2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include <linux/posix_acl.h>
40 #include <linux/falloc.h>
44 #include "transaction.h"
45 #include "btrfs_inode.h"
47 #include "print-tree.h"
49 #include "ordered-data.h"
52 #include "ref-cache.h"
53 #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
);
95 * a very lame attempt at stopping writes when the FS is 85% full. There
96 * are countless ways this is incorrect, but it is better than nothing.
98 int btrfs_check_free_space(struct btrfs_root
*root
, u64 num_required
,
107 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
108 total
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
109 used
= btrfs_super_bytes_used(&root
->fs_info
->super_copy
);
117 if (used
+ root
->fs_info
->delalloc_bytes
+ num_required
> thresh
)
119 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static int noinline
insert_inline_extent(struct btrfs_trans_handle
*trans
,
129 struct btrfs_root
*root
, struct inode
*inode
,
130 u64 start
, size_t size
, size_t compressed_size
,
131 struct page
**compressed_pages
)
133 struct btrfs_key key
;
134 struct btrfs_path
*path
;
135 struct extent_buffer
*leaf
;
136 struct page
*page
= NULL
;
139 struct btrfs_file_extent_item
*ei
;
142 size_t cur_size
= size
;
144 unsigned long offset
;
145 int use_compress
= 0;
147 if (compressed_size
&& compressed_pages
) {
149 cur_size
= compressed_size
;
152 path
= btrfs_alloc_path(); if (!path
)
155 btrfs_set_trans_block_group(trans
, inode
);
157 key
.objectid
= inode
->i_ino
;
159 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
160 inode_add_bytes(inode
, size
);
161 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
163 inode_add_bytes(inode
, size
);
164 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
169 printk("got bad ret %d\n", ret
);
172 leaf
= path
->nodes
[0];
173 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
174 struct btrfs_file_extent_item
);
175 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
176 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
177 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
178 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
179 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
180 ptr
= btrfs_file_extent_inline_start(ei
);
185 while(compressed_size
> 0) {
186 cpage
= compressed_pages
[i
];
187 cur_size
= min_t(unsigned long, compressed_size
,
191 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
196 compressed_size
-= cur_size
;
198 btrfs_set_file_extent_compression(leaf
, ei
,
199 BTRFS_COMPRESS_ZLIB
);
201 page
= find_get_page(inode
->i_mapping
,
202 start
>> PAGE_CACHE_SHIFT
);
203 btrfs_set_file_extent_compression(leaf
, ei
, 0);
204 kaddr
= kmap_atomic(page
, KM_USER0
);
205 offset
= start
& (PAGE_CACHE_SIZE
- 1);
206 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
207 kunmap_atomic(kaddr
, KM_USER0
);
208 page_cache_release(page
);
210 btrfs_mark_buffer_dirty(leaf
);
211 btrfs_free_path(path
);
213 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
214 btrfs_update_inode(trans
, root
, inode
);
217 btrfs_free_path(path
);
223 * conditionally insert an inline extent into the file. This
224 * does the checks required to make sure the data is small enough
225 * to fit as an inline extent.
227 static int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
228 struct btrfs_root
*root
,
229 struct inode
*inode
, u64 start
, u64 end
,
230 size_t compressed_size
,
231 struct page
**compressed_pages
)
233 u64 isize
= i_size_read(inode
);
234 u64 actual_end
= min(end
+ 1, isize
);
235 u64 inline_len
= actual_end
- start
;
236 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
237 ~((u64
)root
->sectorsize
- 1);
239 u64 data_len
= inline_len
;
243 data_len
= compressed_size
;
246 actual_end
>= PAGE_CACHE_SIZE
||
247 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
249 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
251 data_len
> root
->fs_info
->max_inline
) {
255 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
256 aligned_end
, start
, &hint_byte
);
259 if (isize
> actual_end
)
260 inline_len
= min_t(u64
, isize
, actual_end
);
261 ret
= insert_inline_extent(trans
, root
, inode
, start
,
262 inline_len
, compressed_size
,
265 btrfs_drop_extent_cache(inode
, start
, aligned_end
, 0);
269 struct async_extent
{
274 unsigned long nr_pages
;
275 struct list_head list
;
280 struct btrfs_root
*root
;
281 struct page
*locked_page
;
284 struct list_head extents
;
285 struct btrfs_work work
;
288 static noinline
int add_async_extent(struct async_cow
*cow
,
289 u64 start
, u64 ram_size
,
292 unsigned long nr_pages
)
294 struct async_extent
*async_extent
;
296 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
297 async_extent
->start
= start
;
298 async_extent
->ram_size
= ram_size
;
299 async_extent
->compressed_size
= compressed_size
;
300 async_extent
->pages
= pages
;
301 async_extent
->nr_pages
= nr_pages
;
302 list_add_tail(&async_extent
->list
, &cow
->extents
);
307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
322 static noinline
int compress_file_range(struct inode
*inode
,
323 struct page
*locked_page
,
325 struct async_cow
*async_cow
,
328 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
329 struct btrfs_trans_handle
*trans
;
333 u64 blocksize
= root
->sectorsize
;
336 struct page
**pages
= NULL
;
337 unsigned long nr_pages
;
338 unsigned long nr_pages_ret
= 0;
339 unsigned long total_compressed
= 0;
340 unsigned long total_in
= 0;
341 unsigned long max_compressed
= 128 * 1024;
342 unsigned long max_uncompressed
= 128 * 1024;
350 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
351 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
353 actual_end
= min_t(u64
, i_size_read(inode
), end
+ 1);
354 total_compressed
= actual_end
- start
;
356 /* we want to make sure that amount of ram required to uncompress
357 * an extent is reasonable, so we limit the total size in ram
358 * of a compressed extent to 128k. This is a crucial number
359 * because it also controls how easily we can spread reads across
360 * cpus for decompression.
362 * We also want to make sure the amount of IO required to do
363 * a random read is reasonably small, so we limit the size of
364 * a compressed extent to 128k.
366 total_compressed
= min(total_compressed
, max_uncompressed
);
367 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
368 num_bytes
= max(blocksize
, num_bytes
);
369 disk_num_bytes
= num_bytes
;
374 * we do compression for mount -o compress and when the
375 * inode has not been flagged as nocompress. This flag can
376 * change at any time if we discover bad compression ratios.
378 if (!btrfs_test_flag(inode
, NOCOMPRESS
) &&
379 btrfs_test_opt(root
, COMPRESS
)) {
381 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
383 ret
= btrfs_zlib_compress_pages(inode
->i_mapping
, start
,
384 total_compressed
, pages
,
385 nr_pages
, &nr_pages_ret
,
391 unsigned long offset
= total_compressed
&
392 (PAGE_CACHE_SIZE
- 1);
393 struct page
*page
= pages
[nr_pages_ret
- 1];
396 /* zero the tail end of the last page, we might be
397 * sending it down to disk
400 kaddr
= kmap_atomic(page
, KM_USER0
);
401 memset(kaddr
+ offset
, 0,
402 PAGE_CACHE_SIZE
- offset
);
403 kunmap_atomic(kaddr
, KM_USER0
);
409 trans
= btrfs_join_transaction(root
, 1);
411 btrfs_set_trans_block_group(trans
, inode
);
413 /* lets try to make an inline extent */
414 if (ret
|| total_in
< (actual_end
- start
)) {
415 /* we didn't compress the entire range, try
416 * to make an uncompressed inline extent.
418 ret
= cow_file_range_inline(trans
, root
, inode
,
419 start
, end
, 0, NULL
);
421 /* try making a compressed inline extent */
422 ret
= cow_file_range_inline(trans
, root
, inode
,
424 total_compressed
, pages
);
426 btrfs_end_transaction(trans
, root
);
429 * inline extent creation worked, we don't need
430 * to create any more async work items. Unlock
431 * and free up our temp pages.
433 extent_clear_unlock_delalloc(inode
,
434 &BTRFS_I(inode
)->io_tree
,
435 start
, end
, NULL
, 1, 0,
444 * we aren't doing an inline extent round the compressed size
445 * up to a block size boundary so the allocator does sane
448 total_compressed
= (total_compressed
+ blocksize
- 1) &
452 * one last check to make sure the compression is really a
453 * win, compare the page count read with the blocks on disk
455 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
456 ~(PAGE_CACHE_SIZE
- 1);
457 if (total_compressed
>= total_in
) {
460 disk_num_bytes
= total_compressed
;
461 num_bytes
= total_in
;
464 if (!will_compress
&& pages
) {
466 * the compression code ran but failed to make things smaller,
467 * free any pages it allocated and our page pointer array
469 for (i
= 0; i
< nr_pages_ret
; i
++) {
470 WARN_ON(pages
[i
]->mapping
);
471 page_cache_release(pages
[i
]);
475 total_compressed
= 0;
478 /* flag the file so we don't compress in the future */
479 btrfs_set_flag(inode
, NOCOMPRESS
);
484 /* the async work queues will take care of doing actual
485 * allocation on disk for these compressed pages,
486 * and will submit them to the elevator.
488 add_async_extent(async_cow
, start
, num_bytes
,
489 total_compressed
, pages
, nr_pages_ret
);
491 if (start
+ num_bytes
< end
) {
499 * No compression, but we still need to write the pages in
500 * the file we've been given so far. redirty the locked
501 * page if it corresponds to our extent and set things up
502 * for the async work queue to run cow_file_range to do
503 * the normal delalloc dance
505 if (page_offset(locked_page
) >= start
&&
506 page_offset(locked_page
) <= end
) {
507 __set_page_dirty_nobuffers(locked_page
);
508 /* unlocked later on in the async handlers */
510 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0);
518 for (i
= 0; i
< nr_pages_ret
; i
++) {
519 WARN_ON(pages
[i
]->mapping
);
520 page_cache_release(pages
[i
]);
529 * phase two of compressed writeback. This is the ordered portion
530 * of the code, which only gets called in the order the work was
531 * queued. We walk all the async extents created by compress_file_range
532 * and send them down to the disk.
534 static noinline
int submit_compressed_extents(struct inode
*inode
,
535 struct async_cow
*async_cow
)
537 struct async_extent
*async_extent
;
539 struct btrfs_trans_handle
*trans
;
540 struct btrfs_key ins
;
541 struct extent_map
*em
;
542 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
543 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
544 struct extent_io_tree
*io_tree
;
547 if (list_empty(&async_cow
->extents
))
550 trans
= btrfs_join_transaction(root
, 1);
552 while(!list_empty(&async_cow
->extents
)) {
553 async_extent
= list_entry(async_cow
->extents
.next
,
554 struct async_extent
, list
);
555 list_del(&async_extent
->list
);
557 io_tree
= &BTRFS_I(inode
)->io_tree
;
559 /* did the compression code fall back to uncompressed IO? */
560 if (!async_extent
->pages
) {
561 int page_started
= 0;
562 unsigned long nr_written
= 0;
564 lock_extent(io_tree
, async_extent
->start
,
565 async_extent
->start
+ async_extent
->ram_size
- 1,
568 /* allocate blocks */
569 cow_file_range(inode
, async_cow
->locked_page
,
571 async_extent
->start
+
572 async_extent
->ram_size
- 1,
573 &page_started
, &nr_written
, 0);
576 * if page_started, cow_file_range inserted an
577 * inline extent and took care of all the unlocking
578 * and IO for us. Otherwise, we need to submit
579 * all those pages down to the drive.
582 extent_write_locked_range(io_tree
,
583 inode
, async_extent
->start
,
584 async_extent
->start
+
585 async_extent
->ram_size
- 1,
593 lock_extent(io_tree
, async_extent
->start
,
594 async_extent
->start
+ async_extent
->ram_size
- 1,
597 * here we're doing allocation and writeback of the
600 btrfs_drop_extent_cache(inode
, async_extent
->start
,
601 async_extent
->start
+
602 async_extent
->ram_size
- 1, 0);
604 ret
= btrfs_reserve_extent(trans
, root
,
605 async_extent
->compressed_size
,
606 async_extent
->compressed_size
,
610 em
= alloc_extent_map(GFP_NOFS
);
611 em
->start
= async_extent
->start
;
612 em
->len
= async_extent
->ram_size
;
613 em
->orig_start
= em
->start
;
615 em
->block_start
= ins
.objectid
;
616 em
->block_len
= ins
.offset
;
617 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
618 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
619 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
622 spin_lock(&em_tree
->lock
);
623 ret
= add_extent_mapping(em_tree
, em
);
624 spin_unlock(&em_tree
->lock
);
625 if (ret
!= -EEXIST
) {
629 btrfs_drop_extent_cache(inode
, async_extent
->start
,
630 async_extent
->start
+
631 async_extent
->ram_size
- 1, 0);
634 ret
= btrfs_add_ordered_extent(inode
, async_extent
->start
,
636 async_extent
->ram_size
,
638 BTRFS_ORDERED_COMPRESSED
);
641 btrfs_end_transaction(trans
, root
);
644 * clear dirty, set writeback and unlock the pages.
646 extent_clear_unlock_delalloc(inode
,
647 &BTRFS_I(inode
)->io_tree
,
649 async_extent
->start
+
650 async_extent
->ram_size
- 1,
651 NULL
, 1, 1, 0, 1, 1, 0);
653 ret
= btrfs_submit_compressed_write(inode
,
655 async_extent
->ram_size
,
657 ins
.offset
, async_extent
->pages
,
658 async_extent
->nr_pages
);
661 trans
= btrfs_join_transaction(root
, 1);
662 alloc_hint
= ins
.objectid
+ ins
.offset
;
667 btrfs_end_transaction(trans
, root
);
672 * when extent_io.c finds a delayed allocation range in the file,
673 * the call backs end up in this code. The basic idea is to
674 * allocate extents on disk for the range, and create ordered data structs
675 * in ram to track those extents.
677 * locked_page is the page that writepage had locked already. We use
678 * it to make sure we don't do extra locks or unlocks.
680 * *page_started is set to one if we unlock locked_page and do everything
681 * required to start IO on it. It may be clean and already done with
684 static noinline
int cow_file_range(struct inode
*inode
,
685 struct page
*locked_page
,
686 u64 start
, u64 end
, int *page_started
,
687 unsigned long *nr_written
,
690 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
691 struct btrfs_trans_handle
*trans
;
694 unsigned long ram_size
;
697 u64 blocksize
= root
->sectorsize
;
699 struct btrfs_key ins
;
700 struct extent_map
*em
;
701 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
704 trans
= btrfs_join_transaction(root
, 1);
706 btrfs_set_trans_block_group(trans
, inode
);
708 actual_end
= min_t(u64
, i_size_read(inode
), end
+ 1);
710 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
711 num_bytes
= max(blocksize
, num_bytes
);
712 disk_num_bytes
= num_bytes
;
716 /* lets try to make an inline extent */
717 ret
= cow_file_range_inline(trans
, root
, inode
,
718 start
, end
, 0, NULL
);
720 extent_clear_unlock_delalloc(inode
,
721 &BTRFS_I(inode
)->io_tree
,
722 start
, end
, NULL
, 1, 1,
724 *nr_written
= *nr_written
+
725 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
732 BUG_ON(disk_num_bytes
>
733 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
735 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
737 while(disk_num_bytes
> 0) {
738 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
739 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
740 root
->sectorsize
, 0, alloc_hint
,
745 em
= alloc_extent_map(GFP_NOFS
);
747 em
->orig_start
= em
->start
;
749 ram_size
= ins
.offset
;
750 em
->len
= ins
.offset
;
752 em
->block_start
= ins
.objectid
;
753 em
->block_len
= ins
.offset
;
754 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
755 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
758 spin_lock(&em_tree
->lock
);
759 ret
= add_extent_mapping(em_tree
, em
);
760 spin_unlock(&em_tree
->lock
);
761 if (ret
!= -EEXIST
) {
765 btrfs_drop_extent_cache(inode
, start
,
766 start
+ ram_size
- 1, 0);
769 cur_alloc_size
= ins
.offset
;
770 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
771 ram_size
, cur_alloc_size
, 0);
774 if (disk_num_bytes
< cur_alloc_size
) {
775 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes
,
779 /* we're not doing compressed IO, don't unlock the first
780 * page (which the caller expects to stay locked), don't
781 * clear any dirty bits and don't set any writeback bits
783 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
784 start
, start
+ ram_size
- 1,
785 locked_page
, unlock
, 1,
787 disk_num_bytes
-= cur_alloc_size
;
788 num_bytes
-= cur_alloc_size
;
789 alloc_hint
= ins
.objectid
+ ins
.offset
;
790 start
+= cur_alloc_size
;
794 btrfs_end_transaction(trans
, root
);
800 * work queue call back to started compression on a file and pages
802 static noinline
void async_cow_start(struct btrfs_work
*work
)
804 struct async_cow
*async_cow
;
806 async_cow
= container_of(work
, struct async_cow
, work
);
808 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
809 async_cow
->start
, async_cow
->end
, async_cow
,
812 async_cow
->inode
= NULL
;
816 * work queue call back to submit previously compressed pages
818 static noinline
void async_cow_submit(struct btrfs_work
*work
)
820 struct async_cow
*async_cow
;
821 struct btrfs_root
*root
;
822 unsigned long nr_pages
;
824 async_cow
= container_of(work
, struct async_cow
, work
);
826 root
= async_cow
->root
;
827 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
830 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
832 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
834 waitqueue_active(&root
->fs_info
->async_submit_wait
))
835 wake_up(&root
->fs_info
->async_submit_wait
);
837 if (async_cow
->inode
) {
838 submit_compressed_extents(async_cow
->inode
, async_cow
);
842 static noinline
void async_cow_free(struct btrfs_work
*work
)
844 struct async_cow
*async_cow
;
845 async_cow
= container_of(work
, struct async_cow
, work
);
849 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
850 u64 start
, u64 end
, int *page_started
,
851 unsigned long *nr_written
)
853 struct async_cow
*async_cow
;
854 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
855 unsigned long nr_pages
;
857 int limit
= 10 * 1024 * 1042;
859 if (!btrfs_test_opt(root
, COMPRESS
)) {
860 return cow_file_range(inode
, locked_page
, start
, end
,
861 page_started
, nr_written
, 1);
864 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
|
865 EXTENT_DELALLOC
, 1, 0, GFP_NOFS
);
867 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
868 async_cow
->inode
= inode
;
869 async_cow
->root
= root
;
870 async_cow
->locked_page
= locked_page
;
871 async_cow
->start
= start
;
873 if (btrfs_test_flag(inode
, NOCOMPRESS
))
876 cur_end
= min(end
, start
+ 512 * 1024 - 1);
878 async_cow
->end
= cur_end
;
879 INIT_LIST_HEAD(&async_cow
->extents
);
881 async_cow
->work
.func
= async_cow_start
;
882 async_cow
->work
.ordered_func
= async_cow_submit
;
883 async_cow
->work
.ordered_free
= async_cow_free
;
884 async_cow
->work
.flags
= 0;
886 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
888 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
890 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
893 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
894 wait_event(root
->fs_info
->async_submit_wait
,
895 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
899 while(atomic_read(&root
->fs_info
->async_submit_draining
) &&
900 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
901 wait_event(root
->fs_info
->async_submit_wait
,
902 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
906 *nr_written
+= nr_pages
;
914 * when nowcow writeback call back. This checks for snapshots or COW copies
915 * of the extents that exist in the file, and COWs the file as required.
917 * If no cow copies or snapshots exist, we write directly to the existing
920 static int run_delalloc_nocow(struct inode
*inode
, struct page
*locked_page
,
921 u64 start
, u64 end
, int *page_started
, int force
,
922 unsigned long *nr_written
)
924 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
925 struct btrfs_trans_handle
*trans
;
926 struct extent_buffer
*leaf
;
927 struct btrfs_path
*path
;
928 struct btrfs_file_extent_item
*fi
;
929 struct btrfs_key found_key
;
941 path
= btrfs_alloc_path();
943 trans
= btrfs_join_transaction(root
, 1);
949 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
952 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
953 leaf
= path
->nodes
[0];
954 btrfs_item_key_to_cpu(leaf
, &found_key
,
956 if (found_key
.objectid
== inode
->i_ino
&&
957 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
962 leaf
= path
->nodes
[0];
963 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
964 ret
= btrfs_next_leaf(root
, path
);
969 leaf
= path
->nodes
[0];
974 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
976 if (found_key
.objectid
> inode
->i_ino
||
977 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
978 found_key
.offset
> end
)
981 if (found_key
.offset
> cur_offset
) {
982 extent_end
= found_key
.offset
;
986 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
987 struct btrfs_file_extent_item
);
988 extent_type
= btrfs_file_extent_type(leaf
, fi
);
990 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
991 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
992 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
993 extent_end
= found_key
.offset
+
994 btrfs_file_extent_num_bytes(leaf
, fi
);
995 if (extent_end
<= start
) {
999 if (btrfs_file_extent_compression(leaf
, fi
) ||
1000 btrfs_file_extent_encryption(leaf
, fi
) ||
1001 btrfs_file_extent_other_encoding(leaf
, fi
))
1003 if (disk_bytenr
== 0)
1005 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1007 if (btrfs_cross_ref_exist(trans
, root
, disk_bytenr
))
1009 if (btrfs_extent_readonly(root
, disk_bytenr
))
1011 disk_bytenr
+= btrfs_file_extent_offset(leaf
, fi
);
1013 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1014 extent_end
= found_key
.offset
+
1015 btrfs_file_extent_inline_len(leaf
, fi
);
1016 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1021 if (extent_end
<= start
) {
1026 if (cow_start
== (u64
)-1)
1027 cow_start
= cur_offset
;
1028 cur_offset
= extent_end
;
1029 if (cur_offset
> end
)
1035 btrfs_release_path(root
, path
);
1036 if (cow_start
!= (u64
)-1) {
1037 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1038 found_key
.offset
- 1, page_started
,
1041 cow_start
= (u64
)-1;
1044 disk_bytenr
+= cur_offset
- found_key
.offset
;
1045 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1046 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1047 struct extent_map
*em
;
1048 struct extent_map_tree
*em_tree
;
1049 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1050 em
= alloc_extent_map(GFP_NOFS
);
1051 em
->start
= cur_offset
;
1052 em
->orig_start
= em
->start
;
1053 em
->len
= num_bytes
;
1054 em
->block_len
= num_bytes
;
1055 em
->block_start
= disk_bytenr
;
1056 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1057 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1059 spin_lock(&em_tree
->lock
);
1060 ret
= add_extent_mapping(em_tree
, em
);
1061 spin_unlock(&em_tree
->lock
);
1062 if (ret
!= -EEXIST
) {
1063 free_extent_map(em
);
1066 btrfs_drop_extent_cache(inode
, em
->start
,
1067 em
->start
+ em
->len
- 1, 0);
1069 type
= BTRFS_ORDERED_PREALLOC
;
1071 type
= BTRFS_ORDERED_NOCOW
;
1074 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1075 num_bytes
, num_bytes
, type
);
1078 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1079 cur_offset
, cur_offset
+ num_bytes
- 1,
1080 locked_page
, 1, 1, 1, 0, 0, 0);
1081 cur_offset
= extent_end
;
1082 if (cur_offset
> end
)
1085 btrfs_release_path(root
, path
);
1087 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1088 cow_start
= cur_offset
;
1089 if (cow_start
!= (u64
)-1) {
1090 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1091 page_started
, nr_written
, 1);
1095 ret
= btrfs_end_transaction(trans
, root
);
1097 btrfs_free_path(path
);
1102 * extent_io.c call back to do delayed allocation processing
1104 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1105 u64 start
, u64 end
, int *page_started
,
1106 unsigned long *nr_written
)
1108 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1111 if (btrfs_test_opt(root
, NODATACOW
) ||
1112 btrfs_test_flag(inode
, NODATACOW
))
1113 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1114 page_started
, 1, nr_written
);
1115 else if (btrfs_test_flag(inode
, PREALLOC
))
1116 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1117 page_started
, 0, nr_written
);
1119 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1120 page_started
, nr_written
);
1126 * extent_io.c set_bit_hook, used to track delayed allocation
1127 * bytes in this file, and to maintain the list of inodes that
1128 * have pending delalloc work to be done.
1130 static int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1131 unsigned long old
, unsigned long bits
)
1133 unsigned long flags
;
1134 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1135 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1136 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
1137 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
1138 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
1139 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1140 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1141 &root
->fs_info
->delalloc_inodes
);
1143 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
1149 * extent_io.c clear_bit_hook, see set_bit_hook for why
1151 static int btrfs_clear_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1152 unsigned long old
, unsigned long bits
)
1154 if ((old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1155 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1156 unsigned long flags
;
1158 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
1159 if (end
- start
+ 1 > root
->fs_info
->delalloc_bytes
) {
1160 printk("warning: delalloc account %Lu %Lu\n",
1161 end
- start
+ 1, root
->fs_info
->delalloc_bytes
);
1162 root
->fs_info
->delalloc_bytes
= 0;
1163 BTRFS_I(inode
)->delalloc_bytes
= 0;
1165 root
->fs_info
->delalloc_bytes
-= end
- start
+ 1;
1166 BTRFS_I(inode
)->delalloc_bytes
-= end
- start
+ 1;
1168 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1169 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1170 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1172 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
1178 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1179 * we don't create bios that span stripes or chunks
1181 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1182 size_t size
, struct bio
*bio
,
1183 unsigned long bio_flags
)
1185 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1186 struct btrfs_mapping_tree
*map_tree
;
1187 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1192 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1195 length
= bio
->bi_size
;
1196 map_tree
= &root
->fs_info
->mapping_tree
;
1197 map_length
= length
;
1198 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1199 &map_length
, NULL
, 0);
1201 if (map_length
< length
+ size
) {
1208 * in order to insert checksums into the metadata in large chunks,
1209 * we wait until bio submission time. All the pages in the bio are
1210 * checksummed and sums are attached onto the ordered extent record.
1212 * At IO completion time the cums attached on the ordered extent record
1213 * are inserted into the btree
1215 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
, struct bio
*bio
,
1216 int mirror_num
, unsigned long bio_flags
)
1218 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1221 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1227 * in order to insert checksums into the metadata in large chunks,
1228 * we wait until bio submission time. All the pages in the bio are
1229 * checksummed and sums are attached onto the ordered extent record.
1231 * At IO completion time the cums attached on the ordered extent record
1232 * are inserted into the btree
1234 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1235 int mirror_num
, unsigned long bio_flags
)
1237 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1238 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1242 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1243 * or reading the csums from the tree before a read
1245 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1246 int mirror_num
, unsigned long bio_flags
)
1248 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1252 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1255 skip_sum
= btrfs_test_opt(root
, NODATASUM
) ||
1256 btrfs_test_flag(inode
, NODATASUM
);
1258 if (!(rw
& (1 << BIO_RW
))) {
1259 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1260 return btrfs_submit_compressed_read(inode
, bio
,
1261 mirror_num
, bio_flags
);
1262 } else if (!skip_sum
)
1263 btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1265 } else if (!skip_sum
) {
1266 /* we're doing a write, do the async checksumming */
1267 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1268 inode
, rw
, bio
, mirror_num
,
1269 bio_flags
, __btrfs_submit_bio_start
,
1270 __btrfs_submit_bio_done
);
1274 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1278 * given a list of ordered sums record them in the inode. This happens
1279 * at IO completion time based on sums calculated at bio submission time.
1281 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1282 struct inode
*inode
, u64 file_offset
,
1283 struct list_head
*list
)
1285 struct list_head
*cur
;
1286 struct btrfs_ordered_sum
*sum
;
1288 btrfs_set_trans_block_group(trans
, inode
);
1289 list_for_each(cur
, list
) {
1290 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
1291 btrfs_csum_file_blocks(trans
,
1292 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1297 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
1299 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0) {
1302 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1306 /* see btrfs_writepage_start_hook for details on why this is required */
1307 struct btrfs_writepage_fixup
{
1309 struct btrfs_work work
;
1312 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1314 struct btrfs_writepage_fixup
*fixup
;
1315 struct btrfs_ordered_extent
*ordered
;
1317 struct inode
*inode
;
1321 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1325 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1326 ClearPageChecked(page
);
1330 inode
= page
->mapping
->host
;
1331 page_start
= page_offset(page
);
1332 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1334 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1336 /* already ordered? We're done */
1337 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1338 EXTENT_ORDERED
, 0)) {
1342 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1344 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
1345 page_end
, GFP_NOFS
);
1347 btrfs_start_ordered_extent(inode
, ordered
, 1);
1351 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
1352 ClearPageChecked(page
);
1354 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1357 page_cache_release(page
);
1361 * There are a few paths in the higher layers of the kernel that directly
1362 * set the page dirty bit without asking the filesystem if it is a
1363 * good idea. This causes problems because we want to make sure COW
1364 * properly happens and the data=ordered rules are followed.
1366 * In our case any range that doesn't have the ORDERED bit set
1367 * hasn't been properly setup for IO. We kick off an async process
1368 * to fix it up. The async helper will wait for ordered extents, set
1369 * the delalloc bit and make it safe to write the page.
1371 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1373 struct inode
*inode
= page
->mapping
->host
;
1374 struct btrfs_writepage_fixup
*fixup
;
1375 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1378 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1383 if (PageChecked(page
))
1386 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1390 SetPageChecked(page
);
1391 page_cache_get(page
);
1392 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1394 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1398 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1399 struct inode
*inode
, u64 file_pos
,
1400 u64 disk_bytenr
, u64 disk_num_bytes
,
1401 u64 num_bytes
, u64 ram_bytes
,
1402 u8 compression
, u8 encryption
,
1403 u16 other_encoding
, int extent_type
)
1405 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1406 struct btrfs_file_extent_item
*fi
;
1407 struct btrfs_path
*path
;
1408 struct extent_buffer
*leaf
;
1409 struct btrfs_key ins
;
1413 path
= btrfs_alloc_path();
1416 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1417 file_pos
+ num_bytes
, file_pos
, &hint
);
1420 ins
.objectid
= inode
->i_ino
;
1421 ins
.offset
= file_pos
;
1422 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1423 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1425 leaf
= path
->nodes
[0];
1426 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1427 struct btrfs_file_extent_item
);
1428 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1429 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1430 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1431 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1432 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1433 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1434 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1435 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1436 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1437 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1438 btrfs_mark_buffer_dirty(leaf
);
1440 inode_add_bytes(inode
, num_bytes
);
1441 btrfs_drop_extent_cache(inode
, file_pos
, file_pos
+ num_bytes
- 1, 0);
1443 ins
.objectid
= disk_bytenr
;
1444 ins
.offset
= disk_num_bytes
;
1445 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1446 ret
= btrfs_alloc_reserved_extent(trans
, root
, leaf
->start
,
1447 root
->root_key
.objectid
,
1448 trans
->transid
, inode
->i_ino
, &ins
);
1451 btrfs_free_path(path
);
1455 /* as ordered data IO finishes, this gets called so we can finish
1456 * an ordered extent if the range of bytes in the file it covers are
1459 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1461 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1462 struct btrfs_trans_handle
*trans
;
1463 struct btrfs_ordered_extent
*ordered_extent
;
1464 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1468 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1472 trans
= btrfs_join_transaction(root
, 1);
1474 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1475 BUG_ON(!ordered_extent
);
1476 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1479 lock_extent(io_tree
, ordered_extent
->file_offset
,
1480 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1483 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1485 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1487 ret
= btrfs_mark_extent_written(trans
, root
, inode
,
1488 ordered_extent
->file_offset
,
1489 ordered_extent
->file_offset
+
1490 ordered_extent
->len
);
1493 ret
= insert_reserved_file_extent(trans
, inode
,
1494 ordered_extent
->file_offset
,
1495 ordered_extent
->start
,
1496 ordered_extent
->disk_len
,
1497 ordered_extent
->len
,
1498 ordered_extent
->len
,
1500 BTRFS_FILE_EXTENT_REG
);
1503 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1504 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1507 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1508 &ordered_extent
->list
);
1510 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1511 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1512 btrfs_update_inode(trans
, root
, inode
);
1513 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1514 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1517 btrfs_put_ordered_extent(ordered_extent
);
1518 /* once for the tree */
1519 btrfs_put_ordered_extent(ordered_extent
);
1521 btrfs_end_transaction(trans
, root
);
1525 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1526 struct extent_state
*state
, int uptodate
)
1528 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1532 * When IO fails, either with EIO or csum verification fails, we
1533 * try other mirrors that might have a good copy of the data. This
1534 * io_failure_record is used to record state as we go through all the
1535 * mirrors. If another mirror has good data, the page is set up to date
1536 * and things continue. If a good mirror can't be found, the original
1537 * bio end_io callback is called to indicate things have failed.
1539 struct io_failure_record
{
1544 unsigned long bio_flags
;
1548 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1549 struct page
*page
, u64 start
, u64 end
,
1550 struct extent_state
*state
)
1552 struct io_failure_record
*failrec
= NULL
;
1554 struct extent_map
*em
;
1555 struct inode
*inode
= page
->mapping
->host
;
1556 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1557 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1564 ret
= get_state_private(failure_tree
, start
, &private);
1566 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1569 failrec
->start
= start
;
1570 failrec
->len
= end
- start
+ 1;
1571 failrec
->last_mirror
= 0;
1572 failrec
->bio_flags
= 0;
1574 spin_lock(&em_tree
->lock
);
1575 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1576 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1577 free_extent_map(em
);
1580 spin_unlock(&em_tree
->lock
);
1582 if (!em
|| IS_ERR(em
)) {
1586 logical
= start
- em
->start
;
1587 logical
= em
->block_start
+ logical
;
1588 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1589 logical
= em
->block_start
;
1590 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1592 failrec
->logical
= logical
;
1593 free_extent_map(em
);
1594 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1595 EXTENT_DIRTY
, GFP_NOFS
);
1596 set_state_private(failure_tree
, start
,
1597 (u64
)(unsigned long)failrec
);
1599 failrec
= (struct io_failure_record
*)(unsigned long)private;
1601 num_copies
= btrfs_num_copies(
1602 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1603 failrec
->logical
, failrec
->len
);
1604 failrec
->last_mirror
++;
1606 spin_lock_irq(&BTRFS_I(inode
)->io_tree
.lock
);
1607 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1610 if (state
&& state
->start
!= failrec
->start
)
1612 spin_unlock_irq(&BTRFS_I(inode
)->io_tree
.lock
);
1614 if (!state
|| failrec
->last_mirror
> num_copies
) {
1615 set_state_private(failure_tree
, failrec
->start
, 0);
1616 clear_extent_bits(failure_tree
, failrec
->start
,
1617 failrec
->start
+ failrec
->len
- 1,
1618 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1622 bio
= bio_alloc(GFP_NOFS
, 1);
1623 bio
->bi_private
= state
;
1624 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1625 bio
->bi_sector
= failrec
->logical
>> 9;
1626 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1629 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1630 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1635 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1636 failrec
->last_mirror
,
1637 failrec
->bio_flags
);
1642 * each time an IO finishes, we do a fast check in the IO failure tree
1643 * to see if we need to process or clean up an io_failure_record
1645 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1648 u64 private_failure
;
1649 struct io_failure_record
*failure
;
1653 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1654 (u64
)-1, 1, EXTENT_DIRTY
)) {
1655 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1656 start
, &private_failure
);
1658 failure
= (struct io_failure_record
*)(unsigned long)
1660 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1662 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1664 failure
->start
+ failure
->len
- 1,
1665 EXTENT_DIRTY
| EXTENT_LOCKED
,
1674 * when reads are done, we need to check csums to verify the data is correct
1675 * if there's a match, we allow the bio to finish. If not, we go through
1676 * the io_failure_record routines to find good copies
1678 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1679 struct extent_state
*state
)
1681 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1682 struct inode
*inode
= page
->mapping
->host
;
1683 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1685 u64
private = ~(u32
)0;
1687 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1689 unsigned long flags
;
1691 if (PageChecked(page
)) {
1692 ClearPageChecked(page
);
1695 if (btrfs_test_opt(root
, NODATASUM
) ||
1696 btrfs_test_flag(inode
, NODATASUM
))
1699 if (state
&& state
->start
== start
) {
1700 private = state
->private;
1703 ret
= get_state_private(io_tree
, start
, &private);
1705 local_irq_save(flags
);
1706 kaddr
= kmap_atomic(page
, KM_IRQ0
);
1710 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1711 btrfs_csum_final(csum
, (char *)&csum
);
1712 if (csum
!= private) {
1715 kunmap_atomic(kaddr
, KM_IRQ0
);
1716 local_irq_restore(flags
);
1718 /* if the io failure tree for this inode is non-empty,
1719 * check to see if we've recovered from a failed IO
1721 btrfs_clean_io_failures(inode
, start
);
1725 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1726 page
->mapping
->host
->i_ino
, (unsigned long long)start
, csum
,
1728 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1729 flush_dcache_page(page
);
1730 kunmap_atomic(kaddr
, KM_IRQ0
);
1731 local_irq_restore(flags
);
1738 * This creates an orphan entry for the given inode in case something goes
1739 * wrong in the middle of an unlink/truncate.
1741 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1743 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1746 spin_lock(&root
->list_lock
);
1748 /* already on the orphan list, we're good */
1749 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1750 spin_unlock(&root
->list_lock
);
1754 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1756 spin_unlock(&root
->list_lock
);
1759 * insert an orphan item to track this unlinked/truncated file
1761 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1767 * We have done the truncate/delete so we can go ahead and remove the orphan
1768 * item for this particular inode.
1770 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1772 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1775 spin_lock(&root
->list_lock
);
1777 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1778 spin_unlock(&root
->list_lock
);
1782 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1784 spin_unlock(&root
->list_lock
);
1788 spin_unlock(&root
->list_lock
);
1790 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1796 * this cleans up any orphans that may be left on the list from the last use
1799 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1801 struct btrfs_path
*path
;
1802 struct extent_buffer
*leaf
;
1803 struct btrfs_item
*item
;
1804 struct btrfs_key key
, found_key
;
1805 struct btrfs_trans_handle
*trans
;
1806 struct inode
*inode
;
1807 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1809 path
= btrfs_alloc_path();
1814 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1815 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1816 key
.offset
= (u64
)-1;
1820 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1822 printk(KERN_ERR
"Error searching slot for orphan: %d"
1828 * if ret == 0 means we found what we were searching for, which
1829 * is weird, but possible, so only screw with path if we didnt
1830 * find the key and see if we have stuff that matches
1833 if (path
->slots
[0] == 0)
1838 /* pull out the item */
1839 leaf
= path
->nodes
[0];
1840 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1841 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1843 /* make sure the item matches what we want */
1844 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1846 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1849 /* release the path since we're done with it */
1850 btrfs_release_path(root
, path
);
1853 * this is where we are basically btrfs_lookup, without the
1854 * crossing root thing. we store the inode number in the
1855 * offset of the orphan item.
1857 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
1858 found_key
.offset
, root
);
1862 if (inode
->i_state
& I_NEW
) {
1863 BTRFS_I(inode
)->root
= root
;
1865 /* have to set the location manually */
1866 BTRFS_I(inode
)->location
.objectid
= inode
->i_ino
;
1867 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
1868 BTRFS_I(inode
)->location
.offset
= 0;
1870 btrfs_read_locked_inode(inode
);
1871 unlock_new_inode(inode
);
1875 * add this inode to the orphan list so btrfs_orphan_del does
1876 * the proper thing when we hit it
1878 spin_lock(&root
->list_lock
);
1879 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1880 spin_unlock(&root
->list_lock
);
1883 * if this is a bad inode, means we actually succeeded in
1884 * removing the inode, but not the orphan record, which means
1885 * we need to manually delete the orphan since iput will just
1886 * do a destroy_inode
1888 if (is_bad_inode(inode
)) {
1889 trans
= btrfs_start_transaction(root
, 1);
1890 btrfs_orphan_del(trans
, inode
);
1891 btrfs_end_transaction(trans
, root
);
1896 /* if we have links, this was a truncate, lets do that */
1897 if (inode
->i_nlink
) {
1899 btrfs_truncate(inode
);
1904 /* this will do delete_inode and everything for us */
1909 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
1911 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
1913 btrfs_free_path(path
);
1917 * read an inode from the btree into the in-memory inode
1919 void btrfs_read_locked_inode(struct inode
*inode
)
1921 struct btrfs_path
*path
;
1922 struct extent_buffer
*leaf
;
1923 struct btrfs_inode_item
*inode_item
;
1924 struct btrfs_timespec
*tspec
;
1925 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1926 struct btrfs_key location
;
1927 u64 alloc_group_block
;
1931 path
= btrfs_alloc_path();
1933 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
1935 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
1939 leaf
= path
->nodes
[0];
1940 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1941 struct btrfs_inode_item
);
1943 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
1944 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
1945 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
1946 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
1947 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
1949 tspec
= btrfs_inode_atime(inode_item
);
1950 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1951 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1953 tspec
= btrfs_inode_mtime(inode_item
);
1954 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1955 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1957 tspec
= btrfs_inode_ctime(inode_item
);
1958 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1959 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1961 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
1962 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
1963 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
1964 inode
->i_generation
= BTRFS_I(inode
)->generation
;
1966 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
1968 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1969 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
1971 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
1972 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
1973 alloc_group_block
, 0);
1974 btrfs_free_path(path
);
1977 switch (inode
->i_mode
& S_IFMT
) {
1979 inode
->i_mapping
->a_ops
= &btrfs_aops
;
1980 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
1981 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
1982 inode
->i_fop
= &btrfs_file_operations
;
1983 inode
->i_op
= &btrfs_file_inode_operations
;
1986 inode
->i_fop
= &btrfs_dir_file_operations
;
1987 if (root
== root
->fs_info
->tree_root
)
1988 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
1990 inode
->i_op
= &btrfs_dir_inode_operations
;
1993 inode
->i_op
= &btrfs_symlink_inode_operations
;
1994 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
1995 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
1998 init_special_inode(inode
, inode
->i_mode
, rdev
);
2004 btrfs_free_path(path
);
2005 make_bad_inode(inode
);
2009 * given a leaf and an inode, copy the inode fields into the leaf
2011 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2012 struct extent_buffer
*leaf
,
2013 struct btrfs_inode_item
*item
,
2014 struct inode
*inode
)
2016 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2017 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2018 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2019 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2020 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2022 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2023 inode
->i_atime
.tv_sec
);
2024 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2025 inode
->i_atime
.tv_nsec
);
2027 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2028 inode
->i_mtime
.tv_sec
);
2029 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2030 inode
->i_mtime
.tv_nsec
);
2032 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2033 inode
->i_ctime
.tv_sec
);
2034 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2035 inode
->i_ctime
.tv_nsec
);
2037 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2038 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2039 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2040 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2041 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2042 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2043 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2047 * copy everything in the in-memory inode into the btree.
2049 int noinline
btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2050 struct btrfs_root
*root
,
2051 struct inode
*inode
)
2053 struct btrfs_inode_item
*inode_item
;
2054 struct btrfs_path
*path
;
2055 struct extent_buffer
*leaf
;
2058 path
= btrfs_alloc_path();
2060 ret
= btrfs_lookup_inode(trans
, root
, path
,
2061 &BTRFS_I(inode
)->location
, 1);
2068 leaf
= path
->nodes
[0];
2069 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2070 struct btrfs_inode_item
);
2072 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2073 btrfs_mark_buffer_dirty(leaf
);
2074 btrfs_set_inode_last_trans(trans
, inode
);
2077 btrfs_free_path(path
);
2083 * unlink helper that gets used here in inode.c and in the tree logging
2084 * recovery code. It remove a link in a directory with a given name, and
2085 * also drops the back refs in the inode to the directory
2087 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2088 struct btrfs_root
*root
,
2089 struct inode
*dir
, struct inode
*inode
,
2090 const char *name
, int name_len
)
2092 struct btrfs_path
*path
;
2094 struct extent_buffer
*leaf
;
2095 struct btrfs_dir_item
*di
;
2096 struct btrfs_key key
;
2099 path
= btrfs_alloc_path();
2105 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2106 name
, name_len
, -1);
2115 leaf
= path
->nodes
[0];
2116 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2117 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2120 btrfs_release_path(root
, path
);
2122 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2124 dir
->i_ino
, &index
);
2126 printk("failed to delete reference to %.*s, "
2127 "inode %lu parent %lu\n", name_len
, name
,
2128 inode
->i_ino
, dir
->i_ino
);
2132 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2133 index
, name
, name_len
, -1);
2142 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2143 btrfs_release_path(root
, path
);
2145 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2147 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2149 BTRFS_I(dir
)->log_dirty_trans
= trans
->transid
;
2151 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2155 btrfs_free_path(path
);
2159 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2160 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2161 btrfs_update_inode(trans
, root
, dir
);
2162 btrfs_drop_nlink(inode
);
2163 ret
= btrfs_update_inode(trans
, root
, inode
);
2164 dir
->i_sb
->s_dirt
= 1;
2169 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2171 struct btrfs_root
*root
;
2172 struct btrfs_trans_handle
*trans
;
2173 struct inode
*inode
= dentry
->d_inode
;
2175 unsigned long nr
= 0;
2177 root
= BTRFS_I(dir
)->root
;
2179 ret
= btrfs_check_free_space(root
, 1, 1);
2183 trans
= btrfs_start_transaction(root
, 1);
2185 btrfs_set_trans_block_group(trans
, dir
);
2186 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2187 dentry
->d_name
.name
, dentry
->d_name
.len
);
2189 if (inode
->i_nlink
== 0)
2190 ret
= btrfs_orphan_add(trans
, inode
);
2192 nr
= trans
->blocks_used
;
2194 btrfs_end_transaction_throttle(trans
, root
);
2196 btrfs_btree_balance_dirty(root
, nr
);
2200 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2202 struct inode
*inode
= dentry
->d_inode
;
2205 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2206 struct btrfs_trans_handle
*trans
;
2207 unsigned long nr
= 0;
2210 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2211 * the root of a subvolume or snapshot
2213 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2214 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
2218 ret
= btrfs_check_free_space(root
, 1, 1);
2222 trans
= btrfs_start_transaction(root
, 1);
2223 btrfs_set_trans_block_group(trans
, dir
);
2225 err
= btrfs_orphan_add(trans
, inode
);
2229 /* now the directory is empty */
2230 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2231 dentry
->d_name
.name
, dentry
->d_name
.len
);
2233 btrfs_i_size_write(inode
, 0);
2237 nr
= trans
->blocks_used
;
2238 ret
= btrfs_end_transaction_throttle(trans
, root
);
2240 btrfs_btree_balance_dirty(root
, nr
);
2249 * when truncating bytes in a file, it is possible to avoid reading
2250 * the leaves that contain only checksum items. This can be the
2251 * majority of the IO required to delete a large file, but it must
2252 * be done carefully.
2254 * The keys in the level just above the leaves are checked to make sure
2255 * the lowest key in a given leaf is a csum key, and starts at an offset
2256 * after the new size.
2258 * Then the key for the next leaf is checked to make sure it also has
2259 * a checksum item for the same file. If it does, we know our target leaf
2260 * contains only checksum items, and it can be safely freed without reading
2263 * This is just an optimization targeted at large files. It may do
2264 * nothing. It will return 0 unless things went badly.
2266 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
2267 struct btrfs_root
*root
,
2268 struct btrfs_path
*path
,
2269 struct inode
*inode
, u64 new_size
)
2271 struct btrfs_key key
;
2274 struct btrfs_key found_key
;
2275 struct btrfs_key other_key
;
2276 struct btrfs_leaf_ref
*ref
;
2280 path
->lowest_level
= 1;
2281 key
.objectid
= inode
->i_ino
;
2282 key
.type
= BTRFS_CSUM_ITEM_KEY
;
2283 key
.offset
= new_size
;
2285 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2289 if (path
->nodes
[1] == NULL
) {
2294 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
2295 nritems
= btrfs_header_nritems(path
->nodes
[1]);
2300 if (path
->slots
[1] >= nritems
)
2303 /* did we find a key greater than anything we want to delete? */
2304 if (found_key
.objectid
> inode
->i_ino
||
2305 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
2308 /* we check the next key in the node to make sure the leave contains
2309 * only checksum items. This comparison doesn't work if our
2310 * leaf is the last one in the node
2312 if (path
->slots
[1] + 1 >= nritems
) {
2314 /* search forward from the last key in the node, this
2315 * will bring us into the next node in the tree
2317 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
2319 /* unlikely, but we inc below, so check to be safe */
2320 if (found_key
.offset
== (u64
)-1)
2323 /* search_forward needs a path with locks held, do the
2324 * search again for the original key. It is possible
2325 * this will race with a balance and return a path that
2326 * we could modify, but this drop is just an optimization
2327 * and is allowed to miss some leaves.
2329 btrfs_release_path(root
, path
);
2332 /* setup a max key for search_forward */
2333 other_key
.offset
= (u64
)-1;
2334 other_key
.type
= key
.type
;
2335 other_key
.objectid
= key
.objectid
;
2337 path
->keep_locks
= 1;
2338 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
2340 path
->keep_locks
= 0;
2341 if (ret
|| found_key
.objectid
!= key
.objectid
||
2342 found_key
.type
!= key
.type
) {
2347 key
.offset
= found_key
.offset
;
2348 btrfs_release_path(root
, path
);
2353 /* we know there's one more slot after us in the tree,
2354 * read that key so we can verify it is also a checksum item
2356 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
2358 if (found_key
.objectid
< inode
->i_ino
)
2361 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
2365 * if the key for the next leaf isn't a csum key from this objectid,
2366 * we can't be sure there aren't good items inside this leaf.
2369 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
2372 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
2373 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
2375 * it is safe to delete this leaf, it contains only
2376 * csum items from this inode at an offset >= new_size
2378 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
2381 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
2382 ref
= btrfs_alloc_leaf_ref(root
, 0);
2384 ref
->root_gen
= root
->root_key
.offset
;
2385 ref
->bytenr
= leaf_start
;
2387 ref
->generation
= leaf_gen
;
2390 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
2392 btrfs_free_leaf_ref(root
, ref
);
2398 btrfs_release_path(root
, path
);
2400 if (other_key
.objectid
== inode
->i_ino
&&
2401 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
2402 key
.offset
= other_key
.offset
;
2408 /* fixup any changes we've made to the path */
2409 path
->lowest_level
= 0;
2410 path
->keep_locks
= 0;
2411 btrfs_release_path(root
, path
);
2418 * this can truncate away extent items, csum items and directory items.
2419 * It starts at a high offset and removes keys until it can't find
2420 * any higher than new_size
2422 * csum items that cross the new i_size are truncated to the new size
2425 * min_type is the minimum key type to truncate down to. If set to 0, this
2426 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2428 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2429 struct btrfs_root
*root
,
2430 struct inode
*inode
,
2431 u64 new_size
, u32 min_type
)
2434 struct btrfs_path
*path
;
2435 struct btrfs_key key
;
2436 struct btrfs_key found_key
;
2438 struct extent_buffer
*leaf
;
2439 struct btrfs_file_extent_item
*fi
;
2440 u64 extent_start
= 0;
2441 u64 extent_num_bytes
= 0;
2447 int pending_del_nr
= 0;
2448 int pending_del_slot
= 0;
2449 int extent_type
= -1;
2451 u64 mask
= root
->sectorsize
- 1;
2454 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2455 path
= btrfs_alloc_path();
2459 /* FIXME, add redo link to tree so we don't leak on crash */
2460 key
.objectid
= inode
->i_ino
;
2461 key
.offset
= (u64
)-1;
2464 btrfs_init_path(path
);
2467 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2472 /* there are no items in the tree for us to truncate, we're
2475 if (path
->slots
[0] == 0) {
2484 leaf
= path
->nodes
[0];
2485 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2486 found_type
= btrfs_key_type(&found_key
);
2489 if (found_key
.objectid
!= inode
->i_ino
)
2492 if (found_type
< min_type
)
2495 item_end
= found_key
.offset
;
2496 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2497 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2498 struct btrfs_file_extent_item
);
2499 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2500 encoding
= btrfs_file_extent_compression(leaf
, fi
);
2501 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
2502 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
2504 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2506 btrfs_file_extent_num_bytes(leaf
, fi
);
2507 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2508 item_end
+= btrfs_file_extent_inline_len(leaf
,
2513 if (item_end
< new_size
) {
2514 if (found_type
== BTRFS_DIR_ITEM_KEY
) {
2515 found_type
= BTRFS_INODE_ITEM_KEY
;
2516 } else if (found_type
== BTRFS_EXTENT_ITEM_KEY
) {
2517 found_type
= BTRFS_EXTENT_DATA_KEY
;
2518 } else if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2519 found_type
= BTRFS_XATTR_ITEM_KEY
;
2520 } else if (found_type
== BTRFS_XATTR_ITEM_KEY
) {
2521 found_type
= BTRFS_INODE_REF_KEY
;
2522 } else if (found_type
) {
2527 btrfs_set_key_type(&key
, found_type
);
2530 if (found_key
.offset
>= new_size
)
2536 /* FIXME, shrink the extent if the ref count is only 1 */
2537 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2540 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2542 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2543 if (!del_item
&& !encoding
) {
2544 u64 orig_num_bytes
=
2545 btrfs_file_extent_num_bytes(leaf
, fi
);
2546 extent_num_bytes
= new_size
-
2547 found_key
.offset
+ root
->sectorsize
- 1;
2548 extent_num_bytes
= extent_num_bytes
&
2549 ~((u64
)root
->sectorsize
- 1);
2550 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2552 num_dec
= (orig_num_bytes
-
2554 if (root
->ref_cows
&& extent_start
!= 0)
2555 inode_sub_bytes(inode
, num_dec
);
2556 btrfs_mark_buffer_dirty(leaf
);
2559 btrfs_file_extent_disk_num_bytes(leaf
,
2561 /* FIXME blocksize != 4096 */
2562 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2563 if (extent_start
!= 0) {
2566 inode_sub_bytes(inode
, num_dec
);
2568 root_gen
= btrfs_header_generation(leaf
);
2569 root_owner
= btrfs_header_owner(leaf
);
2571 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2573 * we can't truncate inline items that have had
2577 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2578 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2579 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2580 u32 size
= new_size
- found_key
.offset
;
2582 if (root
->ref_cows
) {
2583 inode_sub_bytes(inode
, item_end
+ 1 -
2587 btrfs_file_extent_calc_inline_size(size
);
2588 ret
= btrfs_truncate_item(trans
, root
, path
,
2591 } else if (root
->ref_cows
) {
2592 inode_sub_bytes(inode
, item_end
+ 1 -
2598 if (!pending_del_nr
) {
2599 /* no pending yet, add ourselves */
2600 pending_del_slot
= path
->slots
[0];
2602 } else if (pending_del_nr
&&
2603 path
->slots
[0] + 1 == pending_del_slot
) {
2604 /* hop on the pending chunk */
2606 pending_del_slot
= path
->slots
[0];
2608 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path
->slots
[0], pending_del_nr
, pending_del_slot
);
2614 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2616 leaf
->start
, root_owner
,
2617 root_gen
, inode
->i_ino
, 0);
2621 if (path
->slots
[0] == 0) {
2624 btrfs_release_path(root
, path
);
2629 if (pending_del_nr
&&
2630 path
->slots
[0] + 1 != pending_del_slot
) {
2631 struct btrfs_key debug
;
2633 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2635 ret
= btrfs_del_items(trans
, root
, path
,
2640 btrfs_release_path(root
, path
);
2646 if (pending_del_nr
) {
2647 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2650 btrfs_free_path(path
);
2651 inode
->i_sb
->s_dirt
= 1;
2656 * taken from block_truncate_page, but does cow as it zeros out
2657 * any bytes left in the last page in the file.
2659 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2661 struct inode
*inode
= mapping
->host
;
2662 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2663 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2664 struct btrfs_ordered_extent
*ordered
;
2666 u32 blocksize
= root
->sectorsize
;
2667 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2668 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2674 if ((offset
& (blocksize
- 1)) == 0)
2679 page
= grab_cache_page(mapping
, index
);
2683 page_start
= page_offset(page
);
2684 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2686 if (!PageUptodate(page
)) {
2687 ret
= btrfs_readpage(NULL
, page
);
2689 if (page
->mapping
!= mapping
) {
2691 page_cache_release(page
);
2694 if (!PageUptodate(page
)) {
2699 wait_on_page_writeback(page
);
2701 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2702 set_page_extent_mapped(page
);
2704 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2706 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2708 page_cache_release(page
);
2709 btrfs_start_ordered_extent(inode
, ordered
, 1);
2710 btrfs_put_ordered_extent(ordered
);
2714 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2716 if (offset
!= PAGE_CACHE_SIZE
) {
2718 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2719 flush_dcache_page(page
);
2722 ClearPageChecked(page
);
2723 set_page_dirty(page
);
2724 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2728 page_cache_release(page
);
2733 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
2735 struct btrfs_trans_handle
*trans
;
2736 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2737 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2738 struct extent_map
*em
;
2739 u64 mask
= root
->sectorsize
- 1;
2740 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2741 u64 block_end
= (size
+ mask
) & ~mask
;
2747 if (size
<= hole_start
)
2750 err
= btrfs_check_free_space(root
, 1, 0);
2754 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2757 struct btrfs_ordered_extent
*ordered
;
2758 btrfs_wait_ordered_range(inode
, hole_start
,
2759 block_end
- hole_start
);
2760 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2761 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2764 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2765 btrfs_put_ordered_extent(ordered
);
2768 trans
= btrfs_start_transaction(root
, 1);
2769 btrfs_set_trans_block_group(trans
, inode
);
2771 cur_offset
= hole_start
;
2773 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2774 block_end
- cur_offset
, 0);
2775 BUG_ON(IS_ERR(em
) || !em
);
2776 last_byte
= min(extent_map_end(em
), block_end
);
2777 last_byte
= (last_byte
+ mask
) & ~mask
;
2778 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2780 hole_size
= last_byte
- cur_offset
;
2781 err
= btrfs_drop_extents(trans
, root
, inode
,
2783 cur_offset
+ hole_size
,
2784 cur_offset
, &hint_byte
);
2787 err
= btrfs_insert_file_extent(trans
, root
,
2788 inode
->i_ino
, cur_offset
, 0,
2789 0, hole_size
, 0, hole_size
,
2791 btrfs_drop_extent_cache(inode
, hole_start
,
2794 free_extent_map(em
);
2795 cur_offset
= last_byte
;
2796 if (err
|| cur_offset
>= block_end
)
2800 btrfs_end_transaction(trans
, root
);
2801 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2805 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2807 struct inode
*inode
= dentry
->d_inode
;
2810 err
= inode_change_ok(inode
, attr
);
2814 if (S_ISREG(inode
->i_mode
) &&
2815 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
> inode
->i_size
) {
2816 err
= btrfs_cont_expand(inode
, attr
->ia_size
);
2821 err
= inode_setattr(inode
, attr
);
2823 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2824 err
= btrfs_acl_chmod(inode
);
2828 void btrfs_delete_inode(struct inode
*inode
)
2830 struct btrfs_trans_handle
*trans
;
2831 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2835 truncate_inode_pages(&inode
->i_data
, 0);
2836 if (is_bad_inode(inode
)) {
2837 btrfs_orphan_del(NULL
, inode
);
2840 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
2842 btrfs_i_size_write(inode
, 0);
2843 trans
= btrfs_start_transaction(root
, 1);
2845 btrfs_set_trans_block_group(trans
, inode
);
2846 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
2848 btrfs_orphan_del(NULL
, inode
);
2849 goto no_delete_lock
;
2852 btrfs_orphan_del(trans
, inode
);
2854 nr
= trans
->blocks_used
;
2857 btrfs_end_transaction(trans
, root
);
2858 btrfs_btree_balance_dirty(root
, nr
);
2862 nr
= trans
->blocks_used
;
2863 btrfs_end_transaction(trans
, root
);
2864 btrfs_btree_balance_dirty(root
, nr
);
2870 * this returns the key found in the dir entry in the location pointer.
2871 * If no dir entries were found, location->objectid is 0.
2873 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
2874 struct btrfs_key
*location
)
2876 const char *name
= dentry
->d_name
.name
;
2877 int namelen
= dentry
->d_name
.len
;
2878 struct btrfs_dir_item
*di
;
2879 struct btrfs_path
*path
;
2880 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2883 path
= btrfs_alloc_path();
2886 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
2890 if (!di
|| IS_ERR(di
)) {
2893 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
2895 btrfs_free_path(path
);
2898 location
->objectid
= 0;
2903 * when we hit a tree root in a directory, the btrfs part of the inode
2904 * needs to be changed to reflect the root directory of the tree root. This
2905 * is kind of like crossing a mount point.
2907 static int fixup_tree_root_location(struct btrfs_root
*root
,
2908 struct btrfs_key
*location
,
2909 struct btrfs_root
**sub_root
,
2910 struct dentry
*dentry
)
2912 struct btrfs_root_item
*ri
;
2914 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
2916 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
2919 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
2920 dentry
->d_name
.name
,
2921 dentry
->d_name
.len
);
2922 if (IS_ERR(*sub_root
))
2923 return PTR_ERR(*sub_root
);
2925 ri
= &(*sub_root
)->root_item
;
2926 location
->objectid
= btrfs_root_dirid(ri
);
2927 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2928 location
->offset
= 0;
2933 static noinline
void init_btrfs_i(struct inode
*inode
)
2935 struct btrfs_inode
*bi
= BTRFS_I(inode
);
2938 bi
->i_default_acl
= NULL
;
2943 bi
->logged_trans
= 0;
2944 bi
->delalloc_bytes
= 0;
2945 bi
->disk_i_size
= 0;
2947 bi
->index_cnt
= (u64
)-1;
2948 bi
->log_dirty_trans
= 0;
2949 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
2950 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
2951 inode
->i_mapping
, GFP_NOFS
);
2952 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
2953 inode
->i_mapping
, GFP_NOFS
);
2954 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
2955 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
2956 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
2957 mutex_init(&BTRFS_I(inode
)->log_mutex
);
2960 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
2962 struct btrfs_iget_args
*args
= p
;
2963 inode
->i_ino
= args
->ino
;
2964 init_btrfs_i(inode
);
2965 BTRFS_I(inode
)->root
= args
->root
;
2969 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
2971 struct btrfs_iget_args
*args
= opaque
;
2972 return (args
->ino
== inode
->i_ino
&&
2973 args
->root
== BTRFS_I(inode
)->root
);
2976 struct inode
*btrfs_ilookup(struct super_block
*s
, u64 objectid
,
2977 struct btrfs_root
*root
, int wait
)
2979 struct inode
*inode
;
2980 struct btrfs_iget_args args
;
2981 args
.ino
= objectid
;
2985 inode
= ilookup5(s
, objectid
, btrfs_find_actor
,
2988 inode
= ilookup5_nowait(s
, objectid
, btrfs_find_actor
,
2994 struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 objectid
,
2995 struct btrfs_root
*root
)
2997 struct inode
*inode
;
2998 struct btrfs_iget_args args
;
2999 args
.ino
= objectid
;
3002 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3003 btrfs_init_locked_inode
,
3008 /* Get an inode object given its location and corresponding root.
3009 * Returns in *is_new if the inode was read from disk
3011 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3012 struct btrfs_root
*root
, int *is_new
)
3014 struct inode
*inode
;
3016 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3018 return ERR_PTR(-EACCES
);
3020 if (inode
->i_state
& I_NEW
) {
3021 BTRFS_I(inode
)->root
= root
;
3022 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3023 btrfs_read_locked_inode(inode
);
3024 unlock_new_inode(inode
);
3035 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3037 struct inode
* inode
;
3038 struct btrfs_inode
*bi
= BTRFS_I(dir
);
3039 struct btrfs_root
*root
= bi
->root
;
3040 struct btrfs_root
*sub_root
= root
;
3041 struct btrfs_key location
;
3044 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3045 return ERR_PTR(-ENAMETOOLONG
);
3047 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3050 return ERR_PTR(ret
);
3053 if (location
.objectid
) {
3054 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
3057 return ERR_PTR(ret
);
3059 return ERR_PTR(-ENOENT
);
3060 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, &new);
3062 return ERR_CAST(inode
);
3067 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
3068 struct nameidata
*nd
)
3070 struct inode
*inode
;
3072 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3073 return ERR_PTR(-ENAMETOOLONG
);
3075 inode
= btrfs_lookup_dentry(dir
, dentry
);
3077 return ERR_CAST(inode
);
3079 return d_splice_alias(inode
, dentry
);
3082 static unsigned char btrfs_filetype_table
[] = {
3083 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
3086 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
3089 struct inode
*inode
= filp
->f_dentry
->d_inode
;
3090 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3091 struct btrfs_item
*item
;
3092 struct btrfs_dir_item
*di
;
3093 struct btrfs_key key
;
3094 struct btrfs_key found_key
;
3095 struct btrfs_path
*path
;
3098 struct extent_buffer
*leaf
;
3101 unsigned char d_type
;
3106 int key_type
= BTRFS_DIR_INDEX_KEY
;
3111 /* FIXME, use a real flag for deciding about the key type */
3112 if (root
->fs_info
->tree_root
== root
)
3113 key_type
= BTRFS_DIR_ITEM_KEY
;
3115 /* special case for "." */
3116 if (filp
->f_pos
== 0) {
3117 over
= filldir(dirent
, ".", 1,
3124 /* special case for .., just use the back ref */
3125 if (filp
->f_pos
== 1) {
3126 u64 pino
= parent_ino(filp
->f_path
.dentry
);
3127 over
= filldir(dirent
, "..", 2,
3133 path
= btrfs_alloc_path();
3136 btrfs_set_key_type(&key
, key_type
);
3137 key
.offset
= filp
->f_pos
;
3138 key
.objectid
= inode
->i_ino
;
3140 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3146 leaf
= path
->nodes
[0];
3147 nritems
= btrfs_header_nritems(leaf
);
3148 slot
= path
->slots
[0];
3149 if (advance
|| slot
>= nritems
) {
3150 if (slot
>= nritems
- 1) {
3151 ret
= btrfs_next_leaf(root
, path
);
3154 leaf
= path
->nodes
[0];
3155 nritems
= btrfs_header_nritems(leaf
);
3156 slot
= path
->slots
[0];
3164 item
= btrfs_item_nr(leaf
, slot
);
3165 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3167 if (found_key
.objectid
!= key
.objectid
)
3169 if (btrfs_key_type(&found_key
) != key_type
)
3171 if (found_key
.offset
< filp
->f_pos
)
3174 filp
->f_pos
= found_key
.offset
;
3176 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
3178 di_total
= btrfs_item_size(leaf
, item
);
3180 while (di_cur
< di_total
) {
3181 struct btrfs_key location
;
3183 name_len
= btrfs_dir_name_len(leaf
, di
);
3184 if (name_len
<= sizeof(tmp_name
)) {
3185 name_ptr
= tmp_name
;
3187 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
3193 read_extent_buffer(leaf
, name_ptr
,
3194 (unsigned long)(di
+ 1), name_len
);
3196 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
3197 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
3199 /* is this a reference to our own snapshot? If so
3202 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
3203 location
.objectid
== root
->root_key
.objectid
) {
3207 over
= filldir(dirent
, name_ptr
, name_len
,
3208 found_key
.offset
, location
.objectid
,
3212 if (name_ptr
!= tmp_name
)
3217 di_len
= btrfs_dir_name_len(leaf
, di
) +
3218 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
3220 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
3224 /* Reached end of directory/root. Bump pos past the last item. */
3225 if (key_type
== BTRFS_DIR_INDEX_KEY
)
3226 filp
->f_pos
= INT_LIMIT(typeof(filp
->f_pos
));
3232 btrfs_free_path(path
);
3236 int btrfs_write_inode(struct inode
*inode
, int wait
)
3238 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3239 struct btrfs_trans_handle
*trans
;
3242 if (root
->fs_info
->btree_inode
== inode
)
3246 trans
= btrfs_join_transaction(root
, 1);
3247 btrfs_set_trans_block_group(trans
, inode
);
3248 ret
= btrfs_commit_transaction(trans
, root
);
3254 * This is somewhat expensive, updating the tree every time the
3255 * inode changes. But, it is most likely to find the inode in cache.
3256 * FIXME, needs more benchmarking...there are no reasons other than performance
3257 * to keep or drop this code.
3259 void btrfs_dirty_inode(struct inode
*inode
)
3261 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3262 struct btrfs_trans_handle
*trans
;
3264 trans
= btrfs_join_transaction(root
, 1);
3265 btrfs_set_trans_block_group(trans
, inode
);
3266 btrfs_update_inode(trans
, root
, inode
);
3267 btrfs_end_transaction(trans
, root
);
3271 * find the highest existing sequence number in a directory
3272 * and then set the in-memory index_cnt variable to reflect
3273 * free sequence numbers
3275 static int btrfs_set_inode_index_count(struct inode
*inode
)
3277 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3278 struct btrfs_key key
, found_key
;
3279 struct btrfs_path
*path
;
3280 struct extent_buffer
*leaf
;
3283 key
.objectid
= inode
->i_ino
;
3284 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
3285 key
.offset
= (u64
)-1;
3287 path
= btrfs_alloc_path();
3291 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3294 /* FIXME: we should be able to handle this */
3300 * MAGIC NUMBER EXPLANATION:
3301 * since we search a directory based on f_pos we have to start at 2
3302 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3303 * else has to start at 2
3305 if (path
->slots
[0] == 0) {
3306 BTRFS_I(inode
)->index_cnt
= 2;
3312 leaf
= path
->nodes
[0];
3313 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3315 if (found_key
.objectid
!= inode
->i_ino
||
3316 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
3317 BTRFS_I(inode
)->index_cnt
= 2;
3321 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
3323 btrfs_free_path(path
);
3328 * helper to find a free sequence number in a given directory. This current
3329 * code is very simple, later versions will do smarter things in the btree
3331 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
3335 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
3336 ret
= btrfs_set_inode_index_count(dir
);
3342 *index
= BTRFS_I(dir
)->index_cnt
;
3343 BTRFS_I(dir
)->index_cnt
++;
3348 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
3349 struct btrfs_root
*root
,
3351 const char *name
, int name_len
,
3352 u64 ref_objectid
, u64 objectid
,
3353 u64 alloc_hint
, int mode
, u64
*index
)
3355 struct inode
*inode
;
3356 struct btrfs_inode_item
*inode_item
;
3357 struct btrfs_key
*location
;
3358 struct btrfs_path
*path
;
3359 struct btrfs_inode_ref
*ref
;
3360 struct btrfs_key key
[2];
3366 path
= btrfs_alloc_path();
3369 inode
= new_inode(root
->fs_info
->sb
);
3371 return ERR_PTR(-ENOMEM
);
3374 ret
= btrfs_set_inode_index(dir
, index
);
3376 return ERR_PTR(ret
);
3379 * index_cnt is ignored for everything but a dir,
3380 * btrfs_get_inode_index_count has an explanation for the magic
3383 init_btrfs_i(inode
);
3384 BTRFS_I(inode
)->index_cnt
= 2;
3385 BTRFS_I(inode
)->root
= root
;
3386 BTRFS_I(inode
)->generation
= trans
->transid
;
3392 BTRFS_I(inode
)->block_group
=
3393 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
3395 key
[0].objectid
= objectid
;
3396 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
3399 key
[1].objectid
= objectid
;
3400 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
3401 key
[1].offset
= ref_objectid
;
3403 sizes
[0] = sizeof(struct btrfs_inode_item
);
3404 sizes
[1] = name_len
+ sizeof(*ref
);
3406 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
3410 if (objectid
> root
->highest_inode
)
3411 root
->highest_inode
= objectid
;
3413 inode
->i_uid
= current_fsuid();
3414 inode
->i_gid
= current_fsgid();
3415 inode
->i_mode
= mode
;
3416 inode
->i_ino
= objectid
;
3417 inode_set_bytes(inode
, 0);
3418 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3419 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3420 struct btrfs_inode_item
);
3421 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
3423 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3424 struct btrfs_inode_ref
);
3425 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
3426 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
3427 ptr
= (unsigned long)(ref
+ 1);
3428 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
3430 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3431 btrfs_free_path(path
);
3433 location
= &BTRFS_I(inode
)->location
;
3434 location
->objectid
= objectid
;
3435 location
->offset
= 0;
3436 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3438 insert_inode_hash(inode
);
3442 BTRFS_I(dir
)->index_cnt
--;
3443 btrfs_free_path(path
);
3444 return ERR_PTR(ret
);
3447 static inline u8
btrfs_inode_type(struct inode
*inode
)
3449 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
3453 * utility function to add 'inode' into 'parent_inode' with
3454 * a give name and a given sequence number.
3455 * if 'add_backref' is true, also insert a backref from the
3456 * inode to the parent directory.
3458 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3459 struct inode
*parent_inode
, struct inode
*inode
,
3460 const char *name
, int name_len
, int add_backref
, u64 index
)
3463 struct btrfs_key key
;
3464 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3466 key
.objectid
= inode
->i_ino
;
3467 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3470 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3471 parent_inode
->i_ino
,
3472 &key
, btrfs_inode_type(inode
),
3476 ret
= btrfs_insert_inode_ref(trans
, root
,
3479 parent_inode
->i_ino
,
3482 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3484 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3485 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3490 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3491 struct dentry
*dentry
, struct inode
*inode
,
3492 int backref
, u64 index
)
3494 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3495 inode
, dentry
->d_name
.name
,
3496 dentry
->d_name
.len
, backref
, index
);
3498 d_instantiate(dentry
, inode
);
3506 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3507 int mode
, dev_t rdev
)
3509 struct btrfs_trans_handle
*trans
;
3510 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3511 struct inode
*inode
= NULL
;
3515 unsigned long nr
= 0;
3518 if (!new_valid_dev(rdev
))
3521 err
= btrfs_check_free_space(root
, 1, 0);
3525 trans
= btrfs_start_transaction(root
, 1);
3526 btrfs_set_trans_block_group(trans
, dir
);
3528 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3534 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3536 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3537 BTRFS_I(dir
)->block_group
, mode
, &index
);
3538 err
= PTR_ERR(inode
);
3542 err
= btrfs_init_acl(inode
, dir
);
3548 btrfs_set_trans_block_group(trans
, inode
);
3549 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3553 inode
->i_op
= &btrfs_special_inode_operations
;
3554 init_special_inode(inode
, inode
->i_mode
, rdev
);
3555 btrfs_update_inode(trans
, root
, inode
);
3557 dir
->i_sb
->s_dirt
= 1;
3558 btrfs_update_inode_block_group(trans
, inode
);
3559 btrfs_update_inode_block_group(trans
, dir
);
3561 nr
= trans
->blocks_used
;
3562 btrfs_end_transaction_throttle(trans
, root
);
3565 inode_dec_link_count(inode
);
3568 btrfs_btree_balance_dirty(root
, nr
);
3572 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3573 int mode
, struct nameidata
*nd
)
3575 struct btrfs_trans_handle
*trans
;
3576 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3577 struct inode
*inode
= NULL
;
3580 unsigned long nr
= 0;
3584 err
= btrfs_check_free_space(root
, 1, 0);
3587 trans
= btrfs_start_transaction(root
, 1);
3588 btrfs_set_trans_block_group(trans
, dir
);
3590 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3596 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3598 dentry
->d_parent
->d_inode
->i_ino
,
3599 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3601 err
= PTR_ERR(inode
);
3605 err
= btrfs_init_acl(inode
, dir
);
3611 btrfs_set_trans_block_group(trans
, inode
);
3612 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3616 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3617 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3618 inode
->i_fop
= &btrfs_file_operations
;
3619 inode
->i_op
= &btrfs_file_inode_operations
;
3620 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3622 dir
->i_sb
->s_dirt
= 1;
3623 btrfs_update_inode_block_group(trans
, inode
);
3624 btrfs_update_inode_block_group(trans
, dir
);
3626 nr
= trans
->blocks_used
;
3627 btrfs_end_transaction_throttle(trans
, root
);
3630 inode_dec_link_count(inode
);
3633 btrfs_btree_balance_dirty(root
, nr
);
3637 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3638 struct dentry
*dentry
)
3640 struct btrfs_trans_handle
*trans
;
3641 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3642 struct inode
*inode
= old_dentry
->d_inode
;
3644 unsigned long nr
= 0;
3648 if (inode
->i_nlink
== 0)
3651 btrfs_inc_nlink(inode
);
3652 err
= btrfs_check_free_space(root
, 1, 0);
3655 err
= btrfs_set_inode_index(dir
, &index
);
3659 trans
= btrfs_start_transaction(root
, 1);
3661 btrfs_set_trans_block_group(trans
, dir
);
3662 atomic_inc(&inode
->i_count
);
3664 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3669 dir
->i_sb
->s_dirt
= 1;
3670 btrfs_update_inode_block_group(trans
, dir
);
3671 err
= btrfs_update_inode(trans
, root
, inode
);
3676 nr
= trans
->blocks_used
;
3677 btrfs_end_transaction_throttle(trans
, root
);
3680 inode_dec_link_count(inode
);
3683 btrfs_btree_balance_dirty(root
, nr
);
3687 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3689 struct inode
*inode
= NULL
;
3690 struct btrfs_trans_handle
*trans
;
3691 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3693 int drop_on_err
= 0;
3696 unsigned long nr
= 1;
3698 err
= btrfs_check_free_space(root
, 1, 0);
3702 trans
= btrfs_start_transaction(root
, 1);
3703 btrfs_set_trans_block_group(trans
, dir
);
3705 if (IS_ERR(trans
)) {
3706 err
= PTR_ERR(trans
);
3710 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3716 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3718 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3719 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3721 if (IS_ERR(inode
)) {
3722 err
= PTR_ERR(inode
);
3728 err
= btrfs_init_acl(inode
, dir
);
3732 inode
->i_op
= &btrfs_dir_inode_operations
;
3733 inode
->i_fop
= &btrfs_dir_file_operations
;
3734 btrfs_set_trans_block_group(trans
, inode
);
3736 btrfs_i_size_write(inode
, 0);
3737 err
= btrfs_update_inode(trans
, root
, inode
);
3741 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3742 inode
, dentry
->d_name
.name
,
3743 dentry
->d_name
.len
, 0, index
);
3747 d_instantiate(dentry
, inode
);
3749 dir
->i_sb
->s_dirt
= 1;
3750 btrfs_update_inode_block_group(trans
, inode
);
3751 btrfs_update_inode_block_group(trans
, dir
);
3754 nr
= trans
->blocks_used
;
3755 btrfs_end_transaction_throttle(trans
, root
);
3760 btrfs_btree_balance_dirty(root
, nr
);
3764 /* helper for btfs_get_extent. Given an existing extent in the tree,
3765 * and an extent that you want to insert, deal with overlap and insert
3766 * the new extent into the tree.
3768 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3769 struct extent_map
*existing
,
3770 struct extent_map
*em
,
3771 u64 map_start
, u64 map_len
)
3775 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3776 start_diff
= map_start
- em
->start
;
3777 em
->start
= map_start
;
3779 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3780 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3781 em
->block_start
+= start_diff
;
3782 em
->block_len
-= start_diff
;
3784 return add_extent_mapping(em_tree
, em
);
3787 static noinline
int uncompress_inline(struct btrfs_path
*path
,
3788 struct inode
*inode
, struct page
*page
,
3789 size_t pg_offset
, u64 extent_offset
,
3790 struct btrfs_file_extent_item
*item
)
3793 struct extent_buffer
*leaf
= path
->nodes
[0];
3796 unsigned long inline_size
;
3799 WARN_ON(pg_offset
!= 0);
3800 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
3801 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
3802 btrfs_item_nr(leaf
, path
->slots
[0]));
3803 tmp
= kmalloc(inline_size
, GFP_NOFS
);
3804 ptr
= btrfs_file_extent_inline_start(item
);
3806 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
3808 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
3809 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
3810 inline_size
, max_size
);
3812 char *kaddr
= kmap_atomic(page
, KM_USER0
);
3813 unsigned long copy_size
= min_t(u64
,
3814 PAGE_CACHE_SIZE
- pg_offset
,
3815 max_size
- extent_offset
);
3816 memset(kaddr
+ pg_offset
, 0, copy_size
);
3817 kunmap_atomic(kaddr
, KM_USER0
);
3824 * a bit scary, this does extent mapping from logical file offset to the disk.
3825 * the ugly parts come from merging extents from the disk with the
3826 * in-ram representation. This gets more complex because of the data=ordered code,
3827 * where the in-ram extents might be locked pending data=ordered completion.
3829 * This also copies inline extents directly into the page.
3831 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
3832 size_t pg_offset
, u64 start
, u64 len
,
3838 u64 extent_start
= 0;
3840 u64 objectid
= inode
->i_ino
;
3842 struct btrfs_path
*path
= NULL
;
3843 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3844 struct btrfs_file_extent_item
*item
;
3845 struct extent_buffer
*leaf
;
3846 struct btrfs_key found_key
;
3847 struct extent_map
*em
= NULL
;
3848 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3849 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3850 struct btrfs_trans_handle
*trans
= NULL
;
3854 spin_lock(&em_tree
->lock
);
3855 em
= lookup_extent_mapping(em_tree
, start
, len
);
3857 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3858 spin_unlock(&em_tree
->lock
);
3861 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
3862 free_extent_map(em
);
3863 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
3864 free_extent_map(em
);
3868 em
= alloc_extent_map(GFP_NOFS
);
3873 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3874 em
->start
= EXTENT_MAP_HOLE
;
3875 em
->orig_start
= EXTENT_MAP_HOLE
;
3877 em
->block_len
= (u64
)-1;
3880 path
= btrfs_alloc_path();
3884 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3885 objectid
, start
, trans
!= NULL
);
3892 if (path
->slots
[0] == 0)
3897 leaf
= path
->nodes
[0];
3898 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3899 struct btrfs_file_extent_item
);
3900 /* are we inside the extent that was found? */
3901 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3902 found_type
= btrfs_key_type(&found_key
);
3903 if (found_key
.objectid
!= objectid
||
3904 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3908 found_type
= btrfs_file_extent_type(leaf
, item
);
3909 extent_start
= found_key
.offset
;
3910 compressed
= btrfs_file_extent_compression(leaf
, item
);
3911 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3912 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3913 extent_end
= extent_start
+
3914 btrfs_file_extent_num_bytes(leaf
, item
);
3915 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3917 size
= btrfs_file_extent_inline_len(leaf
, item
);
3918 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
3919 ~((u64
)root
->sectorsize
- 1);
3922 if (start
>= extent_end
) {
3924 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3925 ret
= btrfs_next_leaf(root
, path
);
3932 leaf
= path
->nodes
[0];
3934 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3935 if (found_key
.objectid
!= objectid
||
3936 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3938 if (start
+ len
<= found_key
.offset
)
3941 em
->len
= found_key
.offset
- start
;
3945 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3946 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3947 em
->start
= extent_start
;
3948 em
->len
= extent_end
- extent_start
;
3949 em
->orig_start
= extent_start
-
3950 btrfs_file_extent_offset(leaf
, item
);
3951 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
3953 em
->block_start
= EXTENT_MAP_HOLE
;
3957 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3958 em
->block_start
= bytenr
;
3959 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
3962 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
3963 em
->block_start
= bytenr
;
3964 em
->block_len
= em
->len
;
3965 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
3966 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
3969 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3973 size_t extent_offset
;
3976 em
->block_start
= EXTENT_MAP_INLINE
;
3977 if (!page
|| create
) {
3978 em
->start
= extent_start
;
3979 em
->len
= extent_end
- extent_start
;
3983 size
= btrfs_file_extent_inline_len(leaf
, item
);
3984 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
3985 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
3986 size
- extent_offset
);
3987 em
->start
= extent_start
+ extent_offset
;
3988 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
3989 ~((u64
)root
->sectorsize
- 1);
3990 em
->orig_start
= EXTENT_MAP_INLINE
;
3992 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3993 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
3994 if (create
== 0 && !PageUptodate(page
)) {
3995 if (btrfs_file_extent_compression(leaf
, item
) ==
3996 BTRFS_COMPRESS_ZLIB
) {
3997 ret
= uncompress_inline(path
, inode
, page
,
3999 extent_offset
, item
);
4003 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4007 flush_dcache_page(page
);
4008 } else if (create
&& PageUptodate(page
)) {
4011 free_extent_map(em
);
4013 btrfs_release_path(root
, path
);
4014 trans
= btrfs_join_transaction(root
, 1);
4018 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4021 btrfs_mark_buffer_dirty(leaf
);
4023 set_extent_uptodate(io_tree
, em
->start
,
4024 extent_map_end(em
) - 1, GFP_NOFS
);
4027 printk("unkknown found_type %d\n", found_type
);
4034 em
->block_start
= EXTENT_MAP_HOLE
;
4035 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
4037 btrfs_release_path(root
, path
);
4038 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
4039 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em
->start
, em
->len
, start
, len
);
4045 spin_lock(&em_tree
->lock
);
4046 ret
= add_extent_mapping(em_tree
, em
);
4047 /* it is possible that someone inserted the extent into the tree
4048 * while we had the lock dropped. It is also possible that
4049 * an overlapping map exists in the tree
4051 if (ret
== -EEXIST
) {
4052 struct extent_map
*existing
;
4056 existing
= lookup_extent_mapping(em_tree
, start
, len
);
4057 if (existing
&& (existing
->start
> start
||
4058 existing
->start
+ existing
->len
<= start
)) {
4059 free_extent_map(existing
);
4063 existing
= lookup_extent_mapping(em_tree
, em
->start
,
4066 err
= merge_extent_mapping(em_tree
, existing
,
4069 free_extent_map(existing
);
4071 free_extent_map(em
);
4076 printk("failing to insert %Lu %Lu\n",
4078 free_extent_map(em
);
4082 free_extent_map(em
);
4087 spin_unlock(&em_tree
->lock
);
4090 btrfs_free_path(path
);
4092 ret
= btrfs_end_transaction(trans
, root
);
4098 free_extent_map(em
);
4100 return ERR_PTR(err
);
4105 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
4106 const struct iovec
*iov
, loff_t offset
,
4107 unsigned long nr_segs
)
4112 static sector_t
btrfs_bmap(struct address_space
*mapping
, sector_t iblock
)
4114 return extent_bmap(mapping
, iblock
, btrfs_get_extent
);
4117 int btrfs_readpage(struct file
*file
, struct page
*page
)
4119 struct extent_io_tree
*tree
;
4120 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4121 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
4124 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
4126 struct extent_io_tree
*tree
;
4129 if (current
->flags
& PF_MEMALLOC
) {
4130 redirty_page_for_writepage(wbc
, page
);
4134 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4135 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
4138 int btrfs_writepages(struct address_space
*mapping
,
4139 struct writeback_control
*wbc
)
4141 struct extent_io_tree
*tree
;
4143 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4144 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
4148 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
4149 struct list_head
*pages
, unsigned nr_pages
)
4151 struct extent_io_tree
*tree
;
4152 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4153 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
4156 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4158 struct extent_io_tree
*tree
;
4159 struct extent_map_tree
*map
;
4162 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4163 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
4164 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
4166 ClearPagePrivate(page
);
4167 set_page_private(page
, 0);
4168 page_cache_release(page
);
4173 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4175 if (PageWriteback(page
) || PageDirty(page
))
4177 return __btrfs_releasepage(page
, gfp_flags
);
4180 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
4182 struct extent_io_tree
*tree
;
4183 struct btrfs_ordered_extent
*ordered
;
4184 u64 page_start
= page_offset(page
);
4185 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4187 wait_on_page_writeback(page
);
4188 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4190 btrfs_releasepage(page
, GFP_NOFS
);
4194 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4195 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
4199 * IO on this page will never be started, so we need
4200 * to account for any ordered extents now
4202 clear_extent_bit(tree
, page_start
, page_end
,
4203 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4204 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
4205 btrfs_finish_ordered_io(page
->mapping
->host
,
4206 page_start
, page_end
);
4207 btrfs_put_ordered_extent(ordered
);
4208 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4210 clear_extent_bit(tree
, page_start
, page_end
,
4211 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4214 __btrfs_releasepage(page
, GFP_NOFS
);
4216 ClearPageChecked(page
);
4217 if (PagePrivate(page
)) {
4218 ClearPagePrivate(page
);
4219 set_page_private(page
, 0);
4220 page_cache_release(page
);
4225 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4226 * called from a page fault handler when a page is first dirtied. Hence we must
4227 * be careful to check for EOF conditions here. We set the page up correctly
4228 * for a written page which means we get ENOSPC checking when writing into
4229 * holes and correct delalloc and unwritten extent mapping on filesystems that
4230 * support these features.
4232 * We are not allowed to take the i_mutex here so we have to play games to
4233 * protect against truncate races as the page could now be beyond EOF. Because
4234 * vmtruncate() writes the inode size before removing pages, once we have the
4235 * page lock we can determine safely if the page is beyond EOF. If it is not
4236 * beyond EOF, then the page is guaranteed safe against truncation until we
4239 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4241 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
4242 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4243 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4244 struct btrfs_ordered_extent
*ordered
;
4246 unsigned long zero_start
;
4252 ret
= btrfs_check_free_space(root
, PAGE_CACHE_SIZE
, 0);
4259 size
= i_size_read(inode
);
4260 page_start
= page_offset(page
);
4261 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4263 if ((page
->mapping
!= inode
->i_mapping
) ||
4264 (page_start
>= size
)) {
4265 /* page got truncated out from underneath us */
4268 wait_on_page_writeback(page
);
4270 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4271 set_page_extent_mapped(page
);
4274 * we can't set the delalloc bits if there are pending ordered
4275 * extents. Drop our locks and wait for them to finish
4277 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4279 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4281 btrfs_start_ordered_extent(inode
, ordered
, 1);
4282 btrfs_put_ordered_extent(ordered
);
4286 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
4289 /* page is wholly or partially inside EOF */
4290 if (page_start
+ PAGE_CACHE_SIZE
> size
)
4291 zero_start
= size
& ~PAGE_CACHE_MASK
;
4293 zero_start
= PAGE_CACHE_SIZE
;
4295 if (zero_start
!= PAGE_CACHE_SIZE
) {
4297 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
4298 flush_dcache_page(page
);
4301 ClearPageChecked(page
);
4302 set_page_dirty(page
);
4303 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4311 static void btrfs_truncate(struct inode
*inode
)
4313 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4315 struct btrfs_trans_handle
*trans
;
4317 u64 mask
= root
->sectorsize
- 1;
4319 if (!S_ISREG(inode
->i_mode
))
4321 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4324 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
4325 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
4327 trans
= btrfs_start_transaction(root
, 1);
4328 btrfs_set_trans_block_group(trans
, inode
);
4329 btrfs_i_size_write(inode
, inode
->i_size
);
4331 ret
= btrfs_orphan_add(trans
, inode
);
4334 /* FIXME, add redo link to tree so we don't leak on crash */
4335 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
4336 BTRFS_EXTENT_DATA_KEY
);
4337 btrfs_update_inode(trans
, root
, inode
);
4339 ret
= btrfs_orphan_del(trans
, inode
);
4343 nr
= trans
->blocks_used
;
4344 ret
= btrfs_end_transaction_throttle(trans
, root
);
4346 btrfs_btree_balance_dirty(root
, nr
);
4350 * create a new subvolume directory/inode (helper for the ioctl).
4352 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
4353 struct btrfs_root
*new_root
, struct dentry
*dentry
,
4354 u64 new_dirid
, u64 alloc_hint
)
4356 struct inode
*inode
;
4360 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
4361 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
4363 return PTR_ERR(inode
);
4364 inode
->i_op
= &btrfs_dir_inode_operations
;
4365 inode
->i_fop
= &btrfs_dir_file_operations
;
4368 btrfs_i_size_write(inode
, 0);
4370 error
= btrfs_update_inode(trans
, new_root
, inode
);
4374 d_instantiate(dentry
, inode
);
4378 /* helper function for file defrag and space balancing. This
4379 * forces readahead on a given range of bytes in an inode
4381 unsigned long btrfs_force_ra(struct address_space
*mapping
,
4382 struct file_ra_state
*ra
, struct file
*file
,
4383 pgoff_t offset
, pgoff_t last_index
)
4385 pgoff_t req_size
= last_index
- offset
+ 1;
4387 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
4388 return offset
+ req_size
;
4391 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
4393 struct btrfs_inode
*ei
;
4395 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
4399 ei
->logged_trans
= 0;
4400 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
4401 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
4402 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
4403 INIT_LIST_HEAD(&ei
->i_orphan
);
4404 return &ei
->vfs_inode
;
4407 void btrfs_destroy_inode(struct inode
*inode
)
4409 struct btrfs_ordered_extent
*ordered
;
4410 WARN_ON(!list_empty(&inode
->i_dentry
));
4411 WARN_ON(inode
->i_data
.nrpages
);
4413 if (BTRFS_I(inode
)->i_acl
&&
4414 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
4415 posix_acl_release(BTRFS_I(inode
)->i_acl
);
4416 if (BTRFS_I(inode
)->i_default_acl
&&
4417 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
4418 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
4420 spin_lock(&BTRFS_I(inode
)->root
->list_lock
);
4421 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
4422 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
4423 " list\n", inode
->i_ino
);
4426 spin_unlock(&BTRFS_I(inode
)->root
->list_lock
);
4429 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
4433 printk("found ordered extent %Lu %Lu\n",
4434 ordered
->file_offset
, ordered
->len
);
4435 btrfs_remove_ordered_extent(inode
, ordered
);
4436 btrfs_put_ordered_extent(ordered
);
4437 btrfs_put_ordered_extent(ordered
);
4440 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4441 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4444 static void init_once(void *foo
)
4446 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4448 inode_init_once(&ei
->vfs_inode
);
4451 void btrfs_destroy_cachep(void)
4453 if (btrfs_inode_cachep
)
4454 kmem_cache_destroy(btrfs_inode_cachep
);
4455 if (btrfs_trans_handle_cachep
)
4456 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4457 if (btrfs_transaction_cachep
)
4458 kmem_cache_destroy(btrfs_transaction_cachep
);
4459 if (btrfs_bit_radix_cachep
)
4460 kmem_cache_destroy(btrfs_bit_radix_cachep
);
4461 if (btrfs_path_cachep
)
4462 kmem_cache_destroy(btrfs_path_cachep
);
4465 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
4466 unsigned long extra_flags
,
4467 void (*ctor
)(void *))
4469 return kmem_cache_create(name
, size
, 0, (SLAB_RECLAIM_ACCOUNT
|
4470 SLAB_MEM_SPREAD
| extra_flags
), ctor
);
4473 int btrfs_init_cachep(void)
4475 btrfs_inode_cachep
= btrfs_cache_create("btrfs_inode_cache",
4476 sizeof(struct btrfs_inode
),
4478 if (!btrfs_inode_cachep
)
4480 btrfs_trans_handle_cachep
=
4481 btrfs_cache_create("btrfs_trans_handle_cache",
4482 sizeof(struct btrfs_trans_handle
),
4484 if (!btrfs_trans_handle_cachep
)
4486 btrfs_transaction_cachep
= btrfs_cache_create("btrfs_transaction_cache",
4487 sizeof(struct btrfs_transaction
),
4489 if (!btrfs_transaction_cachep
)
4491 btrfs_path_cachep
= btrfs_cache_create("btrfs_path_cache",
4492 sizeof(struct btrfs_path
),
4494 if (!btrfs_path_cachep
)
4496 btrfs_bit_radix_cachep
= btrfs_cache_create("btrfs_radix", 256,
4497 SLAB_DESTROY_BY_RCU
, NULL
);
4498 if (!btrfs_bit_radix_cachep
)
4502 btrfs_destroy_cachep();
4506 static int btrfs_getattr(struct vfsmount
*mnt
,
4507 struct dentry
*dentry
, struct kstat
*stat
)
4509 struct inode
*inode
= dentry
->d_inode
;
4510 generic_fillattr(inode
, stat
);
4511 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
4512 stat
->blksize
= PAGE_CACHE_SIZE
;
4513 stat
->blocks
= (inode_get_bytes(inode
) +
4514 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4518 static int btrfs_rename(struct inode
* old_dir
, struct dentry
*old_dentry
,
4519 struct inode
* new_dir
,struct dentry
*new_dentry
)
4521 struct btrfs_trans_handle
*trans
;
4522 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4523 struct inode
*new_inode
= new_dentry
->d_inode
;
4524 struct inode
*old_inode
= old_dentry
->d_inode
;
4525 struct timespec ctime
= CURRENT_TIME
;
4529 /* we're not allowed to rename between subvolumes */
4530 if (BTRFS_I(old_inode
)->root
->root_key
.objectid
!=
4531 BTRFS_I(new_dir
)->root
->root_key
.objectid
)
4534 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4535 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4539 /* to rename a snapshot or subvolume, we need to juggle the
4540 * backrefs. This isn't coded yet
4542 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
4545 ret
= btrfs_check_free_space(root
, 1, 0);
4549 trans
= btrfs_start_transaction(root
, 1);
4551 btrfs_set_trans_block_group(trans
, new_dir
);
4553 btrfs_inc_nlink(old_dentry
->d_inode
);
4554 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4555 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4556 old_inode
->i_ctime
= ctime
;
4558 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4559 old_dentry
->d_name
.name
,
4560 old_dentry
->d_name
.len
);
4565 new_inode
->i_ctime
= CURRENT_TIME
;
4566 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4567 new_dentry
->d_inode
,
4568 new_dentry
->d_name
.name
,
4569 new_dentry
->d_name
.len
);
4572 if (new_inode
->i_nlink
== 0) {
4573 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4579 ret
= btrfs_set_inode_index(new_dir
, &index
);
4583 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4584 old_inode
, new_dentry
->d_name
.name
,
4585 new_dentry
->d_name
.len
, 1, index
);
4590 btrfs_end_transaction_throttle(trans
, root
);
4596 * some fairly slow code that needs optimization. This walks the list
4597 * of all the inodes with pending delalloc and forces them to disk.
4599 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4601 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4602 struct btrfs_inode
*binode
;
4603 struct inode
*inode
;
4604 unsigned long flags
;
4606 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
4609 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
4610 while(!list_empty(head
)) {
4611 binode
= list_entry(head
->next
, struct btrfs_inode
,
4613 inode
= igrab(&binode
->vfs_inode
);
4615 list_del_init(&binode
->delalloc_inodes
);
4616 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
4618 filemap_flush(inode
->i_mapping
);
4622 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
4624 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
4626 /* the filemap_flush will queue IO into the worker threads, but
4627 * we have to make sure the IO is actually started and that
4628 * ordered extents get created before we return
4630 atomic_inc(&root
->fs_info
->async_submit_draining
);
4631 while(atomic_read(&root
->fs_info
->nr_async_submits
) ||
4632 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
4633 wait_event(root
->fs_info
->async_submit_wait
,
4634 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
4635 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
4637 atomic_dec(&root
->fs_info
->async_submit_draining
);
4641 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4642 const char *symname
)
4644 struct btrfs_trans_handle
*trans
;
4645 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4646 struct btrfs_path
*path
;
4647 struct btrfs_key key
;
4648 struct inode
*inode
= NULL
;
4656 struct btrfs_file_extent_item
*ei
;
4657 struct extent_buffer
*leaf
;
4658 unsigned long nr
= 0;
4660 name_len
= strlen(symname
) + 1;
4661 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4662 return -ENAMETOOLONG
;
4664 err
= btrfs_check_free_space(root
, 1, 0);
4668 trans
= btrfs_start_transaction(root
, 1);
4669 btrfs_set_trans_block_group(trans
, dir
);
4671 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4677 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4679 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4680 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4682 err
= PTR_ERR(inode
);
4686 err
= btrfs_init_acl(inode
, dir
);
4692 btrfs_set_trans_block_group(trans
, inode
);
4693 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4697 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4698 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4699 inode
->i_fop
= &btrfs_file_operations
;
4700 inode
->i_op
= &btrfs_file_inode_operations
;
4701 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4703 dir
->i_sb
->s_dirt
= 1;
4704 btrfs_update_inode_block_group(trans
, inode
);
4705 btrfs_update_inode_block_group(trans
, dir
);
4709 path
= btrfs_alloc_path();
4711 key
.objectid
= inode
->i_ino
;
4713 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
4714 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
4715 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
4721 leaf
= path
->nodes
[0];
4722 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
4723 struct btrfs_file_extent_item
);
4724 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
4725 btrfs_set_file_extent_type(leaf
, ei
,
4726 BTRFS_FILE_EXTENT_INLINE
);
4727 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
4728 btrfs_set_file_extent_compression(leaf
, ei
, 0);
4729 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
4730 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
4732 ptr
= btrfs_file_extent_inline_start(ei
);
4733 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
4734 btrfs_mark_buffer_dirty(leaf
);
4735 btrfs_free_path(path
);
4737 inode
->i_op
= &btrfs_symlink_inode_operations
;
4738 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
4739 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4740 inode_set_bytes(inode
, name_len
);
4741 btrfs_i_size_write(inode
, name_len
- 1);
4742 err
= btrfs_update_inode(trans
, root
, inode
);
4747 nr
= trans
->blocks_used
;
4748 btrfs_end_transaction_throttle(trans
, root
);
4751 inode_dec_link_count(inode
);
4754 btrfs_btree_balance_dirty(root
, nr
);
4758 static int prealloc_file_range(struct inode
*inode
, u64 start
, u64 end
,
4759 u64 alloc_hint
, int mode
)
4761 struct btrfs_trans_handle
*trans
;
4762 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4763 struct btrfs_key ins
;
4765 u64 cur_offset
= start
;
4766 u64 num_bytes
= end
- start
;
4769 trans
= btrfs_join_transaction(root
, 1);
4771 btrfs_set_trans_block_group(trans
, inode
);
4773 while (num_bytes
> 0) {
4774 alloc_size
= min(num_bytes
, root
->fs_info
->max_extent
);
4775 ret
= btrfs_reserve_extent(trans
, root
, alloc_size
,
4776 root
->sectorsize
, 0, alloc_hint
,
4782 ret
= insert_reserved_file_extent(trans
, inode
,
4783 cur_offset
, ins
.objectid
,
4784 ins
.offset
, ins
.offset
,
4785 ins
.offset
, 0, 0, 0,
4786 BTRFS_FILE_EXTENT_PREALLOC
);
4788 num_bytes
-= ins
.offset
;
4789 cur_offset
+= ins
.offset
;
4790 alloc_hint
= ins
.objectid
+ ins
.offset
;
4793 if (cur_offset
> start
) {
4794 inode
->i_ctime
= CURRENT_TIME
;
4795 btrfs_set_flag(inode
, PREALLOC
);
4796 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4797 cur_offset
> i_size_read(inode
))
4798 btrfs_i_size_write(inode
, cur_offset
);
4799 ret
= btrfs_update_inode(trans
, root
, inode
);
4803 btrfs_end_transaction(trans
, root
);
4807 static long btrfs_fallocate(struct inode
*inode
, int mode
,
4808 loff_t offset
, loff_t len
)
4815 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
4816 struct extent_map
*em
;
4819 alloc_start
= offset
& ~mask
;
4820 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
4822 mutex_lock(&inode
->i_mutex
);
4823 if (alloc_start
> inode
->i_size
) {
4824 ret
= btrfs_cont_expand(inode
, alloc_start
);
4830 struct btrfs_ordered_extent
*ordered
;
4831 lock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
,
4832 alloc_end
- 1, GFP_NOFS
);
4833 ordered
= btrfs_lookup_first_ordered_extent(inode
,
4836 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
4837 ordered
->file_offset
< alloc_end
) {
4838 btrfs_put_ordered_extent(ordered
);
4839 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4840 alloc_start
, alloc_end
- 1, GFP_NOFS
);
4841 btrfs_wait_ordered_range(inode
, alloc_start
,
4842 alloc_end
- alloc_start
);
4845 btrfs_put_ordered_extent(ordered
);
4850 cur_offset
= alloc_start
;
4852 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4853 alloc_end
- cur_offset
, 0);
4854 BUG_ON(IS_ERR(em
) || !em
);
4855 last_byte
= min(extent_map_end(em
), alloc_end
);
4856 last_byte
= (last_byte
+ mask
) & ~mask
;
4857 if (em
->block_start
== EXTENT_MAP_HOLE
) {
4858 ret
= prealloc_file_range(inode
, cur_offset
,
4859 last_byte
, alloc_hint
, mode
);
4861 free_extent_map(em
);
4865 if (em
->block_start
<= EXTENT_MAP_LAST_BYTE
)
4866 alloc_hint
= em
->block_start
;
4867 free_extent_map(em
);
4869 cur_offset
= last_byte
;
4870 if (cur_offset
>= alloc_end
) {
4875 unlock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, alloc_end
- 1,
4878 mutex_unlock(&inode
->i_mutex
);
4882 static int btrfs_set_page_dirty(struct page
*page
)
4884 return __set_page_dirty_nobuffers(page
);
4887 static int btrfs_permission(struct inode
*inode
, int mask
)
4889 if (btrfs_test_flag(inode
, READONLY
) && (mask
& MAY_WRITE
))
4891 return generic_permission(inode
, mask
, btrfs_check_acl
);
4894 static struct inode_operations btrfs_dir_inode_operations
= {
4895 .getattr
= btrfs_getattr
,
4896 .lookup
= btrfs_lookup
,
4897 .create
= btrfs_create
,
4898 .unlink
= btrfs_unlink
,
4900 .mkdir
= btrfs_mkdir
,
4901 .rmdir
= btrfs_rmdir
,
4902 .rename
= btrfs_rename
,
4903 .symlink
= btrfs_symlink
,
4904 .setattr
= btrfs_setattr
,
4905 .mknod
= btrfs_mknod
,
4906 .setxattr
= btrfs_setxattr
,
4907 .getxattr
= btrfs_getxattr
,
4908 .listxattr
= btrfs_listxattr
,
4909 .removexattr
= btrfs_removexattr
,
4910 .permission
= btrfs_permission
,
4912 static struct inode_operations btrfs_dir_ro_inode_operations
= {
4913 .lookup
= btrfs_lookup
,
4914 .permission
= btrfs_permission
,
4916 static struct file_operations btrfs_dir_file_operations
= {
4917 .llseek
= generic_file_llseek
,
4918 .read
= generic_read_dir
,
4919 .readdir
= btrfs_real_readdir
,
4920 .unlocked_ioctl
= btrfs_ioctl
,
4921 #ifdef CONFIG_COMPAT
4922 .compat_ioctl
= btrfs_ioctl
,
4924 .release
= btrfs_release_file
,
4925 .fsync
= btrfs_sync_file
,
4928 static struct extent_io_ops btrfs_extent_io_ops
= {
4929 .fill_delalloc
= run_delalloc_range
,
4930 .submit_bio_hook
= btrfs_submit_bio_hook
,
4931 .merge_bio_hook
= btrfs_merge_bio_hook
,
4932 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
4933 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
4934 .writepage_start_hook
= btrfs_writepage_start_hook
,
4935 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
4936 .set_bit_hook
= btrfs_set_bit_hook
,
4937 .clear_bit_hook
= btrfs_clear_bit_hook
,
4940 static struct address_space_operations btrfs_aops
= {
4941 .readpage
= btrfs_readpage
,
4942 .writepage
= btrfs_writepage
,
4943 .writepages
= btrfs_writepages
,
4944 .readpages
= btrfs_readpages
,
4945 .sync_page
= block_sync_page
,
4947 .direct_IO
= btrfs_direct_IO
,
4948 .invalidatepage
= btrfs_invalidatepage
,
4949 .releasepage
= btrfs_releasepage
,
4950 .set_page_dirty
= btrfs_set_page_dirty
,
4953 static struct address_space_operations btrfs_symlink_aops
= {
4954 .readpage
= btrfs_readpage
,
4955 .writepage
= btrfs_writepage
,
4956 .invalidatepage
= btrfs_invalidatepage
,
4957 .releasepage
= btrfs_releasepage
,
4960 static struct inode_operations btrfs_file_inode_operations
= {
4961 .truncate
= btrfs_truncate
,
4962 .getattr
= btrfs_getattr
,
4963 .setattr
= btrfs_setattr
,
4964 .setxattr
= btrfs_setxattr
,
4965 .getxattr
= btrfs_getxattr
,
4966 .listxattr
= btrfs_listxattr
,
4967 .removexattr
= btrfs_removexattr
,
4968 .permission
= btrfs_permission
,
4969 .fallocate
= btrfs_fallocate
,
4971 static struct inode_operations btrfs_special_inode_operations
= {
4972 .getattr
= btrfs_getattr
,
4973 .setattr
= btrfs_setattr
,
4974 .permission
= btrfs_permission
,
4975 .setxattr
= btrfs_setxattr
,
4976 .getxattr
= btrfs_getxattr
,
4977 .listxattr
= btrfs_listxattr
,
4978 .removexattr
= btrfs_removexattr
,
4980 static struct inode_operations btrfs_symlink_inode_operations
= {
4981 .readlink
= generic_readlink
,
4982 .follow_link
= page_follow_link_light
,
4983 .put_link
= page_put_link
,
4984 .permission
= btrfs_permission
,