2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include <linux/posix_acl.h>
40 #include <linux/falloc.h>
44 #include "transaction.h"
45 #include "btrfs_inode.h"
47 #include "print-tree.h"
49 #include "ordered-data.h"
52 #include "ref-cache.h"
53 #include "compression.h"
55 struct btrfs_iget_args
{
57 struct btrfs_root
*root
;
60 static struct inode_operations btrfs_dir_inode_operations
;
61 static struct inode_operations btrfs_symlink_inode_operations
;
62 static struct inode_operations btrfs_dir_ro_inode_operations
;
63 static struct inode_operations btrfs_special_inode_operations
;
64 static struct inode_operations btrfs_file_inode_operations
;
65 static struct address_space_operations btrfs_aops
;
66 static struct address_space_operations btrfs_symlink_aops
;
67 static struct file_operations btrfs_dir_file_operations
;
68 static struct extent_io_ops btrfs_extent_io_ops
;
70 static struct kmem_cache
*btrfs_inode_cachep
;
71 struct kmem_cache
*btrfs_trans_handle_cachep
;
72 struct kmem_cache
*btrfs_transaction_cachep
;
73 struct kmem_cache
*btrfs_bit_radix_cachep
;
74 struct kmem_cache
*btrfs_path_cachep
;
77 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
78 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
79 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
80 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
81 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
82 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
83 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
84 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
87 static void btrfs_truncate(struct inode
*inode
);
88 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
89 static noinline
int cow_file_range(struct inode
*inode
,
90 struct page
*locked_page
,
91 u64 start
, u64 end
, int *page_started
,
92 unsigned long *nr_written
, int unlock
);
95 * a very lame attempt at stopping writes when the FS is 85% full. There
96 * are countless ways this is incorrect, but it is better than nothing.
98 int btrfs_check_free_space(struct btrfs_root
*root
, u64 num_required
,
107 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
108 total
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
109 used
= btrfs_super_bytes_used(&root
->fs_info
->super_copy
);
117 if (used
+ root
->fs_info
->delalloc_bytes
+ num_required
> thresh
)
119 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static int noinline
insert_inline_extent(struct btrfs_trans_handle
*trans
,
129 struct btrfs_root
*root
, struct inode
*inode
,
130 u64 start
, size_t size
, size_t compressed_size
,
131 struct page
**compressed_pages
)
133 struct btrfs_key key
;
134 struct btrfs_path
*path
;
135 struct extent_buffer
*leaf
;
136 struct page
*page
= NULL
;
139 struct btrfs_file_extent_item
*ei
;
142 size_t cur_size
= size
;
144 unsigned long offset
;
145 int use_compress
= 0;
147 if (compressed_size
&& compressed_pages
) {
149 cur_size
= compressed_size
;
152 path
= btrfs_alloc_path(); if (!path
)
155 btrfs_set_trans_block_group(trans
, inode
);
157 key
.objectid
= inode
->i_ino
;
159 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
160 inode_add_bytes(inode
, size
);
161 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
163 inode_add_bytes(inode
, size
);
164 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
169 printk("got bad ret %d\n", ret
);
172 leaf
= path
->nodes
[0];
173 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
174 struct btrfs_file_extent_item
);
175 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
176 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
177 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
178 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
179 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
180 ptr
= btrfs_file_extent_inline_start(ei
);
185 while(compressed_size
> 0) {
186 cpage
= compressed_pages
[i
];
187 cur_size
= min_t(unsigned long, compressed_size
,
191 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
196 compressed_size
-= cur_size
;
198 btrfs_set_file_extent_compression(leaf
, ei
,
199 BTRFS_COMPRESS_ZLIB
);
201 page
= find_get_page(inode
->i_mapping
,
202 start
>> PAGE_CACHE_SHIFT
);
203 btrfs_set_file_extent_compression(leaf
, ei
, 0);
204 kaddr
= kmap_atomic(page
, KM_USER0
);
205 offset
= start
& (PAGE_CACHE_SIZE
- 1);
206 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
207 kunmap_atomic(kaddr
, KM_USER0
);
208 page_cache_release(page
);
210 btrfs_mark_buffer_dirty(leaf
);
211 btrfs_free_path(path
);
213 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
214 btrfs_update_inode(trans
, root
, inode
);
217 btrfs_free_path(path
);
223 * conditionally insert an inline extent into the file. This
224 * does the checks required to make sure the data is small enough
225 * to fit as an inline extent.
227 static int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
228 struct btrfs_root
*root
,
229 struct inode
*inode
, u64 start
, u64 end
,
230 size_t compressed_size
,
231 struct page
**compressed_pages
)
233 u64 isize
= i_size_read(inode
);
234 u64 actual_end
= min(end
+ 1, isize
);
235 u64 inline_len
= actual_end
- start
;
236 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
237 ~((u64
)root
->sectorsize
- 1);
239 u64 data_len
= inline_len
;
243 data_len
= compressed_size
;
246 actual_end
>= PAGE_CACHE_SIZE
||
247 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
249 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
251 data_len
> root
->fs_info
->max_inline
) {
255 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
256 aligned_end
, start
, &hint_byte
);
259 if (isize
> actual_end
)
260 inline_len
= min_t(u64
, isize
, actual_end
);
261 ret
= insert_inline_extent(trans
, root
, inode
, start
,
262 inline_len
, compressed_size
,
265 btrfs_drop_extent_cache(inode
, start
, aligned_end
, 0);
269 struct async_extent
{
274 unsigned long nr_pages
;
275 struct list_head list
;
280 struct btrfs_root
*root
;
281 struct page
*locked_page
;
284 struct list_head extents
;
285 struct btrfs_work work
;
288 static noinline
int add_async_extent(struct async_cow
*cow
,
289 u64 start
, u64 ram_size
,
292 unsigned long nr_pages
)
294 struct async_extent
*async_extent
;
296 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
297 async_extent
->start
= start
;
298 async_extent
->ram_size
= ram_size
;
299 async_extent
->compressed_size
= compressed_size
;
300 async_extent
->pages
= pages
;
301 async_extent
->nr_pages
= nr_pages
;
302 list_add_tail(&async_extent
->list
, &cow
->extents
);
307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
322 static noinline
int compress_file_range(struct inode
*inode
,
323 struct page
*locked_page
,
325 struct async_cow
*async_cow
,
328 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
329 struct btrfs_trans_handle
*trans
;
333 u64 blocksize
= root
->sectorsize
;
335 u64 isize
= i_size_read(inode
);
337 struct page
**pages
= NULL
;
338 unsigned long nr_pages
;
339 unsigned long nr_pages_ret
= 0;
340 unsigned long total_compressed
= 0;
341 unsigned long total_in
= 0;
342 unsigned long max_compressed
= 128 * 1024;
343 unsigned long max_uncompressed
= 128 * 1024;
349 actual_end
= min_t(u64
, isize
, end
+ 1);
352 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
353 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
355 total_compressed
= actual_end
- start
;
357 /* we want to make sure that amount of ram required to uncompress
358 * an extent is reasonable, so we limit the total size in ram
359 * of a compressed extent to 128k. This is a crucial number
360 * because it also controls how easily we can spread reads across
361 * cpus for decompression.
363 * We also want to make sure the amount of IO required to do
364 * a random read is reasonably small, so we limit the size of
365 * a compressed extent to 128k.
367 total_compressed
= min(total_compressed
, max_uncompressed
);
368 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
369 num_bytes
= max(blocksize
, num_bytes
);
370 disk_num_bytes
= num_bytes
;
375 * we do compression for mount -o compress and when the
376 * inode has not been flagged as nocompress. This flag can
377 * change at any time if we discover bad compression ratios.
379 if (!btrfs_test_flag(inode
, NOCOMPRESS
) &&
380 btrfs_test_opt(root
, COMPRESS
)) {
382 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
384 ret
= btrfs_zlib_compress_pages(inode
->i_mapping
, start
,
385 total_compressed
, pages
,
386 nr_pages
, &nr_pages_ret
,
392 unsigned long offset
= total_compressed
&
393 (PAGE_CACHE_SIZE
- 1);
394 struct page
*page
= pages
[nr_pages_ret
- 1];
397 /* zero the tail end of the last page, we might be
398 * sending it down to disk
401 kaddr
= kmap_atomic(page
, KM_USER0
);
402 memset(kaddr
+ offset
, 0,
403 PAGE_CACHE_SIZE
- offset
);
404 kunmap_atomic(kaddr
, KM_USER0
);
410 trans
= btrfs_join_transaction(root
, 1);
412 btrfs_set_trans_block_group(trans
, inode
);
414 /* lets try to make an inline extent */
415 if (ret
|| total_in
< (actual_end
- start
)) {
416 /* we didn't compress the entire range, try
417 * to make an uncompressed inline extent.
419 ret
= cow_file_range_inline(trans
, root
, inode
,
420 start
, end
, 0, NULL
);
422 /* try making a compressed inline extent */
423 ret
= cow_file_range_inline(trans
, root
, inode
,
425 total_compressed
, pages
);
427 btrfs_end_transaction(trans
, root
);
430 * inline extent creation worked, we don't need
431 * to create any more async work items. Unlock
432 * and free up our temp pages.
434 extent_clear_unlock_delalloc(inode
,
435 &BTRFS_I(inode
)->io_tree
,
436 start
, end
, NULL
, 1, 0,
445 * we aren't doing an inline extent round the compressed size
446 * up to a block size boundary so the allocator does sane
449 total_compressed
= (total_compressed
+ blocksize
- 1) &
453 * one last check to make sure the compression is really a
454 * win, compare the page count read with the blocks on disk
456 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
457 ~(PAGE_CACHE_SIZE
- 1);
458 if (total_compressed
>= total_in
) {
461 disk_num_bytes
= total_compressed
;
462 num_bytes
= total_in
;
465 if (!will_compress
&& pages
) {
467 * the compression code ran but failed to make things smaller,
468 * free any pages it allocated and our page pointer array
470 for (i
= 0; i
< nr_pages_ret
; i
++) {
471 WARN_ON(pages
[i
]->mapping
);
472 page_cache_release(pages
[i
]);
476 total_compressed
= 0;
479 /* flag the file so we don't compress in the future */
480 btrfs_set_flag(inode
, NOCOMPRESS
);
485 /* the async work queues will take care of doing actual
486 * allocation on disk for these compressed pages,
487 * and will submit them to the elevator.
489 add_async_extent(async_cow
, start
, num_bytes
,
490 total_compressed
, pages
, nr_pages_ret
);
492 if (start
+ num_bytes
< end
&& start
+ num_bytes
< actual_end
) {
500 * No compression, but we still need to write the pages in
501 * the file we've been given so far. redirty the locked
502 * page if it corresponds to our extent and set things up
503 * for the async work queue to run cow_file_range to do
504 * the normal delalloc dance
506 if (page_offset(locked_page
) >= start
&&
507 page_offset(locked_page
) <= end
) {
508 __set_page_dirty_nobuffers(locked_page
);
509 /* unlocked later on in the async handlers */
511 add_async_extent(async_cow
, start
, end
- start
+ 1, 0, NULL
, 0);
519 for (i
= 0; i
< nr_pages_ret
; i
++) {
520 WARN_ON(pages
[i
]->mapping
);
521 page_cache_release(pages
[i
]);
530 * phase two of compressed writeback. This is the ordered portion
531 * of the code, which only gets called in the order the work was
532 * queued. We walk all the async extents created by compress_file_range
533 * and send them down to the disk.
535 static noinline
int submit_compressed_extents(struct inode
*inode
,
536 struct async_cow
*async_cow
)
538 struct async_extent
*async_extent
;
540 struct btrfs_trans_handle
*trans
;
541 struct btrfs_key ins
;
542 struct extent_map
*em
;
543 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
544 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
545 struct extent_io_tree
*io_tree
;
548 if (list_empty(&async_cow
->extents
))
551 trans
= btrfs_join_transaction(root
, 1);
553 while(!list_empty(&async_cow
->extents
)) {
554 async_extent
= list_entry(async_cow
->extents
.next
,
555 struct async_extent
, list
);
556 list_del(&async_extent
->list
);
558 io_tree
= &BTRFS_I(inode
)->io_tree
;
560 /* did the compression code fall back to uncompressed IO? */
561 if (!async_extent
->pages
) {
562 int page_started
= 0;
563 unsigned long nr_written
= 0;
565 lock_extent(io_tree
, async_extent
->start
,
566 async_extent
->start
+ async_extent
->ram_size
- 1,
569 /* allocate blocks */
570 cow_file_range(inode
, async_cow
->locked_page
,
572 async_extent
->start
+
573 async_extent
->ram_size
- 1,
574 &page_started
, &nr_written
, 0);
577 * if page_started, cow_file_range inserted an
578 * inline extent and took care of all the unlocking
579 * and IO for us. Otherwise, we need to submit
580 * all those pages down to the drive.
583 extent_write_locked_range(io_tree
,
584 inode
, async_extent
->start
,
585 async_extent
->start
+
586 async_extent
->ram_size
- 1,
594 lock_extent(io_tree
, async_extent
->start
,
595 async_extent
->start
+ async_extent
->ram_size
- 1,
598 * here we're doing allocation and writeback of the
601 btrfs_drop_extent_cache(inode
, async_extent
->start
,
602 async_extent
->start
+
603 async_extent
->ram_size
- 1, 0);
605 ret
= btrfs_reserve_extent(trans
, root
,
606 async_extent
->compressed_size
,
607 async_extent
->compressed_size
,
611 em
= alloc_extent_map(GFP_NOFS
);
612 em
->start
= async_extent
->start
;
613 em
->len
= async_extent
->ram_size
;
614 em
->orig_start
= em
->start
;
616 em
->block_start
= ins
.objectid
;
617 em
->block_len
= ins
.offset
;
618 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
619 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
620 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
623 spin_lock(&em_tree
->lock
);
624 ret
= add_extent_mapping(em_tree
, em
);
625 spin_unlock(&em_tree
->lock
);
626 if (ret
!= -EEXIST
) {
630 btrfs_drop_extent_cache(inode
, async_extent
->start
,
631 async_extent
->start
+
632 async_extent
->ram_size
- 1, 0);
635 ret
= btrfs_add_ordered_extent(inode
, async_extent
->start
,
637 async_extent
->ram_size
,
639 BTRFS_ORDERED_COMPRESSED
);
642 btrfs_end_transaction(trans
, root
);
645 * clear dirty, set writeback and unlock the pages.
647 extent_clear_unlock_delalloc(inode
,
648 &BTRFS_I(inode
)->io_tree
,
650 async_extent
->start
+
651 async_extent
->ram_size
- 1,
652 NULL
, 1, 1, 0, 1, 1, 0);
654 ret
= btrfs_submit_compressed_write(inode
,
656 async_extent
->ram_size
,
658 ins
.offset
, async_extent
->pages
,
659 async_extent
->nr_pages
);
662 trans
= btrfs_join_transaction(root
, 1);
663 alloc_hint
= ins
.objectid
+ ins
.offset
;
668 btrfs_end_transaction(trans
, root
);
673 * when extent_io.c finds a delayed allocation range in the file,
674 * the call backs end up in this code. The basic idea is to
675 * allocate extents on disk for the range, and create ordered data structs
676 * in ram to track those extents.
678 * locked_page is the page that writepage had locked already. We use
679 * it to make sure we don't do extra locks or unlocks.
681 * *page_started is set to one if we unlock locked_page and do everything
682 * required to start IO on it. It may be clean and already done with
685 static noinline
int cow_file_range(struct inode
*inode
,
686 struct page
*locked_page
,
687 u64 start
, u64 end
, int *page_started
,
688 unsigned long *nr_written
,
691 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
692 struct btrfs_trans_handle
*trans
;
695 unsigned long ram_size
;
698 u64 blocksize
= root
->sectorsize
;
700 u64 isize
= i_size_read(inode
);
701 struct btrfs_key ins
;
702 struct extent_map
*em
;
703 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
706 trans
= btrfs_join_transaction(root
, 1);
708 btrfs_set_trans_block_group(trans
, inode
);
710 actual_end
= min_t(u64
, isize
, end
+ 1);
712 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
713 num_bytes
= max(blocksize
, num_bytes
);
714 disk_num_bytes
= num_bytes
;
718 /* lets try to make an inline extent */
719 ret
= cow_file_range_inline(trans
, root
, inode
,
720 start
, end
, 0, NULL
);
722 extent_clear_unlock_delalloc(inode
,
723 &BTRFS_I(inode
)->io_tree
,
724 start
, end
, NULL
, 1, 1,
726 *nr_written
= *nr_written
+
727 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
734 BUG_ON(disk_num_bytes
>
735 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
737 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
739 while(disk_num_bytes
> 0) {
740 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
741 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
742 root
->sectorsize
, 0, alloc_hint
,
747 em
= alloc_extent_map(GFP_NOFS
);
749 em
->orig_start
= em
->start
;
751 ram_size
= ins
.offset
;
752 em
->len
= ins
.offset
;
754 em
->block_start
= ins
.objectid
;
755 em
->block_len
= ins
.offset
;
756 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
757 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
760 spin_lock(&em_tree
->lock
);
761 ret
= add_extent_mapping(em_tree
, em
);
762 spin_unlock(&em_tree
->lock
);
763 if (ret
!= -EEXIST
) {
767 btrfs_drop_extent_cache(inode
, start
,
768 start
+ ram_size
- 1, 0);
771 cur_alloc_size
= ins
.offset
;
772 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
773 ram_size
, cur_alloc_size
, 0);
776 if (root
->root_key
.objectid
==
777 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
778 ret
= btrfs_reloc_clone_csums(inode
, start
,
783 if (disk_num_bytes
< cur_alloc_size
) {
784 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes
,
788 /* we're not doing compressed IO, don't unlock the first
789 * page (which the caller expects to stay locked), don't
790 * clear any dirty bits and don't set any writeback bits
792 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
793 start
, start
+ ram_size
- 1,
794 locked_page
, unlock
, 1,
796 disk_num_bytes
-= cur_alloc_size
;
797 num_bytes
-= cur_alloc_size
;
798 alloc_hint
= ins
.objectid
+ ins
.offset
;
799 start
+= cur_alloc_size
;
803 btrfs_end_transaction(trans
, root
);
809 * work queue call back to started compression on a file and pages
811 static noinline
void async_cow_start(struct btrfs_work
*work
)
813 struct async_cow
*async_cow
;
815 async_cow
= container_of(work
, struct async_cow
, work
);
817 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
818 async_cow
->start
, async_cow
->end
, async_cow
,
821 async_cow
->inode
= NULL
;
825 * work queue call back to submit previously compressed pages
827 static noinline
void async_cow_submit(struct btrfs_work
*work
)
829 struct async_cow
*async_cow
;
830 struct btrfs_root
*root
;
831 unsigned long nr_pages
;
833 async_cow
= container_of(work
, struct async_cow
, work
);
835 root
= async_cow
->root
;
836 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
839 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
841 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
843 waitqueue_active(&root
->fs_info
->async_submit_wait
))
844 wake_up(&root
->fs_info
->async_submit_wait
);
846 if (async_cow
->inode
) {
847 submit_compressed_extents(async_cow
->inode
, async_cow
);
851 static noinline
void async_cow_free(struct btrfs_work
*work
)
853 struct async_cow
*async_cow
;
854 async_cow
= container_of(work
, struct async_cow
, work
);
858 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
859 u64 start
, u64 end
, int *page_started
,
860 unsigned long *nr_written
)
862 struct async_cow
*async_cow
;
863 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
864 unsigned long nr_pages
;
866 int limit
= 10 * 1024 * 1042;
868 if (!btrfs_test_opt(root
, COMPRESS
)) {
869 return cow_file_range(inode
, locked_page
, start
, end
,
870 page_started
, nr_written
, 1);
873 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
|
874 EXTENT_DELALLOC
, 1, 0, GFP_NOFS
);
876 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
877 async_cow
->inode
= inode
;
878 async_cow
->root
= root
;
879 async_cow
->locked_page
= locked_page
;
880 async_cow
->start
= start
;
882 if (btrfs_test_flag(inode
, NOCOMPRESS
))
885 cur_end
= min(end
, start
+ 512 * 1024 - 1);
887 async_cow
->end
= cur_end
;
888 INIT_LIST_HEAD(&async_cow
->extents
);
890 async_cow
->work
.func
= async_cow_start
;
891 async_cow
->work
.ordered_func
= async_cow_submit
;
892 async_cow
->work
.ordered_free
= async_cow_free
;
893 async_cow
->work
.flags
= 0;
895 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
897 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
899 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
902 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
903 wait_event(root
->fs_info
->async_submit_wait
,
904 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
908 while(atomic_read(&root
->fs_info
->async_submit_draining
) &&
909 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
910 wait_event(root
->fs_info
->async_submit_wait
,
911 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
915 *nr_written
+= nr_pages
;
922 static int noinline
csum_exist_in_range(struct btrfs_root
*root
,
923 u64 bytenr
, u64 num_bytes
)
926 struct btrfs_ordered_sum
*sums
;
929 ret
= btrfs_lookup_csums_range(root
, bytenr
, bytenr
+ num_bytes
- 1,
931 if (ret
== 0 && list_empty(&list
))
934 while (!list_empty(&list
)) {
935 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
936 list_del(&sums
->list
);
943 * when nowcow writeback call back. This checks for snapshots or COW copies
944 * of the extents that exist in the file, and COWs the file as required.
946 * If no cow copies or snapshots exist, we write directly to the existing
949 static int run_delalloc_nocow(struct inode
*inode
, struct page
*locked_page
,
950 u64 start
, u64 end
, int *page_started
, int force
,
951 unsigned long *nr_written
)
953 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
954 struct btrfs_trans_handle
*trans
;
955 struct extent_buffer
*leaf
;
956 struct btrfs_path
*path
;
957 struct btrfs_file_extent_item
*fi
;
958 struct btrfs_key found_key
;
970 path
= btrfs_alloc_path();
972 trans
= btrfs_join_transaction(root
, 1);
978 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
981 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
982 leaf
= path
->nodes
[0];
983 btrfs_item_key_to_cpu(leaf
, &found_key
,
985 if (found_key
.objectid
== inode
->i_ino
&&
986 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
991 leaf
= path
->nodes
[0];
992 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
993 ret
= btrfs_next_leaf(root
, path
);
998 leaf
= path
->nodes
[0];
1004 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1006 if (found_key
.objectid
> inode
->i_ino
||
1007 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1008 found_key
.offset
> end
)
1011 if (found_key
.offset
> cur_offset
) {
1012 extent_end
= found_key
.offset
;
1016 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1017 struct btrfs_file_extent_item
);
1018 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1020 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1021 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1022 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1023 extent_end
= found_key
.offset
+
1024 btrfs_file_extent_num_bytes(leaf
, fi
);
1025 if (extent_end
<= start
) {
1029 if (disk_bytenr
== 0)
1031 if (btrfs_file_extent_compression(leaf
, fi
) ||
1032 btrfs_file_extent_encryption(leaf
, fi
) ||
1033 btrfs_file_extent_other_encoding(leaf
, fi
))
1035 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1037 if (btrfs_extent_readonly(root
, disk_bytenr
))
1039 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1042 disk_bytenr
+= btrfs_file_extent_offset(leaf
, fi
);
1043 disk_bytenr
+= cur_offset
- found_key
.offset
;
1044 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1046 * force cow if csum exists in the range.
1047 * this ensure that csum for a given extent are
1048 * either valid or do not exist.
1050 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1053 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1054 extent_end
= found_key
.offset
+
1055 btrfs_file_extent_inline_len(leaf
, fi
);
1056 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1061 if (extent_end
<= start
) {
1066 if (cow_start
== (u64
)-1)
1067 cow_start
= cur_offset
;
1068 cur_offset
= extent_end
;
1069 if (cur_offset
> end
)
1075 btrfs_release_path(root
, path
);
1076 if (cow_start
!= (u64
)-1) {
1077 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1078 found_key
.offset
- 1, page_started
,
1081 cow_start
= (u64
)-1;
1084 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1085 struct extent_map
*em
;
1086 struct extent_map_tree
*em_tree
;
1087 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1088 em
= alloc_extent_map(GFP_NOFS
);
1089 em
->start
= cur_offset
;
1090 em
->orig_start
= em
->start
;
1091 em
->len
= num_bytes
;
1092 em
->block_len
= num_bytes
;
1093 em
->block_start
= disk_bytenr
;
1094 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1095 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1097 spin_lock(&em_tree
->lock
);
1098 ret
= add_extent_mapping(em_tree
, em
);
1099 spin_unlock(&em_tree
->lock
);
1100 if (ret
!= -EEXIST
) {
1101 free_extent_map(em
);
1104 btrfs_drop_extent_cache(inode
, em
->start
,
1105 em
->start
+ em
->len
- 1, 0);
1107 type
= BTRFS_ORDERED_PREALLOC
;
1109 type
= BTRFS_ORDERED_NOCOW
;
1112 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1113 num_bytes
, num_bytes
, type
);
1116 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1117 cur_offset
, cur_offset
+ num_bytes
- 1,
1118 locked_page
, 1, 1, 1, 0, 0, 0);
1119 cur_offset
= extent_end
;
1120 if (cur_offset
> end
)
1123 btrfs_release_path(root
, path
);
1125 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1126 cow_start
= cur_offset
;
1127 if (cow_start
!= (u64
)-1) {
1128 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1129 page_started
, nr_written
, 1);
1133 ret
= btrfs_end_transaction(trans
, root
);
1135 btrfs_free_path(path
);
1140 * extent_io.c call back to do delayed allocation processing
1142 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1143 u64 start
, u64 end
, int *page_started
,
1144 unsigned long *nr_written
)
1148 if (btrfs_test_flag(inode
, NODATACOW
))
1149 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1150 page_started
, 1, nr_written
);
1151 else if (btrfs_test_flag(inode
, PREALLOC
))
1152 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1153 page_started
, 0, nr_written
);
1155 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1156 page_started
, nr_written
);
1162 * extent_io.c set_bit_hook, used to track delayed allocation
1163 * bytes in this file, and to maintain the list of inodes that
1164 * have pending delalloc work to be done.
1166 static int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1167 unsigned long old
, unsigned long bits
)
1169 unsigned long flags
;
1170 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1171 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1172 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
1173 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
1174 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
1175 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1176 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1177 &root
->fs_info
->delalloc_inodes
);
1179 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
1185 * extent_io.c clear_bit_hook, see set_bit_hook for why
1187 static int btrfs_clear_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
1188 unsigned long old
, unsigned long bits
)
1190 if ((old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
1191 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1192 unsigned long flags
;
1194 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
1195 if (end
- start
+ 1 > root
->fs_info
->delalloc_bytes
) {
1196 printk("warning: delalloc account %Lu %Lu\n",
1197 end
- start
+ 1, root
->fs_info
->delalloc_bytes
);
1198 root
->fs_info
->delalloc_bytes
= 0;
1199 BTRFS_I(inode
)->delalloc_bytes
= 0;
1201 root
->fs_info
->delalloc_bytes
-= end
- start
+ 1;
1202 BTRFS_I(inode
)->delalloc_bytes
-= end
- start
+ 1;
1204 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1205 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1206 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1208 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
1214 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1215 * we don't create bios that span stripes or chunks
1217 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1218 size_t size
, struct bio
*bio
,
1219 unsigned long bio_flags
)
1221 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1222 struct btrfs_mapping_tree
*map_tree
;
1223 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1228 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1231 length
= bio
->bi_size
;
1232 map_tree
= &root
->fs_info
->mapping_tree
;
1233 map_length
= length
;
1234 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1235 &map_length
, NULL
, 0);
1237 if (map_length
< length
+ size
) {
1244 * in order to insert checksums into the metadata in large chunks,
1245 * we wait until bio submission time. All the pages in the bio are
1246 * checksummed and sums are attached onto the ordered extent record.
1248 * At IO completion time the cums attached on the ordered extent record
1249 * are inserted into the btree
1251 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
, struct bio
*bio
,
1252 int mirror_num
, unsigned long bio_flags
)
1254 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1257 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1263 * in order to insert checksums into the metadata in large chunks,
1264 * we wait until bio submission time. All the pages in the bio are
1265 * checksummed and sums are attached onto the ordered extent record.
1267 * At IO completion time the cums attached on the ordered extent record
1268 * are inserted into the btree
1270 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1271 int mirror_num
, unsigned long bio_flags
)
1273 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1274 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1278 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1279 * or reading the csums from the tree before a read
1281 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1282 int mirror_num
, unsigned long bio_flags
)
1284 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1288 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1291 skip_sum
= btrfs_test_flag(inode
, NODATASUM
);
1293 if (!(rw
& (1 << BIO_RW
))) {
1294 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1295 return btrfs_submit_compressed_read(inode
, bio
,
1296 mirror_num
, bio_flags
);
1297 } else if (!skip_sum
)
1298 btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1300 } else if (!skip_sum
) {
1301 /* csum items have already been cloned */
1302 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1304 /* we're doing a write, do the async checksumming */
1305 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1306 inode
, rw
, bio
, mirror_num
,
1307 bio_flags
, __btrfs_submit_bio_start
,
1308 __btrfs_submit_bio_done
);
1312 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1316 * given a list of ordered sums record them in the inode. This happens
1317 * at IO completion time based on sums calculated at bio submission time.
1319 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1320 struct inode
*inode
, u64 file_offset
,
1321 struct list_head
*list
)
1323 struct list_head
*cur
;
1324 struct btrfs_ordered_sum
*sum
;
1326 btrfs_set_trans_block_group(trans
, inode
);
1327 list_for_each(cur
, list
) {
1328 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
1329 btrfs_csum_file_blocks(trans
,
1330 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1335 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
1337 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0) {
1340 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1344 /* see btrfs_writepage_start_hook for details on why this is required */
1345 struct btrfs_writepage_fixup
{
1347 struct btrfs_work work
;
1350 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1352 struct btrfs_writepage_fixup
*fixup
;
1353 struct btrfs_ordered_extent
*ordered
;
1355 struct inode
*inode
;
1359 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1363 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1364 ClearPageChecked(page
);
1368 inode
= page
->mapping
->host
;
1369 page_start
= page_offset(page
);
1370 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1372 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1374 /* already ordered? We're done */
1375 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1376 EXTENT_ORDERED
, 0)) {
1380 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1382 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
1383 page_end
, GFP_NOFS
);
1385 btrfs_start_ordered_extent(inode
, ordered
, 1);
1389 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
1390 ClearPageChecked(page
);
1392 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
1395 page_cache_release(page
);
1399 * There are a few paths in the higher layers of the kernel that directly
1400 * set the page dirty bit without asking the filesystem if it is a
1401 * good idea. This causes problems because we want to make sure COW
1402 * properly happens and the data=ordered rules are followed.
1404 * In our case any range that doesn't have the ORDERED bit set
1405 * hasn't been properly setup for IO. We kick off an async process
1406 * to fix it up. The async helper will wait for ordered extents, set
1407 * the delalloc bit and make it safe to write the page.
1409 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1411 struct inode
*inode
= page
->mapping
->host
;
1412 struct btrfs_writepage_fixup
*fixup
;
1413 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1416 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
1421 if (PageChecked(page
))
1424 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1428 SetPageChecked(page
);
1429 page_cache_get(page
);
1430 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1432 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1436 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1437 struct inode
*inode
, u64 file_pos
,
1438 u64 disk_bytenr
, u64 disk_num_bytes
,
1439 u64 num_bytes
, u64 ram_bytes
,
1440 u8 compression
, u8 encryption
,
1441 u16 other_encoding
, int extent_type
)
1443 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1444 struct btrfs_file_extent_item
*fi
;
1445 struct btrfs_path
*path
;
1446 struct extent_buffer
*leaf
;
1447 struct btrfs_key ins
;
1451 path
= btrfs_alloc_path();
1454 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1455 file_pos
+ num_bytes
, file_pos
, &hint
);
1458 ins
.objectid
= inode
->i_ino
;
1459 ins
.offset
= file_pos
;
1460 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1461 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1463 leaf
= path
->nodes
[0];
1464 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1465 struct btrfs_file_extent_item
);
1466 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1467 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1468 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1469 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1470 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1471 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1472 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1473 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1474 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1475 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1476 btrfs_mark_buffer_dirty(leaf
);
1478 inode_add_bytes(inode
, num_bytes
);
1479 btrfs_drop_extent_cache(inode
, file_pos
, file_pos
+ num_bytes
- 1, 0);
1481 ins
.objectid
= disk_bytenr
;
1482 ins
.offset
= disk_num_bytes
;
1483 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1484 ret
= btrfs_alloc_reserved_extent(trans
, root
, leaf
->start
,
1485 root
->root_key
.objectid
,
1486 trans
->transid
, inode
->i_ino
, &ins
);
1489 btrfs_free_path(path
);
1493 /* as ordered data IO finishes, this gets called so we can finish
1494 * an ordered extent if the range of bytes in the file it covers are
1497 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1499 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1500 struct btrfs_trans_handle
*trans
;
1501 struct btrfs_ordered_extent
*ordered_extent
;
1502 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1506 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1510 trans
= btrfs_join_transaction(root
, 1);
1512 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1513 BUG_ON(!ordered_extent
);
1514 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1517 lock_extent(io_tree
, ordered_extent
->file_offset
,
1518 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1521 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1523 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1525 ret
= btrfs_mark_extent_written(trans
, root
, inode
,
1526 ordered_extent
->file_offset
,
1527 ordered_extent
->file_offset
+
1528 ordered_extent
->len
);
1531 ret
= insert_reserved_file_extent(trans
, inode
,
1532 ordered_extent
->file_offset
,
1533 ordered_extent
->start
,
1534 ordered_extent
->disk_len
,
1535 ordered_extent
->len
,
1536 ordered_extent
->len
,
1538 BTRFS_FILE_EXTENT_REG
);
1541 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1542 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1545 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1546 &ordered_extent
->list
);
1548 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1549 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1550 btrfs_update_inode(trans
, root
, inode
);
1551 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1552 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1555 btrfs_put_ordered_extent(ordered_extent
);
1556 /* once for the tree */
1557 btrfs_put_ordered_extent(ordered_extent
);
1559 btrfs_end_transaction(trans
, root
);
1563 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1564 struct extent_state
*state
, int uptodate
)
1566 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1570 * When IO fails, either with EIO or csum verification fails, we
1571 * try other mirrors that might have a good copy of the data. This
1572 * io_failure_record is used to record state as we go through all the
1573 * mirrors. If another mirror has good data, the page is set up to date
1574 * and things continue. If a good mirror can't be found, the original
1575 * bio end_io callback is called to indicate things have failed.
1577 struct io_failure_record
{
1582 unsigned long bio_flags
;
1586 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1587 struct page
*page
, u64 start
, u64 end
,
1588 struct extent_state
*state
)
1590 struct io_failure_record
*failrec
= NULL
;
1592 struct extent_map
*em
;
1593 struct inode
*inode
= page
->mapping
->host
;
1594 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1595 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1602 ret
= get_state_private(failure_tree
, start
, &private);
1604 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1607 failrec
->start
= start
;
1608 failrec
->len
= end
- start
+ 1;
1609 failrec
->last_mirror
= 0;
1610 failrec
->bio_flags
= 0;
1612 spin_lock(&em_tree
->lock
);
1613 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1614 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1615 free_extent_map(em
);
1618 spin_unlock(&em_tree
->lock
);
1620 if (!em
|| IS_ERR(em
)) {
1624 logical
= start
- em
->start
;
1625 logical
= em
->block_start
+ logical
;
1626 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1627 logical
= em
->block_start
;
1628 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1630 failrec
->logical
= logical
;
1631 free_extent_map(em
);
1632 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1633 EXTENT_DIRTY
, GFP_NOFS
);
1634 set_state_private(failure_tree
, start
,
1635 (u64
)(unsigned long)failrec
);
1637 failrec
= (struct io_failure_record
*)(unsigned long)private;
1639 num_copies
= btrfs_num_copies(
1640 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1641 failrec
->logical
, failrec
->len
);
1642 failrec
->last_mirror
++;
1644 spin_lock_irq(&BTRFS_I(inode
)->io_tree
.lock
);
1645 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1648 if (state
&& state
->start
!= failrec
->start
)
1650 spin_unlock_irq(&BTRFS_I(inode
)->io_tree
.lock
);
1652 if (!state
|| failrec
->last_mirror
> num_copies
) {
1653 set_state_private(failure_tree
, failrec
->start
, 0);
1654 clear_extent_bits(failure_tree
, failrec
->start
,
1655 failrec
->start
+ failrec
->len
- 1,
1656 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1660 bio
= bio_alloc(GFP_NOFS
, 1);
1661 bio
->bi_private
= state
;
1662 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1663 bio
->bi_sector
= failrec
->logical
>> 9;
1664 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1667 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1668 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1673 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1674 failrec
->last_mirror
,
1675 failrec
->bio_flags
);
1680 * each time an IO finishes, we do a fast check in the IO failure tree
1681 * to see if we need to process or clean up an io_failure_record
1683 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1686 u64 private_failure
;
1687 struct io_failure_record
*failure
;
1691 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1692 (u64
)-1, 1, EXTENT_DIRTY
)) {
1693 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1694 start
, &private_failure
);
1696 failure
= (struct io_failure_record
*)(unsigned long)
1698 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1700 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1702 failure
->start
+ failure
->len
- 1,
1703 EXTENT_DIRTY
| EXTENT_LOCKED
,
1712 * when reads are done, we need to check csums to verify the data is correct
1713 * if there's a match, we allow the bio to finish. If not, we go through
1714 * the io_failure_record routines to find good copies
1716 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1717 struct extent_state
*state
)
1719 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1720 struct inode
*inode
= page
->mapping
->host
;
1721 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1723 u64
private = ~(u32
)0;
1725 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1727 unsigned long flags
;
1729 if (PageChecked(page
)) {
1730 ClearPageChecked(page
);
1733 if (btrfs_test_flag(inode
, NODATASUM
))
1736 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1737 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1)) {
1738 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1743 if (state
&& state
->start
== start
) {
1744 private = state
->private;
1747 ret
= get_state_private(io_tree
, start
, &private);
1749 local_irq_save(flags
);
1750 kaddr
= kmap_atomic(page
, KM_IRQ0
);
1754 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1755 btrfs_csum_final(csum
, (char *)&csum
);
1756 if (csum
!= private) {
1759 kunmap_atomic(kaddr
, KM_IRQ0
);
1760 local_irq_restore(flags
);
1762 /* if the io failure tree for this inode is non-empty,
1763 * check to see if we've recovered from a failed IO
1765 btrfs_clean_io_failures(inode
, start
);
1769 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1770 page
->mapping
->host
->i_ino
, (unsigned long long)start
, csum
,
1772 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1773 flush_dcache_page(page
);
1774 kunmap_atomic(kaddr
, KM_IRQ0
);
1775 local_irq_restore(flags
);
1782 * This creates an orphan entry for the given inode in case something goes
1783 * wrong in the middle of an unlink/truncate.
1785 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1787 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1790 spin_lock(&root
->list_lock
);
1792 /* already on the orphan list, we're good */
1793 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1794 spin_unlock(&root
->list_lock
);
1798 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1800 spin_unlock(&root
->list_lock
);
1803 * insert an orphan item to track this unlinked/truncated file
1805 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1811 * We have done the truncate/delete so we can go ahead and remove the orphan
1812 * item for this particular inode.
1814 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1816 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1819 spin_lock(&root
->list_lock
);
1821 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1822 spin_unlock(&root
->list_lock
);
1826 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1828 spin_unlock(&root
->list_lock
);
1832 spin_unlock(&root
->list_lock
);
1834 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1840 * this cleans up any orphans that may be left on the list from the last use
1843 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1845 struct btrfs_path
*path
;
1846 struct extent_buffer
*leaf
;
1847 struct btrfs_item
*item
;
1848 struct btrfs_key key
, found_key
;
1849 struct btrfs_trans_handle
*trans
;
1850 struct inode
*inode
;
1851 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1853 path
= btrfs_alloc_path();
1858 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1859 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1860 key
.offset
= (u64
)-1;
1864 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1866 printk(KERN_ERR
"Error searching slot for orphan: %d"
1872 * if ret == 0 means we found what we were searching for, which
1873 * is weird, but possible, so only screw with path if we didnt
1874 * find the key and see if we have stuff that matches
1877 if (path
->slots
[0] == 0)
1882 /* pull out the item */
1883 leaf
= path
->nodes
[0];
1884 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1885 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1887 /* make sure the item matches what we want */
1888 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1890 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1893 /* release the path since we're done with it */
1894 btrfs_release_path(root
, path
);
1897 * this is where we are basically btrfs_lookup, without the
1898 * crossing root thing. we store the inode number in the
1899 * offset of the orphan item.
1901 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
1902 found_key
.offset
, root
);
1906 if (inode
->i_state
& I_NEW
) {
1907 BTRFS_I(inode
)->root
= root
;
1909 /* have to set the location manually */
1910 BTRFS_I(inode
)->location
.objectid
= inode
->i_ino
;
1911 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
1912 BTRFS_I(inode
)->location
.offset
= 0;
1914 btrfs_read_locked_inode(inode
);
1915 unlock_new_inode(inode
);
1919 * add this inode to the orphan list so btrfs_orphan_del does
1920 * the proper thing when we hit it
1922 spin_lock(&root
->list_lock
);
1923 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1924 spin_unlock(&root
->list_lock
);
1927 * if this is a bad inode, means we actually succeeded in
1928 * removing the inode, but not the orphan record, which means
1929 * we need to manually delete the orphan since iput will just
1930 * do a destroy_inode
1932 if (is_bad_inode(inode
)) {
1933 trans
= btrfs_start_transaction(root
, 1);
1934 btrfs_orphan_del(trans
, inode
);
1935 btrfs_end_transaction(trans
, root
);
1940 /* if we have links, this was a truncate, lets do that */
1941 if (inode
->i_nlink
) {
1943 btrfs_truncate(inode
);
1948 /* this will do delete_inode and everything for us */
1953 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
1955 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
1957 btrfs_free_path(path
);
1961 * read an inode from the btree into the in-memory inode
1963 void btrfs_read_locked_inode(struct inode
*inode
)
1965 struct btrfs_path
*path
;
1966 struct extent_buffer
*leaf
;
1967 struct btrfs_inode_item
*inode_item
;
1968 struct btrfs_timespec
*tspec
;
1969 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1970 struct btrfs_key location
;
1971 u64 alloc_group_block
;
1975 path
= btrfs_alloc_path();
1977 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
1979 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
1983 leaf
= path
->nodes
[0];
1984 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1985 struct btrfs_inode_item
);
1987 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
1988 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
1989 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
1990 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
1991 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
1993 tspec
= btrfs_inode_atime(inode_item
);
1994 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1995 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1997 tspec
= btrfs_inode_mtime(inode_item
);
1998 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1999 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2001 tspec
= btrfs_inode_ctime(inode_item
);
2002 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2003 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2005 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2006 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2007 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2008 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2010 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2012 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2013 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2015 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2016 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2017 alloc_group_block
, 0);
2018 btrfs_free_path(path
);
2021 switch (inode
->i_mode
& S_IFMT
) {
2023 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2024 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2025 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2026 inode
->i_fop
= &btrfs_file_operations
;
2027 inode
->i_op
= &btrfs_file_inode_operations
;
2030 inode
->i_fop
= &btrfs_dir_file_operations
;
2031 if (root
== root
->fs_info
->tree_root
)
2032 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2034 inode
->i_op
= &btrfs_dir_inode_operations
;
2037 inode
->i_op
= &btrfs_symlink_inode_operations
;
2038 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2039 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2042 init_special_inode(inode
, inode
->i_mode
, rdev
);
2048 btrfs_free_path(path
);
2049 make_bad_inode(inode
);
2053 * given a leaf and an inode, copy the inode fields into the leaf
2055 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2056 struct extent_buffer
*leaf
,
2057 struct btrfs_inode_item
*item
,
2058 struct inode
*inode
)
2060 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2061 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2062 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2063 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2064 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2066 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2067 inode
->i_atime
.tv_sec
);
2068 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2069 inode
->i_atime
.tv_nsec
);
2071 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2072 inode
->i_mtime
.tv_sec
);
2073 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2074 inode
->i_mtime
.tv_nsec
);
2076 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2077 inode
->i_ctime
.tv_sec
);
2078 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2079 inode
->i_ctime
.tv_nsec
);
2081 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2082 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2083 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2084 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2085 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2086 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2087 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2091 * copy everything in the in-memory inode into the btree.
2093 int noinline
btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2094 struct btrfs_root
*root
,
2095 struct inode
*inode
)
2097 struct btrfs_inode_item
*inode_item
;
2098 struct btrfs_path
*path
;
2099 struct extent_buffer
*leaf
;
2102 path
= btrfs_alloc_path();
2104 ret
= btrfs_lookup_inode(trans
, root
, path
,
2105 &BTRFS_I(inode
)->location
, 1);
2112 leaf
= path
->nodes
[0];
2113 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2114 struct btrfs_inode_item
);
2116 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2117 btrfs_mark_buffer_dirty(leaf
);
2118 btrfs_set_inode_last_trans(trans
, inode
);
2121 btrfs_free_path(path
);
2127 * unlink helper that gets used here in inode.c and in the tree logging
2128 * recovery code. It remove a link in a directory with a given name, and
2129 * also drops the back refs in the inode to the directory
2131 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2132 struct btrfs_root
*root
,
2133 struct inode
*dir
, struct inode
*inode
,
2134 const char *name
, int name_len
)
2136 struct btrfs_path
*path
;
2138 struct extent_buffer
*leaf
;
2139 struct btrfs_dir_item
*di
;
2140 struct btrfs_key key
;
2143 path
= btrfs_alloc_path();
2149 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2150 name
, name_len
, -1);
2159 leaf
= path
->nodes
[0];
2160 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2161 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2164 btrfs_release_path(root
, path
);
2166 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2168 dir
->i_ino
, &index
);
2170 printk("failed to delete reference to %.*s, "
2171 "inode %lu parent %lu\n", name_len
, name
,
2172 inode
->i_ino
, dir
->i_ino
);
2176 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2177 index
, name
, name_len
, -1);
2186 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2187 btrfs_release_path(root
, path
);
2189 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2191 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2193 BTRFS_I(dir
)->log_dirty_trans
= trans
->transid
;
2195 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2199 btrfs_free_path(path
);
2203 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2204 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2205 btrfs_update_inode(trans
, root
, dir
);
2206 btrfs_drop_nlink(inode
);
2207 ret
= btrfs_update_inode(trans
, root
, inode
);
2208 dir
->i_sb
->s_dirt
= 1;
2213 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2215 struct btrfs_root
*root
;
2216 struct btrfs_trans_handle
*trans
;
2217 struct inode
*inode
= dentry
->d_inode
;
2219 unsigned long nr
= 0;
2221 root
= BTRFS_I(dir
)->root
;
2223 ret
= btrfs_check_free_space(root
, 1, 1);
2227 trans
= btrfs_start_transaction(root
, 1);
2229 btrfs_set_trans_block_group(trans
, dir
);
2230 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2231 dentry
->d_name
.name
, dentry
->d_name
.len
);
2233 if (inode
->i_nlink
== 0)
2234 ret
= btrfs_orphan_add(trans
, inode
);
2236 nr
= trans
->blocks_used
;
2238 btrfs_end_transaction_throttle(trans
, root
);
2240 btrfs_btree_balance_dirty(root
, nr
);
2244 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2246 struct inode
*inode
= dentry
->d_inode
;
2249 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2250 struct btrfs_trans_handle
*trans
;
2251 unsigned long nr
= 0;
2254 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2255 * the root of a subvolume or snapshot
2257 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2258 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
) {
2262 ret
= btrfs_check_free_space(root
, 1, 1);
2266 trans
= btrfs_start_transaction(root
, 1);
2267 btrfs_set_trans_block_group(trans
, dir
);
2269 err
= btrfs_orphan_add(trans
, inode
);
2273 /* now the directory is empty */
2274 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2275 dentry
->d_name
.name
, dentry
->d_name
.len
);
2277 btrfs_i_size_write(inode
, 0);
2281 nr
= trans
->blocks_used
;
2282 ret
= btrfs_end_transaction_throttle(trans
, root
);
2284 btrfs_btree_balance_dirty(root
, nr
);
2293 * when truncating bytes in a file, it is possible to avoid reading
2294 * the leaves that contain only checksum items. This can be the
2295 * majority of the IO required to delete a large file, but it must
2296 * be done carefully.
2298 * The keys in the level just above the leaves are checked to make sure
2299 * the lowest key in a given leaf is a csum key, and starts at an offset
2300 * after the new size.
2302 * Then the key for the next leaf is checked to make sure it also has
2303 * a checksum item for the same file. If it does, we know our target leaf
2304 * contains only checksum items, and it can be safely freed without reading
2307 * This is just an optimization targeted at large files. It may do
2308 * nothing. It will return 0 unless things went badly.
2310 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
2311 struct btrfs_root
*root
,
2312 struct btrfs_path
*path
,
2313 struct inode
*inode
, u64 new_size
)
2315 struct btrfs_key key
;
2318 struct btrfs_key found_key
;
2319 struct btrfs_key other_key
;
2320 struct btrfs_leaf_ref
*ref
;
2324 path
->lowest_level
= 1;
2325 key
.objectid
= inode
->i_ino
;
2326 key
.type
= BTRFS_CSUM_ITEM_KEY
;
2327 key
.offset
= new_size
;
2329 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2333 if (path
->nodes
[1] == NULL
) {
2338 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
2339 nritems
= btrfs_header_nritems(path
->nodes
[1]);
2344 if (path
->slots
[1] >= nritems
)
2347 /* did we find a key greater than anything we want to delete? */
2348 if (found_key
.objectid
> inode
->i_ino
||
2349 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
2352 /* we check the next key in the node to make sure the leave contains
2353 * only checksum items. This comparison doesn't work if our
2354 * leaf is the last one in the node
2356 if (path
->slots
[1] + 1 >= nritems
) {
2358 /* search forward from the last key in the node, this
2359 * will bring us into the next node in the tree
2361 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
2363 /* unlikely, but we inc below, so check to be safe */
2364 if (found_key
.offset
== (u64
)-1)
2367 /* search_forward needs a path with locks held, do the
2368 * search again for the original key. It is possible
2369 * this will race with a balance and return a path that
2370 * we could modify, but this drop is just an optimization
2371 * and is allowed to miss some leaves.
2373 btrfs_release_path(root
, path
);
2376 /* setup a max key for search_forward */
2377 other_key
.offset
= (u64
)-1;
2378 other_key
.type
= key
.type
;
2379 other_key
.objectid
= key
.objectid
;
2381 path
->keep_locks
= 1;
2382 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
2384 path
->keep_locks
= 0;
2385 if (ret
|| found_key
.objectid
!= key
.objectid
||
2386 found_key
.type
!= key
.type
) {
2391 key
.offset
= found_key
.offset
;
2392 btrfs_release_path(root
, path
);
2397 /* we know there's one more slot after us in the tree,
2398 * read that key so we can verify it is also a checksum item
2400 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
2402 if (found_key
.objectid
< inode
->i_ino
)
2405 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
2409 * if the key for the next leaf isn't a csum key from this objectid,
2410 * we can't be sure there aren't good items inside this leaf.
2413 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
2416 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
2417 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
2419 * it is safe to delete this leaf, it contains only
2420 * csum items from this inode at an offset >= new_size
2422 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
2425 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
2426 ref
= btrfs_alloc_leaf_ref(root
, 0);
2428 ref
->root_gen
= root
->root_key
.offset
;
2429 ref
->bytenr
= leaf_start
;
2431 ref
->generation
= leaf_gen
;
2434 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
2436 btrfs_free_leaf_ref(root
, ref
);
2442 btrfs_release_path(root
, path
);
2444 if (other_key
.objectid
== inode
->i_ino
&&
2445 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
2446 key
.offset
= other_key
.offset
;
2452 /* fixup any changes we've made to the path */
2453 path
->lowest_level
= 0;
2454 path
->keep_locks
= 0;
2455 btrfs_release_path(root
, path
);
2462 * this can truncate away extent items, csum items and directory items.
2463 * It starts at a high offset and removes keys until it can't find
2464 * any higher than new_size
2466 * csum items that cross the new i_size are truncated to the new size
2469 * min_type is the minimum key type to truncate down to. If set to 0, this
2470 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2472 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2473 struct btrfs_root
*root
,
2474 struct inode
*inode
,
2475 u64 new_size
, u32 min_type
)
2478 struct btrfs_path
*path
;
2479 struct btrfs_key key
;
2480 struct btrfs_key found_key
;
2482 struct extent_buffer
*leaf
;
2483 struct btrfs_file_extent_item
*fi
;
2484 u64 extent_start
= 0;
2485 u64 extent_num_bytes
= 0;
2491 int pending_del_nr
= 0;
2492 int pending_del_slot
= 0;
2493 int extent_type
= -1;
2495 u64 mask
= root
->sectorsize
- 1;
2498 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2499 path
= btrfs_alloc_path();
2503 /* FIXME, add redo link to tree so we don't leak on crash */
2504 key
.objectid
= inode
->i_ino
;
2505 key
.offset
= (u64
)-1;
2508 btrfs_init_path(path
);
2511 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2516 /* there are no items in the tree for us to truncate, we're
2519 if (path
->slots
[0] == 0) {
2528 leaf
= path
->nodes
[0];
2529 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2530 found_type
= btrfs_key_type(&found_key
);
2533 if (found_key
.objectid
!= inode
->i_ino
)
2536 if (found_type
< min_type
)
2539 item_end
= found_key
.offset
;
2540 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2541 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2542 struct btrfs_file_extent_item
);
2543 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2544 encoding
= btrfs_file_extent_compression(leaf
, fi
);
2545 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
2546 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
2548 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2550 btrfs_file_extent_num_bytes(leaf
, fi
);
2551 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2552 item_end
+= btrfs_file_extent_inline_len(leaf
,
2557 if (item_end
< new_size
) {
2558 if (found_type
== BTRFS_DIR_ITEM_KEY
) {
2559 found_type
= BTRFS_INODE_ITEM_KEY
;
2560 } else if (found_type
== BTRFS_EXTENT_ITEM_KEY
) {
2561 found_type
= BTRFS_EXTENT_DATA_KEY
;
2562 } else if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2563 found_type
= BTRFS_XATTR_ITEM_KEY
;
2564 } else if (found_type
== BTRFS_XATTR_ITEM_KEY
) {
2565 found_type
= BTRFS_INODE_REF_KEY
;
2566 } else if (found_type
) {
2571 btrfs_set_key_type(&key
, found_type
);
2574 if (found_key
.offset
>= new_size
)
2580 /* FIXME, shrink the extent if the ref count is only 1 */
2581 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2584 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2586 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2587 if (!del_item
&& !encoding
) {
2588 u64 orig_num_bytes
=
2589 btrfs_file_extent_num_bytes(leaf
, fi
);
2590 extent_num_bytes
= new_size
-
2591 found_key
.offset
+ root
->sectorsize
- 1;
2592 extent_num_bytes
= extent_num_bytes
&
2593 ~((u64
)root
->sectorsize
- 1);
2594 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2596 num_dec
= (orig_num_bytes
-
2598 if (root
->ref_cows
&& extent_start
!= 0)
2599 inode_sub_bytes(inode
, num_dec
);
2600 btrfs_mark_buffer_dirty(leaf
);
2603 btrfs_file_extent_disk_num_bytes(leaf
,
2605 /* FIXME blocksize != 4096 */
2606 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2607 if (extent_start
!= 0) {
2610 inode_sub_bytes(inode
, num_dec
);
2612 root_gen
= btrfs_header_generation(leaf
);
2613 root_owner
= btrfs_header_owner(leaf
);
2615 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2617 * we can't truncate inline items that have had
2621 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2622 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2623 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2624 u32 size
= new_size
- found_key
.offset
;
2626 if (root
->ref_cows
) {
2627 inode_sub_bytes(inode
, item_end
+ 1 -
2631 btrfs_file_extent_calc_inline_size(size
);
2632 ret
= btrfs_truncate_item(trans
, root
, path
,
2635 } else if (root
->ref_cows
) {
2636 inode_sub_bytes(inode
, item_end
+ 1 -
2642 if (!pending_del_nr
) {
2643 /* no pending yet, add ourselves */
2644 pending_del_slot
= path
->slots
[0];
2646 } else if (pending_del_nr
&&
2647 path
->slots
[0] + 1 == pending_del_slot
) {
2648 /* hop on the pending chunk */
2650 pending_del_slot
= path
->slots
[0];
2652 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path
->slots
[0], pending_del_nr
, pending_del_slot
);
2658 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2660 leaf
->start
, root_owner
,
2661 root_gen
, inode
->i_ino
, 0);
2665 if (path
->slots
[0] == 0) {
2668 btrfs_release_path(root
, path
);
2673 if (pending_del_nr
&&
2674 path
->slots
[0] + 1 != pending_del_slot
) {
2675 struct btrfs_key debug
;
2677 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2679 ret
= btrfs_del_items(trans
, root
, path
,
2684 btrfs_release_path(root
, path
);
2690 if (pending_del_nr
) {
2691 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2694 btrfs_free_path(path
);
2695 inode
->i_sb
->s_dirt
= 1;
2700 * taken from block_truncate_page, but does cow as it zeros out
2701 * any bytes left in the last page in the file.
2703 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2705 struct inode
*inode
= mapping
->host
;
2706 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2707 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2708 struct btrfs_ordered_extent
*ordered
;
2710 u32 blocksize
= root
->sectorsize
;
2711 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2712 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2718 if ((offset
& (blocksize
- 1)) == 0)
2723 page
= grab_cache_page(mapping
, index
);
2727 page_start
= page_offset(page
);
2728 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2730 if (!PageUptodate(page
)) {
2731 ret
= btrfs_readpage(NULL
, page
);
2733 if (page
->mapping
!= mapping
) {
2735 page_cache_release(page
);
2738 if (!PageUptodate(page
)) {
2743 wait_on_page_writeback(page
);
2745 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2746 set_page_extent_mapped(page
);
2748 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2750 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2752 page_cache_release(page
);
2753 btrfs_start_ordered_extent(inode
, ordered
, 1);
2754 btrfs_put_ordered_extent(ordered
);
2758 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2760 if (offset
!= PAGE_CACHE_SIZE
) {
2762 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2763 flush_dcache_page(page
);
2766 ClearPageChecked(page
);
2767 set_page_dirty(page
);
2768 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2772 page_cache_release(page
);
2777 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
2779 struct btrfs_trans_handle
*trans
;
2780 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2781 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2782 struct extent_map
*em
;
2783 u64 mask
= root
->sectorsize
- 1;
2784 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2785 u64 block_end
= (size
+ mask
) & ~mask
;
2791 if (size
<= hole_start
)
2794 err
= btrfs_check_free_space(root
, 1, 0);
2798 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2801 struct btrfs_ordered_extent
*ordered
;
2802 btrfs_wait_ordered_range(inode
, hole_start
,
2803 block_end
- hole_start
);
2804 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2805 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2808 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2809 btrfs_put_ordered_extent(ordered
);
2812 trans
= btrfs_start_transaction(root
, 1);
2813 btrfs_set_trans_block_group(trans
, inode
);
2815 cur_offset
= hole_start
;
2817 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2818 block_end
- cur_offset
, 0);
2819 BUG_ON(IS_ERR(em
) || !em
);
2820 last_byte
= min(extent_map_end(em
), block_end
);
2821 last_byte
= (last_byte
+ mask
) & ~mask
;
2822 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2824 hole_size
= last_byte
- cur_offset
;
2825 err
= btrfs_drop_extents(trans
, root
, inode
,
2827 cur_offset
+ hole_size
,
2828 cur_offset
, &hint_byte
);
2831 err
= btrfs_insert_file_extent(trans
, root
,
2832 inode
->i_ino
, cur_offset
, 0,
2833 0, hole_size
, 0, hole_size
,
2835 btrfs_drop_extent_cache(inode
, hole_start
,
2838 free_extent_map(em
);
2839 cur_offset
= last_byte
;
2840 if (err
|| cur_offset
>= block_end
)
2844 btrfs_end_transaction(trans
, root
);
2845 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2849 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2851 struct inode
*inode
= dentry
->d_inode
;
2854 err
= inode_change_ok(inode
, attr
);
2858 if (S_ISREG(inode
->i_mode
) &&
2859 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
> inode
->i_size
) {
2860 err
= btrfs_cont_expand(inode
, attr
->ia_size
);
2865 err
= inode_setattr(inode
, attr
);
2867 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2868 err
= btrfs_acl_chmod(inode
);
2872 void btrfs_delete_inode(struct inode
*inode
)
2874 struct btrfs_trans_handle
*trans
;
2875 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2879 truncate_inode_pages(&inode
->i_data
, 0);
2880 if (is_bad_inode(inode
)) {
2881 btrfs_orphan_del(NULL
, inode
);
2884 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
2886 btrfs_i_size_write(inode
, 0);
2887 trans
= btrfs_start_transaction(root
, 1);
2889 btrfs_set_trans_block_group(trans
, inode
);
2890 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
2892 btrfs_orphan_del(NULL
, inode
);
2893 goto no_delete_lock
;
2896 btrfs_orphan_del(trans
, inode
);
2898 nr
= trans
->blocks_used
;
2901 btrfs_end_transaction(trans
, root
);
2902 btrfs_btree_balance_dirty(root
, nr
);
2906 nr
= trans
->blocks_used
;
2907 btrfs_end_transaction(trans
, root
);
2908 btrfs_btree_balance_dirty(root
, nr
);
2914 * this returns the key found in the dir entry in the location pointer.
2915 * If no dir entries were found, location->objectid is 0.
2917 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
2918 struct btrfs_key
*location
)
2920 const char *name
= dentry
->d_name
.name
;
2921 int namelen
= dentry
->d_name
.len
;
2922 struct btrfs_dir_item
*di
;
2923 struct btrfs_path
*path
;
2924 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2927 path
= btrfs_alloc_path();
2930 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
2934 if (!di
|| IS_ERR(di
)) {
2937 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
2939 btrfs_free_path(path
);
2942 location
->objectid
= 0;
2947 * when we hit a tree root in a directory, the btrfs part of the inode
2948 * needs to be changed to reflect the root directory of the tree root. This
2949 * is kind of like crossing a mount point.
2951 static int fixup_tree_root_location(struct btrfs_root
*root
,
2952 struct btrfs_key
*location
,
2953 struct btrfs_root
**sub_root
,
2954 struct dentry
*dentry
)
2956 struct btrfs_root_item
*ri
;
2958 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
2960 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
2963 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
2964 dentry
->d_name
.name
,
2965 dentry
->d_name
.len
);
2966 if (IS_ERR(*sub_root
))
2967 return PTR_ERR(*sub_root
);
2969 ri
= &(*sub_root
)->root_item
;
2970 location
->objectid
= btrfs_root_dirid(ri
);
2971 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2972 location
->offset
= 0;
2977 static noinline
void init_btrfs_i(struct inode
*inode
)
2979 struct btrfs_inode
*bi
= BTRFS_I(inode
);
2982 bi
->i_default_acl
= NULL
;
2987 bi
->logged_trans
= 0;
2988 bi
->delalloc_bytes
= 0;
2989 bi
->disk_i_size
= 0;
2991 bi
->index_cnt
= (u64
)-1;
2992 bi
->log_dirty_trans
= 0;
2993 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
2994 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
2995 inode
->i_mapping
, GFP_NOFS
);
2996 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
2997 inode
->i_mapping
, GFP_NOFS
);
2998 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
2999 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
3000 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
3001 mutex_init(&BTRFS_I(inode
)->log_mutex
);
3004 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3006 struct btrfs_iget_args
*args
= p
;
3007 inode
->i_ino
= args
->ino
;
3008 init_btrfs_i(inode
);
3009 BTRFS_I(inode
)->root
= args
->root
;
3013 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3015 struct btrfs_iget_args
*args
= opaque
;
3016 return (args
->ino
== inode
->i_ino
&&
3017 args
->root
== BTRFS_I(inode
)->root
);
3020 struct inode
*btrfs_ilookup(struct super_block
*s
, u64 objectid
,
3021 struct btrfs_root
*root
, int wait
)
3023 struct inode
*inode
;
3024 struct btrfs_iget_args args
;
3025 args
.ino
= objectid
;
3029 inode
= ilookup5(s
, objectid
, btrfs_find_actor
,
3032 inode
= ilookup5_nowait(s
, objectid
, btrfs_find_actor
,
3038 struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 objectid
,
3039 struct btrfs_root
*root
)
3041 struct inode
*inode
;
3042 struct btrfs_iget_args args
;
3043 args
.ino
= objectid
;
3046 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3047 btrfs_init_locked_inode
,
3052 /* Get an inode object given its location and corresponding root.
3053 * Returns in *is_new if the inode was read from disk
3055 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3056 struct btrfs_root
*root
, int *is_new
)
3058 struct inode
*inode
;
3060 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3062 return ERR_PTR(-EACCES
);
3064 if (inode
->i_state
& I_NEW
) {
3065 BTRFS_I(inode
)->root
= root
;
3066 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3067 btrfs_read_locked_inode(inode
);
3068 unlock_new_inode(inode
);
3079 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3081 struct inode
* inode
;
3082 struct btrfs_inode
*bi
= BTRFS_I(dir
);
3083 struct btrfs_root
*root
= bi
->root
;
3084 struct btrfs_root
*sub_root
= root
;
3085 struct btrfs_key location
;
3088 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3089 return ERR_PTR(-ENAMETOOLONG
);
3091 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3094 return ERR_PTR(ret
);
3097 if (location
.objectid
) {
3098 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
3101 return ERR_PTR(ret
);
3103 return ERR_PTR(-ENOENT
);
3104 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, &new);
3106 return ERR_CAST(inode
);
3111 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
3112 struct nameidata
*nd
)
3114 struct inode
*inode
;
3116 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3117 return ERR_PTR(-ENAMETOOLONG
);
3119 inode
= btrfs_lookup_dentry(dir
, dentry
);
3121 return ERR_CAST(inode
);
3123 return d_splice_alias(inode
, dentry
);
3126 static unsigned char btrfs_filetype_table
[] = {
3127 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
3130 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
3133 struct inode
*inode
= filp
->f_dentry
->d_inode
;
3134 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3135 struct btrfs_item
*item
;
3136 struct btrfs_dir_item
*di
;
3137 struct btrfs_key key
;
3138 struct btrfs_key found_key
;
3139 struct btrfs_path
*path
;
3142 struct extent_buffer
*leaf
;
3145 unsigned char d_type
;
3150 int key_type
= BTRFS_DIR_INDEX_KEY
;
3155 /* FIXME, use a real flag for deciding about the key type */
3156 if (root
->fs_info
->tree_root
== root
)
3157 key_type
= BTRFS_DIR_ITEM_KEY
;
3159 /* special case for "." */
3160 if (filp
->f_pos
== 0) {
3161 over
= filldir(dirent
, ".", 1,
3168 /* special case for .., just use the back ref */
3169 if (filp
->f_pos
== 1) {
3170 u64 pino
= parent_ino(filp
->f_path
.dentry
);
3171 over
= filldir(dirent
, "..", 2,
3177 path
= btrfs_alloc_path();
3180 btrfs_set_key_type(&key
, key_type
);
3181 key
.offset
= filp
->f_pos
;
3182 key
.objectid
= inode
->i_ino
;
3184 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3190 leaf
= path
->nodes
[0];
3191 nritems
= btrfs_header_nritems(leaf
);
3192 slot
= path
->slots
[0];
3193 if (advance
|| slot
>= nritems
) {
3194 if (slot
>= nritems
- 1) {
3195 ret
= btrfs_next_leaf(root
, path
);
3198 leaf
= path
->nodes
[0];
3199 nritems
= btrfs_header_nritems(leaf
);
3200 slot
= path
->slots
[0];
3208 item
= btrfs_item_nr(leaf
, slot
);
3209 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3211 if (found_key
.objectid
!= key
.objectid
)
3213 if (btrfs_key_type(&found_key
) != key_type
)
3215 if (found_key
.offset
< filp
->f_pos
)
3218 filp
->f_pos
= found_key
.offset
;
3220 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
3222 di_total
= btrfs_item_size(leaf
, item
);
3224 while (di_cur
< di_total
) {
3225 struct btrfs_key location
;
3227 name_len
= btrfs_dir_name_len(leaf
, di
);
3228 if (name_len
<= sizeof(tmp_name
)) {
3229 name_ptr
= tmp_name
;
3231 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
3237 read_extent_buffer(leaf
, name_ptr
,
3238 (unsigned long)(di
+ 1), name_len
);
3240 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
3241 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
3243 /* is this a reference to our own snapshot? If so
3246 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
3247 location
.objectid
== root
->root_key
.objectid
) {
3251 over
= filldir(dirent
, name_ptr
, name_len
,
3252 found_key
.offset
, location
.objectid
,
3256 if (name_ptr
!= tmp_name
)
3261 di_len
= btrfs_dir_name_len(leaf
, di
) +
3262 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
3264 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
3268 /* Reached end of directory/root. Bump pos past the last item. */
3269 if (key_type
== BTRFS_DIR_INDEX_KEY
)
3270 filp
->f_pos
= INT_LIMIT(typeof(filp
->f_pos
));
3276 btrfs_free_path(path
);
3280 int btrfs_write_inode(struct inode
*inode
, int wait
)
3282 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3283 struct btrfs_trans_handle
*trans
;
3286 if (root
->fs_info
->btree_inode
== inode
)
3290 trans
= btrfs_join_transaction(root
, 1);
3291 btrfs_set_trans_block_group(trans
, inode
);
3292 ret
= btrfs_commit_transaction(trans
, root
);
3298 * This is somewhat expensive, updating the tree every time the
3299 * inode changes. But, it is most likely to find the inode in cache.
3300 * FIXME, needs more benchmarking...there are no reasons other than performance
3301 * to keep or drop this code.
3303 void btrfs_dirty_inode(struct inode
*inode
)
3305 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3306 struct btrfs_trans_handle
*trans
;
3308 trans
= btrfs_join_transaction(root
, 1);
3309 btrfs_set_trans_block_group(trans
, inode
);
3310 btrfs_update_inode(trans
, root
, inode
);
3311 btrfs_end_transaction(trans
, root
);
3315 * find the highest existing sequence number in a directory
3316 * and then set the in-memory index_cnt variable to reflect
3317 * free sequence numbers
3319 static int btrfs_set_inode_index_count(struct inode
*inode
)
3321 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3322 struct btrfs_key key
, found_key
;
3323 struct btrfs_path
*path
;
3324 struct extent_buffer
*leaf
;
3327 key
.objectid
= inode
->i_ino
;
3328 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
3329 key
.offset
= (u64
)-1;
3331 path
= btrfs_alloc_path();
3335 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3338 /* FIXME: we should be able to handle this */
3344 * MAGIC NUMBER EXPLANATION:
3345 * since we search a directory based on f_pos we have to start at 2
3346 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3347 * else has to start at 2
3349 if (path
->slots
[0] == 0) {
3350 BTRFS_I(inode
)->index_cnt
= 2;
3356 leaf
= path
->nodes
[0];
3357 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3359 if (found_key
.objectid
!= inode
->i_ino
||
3360 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
3361 BTRFS_I(inode
)->index_cnt
= 2;
3365 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
3367 btrfs_free_path(path
);
3372 * helper to find a free sequence number in a given directory. This current
3373 * code is very simple, later versions will do smarter things in the btree
3375 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
3379 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
3380 ret
= btrfs_set_inode_index_count(dir
);
3386 *index
= BTRFS_I(dir
)->index_cnt
;
3387 BTRFS_I(dir
)->index_cnt
++;
3392 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
3393 struct btrfs_root
*root
,
3395 const char *name
, int name_len
,
3396 u64 ref_objectid
, u64 objectid
,
3397 u64 alloc_hint
, int mode
, u64
*index
)
3399 struct inode
*inode
;
3400 struct btrfs_inode_item
*inode_item
;
3401 struct btrfs_key
*location
;
3402 struct btrfs_path
*path
;
3403 struct btrfs_inode_ref
*ref
;
3404 struct btrfs_key key
[2];
3410 path
= btrfs_alloc_path();
3413 inode
= new_inode(root
->fs_info
->sb
);
3415 return ERR_PTR(-ENOMEM
);
3418 ret
= btrfs_set_inode_index(dir
, index
);
3420 return ERR_PTR(ret
);
3423 * index_cnt is ignored for everything but a dir,
3424 * btrfs_get_inode_index_count has an explanation for the magic
3427 init_btrfs_i(inode
);
3428 BTRFS_I(inode
)->index_cnt
= 2;
3429 BTRFS_I(inode
)->root
= root
;
3430 BTRFS_I(inode
)->generation
= trans
->transid
;
3436 BTRFS_I(inode
)->block_group
=
3437 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
3438 if ((mode
& S_IFREG
)) {
3439 if (btrfs_test_opt(root
, NODATASUM
))
3440 btrfs_set_flag(inode
, NODATASUM
);
3441 if (btrfs_test_opt(root
, NODATACOW
))
3442 btrfs_set_flag(inode
, NODATACOW
);
3445 key
[0].objectid
= objectid
;
3446 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
3449 key
[1].objectid
= objectid
;
3450 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
3451 key
[1].offset
= ref_objectid
;
3453 sizes
[0] = sizeof(struct btrfs_inode_item
);
3454 sizes
[1] = name_len
+ sizeof(*ref
);
3456 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
3460 if (objectid
> root
->highest_inode
)
3461 root
->highest_inode
= objectid
;
3463 inode
->i_uid
= current_fsuid();
3464 inode
->i_gid
= current_fsgid();
3465 inode
->i_mode
= mode
;
3466 inode
->i_ino
= objectid
;
3467 inode_set_bytes(inode
, 0);
3468 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3469 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3470 struct btrfs_inode_item
);
3471 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
3473 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3474 struct btrfs_inode_ref
);
3475 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
3476 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
3477 ptr
= (unsigned long)(ref
+ 1);
3478 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
3480 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3481 btrfs_free_path(path
);
3483 location
= &BTRFS_I(inode
)->location
;
3484 location
->objectid
= objectid
;
3485 location
->offset
= 0;
3486 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3488 insert_inode_hash(inode
);
3492 BTRFS_I(dir
)->index_cnt
--;
3493 btrfs_free_path(path
);
3494 return ERR_PTR(ret
);
3497 static inline u8
btrfs_inode_type(struct inode
*inode
)
3499 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
3503 * utility function to add 'inode' into 'parent_inode' with
3504 * a give name and a given sequence number.
3505 * if 'add_backref' is true, also insert a backref from the
3506 * inode to the parent directory.
3508 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3509 struct inode
*parent_inode
, struct inode
*inode
,
3510 const char *name
, int name_len
, int add_backref
, u64 index
)
3513 struct btrfs_key key
;
3514 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3516 key
.objectid
= inode
->i_ino
;
3517 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3520 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3521 parent_inode
->i_ino
,
3522 &key
, btrfs_inode_type(inode
),
3526 ret
= btrfs_insert_inode_ref(trans
, root
,
3529 parent_inode
->i_ino
,
3532 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3534 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3535 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3540 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3541 struct dentry
*dentry
, struct inode
*inode
,
3542 int backref
, u64 index
)
3544 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3545 inode
, dentry
->d_name
.name
,
3546 dentry
->d_name
.len
, backref
, index
);
3548 d_instantiate(dentry
, inode
);
3556 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3557 int mode
, dev_t rdev
)
3559 struct btrfs_trans_handle
*trans
;
3560 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3561 struct inode
*inode
= NULL
;
3565 unsigned long nr
= 0;
3568 if (!new_valid_dev(rdev
))
3571 err
= btrfs_check_free_space(root
, 1, 0);
3575 trans
= btrfs_start_transaction(root
, 1);
3576 btrfs_set_trans_block_group(trans
, dir
);
3578 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3584 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3586 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3587 BTRFS_I(dir
)->block_group
, mode
, &index
);
3588 err
= PTR_ERR(inode
);
3592 err
= btrfs_init_acl(inode
, dir
);
3598 btrfs_set_trans_block_group(trans
, inode
);
3599 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3603 inode
->i_op
= &btrfs_special_inode_operations
;
3604 init_special_inode(inode
, inode
->i_mode
, rdev
);
3605 btrfs_update_inode(trans
, root
, inode
);
3607 dir
->i_sb
->s_dirt
= 1;
3608 btrfs_update_inode_block_group(trans
, inode
);
3609 btrfs_update_inode_block_group(trans
, dir
);
3611 nr
= trans
->blocks_used
;
3612 btrfs_end_transaction_throttle(trans
, root
);
3615 inode_dec_link_count(inode
);
3618 btrfs_btree_balance_dirty(root
, nr
);
3622 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3623 int mode
, struct nameidata
*nd
)
3625 struct btrfs_trans_handle
*trans
;
3626 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3627 struct inode
*inode
= NULL
;
3630 unsigned long nr
= 0;
3634 err
= btrfs_check_free_space(root
, 1, 0);
3637 trans
= btrfs_start_transaction(root
, 1);
3638 btrfs_set_trans_block_group(trans
, dir
);
3640 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3646 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3648 dentry
->d_parent
->d_inode
->i_ino
,
3649 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3651 err
= PTR_ERR(inode
);
3655 err
= btrfs_init_acl(inode
, dir
);
3661 btrfs_set_trans_block_group(trans
, inode
);
3662 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3666 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3667 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3668 inode
->i_fop
= &btrfs_file_operations
;
3669 inode
->i_op
= &btrfs_file_inode_operations
;
3670 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3672 dir
->i_sb
->s_dirt
= 1;
3673 btrfs_update_inode_block_group(trans
, inode
);
3674 btrfs_update_inode_block_group(trans
, dir
);
3676 nr
= trans
->blocks_used
;
3677 btrfs_end_transaction_throttle(trans
, root
);
3680 inode_dec_link_count(inode
);
3683 btrfs_btree_balance_dirty(root
, nr
);
3687 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3688 struct dentry
*dentry
)
3690 struct btrfs_trans_handle
*trans
;
3691 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3692 struct inode
*inode
= old_dentry
->d_inode
;
3694 unsigned long nr
= 0;
3698 if (inode
->i_nlink
== 0)
3701 btrfs_inc_nlink(inode
);
3702 err
= btrfs_check_free_space(root
, 1, 0);
3705 err
= btrfs_set_inode_index(dir
, &index
);
3709 trans
= btrfs_start_transaction(root
, 1);
3711 btrfs_set_trans_block_group(trans
, dir
);
3712 atomic_inc(&inode
->i_count
);
3714 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3719 dir
->i_sb
->s_dirt
= 1;
3720 btrfs_update_inode_block_group(trans
, dir
);
3721 err
= btrfs_update_inode(trans
, root
, inode
);
3726 nr
= trans
->blocks_used
;
3727 btrfs_end_transaction_throttle(trans
, root
);
3730 inode_dec_link_count(inode
);
3733 btrfs_btree_balance_dirty(root
, nr
);
3737 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3739 struct inode
*inode
= NULL
;
3740 struct btrfs_trans_handle
*trans
;
3741 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3743 int drop_on_err
= 0;
3746 unsigned long nr
= 1;
3748 err
= btrfs_check_free_space(root
, 1, 0);
3752 trans
= btrfs_start_transaction(root
, 1);
3753 btrfs_set_trans_block_group(trans
, dir
);
3755 if (IS_ERR(trans
)) {
3756 err
= PTR_ERR(trans
);
3760 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3766 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3768 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3769 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3771 if (IS_ERR(inode
)) {
3772 err
= PTR_ERR(inode
);
3778 err
= btrfs_init_acl(inode
, dir
);
3782 inode
->i_op
= &btrfs_dir_inode_operations
;
3783 inode
->i_fop
= &btrfs_dir_file_operations
;
3784 btrfs_set_trans_block_group(trans
, inode
);
3786 btrfs_i_size_write(inode
, 0);
3787 err
= btrfs_update_inode(trans
, root
, inode
);
3791 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3792 inode
, dentry
->d_name
.name
,
3793 dentry
->d_name
.len
, 0, index
);
3797 d_instantiate(dentry
, inode
);
3799 dir
->i_sb
->s_dirt
= 1;
3800 btrfs_update_inode_block_group(trans
, inode
);
3801 btrfs_update_inode_block_group(trans
, dir
);
3804 nr
= trans
->blocks_used
;
3805 btrfs_end_transaction_throttle(trans
, root
);
3810 btrfs_btree_balance_dirty(root
, nr
);
3814 /* helper for btfs_get_extent. Given an existing extent in the tree,
3815 * and an extent that you want to insert, deal with overlap and insert
3816 * the new extent into the tree.
3818 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3819 struct extent_map
*existing
,
3820 struct extent_map
*em
,
3821 u64 map_start
, u64 map_len
)
3825 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3826 start_diff
= map_start
- em
->start
;
3827 em
->start
= map_start
;
3829 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3830 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3831 em
->block_start
+= start_diff
;
3832 em
->block_len
-= start_diff
;
3834 return add_extent_mapping(em_tree
, em
);
3837 static noinline
int uncompress_inline(struct btrfs_path
*path
,
3838 struct inode
*inode
, struct page
*page
,
3839 size_t pg_offset
, u64 extent_offset
,
3840 struct btrfs_file_extent_item
*item
)
3843 struct extent_buffer
*leaf
= path
->nodes
[0];
3846 unsigned long inline_size
;
3849 WARN_ON(pg_offset
!= 0);
3850 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
3851 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
3852 btrfs_item_nr(leaf
, path
->slots
[0]));
3853 tmp
= kmalloc(inline_size
, GFP_NOFS
);
3854 ptr
= btrfs_file_extent_inline_start(item
);
3856 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
3858 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
3859 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
3860 inline_size
, max_size
);
3862 char *kaddr
= kmap_atomic(page
, KM_USER0
);
3863 unsigned long copy_size
= min_t(u64
,
3864 PAGE_CACHE_SIZE
- pg_offset
,
3865 max_size
- extent_offset
);
3866 memset(kaddr
+ pg_offset
, 0, copy_size
);
3867 kunmap_atomic(kaddr
, KM_USER0
);
3874 * a bit scary, this does extent mapping from logical file offset to the disk.
3875 * the ugly parts come from merging extents from the disk with the
3876 * in-ram representation. This gets more complex because of the data=ordered code,
3877 * where the in-ram extents might be locked pending data=ordered completion.
3879 * This also copies inline extents directly into the page.
3881 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
3882 size_t pg_offset
, u64 start
, u64 len
,
3888 u64 extent_start
= 0;
3890 u64 objectid
= inode
->i_ino
;
3892 struct btrfs_path
*path
= NULL
;
3893 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3894 struct btrfs_file_extent_item
*item
;
3895 struct extent_buffer
*leaf
;
3896 struct btrfs_key found_key
;
3897 struct extent_map
*em
= NULL
;
3898 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3899 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3900 struct btrfs_trans_handle
*trans
= NULL
;
3904 spin_lock(&em_tree
->lock
);
3905 em
= lookup_extent_mapping(em_tree
, start
, len
);
3907 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3908 spin_unlock(&em_tree
->lock
);
3911 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
3912 free_extent_map(em
);
3913 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
3914 free_extent_map(em
);
3918 em
= alloc_extent_map(GFP_NOFS
);
3923 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3924 em
->start
= EXTENT_MAP_HOLE
;
3925 em
->orig_start
= EXTENT_MAP_HOLE
;
3927 em
->block_len
= (u64
)-1;
3930 path
= btrfs_alloc_path();
3934 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3935 objectid
, start
, trans
!= NULL
);
3942 if (path
->slots
[0] == 0)
3947 leaf
= path
->nodes
[0];
3948 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3949 struct btrfs_file_extent_item
);
3950 /* are we inside the extent that was found? */
3951 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3952 found_type
= btrfs_key_type(&found_key
);
3953 if (found_key
.objectid
!= objectid
||
3954 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3958 found_type
= btrfs_file_extent_type(leaf
, item
);
3959 extent_start
= found_key
.offset
;
3960 compressed
= btrfs_file_extent_compression(leaf
, item
);
3961 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3962 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3963 extent_end
= extent_start
+
3964 btrfs_file_extent_num_bytes(leaf
, item
);
3965 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3967 size
= btrfs_file_extent_inline_len(leaf
, item
);
3968 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
3969 ~((u64
)root
->sectorsize
- 1);
3972 if (start
>= extent_end
) {
3974 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3975 ret
= btrfs_next_leaf(root
, path
);
3982 leaf
= path
->nodes
[0];
3984 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3985 if (found_key
.objectid
!= objectid
||
3986 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3988 if (start
+ len
<= found_key
.offset
)
3991 em
->len
= found_key
.offset
- start
;
3995 if (found_type
== BTRFS_FILE_EXTENT_REG
||
3996 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
3997 em
->start
= extent_start
;
3998 em
->len
= extent_end
- extent_start
;
3999 em
->orig_start
= extent_start
-
4000 btrfs_file_extent_offset(leaf
, item
);
4001 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
4003 em
->block_start
= EXTENT_MAP_HOLE
;
4007 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4008 em
->block_start
= bytenr
;
4009 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
4012 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
4013 em
->block_start
= bytenr
;
4014 em
->block_len
= em
->len
;
4015 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
4016 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
4019 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
4023 size_t extent_offset
;
4026 em
->block_start
= EXTENT_MAP_INLINE
;
4027 if (!page
|| create
) {
4028 em
->start
= extent_start
;
4029 em
->len
= extent_end
- extent_start
;
4033 size
= btrfs_file_extent_inline_len(leaf
, item
);
4034 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
4035 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
4036 size
- extent_offset
);
4037 em
->start
= extent_start
+ extent_offset
;
4038 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
4039 ~((u64
)root
->sectorsize
- 1);
4040 em
->orig_start
= EXTENT_MAP_INLINE
;
4042 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
4043 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
4044 if (create
== 0 && !PageUptodate(page
)) {
4045 if (btrfs_file_extent_compression(leaf
, item
) ==
4046 BTRFS_COMPRESS_ZLIB
) {
4047 ret
= uncompress_inline(path
, inode
, page
,
4049 extent_offset
, item
);
4053 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4057 flush_dcache_page(page
);
4058 } else if (create
&& PageUptodate(page
)) {
4061 free_extent_map(em
);
4063 btrfs_release_path(root
, path
);
4064 trans
= btrfs_join_transaction(root
, 1);
4068 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
4071 btrfs_mark_buffer_dirty(leaf
);
4073 set_extent_uptodate(io_tree
, em
->start
,
4074 extent_map_end(em
) - 1, GFP_NOFS
);
4077 printk("unkknown found_type %d\n", found_type
);
4084 em
->block_start
= EXTENT_MAP_HOLE
;
4085 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
4087 btrfs_release_path(root
, path
);
4088 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
4089 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em
->start
, em
->len
, start
, len
);
4095 spin_lock(&em_tree
->lock
);
4096 ret
= add_extent_mapping(em_tree
, em
);
4097 /* it is possible that someone inserted the extent into the tree
4098 * while we had the lock dropped. It is also possible that
4099 * an overlapping map exists in the tree
4101 if (ret
== -EEXIST
) {
4102 struct extent_map
*existing
;
4106 existing
= lookup_extent_mapping(em_tree
, start
, len
);
4107 if (existing
&& (existing
->start
> start
||
4108 existing
->start
+ existing
->len
<= start
)) {
4109 free_extent_map(existing
);
4113 existing
= lookup_extent_mapping(em_tree
, em
->start
,
4116 err
= merge_extent_mapping(em_tree
, existing
,
4119 free_extent_map(existing
);
4121 free_extent_map(em
);
4126 printk("failing to insert %Lu %Lu\n",
4128 free_extent_map(em
);
4132 free_extent_map(em
);
4137 spin_unlock(&em_tree
->lock
);
4140 btrfs_free_path(path
);
4142 ret
= btrfs_end_transaction(trans
, root
);
4148 free_extent_map(em
);
4150 return ERR_PTR(err
);
4155 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
4156 const struct iovec
*iov
, loff_t offset
,
4157 unsigned long nr_segs
)
4162 static sector_t
btrfs_bmap(struct address_space
*mapping
, sector_t iblock
)
4164 return extent_bmap(mapping
, iblock
, btrfs_get_extent
);
4167 int btrfs_readpage(struct file
*file
, struct page
*page
)
4169 struct extent_io_tree
*tree
;
4170 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4171 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
4174 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
4176 struct extent_io_tree
*tree
;
4179 if (current
->flags
& PF_MEMALLOC
) {
4180 redirty_page_for_writepage(wbc
, page
);
4184 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4185 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
4188 int btrfs_writepages(struct address_space
*mapping
,
4189 struct writeback_control
*wbc
)
4191 struct extent_io_tree
*tree
;
4193 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4194 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
4198 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
4199 struct list_head
*pages
, unsigned nr_pages
)
4201 struct extent_io_tree
*tree
;
4202 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
4203 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
4206 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4208 struct extent_io_tree
*tree
;
4209 struct extent_map_tree
*map
;
4212 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4213 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
4214 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
4216 ClearPagePrivate(page
);
4217 set_page_private(page
, 0);
4218 page_cache_release(page
);
4223 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
4225 if (PageWriteback(page
) || PageDirty(page
))
4227 return __btrfs_releasepage(page
, gfp_flags
);
4230 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
4232 struct extent_io_tree
*tree
;
4233 struct btrfs_ordered_extent
*ordered
;
4234 u64 page_start
= page_offset(page
);
4235 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4237 wait_on_page_writeback(page
);
4238 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
4240 btrfs_releasepage(page
, GFP_NOFS
);
4244 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4245 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
4249 * IO on this page will never be started, so we need
4250 * to account for any ordered extents now
4252 clear_extent_bit(tree
, page_start
, page_end
,
4253 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4254 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
4255 btrfs_finish_ordered_io(page
->mapping
->host
,
4256 page_start
, page_end
);
4257 btrfs_put_ordered_extent(ordered
);
4258 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
4260 clear_extent_bit(tree
, page_start
, page_end
,
4261 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
4264 __btrfs_releasepage(page
, GFP_NOFS
);
4266 ClearPageChecked(page
);
4267 if (PagePrivate(page
)) {
4268 ClearPagePrivate(page
);
4269 set_page_private(page
, 0);
4270 page_cache_release(page
);
4275 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4276 * called from a page fault handler when a page is first dirtied. Hence we must
4277 * be careful to check for EOF conditions here. We set the page up correctly
4278 * for a written page which means we get ENOSPC checking when writing into
4279 * holes and correct delalloc and unwritten extent mapping on filesystems that
4280 * support these features.
4282 * We are not allowed to take the i_mutex here so we have to play games to
4283 * protect against truncate races as the page could now be beyond EOF. Because
4284 * vmtruncate() writes the inode size before removing pages, once we have the
4285 * page lock we can determine safely if the page is beyond EOF. If it is not
4286 * beyond EOF, then the page is guaranteed safe against truncation until we
4289 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
4291 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
4292 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4293 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4294 struct btrfs_ordered_extent
*ordered
;
4296 unsigned long zero_start
;
4302 ret
= btrfs_check_free_space(root
, PAGE_CACHE_SIZE
, 0);
4309 size
= i_size_read(inode
);
4310 page_start
= page_offset(page
);
4311 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4313 if ((page
->mapping
!= inode
->i_mapping
) ||
4314 (page_start
>= size
)) {
4315 /* page got truncated out from underneath us */
4318 wait_on_page_writeback(page
);
4320 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4321 set_page_extent_mapped(page
);
4324 * we can't set the delalloc bits if there are pending ordered
4325 * extents. Drop our locks and wait for them to finish
4327 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4329 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4331 btrfs_start_ordered_extent(inode
, ordered
, 1);
4332 btrfs_put_ordered_extent(ordered
);
4336 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
4339 /* page is wholly or partially inside EOF */
4340 if (page_start
+ PAGE_CACHE_SIZE
> size
)
4341 zero_start
= size
& ~PAGE_CACHE_MASK
;
4343 zero_start
= PAGE_CACHE_SIZE
;
4345 if (zero_start
!= PAGE_CACHE_SIZE
) {
4347 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
4348 flush_dcache_page(page
);
4351 ClearPageChecked(page
);
4352 set_page_dirty(page
);
4353 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
4361 static void btrfs_truncate(struct inode
*inode
)
4363 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4365 struct btrfs_trans_handle
*trans
;
4367 u64 mask
= root
->sectorsize
- 1;
4369 if (!S_ISREG(inode
->i_mode
))
4371 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
4374 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
4375 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
4377 trans
= btrfs_start_transaction(root
, 1);
4378 btrfs_set_trans_block_group(trans
, inode
);
4379 btrfs_i_size_write(inode
, inode
->i_size
);
4381 ret
= btrfs_orphan_add(trans
, inode
);
4384 /* FIXME, add redo link to tree so we don't leak on crash */
4385 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
4386 BTRFS_EXTENT_DATA_KEY
);
4387 btrfs_update_inode(trans
, root
, inode
);
4389 ret
= btrfs_orphan_del(trans
, inode
);
4393 nr
= trans
->blocks_used
;
4394 ret
= btrfs_end_transaction_throttle(trans
, root
);
4396 btrfs_btree_balance_dirty(root
, nr
);
4400 * create a new subvolume directory/inode (helper for the ioctl).
4402 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
4403 struct btrfs_root
*new_root
, struct dentry
*dentry
,
4404 u64 new_dirid
, u64 alloc_hint
)
4406 struct inode
*inode
;
4410 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
4411 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
4413 return PTR_ERR(inode
);
4414 inode
->i_op
= &btrfs_dir_inode_operations
;
4415 inode
->i_fop
= &btrfs_dir_file_operations
;
4418 btrfs_i_size_write(inode
, 0);
4420 error
= btrfs_update_inode(trans
, new_root
, inode
);
4424 d_instantiate(dentry
, inode
);
4428 /* helper function for file defrag and space balancing. This
4429 * forces readahead on a given range of bytes in an inode
4431 unsigned long btrfs_force_ra(struct address_space
*mapping
,
4432 struct file_ra_state
*ra
, struct file
*file
,
4433 pgoff_t offset
, pgoff_t last_index
)
4435 pgoff_t req_size
= last_index
- offset
+ 1;
4437 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
4438 return offset
+ req_size
;
4441 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
4443 struct btrfs_inode
*ei
;
4445 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
4449 ei
->logged_trans
= 0;
4450 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
4451 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
4452 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
4453 INIT_LIST_HEAD(&ei
->i_orphan
);
4454 return &ei
->vfs_inode
;
4457 void btrfs_destroy_inode(struct inode
*inode
)
4459 struct btrfs_ordered_extent
*ordered
;
4460 WARN_ON(!list_empty(&inode
->i_dentry
));
4461 WARN_ON(inode
->i_data
.nrpages
);
4463 if (BTRFS_I(inode
)->i_acl
&&
4464 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
4465 posix_acl_release(BTRFS_I(inode
)->i_acl
);
4466 if (BTRFS_I(inode
)->i_default_acl
&&
4467 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
4468 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
4470 spin_lock(&BTRFS_I(inode
)->root
->list_lock
);
4471 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
4472 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
4473 " list\n", inode
->i_ino
);
4476 spin_unlock(&BTRFS_I(inode
)->root
->list_lock
);
4479 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
4483 printk("found ordered extent %Lu %Lu\n",
4484 ordered
->file_offset
, ordered
->len
);
4485 btrfs_remove_ordered_extent(inode
, ordered
);
4486 btrfs_put_ordered_extent(ordered
);
4487 btrfs_put_ordered_extent(ordered
);
4490 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4491 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4494 static void init_once(void *foo
)
4496 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4498 inode_init_once(&ei
->vfs_inode
);
4501 void btrfs_destroy_cachep(void)
4503 if (btrfs_inode_cachep
)
4504 kmem_cache_destroy(btrfs_inode_cachep
);
4505 if (btrfs_trans_handle_cachep
)
4506 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4507 if (btrfs_transaction_cachep
)
4508 kmem_cache_destroy(btrfs_transaction_cachep
);
4509 if (btrfs_bit_radix_cachep
)
4510 kmem_cache_destroy(btrfs_bit_radix_cachep
);
4511 if (btrfs_path_cachep
)
4512 kmem_cache_destroy(btrfs_path_cachep
);
4515 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
4516 unsigned long extra_flags
,
4517 void (*ctor
)(void *))
4519 return kmem_cache_create(name
, size
, 0, (SLAB_RECLAIM_ACCOUNT
|
4520 SLAB_MEM_SPREAD
| extra_flags
), ctor
);
4523 int btrfs_init_cachep(void)
4525 btrfs_inode_cachep
= btrfs_cache_create("btrfs_inode_cache",
4526 sizeof(struct btrfs_inode
),
4528 if (!btrfs_inode_cachep
)
4530 btrfs_trans_handle_cachep
=
4531 btrfs_cache_create("btrfs_trans_handle_cache",
4532 sizeof(struct btrfs_trans_handle
),
4534 if (!btrfs_trans_handle_cachep
)
4536 btrfs_transaction_cachep
= btrfs_cache_create("btrfs_transaction_cache",
4537 sizeof(struct btrfs_transaction
),
4539 if (!btrfs_transaction_cachep
)
4541 btrfs_path_cachep
= btrfs_cache_create("btrfs_path_cache",
4542 sizeof(struct btrfs_path
),
4544 if (!btrfs_path_cachep
)
4546 btrfs_bit_radix_cachep
= btrfs_cache_create("btrfs_radix", 256,
4547 SLAB_DESTROY_BY_RCU
, NULL
);
4548 if (!btrfs_bit_radix_cachep
)
4552 btrfs_destroy_cachep();
4556 static int btrfs_getattr(struct vfsmount
*mnt
,
4557 struct dentry
*dentry
, struct kstat
*stat
)
4559 struct inode
*inode
= dentry
->d_inode
;
4560 generic_fillattr(inode
, stat
);
4561 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
4562 stat
->blksize
= PAGE_CACHE_SIZE
;
4563 stat
->blocks
= (inode_get_bytes(inode
) +
4564 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4568 static int btrfs_rename(struct inode
* old_dir
, struct dentry
*old_dentry
,
4569 struct inode
* new_dir
,struct dentry
*new_dentry
)
4571 struct btrfs_trans_handle
*trans
;
4572 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4573 struct inode
*new_inode
= new_dentry
->d_inode
;
4574 struct inode
*old_inode
= old_dentry
->d_inode
;
4575 struct timespec ctime
= CURRENT_TIME
;
4579 /* we're not allowed to rename between subvolumes */
4580 if (BTRFS_I(old_inode
)->root
->root_key
.objectid
!=
4581 BTRFS_I(new_dir
)->root
->root_key
.objectid
)
4584 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4585 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4589 /* to rename a snapshot or subvolume, we need to juggle the
4590 * backrefs. This isn't coded yet
4592 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
4595 ret
= btrfs_check_free_space(root
, 1, 0);
4599 trans
= btrfs_start_transaction(root
, 1);
4601 btrfs_set_trans_block_group(trans
, new_dir
);
4603 btrfs_inc_nlink(old_dentry
->d_inode
);
4604 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4605 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4606 old_inode
->i_ctime
= ctime
;
4608 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4609 old_dentry
->d_name
.name
,
4610 old_dentry
->d_name
.len
);
4615 new_inode
->i_ctime
= CURRENT_TIME
;
4616 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4617 new_dentry
->d_inode
,
4618 new_dentry
->d_name
.name
,
4619 new_dentry
->d_name
.len
);
4622 if (new_inode
->i_nlink
== 0) {
4623 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4629 ret
= btrfs_set_inode_index(new_dir
, &index
);
4633 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4634 old_inode
, new_dentry
->d_name
.name
,
4635 new_dentry
->d_name
.len
, 1, index
);
4640 btrfs_end_transaction_throttle(trans
, root
);
4646 * some fairly slow code that needs optimization. This walks the list
4647 * of all the inodes with pending delalloc and forces them to disk.
4649 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4651 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4652 struct btrfs_inode
*binode
;
4653 struct inode
*inode
;
4654 unsigned long flags
;
4656 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
4659 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
4660 while(!list_empty(head
)) {
4661 binode
= list_entry(head
->next
, struct btrfs_inode
,
4663 inode
= igrab(&binode
->vfs_inode
);
4665 list_del_init(&binode
->delalloc_inodes
);
4666 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
4668 filemap_flush(inode
->i_mapping
);
4672 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
4674 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
4676 /* the filemap_flush will queue IO into the worker threads, but
4677 * we have to make sure the IO is actually started and that
4678 * ordered extents get created before we return
4680 atomic_inc(&root
->fs_info
->async_submit_draining
);
4681 while(atomic_read(&root
->fs_info
->nr_async_submits
) ||
4682 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
4683 wait_event(root
->fs_info
->async_submit_wait
,
4684 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
4685 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
4687 atomic_dec(&root
->fs_info
->async_submit_draining
);
4691 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4692 const char *symname
)
4694 struct btrfs_trans_handle
*trans
;
4695 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4696 struct btrfs_path
*path
;
4697 struct btrfs_key key
;
4698 struct inode
*inode
= NULL
;
4706 struct btrfs_file_extent_item
*ei
;
4707 struct extent_buffer
*leaf
;
4708 unsigned long nr
= 0;
4710 name_len
= strlen(symname
) + 1;
4711 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4712 return -ENAMETOOLONG
;
4714 err
= btrfs_check_free_space(root
, 1, 0);
4718 trans
= btrfs_start_transaction(root
, 1);
4719 btrfs_set_trans_block_group(trans
, dir
);
4721 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4727 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4729 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4730 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4732 err
= PTR_ERR(inode
);
4736 err
= btrfs_init_acl(inode
, dir
);
4742 btrfs_set_trans_block_group(trans
, inode
);
4743 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4747 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4748 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4749 inode
->i_fop
= &btrfs_file_operations
;
4750 inode
->i_op
= &btrfs_file_inode_operations
;
4751 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4753 dir
->i_sb
->s_dirt
= 1;
4754 btrfs_update_inode_block_group(trans
, inode
);
4755 btrfs_update_inode_block_group(trans
, dir
);
4759 path
= btrfs_alloc_path();
4761 key
.objectid
= inode
->i_ino
;
4763 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
4764 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
4765 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
4771 leaf
= path
->nodes
[0];
4772 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
4773 struct btrfs_file_extent_item
);
4774 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
4775 btrfs_set_file_extent_type(leaf
, ei
,
4776 BTRFS_FILE_EXTENT_INLINE
);
4777 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
4778 btrfs_set_file_extent_compression(leaf
, ei
, 0);
4779 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
4780 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
4782 ptr
= btrfs_file_extent_inline_start(ei
);
4783 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
4784 btrfs_mark_buffer_dirty(leaf
);
4785 btrfs_free_path(path
);
4787 inode
->i_op
= &btrfs_symlink_inode_operations
;
4788 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
4789 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4790 inode_set_bytes(inode
, name_len
);
4791 btrfs_i_size_write(inode
, name_len
- 1);
4792 err
= btrfs_update_inode(trans
, root
, inode
);
4797 nr
= trans
->blocks_used
;
4798 btrfs_end_transaction_throttle(trans
, root
);
4801 inode_dec_link_count(inode
);
4804 btrfs_btree_balance_dirty(root
, nr
);
4808 static int prealloc_file_range(struct inode
*inode
, u64 start
, u64 end
,
4809 u64 alloc_hint
, int mode
)
4811 struct btrfs_trans_handle
*trans
;
4812 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4813 struct btrfs_key ins
;
4815 u64 cur_offset
= start
;
4816 u64 num_bytes
= end
- start
;
4819 trans
= btrfs_join_transaction(root
, 1);
4821 btrfs_set_trans_block_group(trans
, inode
);
4823 while (num_bytes
> 0) {
4824 alloc_size
= min(num_bytes
, root
->fs_info
->max_extent
);
4825 ret
= btrfs_reserve_extent(trans
, root
, alloc_size
,
4826 root
->sectorsize
, 0, alloc_hint
,
4832 ret
= insert_reserved_file_extent(trans
, inode
,
4833 cur_offset
, ins
.objectid
,
4834 ins
.offset
, ins
.offset
,
4835 ins
.offset
, 0, 0, 0,
4836 BTRFS_FILE_EXTENT_PREALLOC
);
4838 num_bytes
-= ins
.offset
;
4839 cur_offset
+= ins
.offset
;
4840 alloc_hint
= ins
.objectid
+ ins
.offset
;
4843 if (cur_offset
> start
) {
4844 inode
->i_ctime
= CURRENT_TIME
;
4845 btrfs_set_flag(inode
, PREALLOC
);
4846 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
4847 cur_offset
> i_size_read(inode
))
4848 btrfs_i_size_write(inode
, cur_offset
);
4849 ret
= btrfs_update_inode(trans
, root
, inode
);
4853 btrfs_end_transaction(trans
, root
);
4857 static long btrfs_fallocate(struct inode
*inode
, int mode
,
4858 loff_t offset
, loff_t len
)
4865 u64 mask
= BTRFS_I(inode
)->root
->sectorsize
- 1;
4866 struct extent_map
*em
;
4869 alloc_start
= offset
& ~mask
;
4870 alloc_end
= (offset
+ len
+ mask
) & ~mask
;
4872 mutex_lock(&inode
->i_mutex
);
4873 if (alloc_start
> inode
->i_size
) {
4874 ret
= btrfs_cont_expand(inode
, alloc_start
);
4880 struct btrfs_ordered_extent
*ordered
;
4881 lock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
,
4882 alloc_end
- 1, GFP_NOFS
);
4883 ordered
= btrfs_lookup_first_ordered_extent(inode
,
4886 ordered
->file_offset
+ ordered
->len
> alloc_start
&&
4887 ordered
->file_offset
< alloc_end
) {
4888 btrfs_put_ordered_extent(ordered
);
4889 unlock_extent(&BTRFS_I(inode
)->io_tree
,
4890 alloc_start
, alloc_end
- 1, GFP_NOFS
);
4891 btrfs_wait_ordered_range(inode
, alloc_start
,
4892 alloc_end
- alloc_start
);
4895 btrfs_put_ordered_extent(ordered
);
4900 cur_offset
= alloc_start
;
4902 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4903 alloc_end
- cur_offset
, 0);
4904 BUG_ON(IS_ERR(em
) || !em
);
4905 last_byte
= min(extent_map_end(em
), alloc_end
);
4906 last_byte
= (last_byte
+ mask
) & ~mask
;
4907 if (em
->block_start
== EXTENT_MAP_HOLE
) {
4908 ret
= prealloc_file_range(inode
, cur_offset
,
4909 last_byte
, alloc_hint
, mode
);
4911 free_extent_map(em
);
4915 if (em
->block_start
<= EXTENT_MAP_LAST_BYTE
)
4916 alloc_hint
= em
->block_start
;
4917 free_extent_map(em
);
4919 cur_offset
= last_byte
;
4920 if (cur_offset
>= alloc_end
) {
4925 unlock_extent(&BTRFS_I(inode
)->io_tree
, alloc_start
, alloc_end
- 1,
4928 mutex_unlock(&inode
->i_mutex
);
4932 static int btrfs_set_page_dirty(struct page
*page
)
4934 return __set_page_dirty_nobuffers(page
);
4937 static int btrfs_permission(struct inode
*inode
, int mask
)
4939 if (btrfs_test_flag(inode
, READONLY
) && (mask
& MAY_WRITE
))
4941 return generic_permission(inode
, mask
, btrfs_check_acl
);
4944 static struct inode_operations btrfs_dir_inode_operations
= {
4945 .getattr
= btrfs_getattr
,
4946 .lookup
= btrfs_lookup
,
4947 .create
= btrfs_create
,
4948 .unlink
= btrfs_unlink
,
4950 .mkdir
= btrfs_mkdir
,
4951 .rmdir
= btrfs_rmdir
,
4952 .rename
= btrfs_rename
,
4953 .symlink
= btrfs_symlink
,
4954 .setattr
= btrfs_setattr
,
4955 .mknod
= btrfs_mknod
,
4956 .setxattr
= btrfs_setxattr
,
4957 .getxattr
= btrfs_getxattr
,
4958 .listxattr
= btrfs_listxattr
,
4959 .removexattr
= btrfs_removexattr
,
4960 .permission
= btrfs_permission
,
4962 static struct inode_operations btrfs_dir_ro_inode_operations
= {
4963 .lookup
= btrfs_lookup
,
4964 .permission
= btrfs_permission
,
4966 static struct file_operations btrfs_dir_file_operations
= {
4967 .llseek
= generic_file_llseek
,
4968 .read
= generic_read_dir
,
4969 .readdir
= btrfs_real_readdir
,
4970 .unlocked_ioctl
= btrfs_ioctl
,
4971 #ifdef CONFIG_COMPAT
4972 .compat_ioctl
= btrfs_ioctl
,
4974 .release
= btrfs_release_file
,
4975 .fsync
= btrfs_sync_file
,
4978 static struct extent_io_ops btrfs_extent_io_ops
= {
4979 .fill_delalloc
= run_delalloc_range
,
4980 .submit_bio_hook
= btrfs_submit_bio_hook
,
4981 .merge_bio_hook
= btrfs_merge_bio_hook
,
4982 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
4983 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
4984 .writepage_start_hook
= btrfs_writepage_start_hook
,
4985 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
4986 .set_bit_hook
= btrfs_set_bit_hook
,
4987 .clear_bit_hook
= btrfs_clear_bit_hook
,
4990 static struct address_space_operations btrfs_aops
= {
4991 .readpage
= btrfs_readpage
,
4992 .writepage
= btrfs_writepage
,
4993 .writepages
= btrfs_writepages
,
4994 .readpages
= btrfs_readpages
,
4995 .sync_page
= block_sync_page
,
4997 .direct_IO
= btrfs_direct_IO
,
4998 .invalidatepage
= btrfs_invalidatepage
,
4999 .releasepage
= btrfs_releasepage
,
5000 .set_page_dirty
= btrfs_set_page_dirty
,
5003 static struct address_space_operations btrfs_symlink_aops
= {
5004 .readpage
= btrfs_readpage
,
5005 .writepage
= btrfs_writepage
,
5006 .invalidatepage
= btrfs_invalidatepage
,
5007 .releasepage
= btrfs_releasepage
,
5010 static struct inode_operations btrfs_file_inode_operations
= {
5011 .truncate
= btrfs_truncate
,
5012 .getattr
= btrfs_getattr
,
5013 .setattr
= btrfs_setattr
,
5014 .setxattr
= btrfs_setxattr
,
5015 .getxattr
= btrfs_getxattr
,
5016 .listxattr
= btrfs_listxattr
,
5017 .removexattr
= btrfs_removexattr
,
5018 .permission
= btrfs_permission
,
5019 .fallocate
= btrfs_fallocate
,
5021 static struct inode_operations btrfs_special_inode_operations
= {
5022 .getattr
= btrfs_getattr
,
5023 .setattr
= btrfs_setattr
,
5024 .permission
= btrfs_permission
,
5025 .setxattr
= btrfs_setxattr
,
5026 .getxattr
= btrfs_getxattr
,
5027 .listxattr
= btrfs_listxattr
,
5028 .removexattr
= btrfs_removexattr
,
5030 static struct inode_operations btrfs_symlink_inode_operations
= {
5031 .readlink
= generic_readlink
,
5032 .follow_link
= page_follow_link_light
,
5033 .put_link
= page_put_link
,
5034 .permission
= btrfs_permission
,