookup5(s
, objectid
, btrfs_find_actor
,
3044 inode
= ilookup5_nowait(s
, objectid
, btrfs_find_actor
,
3050 struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 objectid
,
3051 struct btrfs_root
*root
)
3053 struct inode
*inode
;
3054 struct btrfs_iget_args args
;
3055 args
.ino
= objectid
;
3058 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3059 btrfs_init_locked_inode
,
3064 /* Get an inode object given its location and corresponding root.
3065 * Returns in *is_new if the inode was read from disk
3067 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3068 struct btrfs_root
*root
, int *is_new
)
3070 struct inode
*inode
;
3072 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3074 return ERR_PTR(-EACCES
);
3076 if (inode
->i_state
& I_NEW
) {
3077 BTRFS_I(inode
)->root
= root
;
3078 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3079 btrfs_read_locked_inode(inode
);
3080 unlock_new_inode(inode
);
3091 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3093 struct inode
*inode
;
3094 struct btrfs_inode
*bi
= BTRFS_I(dir
);
3095 struct btrfs_root
*root
= bi
->root
;
3096 struct btrfs_root
*sub_root
= root
;
3097 struct btrfs_key location
;
3100 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3101 return ERR_PTR(-ENAMETOOLONG
);
3103 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3106 return ERR_PTR(ret
);
3109 if (location
.objectid
) {
3110 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
3113 return ERR_PTR(ret
);
3115 return ERR_PTR(-ENOENT
);
3116 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, &new);
3118 return ERR_CAST(inode
);
3123 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
3124 struct nameidata
*nd
)
3126 struct inode
*inode
;
3128 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3129 return ERR_PTR(-ENAMETOOLONG
);
3131 inode
= btrfs_lookup_dentry(dir
, dentry
);
3133 return ERR_CAST(inode
);
3135 return d_splice_alias(inode
, dentry
);
3138 static unsigned char btrfs_filetype_table
[] = {
3139 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
3142 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
3145 struct inode
*inode
= filp
->f_dentry
->d_inode
;
3146 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3147 struct btrfs_item
*item
;
3148 struct btrfs_dir_item
*di
;
3149 struct btrfs_key key
;
3150 struct btrfs_key found_key
;
3151 struct btrfs_path
*path
;
3154 struct extent_buffer
*leaf
;
3157 unsigned char d_type
;
3162 int key_type
= BTRFS_DIR_INDEX_KEY
;
3167 /* FIXME, use a real flag for deciding about the key type */
3168 if (root
->fs_info
->tree_root
== root
)
3169 key_type
= BTRFS_DIR_ITEM_KEY
;
3171 /* special case for "." */
3172 if (filp
->f_pos
== 0) {
3173 over
= filldir(dirent
, ".", 1,
3180 /* special case for .., just use the back ref */
3181 if (filp
->f_pos
== 1) {
3182 u64 pino
= parent_ino(filp
->f_path
.dentry
);
3183 over
= filldir(dirent
, "..", 2,
3189 path
= btrfs_alloc_path();
3192 btrfs_set_key_type(&key
, key_type
);
3193 key
.offset
= filp
->f_pos
;
3194 key
.objectid
= inode
->i_ino
;
3196 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3202 leaf
= path
->nodes
[0];
3203 nritems
= btrfs_header_nritems(leaf
);
3204 slot
= path
->slots
[0];
3205 if (advance
|| slot
>= nritems
) {
3206 if (slot
>= nritems
- 1) {
3207 ret
= btrfs_next_leaf(root
, path
);
3210 leaf
= path
->nodes
[0];
3211 nritems
= btrfs_header_nritems(leaf
);
3212 slot
= path
->slots
[0];
3220 item
= btrfs_item_nr(leaf
, slot
);
3221 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3223 if (found_key
.objectid
!= key
.objectid
)
3225 if (btrfs_key_type(&found_key
) != key_type
)
3227 if (found_key
.offset
< filp
->f_pos
)
3230 filp
->f_pos
= found_key
.offset
;
3232 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
3234 di_total
= btrfs_item_size(leaf
, item
);
3236 while (di_cur
< di_total
) {
3237 struct btrfs_key location
;
3239 name_len
= btrfs_dir_name_len(leaf
, di
);
3240 if (name_len
<= sizeof(tmp_name
)) {
3241 name_ptr
= tmp_name
;
3243 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
3249 read_extent_buffer(leaf
, name_ptr
,
3250 (unsigned long)(di
+ 1), name_len
);
3252 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
3253 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
3255 /* is this a reference to our own snapshot? If so
3258 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
3259 location
.objectid
== root
->root_key
.objectid
) {
3263 over
= filldir(dirent
, name_ptr
, name_len
,
3264 found_key
.offset
, location
.objectid
,
3268 if (name_ptr
!= tmp_name
)
3273 di_len
= btrfs_dir_name_len(leaf
, di
) +
3274 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
3276 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
3280 /* Reached end of directory/root. Bump pos past the last item. */
3281 if (key_type
== BTRFS_DIR_INDEX_KEY
)
3282 filp
->f_pos
= INT_LIMIT(off_t
);
3288 btrfs_free_path(path
);
3292 int btrfs_write_inode(struct inode
*inode
, int wait
)
3294 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3295 struct btrfs_trans_handle
*trans
;
3298 if (root
->fs_info
->btree_inode
== inode
)
3302 trans
= btrfs_join_transaction(root
, 1);
3303 btrfs_set_trans_block_group(trans
, inode
);
3304 ret
= btrfs_commit_transaction(trans
, root
);
3310 * This is somewhat expensive, updating the tree every time the
3311 * inode changes. But, it is most likely to find the inode in cache.
3312 * FIXME, needs more benchmarking...there are no reasons other than performance
3313 * to keep or drop this code.
3315 void btrfs_dirty_inode(struct inode
*inode
)
3317 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3318 struct btrfs_trans_handle
*trans
;
3320 trans
= btrfs_join_transaction(root
, 1);
3321 btrfs_set_trans_block_group(trans
, inode
);
3322 btrfs_update_inode(trans
, root
, inode
);
3323 btrfs_end_transaction(trans
, root
);
3327 * find the highest existing sequence number in a directory
3328 * and then set the in-memory index_cnt variable to reflect
3329 * free sequence numbers
3331 static int btrfs_set_inode_index_count(struct inode
*inode
)
3333 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3334 struct btrfs_key key
, found_key
;
3335 struct btrfs_path
*path
;
3336 struct extent_buffer
*leaf
;
3339 key
.objectid
= inode
->i_ino
;
3340 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
3341 key
.offset
= (u64
)-1;
3343 path
= btrfs_alloc_path();
3347 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3350 /* FIXME: we should be able to handle this */
3356 * MAGIC NUMBER EXPLANATION:
3357 * since we search a directory based on f_pos we have to start at 2
3358 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3359 * else has to start at 2
3361 if (path
->slots
[0] == 0) {
3362 BTRFS_I(inode
)->index_cnt
= 2;
3368 leaf
= path
->nodes
[0];
3369 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3371 if (found_key
.objectid
!= inode
->i_ino
||
3372 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
3373 BTRFS_I(inode
)->index_cnt
= 2;
3377 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
3379 btrfs_free_path(path
);
3384 * helper to find a free sequence number in a given directory. This current
3385 * code is very simple, later versions will do smarter things in the btree
3387 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
3391 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
3392 ret
= btrfs_set_inode_index_count(dir
);
3397 *index
= BTRFS_I(dir
)->index_cnt
;
3398 BTRFS_I(dir
)->index_cnt
++;
3403 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
3404 struct btrfs_root
*root
,
3406 const char *name
, int name_len
,
3407 u64 ref_objectid
, u64 objectid
,
3408 u64 alloc_hint
, int mode
, u64
*index
)
3410 struct inode
*inode
;
3411 struct btrfs_inode_item
*inode_item
;
3412 struct btrfs_key
*location
;
3413 struct btrfs_path
*path
;
3414 struct btrfs_inode_ref
*ref
;
3415 struct btrfs_key key
[2];
3421 path
= btrfs_alloc_path();
3424 inode
= new_inode(root
->fs_info
->sb
);
3426 return ERR_PTR(-ENOMEM
);
3429 ret
= btrfs_set_inode_index(dir
, index
);
3431 return ERR_PTR(ret
);
3434 * index_cnt is ignored for everything but a dir,
3435 * btrfs_get_inode_index_count has an explanation for the magic
3438 init_btrfs_i(inode
);
3439 BTRFS_I(inode
)->index_cnt
= 2;
3440 BTRFS_I(inode
)->root
= root
;
3441 BTRFS_I(inode
)->generation
= trans
->transid
;
3447 BTRFS_I(inode
)->block_group
=
3448 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
3449 if ((mode
& S_IFREG
)) {
3450 if (btrfs_test_opt(root
, NODATASUM
))
3451 btrfs_set_flag(inode
, NODATASUM
);
3452 if (btrfs_test_opt(root
, NODATACOW
))
3453 btrfs_set_flag(inode
, NODATACOW
);
3456 key
[0].objectid
= objectid
;
3457 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
3460 key
[1].objectid
= objectid
;
3461 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
3462 key
[1].offset
= ref_objectid
;
3464 sizes
[0] = sizeof(struct btrfs_inode_item
);
3465 sizes
[1] = name_len
+ sizeof(*ref
);
3467 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
3471 if (objectid
> root
->highest_inode
)
3472 root
->highest_inode
= objectid
;
3474 inode
->i_uid
= current_fsuid();
3475 inode
->i_gid
= current_fsgid();
3476 inode
->i_mode
= mode
;
3477 inode
->i_ino
= objectid
;
3478 inode_set_bytes(inode
, 0);
3479 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3480 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3481 struct btrfs_inode_item
);
3482 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
3484 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3485 struct btrfs_inode_ref
);
3486 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
3487 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
3488 ptr
= (unsigned long)(ref
+ 1);
3489 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
3491 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3492 btrfs_free_path(path
);
3494 location
= &BTRFS_I(inode
)->location
;
3495 location
->objectid
= objectid
;
3496 location
->offset
= 0;
3497 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3499 insert_inode_hash(inode
);
3503 BTRFS_I(dir
)->index_cnt
--;
3504 btrfs_free_path(path
);
3505 return ERR_PTR(ret
);
3508 static inline u8
btrfs_inode_type(struct inode
*inode
)
3510 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
3514 * utility function to add 'inode' into 'parent_inode' with
3515 * a give name and a given sequence number.
3516 * if 'add_backref' is true, also insert a backref from the
3517 * inode to the parent directory.
3519 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3520 struct inode
*parent_inode
, struct inode
*inode
,
3521 const char *name
, int name_len
, int add_backref
, u64 index
)
3524 struct btrfs_key key
;
3525 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3527 key
.objectid
= inode
->i_ino
;
3528 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3531 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3532 parent_inode
->i_ino
,
3533 &key
, btrfs_inode_type(inode
),
3537 ret
= btrfs_insert_inode_ref(trans
, root
,
3540 parent_inode
->i_ino
,
3543 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3545 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3546 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3551 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3552 struct dentry
*dentry
, struct inode
*inode
,
3553 int backref
, u64 index
)
3555 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3556 inode
, dentry
->d_name
.name
,
3557 dentry
->d_name
.len
, backref
, index
);
3559 d_instantiate(dentry
, inode
);
3567 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3568 int mode
, dev_t rdev
)
3570 struct btrfs_trans_handle
*trans
;
3571 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3572 struct inode
*inode
= NULL
;
3576 unsigned long nr
= 0;
3579 if (!new_valid_dev(rdev
))
3582 err
= btrfs_check_free_space(root
, 1, 0);
3586 trans
= btrfs_start_transaction(root
, 1);
3587 btrfs_set_trans_block_group(trans
, dir
);
3589 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3595 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3597 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3598 BTRFS_I(dir
)->block_group
, mode
, &index
);
3599 err
= PTR_ERR(inode
);
3603 err
= btrfs_init_inode_security(inode
, dir
);
3609 btrfs_set_trans_block_group(trans
, inode
);
3610 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3614 inode
->i_op
= &btrfs_special_inode_operations
;
3615 init_special_inode(inode
, inode
->i_mode
, rdev
);
3616 btrfs_update_inode(trans
, root
, inode
);
3618 dir
->i_sb
->s_dirt
= 1;
3619 btrfs_update_inode_block_group(trans
, inode
);
3620 btrfs_update_inode_block_group(trans
, dir
);
3622 nr
= trans
->blocks_used
;
3623 btrfs_end_transaction_throttle(trans
, root
);
3626 inode_dec_link_count(inode
);
3629 btrfs_btree_balance_dirty(root
, nr
);
3633 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3634 int mode
, struct nameidata
*nd
)
3636 struct btrfs_trans_handle
*trans
;
3637 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3638 struct inode
*inode
= NULL
;
3641 unsigned long nr
= 0;
3645 err
= btrfs_check_free_space(root
, 1, 0);
3648 trans
= btrfs_start_transaction(root
, 1);
3649 btrfs_set_trans_block_group(trans
, dir
);
3651 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3657 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3659 dentry
->d_parent
->d_inode
->i_ino
,
3660 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3662 err
= PTR_ERR(inode
);
3666 err
= btrfs_init_inode_security(inode
, dir
);
3672 btrfs_set_trans_block_group(trans
, inode
);
3673 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3677 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3678 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3679 inode
->i_fop
= &btrfs_file_operations
;
3680 inode
->i_op
= &btrfs_file_inode_operations
;
3681 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3683 dir
->i_sb
->s_dirt
= 1;
3684 btrfs_update_inode_block_group(trans
, inode
);
3685 btrfs_update_inode_block_group(trans
, dir
);
3687 nr
= trans
->blocks_used
;
3688 btrfs_end_transaction_throttle(trans
, root
);
3691 inode_dec_link_count(inode
);
3694 btrfs_btree_balance_dirty(root
, nr
);
3698 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3699 struct dentry
*dentry
)
3701 struct btrfs_trans_handle
*trans
;
3702 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3703 struct inode
*inode
= old_dentry
->d_inode
;
3705 unsigned long nr
= 0;
3709 if (inode
->i_nlink
== 0)
3712 btrfs_inc_nlink(inode
);
3713 err
= btrfs_check_free_space(root
, 1, 0);
3716 err
= btrfs_set_inode_index(dir
, &index
);
3720 trans
= btrfs_start_transaction(root
, 1);
3722 btrfs_set_trans_block_group(trans
, dir
);
3723 atomic_inc(&inode
->i_count
);
3725 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3730 dir
->i_sb
->s_dirt
= 1;
3731 btrfs_update_inode_block_group(trans
, dir
);
3732 err
= btrfs_update_inode(trans
, root
, inode
);
3737 nr
= trans
->blocks_used
;
3738 btrfs_end_transaction_throttle(trans
, root
);
3741 inode_dec_link_count(inode
);
3744 btrfs_btree_balance_dirty(root
, nr
);
3748 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3750 struct inode
*inode
= NULL
;
3751 struct btrfs_trans_handle
*trans
;
3752 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3754 int drop_on_err
= 0;
3757 unsigned long nr
= 1;
3759 err
= btrfs_check_free_space(root
, 1, 0);
3763 trans
= btrfs_start_transaction(root
, 1);
3764 btrfs_set_trans_block_group(trans
, dir
);
3766 if (IS_ERR(trans
)) {
3767 err
= PTR_ERR(trans
);
3771 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3777 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3779 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3780 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3782 if (IS_ERR(inode
)) {
3783 err
= PTR_ERR(inode
);
3789 err
= btrfs_init_inode_security(inode
, dir
);
3793 inode
->i_op
= &btrfs_dir_inode_operations
;
3794 inode
->i_fop
= &btrfs_dir_file_operations
;
3795 btrfs_set_trans_block_group(trans
, inode
);
3797 btrfs_i_size_write(inode
, 0);
3798 err
= btrfs_update_inode(trans
, root
, inode
);
3802 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3803 inode
, dentry
->d_name
.name
,
3804 dentry
->d_name
.len
, 0, index
);
3808 d_instantiate(dentry
, inode
);
3810 dir
->i_sb
->s_dirt
= 1;
3811 btrfs_update_inode_block_group(trans
, inode
);
3812 btrfs_update_inode_block_group(trans
, dir
);
3815 nr
= trans
->blocks_used
;
3816 btrfs_end_transaction_throttle(trans
, root
);
3821 btrfs_btree_balance_dirty(root
, nr
);
3825 /* helper for btfs_get_extent. Given an existing extent in the tree,
3826 * and an extent that you want to insert, deal with overlap and insert
3827 * the new extent into the tree.
3829 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3830 struct extent_map
*existing
,
3831 struct extent_map
*em
,
3832 u64 map_start
, u64 map_len
)
3836 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3837 start_diff
= map_start
- em
->start
;
3838 em
->start
= map_start
;
3840 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3841 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3842 em
->block_start
+= start_diff
;
3843 em
->block_len
-= start_diff
;
3845 return add_extent_mapping(em_tree
, em
);
3848 static noinline
int uncompress_inline(struct btrfs_path
*path
,
3849 struct inode
*inode
, struct page
*page
,
3850 size_t pg_offset
, u64 extent_offset
,
3851 struct btrfs_file_extent_item
*item
)
3854 struct extent_buffer
*leaf
= path
->nodes
[0];
3857 unsigned long inline_size
;
3860 WARN_ON(pg_offset
!= 0);
3861 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
3862 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
3863 btrfs_item_nr(leaf
, path
->slots
[0]));
3864 tmp
= kmalloc(inline_size
, GFP_NOFS
);
3865 ptr
= btrfs_file_extent_inline_start(item
);
3867 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
3869 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
3870 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
3871 inline_size
, max_size
);
3873 char *kaddr
= kmap_atomic(page
, KM_USER0
);
3874 unsigned long copy_size
= min_t(u64
,
3875 PAGE_CACHE_SIZE
- pg_offset
,
3876 max_size
- extent_offset
);
3877 memset(kaddr
+ pg_offset
, 0, copy_size
);
3878 kunmap_atomic(kaddr
, KM_USER0
);
3885 * a bit scary, this does extent mapping from logical file offset to the disk.
3886 * the ugly parts come from merging extents from the disk with the in-ram
3887 * representation. This gets more complex because of the data=ordered code,
3888 * where the in-ram extents might be locked pending data=ordered completion.
3890 * This also copies inline extents directly into the page.
3893 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
3894 size_t pg_offset
, u64 start
, u64 len
,
3900 u64 extent_start
= 0;
3902 u64 objectid
= inode
->i_ino
;
3904 struct btrfs_path
*path
= NULL
;
3905 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3906 struct btrfs_file_extent_item
*item
;
3907 struct extent_buffer
*leaf
;
3908 struct btrfs_key found_key
;
3909 struct extent_map
*em
= NULL
;
3910 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3911 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3912 struct btrfs_trans_handle
*trans
= NULL
;
3916 spin_lock(&em_tree
->lock
);
3917 em
= lookup_extent_mapping(em_tree
, start
, len
);
3919 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3920 spin_unlock(&em_tree
->lock
);
3923 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
3924 free_extent_map(em
);
3925 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
3926 free_extent_map(em
);
3930 em
= alloc_extent_map(GFP_NOFS
);
3935 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3936 em
->start
= EXTENT_MAP_HOLE
;
3937 em
->orig_start
= EXTENT_MAP_HOLE
;
3939 em
->block_len
= (u64
)-1;
3942 path
= btrfs_alloc_path();
3946 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3947 objectid
, start
, trans
!= NULL
);
3954 if (path
->slots
[0] == 0)
3959 leaf
= path
->nodes
[0];
3960 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3961 struct btrfs_file_extent_item
);
3962 /* are we inside the extent that was found? */
3963 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3964 found_type
= btrfs_key_type(&found_key
);
3965 if (found_key
.objectid
!= objectid
||
3966 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3970 found_type
= btrfs_file_extent_type(leaf
, item
);
3971 extent_start
= found_key
.offset
;
3972 compressed
= btrfs_file_extent_compression(leaf
, item
);
3973 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3974 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3975 extent_end
= extent_start
+
3976 btrfs_file_extent_num_bytes(leaf
, item
);
3977 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3979 size
= btrfs_file_extent_inline_len(leaf
, item
);
3980 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
3981 ~((u64
)root
->sectorsize
- 1);
3984 if (start
>= extent_end
) {
3986 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3987 ret
= btrfs_next_leaf(root
, path
);
3994 leaf
= path
->nodes
[0];
3996 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3997 if (found_key
.objectid
!= objectid
||
3998 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
4000 if (start
+ len
<= found_key
.offset
)
4003 em
->len
= found_key
.offset
- start
;
4007 if (found_type
== BTRFS_FILE_EXTENT_REG
||
4008 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4009 em
->start
= extent_start
;
4010 em
->len
= extent_end
- extent_start
;
4011 em
->orig_start
= extent_start
-
4012 btrfs_file_extent_offset(leaf
, item
);
4013 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
4015 em
->block_start
= EXTENT_MAP_HOLE
;
4019 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4020 em
->block_start
= bytenr
;
4021 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
4024 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
4025 em
->block_start
= bytenr
;
4026 em
->block_len
= em
->len
;
4027 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
4028 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
4031 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4035 size_t extent_offset
;
4038 em
->block_start
= EXTENT_MAP_INLINE
;
4039 if (!page
|| create
) {
4040 em
->start
= extent_start
;
4041 em
->len
= extent_end
- extent_start
;
4045 size
= btrfs_file_extent_inline_len(leaf
, item
);
4046 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
4047 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
4048 size
- extent_offset
);
4049 em
->start
= extent_start
+ extent_offset
;
4050 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
4051 ~((u64
)root
->sectorsize
- 1);
4052 em
->orig_start
= EXTENT_MAP_INLINE
;
4054 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4055 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
4056 if (create
== 0 && !PageUptodate(page
)) {
4057 if (btrfs_file_extent_compression(leaf
, item
) ==
4058 BTRFS_COMPRESS_ZLIB
) {
4059 ret
= uncompress_inline(path
, inode
, page
,
4061 extent_offset
, item
);
4065 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4069 flush_dcache_page(page
);
4070 } else if (create
&& PageUptodate(page
)) {
4073 free_extent_map(em
);
4075 btrfs_release_path(root
, path
);
4076 trans
= btrfs_join_transaction(root
, 1);
4080 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4083 btrfs_mark_buffer_dirty(leaf
);
4085 set_extent_uptodate(io_tree
, em
->start
,
4086 extent_map_end(em
) - 1, GFP_NOFS
);
4089 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
4096 em
->block_start
= EXTENT_MAP_HOLE
;
4097 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
4099 btrfs_release_path(root
, path
);
4100 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
4101 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
4102 "[%llu %llu]\n", (unsigned long long)em
->start
,
4103 (unsigned long long)em
->len
,
4104 (unsigned long long)start
,
4105 (unsigned long long)len
);
4111 spin_lock(&em_tree
->lock
);
4112 ret
= add_extent_mapping(em_tree
, em
);
4113 /* it is possible that someone inserted the extent into the tree
4114 * while we had the lock dropped. It is also possible that
4115 * an overlapping map exists in the tree
4117 if (ret
== -EEXIST
) {
4118 struct extent_map
*existing
;
4122 existing
= lookup_extent_mapping(em_tree
, start
, len
);
4123 if (existing
&& (existing
->start
> start
||
4124 existing
->start
+ existing
->len
<= start
)) {
4125 free_extent_map(existing
);
4129 existing
= lookup_extent_mapping(em_tree
, em
->start
,
4132 err
= merge_extent_mapping(em_tree
, existing
,
4135 free_extent_map(existing
);
4137 free_extent_map(em
);
4142 free_extent_map(em
);
4146 free_extent_map(em
);
4151 spin_unlock(&em_tree
->lock
);
4154 btrfs_free_path(path
);
4156 ret
= btrfs_end_transaction(trans
, root
);
4161 free_extent_map(em
);
4163 return ERR_PTR(err
);
4168 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
4169 const struct iovec
*iov
, loff_t offset
,
4170 unsigned long nr_segs
)
4175 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
4176 __u64 start
, __u64 len
)
4178 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent
);
4181 int btrfs_readpage(struct file
*file
, struct page
*page
)
4183 struct extent_io_tree
*tree
;
4184 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4185 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
4188 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
4190 struct extent_io_tree
*tree
;
4193 if (current
->flags
& PF_MEMALLOC
) {
4194 redirty_page_for_writepage(wbc
, page
);
4198 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4199 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
4202 int btrfs_writepages(struct address_space
*mapping
,
4203 struct writeback_control
*wbc
)
4205 struct extent_io_tree
*tree
;
4207 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4208 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
4212 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
4213 struct list_head
*pages
, unsigned nr_pages
)
4215 struct extent_io_tree
*tree
;
4216 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4217 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
4220 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4222 struct extent_io_tree
*tree
;
4223 struct extent_map_tree
*map
;
4226 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4227 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
4228 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
4230 ClearPagePrivate(page
);
4231 set_page_private(page
, 0);
4232 page_cache_release(page
);
4237 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4239 if (PageWriteback(page
) || PageDirty(page
))
4241 return __btrfs_releasepage(page
, gfp_flags
);
4244 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
4246 struct extent_io_tree
*tree
;
4247 struct btrfs_ordered_extent
*ordered
;
4248 u64 page_start
= page_offset(page
);
4249 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4251 wait_on_page_writeback(page
);
4252 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4254 btrfs_releasepage(page
, GFP_NOFS
);
4258 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4259 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
4263 * IO on this page will never be started, so we need
4264 * to account for any ordered extents now
4266 clear_extent_bit(tree
, page_start
, page_end
,
4267 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4268 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
4269 btrfs_finish_ordered_io(page
->mapping
->host
,
4270 page_start
, page_end
);
4271 btrfs_put_ordered_extent(ordered
);
4272 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4274 clear_extent_bit(tree
, page_start
, page_end
,
4275 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4278 __btrfs_releasepage(page
, GFP_NOFS
);
4280 ClearPageChecked(page
);
4281 if (PagePrivate(page
)) {
4282 ClearPagePrivate(page
);
4283 set_page_private(page
, 0);
4284 page_cache_release(page
);
4289 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4290 * called from a page fault handler when a page is first dirtied. Hence we must
4291 * be careful to check for EOF conditions here. We set the page up correctly
4292 * for a written page which means we get ENOSPC checking when writing into
4293 * holes and correct delalloc and unwritten extent mapping on filesystems that
4294 * support these features.
4296 * We are not allowed to take the i_mutex here so we have to play games to
4297 * protect against truncate races as the page could now be beyond EOF. Because
4298 * vmtruncate() writes the inode size before removing pages, once we have the
4299 * page lock we can determine safely if the page is beyond EOF. If it is not
4300 * beyond EOF, then the page is guaranteed safe against truncation until we
4303 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4305 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
4306 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4307 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4308 struct btrfs_ordered_extent
*ordered
;
4310 unsigned long zero_start
;
4316 ret
= btrfs_check_free_space(root
, PAGE_CACHE_SIZE
, 0);
4323 size
= i_size_read(inode
);
4324 page_start
= page_offset(page
);
4325 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4327 if ((page
->mapping
!= inode
->i_mapping
) ||
4328 (page_start
>= size
)) {
4329 /* page got truncated out from underneath us */
4332 wait_on_page_writeback(page
);
4334 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4335 set_page_extent_mapped(page
);
4338 * we can't set the delalloc bits if there are pending ordered
4339 * extents. Drop our locks and wait for them to finish
4341 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4343 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4345 btrfs_start_ordered_extent(inode
, ordered
, 1);
4346 btrfs_put_ordered_extent(ordered
);
4350 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
4353 /* page is wholly or partially inside EOF */
4354 if (page_start
+ PAGE_CACHE_SIZE
> size
)
4355 zero_start
= size
& ~PAGE_CACHE_MASK
;
4357 zero_start
= PAGE_CACHE_SIZE
;
4359 if (zero_start
!= PAGE_CACHE_SIZE
) {
4361 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
4362 flush_dcache_page(page
);
4365 ClearPageChecked(page
);
4366 set_page_dirty(page
);
4367 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4375 static void btrfs_truncate(struct inode
*inode
)
4377 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4379 struct btrfs_trans_handle
*trans
;
4381 u64 mask
= root
->sectorsize
- 1;
4383 if (!S_ISREG(inode
->i_mode
))
4385 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4388 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
4389 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
4391 trans
= btrfs_start_transaction(root
, 1);
4392 btrfs_set_trans_block_group(trans
, inode
);
4393 btrfs_i_size_write(inode
, inode
->i_size
);
4395 ret
= btrfs_orphan_add(trans
, inode
);
4398 /* FIXME, add redo link to tree so we don't leak on crash */
4399 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
4400 BTRFS_EXTENT_DATA_KEY
);
4401 btrfs_update_inode(trans
, root
, inode
);
4403 ret
= btrfs_orphan_del(trans
, inode
);
4407 nr
= trans
->blocks_used
;
4408 ret
= btrfs_end_transaction_throttle(trans
, root
);
4410 btrfs_btree_balance_dirty(root
, nr
);
4414 * create a new subvolume directory/inode (helper for the ioctl).
4416 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
4417 struct btrfs_root
*new_root
, struct dentry
*dentry
,
4418 u64 new_dirid
, u64 alloc_hint
)
4420 struct inode
*inode
;
4424 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
4425 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
4427 return PTR_ERR(inode
);
4428 inode
->i_op
= &btrfs_dir_inode_operations
;
4429 inode
->i_fop
= &btrfs_dir_file_operations
;
4432 btrfs_i_size_write(inode
, 0);
4434 error
= btrfs_update_inode(trans
, new_root
, inode
);
4438 d_instantiate(dentry
, inode
);
4442 /* helper function for file defrag and space balancing. This
4443 * forces readahead on a given range of bytes in an inode
4445 unsigned long btrfs_force_ra(struct address_space
*mapping
,
4446 struct file_ra_state
*ra
, struct file
*file
,
4447 pgoff_t offset
, pgoff_t last_index
)
4449 pgoff_t req_size
= last_index
- offset
+ 1;
4451 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
4452 return offset
+ req_size
;
4455 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
4457 struct btrfs_inode
*ei
;
4459 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
4463 ei
->logged_trans
= 0;
4464 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
4465 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
4466 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
4467 INIT_LIST_HEAD(&ei
->i_orphan
);
4468 return &ei
->vfs_inode
;
4471 void btrfs_destroy_inode(struct inode
*inode
)
4473 struct btrfs_ordered_extent
*ordered
;
4474 WARN_ON(!list_empty(&inode
->i_dentry
));
4475 WARN_ON(inode
->i_data
.nrpages
);
4477 if (BTRFS_I(inode
)->i_acl
&&
4478 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
4479 posix_acl_release(BTRFS_I(inode
)->i_acl
);
4480 if (BTRFS_I(inode
)->i_default_acl
&&
4481 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
4482 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
4484 spin_lock(&BTRFS_I(inode
)->root
->list_lock
);
4485 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
4486 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
4487 " list\n", inode
->i_ino
);
4490 spin_unlock(&BTRFS_I(inode
)->root
->list_lock
);
4493 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
4497 printk(KERN_ERR
"btrfs found ordered "
4498 "extent %llu %llu on inode cleanup\n",
4499 (unsigned long long)ordered
->file_offset
,
4500 (unsigned long long)ordered
->len
);
4501 btrfs_remove_ordered_extent(inode
, ordered
);
4502 btrfs_put_ordered_extent(ordered
);
4503 btrfs_put_ordered_extent(ordered
);
4506 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4507 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4510 static void init_once(void *foo
)
4512 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4514 inode_init_once(&ei
->vfs_inode
);
4517 void btrfs_destroy_cachep(void)
4519 if (btrfs_inode_cachep
)
4520 kmem_cache_destroy(btrfs_inode_cachep
);
4521 if (btrfs_trans_handle_cachep
)
4522 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4523 if (btrfs_transaction_cachep
)
4524 kmem_cache_destroy(btrfs_transaction_cachep
);
4525 if (btrfs_bit_radix_cachep
)
4526 kmem_cache_destroy(btrfs_bit_radix_cachep
);
4527 if (btrfs_path_cachep
)
4528 kmem_cache_destroy(btrfs_path_cachep
);
4531 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
4532 unsigned long extra_flags
,
4533 void (*ctor
)(void *))
4535 return kmem_cache_create(name
, size
, 0, (SLAB_RECLAIM_ACCOUNT
|
4536 SLAB_MEM_SPREAD
| extra_flags
), ctor
);
4539 int btrfs_init_cachep(void)
4541 btrfs_inode_cachep
= btrfs_cache_create("btrfs_inode_cache",
4542 sizeof(struct btrfs_inode
),
4544 if (!btrfs_inode_cachep
)
4546 btrfs_trans_handle_cachep
=
4547 btrfs_cache_create("btrfs_trans_handle_cache",
4548 sizeof(struct btrfs_trans_handle
),
4550 if (!btrfs_trans_handle_cachep
)
4552 btrfs_transaction_cachep
= btrfs_cache_create("btrfs_transaction_cache",
4553 sizeof(struct btrfs_transaction
),
4555 if (!btrfs_transaction_cachep
)
4557 btrfs_path_cachep
= btrfs_cache_create("btrfs_path_cache",
4558 sizeof(struct btrfs_path
),
4560 if (!btrfs_path_cachep
)
4562 btrfs_bit_radix_cachep
= btrfs_cache_create("btrfs_radix", 256,
4563 SLAB_DESTROY_BY_RCU
, NULL
);
4564 if (!btrfs_bit_radix_cachep
)
4568 btrfs_destroy_cachep();
4572 static int btrfs_getattr(struct vfsmount
*mnt
,
4573 struct dentry
*dentry
, struct kstat
*stat
)
4575 struct inode
*inode
= dentry
->d_inode
;
4576 generic_fillattr(inode
, stat
);
4577 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
4578 stat
->blksize
= PAGE_CACHE_SIZE
;
4579 stat
->blocks
= (inode_get_bytes(inode
) +
4580 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4584 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4585 struct inode
*new_dir
, struct dentry
*new_dentry
)
4587 struct btrfs_trans_handle
*trans
;
4588 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4589 struct inode
*new_inode
= new_dentry
->d_inode
;
4590 struct inode
*old_inode
= old_dentry
->d_inode
;
4591 struct timespec ctime
= CURRENT_TIME
;
4595 /* we're not allowed to rename between subvolumes */
4596 if (BTRFS_I(old_inode
)->root
->root_key
.objectid
!=
4597 BTRFS_I(new_dir
)->root
->root_key
.objectid
)
4600 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4601 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4605 /* to rename a snapshot or subvolume, we need to juggle the
4606 * backrefs. This isn't coded yet
4608 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
4611 ret
= btrfs_check_free_space(root
, 1, 0);
4615 trans
= btrfs_start_transaction(root
, 1);
4617 btrfs_set_trans_block_group(trans
, new_dir
);
4619 btrfs_inc_nlink(old_dentry
->d_inode
);
4620 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4621 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4622 old_inode
->i_ctime
= ctime
;
4624 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4625 old_dentry
->d_name
.name
,
4626 old_dentry
->d_name
.len
);
4631 new_inode
->i_ctime
= CURRENT_TIME
;
4632 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4633 new_dentry
->d_inode
,
4634 new_dentry
->d_name
.name
,
4635 new_dentry
->d_name
.len
);
4638 if (new_inode
->i_nlink
== 0) {
4639 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4645 ret
= btrfs_set_inode_index(new_dir
, &index
);
4649 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4650 old_inode
, new_dentry
->d_name
.name
,
4651 new_dentry
->d_name
.len
, 1, index
);
4656 btrfs_end_transaction_throttle(trans
, root
);
4662 * some fairly slow code that needs optimization. This walks the list
4663 * of all the inodes with pending delalloc and forces them to disk.
4665 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4667 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4668 struct btrfs_inode
*binode
;
4669 struct inode
*inode
;
4671 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
4674 spin_lock(&root
->fs_info
->delalloc_lock
);
4675 while (!list_empty(head
)) {
4676 binode
= list_entry(head
->next
, struct btrfs_inode
,
4678 inode
= igrab(&binode
->vfs_inode
);
4680 list_del_init(&binode
->delalloc_inodes
);
4681 spin_unlock(&root
->fs_info
->delalloc_lock
);
4683 filemap_flush(inode
->i_mapping
);
4687 spin_lock(&root
->fs_info
->delalloc_lock
);
4689 spin_unlock(&root
->fs_info
->delalloc_lock
);
4691 /* the filemap_flush will queue IO into the worker threads, but
4692 * we have to make sure the IO is actually started and that
4693 * ordered extents get created before we return
4695 atomic_inc(&root
->fs_info
->async_submit_draining
);
4696 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
4697 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
4698 wait_event(root
->fs_info
->async_submit_wait
,
4699 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
4700 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
4702 atomic_dec(&root
->fs_info
->async_submit_draining
);
4706 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4707 const char *symname
)
4709 struct btrfs_trans_handle
*trans
;
4710 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4711 struct btrfs_path
*path
;
4712 struct btrfs_key key
;
4713 struct inode
*inode
= NULL
;
4721 struct btrfs_file_extent_item
*ei
;
4722 struct extent_buffer
*leaf
;
4723 unsigned long nr
= 0;
4725 name_len
= strlen(symname
) + 1;
4726 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4727 return -ENAMETOOLONG
;
4729 err
= btrfs_check_free_space(root
, 1, 0);
4733 trans
= btrfs_start_transaction(root
, 1);
4734 btrfs_set_trans_block_group(trans
, dir
);
4736 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4742 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4744 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4745 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4747 err
= PTR_ERR(inode
);
4751 err
= btrfs_init_inode_security(inode
, dir
);
4757 btrfs_set_trans_block_group(trans
, inode
);
4758 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4762 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4763 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4764 inode
->i_fop
= &btrfs_file_operations
;
4765 inode
->i_op
= &btrfs_file_inode_operations
;
4766 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4768 dir
->i_sb
->s_dirt
= 1;
4769 btrfs_update_inode_block_group(trans
, inode
);
4770 btrfs_update_inode_block_group(trans
, dir
);
4774 path
= btrfs_alloc_path();
4776 key
.objectid
= inode
->i_ino
;
4778 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
4779 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
4780 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
4786 leaf
= path
->nodes
[0];
4787 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
4788 struct btrfs_file_extent_item
);
4789 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
4790 btrfs_set_file_extent_type(leaf
, ei
,
4791 BTRFS_FILE_EXTENT_INLINE
);
4792 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
4793 btrfs_set_file_extent_compression(leaf
, ei
, 0);
4794 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
4795 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
4797 ptr
= btrfs_file_extent_inline_start(ei
);
4798 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
4799 btrfs_mark_buffer_dirty(leaf
);
4800 btrfs_free_path(path
);
4802 inode
->i_op
= &btrfs_symlink_inode_operations
;
4803 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
4804 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4805 inode_set_bytes(inode
, name_len
);
4806 btrfs_i_size_write(inode
, name_len
- 1);
4807 err
= btrfs_update_inode(trans
, root
, inode
);
4812 nr
= trans
->blocks_used
;
4813 btrfs_end_transaction_throttle(trans
, root
);
4816 inode_dec_link_count(inode
);
4819 btrfs_btree_balance_dirty(root
, nr
);
4823 static int prealloc_file_range(struct inode
*inode
, u64 start
, u64 end
,
4824 u64 alloc_hint
, int mode
)
4826 struct btrfs_trans_handle
*trans
;
4827 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4828 struct btrfs_key ins
;
4830 u64 cur_offset
= start
;
4831 u64 num_bytes
= end
- start
;
4834 trans
= btrfs_join_transaction(root
, 1);
4836 btrfs_set_trans_block_group(trans
, inode
);
4838 while (num_bytes
> 0) {
4839 alloc_size
= min(num_bytes
, root
->fs_info
->max_extent
);
4840 ret
= btrfs_reserve_extent(trans
, root
, alloc_size
,
4841 root
->sectorsize
, 0, alloc_hint
,
4847 ret
= insert_reserved_file_extent(trans
, inode
,
4848 cur_offset
, ins
.objectid
,
4849 ins
.offset
, ins
.offset
,
4850 ins
.offset
, 0, 0, 0,
4851 BTRFS_FILE_EXTENT_PREALLOC
);
4853 num_bytes
-= ins
.offset
;
4854 cur_offset
+= ins
.offset
;
4855 alloc_hint
= ins
.objectid
+ ins
.offset
;
4858 if (cur_offset
> start
) {
4859 inode
->i_ctime
= CURRENT_TIME
;
4860 btrfs_set_flag(inode
, PREALLOC
);
4861 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4862 cur_offset
> i_size_read(inode
))
4863 btrfs_i_size_write(inode
, cur_offset
);
4864 ret
= btrfs_update_inode(trans
, root
, inode
);
4868 btrfs_end_transaction(trans
, root
);
4872 static long btrfs_fallocate(struct inode
*inode
, int mode
,
4873 loff_t offset
, loff_t len
)
4880 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
4881 struct extent_map
*em
;
4884 alloc_start
= offset
& ~mask
;
4885 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
4887 mutex_lock(&inode
->i_mutex
);
4888 if (alloc_start
> inode
->i_size
) {
4889 ret
= btrfs_cont_expand(inode
, alloc_start
);
4895 struct btrfs_ordered_extent
*ordered
;
4896 lock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
,
4897 alloc_end
- 1, GFP_NOFS
);
4898 ordered
= btrfs_lookup_first_ordered_extent(inode
,
4901 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
4902 ordered
->file_offset
< alloc_end
) {
4903 btrfs_put_ordered_extent(ordered
);
4904 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4905 alloc_start
, alloc_end
- 1, GFP_NOFS
);
4906 btrfs_wait_ordered_range(inode
, alloc_start
,
4907 alloc_end
- alloc_start
);
4910 btrfs_put_ordered_extent(ordered
);
4915 cur_offset
= alloc_start
;
4917 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4918 alloc_end
- cur_offset
, 0);
4919 BUG_ON(IS_ERR(em
) || !em
);
4920 last_byte
= min(extent_map_end(em
), alloc_end
);
4921 last_byte
= (last_byte
+ mask
) & ~mask
;
4922 if (em
->block_start
== EXTENT_MAP_HOLE
) {
4923 ret
= prealloc_file_range(inode
, cur_offset
,
4924 last_byte
, alloc_hint
, mode
);
4926 free_extent_map(em
);
4930 if (em
->block_start
<= EXTENT_MAP_LAST_BYTE
)
4931 alloc_hint
= em
->block_start
;
4932 free_extent_map(em
);
4934 cur_offset
= last_byte
;
4935 if (cur_offset
>= alloc_end
) {
4940 unlock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, alloc_end
- 1,
4943 mutex_unlock(&inode
->i_mutex
);
4947 static int btrfs_set_page_dirty(struct page
*page
)
4949 return __set_page_dirty_nobuffers(page
);
4952 static int btrfs_permission(struct inode
*inode
, int mask
)
4954 if (btrfs_test_flag(inode
, READONLY
) && (mask
& MAY_WRITE
))
4956 return generic_permission(inode
, mask
, btrfs_check_acl
);
4959 static struct inode_operations btrfs_dir_inode_operations
= {
4960 .getattr
= btrfs_getattr
,
4961 .lookup
= btrfs_lookup
,
4962 .create
= btrfs_create
,
4963 .unlink
= btrfs_unlink
,
4965 .mkdir
= btrfs_mkdir
,
4966 .rmdir
= btrfs_rmdir
,
4967 .rename
= btrfs_rename
,
4968 .symlink
= btrfs_symlink
,
4969 .setattr
= btrfs_setattr
,
4970 .mknod
= btrfs_mknod
,
4971 .setxattr
= btrfs_setxattr
,
4972 .getxattr
= btrfs_getxattr
,
4973 .listxattr
= btrfs_listxattr
,
4974 .removexattr
= btrfs_removexattr
,
4975 .permission
= btrfs_permission
,
4977 static struct inode_operations btrfs_dir_ro_inode_operations
= {
4978 .lookup
= btrfs_lookup
,
4979 .permission
= btrfs_permission
,
4981 static struct file_operations btrfs_dir_file_operations
= {
4982 .llseek
= generic_file_llseek
,
4983 .read
= generic_read_dir
,
4984 .readdir
= btrfs_real_readdir
,
4985 .unlocked_ioctl
= btrfs_ioctl
,
4986 #ifdef CONFIG_COMPAT
4987 .compat_ioctl
= btrfs_ioctl
,
4989 .release
= btrfs_release_file
,
4990 .fsync
= btrfs_sync_file
,
4993 static struct extent_io_ops btrfs_extent_io_ops
= {
4994 .fill_delalloc
= run_delalloc_range
,
4995 .submit_bio_hook
= btrfs_submit_bio_hook
,
4996 .merge_bio_hook
= btrfs_merge_bio_hook
,
4997 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
4998 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
4999 .writepage_start_hook
= btrfs_writepage_start_hook
,
5000 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
5001 .set_bit_hook
= btrfs_set_bit_hook
,
5002 .clear_bit_hook
= btrfs_clear_bit_hook
,
5006 * btrfs doesn't support the bmap operation because swapfiles
5007 * use bmap to make a mapping of extents in the file. They assume
5008 * these extents won't change over the life of the file and they
5009 * use the bmap result to do IO directly to the drive.
5011 * the btrfs bmap call would return logical addresses that aren't
5012 * suitable for IO and they also will change frequently as COW
5013 * operations happen. So, swapfile + btrfs == corruption.
5015 * For now we're avoiding this by dropping bmap.
5017 static struct address_space_operations btrfs_aops
= {
5018 .readpage
= btrfs_readpage
,
5019 .writepage
= btrfs_writepage
,
5020 .writepages
= btrfs_writepages
,
5021 .readpages
= btrfs_readpages
,
5022 .sync_page
= block_sync_page
,
5023 .direct_IO
= btrfs_direct_IO
,
5024 .invalidatepage
= btrfs_invalidatepage
,
5025 .releasepage
= btrfs_releasepage
,
5026 .set_page_dirty
= btrfs_set_page_dirty
,
5029 static struct address_space_operations btrfs_symlink_aops
= {
5030 .readpage
= btrfs_readpage
,
5031 .writepage
= btrfs_writepage
,
5032 .invalidatepage
= btrfs_invalidatepage
,
5033 .releasepage
= btrfs_releasepage
,
5036 static struct inode_operations btrfs_file_inode_operations
= {
5037 .truncate
= btrfs_truncate
,
5038 .getattr
= btrfs_getattr
,
5039 .setattr
= btrfs_setattr
,
5040 .setxattr
= btrfs_setxattr
,
5041 .getxattr
= btrfs_getxattr
,
5042 .listxattr
= btrfs_listxattr
,
5043 .removexattr
= btrfs_removexattr
,
5044 .permission
= btrfs_permission
,
5045 .fallocate
= btrfs_fallocate
,
5046 .fiemap
= btrfs_fiemap
,
5048 static struct inode_operations btrfs_special_inode_operations
= {
5049 .getattr
= btrfs_getattr
,
5050 .setattr
= btrfs_setattr
,
5051 .permission
= btrfs_permission
,
5052 .setxattr
= btrfs_setxattr
,
5053 .getxattr
= btrfs_getxattr
,
5054 .listxattr
= btrfs_listxattr
,
5055 .removexattr
= btrfs_removexattr
,
5057 static struct inode_operations btrfs_symlink_inode_operations
= {
5058 .readlink
= generic_readlink
,
5059 .follow_link
= page_follow_link_light
,
5060 .put_link
= page_put_link
,
5061 .permission
= btrfs_permission
,
5062 .setxattr
= btrfs_setxattr
,
5063 .getxattr
= btrfs_getxattr
,
5064 .listxattr
= btrfs_listxattr
,
5065 .removexattr
= btrfs_removexattr
,