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>
42 #include "transaction.h"
43 #include "btrfs_inode.h"
45 #include "print-tree.h"
47 #include "ordered-data.h"
51 #include "ref-cache.h"
52 #include "compression.h"
54 struct btrfs_iget_args
{
56 struct btrfs_root
*root
;
59 static struct inode_operations btrfs_dir_inode_operations
;
60 static struct inode_operations btrfs_symlink_inode_operations
;
61 static struct inode_operations btrfs_dir_ro_inode_operations
;
62 static struct inode_operations btrfs_special_inode_operations
;
63 static struct inode_operations btrfs_file_inode_operations
;
64 static struct address_space_operations btrfs_aops
;
65 static struct address_space_operations btrfs_symlink_aops
;
66 static struct file_operations btrfs_dir_file_operations
;
67 static struct extent_io_ops btrfs_extent_io_ops
;
69 static struct kmem_cache
*btrfs_inode_cachep
;
70 struct kmem_cache
*btrfs_trans_handle_cachep
;
71 struct kmem_cache
*btrfs_transaction_cachep
;
72 struct kmem_cache
*btrfs_bit_radix_cachep
;
73 struct kmem_cache
*btrfs_path_cachep
;
76 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
77 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
78 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
79 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
80 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
81 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
82 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
83 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
86 static void btrfs_truncate(struct inode
*inode
);
87 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
90 * a very lame attempt at stopping writes when the FS is 85% full. There
91 * are countless ways this is incorrect, but it is better than nothing.
93 int btrfs_check_free_space(struct btrfs_root
*root
, u64 num_required
,
102 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
103 total
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
104 used
= btrfs_super_bytes_used(&root
->fs_info
->super_copy
);
112 if (used
+ root
->fs_info
->delalloc_bytes
+ num_required
> thresh
)
114 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
119 * this does all the hard work for inserting an inline extent into
120 * the btree. The caller should have done a btrfs_drop_extents so that
121 * no overlapping inline items exist in the btree
123 static int noinline
insert_inline_extent(struct btrfs_trans_handle
*trans
,
124 struct btrfs_root
*root
, struct inode
*inode
,
125 u64 start
, size_t size
, size_t compressed_size
,
126 struct page
**compressed_pages
)
128 struct btrfs_key key
;
129 struct btrfs_path
*path
;
130 struct extent_buffer
*leaf
;
131 struct page
*page
= NULL
;
134 struct btrfs_file_extent_item
*ei
;
137 size_t cur_size
= size
;
139 unsigned long offset
;
140 int use_compress
= 0;
142 if (compressed_size
&& compressed_pages
) {
144 cur_size
= compressed_size
;
147 path
= btrfs_alloc_path(); if (!path
)
150 btrfs_set_trans_block_group(trans
, inode
);
152 key
.objectid
= inode
->i_ino
;
154 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
155 inode_add_bytes(inode
, size
);
156 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
158 inode_add_bytes(inode
, size
);
159 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
164 printk("got bad ret %d\n", ret
);
167 leaf
= path
->nodes
[0];
168 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
169 struct btrfs_file_extent_item
);
170 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
171 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
172 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
173 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
174 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
175 ptr
= btrfs_file_extent_inline_start(ei
);
180 while(compressed_size
> 0) {
181 cpage
= compressed_pages
[i
];
182 cur_size
= min(compressed_size
,
186 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
191 compressed_size
-= cur_size
;
193 btrfs_set_file_extent_compression(leaf
, ei
,
194 BTRFS_COMPRESS_ZLIB
);
196 page
= find_get_page(inode
->i_mapping
,
197 start
>> PAGE_CACHE_SHIFT
);
198 btrfs_set_file_extent_compression(leaf
, ei
, 0);
199 kaddr
= kmap_atomic(page
, KM_USER0
);
200 offset
= start
& (PAGE_CACHE_SIZE
- 1);
201 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
202 kunmap_atomic(kaddr
, KM_USER0
);
203 page_cache_release(page
);
205 btrfs_mark_buffer_dirty(leaf
);
206 btrfs_free_path(path
);
208 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
209 btrfs_update_inode(trans
, root
, inode
);
212 btrfs_free_path(path
);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
223 struct btrfs_root
*root
,
224 struct inode
*inode
, u64 start
, u64 end
,
225 size_t compressed_size
,
226 struct page
**compressed_pages
)
228 u64 isize
= i_size_read(inode
);
229 u64 actual_end
= min(end
+ 1, isize
);
230 u64 inline_len
= actual_end
- start
;
231 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
232 ~((u64
)root
->sectorsize
- 1);
234 u64 data_len
= inline_len
;
238 data_len
= compressed_size
;
241 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
243 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
245 data_len
> root
->fs_info
->max_inline
) {
249 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
250 aligned_end
, aligned_end
, &hint_byte
);
253 if (isize
> actual_end
)
254 inline_len
= min_t(u64
, isize
, actual_end
);
255 ret
= insert_inline_extent(trans
, root
, inode
, start
,
256 inline_len
, compressed_size
,
259 btrfs_drop_extent_cache(inode
, start
, aligned_end
, 0);
264 * when extent_io.c finds a delayed allocation range in the file,
265 * the call backs end up in this code. The basic idea is to
266 * allocate extents on disk for the range, and create ordered data structs
267 * in ram to track those extents.
269 * locked_page is the page that writepage had locked already. We use
270 * it to make sure we don't do extra locks or unlocks.
272 * *page_started is set to one if we unlock locked_page and do everything
273 * required to start IO on it. It may be clean and already done with
276 static int cow_file_range(struct inode
*inode
, struct page
*locked_page
,
277 u64 start
, u64 end
, int *page_started
)
279 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
280 struct btrfs_trans_handle
*trans
;
283 unsigned long ram_size
;
287 u64 blocksize
= root
->sectorsize
;
289 struct btrfs_key ins
;
290 struct extent_map
*em
;
291 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
293 struct page
**pages
= NULL
;
294 unsigned long nr_pages
;
295 unsigned long nr_pages_ret
= 0;
296 unsigned long total_compressed
= 0;
297 unsigned long total_in
= 0;
298 unsigned long max_compressed
= 128 * 1024;
299 unsigned long max_uncompressed
= 256 * 1024;
304 trans
= btrfs_join_transaction(root
, 1);
306 btrfs_set_trans_block_group(trans
, inode
);
310 * compression made this loop a bit ugly, but the basic idea is to
311 * compress some pages but keep the total size of the compressed
312 * extent relatively small. If compression is off, this goto target
317 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
318 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
320 actual_end
= min_t(u64
, i_size_read(inode
), end
+ 1);
321 total_compressed
= actual_end
- start
;
323 /* we want to make sure that amount of ram required to uncompress
324 * an extent is reasonable, so we limit the total size in ram
325 * of a compressed extent to 256k
327 total_compressed
= min(total_compressed
, max_uncompressed
);
328 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
329 num_bytes
= max(blocksize
, num_bytes
);
330 disk_num_bytes
= num_bytes
;
334 /* we do compression for mount -o compress and when the
335 * inode has not been flagged as nocompress
337 if (!btrfs_test_flag(inode
, NOCOMPRESS
) &&
338 btrfs_test_opt(root
, COMPRESS
)) {
340 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
342 /* we want to make sure the amount of IO required to satisfy
343 * a random read is reasonably small, so we limit the size
344 * of a compressed extent to 128k
346 ret
= btrfs_zlib_compress_pages(inode
->i_mapping
, start
,
347 total_compressed
, pages
,
348 nr_pages
, &nr_pages_ret
,
354 unsigned long offset
= total_compressed
&
355 (PAGE_CACHE_SIZE
- 1);
356 struct page
*page
= pages
[nr_pages_ret
- 1];
359 /* zero the tail end of the last page, we might be
360 * sending it down to disk
363 kaddr
= kmap_atomic(page
, KM_USER0
);
364 memset(kaddr
+ offset
, 0,
365 PAGE_CACHE_SIZE
- offset
);
366 kunmap_atomic(kaddr
, KM_USER0
);
372 /* lets try to make an inline extent */
373 if (ret
|| total_in
< (end
- start
+ 1)) {
374 /* we didn't compress the entire range, try
375 * to make an uncompressed inline extent. This
376 * is almost sure to fail, but maybe inline sizes
377 * will get bigger later
379 ret
= cow_file_range_inline(trans
, root
, inode
,
380 start
, end
, 0, NULL
);
382 ret
= cow_file_range_inline(trans
, root
, inode
,
384 total_compressed
, pages
);
387 extent_clear_unlock_delalloc(inode
,
388 &BTRFS_I(inode
)->io_tree
,
399 * we aren't doing an inline extent round the compressed size
400 * up to a block size boundary so the allocator does sane
403 total_compressed
= (total_compressed
+ blocksize
- 1) &
407 * one last check to make sure the compression is really a
408 * win, compare the page count read with the blocks on disk
410 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
411 ~(PAGE_CACHE_SIZE
- 1);
412 if (total_compressed
>= total_in
) {
415 disk_num_bytes
= total_compressed
;
416 num_bytes
= total_in
;
419 if (!will_compress
&& pages
) {
421 * the compression code ran but failed to make things smaller,
422 * free any pages it allocated and our page pointer array
424 for (i
= 0; i
< nr_pages_ret
; i
++) {
425 page_cache_release(pages
[i
]);
429 total_compressed
= 0;
432 /* flag the file so we don't compress in the future */
433 btrfs_set_flag(inode
, NOCOMPRESS
);
436 BUG_ON(disk_num_bytes
>
437 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
439 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
441 while(disk_num_bytes
> 0) {
442 unsigned long min_bytes
;
445 * the max size of a compressed extent is pretty small,
446 * make the code a little less complex by forcing
447 * the allocator to find a whole compressed extent at once
450 min_bytes
= disk_num_bytes
;
452 min_bytes
= root
->sectorsize
;
454 cur_alloc_size
= min(disk_num_bytes
, root
->fs_info
->max_extent
);
455 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
456 min_bytes
, 0, alloc_hint
,
460 goto free_pages_out_fail
;
462 em
= alloc_extent_map(GFP_NOFS
);
466 ram_size
= num_bytes
;
469 /* ramsize == disk size */
470 ram_size
= ins
.offset
;
471 em
->len
= ins
.offset
;
474 em
->block_start
= ins
.objectid
;
475 em
->block_len
= ins
.offset
;
476 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
477 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
480 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
483 spin_lock(&em_tree
->lock
);
484 ret
= add_extent_mapping(em_tree
, em
);
485 spin_unlock(&em_tree
->lock
);
486 if (ret
!= -EEXIST
) {
490 btrfs_drop_extent_cache(inode
, start
,
491 start
+ ram_size
- 1, 0);
494 cur_alloc_size
= ins
.offset
;
495 ordered_type
= will_compress
? BTRFS_ORDERED_COMPRESSED
: 0;
496 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
497 ram_size
, cur_alloc_size
,
501 if (disk_num_bytes
< cur_alloc_size
) {
502 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes
,
509 * we're doing compression, we and we need to
510 * submit the compressed extents down to the device.
512 * We lock down all the file pages, clearing their
513 * dirty bits and setting them writeback. Everyone
514 * that wants to modify the page will wait on the
515 * ordered extent above.
517 * The writeback bits on the file pages are
518 * cleared when the compressed pages are on disk
520 btrfs_end_transaction(trans
, root
);
522 if (start
<= page_offset(locked_page
) &&
523 page_offset(locked_page
) < start
+ ram_size
) {
527 extent_clear_unlock_delalloc(inode
,
528 &BTRFS_I(inode
)->io_tree
,
530 start
+ ram_size
- 1,
533 ret
= btrfs_submit_compressed_write(inode
, start
,
534 ram_size
, ins
.objectid
,
535 cur_alloc_size
, pages
,
539 trans
= btrfs_join_transaction(root
, 1);
540 if (start
+ ram_size
< end
) {
542 alloc_hint
= ins
.objectid
+ ins
.offset
;
543 /* pages will be freed at end_bio time */
547 /* we've written everything, time to go */
551 /* we're not doing compressed IO, don't unlock the first
552 * page (which the caller expects to stay locked), don't
553 * clear any dirty bits and don't set any writeback bits
555 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
556 start
, start
+ ram_size
- 1,
557 locked_page
, 0, 0, 0);
558 disk_num_bytes
-= cur_alloc_size
;
559 num_bytes
-= cur_alloc_size
;
560 alloc_hint
= ins
.objectid
+ ins
.offset
;
561 start
+= cur_alloc_size
;
566 btrfs_end_transaction(trans
, root
);
571 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
572 start
, end
, locked_page
, 0, 0, 0);
574 for (i
= 0; i
< nr_pages_ret
; i
++)
575 page_cache_release(pages
[i
]);
583 * when nowcow writeback call back. This checks for snapshots or COW copies
584 * of the extents that exist in the file, and COWs the file as required.
586 * If no cow copies or snapshots exist, we write directly to the existing
589 static int run_delalloc_nocow(struct inode
*inode
, struct page
*locked_page
,
590 u64 start
, u64 end
, int *page_started
)
592 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
593 struct btrfs_trans_handle
*trans
;
594 struct extent_buffer
*leaf
;
595 struct btrfs_path
*path
;
596 struct btrfs_file_extent_item
*fi
;
597 struct btrfs_key found_key
;
608 path
= btrfs_alloc_path();
610 trans
= btrfs_join_transaction(root
, 1);
616 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
619 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
620 leaf
= path
->nodes
[0];
621 btrfs_item_key_to_cpu(leaf
, &found_key
,
623 if (found_key
.objectid
== inode
->i_ino
&&
624 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
629 leaf
= path
->nodes
[0];
630 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
631 ret
= btrfs_next_leaf(root
, path
);
636 leaf
= path
->nodes
[0];
641 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
643 if (found_key
.objectid
> inode
->i_ino
||
644 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
645 found_key
.offset
> end
)
648 if (found_key
.offset
> cur_offset
) {
649 extent_end
= found_key
.offset
;
653 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
654 struct btrfs_file_extent_item
);
655 extent_type
= btrfs_file_extent_type(leaf
, fi
);
657 if (extent_type
== BTRFS_FILE_EXTENT_REG
) {
658 struct btrfs_block_group_cache
*block_group
;
659 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
660 extent_end
= found_key
.offset
+
661 btrfs_file_extent_num_bytes(leaf
, fi
);
662 if (extent_end
<= start
) {
666 if (btrfs_file_extent_compression(leaf
, fi
) ||
667 btrfs_file_extent_encryption(leaf
, fi
) ||
668 btrfs_file_extent_other_encoding(leaf
, fi
))
670 if (disk_bytenr
== 0)
672 if (btrfs_cross_ref_exist(trans
, root
, disk_bytenr
))
674 block_group
= btrfs_lookup_block_group(root
->fs_info
,
676 if (!block_group
|| block_group
->ro
)
678 disk_bytenr
+= btrfs_file_extent_offset(leaf
, fi
);
680 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
681 extent_end
= found_key
.offset
+
682 btrfs_file_extent_inline_len(leaf
, fi
);
683 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
688 if (extent_end
<= start
) {
693 if (cow_start
== (u64
)-1)
694 cow_start
= cur_offset
;
695 cur_offset
= extent_end
;
696 if (cur_offset
> end
)
702 btrfs_release_path(root
, path
);
703 if (cow_start
!= (u64
)-1) {
704 ret
= cow_file_range(inode
, locked_page
, cow_start
,
705 found_key
.offset
- 1, page_started
);
710 disk_bytenr
+= cur_offset
- found_key
.offset
;
711 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
713 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
714 num_bytes
, num_bytes
,
715 BTRFS_ORDERED_NOCOW
);
716 cur_offset
= extent_end
;
717 if (cur_offset
> end
)
720 btrfs_release_path(root
, path
);
722 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
723 cow_start
= cur_offset
;
724 if (cow_start
!= (u64
)-1) {
725 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
730 ret
= btrfs_end_transaction(trans
, root
);
732 btrfs_free_path(path
);
737 * extent_io.c call back to do delayed allocation processing
739 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
740 u64 start
, u64 end
, int *page_started
)
742 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
745 if (btrfs_test_opt(root
, NODATACOW
) ||
746 btrfs_test_flag(inode
, NODATACOW
))
747 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
750 ret
= cow_file_range(inode
, locked_page
, start
, end
,
757 * extent_io.c set_bit_hook, used to track delayed allocation
758 * bytes in this file, and to maintain the list of inodes that
759 * have pending delalloc work to be done.
761 int btrfs_set_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
762 unsigned long old
, unsigned long bits
)
765 if (!(old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
766 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
767 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
768 BTRFS_I(inode
)->delalloc_bytes
+= end
- start
+ 1;
769 root
->fs_info
->delalloc_bytes
+= end
- start
+ 1;
770 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
771 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
772 &root
->fs_info
->delalloc_inodes
);
774 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
780 * extent_io.c clear_bit_hook, see set_bit_hook for why
782 int btrfs_clear_bit_hook(struct inode
*inode
, u64 start
, u64 end
,
783 unsigned long old
, unsigned long bits
)
785 if ((old
& EXTENT_DELALLOC
) && (bits
& EXTENT_DELALLOC
)) {
786 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
789 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
790 if (end
- start
+ 1 > root
->fs_info
->delalloc_bytes
) {
791 printk("warning: delalloc account %Lu %Lu\n",
792 end
- start
+ 1, root
->fs_info
->delalloc_bytes
);
793 root
->fs_info
->delalloc_bytes
= 0;
794 BTRFS_I(inode
)->delalloc_bytes
= 0;
796 root
->fs_info
->delalloc_bytes
-= end
- start
+ 1;
797 BTRFS_I(inode
)->delalloc_bytes
-= end
- start
+ 1;
799 if (BTRFS_I(inode
)->delalloc_bytes
== 0 &&
800 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
801 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
803 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
809 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
810 * we don't create bios that span stripes or chunks
812 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
813 size_t size
, struct bio
*bio
,
814 unsigned long bio_flags
)
816 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
817 struct btrfs_mapping_tree
*map_tree
;
818 u64 logical
= (u64
)bio
->bi_sector
<< 9;
823 length
= bio
->bi_size
;
824 map_tree
= &root
->fs_info
->mapping_tree
;
826 ret
= btrfs_map_block(map_tree
, READ
, logical
,
827 &map_length
, NULL
, 0);
829 if (map_length
< length
+ size
) {
836 * in order to insert checksums into the metadata in large chunks,
837 * we wait until bio submission time. All the pages in the bio are
838 * checksummed and sums are attached onto the ordered extent record.
840 * At IO completion time the cums attached on the ordered extent record
841 * are inserted into the btree
843 int __btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
844 int mirror_num
, unsigned long bio_flags
)
846 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
849 ret
= btrfs_csum_one_bio(root
, inode
, bio
);
852 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
856 * extent_io.c submission hook. This does the right thing for csum calculation on write,
857 * or reading the csums from the tree before a read
859 int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
860 int mirror_num
, unsigned long bio_flags
)
862 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
866 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
869 skip_sum
= btrfs_test_opt(root
, NODATASUM
) ||
870 btrfs_test_flag(inode
, NODATASUM
);
872 if (!(rw
& (1 << BIO_RW
))) {
874 btrfs_lookup_bio_sums(root
, inode
, bio
);
876 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
877 return btrfs_submit_compressed_read(inode
, bio
,
878 mirror_num
, bio_flags
);
880 } else if (!skip_sum
) {
881 /* we're doing a write, do the async checksumming */
882 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
883 inode
, rw
, bio
, mirror_num
,
884 bio_flags
, __btrfs_submit_bio_hook
);
888 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
892 * given a list of ordered sums record them in the inode. This happens
893 * at IO completion time based on sums calculated at bio submission time.
895 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
896 struct inode
*inode
, u64 file_offset
,
897 struct list_head
*list
)
899 struct list_head
*cur
;
900 struct btrfs_ordered_sum
*sum
;
902 btrfs_set_trans_block_group(trans
, inode
);
903 list_for_each(cur
, list
) {
904 sum
= list_entry(cur
, struct btrfs_ordered_sum
, list
);
905 btrfs_csum_file_blocks(trans
, BTRFS_I(inode
)->root
,
911 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
)
913 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
917 /* see btrfs_writepage_start_hook for details on why this is required */
918 struct btrfs_writepage_fixup
{
920 struct btrfs_work work
;
923 void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
925 struct btrfs_writepage_fixup
*fixup
;
926 struct btrfs_ordered_extent
*ordered
;
932 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
936 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
937 ClearPageChecked(page
);
941 inode
= page
->mapping
->host
;
942 page_start
= page_offset(page
);
943 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
945 lock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
947 /* already ordered? We're done */
948 if (test_range_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
949 EXTENT_ORDERED
, 0)) {
953 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
955 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
,
958 btrfs_start_ordered_extent(inode
, ordered
, 1);
962 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
963 ClearPageChecked(page
);
965 unlock_extent(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, GFP_NOFS
);
968 page_cache_release(page
);
972 * There are a few paths in the higher layers of the kernel that directly
973 * set the page dirty bit without asking the filesystem if it is a
974 * good idea. This causes problems because we want to make sure COW
975 * properly happens and the data=ordered rules are followed.
977 * In our case any range that doesn't have the ORDERED bit set
978 * hasn't been properly setup for IO. We kick off an async process
979 * to fix it up. The async helper will wait for ordered extents, set
980 * the delalloc bit and make it safe to write the page.
982 int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
984 struct inode
*inode
= page
->mapping
->host
;
985 struct btrfs_writepage_fixup
*fixup
;
986 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
989 ret
= test_range_bit(&BTRFS_I(inode
)->io_tree
, start
, end
,
994 if (PageChecked(page
))
997 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1001 SetPageChecked(page
);
1002 page_cache_get(page
);
1003 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1005 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1009 /* as ordered data IO finishes, this gets called so we can finish
1010 * an ordered extent if the range of bytes in the file it covers are
1013 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1015 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1016 struct btrfs_trans_handle
*trans
;
1017 struct btrfs_ordered_extent
*ordered_extent
;
1018 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1019 struct btrfs_file_extent_item
*extent_item
;
1020 struct btrfs_path
*path
= NULL
;
1021 struct extent_buffer
*leaf
;
1023 struct list_head list
;
1024 struct btrfs_key ins
;
1027 ret
= btrfs_dec_test_ordered_pending(inode
, start
, end
- start
+ 1);
1031 trans
= btrfs_join_transaction(root
, 1);
1033 ordered_extent
= btrfs_lookup_ordered_extent(inode
, start
);
1034 BUG_ON(!ordered_extent
);
1035 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
))
1038 path
= btrfs_alloc_path();
1041 lock_extent(io_tree
, ordered_extent
->file_offset
,
1042 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1045 INIT_LIST_HEAD(&list
);
1047 ret
= btrfs_drop_extents(trans
, root
, inode
,
1048 ordered_extent
->file_offset
,
1049 ordered_extent
->file_offset
+
1050 ordered_extent
->len
,
1051 ordered_extent
->file_offset
, &alloc_hint
);
1054 ins
.objectid
= inode
->i_ino
;
1055 ins
.offset
= ordered_extent
->file_offset
;
1056 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1057 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
1058 sizeof(*extent_item
));
1060 leaf
= path
->nodes
[0];
1061 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1062 struct btrfs_file_extent_item
);
1063 btrfs_set_file_extent_generation(leaf
, extent_item
, trans
->transid
);
1064 btrfs_set_file_extent_type(leaf
, extent_item
, BTRFS_FILE_EXTENT_REG
);
1065 btrfs_set_file_extent_disk_bytenr(leaf
, extent_item
,
1066 ordered_extent
->start
);
1067 btrfs_set_file_extent_disk_num_bytes(leaf
, extent_item
,
1068 ordered_extent
->disk_len
);
1069 btrfs_set_file_extent_offset(leaf
, extent_item
, 0);
1071 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1072 btrfs_set_file_extent_compression(leaf
, extent_item
, 1);
1074 btrfs_set_file_extent_compression(leaf
, extent_item
, 0);
1075 btrfs_set_file_extent_encryption(leaf
, extent_item
, 0);
1076 btrfs_set_file_extent_other_encoding(leaf
, extent_item
, 0);
1078 /* ram bytes = extent_num_bytes for now */
1079 btrfs_set_file_extent_num_bytes(leaf
, extent_item
,
1080 ordered_extent
->len
);
1081 btrfs_set_file_extent_ram_bytes(leaf
, extent_item
,
1082 ordered_extent
->len
);
1083 btrfs_mark_buffer_dirty(leaf
);
1085 btrfs_drop_extent_cache(inode
, ordered_extent
->file_offset
,
1086 ordered_extent
->file_offset
+
1087 ordered_extent
->len
- 1, 0);
1089 ins
.objectid
= ordered_extent
->start
;
1090 ins
.offset
= ordered_extent
->disk_len
;
1091 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1092 ret
= btrfs_alloc_reserved_extent(trans
, root
, leaf
->start
,
1093 root
->root_key
.objectid
,
1094 trans
->transid
, inode
->i_ino
, &ins
);
1096 btrfs_release_path(root
, path
);
1098 inode_add_bytes(inode
, ordered_extent
->len
);
1099 unlock_extent(io_tree
, ordered_extent
->file_offset
,
1100 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1103 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1104 &ordered_extent
->list
);
1106 mutex_lock(&BTRFS_I(inode
)->extent_mutex
);
1107 btrfs_ordered_update_i_size(inode
, ordered_extent
);
1108 btrfs_update_inode(trans
, root
, inode
);
1109 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1110 mutex_unlock(&BTRFS_I(inode
)->extent_mutex
);
1113 btrfs_put_ordered_extent(ordered_extent
);
1114 /* once for the tree */
1115 btrfs_put_ordered_extent(ordered_extent
);
1117 btrfs_end_transaction(trans
, root
);
1119 btrfs_free_path(path
);
1123 int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1124 struct extent_state
*state
, int uptodate
)
1126 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1130 * When IO fails, either with EIO or csum verification fails, we
1131 * try other mirrors that might have a good copy of the data. This
1132 * io_failure_record is used to record state as we go through all the
1133 * mirrors. If another mirror has good data, the page is set up to date
1134 * and things continue. If a good mirror can't be found, the original
1135 * bio end_io callback is called to indicate things have failed.
1137 struct io_failure_record
{
1145 int btrfs_io_failed_hook(struct bio
*failed_bio
,
1146 struct page
*page
, u64 start
, u64 end
,
1147 struct extent_state
*state
)
1149 struct io_failure_record
*failrec
= NULL
;
1151 struct extent_map
*em
;
1152 struct inode
*inode
= page
->mapping
->host
;
1153 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1154 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1160 unsigned long bio_flags
= 0;
1162 ret
= get_state_private(failure_tree
, start
, &private);
1164 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1167 failrec
->start
= start
;
1168 failrec
->len
= end
- start
+ 1;
1169 failrec
->last_mirror
= 0;
1171 spin_lock(&em_tree
->lock
);
1172 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1173 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1174 free_extent_map(em
);
1177 spin_unlock(&em_tree
->lock
);
1179 if (!em
|| IS_ERR(em
)) {
1183 logical
= start
- em
->start
;
1184 logical
= em
->block_start
+ logical
;
1185 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
))
1186 bio_flags
= EXTENT_BIO_COMPRESSED
;
1187 failrec
->logical
= logical
;
1188 free_extent_map(em
);
1189 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1190 EXTENT_DIRTY
, GFP_NOFS
);
1191 set_state_private(failure_tree
, start
,
1192 (u64
)(unsigned long)failrec
);
1194 failrec
= (struct io_failure_record
*)(unsigned long)private;
1196 num_copies
= btrfs_num_copies(
1197 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1198 failrec
->logical
, failrec
->len
);
1199 failrec
->last_mirror
++;
1201 spin_lock_irq(&BTRFS_I(inode
)->io_tree
.lock
);
1202 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1205 if (state
&& state
->start
!= failrec
->start
)
1207 spin_unlock_irq(&BTRFS_I(inode
)->io_tree
.lock
);
1209 if (!state
|| failrec
->last_mirror
> num_copies
) {
1210 set_state_private(failure_tree
, failrec
->start
, 0);
1211 clear_extent_bits(failure_tree
, failrec
->start
,
1212 failrec
->start
+ failrec
->len
- 1,
1213 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1217 bio
= bio_alloc(GFP_NOFS
, 1);
1218 bio
->bi_private
= state
;
1219 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1220 bio
->bi_sector
= failrec
->logical
>> 9;
1221 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1223 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1224 if (failed_bio
->bi_rw
& (1 << BIO_RW
))
1229 BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1230 failrec
->last_mirror
,
1236 * each time an IO finishes, we do a fast check in the IO failure tree
1237 * to see if we need to process or clean up an io_failure_record
1239 int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1242 u64 private_failure
;
1243 struct io_failure_record
*failure
;
1247 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1248 (u64
)-1, 1, EXTENT_DIRTY
)) {
1249 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1250 start
, &private_failure
);
1252 failure
= (struct io_failure_record
*)(unsigned long)
1254 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1256 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1258 failure
->start
+ failure
->len
- 1,
1259 EXTENT_DIRTY
| EXTENT_LOCKED
,
1268 * when reads are done, we need to check csums to verify the data is correct
1269 * if there's a match, we allow the bio to finish. If not, we go through
1270 * the io_failure_record routines to find good copies
1272 int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1273 struct extent_state
*state
)
1275 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1276 struct inode
*inode
= page
->mapping
->host
;
1277 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1279 u64
private = ~(u32
)0;
1281 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1283 unsigned long flags
;
1285 if (btrfs_test_opt(root
, NODATASUM
) ||
1286 btrfs_test_flag(inode
, NODATASUM
))
1288 if (state
&& state
->start
== start
) {
1289 private = state
->private;
1292 ret
= get_state_private(io_tree
, start
, &private);
1294 local_irq_save(flags
);
1295 kaddr
= kmap_atomic(page
, KM_IRQ0
);
1299 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1300 btrfs_csum_final(csum
, (char *)&csum
);
1301 if (csum
!= private) {
1304 kunmap_atomic(kaddr
, KM_IRQ0
);
1305 local_irq_restore(flags
);
1307 /* if the io failure tree for this inode is non-empty,
1308 * check to see if we've recovered from a failed IO
1310 btrfs_clean_io_failures(inode
, start
);
1314 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1315 page
->mapping
->host
->i_ino
, (unsigned long long)start
, csum
,
1317 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1318 flush_dcache_page(page
);
1319 kunmap_atomic(kaddr
, KM_IRQ0
);
1320 local_irq_restore(flags
);
1327 * This creates an orphan entry for the given inode in case something goes
1328 * wrong in the middle of an unlink/truncate.
1330 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1332 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1335 spin_lock(&root
->list_lock
);
1337 /* already on the orphan list, we're good */
1338 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1339 spin_unlock(&root
->list_lock
);
1343 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1345 spin_unlock(&root
->list_lock
);
1348 * insert an orphan item to track this unlinked/truncated file
1350 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
1356 * We have done the truncate/delete so we can go ahead and remove the orphan
1357 * item for this particular inode.
1359 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
1361 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1364 spin_lock(&root
->list_lock
);
1366 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
1367 spin_unlock(&root
->list_lock
);
1371 list_del_init(&BTRFS_I(inode
)->i_orphan
);
1373 spin_unlock(&root
->list_lock
);
1377 spin_unlock(&root
->list_lock
);
1379 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
1385 * this cleans up any orphans that may be left on the list from the last use
1388 void btrfs_orphan_cleanup(struct btrfs_root
*root
)
1390 struct btrfs_path
*path
;
1391 struct extent_buffer
*leaf
;
1392 struct btrfs_item
*item
;
1393 struct btrfs_key key
, found_key
;
1394 struct btrfs_trans_handle
*trans
;
1395 struct inode
*inode
;
1396 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
1398 /* don't do orphan cleanup if the fs is readonly. */
1399 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
1402 path
= btrfs_alloc_path();
1407 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
1408 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
1409 key
.offset
= (u64
)-1;
1413 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1415 printk(KERN_ERR
"Error searching slot for orphan: %d"
1421 * if ret == 0 means we found what we were searching for, which
1422 * is weird, but possible, so only screw with path if we didnt
1423 * find the key and see if we have stuff that matches
1426 if (path
->slots
[0] == 0)
1431 /* pull out the item */
1432 leaf
= path
->nodes
[0];
1433 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
1434 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1436 /* make sure the item matches what we want */
1437 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
1439 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
1442 /* release the path since we're done with it */
1443 btrfs_release_path(root
, path
);
1446 * this is where we are basically btrfs_lookup, without the
1447 * crossing root thing. we store the inode number in the
1448 * offset of the orphan item.
1450 inode
= btrfs_iget_locked(root
->fs_info
->sb
,
1451 found_key
.offset
, root
);
1455 if (inode
->i_state
& I_NEW
) {
1456 BTRFS_I(inode
)->root
= root
;
1458 /* have to set the location manually */
1459 BTRFS_I(inode
)->location
.objectid
= inode
->i_ino
;
1460 BTRFS_I(inode
)->location
.type
= BTRFS_INODE_ITEM_KEY
;
1461 BTRFS_I(inode
)->location
.offset
= 0;
1463 btrfs_read_locked_inode(inode
);
1464 unlock_new_inode(inode
);
1468 * add this inode to the orphan list so btrfs_orphan_del does
1469 * the proper thing when we hit it
1471 spin_lock(&root
->list_lock
);
1472 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
1473 spin_unlock(&root
->list_lock
);
1476 * if this is a bad inode, means we actually succeeded in
1477 * removing the inode, but not the orphan record, which means
1478 * we need to manually delete the orphan since iput will just
1479 * do a destroy_inode
1481 if (is_bad_inode(inode
)) {
1482 trans
= btrfs_start_transaction(root
, 1);
1483 btrfs_orphan_del(trans
, inode
);
1484 btrfs_end_transaction(trans
, root
);
1489 /* if we have links, this was a truncate, lets do that */
1490 if (inode
->i_nlink
) {
1492 btrfs_truncate(inode
);
1497 /* this will do delete_inode and everything for us */
1502 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
1504 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
1506 btrfs_free_path(path
);
1510 * read an inode from the btree into the in-memory inode
1512 void btrfs_read_locked_inode(struct inode
*inode
)
1514 struct btrfs_path
*path
;
1515 struct extent_buffer
*leaf
;
1516 struct btrfs_inode_item
*inode_item
;
1517 struct btrfs_timespec
*tspec
;
1518 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1519 struct btrfs_key location
;
1520 u64 alloc_group_block
;
1524 path
= btrfs_alloc_path();
1526 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
1528 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
1532 leaf
= path
->nodes
[0];
1533 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1534 struct btrfs_inode_item
);
1536 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
1537 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
1538 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
1539 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
1540 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
1542 tspec
= btrfs_inode_atime(inode_item
);
1543 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1544 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1546 tspec
= btrfs_inode_mtime(inode_item
);
1547 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1548 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1550 tspec
= btrfs_inode_ctime(inode_item
);
1551 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
1552 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
1554 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
1555 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
1556 inode
->i_generation
= BTRFS_I(inode
)->generation
;
1558 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
1560 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
1562 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
1563 BTRFS_I(inode
)->block_group
= btrfs_lookup_block_group(root
->fs_info
,
1565 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
1566 if (!BTRFS_I(inode
)->block_group
) {
1567 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
,
1569 BTRFS_BLOCK_GROUP_METADATA
, 0);
1571 btrfs_free_path(path
);
1574 switch (inode
->i_mode
& S_IFMT
) {
1576 inode
->i_mapping
->a_ops
= &btrfs_aops
;
1577 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
1578 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
1579 inode
->i_fop
= &btrfs_file_operations
;
1580 inode
->i_op
= &btrfs_file_inode_operations
;
1583 inode
->i_fop
= &btrfs_dir_file_operations
;
1584 if (root
== root
->fs_info
->tree_root
)
1585 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
1587 inode
->i_op
= &btrfs_dir_inode_operations
;
1590 inode
->i_op
= &btrfs_symlink_inode_operations
;
1591 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
1592 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
1595 init_special_inode(inode
, inode
->i_mode
, rdev
);
1601 btrfs_free_path(path
);
1602 make_bad_inode(inode
);
1606 * given a leaf and an inode, copy the inode fields into the leaf
1608 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
1609 struct extent_buffer
*leaf
,
1610 struct btrfs_inode_item
*item
,
1611 struct inode
*inode
)
1613 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
1614 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
1615 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
1616 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
1617 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
1619 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
1620 inode
->i_atime
.tv_sec
);
1621 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
1622 inode
->i_atime
.tv_nsec
);
1624 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
1625 inode
->i_mtime
.tv_sec
);
1626 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
1627 inode
->i_mtime
.tv_nsec
);
1629 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
1630 inode
->i_ctime
.tv_sec
);
1631 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
1632 inode
->i_ctime
.tv_nsec
);
1634 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
1635 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
1636 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
1637 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
1638 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
1639 btrfs_set_inode_block_group(leaf
, item
,
1640 BTRFS_I(inode
)->block_group
->key
.objectid
);
1644 * copy everything in the in-memory inode into the btree.
1646 int noinline
btrfs_update_inode(struct btrfs_trans_handle
*trans
,
1647 struct btrfs_root
*root
,
1648 struct inode
*inode
)
1650 struct btrfs_inode_item
*inode_item
;
1651 struct btrfs_path
*path
;
1652 struct extent_buffer
*leaf
;
1655 path
= btrfs_alloc_path();
1657 ret
= btrfs_lookup_inode(trans
, root
, path
,
1658 &BTRFS_I(inode
)->location
, 1);
1665 leaf
= path
->nodes
[0];
1666 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
1667 struct btrfs_inode_item
);
1669 fill_inode_item(trans
, leaf
, inode_item
, inode
);
1670 btrfs_mark_buffer_dirty(leaf
);
1671 btrfs_set_inode_last_trans(trans
, inode
);
1674 btrfs_free_path(path
);
1680 * unlink helper that gets used here in inode.c and in the tree logging
1681 * recovery code. It remove a link in a directory with a given name, and
1682 * also drops the back refs in the inode to the directory
1684 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
1685 struct btrfs_root
*root
,
1686 struct inode
*dir
, struct inode
*inode
,
1687 const char *name
, int name_len
)
1689 struct btrfs_path
*path
;
1691 struct extent_buffer
*leaf
;
1692 struct btrfs_dir_item
*di
;
1693 struct btrfs_key key
;
1696 path
= btrfs_alloc_path();
1702 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
1703 name
, name_len
, -1);
1712 leaf
= path
->nodes
[0];
1713 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
1714 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
1717 btrfs_release_path(root
, path
);
1719 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
1721 dir
->i_ino
, &index
);
1723 printk("failed to delete reference to %.*s, "
1724 "inode %lu parent %lu\n", name_len
, name
,
1725 inode
->i_ino
, dir
->i_ino
);
1729 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
1730 index
, name
, name_len
, -1);
1739 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
1740 btrfs_release_path(root
, path
);
1742 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
1744 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
1746 BTRFS_I(dir
)->log_dirty_trans
= trans
->transid
;
1748 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
1752 btrfs_free_path(path
);
1756 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
1757 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
1758 btrfs_update_inode(trans
, root
, dir
);
1759 btrfs_drop_nlink(inode
);
1760 ret
= btrfs_update_inode(trans
, root
, inode
);
1761 dir
->i_sb
->s_dirt
= 1;
1766 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
1768 struct btrfs_root
*root
;
1769 struct btrfs_trans_handle
*trans
;
1770 struct inode
*inode
= dentry
->d_inode
;
1772 unsigned long nr
= 0;
1774 root
= BTRFS_I(dir
)->root
;
1776 ret
= btrfs_check_free_space(root
, 1, 1);
1780 trans
= btrfs_start_transaction(root
, 1);
1782 btrfs_set_trans_block_group(trans
, dir
);
1783 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
1784 dentry
->d_name
.name
, dentry
->d_name
.len
);
1786 if (inode
->i_nlink
== 0)
1787 ret
= btrfs_orphan_add(trans
, inode
);
1789 nr
= trans
->blocks_used
;
1791 btrfs_end_transaction_throttle(trans
, root
);
1793 btrfs_btree_balance_dirty(root
, nr
);
1797 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
1799 struct inode
*inode
= dentry
->d_inode
;
1802 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
1803 struct btrfs_trans_handle
*trans
;
1804 unsigned long nr
= 0;
1806 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
1810 ret
= btrfs_check_free_space(root
, 1, 1);
1814 trans
= btrfs_start_transaction(root
, 1);
1815 btrfs_set_trans_block_group(trans
, dir
);
1817 err
= btrfs_orphan_add(trans
, inode
);
1821 /* now the directory is empty */
1822 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
1823 dentry
->d_name
.name
, dentry
->d_name
.len
);
1825 btrfs_i_size_write(inode
, 0);
1829 nr
= trans
->blocks_used
;
1830 ret
= btrfs_end_transaction_throttle(trans
, root
);
1832 btrfs_btree_balance_dirty(root
, nr
);
1840 * when truncating bytes in a file, it is possible to avoid reading
1841 * the leaves that contain only checksum items. This can be the
1842 * majority of the IO required to delete a large file, but it must
1843 * be done carefully.
1845 * The keys in the level just above the leaves are checked to make sure
1846 * the lowest key in a given leaf is a csum key, and starts at an offset
1847 * after the new size.
1849 * Then the key for the next leaf is checked to make sure it also has
1850 * a checksum item for the same file. If it does, we know our target leaf
1851 * contains only checksum items, and it can be safely freed without reading
1854 * This is just an optimization targeted at large files. It may do
1855 * nothing. It will return 0 unless things went badly.
1857 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
1858 struct btrfs_root
*root
,
1859 struct btrfs_path
*path
,
1860 struct inode
*inode
, u64 new_size
)
1862 struct btrfs_key key
;
1865 struct btrfs_key found_key
;
1866 struct btrfs_key other_key
;
1867 struct btrfs_leaf_ref
*ref
;
1871 path
->lowest_level
= 1;
1872 key
.objectid
= inode
->i_ino
;
1873 key
.type
= BTRFS_CSUM_ITEM_KEY
;
1874 key
.offset
= new_size
;
1876 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1880 if (path
->nodes
[1] == NULL
) {
1885 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
1886 nritems
= btrfs_header_nritems(path
->nodes
[1]);
1891 if (path
->slots
[1] >= nritems
)
1894 /* did we find a key greater than anything we want to delete? */
1895 if (found_key
.objectid
> inode
->i_ino
||
1896 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
1899 /* we check the next key in the node to make sure the leave contains
1900 * only checksum items. This comparison doesn't work if our
1901 * leaf is the last one in the node
1903 if (path
->slots
[1] + 1 >= nritems
) {
1905 /* search forward from the last key in the node, this
1906 * will bring us into the next node in the tree
1908 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
1910 /* unlikely, but we inc below, so check to be safe */
1911 if (found_key
.offset
== (u64
)-1)
1914 /* search_forward needs a path with locks held, do the
1915 * search again for the original key. It is possible
1916 * this will race with a balance and return a path that
1917 * we could modify, but this drop is just an optimization
1918 * and is allowed to miss some leaves.
1920 btrfs_release_path(root
, path
);
1923 /* setup a max key for search_forward */
1924 other_key
.offset
= (u64
)-1;
1925 other_key
.type
= key
.type
;
1926 other_key
.objectid
= key
.objectid
;
1928 path
->keep_locks
= 1;
1929 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
1931 path
->keep_locks
= 0;
1932 if (ret
|| found_key
.objectid
!= key
.objectid
||
1933 found_key
.type
!= key
.type
) {
1938 key
.offset
= found_key
.offset
;
1939 btrfs_release_path(root
, path
);
1944 /* we know there's one more slot after us in the tree,
1945 * read that key so we can verify it is also a checksum item
1947 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
1949 if (found_key
.objectid
< inode
->i_ino
)
1952 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
1956 * if the key for the next leaf isn't a csum key from this objectid,
1957 * we can't be sure there aren't good items inside this leaf.
1960 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
1963 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
1964 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
1966 * it is safe to delete this leaf, it contains only
1967 * csum items from this inode at an offset >= new_size
1969 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
1972 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
1973 ref
= btrfs_alloc_leaf_ref(root
, 0);
1975 ref
->root_gen
= root
->root_key
.offset
;
1976 ref
->bytenr
= leaf_start
;
1978 ref
->generation
= leaf_gen
;
1981 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
1983 btrfs_free_leaf_ref(root
, ref
);
1989 btrfs_release_path(root
, path
);
1991 if (other_key
.objectid
== inode
->i_ino
&&
1992 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
1993 key
.offset
= other_key
.offset
;
1999 /* fixup any changes we've made to the path */
2000 path
->lowest_level
= 0;
2001 path
->keep_locks
= 0;
2002 btrfs_release_path(root
, path
);
2007 * this can truncate away extent items, csum items and directory items.
2008 * It starts at a high offset and removes keys until it can't find
2009 * any higher than new_size
2011 * csum items that cross the new i_size are truncated to the new size
2014 * min_type is the minimum key type to truncate down to. If set to 0, this
2015 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2017 noinline
int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
2018 struct btrfs_root
*root
,
2019 struct inode
*inode
,
2020 u64 new_size
, u32 min_type
)
2023 struct btrfs_path
*path
;
2024 struct btrfs_key key
;
2025 struct btrfs_key found_key
;
2027 struct extent_buffer
*leaf
;
2028 struct btrfs_file_extent_item
*fi
;
2029 u64 extent_start
= 0;
2030 u64 extent_num_bytes
= 0;
2036 int pending_del_nr
= 0;
2037 int pending_del_slot
= 0;
2038 int extent_type
= -1;
2039 u64 mask
= root
->sectorsize
- 1;
2042 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
2043 path
= btrfs_alloc_path();
2047 /* FIXME, add redo link to tree so we don't leak on crash */
2048 key
.objectid
= inode
->i_ino
;
2049 key
.offset
= (u64
)-1;
2052 btrfs_init_path(path
);
2054 ret
= drop_csum_leaves(trans
, root
, path
, inode
, new_size
);
2058 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
2063 /* there are no items in the tree for us to truncate, we're
2066 if (path
->slots
[0] == 0) {
2075 leaf
= path
->nodes
[0];
2076 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2077 found_type
= btrfs_key_type(&found_key
);
2079 if (found_key
.objectid
!= inode
->i_ino
)
2082 if (found_type
< min_type
)
2085 item_end
= found_key
.offset
;
2086 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2087 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2088 struct btrfs_file_extent_item
);
2089 extent_type
= btrfs_file_extent_type(leaf
, fi
);
2090 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2092 btrfs_file_extent_num_bytes(leaf
, fi
);
2093 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2094 item_end
+= btrfs_file_extent_inline_len(leaf
,
2099 if (found_type
== BTRFS_CSUM_ITEM_KEY
) {
2100 ret
= btrfs_csum_truncate(trans
, root
, path
,
2104 if (item_end
< new_size
) {
2105 if (found_type
== BTRFS_DIR_ITEM_KEY
) {
2106 found_type
= BTRFS_INODE_ITEM_KEY
;
2107 } else if (found_type
== BTRFS_EXTENT_ITEM_KEY
) {
2108 found_type
= BTRFS_CSUM_ITEM_KEY
;
2109 } else if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
2110 found_type
= BTRFS_XATTR_ITEM_KEY
;
2111 } else if (found_type
== BTRFS_XATTR_ITEM_KEY
) {
2112 found_type
= BTRFS_INODE_REF_KEY
;
2113 } else if (found_type
) {
2118 btrfs_set_key_type(&key
, found_type
);
2121 if (found_key
.offset
>= new_size
)
2127 /* FIXME, shrink the extent if the ref count is only 1 */
2128 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
2131 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
2133 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
2135 u64 orig_num_bytes
=
2136 btrfs_file_extent_num_bytes(leaf
, fi
);
2137 extent_num_bytes
= new_size
-
2138 found_key
.offset
+ root
->sectorsize
- 1;
2139 extent_num_bytes
= extent_num_bytes
&
2140 ~((u64
)root
->sectorsize
- 1);
2141 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2143 num_dec
= (orig_num_bytes
-
2145 if (root
->ref_cows
&& extent_start
!= 0)
2146 inode_sub_bytes(inode
, num_dec
);
2147 btrfs_mark_buffer_dirty(leaf
);
2150 btrfs_file_extent_disk_num_bytes(leaf
,
2152 /* FIXME blocksize != 4096 */
2153 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
2154 if (extent_start
!= 0) {
2157 inode_sub_bytes(inode
, num_dec
);
2159 root_gen
= btrfs_header_generation(leaf
);
2160 root_owner
= btrfs_header_owner(leaf
);
2162 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
2164 * we can't truncate inline items that have had
2168 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
2169 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
2170 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
2171 u32 size
= new_size
- found_key
.offset
;
2173 if (root
->ref_cows
) {
2174 inode_sub_bytes(inode
, item_end
+ 1 -
2178 btrfs_file_extent_calc_inline_size(size
);
2179 ret
= btrfs_truncate_item(trans
, root
, path
,
2182 } else if (root
->ref_cows
) {
2183 inode_sub_bytes(inode
, item_end
+ 1 -
2189 if (!pending_del_nr
) {
2190 /* no pending yet, add ourselves */
2191 pending_del_slot
= path
->slots
[0];
2193 } else if (pending_del_nr
&&
2194 path
->slots
[0] + 1 == pending_del_slot
) {
2195 /* hop on the pending chunk */
2197 pending_del_slot
= path
->slots
[0];
2199 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path
->slots
[0], pending_del_nr
, pending_del_slot
);
2205 ret
= btrfs_free_extent(trans
, root
, extent_start
,
2207 leaf
->start
, root_owner
,
2208 root_gen
, inode
->i_ino
, 0);
2212 if (path
->slots
[0] == 0) {
2215 btrfs_release_path(root
, path
);
2220 if (pending_del_nr
&&
2221 path
->slots
[0] + 1 != pending_del_slot
) {
2222 struct btrfs_key debug
;
2224 btrfs_item_key_to_cpu(path
->nodes
[0], &debug
,
2226 ret
= btrfs_del_items(trans
, root
, path
,
2231 btrfs_release_path(root
, path
);
2237 if (pending_del_nr
) {
2238 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
2241 btrfs_free_path(path
);
2242 inode
->i_sb
->s_dirt
= 1;
2247 * taken from block_truncate_page, but does cow as it zeros out
2248 * any bytes left in the last page in the file.
2250 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
2252 struct inode
*inode
= mapping
->host
;
2253 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2254 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2255 struct btrfs_ordered_extent
*ordered
;
2257 u32 blocksize
= root
->sectorsize
;
2258 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
2259 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
2265 if ((offset
& (blocksize
- 1)) == 0)
2270 page
= grab_cache_page(mapping
, index
);
2274 page_start
= page_offset(page
);
2275 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
2277 if (!PageUptodate(page
)) {
2278 ret
= btrfs_readpage(NULL
, page
);
2280 if (page
->mapping
!= mapping
) {
2282 page_cache_release(page
);
2285 if (!PageUptodate(page
)) {
2290 wait_on_page_writeback(page
);
2292 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2293 set_page_extent_mapped(page
);
2295 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
2297 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2299 page_cache_release(page
);
2300 btrfs_start_ordered_extent(inode
, ordered
, 1);
2301 btrfs_put_ordered_extent(ordered
);
2305 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
2307 if (offset
!= PAGE_CACHE_SIZE
) {
2309 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
2310 flush_dcache_page(page
);
2313 ClearPageChecked(page
);
2314 set_page_dirty(page
);
2315 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
2319 page_cache_release(page
);
2324 int btrfs_cont_expand(struct inode
*inode
, loff_t size
)
2326 struct btrfs_trans_handle
*trans
;
2327 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2328 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2329 struct extent_map
*em
;
2330 u64 mask
= root
->sectorsize
- 1;
2331 u64 hole_start
= (inode
->i_size
+ mask
) & ~mask
;
2332 u64 block_end
= (size
+ mask
) & ~mask
;
2338 if (size
<= hole_start
)
2341 err
= btrfs_check_free_space(root
, 1, 0);
2345 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
2348 struct btrfs_ordered_extent
*ordered
;
2349 btrfs_wait_ordered_range(inode
, hole_start
,
2350 block_end
- hole_start
);
2351 lock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2352 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
2355 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2356 btrfs_put_ordered_extent(ordered
);
2359 trans
= btrfs_start_transaction(root
, 1);
2360 btrfs_set_trans_block_group(trans
, inode
);
2362 cur_offset
= hole_start
;
2364 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
2365 block_end
- cur_offset
, 0);
2366 BUG_ON(IS_ERR(em
) || !em
);
2367 last_byte
= min(extent_map_end(em
), block_end
);
2368 last_byte
= (last_byte
+ mask
) & ~mask
;
2369 if (test_bit(EXTENT_FLAG_VACANCY
, &em
->flags
)) {
2370 hole_size
= last_byte
- cur_offset
;
2371 err
= btrfs_insert_file_extent(trans
, root
,
2372 inode
->i_ino
, cur_offset
, 0,
2373 0, hole_size
, 0, hole_size
,
2375 btrfs_drop_extent_cache(inode
, hole_start
,
2378 free_extent_map(em
);
2379 cur_offset
= last_byte
;
2380 if (err
|| cur_offset
>= block_end
)
2384 btrfs_end_transaction(trans
, root
);
2385 unlock_extent(io_tree
, hole_start
, block_end
- 1, GFP_NOFS
);
2389 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
2391 struct inode
*inode
= dentry
->d_inode
;
2394 err
= inode_change_ok(inode
, attr
);
2398 if (S_ISREG(inode
->i_mode
) &&
2399 attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
> inode
->i_size
) {
2400 err
= btrfs_cont_expand(inode
, attr
->ia_size
);
2405 err
= inode_setattr(inode
, attr
);
2407 if (!err
&& ((attr
->ia_valid
& ATTR_MODE
)))
2408 err
= btrfs_acl_chmod(inode
);
2412 void btrfs_delete_inode(struct inode
*inode
)
2414 struct btrfs_trans_handle
*trans
;
2415 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2419 truncate_inode_pages(&inode
->i_data
, 0);
2420 if (is_bad_inode(inode
)) {
2421 btrfs_orphan_del(NULL
, inode
);
2424 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
2426 btrfs_i_size_write(inode
, 0);
2427 trans
= btrfs_start_transaction(root
, 1);
2429 btrfs_set_trans_block_group(trans
, inode
);
2430 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
, 0);
2432 btrfs_orphan_del(NULL
, inode
);
2433 goto no_delete_lock
;
2436 btrfs_orphan_del(trans
, inode
);
2438 nr
= trans
->blocks_used
;
2441 btrfs_end_transaction(trans
, root
);
2442 btrfs_btree_balance_dirty(root
, nr
);
2446 nr
= trans
->blocks_used
;
2447 btrfs_end_transaction(trans
, root
);
2448 btrfs_btree_balance_dirty(root
, nr
);
2454 * this returns the key found in the dir entry in the location pointer.
2455 * If no dir entries were found, location->objectid is 0.
2457 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
2458 struct btrfs_key
*location
)
2460 const char *name
= dentry
->d_name
.name
;
2461 int namelen
= dentry
->d_name
.len
;
2462 struct btrfs_dir_item
*di
;
2463 struct btrfs_path
*path
;
2464 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2467 path
= btrfs_alloc_path();
2470 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
2474 if (!di
|| IS_ERR(di
)) {
2477 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
2479 btrfs_free_path(path
);
2482 location
->objectid
= 0;
2487 * when we hit a tree root in a directory, the btrfs part of the inode
2488 * needs to be changed to reflect the root directory of the tree root. This
2489 * is kind of like crossing a mount point.
2491 static int fixup_tree_root_location(struct btrfs_root
*root
,
2492 struct btrfs_key
*location
,
2493 struct btrfs_root
**sub_root
,
2494 struct dentry
*dentry
)
2496 struct btrfs_root_item
*ri
;
2498 if (btrfs_key_type(location
) != BTRFS_ROOT_ITEM_KEY
)
2500 if (location
->objectid
== BTRFS_ROOT_TREE_OBJECTID
)
2503 *sub_root
= btrfs_read_fs_root(root
->fs_info
, location
,
2504 dentry
->d_name
.name
,
2505 dentry
->d_name
.len
);
2506 if (IS_ERR(*sub_root
))
2507 return PTR_ERR(*sub_root
);
2509 ri
= &(*sub_root
)->root_item
;
2510 location
->objectid
= btrfs_root_dirid(ri
);
2511 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
2512 location
->offset
= 0;
2517 static noinline
void init_btrfs_i(struct inode
*inode
)
2519 struct btrfs_inode
*bi
= BTRFS_I(inode
);
2522 bi
->i_default_acl
= NULL
;
2526 bi
->logged_trans
= 0;
2527 bi
->delalloc_bytes
= 0;
2528 bi
->disk_i_size
= 0;
2530 bi
->index_cnt
= (u64
)-1;
2531 bi
->log_dirty_trans
= 0;
2532 extent_map_tree_init(&BTRFS_I(inode
)->extent_tree
, GFP_NOFS
);
2533 extent_io_tree_init(&BTRFS_I(inode
)->io_tree
,
2534 inode
->i_mapping
, GFP_NOFS
);
2535 extent_io_tree_init(&BTRFS_I(inode
)->io_failure_tree
,
2536 inode
->i_mapping
, GFP_NOFS
);
2537 INIT_LIST_HEAD(&BTRFS_I(inode
)->delalloc_inodes
);
2538 btrfs_ordered_inode_tree_init(&BTRFS_I(inode
)->ordered_tree
);
2539 mutex_init(&BTRFS_I(inode
)->csum_mutex
);
2540 mutex_init(&BTRFS_I(inode
)->extent_mutex
);
2541 mutex_init(&BTRFS_I(inode
)->log_mutex
);
2544 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
2546 struct btrfs_iget_args
*args
= p
;
2547 inode
->i_ino
= args
->ino
;
2548 init_btrfs_i(inode
);
2549 BTRFS_I(inode
)->root
= args
->root
;
2553 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
2555 struct btrfs_iget_args
*args
= opaque
;
2556 return (args
->ino
== inode
->i_ino
&&
2557 args
->root
== BTRFS_I(inode
)->root
);
2560 struct inode
*btrfs_ilookup(struct super_block
*s
, u64 objectid
,
2561 struct btrfs_root
*root
, int wait
)
2563 struct inode
*inode
;
2564 struct btrfs_iget_args args
;
2565 args
.ino
= objectid
;
2569 inode
= ilookup5(s
, objectid
, btrfs_find_actor
,
2572 inode
= ilookup5_nowait(s
, objectid
, btrfs_find_actor
,
2578 struct inode
*btrfs_iget_locked(struct super_block
*s
, u64 objectid
,
2579 struct btrfs_root
*root
)
2581 struct inode
*inode
;
2582 struct btrfs_iget_args args
;
2583 args
.ino
= objectid
;
2586 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
2587 btrfs_init_locked_inode
,
2592 /* Get an inode object given its location and corresponding root.
2593 * Returns in *is_new if the inode was read from disk
2595 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
2596 struct btrfs_root
*root
, int *is_new
)
2598 struct inode
*inode
;
2600 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
2602 return ERR_PTR(-EACCES
);
2604 if (inode
->i_state
& I_NEW
) {
2605 BTRFS_I(inode
)->root
= root
;
2606 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
2607 btrfs_read_locked_inode(inode
);
2608 unlock_new_inode(inode
);
2619 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
2620 struct nameidata
*nd
)
2622 struct inode
* inode
;
2623 struct btrfs_inode
*bi
= BTRFS_I(dir
);
2624 struct btrfs_root
*root
= bi
->root
;
2625 struct btrfs_root
*sub_root
= root
;
2626 struct btrfs_key location
;
2627 int ret
, new, do_orphan
= 0;
2629 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
2630 return ERR_PTR(-ENAMETOOLONG
);
2632 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
2635 return ERR_PTR(ret
);
2638 if (location
.objectid
) {
2639 ret
= fixup_tree_root_location(root
, &location
, &sub_root
,
2642 return ERR_PTR(ret
);
2644 return ERR_PTR(-ENOENT
);
2645 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, &new);
2647 return ERR_CAST(inode
);
2649 /* the inode and parent dir are two different roots */
2650 if (new && root
!= sub_root
) {
2652 sub_root
->inode
= inode
;
2657 if (unlikely(do_orphan
))
2658 btrfs_orphan_cleanup(sub_root
);
2660 return d_splice_alias(inode
, dentry
);
2663 static unsigned char btrfs_filetype_table
[] = {
2664 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
2667 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
2670 struct inode
*inode
= filp
->f_dentry
->d_inode
;
2671 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2672 struct btrfs_item
*item
;
2673 struct btrfs_dir_item
*di
;
2674 struct btrfs_key key
;
2675 struct btrfs_key found_key
;
2676 struct btrfs_path
*path
;
2679 struct extent_buffer
*leaf
;
2682 unsigned char d_type
;
2687 int key_type
= BTRFS_DIR_INDEX_KEY
;
2692 /* FIXME, use a real flag for deciding about the key type */
2693 if (root
->fs_info
->tree_root
== root
)
2694 key_type
= BTRFS_DIR_ITEM_KEY
;
2696 /* special case for "." */
2697 if (filp
->f_pos
== 0) {
2698 over
= filldir(dirent
, ".", 1,
2705 /* special case for .., just use the back ref */
2706 if (filp
->f_pos
== 1) {
2707 u64 pino
= parent_ino(filp
->f_path
.dentry
);
2708 over
= filldir(dirent
, "..", 2,
2715 path
= btrfs_alloc_path();
2718 btrfs_set_key_type(&key
, key_type
);
2719 key
.offset
= filp
->f_pos
;
2720 key
.objectid
= inode
->i_ino
;
2722 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2728 leaf
= path
->nodes
[0];
2729 nritems
= btrfs_header_nritems(leaf
);
2730 slot
= path
->slots
[0];
2731 if (advance
|| slot
>= nritems
) {
2732 if (slot
>= nritems
- 1) {
2733 ret
= btrfs_next_leaf(root
, path
);
2736 leaf
= path
->nodes
[0];
2737 nritems
= btrfs_header_nritems(leaf
);
2738 slot
= path
->slots
[0];
2745 item
= btrfs_item_nr(leaf
, slot
);
2746 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2748 if (found_key
.objectid
!= key
.objectid
)
2750 if (btrfs_key_type(&found_key
) != key_type
)
2752 if (found_key
.offset
< filp
->f_pos
)
2755 filp
->f_pos
= found_key
.offset
;
2757 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
2759 di_total
= btrfs_item_size(leaf
, item
);
2761 while (di_cur
< di_total
) {
2762 struct btrfs_key location
;
2764 name_len
= btrfs_dir_name_len(leaf
, di
);
2765 if (name_len
<= sizeof(tmp_name
)) {
2766 name_ptr
= tmp_name
;
2768 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
2774 read_extent_buffer(leaf
, name_ptr
,
2775 (unsigned long)(di
+ 1), name_len
);
2777 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
2778 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
2779 over
= filldir(dirent
, name_ptr
, name_len
,
2780 found_key
.offset
, location
.objectid
,
2783 if (name_ptr
!= tmp_name
)
2789 di_len
= btrfs_dir_name_len(leaf
, di
) +
2790 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
2792 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
2796 /* Reached end of directory/root. Bump pos past the last item. */
2797 if (key_type
== BTRFS_DIR_INDEX_KEY
)
2798 filp
->f_pos
= INT_LIMIT(typeof(filp
->f_pos
));
2804 btrfs_free_path(path
);
2808 int btrfs_write_inode(struct inode
*inode
, int wait
)
2810 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2811 struct btrfs_trans_handle
*trans
;
2814 if (root
->fs_info
->closing
> 1)
2818 trans
= btrfs_join_transaction(root
, 1);
2819 btrfs_set_trans_block_group(trans
, inode
);
2820 ret
= btrfs_commit_transaction(trans
, root
);
2826 * This is somewhat expensive, updating the tree every time the
2827 * inode changes. But, it is most likely to find the inode in cache.
2828 * FIXME, needs more benchmarking...there are no reasons other than performance
2829 * to keep or drop this code.
2831 void btrfs_dirty_inode(struct inode
*inode
)
2833 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2834 struct btrfs_trans_handle
*trans
;
2836 trans
= btrfs_join_transaction(root
, 1);
2837 btrfs_set_trans_block_group(trans
, inode
);
2838 btrfs_update_inode(trans
, root
, inode
);
2839 btrfs_end_transaction(trans
, root
);
2843 * find the highest existing sequence number in a directory
2844 * and then set the in-memory index_cnt variable to reflect
2845 * free sequence numbers
2847 static int btrfs_set_inode_index_count(struct inode
*inode
)
2849 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2850 struct btrfs_key key
, found_key
;
2851 struct btrfs_path
*path
;
2852 struct extent_buffer
*leaf
;
2855 key
.objectid
= inode
->i_ino
;
2856 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
2857 key
.offset
= (u64
)-1;
2859 path
= btrfs_alloc_path();
2863 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2866 /* FIXME: we should be able to handle this */
2872 * MAGIC NUMBER EXPLANATION:
2873 * since we search a directory based on f_pos we have to start at 2
2874 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2875 * else has to start at 2
2877 if (path
->slots
[0] == 0) {
2878 BTRFS_I(inode
)->index_cnt
= 2;
2884 leaf
= path
->nodes
[0];
2885 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2887 if (found_key
.objectid
!= inode
->i_ino
||
2888 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
2889 BTRFS_I(inode
)->index_cnt
= 2;
2893 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
2895 btrfs_free_path(path
);
2900 * helper to find a free sequence number in a given directory. This current
2901 * code is very simple, later versions will do smarter things in the btree
2903 static int btrfs_set_inode_index(struct inode
*dir
, struct inode
*inode
,
2908 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
2909 ret
= btrfs_set_inode_index_count(dir
);
2915 *index
= BTRFS_I(dir
)->index_cnt
;
2916 BTRFS_I(dir
)->index_cnt
++;
2921 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
2922 struct btrfs_root
*root
,
2924 const char *name
, int name_len
,
2927 struct btrfs_block_group_cache
*group
,
2928 int mode
, u64
*index
)
2930 struct inode
*inode
;
2931 struct btrfs_inode_item
*inode_item
;
2932 struct btrfs_block_group_cache
*new_inode_group
;
2933 struct btrfs_key
*location
;
2934 struct btrfs_path
*path
;
2935 struct btrfs_inode_ref
*ref
;
2936 struct btrfs_key key
[2];
2942 path
= btrfs_alloc_path();
2945 inode
= new_inode(root
->fs_info
->sb
);
2947 return ERR_PTR(-ENOMEM
);
2950 ret
= btrfs_set_inode_index(dir
, inode
, index
);
2952 return ERR_PTR(ret
);
2955 * index_cnt is ignored for everything but a dir,
2956 * btrfs_get_inode_index_count has an explanation for the magic
2959 init_btrfs_i(inode
);
2960 BTRFS_I(inode
)->index_cnt
= 2;
2961 BTRFS_I(inode
)->root
= root
;
2962 BTRFS_I(inode
)->generation
= trans
->transid
;
2968 new_inode_group
= btrfs_find_block_group(root
, group
, 0,
2969 BTRFS_BLOCK_GROUP_METADATA
, owner
);
2970 if (!new_inode_group
) {
2971 printk("find_block group failed\n");
2972 new_inode_group
= group
;
2974 BTRFS_I(inode
)->block_group
= new_inode_group
;
2976 key
[0].objectid
= objectid
;
2977 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
2980 key
[1].objectid
= objectid
;
2981 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
2982 key
[1].offset
= ref_objectid
;
2984 sizes
[0] = sizeof(struct btrfs_inode_item
);
2985 sizes
[1] = name_len
+ sizeof(*ref
);
2987 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
2991 if (objectid
> root
->highest_inode
)
2992 root
->highest_inode
= objectid
;
2994 inode
->i_uid
= current
->fsuid
;
2995 inode
->i_gid
= current
->fsgid
;
2996 inode
->i_mode
= mode
;
2997 inode
->i_ino
= objectid
;
2998 inode_set_bytes(inode
, 0);
2999 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3000 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
3001 struct btrfs_inode_item
);
3002 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
3004 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
3005 struct btrfs_inode_ref
);
3006 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
3007 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
3008 ptr
= (unsigned long)(ref
+ 1);
3009 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
3011 btrfs_mark_buffer_dirty(path
->nodes
[0]);
3012 btrfs_free_path(path
);
3014 location
= &BTRFS_I(inode
)->location
;
3015 location
->objectid
= objectid
;
3016 location
->offset
= 0;
3017 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
3019 insert_inode_hash(inode
);
3023 BTRFS_I(dir
)->index_cnt
--;
3024 btrfs_free_path(path
);
3025 return ERR_PTR(ret
);
3028 static inline u8
btrfs_inode_type(struct inode
*inode
)
3030 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
3034 * utility function to add 'inode' into 'parent_inode' with
3035 * a give name and a given sequence number.
3036 * if 'add_backref' is true, also insert a backref from the
3037 * inode to the parent directory.
3039 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
3040 struct inode
*parent_inode
, struct inode
*inode
,
3041 const char *name
, int name_len
, int add_backref
, u64 index
)
3044 struct btrfs_key key
;
3045 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
3047 key
.objectid
= inode
->i_ino
;
3048 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
3051 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
3052 parent_inode
->i_ino
,
3053 &key
, btrfs_inode_type(inode
),
3057 ret
= btrfs_insert_inode_ref(trans
, root
,
3060 parent_inode
->i_ino
,
3063 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
3065 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
3066 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
3071 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
3072 struct dentry
*dentry
, struct inode
*inode
,
3073 int backref
, u64 index
)
3075 int err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3076 inode
, dentry
->d_name
.name
,
3077 dentry
->d_name
.len
, backref
, index
);
3079 d_instantiate(dentry
, inode
);
3087 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
3088 int mode
, dev_t rdev
)
3090 struct btrfs_trans_handle
*trans
;
3091 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3092 struct inode
*inode
= NULL
;
3096 unsigned long nr
= 0;
3099 if (!new_valid_dev(rdev
))
3102 err
= btrfs_check_free_space(root
, 1, 0);
3106 trans
= btrfs_start_transaction(root
, 1);
3107 btrfs_set_trans_block_group(trans
, dir
);
3109 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3115 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3117 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3118 BTRFS_I(dir
)->block_group
, mode
, &index
);
3119 err
= PTR_ERR(inode
);
3123 err
= btrfs_init_acl(inode
, dir
);
3129 btrfs_set_trans_block_group(trans
, inode
);
3130 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3134 inode
->i_op
= &btrfs_special_inode_operations
;
3135 init_special_inode(inode
, inode
->i_mode
, rdev
);
3136 btrfs_update_inode(trans
, root
, inode
);
3138 dir
->i_sb
->s_dirt
= 1;
3139 btrfs_update_inode_block_group(trans
, inode
);
3140 btrfs_update_inode_block_group(trans
, dir
);
3142 nr
= trans
->blocks_used
;
3143 btrfs_end_transaction_throttle(trans
, root
);
3146 inode_dec_link_count(inode
);
3149 btrfs_btree_balance_dirty(root
, nr
);
3153 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
3154 int mode
, struct nameidata
*nd
)
3156 struct btrfs_trans_handle
*trans
;
3157 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3158 struct inode
*inode
= NULL
;
3161 unsigned long nr
= 0;
3165 err
= btrfs_check_free_space(root
, 1, 0);
3168 trans
= btrfs_start_transaction(root
, 1);
3169 btrfs_set_trans_block_group(trans
, dir
);
3171 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3177 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3179 dentry
->d_parent
->d_inode
->i_ino
,
3180 objectid
, BTRFS_I(dir
)->block_group
, mode
,
3182 err
= PTR_ERR(inode
);
3186 err
= btrfs_init_acl(inode
, dir
);
3192 btrfs_set_trans_block_group(trans
, inode
);
3193 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
3197 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3198 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3199 inode
->i_fop
= &btrfs_file_operations
;
3200 inode
->i_op
= &btrfs_file_inode_operations
;
3201 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3203 dir
->i_sb
->s_dirt
= 1;
3204 btrfs_update_inode_block_group(trans
, inode
);
3205 btrfs_update_inode_block_group(trans
, dir
);
3207 nr
= trans
->blocks_used
;
3208 btrfs_end_transaction_throttle(trans
, root
);
3211 inode_dec_link_count(inode
);
3214 btrfs_btree_balance_dirty(root
, nr
);
3218 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
3219 struct dentry
*dentry
)
3221 struct btrfs_trans_handle
*trans
;
3222 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3223 struct inode
*inode
= old_dentry
->d_inode
;
3225 unsigned long nr
= 0;
3229 if (inode
->i_nlink
== 0)
3232 btrfs_inc_nlink(inode
);
3233 err
= btrfs_check_free_space(root
, 1, 0);
3236 err
= btrfs_set_inode_index(dir
, inode
, &index
);
3240 trans
= btrfs_start_transaction(root
, 1);
3242 btrfs_set_trans_block_group(trans
, dir
);
3243 atomic_inc(&inode
->i_count
);
3245 err
= btrfs_add_nondir(trans
, dentry
, inode
, 1, index
);
3250 dir
->i_sb
->s_dirt
= 1;
3251 btrfs_update_inode_block_group(trans
, dir
);
3252 err
= btrfs_update_inode(trans
, root
, inode
);
3257 nr
= trans
->blocks_used
;
3258 btrfs_end_transaction_throttle(trans
, root
);
3261 inode_dec_link_count(inode
);
3264 btrfs_btree_balance_dirty(root
, nr
);
3268 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
3270 struct inode
*inode
= NULL
;
3271 struct btrfs_trans_handle
*trans
;
3272 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3274 int drop_on_err
= 0;
3277 unsigned long nr
= 1;
3279 err
= btrfs_check_free_space(root
, 1, 0);
3283 trans
= btrfs_start_transaction(root
, 1);
3284 btrfs_set_trans_block_group(trans
, dir
);
3286 if (IS_ERR(trans
)) {
3287 err
= PTR_ERR(trans
);
3291 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
3297 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
3299 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
3300 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
3302 if (IS_ERR(inode
)) {
3303 err
= PTR_ERR(inode
);
3309 err
= btrfs_init_acl(inode
, dir
);
3313 inode
->i_op
= &btrfs_dir_inode_operations
;
3314 inode
->i_fop
= &btrfs_dir_file_operations
;
3315 btrfs_set_trans_block_group(trans
, inode
);
3317 btrfs_i_size_write(inode
, 0);
3318 err
= btrfs_update_inode(trans
, root
, inode
);
3322 err
= btrfs_add_link(trans
, dentry
->d_parent
->d_inode
,
3323 inode
, dentry
->d_name
.name
,
3324 dentry
->d_name
.len
, 0, index
);
3328 d_instantiate(dentry
, inode
);
3330 dir
->i_sb
->s_dirt
= 1;
3331 btrfs_update_inode_block_group(trans
, inode
);
3332 btrfs_update_inode_block_group(trans
, dir
);
3335 nr
= trans
->blocks_used
;
3336 btrfs_end_transaction_throttle(trans
, root
);
3341 btrfs_btree_balance_dirty(root
, nr
);
3345 /* helper for btfs_get_extent. Given an existing extent in the tree,
3346 * and an extent that you want to insert, deal with overlap and insert
3347 * the new extent into the tree.
3349 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
3350 struct extent_map
*existing
,
3351 struct extent_map
*em
,
3352 u64 map_start
, u64 map_len
)
3356 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
3357 start_diff
= map_start
- em
->start
;
3358 em
->start
= map_start
;
3360 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
3361 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
3362 em
->block_start
+= start_diff
;
3363 em
->block_len
-= start_diff
;
3365 return add_extent_mapping(em_tree
, em
);
3368 static noinline
int uncompress_inline(struct btrfs_path
*path
,
3369 struct inode
*inode
, struct page
*page
,
3370 size_t pg_offset
, u64 extent_offset
,
3371 struct btrfs_file_extent_item
*item
)
3374 struct extent_buffer
*leaf
= path
->nodes
[0];
3377 unsigned long inline_size
;
3380 WARN_ON(pg_offset
!= 0);
3381 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
3382 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
3383 btrfs_item_nr(leaf
, path
->slots
[0]));
3384 tmp
= kmalloc(inline_size
, GFP_NOFS
);
3385 ptr
= btrfs_file_extent_inline_start(item
);
3387 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
3389 max_size
= min(PAGE_CACHE_SIZE
, max_size
);
3390 ret
= btrfs_zlib_decompress(tmp
, page
, extent_offset
,
3391 inline_size
, max_size
);
3393 char *kaddr
= kmap_atomic(page
, KM_USER0
);
3394 unsigned long copy_size
= min_t(u64
,
3395 PAGE_CACHE_SIZE
- pg_offset
,
3396 max_size
- extent_offset
);
3397 memset(kaddr
+ pg_offset
, 0, copy_size
);
3398 kunmap_atomic(kaddr
, KM_USER0
);
3405 * a bit scary, this does extent mapping from logical file offset to the disk.
3406 * the ugly parts come from merging extents from the disk with the
3407 * in-ram representation. This gets more complex because of the data=ordered code,
3408 * where the in-ram extents might be locked pending data=ordered completion.
3410 * This also copies inline extents directly into the page.
3412 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
3413 size_t pg_offset
, u64 start
, u64 len
,
3419 u64 extent_start
= 0;
3421 u64 objectid
= inode
->i_ino
;
3423 struct btrfs_path
*path
= NULL
;
3424 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3425 struct btrfs_file_extent_item
*item
;
3426 struct extent_buffer
*leaf
;
3427 struct btrfs_key found_key
;
3428 struct extent_map
*em
= NULL
;
3429 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3430 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3431 struct btrfs_trans_handle
*trans
= NULL
;
3435 spin_lock(&em_tree
->lock
);
3436 em
= lookup_extent_mapping(em_tree
, start
, len
);
3438 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3439 spin_unlock(&em_tree
->lock
);
3442 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
3443 free_extent_map(em
);
3444 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
3445 free_extent_map(em
);
3449 em
= alloc_extent_map(GFP_NOFS
);
3454 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3455 em
->start
= EXTENT_MAP_HOLE
;
3457 em
->block_len
= (u64
)-1;
3460 path
= btrfs_alloc_path();
3464 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3465 objectid
, start
, trans
!= NULL
);
3472 if (path
->slots
[0] == 0)
3477 leaf
= path
->nodes
[0];
3478 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3479 struct btrfs_file_extent_item
);
3480 /* are we inside the extent that was found? */
3481 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3482 found_type
= btrfs_key_type(&found_key
);
3483 if (found_key
.objectid
!= objectid
||
3484 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
3488 found_type
= btrfs_file_extent_type(leaf
, item
);
3489 extent_start
= found_key
.offset
;
3490 compressed
= btrfs_file_extent_compression(leaf
, item
);
3491 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
3492 extent_end
= extent_start
+
3493 btrfs_file_extent_num_bytes(leaf
, item
);
3494 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3496 size
= btrfs_file_extent_inline_len(leaf
, item
);
3497 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
3498 ~((u64
)root
->sectorsize
- 1);
3501 if (start
>= extent_end
) {
3503 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3504 ret
= btrfs_next_leaf(root
, path
);
3511 leaf
= path
->nodes
[0];
3513 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3514 if (found_key
.objectid
!= objectid
||
3515 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
3517 if (start
+ len
<= found_key
.offset
)
3520 em
->len
= found_key
.offset
- start
;
3524 if (found_type
== BTRFS_FILE_EXTENT_REG
) {
3525 em
->start
= extent_start
;
3526 em
->len
= extent_end
- extent_start
;
3527 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
3529 em
->block_start
= EXTENT_MAP_HOLE
;
3533 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3534 em
->block_start
= bytenr
;
3535 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
3538 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
3539 em
->block_start
= bytenr
;
3540 em
->block_len
= em
->len
;
3543 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
3547 size_t extent_offset
;
3550 em
->block_start
= EXTENT_MAP_INLINE
;
3551 if (!page
|| create
) {
3552 em
->start
= extent_start
;
3553 em
->len
= extent_end
- extent_start
;
3557 size
= btrfs_file_extent_inline_len(leaf
, item
);
3558 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
3559 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
3560 size
- extent_offset
);
3561 em
->start
= extent_start
+ extent_offset
;
3562 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
3563 ~((u64
)root
->sectorsize
- 1);
3565 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
3566 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
3567 if (create
== 0 && !PageUptodate(page
)) {
3568 if (btrfs_file_extent_compression(leaf
, item
) ==
3569 BTRFS_COMPRESS_ZLIB
) {
3570 ret
= uncompress_inline(path
, inode
, page
,
3572 extent_offset
, item
);
3576 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
3580 flush_dcache_page(page
);
3581 } else if (create
&& PageUptodate(page
)) {
3584 free_extent_map(em
);
3586 btrfs_release_path(root
, path
);
3587 trans
= btrfs_join_transaction(root
, 1);
3591 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
3594 btrfs_mark_buffer_dirty(leaf
);
3596 set_extent_uptodate(io_tree
, em
->start
,
3597 extent_map_end(em
) - 1, GFP_NOFS
);
3600 printk("unkknown found_type %d\n", found_type
);
3607 em
->block_start
= EXTENT_MAP_HOLE
;
3608 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
3610 btrfs_release_path(root
, path
);
3611 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
3612 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em
->start
, em
->len
, start
, len
);
3618 spin_lock(&em_tree
->lock
);
3619 ret
= add_extent_mapping(em_tree
, em
);
3620 /* it is possible that someone inserted the extent into the tree
3621 * while we had the lock dropped. It is also possible that
3622 * an overlapping map exists in the tree
3624 if (ret
== -EEXIST
) {
3625 struct extent_map
*existing
;
3629 existing
= lookup_extent_mapping(em_tree
, start
, len
);
3630 if (existing
&& (existing
->start
> start
||
3631 existing
->start
+ existing
->len
<= start
)) {
3632 free_extent_map(existing
);
3636 existing
= lookup_extent_mapping(em_tree
, em
->start
,
3639 err
= merge_extent_mapping(em_tree
, existing
,
3642 free_extent_map(existing
);
3644 free_extent_map(em
);
3649 printk("failing to insert %Lu %Lu\n",
3651 free_extent_map(em
);
3655 free_extent_map(em
);
3660 spin_unlock(&em_tree
->lock
);
3663 btrfs_free_path(path
);
3665 ret
= btrfs_end_transaction(trans
, root
);
3671 free_extent_map(em
);
3673 return ERR_PTR(err
);
3678 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
3679 const struct iovec
*iov
, loff_t offset
,
3680 unsigned long nr_segs
)
3685 static sector_t
btrfs_bmap(struct address_space
*mapping
, sector_t iblock
)
3687 return extent_bmap(mapping
, iblock
, btrfs_get_extent
);
3690 int btrfs_readpage(struct file
*file
, struct page
*page
)
3692 struct extent_io_tree
*tree
;
3693 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
3694 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
3697 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
3699 struct extent_io_tree
*tree
;
3702 if (current
->flags
& PF_MEMALLOC
) {
3703 redirty_page_for_writepage(wbc
, page
);
3707 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
3708 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
3711 int btrfs_writepages(struct address_space
*mapping
,
3712 struct writeback_control
*wbc
)
3714 struct extent_io_tree
*tree
;
3715 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3716 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
3720 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
3721 struct list_head
*pages
, unsigned nr_pages
)
3723 struct extent_io_tree
*tree
;
3724 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
3725 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
3728 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
3730 struct extent_io_tree
*tree
;
3731 struct extent_map_tree
*map
;
3734 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
3735 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
3736 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
3738 ClearPagePrivate(page
);
3739 set_page_private(page
, 0);
3740 page_cache_release(page
);
3745 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
3747 if (PageWriteback(page
) || PageDirty(page
))
3749 return __btrfs_releasepage(page
, gfp_flags
);
3752 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
3754 struct extent_io_tree
*tree
;
3755 struct btrfs_ordered_extent
*ordered
;
3756 u64 page_start
= page_offset(page
);
3757 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3759 wait_on_page_writeback(page
);
3760 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
3762 btrfs_releasepage(page
, GFP_NOFS
);
3766 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
3767 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
3771 * IO on this page will never be started, so we need
3772 * to account for any ordered extents now
3774 clear_extent_bit(tree
, page_start
, page_end
,
3775 EXTENT_DIRTY
| EXTENT_DELALLOC
|
3776 EXTENT_LOCKED
, 1, 0, GFP_NOFS
);
3777 btrfs_finish_ordered_io(page
->mapping
->host
,
3778 page_start
, page_end
);
3779 btrfs_put_ordered_extent(ordered
);
3780 lock_extent(tree
, page_start
, page_end
, GFP_NOFS
);
3782 clear_extent_bit(tree
, page_start
, page_end
,
3783 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
3786 __btrfs_releasepage(page
, GFP_NOFS
);
3788 ClearPageChecked(page
);
3789 if (PagePrivate(page
)) {
3790 ClearPagePrivate(page
);
3791 set_page_private(page
, 0);
3792 page_cache_release(page
);
3797 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3798 * called from a page fault handler when a page is first dirtied. Hence we must
3799 * be careful to check for EOF conditions here. We set the page up correctly
3800 * for a written page which means we get ENOSPC checking when writing into
3801 * holes and correct delalloc and unwritten extent mapping on filesystems that
3802 * support these features.
3804 * We are not allowed to take the i_mutex here so we have to play games to
3805 * protect against truncate races as the page could now be beyond EOF. Because
3806 * vmtruncate() writes the inode size before removing pages, once we have the
3807 * page lock we can determine safely if the page is beyond EOF. If it is not
3808 * beyond EOF, then the page is guaranteed safe against truncation until we
3811 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
3813 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
3814 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3815 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3816 struct btrfs_ordered_extent
*ordered
;
3818 unsigned long zero_start
;
3824 ret
= btrfs_check_free_space(root
, PAGE_CACHE_SIZE
, 0);
3831 size
= i_size_read(inode
);
3832 page_start
= page_offset(page
);
3833 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3835 if ((page
->mapping
!= inode
->i_mapping
) ||
3836 (page_start
>= size
)) {
3837 /* page got truncated out from underneath us */
3840 wait_on_page_writeback(page
);
3842 lock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3843 set_page_extent_mapped(page
);
3846 * we can't set the delalloc bits if there are pending ordered
3847 * extents. Drop our locks and wait for them to finish
3849 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3851 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3853 btrfs_start_ordered_extent(inode
, ordered
, 1);
3854 btrfs_put_ordered_extent(ordered
);
3858 btrfs_set_extent_delalloc(inode
, page_start
, page_end
);
3861 /* page is wholly or partially inside EOF */
3862 if (page_start
+ PAGE_CACHE_SIZE
> size
)
3863 zero_start
= size
& ~PAGE_CACHE_MASK
;
3865 zero_start
= PAGE_CACHE_SIZE
;
3867 if (zero_start
!= PAGE_CACHE_SIZE
) {
3869 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
3870 flush_dcache_page(page
);
3873 ClearPageChecked(page
);
3874 set_page_dirty(page
);
3875 unlock_extent(io_tree
, page_start
, page_end
, GFP_NOFS
);
3883 static void btrfs_truncate(struct inode
*inode
)
3885 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3887 struct btrfs_trans_handle
*trans
;
3889 u64 mask
= root
->sectorsize
- 1;
3891 if (!S_ISREG(inode
->i_mode
))
3893 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3896 btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
3897 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
3899 trans
= btrfs_start_transaction(root
, 1);
3900 btrfs_set_trans_block_group(trans
, inode
);
3901 btrfs_i_size_write(inode
, inode
->i_size
);
3903 ret
= btrfs_orphan_add(trans
, inode
);
3906 /* FIXME, add redo link to tree so we don't leak on crash */
3907 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, inode
->i_size
,
3908 BTRFS_EXTENT_DATA_KEY
);
3909 btrfs_update_inode(trans
, root
, inode
);
3911 ret
= btrfs_orphan_del(trans
, inode
);
3915 nr
= trans
->blocks_used
;
3916 ret
= btrfs_end_transaction_throttle(trans
, root
);
3918 btrfs_btree_balance_dirty(root
, nr
);
3922 * Invalidate a single dcache entry at the root of the filesystem.
3923 * Needed after creation of snapshot or subvolume.
3925 void btrfs_invalidate_dcache_root(struct btrfs_root
*root
, char *name
,
3928 struct dentry
*alias
, *entry
;
3931 alias
= d_find_alias(root
->fs_info
->sb
->s_root
->d_inode
);
3935 /* change me if btrfs ever gets a d_hash operation */
3936 qstr
.hash
= full_name_hash(qstr
.name
, qstr
.len
);
3937 entry
= d_lookup(alias
, &qstr
);
3940 d_invalidate(entry
);
3947 * create a new subvolume directory/inode (helper for the ioctl).
3949 int btrfs_create_subvol_root(struct btrfs_root
*new_root
, struct dentry
*dentry
,
3950 struct btrfs_trans_handle
*trans
, u64 new_dirid
,
3951 struct btrfs_block_group_cache
*block_group
)
3953 struct inode
*inode
;
3957 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
3958 new_dirid
, block_group
, S_IFDIR
| 0700, &index
);
3960 return PTR_ERR(inode
);
3961 inode
->i_op
= &btrfs_dir_inode_operations
;
3962 inode
->i_fop
= &btrfs_dir_file_operations
;
3963 new_root
->inode
= inode
;
3966 btrfs_i_size_write(inode
, 0);
3968 error
= btrfs_update_inode(trans
, new_root
, inode
);
3972 d_instantiate(dentry
, inode
);
3976 /* helper function for file defrag and space balancing. This
3977 * forces readahead on a given range of bytes in an inode
3979 unsigned long btrfs_force_ra(struct address_space
*mapping
,
3980 struct file_ra_state
*ra
, struct file
*file
,
3981 pgoff_t offset
, pgoff_t last_index
)
3983 pgoff_t req_size
= last_index
- offset
+ 1;
3985 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
3986 return offset
+ req_size
;
3989 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
3991 struct btrfs_inode
*ei
;
3993 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
3997 ei
->logged_trans
= 0;
3998 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
3999 ei
->i_acl
= BTRFS_ACL_NOT_CACHED
;
4000 ei
->i_default_acl
= BTRFS_ACL_NOT_CACHED
;
4001 INIT_LIST_HEAD(&ei
->i_orphan
);
4002 return &ei
->vfs_inode
;
4005 void btrfs_destroy_inode(struct inode
*inode
)
4007 struct btrfs_ordered_extent
*ordered
;
4008 WARN_ON(!list_empty(&inode
->i_dentry
));
4009 WARN_ON(inode
->i_data
.nrpages
);
4011 if (BTRFS_I(inode
)->i_acl
&&
4012 BTRFS_I(inode
)->i_acl
!= BTRFS_ACL_NOT_CACHED
)
4013 posix_acl_release(BTRFS_I(inode
)->i_acl
);
4014 if (BTRFS_I(inode
)->i_default_acl
&&
4015 BTRFS_I(inode
)->i_default_acl
!= BTRFS_ACL_NOT_CACHED
)
4016 posix_acl_release(BTRFS_I(inode
)->i_default_acl
);
4018 spin_lock(&BTRFS_I(inode
)->root
->list_lock
);
4019 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
4020 printk(KERN_ERR
"BTRFS: inode %lu: inode still on the orphan"
4021 " list\n", inode
->i_ino
);
4024 spin_unlock(&BTRFS_I(inode
)->root
->list_lock
);
4027 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
4031 printk("found ordered extent %Lu %Lu\n",
4032 ordered
->file_offset
, ordered
->len
);
4033 btrfs_remove_ordered_extent(inode
, ordered
);
4034 btrfs_put_ordered_extent(ordered
);
4035 btrfs_put_ordered_extent(ordered
);
4038 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
4039 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
4042 static void init_once(void *foo
)
4044 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
4046 inode_init_once(&ei
->vfs_inode
);
4049 void btrfs_destroy_cachep(void)
4051 if (btrfs_inode_cachep
)
4052 kmem_cache_destroy(btrfs_inode_cachep
);
4053 if (btrfs_trans_handle_cachep
)
4054 kmem_cache_destroy(btrfs_trans_handle_cachep
);
4055 if (btrfs_transaction_cachep
)
4056 kmem_cache_destroy(btrfs_transaction_cachep
);
4057 if (btrfs_bit_radix_cachep
)
4058 kmem_cache_destroy(btrfs_bit_radix_cachep
);
4059 if (btrfs_path_cachep
)
4060 kmem_cache_destroy(btrfs_path_cachep
);
4063 struct kmem_cache
*btrfs_cache_create(const char *name
, size_t size
,
4064 unsigned long extra_flags
,
4065 void (*ctor
)(void *))
4067 return kmem_cache_create(name
, size
, 0, (SLAB_RECLAIM_ACCOUNT
|
4068 SLAB_MEM_SPREAD
| extra_flags
), ctor
);
4071 int btrfs_init_cachep(void)
4073 btrfs_inode_cachep
= btrfs_cache_create("btrfs_inode_cache",
4074 sizeof(struct btrfs_inode
),
4076 if (!btrfs_inode_cachep
)
4078 btrfs_trans_handle_cachep
=
4079 btrfs_cache_create("btrfs_trans_handle_cache",
4080 sizeof(struct btrfs_trans_handle
),
4082 if (!btrfs_trans_handle_cachep
)
4084 btrfs_transaction_cachep
= btrfs_cache_create("btrfs_transaction_cache",
4085 sizeof(struct btrfs_transaction
),
4087 if (!btrfs_transaction_cachep
)
4089 btrfs_path_cachep
= btrfs_cache_create("btrfs_path_cache",
4090 sizeof(struct btrfs_path
),
4092 if (!btrfs_path_cachep
)
4094 btrfs_bit_radix_cachep
= btrfs_cache_create("btrfs_radix", 256,
4095 SLAB_DESTROY_BY_RCU
, NULL
);
4096 if (!btrfs_bit_radix_cachep
)
4100 btrfs_destroy_cachep();
4104 static int btrfs_getattr(struct vfsmount
*mnt
,
4105 struct dentry
*dentry
, struct kstat
*stat
)
4107 struct inode
*inode
= dentry
->d_inode
;
4108 generic_fillattr(inode
, stat
);
4109 stat
->blksize
= PAGE_CACHE_SIZE
;
4110 stat
->blocks
= (inode_get_bytes(inode
) +
4111 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
4115 static int btrfs_rename(struct inode
* old_dir
, struct dentry
*old_dentry
,
4116 struct inode
* new_dir
,struct dentry
*new_dentry
)
4118 struct btrfs_trans_handle
*trans
;
4119 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
4120 struct inode
*new_inode
= new_dentry
->d_inode
;
4121 struct inode
*old_inode
= old_dentry
->d_inode
;
4122 struct timespec ctime
= CURRENT_TIME
;
4126 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
4127 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
) {
4131 ret
= btrfs_check_free_space(root
, 1, 0);
4135 trans
= btrfs_start_transaction(root
, 1);
4137 btrfs_set_trans_block_group(trans
, new_dir
);
4139 btrfs_inc_nlink(old_dentry
->d_inode
);
4140 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
4141 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
4142 old_inode
->i_ctime
= ctime
;
4144 ret
= btrfs_unlink_inode(trans
, root
, old_dir
, old_dentry
->d_inode
,
4145 old_dentry
->d_name
.name
,
4146 old_dentry
->d_name
.len
);
4151 new_inode
->i_ctime
= CURRENT_TIME
;
4152 ret
= btrfs_unlink_inode(trans
, root
, new_dir
,
4153 new_dentry
->d_inode
,
4154 new_dentry
->d_name
.name
,
4155 new_dentry
->d_name
.len
);
4158 if (new_inode
->i_nlink
== 0) {
4159 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
4165 ret
= btrfs_set_inode_index(new_dir
, old_inode
, &index
);
4169 ret
= btrfs_add_link(trans
, new_dentry
->d_parent
->d_inode
,
4170 old_inode
, new_dentry
->d_name
.name
,
4171 new_dentry
->d_name
.len
, 1, index
);
4176 btrfs_end_transaction_throttle(trans
, root
);
4182 * some fairly slow code that needs optimization. This walks the list
4183 * of all the inodes with pending delalloc and forces them to disk.
4185 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
)
4187 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
4188 struct btrfs_inode
*binode
;
4189 struct inode
*inode
;
4190 unsigned long flags
;
4192 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
4193 while(!list_empty(head
)) {
4194 binode
= list_entry(head
->next
, struct btrfs_inode
,
4196 inode
= igrab(&binode
->vfs_inode
);
4198 list_del_init(&binode
->delalloc_inodes
);
4199 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
4201 filemap_flush(inode
->i_mapping
);
4205 spin_lock_irqsave(&root
->fs_info
->delalloc_lock
, flags
);
4207 spin_unlock_irqrestore(&root
->fs_info
->delalloc_lock
, flags
);
4209 /* the filemap_flush will queue IO into the worker threads, but
4210 * we have to make sure the IO is actually started and that
4211 * ordered extents get created before we return
4213 atomic_inc(&root
->fs_info
->async_submit_draining
);
4214 while(atomic_read(&root
->fs_info
->nr_async_submits
)) {
4215 wait_event(root
->fs_info
->async_submit_wait
,
4216 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0));
4218 atomic_dec(&root
->fs_info
->async_submit_draining
);
4222 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
4223 const char *symname
)
4225 struct btrfs_trans_handle
*trans
;
4226 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4227 struct btrfs_path
*path
;
4228 struct btrfs_key key
;
4229 struct inode
*inode
= NULL
;
4237 struct btrfs_file_extent_item
*ei
;
4238 struct extent_buffer
*leaf
;
4239 unsigned long nr
= 0;
4241 name_len
= strlen(symname
) + 1;
4242 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
4243 return -ENAMETOOLONG
;
4245 err
= btrfs_check_free_space(root
, 1, 0);
4249 trans
= btrfs_start_transaction(root
, 1);
4250 btrfs_set_trans_block_group(trans
, dir
);
4252 err
= btrfs_find_free_objectid(trans
, root
, dir
->i_ino
, &objectid
);
4258 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4260 dentry
->d_parent
->d_inode
->i_ino
, objectid
,
4261 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
4263 err
= PTR_ERR(inode
);
4267 err
= btrfs_init_acl(inode
, dir
);
4273 btrfs_set_trans_block_group(trans
, inode
);
4274 err
= btrfs_add_nondir(trans
, dentry
, inode
, 0, index
);
4278 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4279 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4280 inode
->i_fop
= &btrfs_file_operations
;
4281 inode
->i_op
= &btrfs_file_inode_operations
;
4282 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4284 dir
->i_sb
->s_dirt
= 1;
4285 btrfs_update_inode_block_group(trans
, inode
);
4286 btrfs_update_inode_block_group(trans
, dir
);
4290 path
= btrfs_alloc_path();
4292 key
.objectid
= inode
->i_ino
;
4294 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
4295 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
4296 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
4302 leaf
= path
->nodes
[0];
4303 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
4304 struct btrfs_file_extent_item
);
4305 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
4306 btrfs_set_file_extent_type(leaf
, ei
,
4307 BTRFS_FILE_EXTENT_INLINE
);
4308 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
4309 btrfs_set_file_extent_compression(leaf
, ei
, 0);
4310 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
4311 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
4313 ptr
= btrfs_file_extent_inline_start(ei
);
4314 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
4315 btrfs_mark_buffer_dirty(leaf
);
4316 btrfs_free_path(path
);
4318 inode
->i_op
= &btrfs_symlink_inode_operations
;
4319 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
4320 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4321 btrfs_i_size_write(inode
, name_len
- 1);
4322 err
= btrfs_update_inode(trans
, root
, inode
);
4327 nr
= trans
->blocks_used
;
4328 btrfs_end_transaction_throttle(trans
, root
);
4331 inode_dec_link_count(inode
);
4334 btrfs_btree_balance_dirty(root
, nr
);
4338 static int btrfs_set_page_dirty(struct page
*page
)
4340 return __set_page_dirty_nobuffers(page
);
4343 static int btrfs_permission(struct inode
*inode
, int mask
)
4345 if (btrfs_test_flag(inode
, READONLY
) && (mask
& MAY_WRITE
))
4347 return generic_permission(inode
, mask
, btrfs_check_acl
);
4350 static struct inode_operations btrfs_dir_inode_operations
= {
4351 .lookup
= btrfs_lookup
,
4352 .create
= btrfs_create
,
4353 .unlink
= btrfs_unlink
,
4355 .mkdir
= btrfs_mkdir
,
4356 .rmdir
= btrfs_rmdir
,
4357 .rename
= btrfs_rename
,
4358 .symlink
= btrfs_symlink
,
4359 .setattr
= btrfs_setattr
,
4360 .mknod
= btrfs_mknod
,
4361 .setxattr
= btrfs_setxattr
,
4362 .getxattr
= btrfs_getxattr
,
4363 .listxattr
= btrfs_listxattr
,
4364 .removexattr
= btrfs_removexattr
,
4365 .permission
= btrfs_permission
,
4367 static struct inode_operations btrfs_dir_ro_inode_operations
= {
4368 .lookup
= btrfs_lookup
,
4369 .permission
= btrfs_permission
,
4371 static struct file_operations btrfs_dir_file_operations
= {
4372 .llseek
= generic_file_llseek
,
4373 .read
= generic_read_dir
,
4374 .readdir
= btrfs_real_readdir
,
4375 .unlocked_ioctl
= btrfs_ioctl
,
4376 #ifdef CONFIG_COMPAT
4377 .compat_ioctl
= btrfs_ioctl
,
4379 .release
= btrfs_release_file
,
4380 .fsync
= btrfs_sync_file
,
4383 static struct extent_io_ops btrfs_extent_io_ops
= {
4384 .fill_delalloc
= run_delalloc_range
,
4385 .submit_bio_hook
= btrfs_submit_bio_hook
,
4386 .merge_bio_hook
= btrfs_merge_bio_hook
,
4387 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
4388 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
4389 .writepage_start_hook
= btrfs_writepage_start_hook
,
4390 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
4391 .set_bit_hook
= btrfs_set_bit_hook
,
4392 .clear_bit_hook
= btrfs_clear_bit_hook
,
4395 static struct address_space_operations btrfs_aops
= {
4396 .readpage
= btrfs_readpage
,
4397 .writepage
= btrfs_writepage
,
4398 .writepages
= btrfs_writepages
,
4399 .readpages
= btrfs_readpages
,
4400 .sync_page
= block_sync_page
,
4402 .direct_IO
= btrfs_direct_IO
,
4403 .invalidatepage
= btrfs_invalidatepage
,
4404 .releasepage
= btrfs_releasepage
,
4405 .set_page_dirty
= btrfs_set_page_dirty
,
4408 static struct address_space_operations btrfs_symlink_aops
= {
4409 .readpage
= btrfs_readpage
,
4410 .writepage
= btrfs_writepage
,
4411 .invalidatepage
= btrfs_invalidatepage
,
4412 .releasepage
= btrfs_releasepage
,
4415 static struct inode_operations btrfs_file_inode_operations
= {
4416 .truncate
= btrfs_truncate
,
4417 .getattr
= btrfs_getattr
,
4418 .setattr
= btrfs_setattr
,
4419 .setxattr
= btrfs_setxattr
,
4420 .getxattr
= btrfs_getxattr
,
4421 .listxattr
= btrfs_listxattr
,
4422 .removexattr
= btrfs_removexattr
,
4423 .permission
= btrfs_permission
,
4425 static struct inode_operations btrfs_special_inode_operations
= {
4426 .getattr
= btrfs_getattr
,
4427 .setattr
= btrfs_setattr
,
4428 .permission
= btrfs_permission
,
4429 .setxattr
= btrfs_setxattr
,
4430 .getxattr
= btrfs_getxattr
,
4431 .listxattr
= btrfs_listxattr
,
4432 .removexattr
= btrfs_removexattr
,
4434 static struct inode_operations btrfs_symlink_inode_operations
= {
4435 .readlink
= generic_readlink
,
4436 .follow_link
= page_follow_link_light
,
4437 .put_link
= page_put_link
,
4438 .permission
= btrfs_permission
,