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
,
106 spin_lock(&root
->fs_info
->delalloc_lock
);
107 total
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
108 used
= btrfs_super_bytes_used(&root
->fs_info
->super_copy
);
116 if (used
+ root
->fs_info
->delalloc_bytes
+ num_required
> thresh
)
118 spin_unlock(&root
->fs_info
->delalloc_lock
);
123 * this does all the hard work for inserting an inline extent into
124 * the btree. The caller should have done a btrfs_drop_extents so that
125 * no overlapping inline items exist in the btree
127 static int noinline
insert_inline_extent(struct btrfs_trans_handle
*trans
,
128 struct btrfs_root
*root
, struct inode
*inode
,
129 u64 start
, size_t size
, size_t compressed_size
,
130 struct page
**compressed_pages
)
132 struct btrfs_key key
;
133 struct btrfs_path
*path
;
134 struct extent_buffer
*leaf
;
135 struct page
*page
= NULL
;
138 struct btrfs_file_extent_item
*ei
;
141 size_t cur_size
= size
;
143 unsigned long offset
;
144 int use_compress
= 0;
146 if (compressed_size
&& compressed_pages
) {
148 cur_size
= compressed_size
;
151 path
= btrfs_alloc_path(); if (!path
)
154 btrfs_set_trans_block_group(trans
, inode
);
156 key
.objectid
= inode
->i_ino
;
158 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
159 inode_add_bytes(inode
, size
);
160 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
162 inode_add_bytes(inode
, size
);
163 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
168 printk("got bad ret %d\n", ret
);
171 leaf
= path
->nodes
[0];
172 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
173 struct btrfs_file_extent_item
);
174 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
175 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
176 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
177 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
178 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
179 ptr
= btrfs_file_extent_inline_start(ei
);
184 while(compressed_size
> 0) {
185 cpage
= compressed_pages
[i
];
186 cur_size
= min_t(unsigned long, compressed_size
,
190 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
195 compressed_size
-= cur_size
;
197 btrfs_set_file_extent_compression(leaf
, ei
,
198 BTRFS_COMPRESS_ZLIB
);
200 page
= find_get_page(inode
->i_mapping
,
201 start
>> PAGE_CACHE_SHIFT
);
202 btrfs_set_file_extent_compression(leaf
, ei
, 0);
203 kaddr
= kmap_atomic(page
, KM_USER0
);
204 offset
= start
& (PAGE_CACHE_SIZE
- 1);
205 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
206 kunmap_atomic(kaddr
, KM_USER0
);
207 page_cache_release(page
);
209 btrfs_mark_buffer_dirty(leaf
);
210 btrfs_free_path(path
);
212 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
213 btrfs_update_inode(trans
, root
, inode
);
216 btrfs_free_path(path
);
222 * conditionally insert an inline extent into the file. This
223 * does the checks required to make sure the data is small enough
224 * to fit as an inline extent.
226 static int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
227 struct btrfs_root
*root
,
228 struct inode
*inode
, u64 start
, u64 end
,
229 size_t compressed_size
,
230 struct page
**compressed_pages
)
232 u64 isize
= i_size_read(inode
);
233 u64 actual_end
= min(end
+ 1, isize
);
234 u64 inline_len
= actual_end
- start
;
235 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
236 ~((u64
)root
->sectorsize
- 1);
238 u64 data_len
= inline_len
;
242 data_len
= compressed_size
;
245 actual_end
>= PAGE_CACHE_SIZE
||
246 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
248 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
250 data_len
> root
->fs_info
->max_inline
) {
254 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
255 aligned_end
, start
, &hint_byte
);
258 if (isize
> actual_end
)
259 inline_len
= min_t(u64
, isize
, actual_end
);
260 ret
= insert_inline_extent(trans
, root
, inode
, start
,
261 inline_len
, compressed_size
,
264 btrfs_drop_extent_cache(inode
, start
, aligned_end
, 0);
268 struct async_extent
{
273 unsigned long nr_pages
;
274 struct list_head list
;
279 struct btrfs_root
*root
;
280 struct page
*locked_page
;
283 struct list_head extents
;
284 struct btrfs_work work
;
287 static noinline
int add_async_extent(struct async_cow
*cow
,
288 u64 start
, u64 ram_size
,
291 unsigned long nr_pages
)
293 struct async_extent
*async_extent
;
295 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
296 async_extent
->start
= start
;
297 async_extent
->ram_size
= ram_size
;
298 async_extent
->compressed_size
= compressed_size
;
299 async_extent
->pages
= pages
;
300 async_extent
->nr_pages
= nr_pages
;
301 list_add_tail(&async_extent
->list
, &cow
->extents
);
306 * we create compressed extents in two phases. The first
307 * phase compresses a range of pages that have already been
308 * locked (both pages and state bits are locked).
310 * This is done inside an ordered work queue, and the compression
311 * is spread across many cpus. The actual IO submission is step
312 * two, and the ordered work queue takes care of making sure that
313 * happens in the same order things were put onto the queue by
314 * writepages and friends.
316 * If this code finds it can't get good compression, it puts an
317 * entry onto the work queue to write the uncompressed bytes. This
318 * makes sure that both compressed inodes and uncompressed inodes
319 * are written in the same order that pdflush sent them down.
321 static noinline
int compress_file_range(struct inode
*inode
,
322 struct page
*locked_page
,
324 struct async_cow
*async_cow
,
327 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
328 struct btrfs_trans_handle
*trans
;
332 u64 blocksize
= root
->sectorsize
;
334 u64 isize
= i_size_read(inode
);
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;
348 actual_end
= min_t(u64
, isize
, end
+ 1);
351 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
352 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
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
&& start
+ num_bytes
< actual_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 u64 isize
= i_size_read(inode
);
700 struct btrfs_key ins
;
701 struct extent_map
*em
;
702 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
705 trans
= btrfs_join_transaction(root
, 1);
707 btrfs_set_trans_block_group(trans
, inode
);
709 actual_end
= min_t(u64
, isize
, end
+ 1);
711 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
712 num_bytes
= max(blocksize
, num_bytes
);
713 disk_num_bytes
= num_bytes
;
717 /* lets try to make an inline extent */
718 ret
= cow_file_range_inline(trans
, root
, inode
,
719 start
, end
, 0, NULL
);
721 extent_clear_unlock_delalloc(inode
,
722 &BTRFS_I(inode
)->io_tree
,
723 start
, end
, NULL
, 1, 1,
725 *nr_written
= *nr_written
+
726 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
733 BUG_ON(disk_num_bytes
>
734 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
736 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
738 while(disk_num_bytes
> 0) {
739 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
740 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
741 root
->sectorsize
, 0, alloc_hint
,
746 em
= alloc_extent_map(GFP_NOFS
);
748 em
->orig_start
= em
->start
;
750 ram_size
= ins
.offset
;
751 em
->len
= ins
.offset
;
753 em
->block_start
= ins
.objectid
;
754 em
->block_len
= ins
.offset
;
755 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
756 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
759 spin_lock(&em_tree
->lock
);
760 ret
= add_extent_mapping(em_tree
, em
);
761 spin_unlock(&em_tree
->lock
);
762 if (ret
!= -EEXIST
) {
766 btrfs_drop_extent_cache(inode
, start
,
767 start
+ ram_size
- 1, 0);
770 cur_alloc_size
= ins
.offset
;
771 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
772 ram_size
, cur_alloc_size
, 0);
775 if (root
->root_key
.objectid
==
776 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
777 ret
= btrfs_reloc_clone_csums(inode
, start
,
782 if (disk_num_bytes
< cur_alloc_size
) {
783 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes
,
787 /* we're not doing compressed IO, don't unlock the first
788 * page (which the caller expects to stay locked), don't
789 * clear any dirty bits and don't set any writeback bits
791 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
792 start
, start
+ ram_size
- 1,
793 locked_page
, unlock
, 1,
795 disk_num_bytes
-= cur_alloc_size
;
796 num_bytes
-= cur_alloc_size
;
797 alloc_hint
= ins
.objectid
+ ins
.offset
;
798 start
+= cur_alloc_size
;
802 btrfs_end_transaction(trans
, root
);
808 * work queue call back to started compression on a file and pages
810 static noinline
void async_cow_start(struct btrfs_work
*work
)
812 struct async_cow
*async_cow
;
814 async_cow
= container_of(work
, struct async_cow
, work
);
816 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
817 async_cow
->start
, async_cow
->end
, async_cow
,
820 async_cow
->inode
= NULL
;
824 * work queue call back to submit previously compressed pages
826 static noinline
void async_cow_submit(struct btrfs_work
*work
)
828 struct async_cow
*async_cow
;
829 struct btrfs_root
*root
;
830 unsigned long nr_pages
;
832 async_cow
= container_of(work
, struct async_cow
, work
);
834 root
= async_cow
->root
;
835 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
838 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
840 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
842 waitqueue_active(&root
->fs_info
->async_submit_wait
))
843 wake_up(&root
->fs_info
->async_submit_wait
);
845 if (async_cow
->inode
) {
846 submit_compressed_extents(async_cow
->inode
, async_cow
);
850 static noinline
void async_cow_free(struct btrfs_work
*work
)
852 struct async_cow
*async_cow
;
853 async_cow
= container_of(work
, struct async_cow
, work
);
857 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
858 u64 start
, u64 end
, int *page_started
,
859 unsigned long *nr_written
)
861 struct async_cow
*async_cow
;
862 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
863 unsigned long nr_pages
;
865 int limit
= 10 * 1024 * 1042;
867 if (!btrfs_test_opt(root
, COMPRESS
)) {
868 return cow_file_range(inode
, locked_page
, start
, end
,
869 page_started
, nr_written
, 1);
872 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
|
873 EXTENT_DELALLOC
, 1, 0, GFP_NOFS
);
875 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
876 async_cow
->inode
= inode
;
877 async_cow
->root
= root
;
878 async_cow
->locked_page
= locked_page
;
879 async_cow
->start
= start
;
881 if (btrfs_test_flag(inode
, NOCOMPRESS
))
884 cur_end
= min(end
, start
+ 512 * 1024 - 1);
886 async_cow
->end
= cur_end
;
887 INIT_LIST_HEAD(&async_cow
->extents
);
889 async_cow
->work
.func
= async_cow_start
;
890 async_cow
->work
.ordered_func
= async_cow_submit
;
891 async_cow
->work
.ordered_free
= async_cow_free
;
892 async_cow
->work
.flags
= 0;
894 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
896 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
898 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
901 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
902 wait_event(root
->fs_info
->async_submit_wait
,
903 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
907 while(atomic_read(&root
->fs_info
->async_submit_draining
) &&
908 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
909 wait_event(root
->fs_info
->async_submit_wait
,
910 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
914 *nr_written
+= nr_pages
;
921 static int noinline
csum_exist_in_range(struct btrfs_root
*root
,
922 u64 bytenr
, u64 num_bytes
)
925 struct btrfs_ordered_sum
*sums
;
928 ret
= btrfs_lookup_csums_range(root
, bytenr
, bytenr
+ num_bytes
- 1,
930 if (ret
== 0 && list_empty(&list
))
933 while (!list_empty(&list
)) {
934 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
935 list_del(&sums
->list
);
942 * when nowcow writeback call back. This checks for snapshots or COW copies
943 * of the extents that exist in the file, and COWs the file as required.
945 * If no cow copies or snapshots exist, we write directly to the existing
948 static int run_delalloc_nocow(struct inode
*inode
, struct page
*locked_page
,
949 u64 start
, u64 end
, int *page_started
, int force
,
950 unsigned long *nr_written
)
952 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
953 struct btrfs_trans_handle
*trans
;
954 struct extent_buffer
*leaf
;
955 struct btrfs_path
*path
;
956 struct btrfs_file_extent_item
*fi
;
957 struct btrfs_key found_key
;
969 path
= btrfs_alloc_path();
971 trans
= btrfs_join_transaction(root
, 1);
977 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
980 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
981 leaf
= path
->nodes
[0];
982 btrfs_item_key_to_cpu(leaf
, &found_key
,
984 if (found_key
.objectid
== inode
->i_ino
&&
985 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
990 leaf
= path
->nodes
[0];
991 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
992 ret
= btrfs_next_leaf(root
, path
);
997 leaf
= path
->nodes
[0];
1003 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1005 if (found_key
.objectid
> inode
->i_ino
||
1006 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1007 found_key
.offset
> end
)
1010 if (found_key
.offset
> cur_offset
) {
1011 extent_end
= found_key
.offset
;
1015 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1016 struct btrfs_file_extent_item
);
1017 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1019 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1020 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1021 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1022 extent_end
= found_key
.offset
+
1023 btrfs_file_extent_num_bytes(leaf
, fi
);
1024 if (extent_end
<= start
) {
1028 if (disk_bytenr
== 0)
1030 if (btrfs_file_extent_compression(leaf
, fi
) ||
1031 btrfs_file_extent_encryption(leaf
, fi
) ||
1032 btrfs_file_extent_other_encoding(leaf
, fi
))
1034 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1036 if (btrfs_extent_readonly(root
, disk_bytenr
))
1038 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1041 disk_bytenr
+= btrfs_file_extent_offset(leaf
, fi
);
1042 disk_bytenr
+= cur_offset
- found_key
.offset
;
1043 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1045 * force cow if csum exists in the range.
1046 * this ensure that csum for a given extent are
1047 * either valid or do not exist.
1049 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1052 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1053 extent_end
= found_key
.offset
+
1054 btrfs_file_extent_inline_len(leaf
, fi
);
1055 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1060 if (extent_end
<= start
) {
1065 if (cow_start
== (u64
)-1)
1066 cow_start
= cur_offset
;
1067 cur_offset
= extent_end
;
1068 if (cur_offset
> end
)
1074 btrfs_release_path(root
, path
);
1075 if (cow_start
!= (u64
)-1) {
1076 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1077 found_key
.offset
- 1, page_started
,
1080 cow_start
= (u64
)-1;
1083 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1084 struct extent_map
*em
;
1085 struct extent_map_tree
*em_tree
;
1086 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1087 em
= alloc_extent_map(GFP_NOFS
);
1088 em
->start
= cur_offset
;
1089 em
->orig_start
= em
->start
;
1090 em
->len
= num_bytes
;
1091 em
->block_len
= num_bytes
;
1092 em
->block_start
= disk_bytenr
;
1093 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1094 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1096 spin_lock(&em_tree
->lock
);
1097 ret
= add_extent_mapping(em_tree
, em
);
1098 spin_unlock(&em_tree
->lock
);
1099 if (ret
!= -EEXIST
) {
1100 free_extent_map(em
);
1103 btrfs_drop_extent_cache(inode
, em
->start
,
1104 em
->start
+ em
->len
- 1, 0);
1106 type
= BTRFS_ORDERED_PREALLOC
;
1108 type
= BTRFS_ORDERED_NOCOW
;
1111 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1112 num_bytes
, num_bytes
, type
);
1115 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1116 cur_offset
, cur_offset
+ num_bytes
- 1,
1117 locked_page
, 1, 1, 1, 0, 0, 0);
1118 cur_offset
= extent_end
;
1119 if (cur_offset
> end
)
1122 btrfs_release_path(root
, path
);
1124 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1125 cow_start
= cur_offset
;
1126 if (cow_start
!= (u64
)-1) {
1127 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1128 page_started
, nr_written
, 1);
1132 ret
= btrfs_end_transaction(trans
, root
);
1134 btrfs_free_path(path
);
1139 * extent_io.c call back to do delayed allocation processing
1141 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1142 u64 start
, u64 end
, int *page_started
,
1143 unsigned long *nr_written
)
1147 if (btrfs_test_flag(inode
, NODATACOW
))
1148 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1149 page_started
, 1, nr_written
);
1150 else if (btrfs_test_flag(inode
, PREALLOC
))
1151 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1152 page_started
, 0, nr_written
);
1154 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1155 page_started
, nr_written
);
1161 * extent_io.c set_bit_hook, used to track delayed allocation
1162 * bytes in this file, and to maintain the list of inodes that
1163 * have pending delalloc work to be done.
1165 static int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1166 unsigned long old
, unsigned long bits
)
1169 * set_bit and clear bit hooks normally require _irqsave/restore
1170 * but in this case, we are only testeing for the DELALLOC
1171 * bit, which is only set or cleared with irqs on
1173 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1174 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1175 spin_lock(&root
->fs_info
->delalloc_lock
);
1176 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
1177 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
1178 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1179 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1180 &root
->fs_info
->delalloc_inodes
);
1182 spin_unlock(&root
->fs_info
->delalloc_lock
);
1188 * extent_io.c clear_bit_hook, see set_bit_hook for why
1190 static int btrfs_clear_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1191 unsigned long old
, unsigned long bits
)
1194 * set_bit and clear bit hooks normally require _irqsave/restore
1195 * but in this case, we are only testeing for the DELALLOC
1196 * bit, which is only set or cleared with irqs on
1198 if ((old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1199 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1201 spin_lock(&root
->fs_info
->delalloc_lock
);
1202 if (end
- start
+ 1 > root
->fs_info
->delalloc_bytes
) {
1203 printk("warning: delalloc account %Lu %Lu\n",
1204 end
- start
+ 1, root
->fs_info
->delalloc_bytes
);
1205 root
->fs_info
->delalloc_bytes
= 0;
1206 BTRFS_I(inode
)->delalloc_bytes
= 0;
1208 root
->fs_info
->delalloc_bytes
-= end
- start
+ 1;
1209 BTRFS_I(inode
)->delalloc_bytes
-= end
- start
+ 1;
1211 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1212 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1213 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1215 spin_unlock(&root
->fs_info
->delalloc_lock
);
1221 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1222 * we don't create bios that span stripes or chunks
1224 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1225 size_t size
, struct bio
*bio
,
1226 unsigned long bio_flags
)
1228 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1229 struct btrfs_mapping_tree
*map_tree
;
1230 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1235 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1238 length
= bio
->bi_size
;
1239 map_tree
= &root
->fs_info
->mapping_tree
;
1240 map_length
= length
;
1241 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1242 &map_length
, NULL
, 0);
1244 if (map_length
< length
+ size
) {
1251 * in order to insert checksums into the metadata in large chunks,
1252 * we wait until bio submission time. All the pages in the bio are
1253 * checksummed and sums are attached onto the ordered extent record.
1255 * At IO completion time the cums attached on the ordered extent record
1256 * are inserted into the btree
1258 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
, struct bio
*bio
,
1259 int mirror_num
, unsigned long bio_flags
)
1261 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1264 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1270 * in order to insert checksums into the metadata in large chunks,
1271 * we wait until bio submission time. All the pages in the bio are
1272 * checksummed and sums are attached onto the ordered extent record.
1274 * At IO completion time the cums attached on the ordered extent record
1275 * are inserted into the btree
1277 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1278 int mirror_num
, unsigned long bio_flags
)
1280 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1281 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1285 * extent_io.c submission hook. This does the right thing for csum calculation
1286 * on write, or reading the csums from the tree before a read
1288 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1289 int mirror_num
, unsigned long bio_flags
)
1291 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1295 skip_sum
= btrfs_test_flag(inode
, NODATASUM
);
1297 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1300 if (!(rw
& (1 << BIO_RW
))) {
1301 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1302 return btrfs_submit_compressed_read(inode
, bio
,
1303 mirror_num
, bio_flags
);
1304 } else if (!skip_sum
)
1305 btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1307 } else if (!skip_sum
) {
1308 /* csum items have already been cloned */
1309 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1311 /* we're doing a write, do the async checksumming */
1312 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1313 inode
, rw
, bio
, mirror_num
,
1314 bio_flags
, __btrfs_submit_bio_start
,
1315 __btrfs_submit_bio_done
);
1319 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1323 * given a list of ordered sums record them in the inode. This happens
1324 * at IO completion time based on sums calculated at bio submission time.
1326 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1327 struct inode
*inode
, u64 file_offset
,
1328 struct list_head
*list
)
1330 struct list_head
*cur
;
1331 struct btrfs_ordered_sum
*sum
;
1333 btrfs_set_trans_block_group(trans
, inode
);
1334 list_for_each(cur
, list
) {
1335 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
1336 btrfs_csum_file_blocks(trans
,
1337 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1342 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
1344 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0) {
1347 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1351 /* see btrfs_writepage_start_hook for details on why this is required */
1352 struct btrfs_writepage_fixup
{
1354 struct btrfs_work work
;
1357 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1359 struct btrfs_writepage_fixup
*fixup
;
1360 struct btrfs_ordered_extent
*ordered
;
1362 struct inode
*inode
;
1366 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1370 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1371 ClearPageChecked(page
);
1375 inode
= page
->mapping
->host
;
1376 page_start
= page_offset(page
);
1377 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1379 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1381 /* already ordered? We're done */
1382 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1383 EXTENT_ORDERED
, 0)) {
1387 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1389 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
1390 page_end
, GFP_NOFS
);
1392 btrfs_start_ordered_extent(inode
, ordered
, 1);
1396 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
1397 ClearPageChecked(page
);
1399 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1402 page_cache_release(page
);
1406 * There are a few paths in the higher layers of the kernel that directly
1407 * set the page dirty bit without asking the filesystem if it is a
1408 * good idea. This causes problems because we want to make sure COW
1409 * properly happens and the data=ordered rules are followed.
1411 * In our case any range that doesn't have the ORDERED bit set
1412 * hasn't been properly setup for IO. We kick off an async process
1413 * to fix it up. The async helper will wait for ordered extents, set
1414 * the delalloc bit and make it safe to write the page.
1416 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1418 struct inode
*inode
= page
->mapping
->host
;
1419 struct btrfs_writepage_fixup
*fixup
;
1420 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1423 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1428 if (PageChecked(page
))
1431 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1435 SetPageChecked(page
);
1436 page_cache_get(page
);
1437 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1439 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1443 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1444 struct inode
*inode
, u64 file_pos
,
1445 u64 disk_bytenr
, u64 disk_num_bytes
,
1446 u64 num_bytes
, u64 ram_bytes
,
1447 u8 compression
, u8 encryption
,
1448 u16 other_encoding
, int extent_type
)
1450 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1451 struct btrfs_file_extent_item
*fi
;
1452 struct btrfs_path
*path
;
1453 struct extent_buffer
*leaf
;
1454 struct btrfs_key ins
;
1458 path
= btrfs_alloc_path();
1461 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1462 file_pos
+ num_bytes
, file_pos
, &hint
);
1465 ins
.objectid
= inode
->i_ino
;
1466 ins
.offset
= file_pos
;
1467 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1468 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1470 leaf
= path
->nodes
[0];
1471 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1472 struct btrfs_file_extent_item
);
1473 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1474 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1475 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1476 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1477 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1478 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1479 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1480 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1481 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1482 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1483 btrfs_mark_buffer_dirty(leaf
);
1485 inode_add_bytes(inode
, num_bytes
);
1486 btrfs_drop_extent_cache(inode
, file_pos
, file_pos
+ num_bytes
- 1, 0);
1488 ins
.objectid
= disk_bytenr
;
1489 ins
.offset
= disk_num_bytes
;
1490 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1491 ret
= btrfs_alloc_reserved_extent(trans
, root
, leaf
->start
,
1492 root
->root_key
.objectid
,
1493 trans
->transid
, inode
->i_ino
, &ins
);
1496 btrfs_free_path(path
);
1500 /* as ordered data IO finishes, this gets called so we can finish
1501 * an ordered extent if the range of bytes in the file it covers are
1504 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1506 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1507 struct btrfs_trans_handle
*trans
;
1508 struct btrfs_ordered_extent
*ordered_extent
;
1509 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1513 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1517 trans
= btrfs_join_transaction(root
, 1);
1519 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1520 BUG_ON(!ordered_extent
);
1521 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1524 lock_extent(io_tree
, ordered_extent
->file_offset
,
1525 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1528 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1530 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1532 ret
= btrfs_mark_extent_written(trans
, root
, inode
,
1533 ordered_extent
->file_offset
,
1534 ordered_extent
->file_offset
+
1535 ordered_extent
->len
);
1538 ret
= insert_reserved_file_extent(trans
, inode
,
1539 ordered_extent
->file_offset
,
1540 ordered_extent
->start
,
1541 ordered_extent
->disk_len
,
1542 ordered_extent
->len
,
1543 ordered_extent
->len
,
1545 BTRFS_FILE_EXTENT_REG
);
1548 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1549 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1552 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1553 &ordered_extent
->list
);
1555 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1556 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1557 btrfs_update_inode(trans
, root
, inode
);
1558 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1559 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1562 btrfs_put_ordered_extent(ordered_extent
);
1563 /* once for the tree */
1564 btrfs_put_ordered_extent(ordered_extent
);
1566 btrfs_end_transaction(trans
, root
);
1570 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1571 struct extent_state
*state
, int uptodate
)
1573 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1577 * When IO fails, either with EIO or csum verification fails, we
1578 * try other mirrors that might have a good copy of the data. This
1579 * io_failure_record is used to record state as we go through all the
1580 * mirrors. If another mirror has good data, the page is set up to date
1581 * and things continue. If a good mirror can't be found, the original
1582 * bio end_io callback is called to indicate things have failed.
1584 struct io_failure_record
{
1589 unsigned long bio_flags
;
1593 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1594 struct page
*page
, u64 start
, u64 end
,
1595 struct extent_state
*state
)
1597 struct io_failure_record
*failrec
= NULL
;
1599 struct extent_map
*em
;
1600 struct inode
*inode
= page
->mapping
->host
;
1601 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1602 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1609 ret
= get_state_private(failure_tree
, start
, &private);
1611 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1614 failrec
->start
= start
;
1615 failrec
->len
= end
- start
+ 1;
1616 failrec
->last_mirror
= 0;
1617 failrec
->bio_flags
= 0;
1619 spin_lock(&em_tree
->lock
);
1620 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1621 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1622 free_extent_map(em
);
1625 spin_unlock(&em_tree
->lock
);
1627 if (!em
|| IS_ERR(em
)) {
1631 logical
= start
- em
->start
;
1632 logical
= em
->block_start
+ logical
;
1633 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1634 logical
= em
->block_start
;
1635 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1637 failrec
->logical
= logical
;
1638 free_extent_map(em
);
1639 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1640 EXTENT_DIRTY
, GFP_NOFS
);
1641 set_state_private(failure_tree
, start
,
1642 (u64
)(unsigned long)failrec
);
1644 failrec
= (struct io_failure_record
*)(unsigned long)private;
1646 num_copies
= btrfs_num_copies(
1647 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1648 failrec
->logical
, failrec
->len
);
1649 failrec
->last_mirror
++;
1651 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1652 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1655 if (state
&& state
->start
!= failrec
->start
)
1657 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1659 if (!state
|| failrec
->last_mirror
> num_copies
) {
1660 set_state_private(failure_tree
, failrec
->start
, 0);
1661 clear_extent_bits(failure_tree
, failrec
->start
,
1662 failrec
->start
+ failrec
->len
- 1,
1663 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1667 bio
= bio_alloc(GFP_NOFS
, 1);
1668 bio
->bi_private
= state
;
1669 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1670 bio
->bi_sector
= failrec
->logical
>> 9;
1671 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1674 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1675 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1680 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1681 failrec
->last_mirror
,
1682 failrec
->bio_flags
);
1687 * each time an IO finishes, we do a fast check in the IO failure tree
1688 * to see if we need to process or clean up an io_failure_record
1690 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1693 u64 private_failure
;
1694 struct io_failure_record
*failure
;
1698 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1699 (u64
)-1, 1, EXTENT_DIRTY
)) {
1700 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1701 start
, &private_failure
);
1703 failure
= (struct io_failure_record
*)(unsigned long)
1705 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1707 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1709 failure
->start
+ failure
->len
- 1,
1710 EXTENT_DIRTY
| EXTENT_LOCKED
,
1719 * when reads are done, we need to check csums to verify the data is correct
1720 * if there's a match, we allow the bio to finish. If not, we go through
1721 * the io_failure_record routines to find good copies
1723 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1724 struct extent_state
*state
)
1726 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1727 struct inode
*inode
= page
->mapping
->host
;
1728 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1730 u64
private = ~(u32
)0;
1732 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1734 unsigned long flags
;
1736 if (PageChecked(page
)) {
1737 ClearPageChecked(page
);
1740 if (btrfs_test_flag(inode
, NODATASUM
))
1743 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1744 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1)) {
1745 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1750 if (state
&& state
->start
== start
) {
1751 private = state
->private;
1754 ret
= get_state_private(io_tree
, start
, &private);
1756 local_irq_save(flags
);
1757 kaddr
= kmap_atomic(page
, KM_IRQ0
);
1761 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1762 btrfs_csum_final(csum
, (char *)&csum
);
1763 if (csum
!= private) {
1766 kunmap_atomic(kaddr
, KM_IRQ0
);
1767 local_irq_restore(flags
);
1769 /* if the io failure tree for this inode is non-empty,
1770 * check to see if we've recovered from a failed IO
1772 btrfs_clean_io_failures(inode
, start
);
1776 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1777 page
->mapping
->host
->i_ino
, (unsigned long long)start
, csum
,
1779 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1780 flush_dcache_page(page
);
1781 kunmap_atomic(kaddr
, KM_IRQ0
);
1782 local_irq_restore(flags
);
1789 * This creates an orphan entry for the given inode in case something goes
1790 * wrong in the middle of an unlink/truncate.
1792 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1794 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1797 spin_lock(&root
->list_lock
);
1799 /* already on the orphan list, we're good */
1800 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1801 spin_unlock(&root
->list_lock
);
1805 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1807 spin_unlock(&root
->list_lock
);
1810 * insert an orphan item to track this unlinked/truncated file
1812 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1818 * We have done the truncate/delete so we can go ahead and remove the orphan
1819 * item for this particular inode.
1821 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1823 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1826 spin_lock(&root
->list_lock
);
1828 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1829 spin_unlock(&root
->list_lock
);
1833 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1835 spin_unlock(&root
->list_lock
);
1839 spin_unlock(&root
->list_lock
);
1841 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1847 * this cleans up any orphans that may be left on the list from the last use
1850 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1852 struct btrfs_path
*path
;
1853 struct extent_buffer
*leaf
;
1854 struct btrfs_item
*item
;
1855 struct btrfs_key key
, found_key
;
1856 struct btrfs_trans_handle
*trans
;
1857 struct inode
*inode
;
1858 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1860 path
= btrfs_alloc_path();
1865 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1866 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1867 key
.offset
= (u64
)-1;
1871 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1873 printk(KERN_ERR
"Error searching slot for orphan: %d"
1879 * if ret == 0 means we found what we were searching for, which
1880 * is weird, but possible, so only screw with path if we didnt
1881 * find the key and see if we have stuff that matches
1884 if (path
->slots
[0] == 0)
1889 /* pull out the item */
1890 leaf
= path
->nodes
[0];
1891 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1892 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1894 /* make sure the item matches what we want */
1895 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1897 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1900 /* release the path since we're done with it */
1901 btrfs_release_path(root
, path
);
1904 * this is where we are basically btrfs_lookup, without the
1905 * crossing root thing. we store the inode number in the
1906 * offset of the orphan item.
1908 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
1909 found_key
.offset
, root
);
1913 if (inode
->i_state
& I_NEW
) {
1914 BTRFS_I(inode
)->root
= root
;
1916 /* have to set the location manually */
1917 BTRFS_I(inode
)->location
.objectid
= inode
->i_ino
;
1918 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
1919 BTRFS_I(inode
)->location
.offset
= 0;
1921 btrfs_read_locked_inode(inode
);
1922 unlock_new_inode(inode
);
1926 * add this inode to the orphan list so btrfs_orphan_del does
1927 * the proper thing when we hit it
1929 spin_lock(&root
->list_lock
);
1930 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1931 spin_unlock(&root
->list_lock
);
1934 * if this is a bad inode, means we actually succeeded in
1935 * removing the inode, but not the orphan record, which means
1936 * we need to manually delete the orphan since iput will just
1937 * do a destroy_inode
1939 if (is_bad_inode(inode
)) {
1940 trans
= btrfs_start_transaction(root
, 1);
1941 btrfs_orphan_del(trans
, inode
);
1942 btrfs_end_transaction(trans
, root
);
1947 /* if we have links, this was a truncate, lets do that */
1948 if (inode
->i_nlink
) {
1950 btrfs_truncate(inode
);
1955 /* this will do delete_inode and everything for us */
1960 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
1962 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
1964 btrfs_free_path(path
);
1968 * read an inode from the btree into the in-memory inode
1970 void btrfs_read_locked_inode(struct inode
*inode
)
1972 struct btrfs_path
*path
;
1973 struct extent_buffer
*leaf
;
1974 struct btrfs_inode_item
*inode_item
;
1975 struct btrfs_timespec
*tspec
;
1976 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1977 struct btrfs_key location
;
1978 u64 alloc_group_block
;
1982 path
= btrfs_alloc_path();
1984 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
1986 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
1990 leaf
= path
->nodes
[0];
1991 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1992 struct btrfs_inode_item
);
1994 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
1995 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
1996 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
1997 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
1998 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2000 tspec
= btrfs_inode_atime(inode_item
);
2001 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2002 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2004 tspec
= btrfs_inode_mtime(inode_item
);
2005 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2006 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2008 tspec
= btrfs_inode_ctime(inode_item
);
2009 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2010 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2012 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2013 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2014 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2015 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2017 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2019 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2020 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2022 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2023 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2024 alloc_group_block
, 0);
2025 btrfs_free_path(path
);
2028 switch (inode
->i_mode
& S_IFMT
) {
2030 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2031 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2032 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2033 inode
->i_fop
= &btrfs_file_operations
;
2034 inode
->i_op
= &btrfs_file_inode_operations
;
2037 inode
->i_fop
= &btrfs_dir_file_operations
;
2038 if (root
== root
->fs_info
->tree_root
)
2039 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2041 inode
->i_op
= &btrfs_dir_inode_operations
;
2044 inode
->i_op
= &btrfs_symlink_inode_operations
;
2045 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2046 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2049 init_special_inode(inode
, inode
->i_mode
, rdev
);
2055 btrfs_free_path(path
);
2056 make_bad_inode(inode
);
2060 * given a leaf and an inode, copy the inode fields into the leaf
2062 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2063 struct extent_buffer
*leaf
,
2064 struct btrfs_inode_item
*item
,
2065 struct inode
*inode
)
2067 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2068 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2069 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2070 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2071 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2073 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2074 inode
->i_atime
.tv_sec
);
2075 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2076 inode
->i_atime
.tv_nsec
);
2078 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2079 inode
->i_mtime
.tv_sec
);
2080 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2081 inode
->i_mtime
.tv_nsec
);
2083 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2084 inode
->i_ctime
.tv_sec
);
2085 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2086 inode
->i_ctime
.tv_nsec
);
2088 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2089 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2090 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2091 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2092 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2093 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2094 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2098 * copy everything in the in-memory inode into the btree.
2100 int noinline
btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2101 struct btrfs_root
*root
,
2102 struct inode
*inode
)
2104 struct btrfs_inode_item
*inode_item
;
2105 struct btrfs_path
*path
;
2106 struct extent_buffer
*leaf
;
2109 path
= btrfs_alloc_path();
2111 ret
= btrfs_lookup_inode(trans
, root
, path
,
2112 &BTRFS_I(inode
)->location
, 1);
2119 leaf
= path
->nodes
[0];
2120 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2121 struct btrfs_inode_item
);
2123 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2124 btrfs_mark_buffer_dirty(leaf
);
2125 btrfs_set_inode_last_trans(trans
, inode
);
2128 btrfs_free_path(path
);
2134 * unlink helper that gets used here in inode.c and in the tree logging
2135 * recovery code. It remove a link in a directory with a given name, and
2136 * also drops the back refs in the inode to the directory
2138 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2139 struct btrfs_root
*root
,
2140 struct inode
*dir
, struct inode
*inode
,
2141 const char *name
, int name_len
)
2143 struct btrfs_path
*path
;
2145 struct extent_buffer
*leaf
;
2146 struct btrfs_dir_item
*di
;
2147 struct btrfs_key key
;
2150 path
= btrfs_alloc_path();
2156 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2157 name
, name_len
, -1);
2166 leaf
= path
->nodes
[0];
2167 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2168 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2171 btrfs_release_path(root
, path
);
2173 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2175 dir
->i_ino
, &index
);
2177 printk("failed to delete reference to %.*s, "
2178 "inode %lu parent %lu\n", name_len
, name
,
2179 inode
->i_ino
, dir
->i_ino
);
2183 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2184 index
, name
, name_len
, -1);
2193 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2194 btrfs_release_path(root
, path
);
2196 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2198 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2200 BTRFS_I(dir
)->log_dirty_trans
= trans
->transid
;
2202 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2206 btrfs_free_path(path
);
2210 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2211 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2212 btrfs_update_inode(trans
, root
, dir
);
2213 btrfs_drop_nlink(inode
);
2214 ret
= btrfs_update_inode(trans
, root
, inode
);
2215 dir
->i_sb
->s_dirt
= 1;
2220 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2222 struct btrfs_root
*root
;
2223 struct btrfs_trans_handle
*trans
;
2224 struct inode
*inode
= dentry
->d_inode
;
2226 unsigned long nr
= 0;
2228 root
= BTRFS_I(dir
)->root
;
2230 ret
= btrfs_check_free_space(root
, 1, 1);
2234 trans
= btrfs_start_transaction(root
, 1);
2236 btrfs_set_trans_block_group(trans
, dir
);
2237 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2238 dentry
->d_name
.name
, dentry
->d_name
.len
);
2240 if (inode
->i_nlink
== 0)
2241 ret
= btrfs_orphan_add(trans
, inode
);
2243 nr
= trans
->blocks_used
;
2245 btrfs_end_transaction_throttle(trans
, root
);
2247 btrfs_btree_balance_dirty(root
, nr
);
2251 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2253 struct inode
*inode
= dentry
->d_inode
;
2256 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2257 struct btrfs_trans_handle
*trans
;
2258 unsigned long nr
= 0;
2261 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2262 * the root of a subvolume or snapshot
2264 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2265 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
2269 ret
= btrfs_check_free_space(root
, 1, 1);
2273 trans
= btrfs_start_transaction(root
, 1);
2274 btrfs_set_trans_block_group(trans
, dir
);
2276 err
= btrfs_orphan_add(trans
, inode
);
2280 /* now the directory is empty */
2281 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2282 dentry
->d_name
.name
, dentry
->d_name
.len
);
2284 btrfs_i_size_write(inode
, 0);
2288 nr
= trans
->blocks_used
;
2289 ret
= btrfs_end_transaction_throttle(trans
, root
);
2291 btrfs_btree_balance_dirty(root
, nr
);
2300 * when truncating bytes in a file, it is possible to avoid reading
2301 * the leaves that contain only checksum items. This can be the
2302 * majority of the IO required to delete a large file, but it must
2303 * be done carefully.
2305 * The keys in the level just above the leaves are checked to make sure
2306 * the lowest key in a given leaf is a csum key, and starts at an offset
2307 * after the new size.
2309 * Then the key for the next leaf is checked to make sure it also has
2310 * a checksum item for the same file. If it does, we know our target leaf
2311 * contains only checksum items, and it can be safely freed without reading
2314 * This is just an optimization targeted at large files. It may do
2315 * nothing. It will return 0 unless things went badly.
2317 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
2318 struct btrfs_root
*root
,
2319 struct btrfs_path
*path
,
2320 struct inode
*inode
, u64 new_size
)
2322 struct btrfs_key key
;
2325 struct btrfs_key found_key
;
2326 struct btrfs_key other_key
;
2327 struct btrfs_leaf_ref
*ref
;
2331 path
->lowest_level
= 1;
2332 key
.objectid
= inode
->i_ino
;
2333 key
.type
= BTRFS_CSUM_ITEM_KEY
;
2334 key
.offset
= new_size
;
2336 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2340 if (path
->nodes
[1] == NULL
) {
2345 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
2346 nritems
= btrfs_header_nritems(path
->nodes
[1]);
2351 if (path
->slots
[1] >= nritems
)
2354 /* did we find a key greater than anything we want to delete? */
2355 if (found_key
.objectid
> inode
->i_ino
||
2356 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
2359 /* we check the next key in the node to make sure the leave contains
2360 * only checksum items. This comparison doesn't work if our
2361 * leaf is the last one in the node
2363 if (path
->slots
[1] + 1 >= nritems
) {
2365 /* search forward from the last key in the node, this
2366 * will bring us into the next node in the tree
2368 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
2370 /* unlikely, but we inc below, so check to be safe */
2371 if (found_key
.offset
== (u64
)-1)
2374 /* search_forward needs a path with locks held, do the
2375 * search again for the original key. It is possible
2376 * this will race with a balance and return a path that
2377 * we could modify, but this drop is just an optimization
2378 * and is allowed to miss some leaves.
2380 btrfs_release_path(root
, path
);
2383 /* setup a max key for search_forward */
2384 other_key
.offset
= (u64
)-1;
2385 other_key
.type
= key
.type
;
2386 other_key
.objectid
= key
.objectid
;
2388 path
->keep_locks
= 1;
2389 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
2391 path
->keep_locks
= 0;
2392 if (ret
|| found_key
.objectid
!= key
.objectid
||
2393 found_key
.type
!= key
.type
) {
2398 key
.offset
= found_key
.offset
;
2399 btrfs_release_path(root
, path
);
2404 /* we know there's one more slot after us in the tree,
2405 * read that key so we can verify it is also a checksum item
2407 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
2409 if (found_key
.objectid
< inode
->i_ino
)
2412 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
2416 * if the key for the next leaf isn't a csum key from this objectid,
2417 * we can't be sure there aren't good items inside this leaf.
2420 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
2423 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
2424 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
2426 * it is safe to delete this leaf, it contains only
2427 * csum items from this inode at an offset >= new_size
2429 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
2432 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
2433 ref
= btrfs_alloc_leaf_ref(root
, 0);
2435 ref
->root_gen
= root
->root_key
.offset
;
2436 ref
->bytenr
= leaf_start
;
2438 ref
->generation
= leaf_gen
;
2441 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
2443 btrfs_free_leaf_ref(root
, ref
);
2449 btrfs_release_path(root
, path
);
2451 if (other_key
.objectid
== inode
->i_ino
&&
2452 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
2453 key
.offset
= other_key
.offset
;
2459 /* fixup any changes we've made to the path */
2460 path
->lowest_level
= 0;
2461 path
->keep_locks
= 0;
2462 btrfs_release_path(root
, path
);
2469 * this can truncate away extent items, csum items and directory items.
2470 * It starts at a high offset and removes keys until it can't find
2471 * any higher than new_size
2473 * csum items that cross the new i_size are truncated to the new size
2476 * min_type is the minimum key type to truncate down to. If set to 0, this
2477 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2479 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2480 struct btrfs_root
*root
,
2481 struct inode
*inode
,
2482 u64 new_size
, u32 min_type
)
2485 struct btrfs_path
*path
;
2486 struct btrfs_key key
;
2487 struct btrfs_key found_key
;
2489 struct extent_buffer
*leaf
;
2490 struct btrfs_file_extent_item
*fi
;
2491 u64 extent_start
= 0;
2492 u64 extent_num_bytes
= 0;
2498 int pending_del_nr
= 0;
2499 int pending_del_slot
= 0;
2500 int extent_type
= -1;
2502 u64 mask
= root
->sectorsize
- 1;
2505 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2506 path
= btrfs_alloc_path();
2510 /* FIXME, add redo link to tree so we don't leak on crash */
2511 key
.objectid
= inode
->i_ino
;
2512 key
.offset
= (u64
)-1;
2515 btrfs_init_path(path
);
2518 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2523 /* there are no items in the tree for us to truncate, we're
2526 if (path
->slots
[0] == 0) {
2535 leaf
= path
->nodes
[0];
2536 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2537 found_type
= btrfs_key_type(&found_key
);
2540 if (found_key
.objectid
!= inode
->i_ino
)
2543 if (found_type
< min_type
)
2546 item_end
= found_key
.offset
;
2547 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2548 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2549 struct btrfs_file_extent_item
);
2550 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2551 encoding
= btrfs_file_extent_compression(leaf
, fi
);
2552 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
2553 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
2555 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2557 btrfs_file_extent_num_bytes(leaf
, fi
);
2558 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2559 item_end
+= btrfs_file_extent_inline_len(leaf
,
2564 if (item_end
< new_size
) {
2565 if (found_type
== BTRFS_DIR_ITEM_KEY
) {
2566 found_type
= BTRFS_INODE_ITEM_KEY
;
2567 } else if (found_type
== BTRFS_EXTENT_ITEM_KEY
) {
2568 found_type
= BTRFS_EXTENT_DATA_KEY
;
2569 } else if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2570 found_type
= BTRFS_XATTR_ITEM_KEY
;
2571 } else if (found_type
== BTRFS_XATTR_ITEM_KEY
) {
2572 found_type
= BTRFS_INODE_REF_KEY
;
2573 } else if (found_type
) {
2578 btrfs_set_key_type(&key
, found_type
);
2581 if (found_key
.offset
>= new_size
)
2587 /* FIXME, shrink the extent if the ref count is only 1 */
2588 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2591 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2593 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2594 if (!del_item
&& !encoding
) {
2595 u64 orig_num_bytes
=
2596 btrfs_file_extent_num_bytes(leaf
, fi
);
2597 extent_num_bytes
= new_size
-
2598 found_key
.offset
+ root
->sectorsize
- 1;
2599 extent_num_bytes
= extent_num_bytes
&
2600 ~((u64
)root
->sectorsize
- 1);
2601 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2603 num_dec
= (orig_num_bytes
-
2605 if (root
->ref_cows
&& extent_start
!= 0)
2606 inode_sub_bytes(inode
, num_dec
);
2607 btrfs_mark_buffer_dirty(leaf
);
2610 btrfs_file_extent_disk_num_bytes(leaf
,
2612 /* FIXME blocksize != 4096 */
2613 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2614 if (extent_start
!= 0) {
2617 inode_sub_bytes(inode
, num_dec
);
2619 root_gen
= btrfs_header_generation(leaf
);
2620 root_owner
= btrfs_header_owner(leaf
);
2622 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2624 * we can't truncate inline items that have had
2628 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2629 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2630 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2631 u32 size
= new_size
- found_key
.offset
;
2633 if (root
->ref_cows
) {
2634 inode_sub_bytes(inode
, item_end
+ 1 -
2638 btrfs_file_extent_calc_inline_size(size
);
2639 ret
= btrfs_truncate_item(trans
, root
, path
,
2642 } else if (root
->ref_cows
) {
2643 inode_sub_bytes(inode
, item_end
+ 1 -
2649 if (!pending_del_nr
) {
2650 /* no pending yet, add ourselves */
2651 pending_del_slot
= path
->slots
[0];
2653 } else if (pending_del_nr
&&
2654 path
->slots
[0] + 1 == pending_del_slot
) {
2655 /* hop on the pending chunk */
2657 pending_del_slot
= path
->slots
[0];
2659 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path
->slots
[0], pending_del_nr
, pending_del_slot
);
2665 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2667 leaf
->start
, root_owner
,
2668 root_gen
, inode
->i_ino
, 0);
2672 if (path
->slots
[0] == 0) {
2675 btrfs_release_path(root
, path
);
2680 if (pending_del_nr
&&
2681 path
->slots
[0] + 1 != pending_del_slot
) {
2682 struct btrfs_key debug
;
2684 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2686 ret
= btrfs_del_items(trans
, root
, path
,
2691 btrfs_release_path(root
, path
);
2697 if (pending_del_nr
) {
2698 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2701 btrfs_free_path(path
);
2702 inode
->i_sb
->s_dirt
= 1;
2707 * taken from block_truncate_page, but does cow as it zeros out
2708 * any bytes left in the last page in the file.
2710 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2712 struct inode
*inode
= mapping
->host
;
2713 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2714 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2715 struct btrfs_ordered_extent
*ordered
;
2717 u32 blocksize
= root
->sectorsize
;
2718 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2719 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2725 if ((offset
& (blocksize
- 1)) == 0)
2730 page
= grab_cache_page(mapping
, index
);
2734 page_start
= page_offset(page
);
2735 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2737 if (!PageUptodate(page
)) {
2738 ret
= btrfs_readpage(NULL
, page
);
2740 if (page
->mapping
!= mapping
) {
2742 page_cache_release(page
);
2745 if (!PageUptodate(page
)) {
2750 wait_on_page_writeback(page
);
2752 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2753 set_page_extent_mapped(page
);
2755 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2757 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2759 page_cache_release(page
);
2760 btrfs_start_ordered_extent(inode
, ordered
, 1);
2761 btrfs_put_ordered_extent(ordered
);
2765 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2767 if (offset
!= PAGE_CACHE_SIZE
) {
2769 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2770 flush_dcache_page(page
);
2773 ClearPageChecked(page
);
2774 set_page_dirty(page
);
2775 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2779 page_cache_release(page
);
2784 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
2786 struct btrfs_trans_handle
*trans
;
2787 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2788 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2789 struct extent_map
*em
;
2790 u64 mask
= root
->sectorsize
- 1;
2791 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2792 u64 block_end
= (size
+ mask
) & ~mask
;
2798 if (size
<= hole_start
)
2801 err
= btrfs_check_free_space(root
, 1, 0);
2805 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2808 struct btrfs_ordered_extent
*ordered
;
2809 btrfs_wait_ordered_range(inode
, hole_start
,
2810 block_end
- hole_start
);
2811 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2812 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2815 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2816 btrfs_put_ordered_extent(ordered
);
2819 trans
= btrfs_start_transaction(root
, 1);
2820 btrfs_set_trans_block_group(trans
, inode
);
2822 cur_offset
= hole_start
;
2824 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2825 block_end
- cur_offset
, 0);
2826 BUG_ON(IS_ERR(em
) || !em
);
2827 last_byte
= min(extent_map_end(em
), block_end
);
2828 last_byte
= (last_byte
+ mask
) & ~mask
;
2829 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2831 hole_size
= last_byte
- cur_offset
;
2832 err
= btrfs_drop_extents(trans
, root
, inode
,
2834 cur_offset
+ hole_size
,
2835 cur_offset
, &hint_byte
);
2838 err
= btrfs_insert_file_extent(trans
, root
,
2839 inode
->i_ino
, cur_offset
, 0,
2840 0, hole_size
, 0, hole_size
,
2842 btrfs_drop_extent_cache(inode
, hole_start
,
2845 free_extent_map(em
);
2846 cur_offset
= last_byte
;
2847 if (err
|| cur_offset
>= block_end
)
2851 btrfs_end_transaction(trans
, root
);
2852 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2856 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2858 struct inode
*inode
= dentry
->d_inode
;
2861 err
= inode_change_ok(inode
, attr
);
2865 if (S_ISREG(inode
->i_mode
) &&
2866 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
> inode
->i_size
) {
2867 err
= btrfs_cont_expand(inode
, attr
->ia_size
);
2872 err
= inode_setattr(inode
, attr
);
2874 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2875 err
= btrfs_acl_chmod(inode
);
2879 void btrfs_delete_inode(struct inode
*inode
)
2881 struct btrfs_trans_handle
*trans
;
2882 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2886 truncate_inode_pages(&inode
->i_data
, 0);
2887 if (is_bad_inode(inode
)) {
2888 btrfs_orphan_del(NULL
, inode
);
2891 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
2893 btrfs_i_size_write(inode
, 0);
2894 trans
= btrfs_start_transaction(root
, 1);
2896 btrfs_set_trans_block_group(trans
, inode
);
2897 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
2899 btrfs_orphan_del(NULL
, inode
);
2900 goto no_delete_lock
;
2903 btrfs_orphan_del(trans
, inode
);
2905 nr
= trans
->blocks_used
;
2908 btrfs_end_transaction(trans
, root
);
2909 btrfs_btree_balance_dirty(root
, nr
);
2913 nr
= trans
->blocks_used
;
2914 btrfs_end_transaction(trans
, root
);
2915 btrfs_btree_balance_dirty(root
, nr
);
2921 * this returns the key found in the dir entry in the location pointer.
2922 * If no dir entries were found, location->objectid is 0.
2924 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
2925 struct btrfs_key
*location
)
2927 const char *name
= dentry
->d_name
.name
;
2928 int namelen
= dentry
->d_name
.len
;
2929 struct btrfs_dir_item
*di
;
2930 struct btrfs_path
*path
;
2931 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2934 path
= btrfs_alloc_path();
2937 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
2941 if (!di
|| IS_ERR(di
)) {
2944 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
2946 btrfs_free_path(path
);
2949 location
->objectid
= 0;
2954 * when we hit a tree root in a directory, the btrfs part of the inode
2955 * needs to be changed to reflect the root directory of the tree root. This
2956 * is kind of like crossing a mount point.
2958 static int fixup_tree_root_location(struct btrfs_root
*root
,
2959 struct btrfs_key
*location
,
2960 struct btrfs_root
**sub_root
,
2961 struct dentry
*dentry
)
2963 struct btrfs_root_item
*ri
;
2965 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
2967 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
2970 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
2971 dentry
->d_name
.name
,
2972 dentry
->d_name
.len
);
2973 if (IS_ERR(*sub_root
))
2974 return PTR_ERR(*sub_root
);
2976 ri
= &(*sub_root
)->root_item
;
2977 location
->objectid
= btrfs_root_dirid(ri
);
2978 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2979 location
->offset
= 0;
2984 static noinline
void init_btrfs_i(struct inode
*inode
)
2986 struct btrfs_inode
*bi
= BTRFS_I(inode
);
2989 bi
->i_default_acl
= NULL
;
2994 bi
->logged_trans
= 0;
2995 bi
->delalloc_bytes
= 0;
2996 bi
->disk_i_size
= 0;
2998 bi
->index_cnt
= (u64
)-1;
2999 bi
->log_dirty_trans
= 0;
3000 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
3001 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
3002 inode
->i_mapping
, GFP_NOFS
);
3003 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
3004 inode
->i_mapping
, GFP_NOFS
);
3005 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
3006 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
3007 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
3008 mutex_init(&BTRFS_I(inode
)->log_mutex
);
3011 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3013 struct btrfs_iget_args
*args
= p
;
3014 inode
->i_ino
= args
->ino
;
3015 init_btrfs_i(inode
);
3016 BTRFS_I(inode
)->root
= args
->root
;
3020 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3022 struct btrfs_iget_args
*args
= opaque
;
3023 return (args
->ino
== inode
->i_ino
&&
3024 args
->root
== BTRFS_I(inode
)->root
);
3027 struct inode
*btrfs_ilookup(struct super_block
*s
, u64 objectid
,
3028 struct btrfs_root
*root
, int wait
)
3030 struct inode
*inode
;
3031 struct btrfs_iget_args args
;
3032 args
.ino
= objectid
;
3036 inode
= ilookup5(s
, objectid
, btrfs_find_actor
,
3039 inode
= ilookup5_nowait(s
, objectid
, btrfs_find_actor
,
3045 struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 objectid
,
3046 struct btrfs_root
*root
)
3048 struct inode
*inode
;
3049 struct btrfs_iget_args args
;
3050 args
.ino
= objectid
;
3053 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3054 btrfs_init_locked_inode
,
3059 /* Get an inode object given its location and corresponding root.
3060 * Returns in *is_new if the inode was read from disk
3062 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3063 struct btrfs_root
*root
, int *is_new
)
3065 struct inode
*inode
;
3067 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3069 return ERR_PTR(-EACCES
);
3071 if (inode
->i_state
& I_NEW
) {
3072 BTRFS_I(inode
)->root
= root
;
3073 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3074 btrfs_read_locked_inode(inode
);
3075 unlock_new_inode(inode
);
3086 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3088 struct inode
* inode
;
3089 struct btrfs_inode
*bi
= BTRFS_I(dir
);
3090 struct btrfs_root
*root
= bi
->root
;
3091 struct btrfs_root
*sub_root
= root
;
3092 struct btrfs_key location
;
3095 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3096 return ERR_PTR(-ENAMETOOLONG
);
3098 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3101 return ERR_PTR(ret
);
3104 if (location
.objectid
) {
3105 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
3108 return ERR_PTR(ret
);
3110 return ERR_PTR(-ENOENT
);
3111 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, &new);
3113 return ERR_CAST(inode
);
3118 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
3119 struct nameidata
*nd
)
3121 struct inode
*inode
;
3123 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3124 return ERR_PTR(-ENAMETOOLONG
);
3126 inode
= btrfs_lookup_dentry(dir
, dentry
);
3128 return ERR_CAST(inode
);
3130 return d_splice_alias(inode
, dentry
);
3133 static unsigned char btrfs_filetype_table
[] = {
3134 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
3137 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
3140 struct inode
*inode
= filp
->f_dentry
->d_inode
;
3141 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3142 struct btrfs_item
*item
;
3143 struct btrfs_dir_item
*di
;
3144 struct btrfs_key key
;
3145 struct btrfs_key found_key
;
3146 struct btrfs_path
*path
;
3149 struct extent_buffer
*leaf
;
3152 unsigned char d_type
;
3157 int key_type
= BTRFS_DIR_INDEX_KEY
;
3162 /* FIXME, use a real flag for deciding about the key type */
3163 if (root
->fs_info
->tree_root
== root
)
3164 key_type
= BTRFS_DIR_ITEM_KEY
;
3166 /* special case for "." */
3167 if (filp
->f_pos
== 0) {
3168 over
= filldir(dirent
, ".", 1,
3175 /* special case for .., just use the back ref */
3176 if (filp
->f_pos
== 1) {
3177 u64 pino
= parent_ino(filp
->f_path
.dentry
);
3178 over
= filldir(dirent
, "..", 2,
3184 path
= btrfs_alloc_path();
3187 btrfs_set_key_type(&key
, key_type
);
3188 key
.offset
= filp
->f_pos
;
3189 key
.objectid
= inode
->i_ino
;
3191 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3197 leaf
= path
->nodes
[0];
3198 nritems
= btrfs_header_nritems(leaf
);
3199 slot
= path
->slots
[0];
3200 if (advance
|| slot
>= nritems
) {
3201 if (slot
>= nritems
- 1) {
3202 ret
= btrfs_next_leaf(root
, path
);
3205 leaf
= path
->nodes
[0];
3206 nritems
= btrfs_header_nritems(leaf
);
3207 slot
= path
->slots
[0];
3215 item
= btrfs_item_nr(leaf
, slot
);
3216 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3218 if (found_key
.objectid
!= key
.objectid
)
3220 if (btrfs_key_type(&found_key
) != key_type
)
3222 if (found_key
.offset
< filp
->f_pos
)
3225 filp
->f_pos
= found_key
.offset
;
3227 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
3229 di_total
= btrfs_item_size(leaf
, item
);
3231 while (di_cur
< di_total
) {
3232 struct btrfs_key location
;
3234 name_len
= btrfs_dir_name_len(leaf
, di
);
3235 if (name_len
<= sizeof(tmp_name
)) {
3236 name_ptr
= tmp_name
;
3238 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
3244 read_extent_buffer(leaf
, name_ptr
,
3245 (unsigned long)(di
+ 1), name_len
);
3247 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
3248 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
3250 /* is this a reference to our own snapshot? If so
3253 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
3254 location
.objectid
== root
->root_key
.objectid
) {
3258 over
= filldir(dirent
, name_ptr
, name_len
,
3259 found_key
.offset
, location
.objectid
,
3263 if (name_ptr
!= tmp_name
)
3268 di_len
= btrfs_dir_name_len(leaf
, di
) +
3269 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
3271 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
3275 /* Reached end of directory/root. Bump pos past the last item. */
3276 if (key_type
== BTRFS_DIR_INDEX_KEY
)
3277 filp
->f_pos
= INT_LIMIT(typeof(filp
->f_pos
));
3283 btrfs_free_path(path
);
3287 int btrfs_write_inode(struct inode
*inode
, int wait
)
3289 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3290 struct btrfs_trans_handle
*trans
;
3293 if (root
->fs_info
->btree_inode
== inode
)
3297 trans
= btrfs_join_transaction(root
, 1);
3298 btrfs_set_trans_block_group(trans
, inode
);
3299 ret
= btrfs_commit_transaction(trans
, root
);
3305 * This is somewhat expensive, updating the tree every time the
3306 * inode changes. But, it is most likely to find the inode in cache.
3307 * FIXME, needs more benchmarking...there are no reasons other than performance
3308 * to keep or drop this code.
3310 void btrfs_dirty_inode(struct inode
*inode
)
3312 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3313 struct btrfs_trans_handle
*trans
;
3315 trans
= btrfs_join_transaction(root
, 1);
3316 btrfs_set_trans_block_group(trans
, inode
);
3317 btrfs_update_inode(trans
, root
, inode
);
3318 btrfs_end_transaction(trans
, root
);
3322 * find the highest existing sequence number in a directory
3323 * and then set the in-memory index_cnt variable to reflect
3324 * free sequence numbers
3326 static int btrfs_set_inode_index_count(struct inode
*inode
)
3328 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3329 struct btrfs_key key
, found_key
;
3330 struct btrfs_path
*path
;
3331 struct extent_buffer
*leaf
;
3334 key
.objectid
= inode
->i_ino
;
3335 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
3336 key
.offset
= (u64
)-1;
3338 path
= btrfs_alloc_path();
3342 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3345 /* FIXME: we should be able to handle this */
3351 * MAGIC NUMBER EXPLANATION:
3352 * since we search a directory based on f_pos we have to start at 2
3353 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3354 * else has to start at 2
3356 if (path
->slots
[0] == 0) {
3357 BTRFS_I(inode
)->index_cnt
= 2;
3363 leaf
= path
->nodes
[0];
3364 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3366 if (found_key
.objectid
!= inode
->i_ino
||
3367 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
3368 BTRFS_I(inode
)->index_cnt
= 2;
3372 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
3374 btrfs_free_path(path
);
3379 * helper to find a free sequence number in a given directory. This current
3380 * code is very simple, later versions will do smarter things in the btree
3382 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
3386 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
3387 ret
= btrfs_set_inode_index_count(dir
);
3393 *index
= BTRFS_I(dir
)->index_cnt
;
3394 BTRFS_I(dir
)->index_cnt
++;
3399 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
3400 struct btrfs_root
*root
,
3402 const char *name
, int name_len
,
3403 u64 ref_objectid
, u64 objectid
,
3404 u64 alloc_hint
, int mode
, u64
*index
)
3406 struct inode
*inode
;
3407 struct btrfs_inode_item
*inode_item
;
3408 struct btrfs_key
*location
;
3409 struct btrfs_path
*path
;
3410 struct btrfs_inode_ref
*ref
;
3411 struct btrfs_key key
[2];
3417 path
= btrfs_alloc_path();
3420 inode
= new_inode(root
->fs_info
->sb
);
3422 return ERR_PTR(-ENOMEM
);
3425 ret
= btrfs_set_inode_index(dir
, index
);
3427 return ERR_PTR(ret
);
3430 * index_cnt is ignored for everything but a dir,
3431 * btrfs_get_inode_index_count has an explanation for the magic
3434 init_btrfs_i(inode
);
3435 BTRFS_I(inode
)->index_cnt
= 2;
3436 BTRFS_I(inode
)->root
= root
;
3437 BTRFS_I(inode
)->generation
= trans
->transid
;
3443 BTRFS_I(inode
)->block_group
=
3444 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
3445 if ((mode
& S_IFREG
)) {
3446 if (btrfs_test_opt(root
, NODATASUM
))
3447 btrfs_set_flag(inode
, NODATASUM
);
3448 if (btrfs_test_opt(root
, NODATACOW
))
3449 btrfs_set_flag(inode
, NODATACOW
);
3452 key
[0].objectid
= objectid
;
3453 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
3456 key
[1].objectid
= objectid
;
3457 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
3458 key
[1].offset
= ref_objectid
;
3460 sizes
[0] = sizeof(struct btrfs_inode_item
);
3461 sizes
[1] = name_len
+ sizeof(*ref
);
3463 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
3467 if (objectid
> root
->highest_inode
)
3468 root
->highest_inode
= objectid
;
3470 inode
->i_uid
= current_fsuid();
3471 inode
->i_gid
= current_fsgid();
3472 inode
->i_mode
= mode
;
3473 inode
->i_ino
= objectid
;
3474 inode_set_bytes(inode
, 0);
3475 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3476 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3477 struct btrfs_inode_item
);
3478 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
3480 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3481 struct btrfs_inode_ref
);
3482 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
3483 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
3484 ptr
= (unsigned long)(ref
+ 1);
3485 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
3487 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3488 btrfs_free_path(path
);
3490 location
= &BTRFS_I(inode
)->location
;
3491 location
->objectid
= objectid
;
3492 location
->offset
= 0;
3493 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3495 insert_inode_hash(inode
);
3499 BTRFS_I(dir
)->index_cnt
--;
3500 btrfs_free_path(path
);
3501 return ERR_PTR(ret
);
3504 static inline u8
btrfs_inode_type(struct inode
*inode
)
3506 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
3510 * utility function to add 'inode' into 'parent_inode' with
3511 * a give name and a given sequence number.
3512 * if 'add_backref' is true, also insert a backref from the
3513 * inode to the parent directory.
3515 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3516 struct inode
*parent_inode
, struct inode
*inode
,
3517 const char *name
, int name_len
, int add_backref
, u64 index
)
3520 struct btrfs_key key
;
3521 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3523 key
.objectid
= inode
->i_ino
;
3524 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3527 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3528 parent_inode
->i_ino
,
3529 &key
, btrfs_inode_type(inode
),
3533 ret
= btrfs_insert_inode_ref(trans
, root
,
3536 parent_inode
->i_ino
,
3539 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3541 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3542 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3547 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3548 struct dentry
*dentry
, struct inode
*inode
,
3549 int backref
, u64 index
)
3551 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3552 inode
, dentry
->d_name
.name
,
3553 dentry
->d_name
.len
, backref
, index
);
3555 d_instantiate(dentry
, inode
);
3563 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3564 int mode
, dev_t rdev
)
3566 struct btrfs_trans_handle
*trans
;
3567 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3568 struct inode
*inode
= NULL
;
3572 unsigned long nr
= 0;
3575 if (!new_valid_dev(rdev
))
3578 err
= btrfs_check_free_space(root
, 1, 0);
3582 trans
= btrfs_start_transaction(root
, 1);
3583 btrfs_set_trans_block_group(trans
, dir
);
3585 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3591 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3593 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3594 BTRFS_I(dir
)->block_group
, mode
, &index
);
3595 err
= PTR_ERR(inode
);
3599 err
= btrfs_init_acl(inode
, dir
);
3605 btrfs_set_trans_block_group(trans
, inode
);
3606 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3610 inode
->i_op
= &btrfs_special_inode_operations
;
3611 init_special_inode(inode
, inode
->i_mode
, rdev
);
3612 btrfs_update_inode(trans
, root
, inode
);
3614 dir
->i_sb
->s_dirt
= 1;
3615 btrfs_update_inode_block_group(trans
, inode
);
3616 btrfs_update_inode_block_group(trans
, dir
);
3618 nr
= trans
->blocks_used
;
3619 btrfs_end_transaction_throttle(trans
, root
);
3622 inode_dec_link_count(inode
);
3625 btrfs_btree_balance_dirty(root
, nr
);
3629 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3630 int mode
, struct nameidata
*nd
)
3632 struct btrfs_trans_handle
*trans
;
3633 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3634 struct inode
*inode
= NULL
;
3637 unsigned long nr
= 0;
3641 err
= btrfs_check_free_space(root
, 1, 0);
3644 trans
= btrfs_start_transaction(root
, 1);
3645 btrfs_set_trans_block_group(trans
, dir
);
3647 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3653 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3655 dentry
->d_parent
->d_inode
->i_ino
,
3656 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3658 err
= PTR_ERR(inode
);
3662 err
= btrfs_init_acl(inode
, dir
);
3668 btrfs_set_trans_block_group(trans
, inode
);
3669 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3673 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3674 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3675 inode
->i_fop
= &btrfs_file_operations
;
3676 inode
->i_op
= &btrfs_file_inode_operations
;
3677 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3679 dir
->i_sb
->s_dirt
= 1;
3680 btrfs_update_inode_block_group(trans
, inode
);
3681 btrfs_update_inode_block_group(trans
, dir
);
3683 nr
= trans
->blocks_used
;
3684 btrfs_end_transaction_throttle(trans
, root
);
3687 inode_dec_link_count(inode
);
3690 btrfs_btree_balance_dirty(root
, nr
);
3694 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3695 struct dentry
*dentry
)
3697 struct btrfs_trans_handle
*trans
;
3698 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3699 struct inode
*inode
= old_dentry
->d_inode
;
3701 unsigned long nr
= 0;
3705 if (inode
->i_nlink
== 0)
3708 btrfs_inc_nlink(inode
);
3709 err
= btrfs_check_free_space(root
, 1, 0);
3712 err
= btrfs_set_inode_index(dir
, &index
);
3716 trans
= btrfs_start_transaction(root
, 1);
3718 btrfs_set_trans_block_group(trans
, dir
);
3719 atomic_inc(&inode
->i_count
);
3721 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3726 dir
->i_sb
->s_dirt
= 1;
3727 btrfs_update_inode_block_group(trans
, dir
);
3728 err
= btrfs_update_inode(trans
, root
, inode
);
3733 nr
= trans
->blocks_used
;
3734 btrfs_end_transaction_throttle(trans
, root
);
3737 inode_dec_link_count(inode
);
3740 btrfs_btree_balance_dirty(root
, nr
);
3744 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3746 struct inode
*inode
= NULL
;
3747 struct btrfs_trans_handle
*trans
;
3748 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3750 int drop_on_err
= 0;
3753 unsigned long nr
= 1;
3755 err
= btrfs_check_free_space(root
, 1, 0);
3759 trans
= btrfs_start_transaction(root
, 1);
3760 btrfs_set_trans_block_group(trans
, dir
);
3762 if (IS_ERR(trans
)) {
3763 err
= PTR_ERR(trans
);
3767 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3773 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3775 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3776 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3778 if (IS_ERR(inode
)) {
3779 err
= PTR_ERR(inode
);
3785 err
= btrfs_init_acl(inode
, dir
);
3789 inode
->i_op
= &btrfs_dir_inode_operations
;
3790 inode
->i_fop
= &btrfs_dir_file_operations
;
3791 btrfs_set_trans_block_group(trans
, inode
);
3793 btrfs_i_size_write(inode
, 0);
3794 err
= btrfs_update_inode(trans
, root
, inode
);
3798 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3799 inode
, dentry
->d_name
.name
,
3800 dentry
->d_name
.len
, 0, index
);
3804 d_instantiate(dentry
, inode
);
3806 dir
->i_sb
->s_dirt
= 1;
3807 btrfs_update_inode_block_group(trans
, inode
);
3808 btrfs_update_inode_block_group(trans
, dir
);
3811 nr
= trans
->blocks_used
;
3812 btrfs_end_transaction_throttle(trans
, root
);
3817 btrfs_btree_balance_dirty(root
, nr
);
3821 /* helper for btfs_get_extent. Given an existing extent in the tree,
3822 * and an extent that you want to insert, deal with overlap and insert
3823 * the new extent into the tree.
3825 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3826 struct extent_map
*existing
,
3827 struct extent_map
*em
,
3828 u64 map_start
, u64 map_len
)
3832 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3833 start_diff
= map_start
- em
->start
;
3834 em
->start
= map_start
;
3836 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3837 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3838 em
->block_start
+= start_diff
;
3839 em
->block_len
-= start_diff
;
3841 return add_extent_mapping(em_tree
, em
);
3844 static noinline
int uncompress_inline(struct btrfs_path
*path
,
3845 struct inode
*inode
, struct page
*page
,
3846 size_t pg_offset
, u64 extent_offset
,
3847 struct btrfs_file_extent_item
*item
)
3850 struct extent_buffer
*leaf
= path
->nodes
[0];
3853 unsigned long inline_size
;
3856 WARN_ON(pg_offset
!= 0);
3857 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
3858 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
3859 btrfs_item_nr(leaf
, path
->slots
[0]));
3860 tmp
= kmalloc(inline_size
, GFP_NOFS
);
3861 ptr
= btrfs_file_extent_inline_start(item
);
3863 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
3865 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
3866 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
3867 inline_size
, max_size
);
3869 char *kaddr
= kmap_atomic(page
, KM_USER0
);
3870 unsigned long copy_size
= min_t(u64
,
3871 PAGE_CACHE_SIZE
- pg_offset
,
3872 max_size
- extent_offset
);
3873 memset(kaddr
+ pg_offset
, 0, copy_size
);
3874 kunmap_atomic(kaddr
, KM_USER0
);
3881 * a bit scary, this does extent mapping from logical file offset to the disk.
3882 * the ugly parts come from merging extents from the disk with the
3883 * in-ram representation. This gets more complex because of the data=ordered code,
3884 * where the in-ram extents might be locked pending data=ordered completion.
3886 * This also copies inline extents directly into the page.
3888 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
3889 size_t pg_offset
, u64 start
, u64 len
,
3895 u64 extent_start
= 0;
3897 u64 objectid
= inode
->i_ino
;
3899 struct btrfs_path
*path
= NULL
;
3900 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3901 struct btrfs_file_extent_item
*item
;
3902 struct extent_buffer
*leaf
;
3903 struct btrfs_key found_key
;
3904 struct extent_map
*em
= NULL
;
3905 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3906 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3907 struct btrfs_trans_handle
*trans
= NULL
;
3911 spin_lock(&em_tree
->lock
);
3912 em
= lookup_extent_mapping(em_tree
, start
, len
);
3914 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3915 spin_unlock(&em_tree
->lock
);
3918 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
3919 free_extent_map(em
);
3920 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
3921 free_extent_map(em
);
3925 em
= alloc_extent_map(GFP_NOFS
);
3930 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3931 em
->start
= EXTENT_MAP_HOLE
;
3932 em
->orig_start
= EXTENT_MAP_HOLE
;
3934 em
->block_len
= (u64
)-1;
3937 path
= btrfs_alloc_path();
3941 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3942 objectid
, start
, trans
!= NULL
);
3949 if (path
->slots
[0] == 0)
3954 leaf
= path
->nodes
[0];
3955 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3956 struct btrfs_file_extent_item
);
3957 /* are we inside the extent that was found? */
3958 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3959 found_type
= btrfs_key_type(&found_key
);
3960 if (found_key
.objectid
!= objectid
||
3961 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3965 found_type
= btrfs_file_extent_type(leaf
, item
);
3966 extent_start
= found_key
.offset
;
3967 compressed
= btrfs_file_extent_compression(leaf
, item
);
3968 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3969 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3970 extent_end
= extent_start
+
3971 btrfs_file_extent_num_bytes(leaf
, item
);
3972 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3974 size
= btrfs_file_extent_inline_len(leaf
, item
);
3975 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
3976 ~((u64
)root
->sectorsize
- 1);
3979 if (start
>= extent_end
) {
3981 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3982 ret
= btrfs_next_leaf(root
, path
);
3989 leaf
= path
->nodes
[0];
3991 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3992 if (found_key
.objectid
!= objectid
||
3993 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3995 if (start
+ len
<= found_key
.offset
)
3998 em
->len
= found_key
.offset
- start
;
4002 if (found_type
== BTRFS_FILE_EXTENT_REG
||
4003 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
4004 em
->start
= extent_start
;
4005 em
->len
= extent_end
- extent_start
;
4006 em
->orig_start
= extent_start
-
4007 btrfs_file_extent_offset(leaf
, item
);
4008 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
4010 em
->block_start
= EXTENT_MAP_HOLE
;
4014 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4015 em
->block_start
= bytenr
;
4016 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
4019 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
4020 em
->block_start
= bytenr
;
4021 em
->block_len
= em
->len
;
4022 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
4023 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
4026 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4030 size_t extent_offset
;
4033 em
->block_start
= EXTENT_MAP_INLINE
;
4034 if (!page
|| create
) {
4035 em
->start
= extent_start
;
4036 em
->len
= extent_end
- extent_start
;
4040 size
= btrfs_file_extent_inline_len(leaf
, item
);
4041 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
4042 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
4043 size
- extent_offset
);
4044 em
->start
= extent_start
+ extent_offset
;
4045 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
4046 ~((u64
)root
->sectorsize
- 1);
4047 em
->orig_start
= EXTENT_MAP_INLINE
;
4049 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4050 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
4051 if (create
== 0 && !PageUptodate(page
)) {
4052 if (btrfs_file_extent_compression(leaf
, item
) ==
4053 BTRFS_COMPRESS_ZLIB
) {
4054 ret
= uncompress_inline(path
, inode
, page
,
4056 extent_offset
, item
);
4060 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4064 flush_dcache_page(page
);
4065 } else if (create
&& PageUptodate(page
)) {
4068 free_extent_map(em
);
4070 btrfs_release_path(root
, path
);
4071 trans
= btrfs_join_transaction(root
, 1);
4075 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4078 btrfs_mark_buffer_dirty(leaf
);
4080 set_extent_uptodate(io_tree
, em
->start
,
4081 extent_map_end(em
) - 1, GFP_NOFS
);
4084 printk("unkknown found_type %d\n", found_type
);
4091 em
->block_start
= EXTENT_MAP_HOLE
;
4092 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
4094 btrfs_release_path(root
, path
);
4095 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
4096 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em
->start
, em
->len
, start
, len
);
4102 spin_lock(&em_tree
->lock
);
4103 ret
= add_extent_mapping(em_tree
, em
);
4104 /* it is possible that someone inserted the extent into the tree
4105 * while we had the lock dropped. It is also possible that
4106 * an overlapping map exists in the tree
4108 if (ret
== -EEXIST
) {
4109 struct extent_map
*existing
;
4113 existing
= lookup_extent_mapping(em_tree
, start
, len
);
4114 if (existing
&& (existing
->start
> start
||
4115 existing
->start
+ existing
->len
<= start
)) {
4116 free_extent_map(existing
);
4120 existing
= lookup_extent_mapping(em_tree
, em
->start
,
4123 err
= merge_extent_mapping(em_tree
, existing
,
4126 free_extent_map(existing
);
4128 free_extent_map(em
);
4133 printk("failing to insert %Lu %Lu\n",
4135 free_extent_map(em
);
4139 free_extent_map(em
);
4144 spin_unlock(&em_tree
->lock
);
4147 btrfs_free_path(path
);
4149 ret
= btrfs_end_transaction(trans
, root
);
4155 free_extent_map(em
);
4157 return ERR_PTR(err
);
4162 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
4163 const struct iovec
*iov
, loff_t offset
,
4164 unsigned long nr_segs
)
4169 static sector_t
btrfs_bmap(struct address_space
*mapping
, sector_t iblock
)
4171 return extent_bmap(mapping
, iblock
, btrfs_get_extent
);
4174 int btrfs_readpage(struct file
*file
, struct page
*page
)
4176 struct extent_io_tree
*tree
;
4177 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4178 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
4181 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
4183 struct extent_io_tree
*tree
;
4186 if (current
->flags
& PF_MEMALLOC
) {
4187 redirty_page_for_writepage(wbc
, page
);
4191 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4192 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
4195 int btrfs_writepages(struct address_space
*mapping
,
4196 struct writeback_control
*wbc
)
4198 struct extent_io_tree
*tree
;
4200 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4201 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
4205 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
4206 struct list_head
*pages
, unsigned nr_pages
)
4208 struct extent_io_tree
*tree
;
4209 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4210 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
4213 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4215 struct extent_io_tree
*tree
;
4216 struct extent_map_tree
*map
;
4219 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4220 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
4221 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
4223 ClearPagePrivate(page
);
4224 set_page_private(page
, 0);
4225 page_cache_release(page
);
4230 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4232 if (PageWriteback(page
) || PageDirty(page
))
4234 return __btrfs_releasepage(page
, gfp_flags
);
4237 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
4239 struct extent_io_tree
*tree
;
4240 struct btrfs_ordered_extent
*ordered
;
4241 u64 page_start
= page_offset(page
);
4242 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4244 wait_on_page_writeback(page
);
4245 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4247 btrfs_releasepage(page
, GFP_NOFS
);
4251 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4252 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
4256 * IO on this page will never be started, so we need
4257 * to account for any ordered extents now
4259 clear_extent_bit(tree
, page_start
, page_end
,
4260 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4261 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
4262 btrfs_finish_ordered_io(page
->mapping
->host
,
4263 page_start
, page_end
);
4264 btrfs_put_ordered_extent(ordered
);
4265 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4267 clear_extent_bit(tree
, page_start
, page_end
,
4268 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4271 __btrfs_releasepage(page
, GFP_NOFS
);
4273 ClearPageChecked(page
);
4274 if (PagePrivate(page
)) {
4275 ClearPagePrivate(page
);
4276 set_page_private(page
, 0);
4277 page_cache_release(page
);
4282 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4283 * called from a page fault handler when a page is first dirtied. Hence we must
4284 * be careful to check for EOF conditions here. We set the page up correctly
4285 * for a written page which means we get ENOSPC checking when writing into
4286 * holes and correct delalloc and unwritten extent mapping on filesystems that
4287 * support these features.
4289 * We are not allowed to take the i_mutex here so we have to play games to
4290 * protect against truncate races as the page could now be beyond EOF. Because
4291 * vmtruncate() writes the inode size before removing pages, once we have the
4292 * page lock we can determine safely if the page is beyond EOF. If it is not
4293 * beyond EOF, then the page is guaranteed safe against truncation until we
4296 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4298 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
4299 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4300 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4301 struct btrfs_ordered_extent
*ordered
;
4303 unsigned long zero_start
;
4309 ret
= btrfs_check_free_space(root
, PAGE_CACHE_SIZE
, 0);
4316 size
= i_size_read(inode
);
4317 page_start
= page_offset(page
);
4318 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4320 if ((page
->mapping
!= inode
->i_mapping
) ||
4321 (page_start
>= size
)) {
4322 /* page got truncated out from underneath us */
4325 wait_on_page_writeback(page
);
4327 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4328 set_page_extent_mapped(page
);
4331 * we can't set the delalloc bits if there are pending ordered
4332 * extents. Drop our locks and wait for them to finish
4334 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4336 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4338 btrfs_start_ordered_extent(inode
, ordered
, 1);
4339 btrfs_put_ordered_extent(ordered
);
4343 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
4346 /* page is wholly or partially inside EOF */
4347 if (page_start
+ PAGE_CACHE_SIZE
> size
)
4348 zero_start
= size
& ~PAGE_CACHE_MASK
;
4350 zero_start
= PAGE_CACHE_SIZE
;
4352 if (zero_start
!= PAGE_CACHE_SIZE
) {
4354 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
4355 flush_dcache_page(page
);
4358 ClearPageChecked(page
);
4359 set_page_dirty(page
);
4360 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4368 static void btrfs_truncate(struct inode
*inode
)
4370 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4372 struct btrfs_trans_handle
*trans
;
4374 u64 mask
= root
->sectorsize
- 1;
4376 if (!S_ISREG(inode
->i_mode
))
4378 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4381 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
4382 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
4384 trans
= btrfs_start_transaction(root
, 1);
4385 btrfs_set_trans_block_group(trans
, inode
);
4386 btrfs_i_size_write(inode
, inode
->i_size
);
4388 ret
= btrfs_orphan_add(trans
, inode
);
4391 /* FIXME, add redo link to tree so we don't leak on crash */
4392 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
4393 BTRFS_EXTENT_DATA_KEY
);
4394 btrfs_update_inode(trans
, root
, inode
);
4396 ret
= btrfs_orphan_del(trans
, inode
);
4400 nr
= trans
->blocks_used
;
4401 ret
= btrfs_end_transaction_throttle(trans
, root
);
4403 btrfs_btree_balance_dirty(root
, nr
);
4407 * create a new subvolume directory/inode (helper for the ioctl).
4409 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
4410 struct btrfs_root
*new_root
, struct dentry
*dentry
,
4411 u64 new_dirid
, u64 alloc_hint
)
4413 struct inode
*inode
;
4417 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
4418 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
4420 return PTR_ERR(inode
);
4421 inode
->i_op
= &btrfs_dir_inode_operations
;
4422 inode
->i_fop
= &btrfs_dir_file_operations
;
4425 btrfs_i_size_write(inode
, 0);
4427 error
= btrfs_update_inode(trans
, new_root
, inode
);
4431 d_instantiate(dentry
, inode
);
4435 /* helper function for file defrag and space balancing. This
4436 * forces readahead on a given range of bytes in an inode
4438 unsigned long btrfs_force_ra(struct address_space
*mapping
,
4439 struct file_ra_state
*ra
, struct file
*file
,
4440 pgoff_t offset
, pgoff_t last_index
)
4442 pgoff_t req_size
= last_index
- offset
+ 1;
4444 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
4445 return offset
+ req_size
;
4448 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
4450 struct btrfs_inode
*ei
;
4452 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
4456 ei
->logged_trans
= 0;
4457 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
4458 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
4459 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
4460 INIT_LIST_HEAD(&ei
->i_orphan
);
4461 return &ei
->vfs_inode
;
4464 void btrfs_destroy_inode(struct inode
*inode
)
4466 struct btrfs_ordered_extent
*ordered
;
4467 WARN_ON(!list_empty(&inode
->i_dentry
));
4468 WARN_ON(inode
->i_data
.nrpages
);
4470 if (BTRFS_I(inode
)->i_acl
&&
4471 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
4472 posix_acl_release(BTRFS_I(inode
)->i_acl
);
4473 if (BTRFS_I(inode
)->i_default_acl
&&
4474 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
4475 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
4477 spin_lock(&BTRFS_I(inode
)->root
->list_lock
);
4478 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
4479 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
4480 " list\n", inode
->i_ino
);
4483 spin_unlock(&BTRFS_I(inode
)->root
->list_lock
);
4486 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
4490 printk("found ordered extent %Lu %Lu\n",
4491 ordered
->file_offset
, ordered
->len
);
4492 btrfs_remove_ordered_extent(inode
, ordered
);
4493 btrfs_put_ordered_extent(ordered
);
4494 btrfs_put_ordered_extent(ordered
);
4497 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4498 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4501 static void init_once(void *foo
)
4503 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4505 inode_init_once(&ei
->vfs_inode
);
4508 void btrfs_destroy_cachep(void)
4510 if (btrfs_inode_cachep
)
4511 kmem_cache_destroy(btrfs_inode_cachep
);
4512 if (btrfs_trans_handle_cachep
)
4513 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4514 if (btrfs_transaction_cachep
)
4515 kmem_cache_destroy(btrfs_transaction_cachep
);
4516 if (btrfs_bit_radix_cachep
)
4517 kmem_cache_destroy(btrfs_bit_radix_cachep
);
4518 if (btrfs_path_cachep
)
4519 kmem_cache_destroy(btrfs_path_cachep
);
4522 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
4523 unsigned long extra_flags
,
4524 void (*ctor
)(void *))
4526 return kmem_cache_create(name
, size
, 0, (SLAB_RECLAIM_ACCOUNT
|
4527 SLAB_MEM_SPREAD
| extra_flags
), ctor
);
4530 int btrfs_init_cachep(void)
4532 btrfs_inode_cachep
= btrfs_cache_create("btrfs_inode_cache",
4533 sizeof(struct btrfs_inode
),
4535 if (!btrfs_inode_cachep
)
4537 btrfs_trans_handle_cachep
=
4538 btrfs_cache_create("btrfs_trans_handle_cache",
4539 sizeof(struct btrfs_trans_handle
),
4541 if (!btrfs_trans_handle_cachep
)
4543 btrfs_transaction_cachep
= btrfs_cache_create("btrfs_transaction_cache",
4544 sizeof(struct btrfs_transaction
),
4546 if (!btrfs_transaction_cachep
)
4548 btrfs_path_cachep
= btrfs_cache_create("btrfs_path_cache",
4549 sizeof(struct btrfs_path
),
4551 if (!btrfs_path_cachep
)
4553 btrfs_bit_radix_cachep
= btrfs_cache_create("btrfs_radix", 256,
4554 SLAB_DESTROY_BY_RCU
, NULL
);
4555 if (!btrfs_bit_radix_cachep
)
4559 btrfs_destroy_cachep();
4563 static int btrfs_getattr(struct vfsmount
*mnt
,
4564 struct dentry
*dentry
, struct kstat
*stat
)
4566 struct inode
*inode
= dentry
->d_inode
;
4567 generic_fillattr(inode
, stat
);
4568 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
4569 stat
->blksize
= PAGE_CACHE_SIZE
;
4570 stat
->blocks
= (inode_get_bytes(inode
) +
4571 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4575 static int btrfs_rename(struct inode
* old_dir
, struct dentry
*old_dentry
,
4576 struct inode
* new_dir
,struct dentry
*new_dentry
)
4578 struct btrfs_trans_handle
*trans
;
4579 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4580 struct inode
*new_inode
= new_dentry
->d_inode
;
4581 struct inode
*old_inode
= old_dentry
->d_inode
;
4582 struct timespec ctime
= CURRENT_TIME
;
4586 /* we're not allowed to rename between subvolumes */
4587 if (BTRFS_I(old_inode
)->root
->root_key
.objectid
!=
4588 BTRFS_I(new_dir
)->root
->root_key
.objectid
)
4591 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4592 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4596 /* to rename a snapshot or subvolume, we need to juggle the
4597 * backrefs. This isn't coded yet
4599 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
4602 ret
= btrfs_check_free_space(root
, 1, 0);
4606 trans
= btrfs_start_transaction(root
, 1);
4608 btrfs_set_trans_block_group(trans
, new_dir
);
4610 btrfs_inc_nlink(old_dentry
->d_inode
);
4611 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4612 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4613 old_inode
->i_ctime
= ctime
;
4615 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4616 old_dentry
->d_name
.name
,
4617 old_dentry
->d_name
.len
);
4622 new_inode
->i_ctime
= CURRENT_TIME
;
4623 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4624 new_dentry
->d_inode
,
4625 new_dentry
->d_name
.name
,
4626 new_dentry
->d_name
.len
);
4629 if (new_inode
->i_nlink
== 0) {
4630 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4636 ret
= btrfs_set_inode_index(new_dir
, &index
);
4640 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4641 old_inode
, new_dentry
->d_name
.name
,
4642 new_dentry
->d_name
.len
, 1, index
);
4647 btrfs_end_transaction_throttle(trans
, root
);
4653 * some fairly slow code that needs optimization. This walks the list
4654 * of all the inodes with pending delalloc and forces them to disk.
4656 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4658 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4659 struct btrfs_inode
*binode
;
4660 struct inode
*inode
;
4662 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
4665 spin_lock(&root
->fs_info
->delalloc_lock
);
4666 while(!list_empty(head
)) {
4667 binode
= list_entry(head
->next
, struct btrfs_inode
,
4669 inode
= igrab(&binode
->vfs_inode
);
4671 list_del_init(&binode
->delalloc_inodes
);
4672 spin_unlock(&root
->fs_info
->delalloc_lock
);
4674 filemap_flush(inode
->i_mapping
);
4678 spin_lock(&root
->fs_info
->delalloc_lock
);
4680 spin_unlock(&root
->fs_info
->delalloc_lock
);
4682 /* the filemap_flush will queue IO into the worker threads, but
4683 * we have to make sure the IO is actually started and that
4684 * ordered extents get created before we return
4686 atomic_inc(&root
->fs_info
->async_submit_draining
);
4687 while(atomic_read(&root
->fs_info
->nr_async_submits
) ||
4688 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
4689 wait_event(root
->fs_info
->async_submit_wait
,
4690 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
4691 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
4693 atomic_dec(&root
->fs_info
->async_submit_draining
);
4697 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4698 const char *symname
)
4700 struct btrfs_trans_handle
*trans
;
4701 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4702 struct btrfs_path
*path
;
4703 struct btrfs_key key
;
4704 struct inode
*inode
= NULL
;
4712 struct btrfs_file_extent_item
*ei
;
4713 struct extent_buffer
*leaf
;
4714 unsigned long nr
= 0;
4716 name_len
= strlen(symname
) + 1;
4717 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4718 return -ENAMETOOLONG
;
4720 err
= btrfs_check_free_space(root
, 1, 0);
4724 trans
= btrfs_start_transaction(root
, 1);
4725 btrfs_set_trans_block_group(trans
, dir
);
4727 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4733 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4735 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4736 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4738 err
= PTR_ERR(inode
);
4742 err
= btrfs_init_acl(inode
, dir
);
4748 btrfs_set_trans_block_group(trans
, inode
);
4749 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4753 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4754 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4755 inode
->i_fop
= &btrfs_file_operations
;
4756 inode
->i_op
= &btrfs_file_inode_operations
;
4757 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4759 dir
->i_sb
->s_dirt
= 1;
4760 btrfs_update_inode_block_group(trans
, inode
);
4761 btrfs_update_inode_block_group(trans
, dir
);
4765 path
= btrfs_alloc_path();
4767 key
.objectid
= inode
->i_ino
;
4769 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
4770 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
4771 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
4777 leaf
= path
->nodes
[0];
4778 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
4779 struct btrfs_file_extent_item
);
4780 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
4781 btrfs_set_file_extent_type(leaf
, ei
,
4782 BTRFS_FILE_EXTENT_INLINE
);
4783 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
4784 btrfs_set_file_extent_compression(leaf
, ei
, 0);
4785 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
4786 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
4788 ptr
= btrfs_file_extent_inline_start(ei
);
4789 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
4790 btrfs_mark_buffer_dirty(leaf
);
4791 btrfs_free_path(path
);
4793 inode
->i_op
= &btrfs_symlink_inode_operations
;
4794 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
4795 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4796 inode_set_bytes(inode
, name_len
);
4797 btrfs_i_size_write(inode
, name_len
- 1);
4798 err
= btrfs_update_inode(trans
, root
, inode
);
4803 nr
= trans
->blocks_used
;
4804 btrfs_end_transaction_throttle(trans
, root
);
4807 inode_dec_link_count(inode
);
4810 btrfs_btree_balance_dirty(root
, nr
);
4814 static int prealloc_file_range(struct inode
*inode
, u64 start
, u64 end
,
4815 u64 alloc_hint
, int mode
)
4817 struct btrfs_trans_handle
*trans
;
4818 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4819 struct btrfs_key ins
;
4821 u64 cur_offset
= start
;
4822 u64 num_bytes
= end
- start
;
4825 trans
= btrfs_join_transaction(root
, 1);
4827 btrfs_set_trans_block_group(trans
, inode
);
4829 while (num_bytes
> 0) {
4830 alloc_size
= min(num_bytes
, root
->fs_info
->max_extent
);
4831 ret
= btrfs_reserve_extent(trans
, root
, alloc_size
,
4832 root
->sectorsize
, 0, alloc_hint
,
4838 ret
= insert_reserved_file_extent(trans
, inode
,
4839 cur_offset
, ins
.objectid
,
4840 ins
.offset
, ins
.offset
,
4841 ins
.offset
, 0, 0, 0,
4842 BTRFS_FILE_EXTENT_PREALLOC
);
4844 num_bytes
-= ins
.offset
;
4845 cur_offset
+= ins
.offset
;
4846 alloc_hint
= ins
.objectid
+ ins
.offset
;
4849 if (cur_offset
> start
) {
4850 inode
->i_ctime
= CURRENT_TIME
;
4851 btrfs_set_flag(inode
, PREALLOC
);
4852 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4853 cur_offset
> i_size_read(inode
))
4854 btrfs_i_size_write(inode
, cur_offset
);
4855 ret
= btrfs_update_inode(trans
, root
, inode
);
4859 btrfs_end_transaction(trans
, root
);
4863 static long btrfs_fallocate(struct inode
*inode
, int mode
,
4864 loff_t offset
, loff_t len
)
4871 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
4872 struct extent_map
*em
;
4875 alloc_start
= offset
& ~mask
;
4876 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
4878 mutex_lock(&inode
->i_mutex
);
4879 if (alloc_start
> inode
->i_size
) {
4880 ret
= btrfs_cont_expand(inode
, alloc_start
);
4886 struct btrfs_ordered_extent
*ordered
;
4887 lock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
,
4888 alloc_end
- 1, GFP_NOFS
);
4889 ordered
= btrfs_lookup_first_ordered_extent(inode
,
4892 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
4893 ordered
->file_offset
< alloc_end
) {
4894 btrfs_put_ordered_extent(ordered
);
4895 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4896 alloc_start
, alloc_end
- 1, GFP_NOFS
);
4897 btrfs_wait_ordered_range(inode
, alloc_start
,
4898 alloc_end
- alloc_start
);
4901 btrfs_put_ordered_extent(ordered
);
4906 cur_offset
= alloc_start
;
4908 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4909 alloc_end
- cur_offset
, 0);
4910 BUG_ON(IS_ERR(em
) || !em
);
4911 last_byte
= min(extent_map_end(em
), alloc_end
);
4912 last_byte
= (last_byte
+ mask
) & ~mask
;
4913 if (em
->block_start
== EXTENT_MAP_HOLE
) {
4914 ret
= prealloc_file_range(inode
, cur_offset
,
4915 last_byte
, alloc_hint
, mode
);
4917 free_extent_map(em
);
4921 if (em
->block_start
<= EXTENT_MAP_LAST_BYTE
)
4922 alloc_hint
= em
->block_start
;
4923 free_extent_map(em
);
4925 cur_offset
= last_byte
;
4926 if (cur_offset
>= alloc_end
) {
4931 unlock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, alloc_end
- 1,
4934 mutex_unlock(&inode
->i_mutex
);
4938 static int btrfs_set_page_dirty(struct page
*page
)
4940 return __set_page_dirty_nobuffers(page
);
4943 static int btrfs_permission(struct inode
*inode
, int mask
)
4945 if (btrfs_test_flag(inode
, READONLY
) && (mask
& MAY_WRITE
))
4947 return generic_permission(inode
, mask
, btrfs_check_acl
);
4950 static struct inode_operations btrfs_dir_inode_operations
= {
4951 .getattr
= btrfs_getattr
,
4952 .lookup
= btrfs_lookup
,
4953 .create
= btrfs_create
,
4954 .unlink
= btrfs_unlink
,
4956 .mkdir
= btrfs_mkdir
,
4957 .rmdir
= btrfs_rmdir
,
4958 .rename
= btrfs_rename
,
4959 .symlink
= btrfs_symlink
,
4960 .setattr
= btrfs_setattr
,
4961 .mknod
= btrfs_mknod
,
4962 .setxattr
= btrfs_setxattr
,
4963 .getxattr
= btrfs_getxattr
,
4964 .listxattr
= btrfs_listxattr
,
4965 .removexattr
= btrfs_removexattr
,
4966 .permission
= btrfs_permission
,
4968 static struct inode_operations btrfs_dir_ro_inode_operations
= {
4969 .lookup
= btrfs_lookup
,
4970 .permission
= btrfs_permission
,
4972 static struct file_operations btrfs_dir_file_operations
= {
4973 .llseek
= generic_file_llseek
,
4974 .read
= generic_read_dir
,
4975 .readdir
= btrfs_real_readdir
,
4976 .unlocked_ioctl
= btrfs_ioctl
,
4977 #ifdef CONFIG_COMPAT
4978 .compat_ioctl
= btrfs_ioctl
,
4980 .release
= btrfs_release_file
,
4981 .fsync
= btrfs_sync_file
,
4984 static struct extent_io_ops btrfs_extent_io_ops
= {
4985 .fill_delalloc
= run_delalloc_range
,
4986 .submit_bio_hook
= btrfs_submit_bio_hook
,
4987 .merge_bio_hook
= btrfs_merge_bio_hook
,
4988 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
4989 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
4990 .writepage_start_hook
= btrfs_writepage_start_hook
,
4991 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
4992 .set_bit_hook
= btrfs_set_bit_hook
,
4993 .clear_bit_hook
= btrfs_clear_bit_hook
,
4996 static struct address_space_operations btrfs_aops
= {
4997 .readpage
= btrfs_readpage
,
4998 .writepage
= btrfs_writepage
,
4999 .writepages
= btrfs_writepages
,
5000 .readpages
= btrfs_readpages
,
5001 .sync_page
= block_sync_page
,
5003 .direct_IO
= btrfs_direct_IO
,
5004 .invalidatepage
= btrfs_invalidatepage
,
5005 .releasepage
= btrfs_releasepage
,
5006 .set_page_dirty
= btrfs_set_page_dirty
,
5009 static struct address_space_operations btrfs_symlink_aops
= {
5010 .readpage
= btrfs_readpage
,
5011 .writepage
= btrfs_writepage
,
5012 .invalidatepage
= btrfs_invalidatepage
,
5013 .releasepage
= btrfs_releasepage
,
5016 static struct inode_operations btrfs_file_inode_operations
= {
5017 .truncate
= btrfs_truncate
,
5018 .getattr
= btrfs_getattr
,
5019 .setattr
= btrfs_setattr
,
5020 .setxattr
= btrfs_setxattr
,
5021 .getxattr
= btrfs_getxattr
,
5022 .listxattr
= btrfs_listxattr
,
5023 .removexattr
= btrfs_removexattr
,
5024 .permission
= btrfs_permission
,
5025 .fallocate
= btrfs_fallocate
,
5027 static struct inode_operations btrfs_special_inode_operations
= {
5028 .getattr
= btrfs_getattr
,
5029 .setattr
= btrfs_setattr
,
5030 .permission
= btrfs_permission
,
5031 .setxattr
= btrfs_setxattr
,
5032 .getxattr
= btrfs_getxattr
,
5033 .listxattr
= btrfs_listxattr
,
5034 .removexattr
= btrfs_removexattr
,
5036 static struct inode_operations btrfs_symlink_inode_operations
= {
5037 .readlink
= generic_readlink
,
5038 .follow_link
= page_follow_link_light
,
5039 .put_link
= page_put_link
,
5040 .permission
= btrfs_permission
,