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/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args
{
60 struct btrfs_root
*root
;
63 static const struct inode_operations btrfs_dir_inode_operations
;
64 static const struct inode_operations btrfs_symlink_inode_operations
;
65 static const struct inode_operations btrfs_dir_ro_inode_operations
;
66 static const struct inode_operations btrfs_special_inode_operations
;
67 static const struct inode_operations btrfs_file_inode_operations
;
68 static const struct address_space_operations btrfs_aops
;
69 static const struct address_space_operations btrfs_symlink_aops
;
70 static const struct file_operations btrfs_dir_file_operations
;
71 static struct extent_io_ops btrfs_extent_io_ops
;
73 static struct kmem_cache
*btrfs_inode_cachep
;
74 struct kmem_cache
*btrfs_trans_handle_cachep
;
75 struct kmem_cache
*btrfs_transaction_cachep
;
76 struct kmem_cache
*btrfs_path_cachep
;
77 struct kmem_cache
*btrfs_free_space_cachep
;
80 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
81 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
82 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
83 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
84 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
85 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
86 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
87 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
90 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
);
91 static int btrfs_truncate(struct inode
*inode
);
92 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
93 static noinline
int cow_file_range(struct inode
*inode
,
94 struct page
*locked_page
,
95 u64 start
, u64 end
, int *page_started
,
96 unsigned long *nr_written
, int unlock
);
97 static noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
98 struct btrfs_root
*root
, struct inode
*inode
);
100 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
101 struct inode
*inode
, struct inode
*dir
,
102 const struct qstr
*qstr
)
106 err
= btrfs_init_acl(trans
, inode
, dir
);
108 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
117 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
118 struct btrfs_root
*root
, struct inode
*inode
,
119 u64 start
, size_t size
, size_t compressed_size
,
121 struct page
**compressed_pages
)
123 struct btrfs_key key
;
124 struct btrfs_path
*path
;
125 struct extent_buffer
*leaf
;
126 struct page
*page
= NULL
;
129 struct btrfs_file_extent_item
*ei
;
132 size_t cur_size
= size
;
134 unsigned long offset
;
136 if (compressed_size
&& compressed_pages
)
137 cur_size
= compressed_size
;
139 path
= btrfs_alloc_path();
143 path
->leave_spinning
= 1;
145 key
.objectid
= btrfs_ino(inode
);
147 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
148 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
150 inode_add_bytes(inode
, size
);
151 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
158 leaf
= path
->nodes
[0];
159 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
160 struct btrfs_file_extent_item
);
161 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
162 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
163 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
164 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
165 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
166 ptr
= btrfs_file_extent_inline_start(ei
);
168 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
171 while (compressed_size
> 0) {
172 cpage
= compressed_pages
[i
];
173 cur_size
= min_t(unsigned long, compressed_size
,
176 kaddr
= kmap_atomic(cpage
, KM_USER0
);
177 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
178 kunmap_atomic(kaddr
, KM_USER0
);
182 compressed_size
-= cur_size
;
184 btrfs_set_file_extent_compression(leaf
, ei
,
187 page
= find_get_page(inode
->i_mapping
,
188 start
>> PAGE_CACHE_SHIFT
);
189 btrfs_set_file_extent_compression(leaf
, ei
, 0);
190 kaddr
= kmap_atomic(page
, KM_USER0
);
191 offset
= start
& (PAGE_CACHE_SIZE
- 1);
192 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
193 kunmap_atomic(kaddr
, KM_USER0
);
194 page_cache_release(page
);
196 btrfs_mark_buffer_dirty(leaf
);
197 btrfs_free_path(path
);
200 * we're an inline extent, so nobody can
201 * extend the file past i_size without locking
202 * a page we already have locked.
204 * We must do any isize and inode updates
205 * before we unlock the pages. Otherwise we
206 * could end up racing with unlink.
208 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
209 btrfs_update_inode(trans
, root
, inode
);
213 btrfs_free_path(path
);
219 * conditionally insert an inline extent into the file. This
220 * does the checks required to make sure the data is small enough
221 * to fit as an inline extent.
223 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
224 struct btrfs_root
*root
,
225 struct inode
*inode
, u64 start
, u64 end
,
226 size_t compressed_size
, int compress_type
,
227 struct page
**compressed_pages
)
229 u64 isize
= i_size_read(inode
);
230 u64 actual_end
= min(end
+ 1, isize
);
231 u64 inline_len
= actual_end
- start
;
232 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
233 ~((u64
)root
->sectorsize
- 1);
235 u64 data_len
= inline_len
;
239 data_len
= compressed_size
;
242 actual_end
>= PAGE_CACHE_SIZE
||
243 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
245 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
247 data_len
> root
->fs_info
->max_inline
) {
251 ret
= btrfs_drop_extents(trans
, inode
, start
, aligned_end
,
255 if (isize
> actual_end
)
256 inline_len
= min_t(u64
, isize
, actual_end
);
257 ret
= insert_inline_extent(trans
, root
, inode
, start
,
258 inline_len
, compressed_size
,
259 compress_type
, compressed_pages
);
261 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
262 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
266 struct async_extent
{
271 unsigned long nr_pages
;
273 struct list_head list
;
278 struct btrfs_root
*root
;
279 struct page
*locked_page
;
282 struct list_head extents
;
283 struct btrfs_work work
;
286 static noinline
int add_async_extent(struct async_cow
*cow
,
287 u64 start
, u64 ram_size
,
290 unsigned long nr_pages
,
293 struct async_extent
*async_extent
;
295 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
296 BUG_ON(!async_extent
);
297 async_extent
->start
= start
;
298 async_extent
->ram_size
= ram_size
;
299 async_extent
->compressed_size
= compressed_size
;
300 async_extent
->pages
= pages
;
301 async_extent
->nr_pages
= nr_pages
;
302 async_extent
->compress_type
= compress_type
;
303 list_add_tail(&async_extent
->list
, &cow
->extents
);
308 * we create compressed extents in two phases. The first
309 * phase compresses a range of pages that have already been
310 * locked (both pages and state bits are locked).
312 * This is done inside an ordered work queue, and the compression
313 * is spread across many cpus. The actual IO submission is step
314 * two, and the ordered work queue takes care of making sure that
315 * happens in the same order things were put onto the queue by
316 * writepages and friends.
318 * If this code finds it can't get good compression, it puts an
319 * entry onto the work queue to write the uncompressed bytes. This
320 * makes sure that both compressed inodes and uncompressed inodes
321 * are written in the same order that pdflush sent them down.
323 static noinline
int compress_file_range(struct inode
*inode
,
324 struct page
*locked_page
,
326 struct async_cow
*async_cow
,
329 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
330 struct btrfs_trans_handle
*trans
;
332 u64 blocksize
= root
->sectorsize
;
334 u64 isize
= i_size_read(inode
);
336 struct page
**pages
= NULL
;
337 unsigned long nr_pages
;
338 unsigned long nr_pages_ret
= 0;
339 unsigned long total_compressed
= 0;
340 unsigned long total_in
= 0;
341 unsigned long max_compressed
= 128 * 1024;
342 unsigned long max_uncompressed
= 128 * 1024;
345 int compress_type
= root
->fs_info
->compress_type
;
347 /* if this is a small write inside eof, kick off a defragbot */
348 if (end
<= BTRFS_I(inode
)->disk_i_size
&& (end
- start
+ 1) < 16 * 1024)
349 btrfs_add_inode_defrag(NULL
, inode
);
351 actual_end
= min_t(u64
, isize
, end
+ 1);
354 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
355 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
358 * we don't want to send crud past the end of i_size through
359 * compression, that's just a waste of CPU time. So, if the
360 * end of the file is before the start of our current
361 * requested range of bytes, we bail out to the uncompressed
362 * cleanup code that can deal with all of this.
364 * It isn't really the fastest way to fix things, but this is a
365 * very uncommon corner.
367 if (actual_end
<= start
)
368 goto cleanup_and_bail_uncompressed
;
370 total_compressed
= actual_end
- start
;
372 /* we want to make sure that amount of ram required to uncompress
373 * an extent is reasonable, so we limit the total size in ram
374 * of a compressed extent to 128k. This is a crucial number
375 * because it also controls how easily we can spread reads across
376 * cpus for decompression.
378 * We also want to make sure the amount of IO required to do
379 * a random read is reasonably small, so we limit the size of
380 * a compressed extent to 128k.
382 total_compressed
= min(total_compressed
, max_uncompressed
);
383 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
384 num_bytes
= max(blocksize
, num_bytes
);
389 * we do compression for mount -o compress and when the
390 * inode has not been flagged as nocompress. This flag can
391 * change at any time if we discover bad compression ratios.
393 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
394 (btrfs_test_opt(root
, COMPRESS
) ||
395 (BTRFS_I(inode
)->force_compress
) ||
396 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
398 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
400 /* just bail out to the uncompressed code */
404 if (BTRFS_I(inode
)->force_compress
)
405 compress_type
= BTRFS_I(inode
)->force_compress
;
407 ret
= btrfs_compress_pages(compress_type
,
408 inode
->i_mapping
, start
,
409 total_compressed
, pages
,
410 nr_pages
, &nr_pages_ret
,
416 unsigned long offset
= total_compressed
&
417 (PAGE_CACHE_SIZE
- 1);
418 struct page
*page
= pages
[nr_pages_ret
- 1];
421 /* zero the tail end of the last page, we might be
422 * sending it down to disk
425 kaddr
= kmap_atomic(page
, KM_USER0
);
426 memset(kaddr
+ offset
, 0,
427 PAGE_CACHE_SIZE
- offset
);
428 kunmap_atomic(kaddr
, KM_USER0
);
435 trans
= btrfs_join_transaction(root
);
436 BUG_ON(IS_ERR(trans
));
437 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
439 /* lets try to make an inline extent */
440 if (ret
|| total_in
< (actual_end
- start
)) {
441 /* we didn't compress the entire range, try
442 * to make an uncompressed inline extent.
444 ret
= cow_file_range_inline(trans
, root
, inode
,
445 start
, end
, 0, 0, NULL
);
447 /* try making a compressed inline extent */
448 ret
= cow_file_range_inline(trans
, root
, inode
,
451 compress_type
, pages
);
455 * inline extent creation worked, we don't need
456 * to create any more async work items. Unlock
457 * and free up our temp pages.
459 extent_clear_unlock_delalloc(inode
,
460 &BTRFS_I(inode
)->io_tree
,
462 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
463 EXTENT_CLEAR_DELALLOC
|
464 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
466 btrfs_end_transaction(trans
, root
);
469 btrfs_end_transaction(trans
, root
);
474 * we aren't doing an inline extent round the compressed size
475 * up to a block size boundary so the allocator does sane
478 total_compressed
= (total_compressed
+ blocksize
- 1) &
482 * one last check to make sure the compression is really a
483 * win, compare the page count read with the blocks on disk
485 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
486 ~(PAGE_CACHE_SIZE
- 1);
487 if (total_compressed
>= total_in
) {
490 num_bytes
= total_in
;
493 if (!will_compress
&& pages
) {
495 * the compression code ran but failed to make things smaller,
496 * free any pages it allocated and our page pointer array
498 for (i
= 0; i
< nr_pages_ret
; i
++) {
499 WARN_ON(pages
[i
]->mapping
);
500 page_cache_release(pages
[i
]);
504 total_compressed
= 0;
507 /* flag the file so we don't compress in the future */
508 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
509 !(BTRFS_I(inode
)->force_compress
)) {
510 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
516 /* the async work queues will take care of doing actual
517 * allocation on disk for these compressed pages,
518 * and will submit them to the elevator.
520 add_async_extent(async_cow
, start
, num_bytes
,
521 total_compressed
, pages
, nr_pages_ret
,
524 if (start
+ num_bytes
< end
) {
531 cleanup_and_bail_uncompressed
:
533 * No compression, but we still need to write the pages in
534 * the file we've been given so far. redirty the locked
535 * page if it corresponds to our extent and set things up
536 * for the async work queue to run cow_file_range to do
537 * the normal delalloc dance
539 if (page_offset(locked_page
) >= start
&&
540 page_offset(locked_page
) <= end
) {
541 __set_page_dirty_nobuffers(locked_page
);
542 /* unlocked later on in the async handlers */
544 add_async_extent(async_cow
, start
, end
- start
+ 1,
545 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
553 for (i
= 0; i
< nr_pages_ret
; i
++) {
554 WARN_ON(pages
[i
]->mapping
);
555 page_cache_release(pages
[i
]);
563 * phase two of compressed writeback. This is the ordered portion
564 * of the code, which only gets called in the order the work was
565 * queued. We walk all the async extents created by compress_file_range
566 * and send them down to the disk.
568 static noinline
int submit_compressed_extents(struct inode
*inode
,
569 struct async_cow
*async_cow
)
571 struct async_extent
*async_extent
;
573 struct btrfs_trans_handle
*trans
;
574 struct btrfs_key ins
;
575 struct extent_map
*em
;
576 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
577 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
578 struct extent_io_tree
*io_tree
;
581 if (list_empty(&async_cow
->extents
))
585 while (!list_empty(&async_cow
->extents
)) {
586 async_extent
= list_entry(async_cow
->extents
.next
,
587 struct async_extent
, list
);
588 list_del(&async_extent
->list
);
590 io_tree
= &BTRFS_I(inode
)->io_tree
;
593 /* did the compression code fall back to uncompressed IO? */
594 if (!async_extent
->pages
) {
595 int page_started
= 0;
596 unsigned long nr_written
= 0;
598 lock_extent(io_tree
, async_extent
->start
,
599 async_extent
->start
+
600 async_extent
->ram_size
- 1, GFP_NOFS
);
602 /* allocate blocks */
603 ret
= cow_file_range(inode
, async_cow
->locked_page
,
605 async_extent
->start
+
606 async_extent
->ram_size
- 1,
607 &page_started
, &nr_written
, 0);
610 * if page_started, cow_file_range inserted an
611 * inline extent and took care of all the unlocking
612 * and IO for us. Otherwise, we need to submit
613 * all those pages down to the drive.
615 if (!page_started
&& !ret
)
616 extent_write_locked_range(io_tree
,
617 inode
, async_extent
->start
,
618 async_extent
->start
+
619 async_extent
->ram_size
- 1,
627 lock_extent(io_tree
, async_extent
->start
,
628 async_extent
->start
+ async_extent
->ram_size
- 1,
631 trans
= btrfs_join_transaction(root
);
632 BUG_ON(IS_ERR(trans
));
633 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
634 ret
= btrfs_reserve_extent(trans
, root
,
635 async_extent
->compressed_size
,
636 async_extent
->compressed_size
,
639 btrfs_end_transaction(trans
, root
);
643 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
644 WARN_ON(async_extent
->pages
[i
]->mapping
);
645 page_cache_release(async_extent
->pages
[i
]);
647 kfree(async_extent
->pages
);
648 async_extent
->nr_pages
= 0;
649 async_extent
->pages
= NULL
;
650 unlock_extent(io_tree
, async_extent
->start
,
651 async_extent
->start
+
652 async_extent
->ram_size
- 1, GFP_NOFS
);
657 * here we're doing allocation and writeback of the
660 btrfs_drop_extent_cache(inode
, async_extent
->start
,
661 async_extent
->start
+
662 async_extent
->ram_size
- 1, 0);
664 em
= alloc_extent_map();
666 em
->start
= async_extent
->start
;
667 em
->len
= async_extent
->ram_size
;
668 em
->orig_start
= em
->start
;
670 em
->block_start
= ins
.objectid
;
671 em
->block_len
= ins
.offset
;
672 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
673 em
->compress_type
= async_extent
->compress_type
;
674 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
675 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
678 write_lock(&em_tree
->lock
);
679 ret
= add_extent_mapping(em_tree
, em
);
680 write_unlock(&em_tree
->lock
);
681 if (ret
!= -EEXIST
) {
685 btrfs_drop_extent_cache(inode
, async_extent
->start
,
686 async_extent
->start
+
687 async_extent
->ram_size
- 1, 0);
690 ret
= btrfs_add_ordered_extent_compress(inode
,
693 async_extent
->ram_size
,
695 BTRFS_ORDERED_COMPRESSED
,
696 async_extent
->compress_type
);
700 * clear dirty, set writeback and unlock the pages.
702 extent_clear_unlock_delalloc(inode
,
703 &BTRFS_I(inode
)->io_tree
,
705 async_extent
->start
+
706 async_extent
->ram_size
- 1,
707 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
708 EXTENT_CLEAR_UNLOCK
|
709 EXTENT_CLEAR_DELALLOC
|
710 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
712 ret
= btrfs_submit_compressed_write(inode
,
714 async_extent
->ram_size
,
716 ins
.offset
, async_extent
->pages
,
717 async_extent
->nr_pages
);
720 alloc_hint
= ins
.objectid
+ ins
.offset
;
728 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
731 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
732 struct extent_map
*em
;
735 read_lock(&em_tree
->lock
);
736 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
739 * if block start isn't an actual block number then find the
740 * first block in this inode and use that as a hint. If that
741 * block is also bogus then just don't worry about it.
743 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
745 em
= search_extent_mapping(em_tree
, 0, 0);
746 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
747 alloc_hint
= em
->block_start
;
751 alloc_hint
= em
->block_start
;
755 read_unlock(&em_tree
->lock
);
761 * when extent_io.c finds a delayed allocation range in the file,
762 * the call backs end up in this code. The basic idea is to
763 * allocate extents on disk for the range, and create ordered data structs
764 * in ram to track those extents.
766 * locked_page is the page that writepage had locked already. We use
767 * it to make sure we don't do extra locks or unlocks.
769 * *page_started is set to one if we unlock locked_page and do everything
770 * required to start IO on it. It may be clean and already done with
773 static noinline
int cow_file_range(struct inode
*inode
,
774 struct page
*locked_page
,
775 u64 start
, u64 end
, int *page_started
,
776 unsigned long *nr_written
,
779 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
780 struct btrfs_trans_handle
*trans
;
783 unsigned long ram_size
;
786 u64 blocksize
= root
->sectorsize
;
787 struct btrfs_key ins
;
788 struct extent_map
*em
;
789 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
792 BUG_ON(btrfs_is_free_space_inode(root
, inode
));
793 trans
= btrfs_join_transaction(root
);
794 BUG_ON(IS_ERR(trans
));
795 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
797 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
798 num_bytes
= max(blocksize
, num_bytes
);
799 disk_num_bytes
= num_bytes
;
802 /* if this is a small write inside eof, kick off defrag */
803 if (end
<= BTRFS_I(inode
)->disk_i_size
&& num_bytes
< 64 * 1024)
804 btrfs_add_inode_defrag(trans
, inode
);
807 /* lets try to make an inline extent */
808 ret
= cow_file_range_inline(trans
, root
, inode
,
809 start
, end
, 0, 0, NULL
);
811 extent_clear_unlock_delalloc(inode
,
812 &BTRFS_I(inode
)->io_tree
,
814 EXTENT_CLEAR_UNLOCK_PAGE
|
815 EXTENT_CLEAR_UNLOCK
|
816 EXTENT_CLEAR_DELALLOC
|
818 EXTENT_SET_WRITEBACK
|
819 EXTENT_END_WRITEBACK
);
821 *nr_written
= *nr_written
+
822 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
829 BUG_ON(disk_num_bytes
>
830 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
832 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
833 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
835 while (disk_num_bytes
> 0) {
838 cur_alloc_size
= disk_num_bytes
;
839 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
840 root
->sectorsize
, 0, alloc_hint
,
844 em
= alloc_extent_map();
847 em
->orig_start
= em
->start
;
848 ram_size
= ins
.offset
;
849 em
->len
= ins
.offset
;
851 em
->block_start
= ins
.objectid
;
852 em
->block_len
= ins
.offset
;
853 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
854 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
857 write_lock(&em_tree
->lock
);
858 ret
= add_extent_mapping(em_tree
, em
);
859 write_unlock(&em_tree
->lock
);
860 if (ret
!= -EEXIST
) {
864 btrfs_drop_extent_cache(inode
, start
,
865 start
+ ram_size
- 1, 0);
868 cur_alloc_size
= ins
.offset
;
869 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
870 ram_size
, cur_alloc_size
, 0);
873 if (root
->root_key
.objectid
==
874 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
875 ret
= btrfs_reloc_clone_csums(inode
, start
,
880 if (disk_num_bytes
< cur_alloc_size
)
883 /* we're not doing compressed IO, don't unlock the first
884 * page (which the caller expects to stay locked), don't
885 * clear any dirty bits and don't set any writeback bits
887 * Do set the Private2 bit so we know this page was properly
888 * setup for writepage
890 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
891 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
894 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
895 start
, start
+ ram_size
- 1,
897 disk_num_bytes
-= cur_alloc_size
;
898 num_bytes
-= cur_alloc_size
;
899 alloc_hint
= ins
.objectid
+ ins
.offset
;
900 start
+= cur_alloc_size
;
904 btrfs_end_transaction(trans
, root
);
910 * work queue call back to started compression on a file and pages
912 static noinline
void async_cow_start(struct btrfs_work
*work
)
914 struct async_cow
*async_cow
;
916 async_cow
= container_of(work
, struct async_cow
, work
);
918 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
919 async_cow
->start
, async_cow
->end
, async_cow
,
922 async_cow
->inode
= NULL
;
926 * work queue call back to submit previously compressed pages
928 static noinline
void async_cow_submit(struct btrfs_work
*work
)
930 struct async_cow
*async_cow
;
931 struct btrfs_root
*root
;
932 unsigned long nr_pages
;
934 async_cow
= container_of(work
, struct async_cow
, work
);
936 root
= async_cow
->root
;
937 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
940 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
942 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
944 waitqueue_active(&root
->fs_info
->async_submit_wait
))
945 wake_up(&root
->fs_info
->async_submit_wait
);
947 if (async_cow
->inode
)
948 submit_compressed_extents(async_cow
->inode
, async_cow
);
951 static noinline
void async_cow_free(struct btrfs_work
*work
)
953 struct async_cow
*async_cow
;
954 async_cow
= container_of(work
, struct async_cow
, work
);
958 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
959 u64 start
, u64 end
, int *page_started
,
960 unsigned long *nr_written
)
962 struct async_cow
*async_cow
;
963 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
964 unsigned long nr_pages
;
966 int limit
= 10 * 1024 * 1042;
968 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
969 1, 0, NULL
, GFP_NOFS
);
970 while (start
< end
) {
971 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
973 async_cow
->inode
= inode
;
974 async_cow
->root
= root
;
975 async_cow
->locked_page
= locked_page
;
976 async_cow
->start
= start
;
978 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
981 cur_end
= min(end
, start
+ 512 * 1024 - 1);
983 async_cow
->end
= cur_end
;
984 INIT_LIST_HEAD(&async_cow
->extents
);
986 async_cow
->work
.func
= async_cow_start
;
987 async_cow
->work
.ordered_func
= async_cow_submit
;
988 async_cow
->work
.ordered_free
= async_cow_free
;
989 async_cow
->work
.flags
= 0;
991 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
993 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
995 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
998 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
999 wait_event(root
->fs_info
->async_submit_wait
,
1000 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1004 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1005 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1006 wait_event(root
->fs_info
->async_submit_wait
,
1007 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1011 *nr_written
+= nr_pages
;
1012 start
= cur_end
+ 1;
1018 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1019 u64 bytenr
, u64 num_bytes
)
1022 struct btrfs_ordered_sum
*sums
;
1025 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1026 bytenr
+ num_bytes
- 1, &list
, 0);
1027 if (ret
== 0 && list_empty(&list
))
1030 while (!list_empty(&list
)) {
1031 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1032 list_del(&sums
->list
);
1039 * when nowcow writeback call back. This checks for snapshots or COW copies
1040 * of the extents that exist in the file, and COWs the file as required.
1042 * If no cow copies or snapshots exist, we write directly to the existing
1045 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1046 struct page
*locked_page
,
1047 u64 start
, u64 end
, int *page_started
, int force
,
1048 unsigned long *nr_written
)
1050 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1051 struct btrfs_trans_handle
*trans
;
1052 struct extent_buffer
*leaf
;
1053 struct btrfs_path
*path
;
1054 struct btrfs_file_extent_item
*fi
;
1055 struct btrfs_key found_key
;
1068 u64 ino
= btrfs_ino(inode
);
1070 path
= btrfs_alloc_path();
1074 nolock
= btrfs_is_free_space_inode(root
, inode
);
1077 trans
= btrfs_join_transaction_nolock(root
);
1079 trans
= btrfs_join_transaction(root
);
1081 BUG_ON(IS_ERR(trans
));
1082 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1084 cow_start
= (u64
)-1;
1087 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1090 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1091 leaf
= path
->nodes
[0];
1092 btrfs_item_key_to_cpu(leaf
, &found_key
,
1093 path
->slots
[0] - 1);
1094 if (found_key
.objectid
== ino
&&
1095 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1100 leaf
= path
->nodes
[0];
1101 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1102 ret
= btrfs_next_leaf(root
, path
);
1107 leaf
= path
->nodes
[0];
1113 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1115 if (found_key
.objectid
> ino
||
1116 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1117 found_key
.offset
> end
)
1120 if (found_key
.offset
> cur_offset
) {
1121 extent_end
= found_key
.offset
;
1126 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1127 struct btrfs_file_extent_item
);
1128 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1130 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1131 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1132 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1133 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1134 extent_end
= found_key
.offset
+
1135 btrfs_file_extent_num_bytes(leaf
, fi
);
1136 if (extent_end
<= start
) {
1140 if (disk_bytenr
== 0)
1142 if (btrfs_file_extent_compression(leaf
, fi
) ||
1143 btrfs_file_extent_encryption(leaf
, fi
) ||
1144 btrfs_file_extent_other_encoding(leaf
, fi
))
1146 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1148 if (btrfs_extent_readonly(root
, disk_bytenr
))
1150 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1152 extent_offset
, disk_bytenr
))
1154 disk_bytenr
+= extent_offset
;
1155 disk_bytenr
+= cur_offset
- found_key
.offset
;
1156 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1158 * force cow if csum exists in the range.
1159 * this ensure that csum for a given extent are
1160 * either valid or do not exist.
1162 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1165 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1166 extent_end
= found_key
.offset
+
1167 btrfs_file_extent_inline_len(leaf
, fi
);
1168 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1173 if (extent_end
<= start
) {
1178 if (cow_start
== (u64
)-1)
1179 cow_start
= cur_offset
;
1180 cur_offset
= extent_end
;
1181 if (cur_offset
> end
)
1187 btrfs_release_path(path
);
1188 if (cow_start
!= (u64
)-1) {
1189 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1190 found_key
.offset
- 1, page_started
,
1193 cow_start
= (u64
)-1;
1196 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1197 struct extent_map
*em
;
1198 struct extent_map_tree
*em_tree
;
1199 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1200 em
= alloc_extent_map();
1202 em
->start
= cur_offset
;
1203 em
->orig_start
= em
->start
;
1204 em
->len
= num_bytes
;
1205 em
->block_len
= num_bytes
;
1206 em
->block_start
= disk_bytenr
;
1207 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1208 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1210 write_lock(&em_tree
->lock
);
1211 ret
= add_extent_mapping(em_tree
, em
);
1212 write_unlock(&em_tree
->lock
);
1213 if (ret
!= -EEXIST
) {
1214 free_extent_map(em
);
1217 btrfs_drop_extent_cache(inode
, em
->start
,
1218 em
->start
+ em
->len
- 1, 0);
1220 type
= BTRFS_ORDERED_PREALLOC
;
1222 type
= BTRFS_ORDERED_NOCOW
;
1225 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1226 num_bytes
, num_bytes
, type
);
1229 if (root
->root_key
.objectid
==
1230 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1231 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1236 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1237 cur_offset
, cur_offset
+ num_bytes
- 1,
1238 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1239 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1240 EXTENT_SET_PRIVATE2
);
1241 cur_offset
= extent_end
;
1242 if (cur_offset
> end
)
1245 btrfs_release_path(path
);
1247 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1248 cow_start
= cur_offset
;
1249 if (cow_start
!= (u64
)-1) {
1250 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1251 page_started
, nr_written
, 1);
1256 ret
= btrfs_end_transaction_nolock(trans
, root
);
1259 ret
= btrfs_end_transaction(trans
, root
);
1262 btrfs_free_path(path
);
1267 * extent_io.c call back to do delayed allocation processing
1269 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1270 u64 start
, u64 end
, int *page_started
,
1271 unsigned long *nr_written
)
1274 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1276 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1277 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1278 page_started
, 1, nr_written
);
1279 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1280 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1281 page_started
, 0, nr_written
);
1282 else if (!btrfs_test_opt(root
, COMPRESS
) &&
1283 !(BTRFS_I(inode
)->force_compress
) &&
1284 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))
1285 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1286 page_started
, nr_written
, 1);
1288 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1289 page_started
, nr_written
);
1293 static void btrfs_split_extent_hook(struct inode
*inode
,
1294 struct extent_state
*orig
, u64 split
)
1296 /* not delalloc, ignore it */
1297 if (!(orig
->state
& EXTENT_DELALLOC
))
1300 spin_lock(&BTRFS_I(inode
)->lock
);
1301 BTRFS_I(inode
)->outstanding_extents
++;
1302 spin_unlock(&BTRFS_I(inode
)->lock
);
1306 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307 * extents so we can keep track of new extents that are just merged onto old
1308 * extents, such as when we are doing sequential writes, so we can properly
1309 * account for the metadata space we'll need.
1311 static void btrfs_merge_extent_hook(struct inode
*inode
,
1312 struct extent_state
*new,
1313 struct extent_state
*other
)
1315 /* not delalloc, ignore it */
1316 if (!(other
->state
& EXTENT_DELALLOC
))
1319 spin_lock(&BTRFS_I(inode
)->lock
);
1320 BTRFS_I(inode
)->outstanding_extents
--;
1321 spin_unlock(&BTRFS_I(inode
)->lock
);
1325 * extent_io.c set_bit_hook, used to track delayed allocation
1326 * bytes in this file, and to maintain the list of inodes that
1327 * have pending delalloc work to be done.
1329 static void btrfs_set_bit_hook(struct inode
*inode
,
1330 struct extent_state
*state
, int *bits
)
1334 * set_bit and clear bit hooks normally require _irqsave/restore
1335 * but in this case, we are only testing for the DELALLOC
1336 * bit, which is only set or cleared with irqs on
1338 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1339 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1340 u64 len
= state
->end
+ 1 - state
->start
;
1341 bool do_list
= !btrfs_is_free_space_inode(root
, inode
);
1343 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1344 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1346 spin_lock(&BTRFS_I(inode
)->lock
);
1347 BTRFS_I(inode
)->outstanding_extents
++;
1348 spin_unlock(&BTRFS_I(inode
)->lock
);
1351 spin_lock(&root
->fs_info
->delalloc_lock
);
1352 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1353 root
->fs_info
->delalloc_bytes
+= len
;
1354 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1355 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1356 &root
->fs_info
->delalloc_inodes
);
1358 spin_unlock(&root
->fs_info
->delalloc_lock
);
1363 * extent_io.c clear_bit_hook, see set_bit_hook for why
1365 static void btrfs_clear_bit_hook(struct inode
*inode
,
1366 struct extent_state
*state
, int *bits
)
1369 * set_bit and clear bit hooks normally require _irqsave/restore
1370 * but in this case, we are only testing for the DELALLOC
1371 * bit, which is only set or cleared with irqs on
1373 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1374 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1375 u64 len
= state
->end
+ 1 - state
->start
;
1376 bool do_list
= !btrfs_is_free_space_inode(root
, inode
);
1378 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1379 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1380 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1381 spin_lock(&BTRFS_I(inode
)->lock
);
1382 BTRFS_I(inode
)->outstanding_extents
--;
1383 spin_unlock(&BTRFS_I(inode
)->lock
);
1386 if (*bits
& EXTENT_DO_ACCOUNTING
)
1387 btrfs_delalloc_release_metadata(inode
, len
);
1389 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1391 btrfs_free_reserved_data_space(inode
, len
);
1393 spin_lock(&root
->fs_info
->delalloc_lock
);
1394 root
->fs_info
->delalloc_bytes
-= len
;
1395 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1397 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1398 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1399 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1401 spin_unlock(&root
->fs_info
->delalloc_lock
);
1406 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1407 * we don't create bios that span stripes or chunks
1409 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1410 size_t size
, struct bio
*bio
,
1411 unsigned long bio_flags
)
1413 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1414 struct btrfs_mapping_tree
*map_tree
;
1415 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1420 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1423 length
= bio
->bi_size
;
1424 map_tree
= &root
->fs_info
->mapping_tree
;
1425 map_length
= length
;
1426 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1427 &map_length
, NULL
, 0);
1429 if (map_length
< length
+ size
)
1435 * in order to insert checksums into the metadata in large chunks,
1436 * we wait until bio submission time. All the pages in the bio are
1437 * checksummed and sums are attached onto the ordered extent record.
1439 * At IO completion time the cums attached on the ordered extent record
1440 * are inserted into the btree
1442 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1443 struct bio
*bio
, int mirror_num
,
1444 unsigned long bio_flags
,
1447 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1450 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1456 * in order to insert checksums into the metadata in large chunks,
1457 * we wait until bio submission time. All the pages in the bio are
1458 * checksummed and sums are attached onto the ordered extent record.
1460 * At IO completion time the cums attached on the ordered extent record
1461 * are inserted into the btree
1463 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1464 int mirror_num
, unsigned long bio_flags
,
1467 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1468 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1472 * extent_io.c submission hook. This does the right thing for csum calculation
1473 * on write, or reading the csums from the tree before a read
1475 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1476 int mirror_num
, unsigned long bio_flags
,
1479 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1484 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1486 if (btrfs_is_free_space_inode(root
, inode
))
1489 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1492 if (!(rw
& REQ_WRITE
)) {
1493 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1494 return btrfs_submit_compressed_read(inode
, bio
,
1495 mirror_num
, bio_flags
);
1496 } else if (!skip_sum
) {
1497 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1502 } else if (!skip_sum
) {
1503 /* csum items have already been cloned */
1504 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1506 /* we're doing a write, do the async checksumming */
1507 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1508 inode
, rw
, bio
, mirror_num
,
1509 bio_flags
, bio_offset
,
1510 __btrfs_submit_bio_start
,
1511 __btrfs_submit_bio_done
);
1515 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1519 * given a list of ordered sums record them in the inode. This happens
1520 * at IO completion time based on sums calculated at bio submission time.
1522 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1523 struct inode
*inode
, u64 file_offset
,
1524 struct list_head
*list
)
1526 struct btrfs_ordered_sum
*sum
;
1528 list_for_each_entry(sum
, list
, list
) {
1529 btrfs_csum_file_blocks(trans
,
1530 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1535 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1536 struct extent_state
**cached_state
)
1538 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1540 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1541 cached_state
, GFP_NOFS
);
1544 /* see btrfs_writepage_start_hook for details on why this is required */
1545 struct btrfs_writepage_fixup
{
1547 struct btrfs_work work
;
1550 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1552 struct btrfs_writepage_fixup
*fixup
;
1553 struct btrfs_ordered_extent
*ordered
;
1554 struct extent_state
*cached_state
= NULL
;
1556 struct inode
*inode
;
1561 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1565 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1566 ClearPageChecked(page
);
1570 inode
= page
->mapping
->host
;
1571 page_start
= page_offset(page
);
1572 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1574 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1575 &cached_state
, GFP_NOFS
);
1577 /* already ordered? We're done */
1578 if (PagePrivate2(page
))
1581 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1583 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1584 page_end
, &cached_state
, GFP_NOFS
);
1586 btrfs_start_ordered_extent(inode
, ordered
, 1);
1587 btrfs_put_ordered_extent(ordered
);
1591 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1593 mapping_set_error(page
->mapping
, ret
);
1594 end_extent_writepage(page
, ret
, page_start
, page_end
);
1595 ClearPageChecked(page
);
1599 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1600 ClearPageChecked(page
);
1601 set_page_dirty(page
);
1603 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1604 &cached_state
, GFP_NOFS
);
1607 page_cache_release(page
);
1612 * There are a few paths in the higher layers of the kernel that directly
1613 * set the page dirty bit without asking the filesystem if it is a
1614 * good idea. This causes problems because we want to make sure COW
1615 * properly happens and the data=ordered rules are followed.
1617 * In our case any range that doesn't have the ORDERED bit set
1618 * hasn't been properly setup for IO. We kick off an async process
1619 * to fix it up. The async helper will wait for ordered extents, set
1620 * the delalloc bit and make it safe to write the page.
1622 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1624 struct inode
*inode
= page
->mapping
->host
;
1625 struct btrfs_writepage_fixup
*fixup
;
1626 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1628 /* this page is properly in the ordered list */
1629 if (TestClearPagePrivate2(page
))
1632 if (PageChecked(page
))
1635 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1639 SetPageChecked(page
);
1640 page_cache_get(page
);
1641 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1643 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1647 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1648 struct inode
*inode
, u64 file_pos
,
1649 u64 disk_bytenr
, u64 disk_num_bytes
,
1650 u64 num_bytes
, u64 ram_bytes
,
1651 u8 compression
, u8 encryption
,
1652 u16 other_encoding
, int extent_type
)
1654 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1655 struct btrfs_file_extent_item
*fi
;
1656 struct btrfs_path
*path
;
1657 struct extent_buffer
*leaf
;
1658 struct btrfs_key ins
;
1662 path
= btrfs_alloc_path();
1666 path
->leave_spinning
= 1;
1669 * we may be replacing one extent in the tree with another.
1670 * The new extent is pinned in the extent map, and we don't want
1671 * to drop it from the cache until it is completely in the btree.
1673 * So, tell btrfs_drop_extents to leave this extent in the cache.
1674 * the caller is expected to unpin it and allow it to be merged
1677 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1681 ins
.objectid
= btrfs_ino(inode
);
1682 ins
.offset
= file_pos
;
1683 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1684 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1686 leaf
= path
->nodes
[0];
1687 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1688 struct btrfs_file_extent_item
);
1689 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1690 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1691 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1692 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1693 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1694 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1695 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1696 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1697 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1698 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1700 btrfs_unlock_up_safe(path
, 1);
1701 btrfs_set_lock_blocking(leaf
);
1703 btrfs_mark_buffer_dirty(leaf
);
1705 inode_add_bytes(inode
, num_bytes
);
1707 ins
.objectid
= disk_bytenr
;
1708 ins
.offset
= disk_num_bytes
;
1709 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1710 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1711 root
->root_key
.objectid
,
1712 btrfs_ino(inode
), file_pos
, &ins
);
1714 btrfs_free_path(path
);
1720 * helper function for btrfs_finish_ordered_io, this
1721 * just reads in some of the csum leaves to prime them into ram
1722 * before we start the transaction. It limits the amount of btree
1723 * reads required while inside the transaction.
1725 /* as ordered data IO finishes, this gets called so we can finish
1726 * an ordered extent if the range of bytes in the file it covers are
1729 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1731 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1732 struct btrfs_trans_handle
*trans
= NULL
;
1733 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1734 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1735 struct extent_state
*cached_state
= NULL
;
1736 int compress_type
= 0;
1740 ret
= btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1744 BUG_ON(!ordered_extent
);
1746 nolock
= btrfs_is_free_space_inode(root
, inode
);
1748 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1749 BUG_ON(!list_empty(&ordered_extent
->list
));
1750 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1753 trans
= btrfs_join_transaction_nolock(root
);
1755 trans
= btrfs_join_transaction(root
);
1756 BUG_ON(IS_ERR(trans
));
1757 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1758 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
1764 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1765 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1766 0, &cached_state
, GFP_NOFS
);
1769 trans
= btrfs_join_transaction_nolock(root
);
1771 trans
= btrfs_join_transaction(root
);
1772 BUG_ON(IS_ERR(trans
));
1773 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1775 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1776 compress_type
= ordered_extent
->compress_type
;
1777 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1778 BUG_ON(compress_type
);
1779 ret
= btrfs_mark_extent_written(trans
, inode
,
1780 ordered_extent
->file_offset
,
1781 ordered_extent
->file_offset
+
1782 ordered_extent
->len
);
1785 BUG_ON(root
== root
->fs_info
->tree_root
);
1786 ret
= insert_reserved_file_extent(trans
, inode
,
1787 ordered_extent
->file_offset
,
1788 ordered_extent
->start
,
1789 ordered_extent
->disk_len
,
1790 ordered_extent
->len
,
1791 ordered_extent
->len
,
1792 compress_type
, 0, 0,
1793 BTRFS_FILE_EXTENT_REG
);
1794 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1795 ordered_extent
->file_offset
,
1796 ordered_extent
->len
);
1799 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1800 ordered_extent
->file_offset
+
1801 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1803 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1804 &ordered_extent
->list
);
1806 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1807 if (!ret
|| !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1808 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
1813 if (root
!= root
->fs_info
->tree_root
)
1814 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1817 btrfs_end_transaction_nolock(trans
, root
);
1819 btrfs_end_transaction(trans
, root
);
1823 btrfs_put_ordered_extent(ordered_extent
);
1824 /* once for the tree */
1825 btrfs_put_ordered_extent(ordered_extent
);
1830 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1831 struct extent_state
*state
, int uptodate
)
1833 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1835 ClearPagePrivate2(page
);
1836 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1840 * when reads are done, we need to check csums to verify the data is correct
1841 * if there's a match, we allow the bio to finish. If not, the code in
1842 * extent_io.c will try to find good copies for us.
1844 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1845 struct extent_state
*state
)
1847 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1848 struct inode
*inode
= page
->mapping
->host
;
1849 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1851 u64
private = ~(u32
)0;
1853 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1856 if (PageChecked(page
)) {
1857 ClearPageChecked(page
);
1861 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1864 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1865 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1866 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1871 if (state
&& state
->start
== start
) {
1872 private = state
->private;
1875 ret
= get_state_private(io_tree
, start
, &private);
1877 kaddr
= kmap_atomic(page
, KM_USER0
);
1881 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1882 btrfs_csum_final(csum
, (char *)&csum
);
1883 if (csum
!= private)
1886 kunmap_atomic(kaddr
, KM_USER0
);
1891 printk_ratelimited(KERN_INFO
"btrfs csum failed ino %llu off %llu csum %u "
1893 (unsigned long long)btrfs_ino(page
->mapping
->host
),
1894 (unsigned long long)start
, csum
,
1895 (unsigned long long)private);
1896 memset(kaddr
+ offset
, 1, end
- start
+ 1);
1897 flush_dcache_page(page
);
1898 kunmap_atomic(kaddr
, KM_USER0
);
1904 struct delayed_iput
{
1905 struct list_head list
;
1906 struct inode
*inode
;
1909 void btrfs_add_delayed_iput(struct inode
*inode
)
1911 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
1912 struct delayed_iput
*delayed
;
1914 if (atomic_add_unless(&inode
->i_count
, -1, 1))
1917 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
1918 delayed
->inode
= inode
;
1920 spin_lock(&fs_info
->delayed_iput_lock
);
1921 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
1922 spin_unlock(&fs_info
->delayed_iput_lock
);
1925 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
1928 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1929 struct delayed_iput
*delayed
;
1932 spin_lock(&fs_info
->delayed_iput_lock
);
1933 empty
= list_empty(&fs_info
->delayed_iputs
);
1934 spin_unlock(&fs_info
->delayed_iput_lock
);
1938 down_read(&root
->fs_info
->cleanup_work_sem
);
1939 spin_lock(&fs_info
->delayed_iput_lock
);
1940 list_splice_init(&fs_info
->delayed_iputs
, &list
);
1941 spin_unlock(&fs_info
->delayed_iput_lock
);
1943 while (!list_empty(&list
)) {
1944 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
1945 list_del(&delayed
->list
);
1946 iput(delayed
->inode
);
1949 up_read(&root
->fs_info
->cleanup_work_sem
);
1952 enum btrfs_orphan_cleanup_state
{
1953 ORPHAN_CLEANUP_STARTED
= 1,
1954 ORPHAN_CLEANUP_DONE
= 2,
1958 * This is called in transaction commit time. If there are no orphan
1959 * files in the subvolume, it removes orphan item and frees block_rsv
1962 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
1963 struct btrfs_root
*root
)
1965 struct btrfs_block_rsv
*block_rsv
;
1968 if (!list_empty(&root
->orphan_list
) ||
1969 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
1972 spin_lock(&root
->orphan_lock
);
1973 if (!list_empty(&root
->orphan_list
)) {
1974 spin_unlock(&root
->orphan_lock
);
1978 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
1979 spin_unlock(&root
->orphan_lock
);
1983 block_rsv
= root
->orphan_block_rsv
;
1984 root
->orphan_block_rsv
= NULL
;
1985 spin_unlock(&root
->orphan_lock
);
1987 if (root
->orphan_item_inserted
&&
1988 btrfs_root_refs(&root
->root_item
) > 0) {
1989 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
1990 root
->root_key
.objectid
);
1992 root
->orphan_item_inserted
= 0;
1996 WARN_ON(block_rsv
->size
> 0);
1997 btrfs_free_block_rsv(root
, block_rsv
);
2002 * This creates an orphan entry for the given inode in case something goes
2003 * wrong in the middle of an unlink/truncate.
2005 * NOTE: caller of this function should reserve 5 units of metadata for
2008 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2010 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2011 struct btrfs_block_rsv
*block_rsv
= NULL
;
2016 if (!root
->orphan_block_rsv
) {
2017 block_rsv
= btrfs_alloc_block_rsv(root
);
2022 spin_lock(&root
->orphan_lock
);
2023 if (!root
->orphan_block_rsv
) {
2024 root
->orphan_block_rsv
= block_rsv
;
2025 } else if (block_rsv
) {
2026 btrfs_free_block_rsv(root
, block_rsv
);
2030 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2031 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2034 * For proper ENOSPC handling, we should do orphan
2035 * cleanup when mounting. But this introduces backward
2036 * compatibility issue.
2038 if (!xchg(&root
->orphan_item_inserted
, 1))
2046 if (!BTRFS_I(inode
)->orphan_meta_reserved
) {
2047 BTRFS_I(inode
)->orphan_meta_reserved
= 1;
2050 spin_unlock(&root
->orphan_lock
);
2052 /* grab metadata reservation from transaction handle */
2054 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2058 /* insert an orphan item to track this unlinked/truncated file */
2060 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
2061 BUG_ON(ret
&& ret
!= -EEXIST
);
2064 /* insert an orphan item to track subvolume contains orphan files */
2066 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2067 root
->root_key
.objectid
);
2074 * We have done the truncate/delete so we can go ahead and remove the orphan
2075 * item for this particular inode.
2077 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2079 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2080 int delete_item
= 0;
2081 int release_rsv
= 0;
2084 spin_lock(&root
->orphan_lock
);
2085 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2086 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2090 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2091 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2094 spin_unlock(&root
->orphan_lock
);
2096 if (trans
&& delete_item
) {
2097 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
2102 btrfs_orphan_release_metadata(inode
);
2108 * this cleans up any orphans that may be left on the list from the last use
2111 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2113 struct btrfs_path
*path
;
2114 struct extent_buffer
*leaf
;
2115 struct btrfs_key key
, found_key
;
2116 struct btrfs_trans_handle
*trans
;
2117 struct inode
*inode
;
2118 u64 last_objectid
= 0;
2119 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2121 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2124 path
= btrfs_alloc_path();
2131 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2132 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2133 key
.offset
= (u64
)-1;
2136 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2141 * if ret == 0 means we found what we were searching for, which
2142 * is weird, but possible, so only screw with path if we didn't
2143 * find the key and see if we have stuff that matches
2147 if (path
->slots
[0] == 0)
2152 /* pull out the item */
2153 leaf
= path
->nodes
[0];
2154 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2156 /* make sure the item matches what we want */
2157 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2159 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2162 /* release the path since we're done with it */
2163 btrfs_release_path(path
);
2166 * this is where we are basically btrfs_lookup, without the
2167 * crossing root thing. we store the inode number in the
2168 * offset of the orphan item.
2171 if (found_key
.offset
== last_objectid
) {
2172 printk(KERN_ERR
"btrfs: Error removing orphan entry, "
2173 "stopping orphan cleanup\n");
2178 last_objectid
= found_key
.offset
;
2180 found_key
.objectid
= found_key
.offset
;
2181 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2182 found_key
.offset
= 0;
2183 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2184 ret
= PTR_RET(inode
);
2185 if (ret
&& ret
!= -ESTALE
)
2188 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
2189 struct btrfs_root
*dead_root
;
2190 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2191 int is_dead_root
= 0;
2194 * this is an orphan in the tree root. Currently these
2195 * could come from 2 sources:
2196 * a) a snapshot deletion in progress
2197 * b) a free space cache inode
2198 * We need to distinguish those two, as the snapshot
2199 * orphan must not get deleted.
2200 * find_dead_roots already ran before us, so if this
2201 * is a snapshot deletion, we should find the root
2202 * in the dead_roots list
2204 spin_lock(&fs_info
->trans_lock
);
2205 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
2207 if (dead_root
->root_key
.objectid
==
2208 found_key
.objectid
) {
2213 spin_unlock(&fs_info
->trans_lock
);
2215 /* prevent this orphan from being found again */
2216 key
.offset
= found_key
.objectid
- 1;
2221 * Inode is already gone but the orphan item is still there,
2222 * kill the orphan item.
2224 if (ret
== -ESTALE
) {
2225 trans
= btrfs_start_transaction(root
, 1);
2226 if (IS_ERR(trans
)) {
2227 ret
= PTR_ERR(trans
);
2230 ret
= btrfs_del_orphan_item(trans
, root
,
2231 found_key
.objectid
);
2233 btrfs_end_transaction(trans
, root
);
2238 * add this inode to the orphan list so btrfs_orphan_del does
2239 * the proper thing when we hit it
2241 spin_lock(&root
->orphan_lock
);
2242 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2243 spin_unlock(&root
->orphan_lock
);
2245 /* if we have links, this was a truncate, lets do that */
2246 if (inode
->i_nlink
) {
2247 if (!S_ISREG(inode
->i_mode
)) {
2253 ret
= btrfs_truncate(inode
);
2258 /* this will do delete_inode and everything for us */
2263 /* release the path since we're done with it */
2264 btrfs_release_path(path
);
2266 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2268 if (root
->orphan_block_rsv
)
2269 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2272 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2273 trans
= btrfs_join_transaction(root
);
2275 btrfs_end_transaction(trans
, root
);
2279 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2281 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2285 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2286 btrfs_free_path(path
);
2291 * very simple check to peek ahead in the leaf looking for xattrs. If we
2292 * don't find any xattrs, we know there can't be any acls.
2294 * slot is the slot the inode is in, objectid is the objectid of the inode
2296 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2297 int slot
, u64 objectid
)
2299 u32 nritems
= btrfs_header_nritems(leaf
);
2300 struct btrfs_key found_key
;
2304 while (slot
< nritems
) {
2305 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2307 /* we found a different objectid, there must not be acls */
2308 if (found_key
.objectid
!= objectid
)
2311 /* we found an xattr, assume we've got an acl */
2312 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2316 * we found a key greater than an xattr key, there can't
2317 * be any acls later on
2319 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2326 * it goes inode, inode backrefs, xattrs, extents,
2327 * so if there are a ton of hard links to an inode there can
2328 * be a lot of backrefs. Don't waste time searching too hard,
2329 * this is just an optimization
2334 /* we hit the end of the leaf before we found an xattr or
2335 * something larger than an xattr. We have to assume the inode
2342 * read an inode from the btree into the in-memory inode
2344 static void btrfs_read_locked_inode(struct inode
*inode
)
2346 struct btrfs_path
*path
;
2347 struct extent_buffer
*leaf
;
2348 struct btrfs_inode_item
*inode_item
;
2349 struct btrfs_timespec
*tspec
;
2350 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2351 struct btrfs_key location
;
2355 bool filled
= false;
2357 ret
= btrfs_fill_inode(inode
, &rdev
);
2361 path
= btrfs_alloc_path();
2365 path
->leave_spinning
= 1;
2366 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2368 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2372 leaf
= path
->nodes
[0];
2377 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2378 struct btrfs_inode_item
);
2379 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2380 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
2381 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2382 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2383 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2385 tspec
= btrfs_inode_atime(inode_item
);
2386 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2387 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2389 tspec
= btrfs_inode_mtime(inode_item
);
2390 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2391 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2393 tspec
= btrfs_inode_ctime(inode_item
);
2394 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2395 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2397 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2398 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2399 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2400 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2402 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2404 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2405 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2408 * try to precache a NULL acl entry for files that don't have
2409 * any xattrs or acls
2411 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
2414 cache_no_acl(inode
);
2416 btrfs_free_path(path
);
2418 switch (inode
->i_mode
& S_IFMT
) {
2420 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2421 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2422 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2423 inode
->i_fop
= &btrfs_file_operations
;
2424 inode
->i_op
= &btrfs_file_inode_operations
;
2427 inode
->i_fop
= &btrfs_dir_file_operations
;
2428 if (root
== root
->fs_info
->tree_root
)
2429 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2431 inode
->i_op
= &btrfs_dir_inode_operations
;
2434 inode
->i_op
= &btrfs_symlink_inode_operations
;
2435 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2436 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2439 inode
->i_op
= &btrfs_special_inode_operations
;
2440 init_special_inode(inode
, inode
->i_mode
, rdev
);
2444 btrfs_update_iflags(inode
);
2448 btrfs_free_path(path
);
2449 make_bad_inode(inode
);
2453 * given a leaf and an inode, copy the inode fields into the leaf
2455 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2456 struct extent_buffer
*leaf
,
2457 struct btrfs_inode_item
*item
,
2458 struct inode
*inode
)
2460 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2461 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2462 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2463 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2464 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2466 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2467 inode
->i_atime
.tv_sec
);
2468 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2469 inode
->i_atime
.tv_nsec
);
2471 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2472 inode
->i_mtime
.tv_sec
);
2473 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2474 inode
->i_mtime
.tv_nsec
);
2476 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2477 inode
->i_ctime
.tv_sec
);
2478 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2479 inode
->i_ctime
.tv_nsec
);
2481 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2482 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2483 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2484 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2485 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2486 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2487 btrfs_set_inode_block_group(leaf
, item
, 0);
2491 * copy everything in the in-memory inode into the btree.
2493 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
2494 struct btrfs_root
*root
, struct inode
*inode
)
2496 struct btrfs_inode_item
*inode_item
;
2497 struct btrfs_path
*path
;
2498 struct extent_buffer
*leaf
;
2501 path
= btrfs_alloc_path();
2505 path
->leave_spinning
= 1;
2506 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
2514 btrfs_unlock_up_safe(path
, 1);
2515 leaf
= path
->nodes
[0];
2516 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2517 struct btrfs_inode_item
);
2519 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2520 btrfs_mark_buffer_dirty(leaf
);
2521 btrfs_set_inode_last_trans(trans
, inode
);
2524 btrfs_free_path(path
);
2529 * copy everything in the in-memory inode into the btree.
2531 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2532 struct btrfs_root
*root
, struct inode
*inode
)
2537 * If the inode is a free space inode, we can deadlock during commit
2538 * if we put it into the delayed code.
2540 * The data relocation inode should also be directly updated
2543 if (!btrfs_is_free_space_inode(root
, inode
)
2544 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
2545 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
2547 btrfs_set_inode_last_trans(trans
, inode
);
2551 return btrfs_update_inode_item(trans
, root
, inode
);
2554 static noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
2555 struct btrfs_root
*root
, struct inode
*inode
)
2559 ret
= btrfs_update_inode(trans
, root
, inode
);
2561 return btrfs_update_inode_item(trans
, root
, inode
);
2566 * unlink helper that gets used here in inode.c and in the tree logging
2567 * recovery code. It remove a link in a directory with a given name, and
2568 * also drops the back refs in the inode to the directory
2570 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2571 struct btrfs_root
*root
,
2572 struct inode
*dir
, struct inode
*inode
,
2573 const char *name
, int name_len
)
2575 struct btrfs_path
*path
;
2577 struct extent_buffer
*leaf
;
2578 struct btrfs_dir_item
*di
;
2579 struct btrfs_key key
;
2581 u64 ino
= btrfs_ino(inode
);
2582 u64 dir_ino
= btrfs_ino(dir
);
2584 path
= btrfs_alloc_path();
2590 path
->leave_spinning
= 1;
2591 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2592 name
, name_len
, -1);
2601 leaf
= path
->nodes
[0];
2602 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2603 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2606 btrfs_release_path(path
);
2608 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
2611 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2612 "inode %llu parent %llu\n", name_len
, name
,
2613 (unsigned long long)ino
, (unsigned long long)dir_ino
);
2617 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2621 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2623 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2625 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2630 btrfs_free_path(path
);
2634 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2635 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2636 btrfs_update_inode(trans
, root
, dir
);
2641 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2642 struct btrfs_root
*root
,
2643 struct inode
*dir
, struct inode
*inode
,
2644 const char *name
, int name_len
)
2647 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2649 btrfs_drop_nlink(inode
);
2650 ret
= btrfs_update_inode(trans
, root
, inode
);
2656 /* helper to check if there is any shared block in the path */
2657 static int check_path_shared(struct btrfs_root
*root
,
2658 struct btrfs_path
*path
)
2660 struct extent_buffer
*eb
;
2664 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2667 if (!path
->nodes
[level
])
2669 eb
= path
->nodes
[level
];
2670 if (!btrfs_block_can_be_shared(root
, eb
))
2672 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2681 * helper to start transaction for unlink and rmdir.
2683 * unlink and rmdir are special in btrfs, they do not always free space.
2684 * so in enospc case, we should make sure they will free space before
2685 * allowing them to use the global metadata reservation.
2687 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2688 struct dentry
*dentry
)
2690 struct btrfs_trans_handle
*trans
;
2691 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2692 struct btrfs_path
*path
;
2693 struct btrfs_inode_ref
*ref
;
2694 struct btrfs_dir_item
*di
;
2695 struct inode
*inode
= dentry
->d_inode
;
2700 u64 ino
= btrfs_ino(inode
);
2701 u64 dir_ino
= btrfs_ino(dir
);
2704 * 1 for the possible orphan item
2705 * 1 for the dir item
2706 * 1 for the dir index
2707 * 1 for the inode ref
2708 * 1 for the inode ref in the tree log
2709 * 2 for the dir entries in the log
2712 trans
= btrfs_start_transaction(root
, 8);
2713 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2716 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2717 return ERR_PTR(-ENOSPC
);
2719 /* check if there is someone else holds reference */
2720 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2721 return ERR_PTR(-ENOSPC
);
2723 if (atomic_read(&inode
->i_count
) > 2)
2724 return ERR_PTR(-ENOSPC
);
2726 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2727 return ERR_PTR(-ENOSPC
);
2729 path
= btrfs_alloc_path();
2731 root
->fs_info
->enospc_unlink
= 0;
2732 return ERR_PTR(-ENOMEM
);
2735 /* 1 for the orphan item */
2736 trans
= btrfs_start_transaction(root
, 1);
2737 if (IS_ERR(trans
)) {
2738 btrfs_free_path(path
);
2739 root
->fs_info
->enospc_unlink
= 0;
2743 path
->skip_locking
= 1;
2744 path
->search_commit_root
= 1;
2746 ret
= btrfs_lookup_inode(trans
, root
, path
,
2747 &BTRFS_I(dir
)->location
, 0);
2753 if (check_path_shared(root
, path
))
2758 btrfs_release_path(path
);
2760 ret
= btrfs_lookup_inode(trans
, root
, path
,
2761 &BTRFS_I(inode
)->location
, 0);
2767 if (check_path_shared(root
, path
))
2772 btrfs_release_path(path
);
2774 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2775 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2782 if (check_path_shared(root
, path
))
2784 btrfs_release_path(path
);
2792 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2793 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2799 if (check_path_shared(root
, path
))
2805 btrfs_release_path(path
);
2807 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2808 dentry
->d_name
.name
, dentry
->d_name
.len
,
2815 if (check_path_shared(root
, path
))
2817 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2818 btrfs_release_path(path
);
2821 * This is a commit root search, if we can lookup inode item and other
2822 * relative items in the commit root, it means the transaction of
2823 * dir/file creation has been committed, and the dir index item that we
2824 * delay to insert has also been inserted into the commit root. So
2825 * we needn't worry about the delayed insertion of the dir index item
2828 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
2829 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2834 BUG_ON(ret
== -ENOENT
);
2835 if (check_path_shared(root
, path
))
2840 btrfs_free_path(path
);
2841 /* Migrate the orphan reservation over */
2843 err
= btrfs_block_rsv_migrate(trans
->block_rsv
,
2844 &root
->fs_info
->global_block_rsv
,
2845 trans
->bytes_reserved
);
2848 btrfs_end_transaction(trans
, root
);
2849 root
->fs_info
->enospc_unlink
= 0;
2850 return ERR_PTR(err
);
2853 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2857 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2858 struct btrfs_root
*root
)
2860 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2861 btrfs_block_rsv_release(root
, trans
->block_rsv
,
2862 trans
->bytes_reserved
);
2863 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
2864 BUG_ON(!root
->fs_info
->enospc_unlink
);
2865 root
->fs_info
->enospc_unlink
= 0;
2867 btrfs_end_transaction(trans
, root
);
2870 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2872 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2873 struct btrfs_trans_handle
*trans
;
2874 struct inode
*inode
= dentry
->d_inode
;
2876 unsigned long nr
= 0;
2878 trans
= __unlink_start_trans(dir
, dentry
);
2880 return PTR_ERR(trans
);
2882 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2884 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2885 dentry
->d_name
.name
, dentry
->d_name
.len
);
2889 if (inode
->i_nlink
== 0) {
2890 ret
= btrfs_orphan_add(trans
, inode
);
2896 nr
= trans
->blocks_used
;
2897 __unlink_end_trans(trans
, root
);
2898 btrfs_btree_balance_dirty(root
, nr
);
2902 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2903 struct btrfs_root
*root
,
2904 struct inode
*dir
, u64 objectid
,
2905 const char *name
, int name_len
)
2907 struct btrfs_path
*path
;
2908 struct extent_buffer
*leaf
;
2909 struct btrfs_dir_item
*di
;
2910 struct btrfs_key key
;
2913 u64 dir_ino
= btrfs_ino(dir
);
2915 path
= btrfs_alloc_path();
2919 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2920 name
, name_len
, -1);
2921 BUG_ON(IS_ERR_OR_NULL(di
));
2923 leaf
= path
->nodes
[0];
2924 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2925 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
2926 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2928 btrfs_release_path(path
);
2930 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
2931 objectid
, root
->root_key
.objectid
,
2932 dir_ino
, &index
, name
, name_len
);
2934 BUG_ON(ret
!= -ENOENT
);
2935 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
2937 BUG_ON(IS_ERR_OR_NULL(di
));
2939 leaf
= path
->nodes
[0];
2940 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2941 btrfs_release_path(path
);
2944 btrfs_release_path(path
);
2946 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2949 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2950 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2951 ret
= btrfs_update_inode(trans
, root
, dir
);
2954 btrfs_free_path(path
);
2958 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2960 struct inode
*inode
= dentry
->d_inode
;
2962 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2963 struct btrfs_trans_handle
*trans
;
2964 unsigned long nr
= 0;
2966 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
2967 btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
2970 trans
= __unlink_start_trans(dir
, dentry
);
2972 return PTR_ERR(trans
);
2974 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
2975 err
= btrfs_unlink_subvol(trans
, root
, dir
,
2976 BTRFS_I(inode
)->location
.objectid
,
2977 dentry
->d_name
.name
,
2978 dentry
->d_name
.len
);
2982 err
= btrfs_orphan_add(trans
, inode
);
2986 /* now the directory is empty */
2987 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2988 dentry
->d_name
.name
, dentry
->d_name
.len
);
2990 btrfs_i_size_write(inode
, 0);
2992 nr
= trans
->blocks_used
;
2993 __unlink_end_trans(trans
, root
);
2994 btrfs_btree_balance_dirty(root
, nr
);
3000 * this can truncate away extent items, csum items and directory items.
3001 * It starts at a high offset and removes keys until it can't find
3002 * any higher than new_size
3004 * csum items that cross the new i_size are truncated to the new size
3007 * min_type is the minimum key type to truncate down to. If set to 0, this
3008 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3010 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3011 struct btrfs_root
*root
,
3012 struct inode
*inode
,
3013 u64 new_size
, u32 min_type
)
3015 struct btrfs_path
*path
;
3016 struct extent_buffer
*leaf
;
3017 struct btrfs_file_extent_item
*fi
;
3018 struct btrfs_key key
;
3019 struct btrfs_key found_key
;
3020 u64 extent_start
= 0;
3021 u64 extent_num_bytes
= 0;
3022 u64 extent_offset
= 0;
3024 u64 mask
= root
->sectorsize
- 1;
3025 u32 found_type
= (u8
)-1;
3028 int pending_del_nr
= 0;
3029 int pending_del_slot
= 0;
3030 int extent_type
= -1;
3033 u64 ino
= btrfs_ino(inode
);
3035 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3037 path
= btrfs_alloc_path();
3042 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3043 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3046 * This function is also used to drop the items in the log tree before
3047 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3048 * it is used to drop the loged items. So we shouldn't kill the delayed
3051 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3052 btrfs_kill_delayed_inode_items(inode
);
3055 key
.offset
= (u64
)-1;
3059 path
->leave_spinning
= 1;
3060 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3067 /* there are no items in the tree for us to truncate, we're
3070 if (path
->slots
[0] == 0)
3077 leaf
= path
->nodes
[0];
3078 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3079 found_type
= btrfs_key_type(&found_key
);
3081 if (found_key
.objectid
!= ino
)
3084 if (found_type
< min_type
)
3087 item_end
= found_key
.offset
;
3088 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3089 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3090 struct btrfs_file_extent_item
);
3091 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3092 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3094 btrfs_file_extent_num_bytes(leaf
, fi
);
3095 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3096 item_end
+= btrfs_file_extent_inline_len(leaf
,
3101 if (found_type
> min_type
) {
3104 if (item_end
< new_size
)
3106 if (found_key
.offset
>= new_size
)
3112 /* FIXME, shrink the extent if the ref count is only 1 */
3113 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3116 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3118 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3120 u64 orig_num_bytes
=
3121 btrfs_file_extent_num_bytes(leaf
, fi
);
3122 extent_num_bytes
= new_size
-
3123 found_key
.offset
+ root
->sectorsize
- 1;
3124 extent_num_bytes
= extent_num_bytes
&
3125 ~((u64
)root
->sectorsize
- 1);
3126 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3128 num_dec
= (orig_num_bytes
-
3130 if (root
->ref_cows
&& extent_start
!= 0)
3131 inode_sub_bytes(inode
, num_dec
);
3132 btrfs_mark_buffer_dirty(leaf
);
3135 btrfs_file_extent_disk_num_bytes(leaf
,
3137 extent_offset
= found_key
.offset
-
3138 btrfs_file_extent_offset(leaf
, fi
);
3140 /* FIXME blocksize != 4096 */
3141 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3142 if (extent_start
!= 0) {
3145 inode_sub_bytes(inode
, num_dec
);
3148 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3150 * we can't truncate inline items that have had
3154 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3155 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3156 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3157 u32 size
= new_size
- found_key
.offset
;
3159 if (root
->ref_cows
) {
3160 inode_sub_bytes(inode
, item_end
+ 1 -
3164 btrfs_file_extent_calc_inline_size(size
);
3165 ret
= btrfs_truncate_item(trans
, root
, path
,
3167 } else if (root
->ref_cows
) {
3168 inode_sub_bytes(inode
, item_end
+ 1 -
3174 if (!pending_del_nr
) {
3175 /* no pending yet, add ourselves */
3176 pending_del_slot
= path
->slots
[0];
3178 } else if (pending_del_nr
&&
3179 path
->slots
[0] + 1 == pending_del_slot
) {
3180 /* hop on the pending chunk */
3182 pending_del_slot
= path
->slots
[0];
3189 if (found_extent
&& (root
->ref_cows
||
3190 root
== root
->fs_info
->tree_root
)) {
3191 btrfs_set_path_blocking(path
);
3192 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3193 extent_num_bytes
, 0,
3194 btrfs_header_owner(leaf
),
3195 ino
, extent_offset
, 0);
3199 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3202 if (path
->slots
[0] == 0 ||
3203 path
->slots
[0] != pending_del_slot
) {
3204 if (root
->ref_cows
&&
3205 BTRFS_I(inode
)->location
.objectid
!=
3206 BTRFS_FREE_INO_OBJECTID
) {
3210 if (pending_del_nr
) {
3211 ret
= btrfs_del_items(trans
, root
, path
,
3217 btrfs_release_path(path
);
3224 if (pending_del_nr
) {
3225 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3229 btrfs_free_path(path
);
3234 * taken from block_truncate_page, but does cow as it zeros out
3235 * any bytes left in the last page in the file.
3237 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3239 struct inode
*inode
= mapping
->host
;
3240 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3241 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3242 struct btrfs_ordered_extent
*ordered
;
3243 struct extent_state
*cached_state
= NULL
;
3245 u32 blocksize
= root
->sectorsize
;
3246 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3247 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3249 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
3254 if ((offset
& (blocksize
- 1)) == 0)
3256 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3262 page
= find_or_create_page(mapping
, index
, mask
);
3264 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3268 page_start
= page_offset(page
);
3269 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3271 if (!PageUptodate(page
)) {
3272 ret
= btrfs_readpage(NULL
, page
);
3274 if (page
->mapping
!= mapping
) {
3276 page_cache_release(page
);
3279 if (!PageUptodate(page
)) {
3284 wait_on_page_writeback(page
);
3286 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3288 set_page_extent_mapped(page
);
3290 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3292 unlock_extent_cached(io_tree
, page_start
, page_end
,
3293 &cached_state
, GFP_NOFS
);
3295 page_cache_release(page
);
3296 btrfs_start_ordered_extent(inode
, ordered
, 1);
3297 btrfs_put_ordered_extent(ordered
);
3301 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3302 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3303 0, 0, &cached_state
, GFP_NOFS
);
3305 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3308 unlock_extent_cached(io_tree
, page_start
, page_end
,
3309 &cached_state
, GFP_NOFS
);
3314 if (offset
!= PAGE_CACHE_SIZE
) {
3316 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3317 flush_dcache_page(page
);
3320 ClearPageChecked(page
);
3321 set_page_dirty(page
);
3322 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3327 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3329 page_cache_release(page
);
3335 * This function puts in dummy file extents for the area we're creating a hole
3336 * for. So if we are truncating this file to a larger size we need to insert
3337 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3338 * the range between oldsize and size
3340 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3342 struct btrfs_trans_handle
*trans
;
3343 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3344 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3345 struct extent_map
*em
= NULL
;
3346 struct extent_state
*cached_state
= NULL
;
3347 u64 mask
= root
->sectorsize
- 1;
3348 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3349 u64 block_end
= (size
+ mask
) & ~mask
;
3355 if (size
<= hole_start
)
3359 struct btrfs_ordered_extent
*ordered
;
3360 btrfs_wait_ordered_range(inode
, hole_start
,
3361 block_end
- hole_start
);
3362 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3363 &cached_state
, GFP_NOFS
);
3364 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3367 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3368 &cached_state
, GFP_NOFS
);
3369 btrfs_put_ordered_extent(ordered
);
3372 cur_offset
= hole_start
;
3374 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3375 block_end
- cur_offset
, 0);
3376 BUG_ON(IS_ERR_OR_NULL(em
));
3377 last_byte
= min(extent_map_end(em
), block_end
);
3378 last_byte
= (last_byte
+ mask
) & ~mask
;
3379 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3381 hole_size
= last_byte
- cur_offset
;
3383 trans
= btrfs_start_transaction(root
, 3);
3384 if (IS_ERR(trans
)) {
3385 err
= PTR_ERR(trans
);
3389 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3390 cur_offset
+ hole_size
,
3393 btrfs_update_inode(trans
, root
, inode
);
3394 btrfs_end_transaction(trans
, root
);
3398 err
= btrfs_insert_file_extent(trans
, root
,
3399 btrfs_ino(inode
), cur_offset
, 0,
3400 0, hole_size
, 0, hole_size
,
3403 btrfs_update_inode(trans
, root
, inode
);
3404 btrfs_end_transaction(trans
, root
);
3408 btrfs_drop_extent_cache(inode
, hole_start
,
3411 btrfs_update_inode(trans
, root
, inode
);
3412 btrfs_end_transaction(trans
, root
);
3414 free_extent_map(em
);
3416 cur_offset
= last_byte
;
3417 if (cur_offset
>= block_end
)
3421 free_extent_map(em
);
3422 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3427 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3429 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3430 struct btrfs_trans_handle
*trans
;
3431 loff_t oldsize
= i_size_read(inode
);
3434 if (newsize
== oldsize
)
3437 if (newsize
> oldsize
) {
3438 truncate_pagecache(inode
, oldsize
, newsize
);
3439 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3443 trans
= btrfs_start_transaction(root
, 1);
3445 return PTR_ERR(trans
);
3447 i_size_write(inode
, newsize
);
3448 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3449 ret
= btrfs_update_inode(trans
, root
, inode
);
3450 btrfs_end_transaction(trans
, root
);
3454 * We're truncating a file that used to have good data down to
3455 * zero. Make sure it gets into the ordered flush list so that
3456 * any new writes get down to disk quickly.
3459 BTRFS_I(inode
)->ordered_data_close
= 1;
3461 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3462 truncate_setsize(inode
, newsize
);
3463 ret
= btrfs_truncate(inode
);
3469 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3471 struct inode
*inode
= dentry
->d_inode
;
3472 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3475 if (btrfs_root_readonly(root
))
3478 err
= inode_change_ok(inode
, attr
);
3482 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3483 err
= btrfs_setsize(inode
, attr
->ia_size
);
3488 if (attr
->ia_valid
) {
3489 setattr_copy(inode
, attr
);
3490 err
= btrfs_dirty_inode(inode
);
3492 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
3493 err
= btrfs_acl_chmod(inode
);
3499 void btrfs_evict_inode(struct inode
*inode
)
3501 struct btrfs_trans_handle
*trans
;
3502 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3503 struct btrfs_block_rsv
*rsv
, *global_rsv
;
3504 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
3508 trace_btrfs_inode_evict(inode
);
3510 truncate_inode_pages(&inode
->i_data
, 0);
3511 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3512 btrfs_is_free_space_inode(root
, inode
)))
3515 if (is_bad_inode(inode
)) {
3516 btrfs_orphan_del(NULL
, inode
);
3519 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3520 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3522 if (root
->fs_info
->log_root_recovering
) {
3523 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3527 if (inode
->i_nlink
> 0) {
3528 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3532 rsv
= btrfs_alloc_block_rsv(root
);
3534 btrfs_orphan_del(NULL
, inode
);
3537 rsv
->size
= min_size
;
3538 global_rsv
= &root
->fs_info
->global_block_rsv
;
3540 btrfs_i_size_write(inode
, 0);
3543 * This is a bit simpler than btrfs_truncate since
3545 * 1) We've already reserved our space for our orphan item in the
3547 * 2) We're going to delete the inode item, so we don't need to update
3550 * So we just need to reserve some slack space in case we add bytes when
3551 * doing the truncate.
3554 ret
= btrfs_block_rsv_refill_noflush(root
, rsv
, min_size
);
3557 * Try and steal from the global reserve since we will
3558 * likely not use this space anyway, we want to try as
3559 * hard as possible to get this to work.
3562 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, min_size
);
3565 printk(KERN_WARNING
"Could not get space for a "
3566 "delete, will truncate on mount %d\n", ret
);
3567 btrfs_orphan_del(NULL
, inode
);
3568 btrfs_free_block_rsv(root
, rsv
);
3572 trans
= btrfs_start_transaction(root
, 0);
3573 if (IS_ERR(trans
)) {
3574 btrfs_orphan_del(NULL
, inode
);
3575 btrfs_free_block_rsv(root
, rsv
);
3579 trans
->block_rsv
= rsv
;
3581 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3585 nr
= trans
->blocks_used
;
3586 btrfs_end_transaction(trans
, root
);
3588 btrfs_btree_balance_dirty(root
, nr
);
3591 btrfs_free_block_rsv(root
, rsv
);
3594 trans
->block_rsv
= root
->orphan_block_rsv
;
3595 ret
= btrfs_orphan_del(trans
, inode
);
3599 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3600 if (!(root
== root
->fs_info
->tree_root
||
3601 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
3602 btrfs_return_ino(root
, btrfs_ino(inode
));
3604 nr
= trans
->blocks_used
;
3605 btrfs_end_transaction(trans
, root
);
3606 btrfs_btree_balance_dirty(root
, nr
);
3608 end_writeback(inode
);
3613 * this returns the key found in the dir entry in the location pointer.
3614 * If no dir entries were found, location->objectid is 0.
3616 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3617 struct btrfs_key
*location
)
3619 const char *name
= dentry
->d_name
.name
;
3620 int namelen
= dentry
->d_name
.len
;
3621 struct btrfs_dir_item
*di
;
3622 struct btrfs_path
*path
;
3623 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3626 path
= btrfs_alloc_path();
3630 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
3635 if (IS_ERR_OR_NULL(di
))
3638 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3640 btrfs_free_path(path
);
3643 location
->objectid
= 0;
3648 * when we hit a tree root in a directory, the btrfs part of the inode
3649 * needs to be changed to reflect the root directory of the tree root. This
3650 * is kind of like crossing a mount point.
3652 static int fixup_tree_root_location(struct btrfs_root
*root
,
3654 struct dentry
*dentry
,
3655 struct btrfs_key
*location
,
3656 struct btrfs_root
**sub_root
)
3658 struct btrfs_path
*path
;
3659 struct btrfs_root
*new_root
;
3660 struct btrfs_root_ref
*ref
;
3661 struct extent_buffer
*leaf
;
3665 path
= btrfs_alloc_path();
3672 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3673 BTRFS_I(dir
)->root
->root_key
.objectid
,
3674 location
->objectid
);
3681 leaf
= path
->nodes
[0];
3682 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3683 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
3684 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3687 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3688 (unsigned long)(ref
+ 1),
3689 dentry
->d_name
.len
);
3693 btrfs_release_path(path
);
3695 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3696 if (IS_ERR(new_root
)) {
3697 err
= PTR_ERR(new_root
);
3701 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3706 *sub_root
= new_root
;
3707 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3708 location
->type
= BTRFS_INODE_ITEM_KEY
;
3709 location
->offset
= 0;
3712 btrfs_free_path(path
);
3716 static void inode_tree_add(struct inode
*inode
)
3718 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3719 struct btrfs_inode
*entry
;
3721 struct rb_node
*parent
;
3722 u64 ino
= btrfs_ino(inode
);
3724 p
= &root
->inode_tree
.rb_node
;
3727 if (inode_unhashed(inode
))
3730 spin_lock(&root
->inode_lock
);
3733 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3735 if (ino
< btrfs_ino(&entry
->vfs_inode
))
3736 p
= &parent
->rb_left
;
3737 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
3738 p
= &parent
->rb_right
;
3740 WARN_ON(!(entry
->vfs_inode
.i_state
&
3741 (I_WILL_FREE
| I_FREEING
)));
3742 rb_erase(parent
, &root
->inode_tree
);
3743 RB_CLEAR_NODE(parent
);
3744 spin_unlock(&root
->inode_lock
);
3748 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3749 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3750 spin_unlock(&root
->inode_lock
);
3753 static void inode_tree_del(struct inode
*inode
)
3755 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3758 spin_lock(&root
->inode_lock
);
3759 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3760 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3761 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3762 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3764 spin_unlock(&root
->inode_lock
);
3767 * Free space cache has inodes in the tree root, but the tree root has a
3768 * root_refs of 0, so this could end up dropping the tree root as a
3769 * snapshot, so we need the extra !root->fs_info->tree_root check to
3770 * make sure we don't drop it.
3772 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3773 root
!= root
->fs_info
->tree_root
) {
3774 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3775 spin_lock(&root
->inode_lock
);
3776 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3777 spin_unlock(&root
->inode_lock
);
3779 btrfs_add_dead_root(root
);
3783 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3785 struct rb_node
*node
;
3786 struct rb_node
*prev
;
3787 struct btrfs_inode
*entry
;
3788 struct inode
*inode
;
3791 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3793 spin_lock(&root
->inode_lock
);
3795 node
= root
->inode_tree
.rb_node
;
3799 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3801 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
3802 node
= node
->rb_left
;
3803 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
3804 node
= node
->rb_right
;
3810 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
3811 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
3815 prev
= rb_next(prev
);
3819 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3820 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
3821 inode
= igrab(&entry
->vfs_inode
);
3823 spin_unlock(&root
->inode_lock
);
3824 if (atomic_read(&inode
->i_count
) > 1)
3825 d_prune_aliases(inode
);
3827 * btrfs_drop_inode will have it removed from
3828 * the inode cache when its usage count
3833 spin_lock(&root
->inode_lock
);
3837 if (cond_resched_lock(&root
->inode_lock
))
3840 node
= rb_next(node
);
3842 spin_unlock(&root
->inode_lock
);
3846 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
3848 struct btrfs_iget_args
*args
= p
;
3849 inode
->i_ino
= args
->ino
;
3850 BTRFS_I(inode
)->root
= args
->root
;
3851 btrfs_set_inode_space_info(args
->root
, inode
);
3855 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
3857 struct btrfs_iget_args
*args
= opaque
;
3858 return args
->ino
== btrfs_ino(inode
) &&
3859 args
->root
== BTRFS_I(inode
)->root
;
3862 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
3864 struct btrfs_root
*root
)
3866 struct inode
*inode
;
3867 struct btrfs_iget_args args
;
3868 args
.ino
= objectid
;
3871 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
3872 btrfs_init_locked_inode
,
3877 /* Get an inode object given its location and corresponding root.
3878 * Returns in *is_new if the inode was read from disk
3880 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
3881 struct btrfs_root
*root
, int *new)
3883 struct inode
*inode
;
3885 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
3887 return ERR_PTR(-ENOMEM
);
3889 if (inode
->i_state
& I_NEW
) {
3890 BTRFS_I(inode
)->root
= root
;
3891 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
3892 btrfs_read_locked_inode(inode
);
3893 if (!is_bad_inode(inode
)) {
3894 inode_tree_add(inode
);
3895 unlock_new_inode(inode
);
3899 unlock_new_inode(inode
);
3901 inode
= ERR_PTR(-ESTALE
);
3908 static struct inode
*new_simple_dir(struct super_block
*s
,
3909 struct btrfs_key
*key
,
3910 struct btrfs_root
*root
)
3912 struct inode
*inode
= new_inode(s
);
3915 return ERR_PTR(-ENOMEM
);
3917 BTRFS_I(inode
)->root
= root
;
3918 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
3919 BTRFS_I(inode
)->dummy_inode
= 1;
3921 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
3922 inode
->i_op
= &simple_dir_inode_operations
;
3923 inode
->i_fop
= &simple_dir_operations
;
3924 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
3925 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
3930 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
3932 struct inode
*inode
;
3933 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3934 struct btrfs_root
*sub_root
= root
;
3935 struct btrfs_key location
;
3939 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
3940 return ERR_PTR(-ENAMETOOLONG
);
3942 if (unlikely(d_need_lookup(dentry
))) {
3943 memcpy(&location
, dentry
->d_fsdata
, sizeof(struct btrfs_key
));
3944 kfree(dentry
->d_fsdata
);
3945 dentry
->d_fsdata
= NULL
;
3946 /* This thing is hashed, drop it for now */
3949 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
3953 return ERR_PTR(ret
);
3955 if (location
.objectid
== 0)
3958 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
3959 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
3963 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
3965 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
3966 ret
= fixup_tree_root_location(root
, dir
, dentry
,
3967 &location
, &sub_root
);
3970 inode
= ERR_PTR(ret
);
3972 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
3974 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
3976 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
3978 if (!IS_ERR(inode
) && root
!= sub_root
) {
3979 down_read(&root
->fs_info
->cleanup_work_sem
);
3980 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
3981 ret
= btrfs_orphan_cleanup(sub_root
);
3982 up_read(&root
->fs_info
->cleanup_work_sem
);
3984 inode
= ERR_PTR(ret
);
3990 static int btrfs_dentry_delete(const struct dentry
*dentry
)
3992 struct btrfs_root
*root
;
3994 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
3995 dentry
= dentry
->d_parent
;
3997 if (dentry
->d_inode
) {
3998 root
= BTRFS_I(dentry
->d_inode
)->root
;
3999 if (btrfs_root_refs(&root
->root_item
) == 0)
4005 static void btrfs_dentry_release(struct dentry
*dentry
)
4007 if (dentry
->d_fsdata
)
4008 kfree(dentry
->d_fsdata
);
4011 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4012 struct nameidata
*nd
)
4016 ret
= d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
4017 if (unlikely(d_need_lookup(dentry
))) {
4018 spin_lock(&dentry
->d_lock
);
4019 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
4020 spin_unlock(&dentry
->d_lock
);
4025 unsigned char btrfs_filetype_table
[] = {
4026 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4029 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4032 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4033 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4034 struct btrfs_item
*item
;
4035 struct btrfs_dir_item
*di
;
4036 struct btrfs_key key
;
4037 struct btrfs_key found_key
;
4038 struct btrfs_path
*path
;
4039 struct list_head ins_list
;
4040 struct list_head del_list
;
4043 struct extent_buffer
*leaf
;
4045 unsigned char d_type
;
4050 int key_type
= BTRFS_DIR_INDEX_KEY
;
4054 int is_curr
= 0; /* filp->f_pos points to the current index? */
4056 /* FIXME, use a real flag for deciding about the key type */
4057 if (root
->fs_info
->tree_root
== root
)
4058 key_type
= BTRFS_DIR_ITEM_KEY
;
4060 /* special case for "." */
4061 if (filp
->f_pos
== 0) {
4062 over
= filldir(dirent
, ".", 1,
4063 filp
->f_pos
, btrfs_ino(inode
), DT_DIR
);
4068 /* special case for .., just use the back ref */
4069 if (filp
->f_pos
== 1) {
4070 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4071 over
= filldir(dirent
, "..", 2,
4072 filp
->f_pos
, pino
, DT_DIR
);
4077 path
= btrfs_alloc_path();
4083 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4084 INIT_LIST_HEAD(&ins_list
);
4085 INIT_LIST_HEAD(&del_list
);
4086 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
4089 btrfs_set_key_type(&key
, key_type
);
4090 key
.offset
= filp
->f_pos
;
4091 key
.objectid
= btrfs_ino(inode
);
4093 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4098 leaf
= path
->nodes
[0];
4099 slot
= path
->slots
[0];
4100 if (slot
>= btrfs_header_nritems(leaf
)) {
4101 ret
= btrfs_next_leaf(root
, path
);
4109 item
= btrfs_item_nr(leaf
, slot
);
4110 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4112 if (found_key
.objectid
!= key
.objectid
)
4114 if (btrfs_key_type(&found_key
) != key_type
)
4116 if (found_key
.offset
< filp
->f_pos
)
4118 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
4119 btrfs_should_delete_dir_index(&del_list
,
4123 filp
->f_pos
= found_key
.offset
;
4126 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4128 di_total
= btrfs_item_size(leaf
, item
);
4130 while (di_cur
< di_total
) {
4131 struct btrfs_key location
;
4134 if (verify_dir_item(root
, leaf
, di
))
4137 name_len
= btrfs_dir_name_len(leaf
, di
);
4138 if (name_len
<= sizeof(tmp_name
)) {
4139 name_ptr
= tmp_name
;
4141 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4147 read_extent_buffer(leaf
, name_ptr
,
4148 (unsigned long)(di
+ 1), name_len
);
4150 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4151 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4155 q
.hash
= full_name_hash(q
.name
, q
.len
);
4156 tmp
= d_lookup(filp
->f_dentry
, &q
);
4158 struct btrfs_key
*newkey
;
4160 newkey
= kzalloc(sizeof(struct btrfs_key
),
4164 tmp
= d_alloc(filp
->f_dentry
, &q
);
4170 memcpy(newkey
, &location
,
4171 sizeof(struct btrfs_key
));
4172 tmp
->d_fsdata
= newkey
;
4173 tmp
->d_flags
|= DCACHE_NEED_LOOKUP
;
4180 /* is this a reference to our own snapshot? If so
4183 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4184 location
.objectid
== root
->root_key
.objectid
) {
4188 over
= filldir(dirent
, name_ptr
, name_len
,
4189 found_key
.offset
, location
.objectid
,
4193 if (name_ptr
!= tmp_name
)
4198 di_len
= btrfs_dir_name_len(leaf
, di
) +
4199 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4201 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4207 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4210 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
4216 /* Reached end of directory/root. Bump pos past the last item. */
4217 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4219 * 32-bit glibc will use getdents64, but then strtol -
4220 * so the last number we can serve is this.
4222 filp
->f_pos
= 0x7fffffff;
4228 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4229 btrfs_put_delayed_items(&ins_list
, &del_list
);
4230 btrfs_free_path(path
);
4234 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4236 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4237 struct btrfs_trans_handle
*trans
;
4239 bool nolock
= false;
4241 if (BTRFS_I(inode
)->dummy_inode
)
4244 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(root
, inode
))
4247 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4249 trans
= btrfs_join_transaction_nolock(root
);
4251 trans
= btrfs_join_transaction(root
);
4253 return PTR_ERR(trans
);
4255 ret
= btrfs_end_transaction_nolock(trans
, root
);
4257 ret
= btrfs_commit_transaction(trans
, root
);
4263 * This is somewhat expensive, updating the tree every time the
4264 * inode changes. But, it is most likely to find the inode in cache.
4265 * FIXME, needs more benchmarking...there are no reasons other than performance
4266 * to keep or drop this code.
4268 int btrfs_dirty_inode(struct inode
*inode
)
4270 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4271 struct btrfs_trans_handle
*trans
;
4274 if (BTRFS_I(inode
)->dummy_inode
)
4277 trans
= btrfs_join_transaction(root
);
4279 return PTR_ERR(trans
);
4281 ret
= btrfs_update_inode(trans
, root
, inode
);
4282 if (ret
&& ret
== -ENOSPC
) {
4283 /* whoops, lets try again with the full transaction */
4284 btrfs_end_transaction(trans
, root
);
4285 trans
= btrfs_start_transaction(root
, 1);
4287 return PTR_ERR(trans
);
4289 ret
= btrfs_update_inode(trans
, root
, inode
);
4291 btrfs_end_transaction(trans
, root
);
4292 if (BTRFS_I(inode
)->delayed_node
)
4293 btrfs_balance_delayed_items(root
);
4299 * This is a copy of file_update_time. We need this so we can return error on
4300 * ENOSPC for updating the inode in the case of file write and mmap writes.
4302 int btrfs_update_time(struct file
*file
)
4304 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4305 struct timespec now
;
4307 enum { S_MTIME
= 1, S_CTIME
= 2, S_VERSION
= 4 } sync_it
= 0;
4309 /* First try to exhaust all avenues to not sync */
4310 if (IS_NOCMTIME(inode
))
4313 now
= current_fs_time(inode
->i_sb
);
4314 if (!timespec_equal(&inode
->i_mtime
, &now
))
4317 if (!timespec_equal(&inode
->i_ctime
, &now
))
4320 if (IS_I_VERSION(inode
))
4321 sync_it
|= S_VERSION
;
4326 /* Finally allowed to write? Takes lock. */
4327 if (mnt_want_write_file(file
))
4330 /* Only change inode inside the lock region */
4331 if (sync_it
& S_VERSION
)
4332 inode_inc_iversion(inode
);
4333 if (sync_it
& S_CTIME
)
4334 inode
->i_ctime
= now
;
4335 if (sync_it
& S_MTIME
)
4336 inode
->i_mtime
= now
;
4337 ret
= btrfs_dirty_inode(inode
);
4339 mark_inode_dirty_sync(inode
);
4340 mnt_drop_write(file
->f_path
.mnt
);
4345 * find the highest existing sequence number in a directory
4346 * and then set the in-memory index_cnt variable to reflect
4347 * free sequence numbers
4349 static int btrfs_set_inode_index_count(struct inode
*inode
)
4351 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4352 struct btrfs_key key
, found_key
;
4353 struct btrfs_path
*path
;
4354 struct extent_buffer
*leaf
;
4357 key
.objectid
= btrfs_ino(inode
);
4358 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4359 key
.offset
= (u64
)-1;
4361 path
= btrfs_alloc_path();
4365 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4368 /* FIXME: we should be able to handle this */
4374 * MAGIC NUMBER EXPLANATION:
4375 * since we search a directory based on f_pos we have to start at 2
4376 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4377 * else has to start at 2
4379 if (path
->slots
[0] == 0) {
4380 BTRFS_I(inode
)->index_cnt
= 2;
4386 leaf
= path
->nodes
[0];
4387 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4389 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4390 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4391 BTRFS_I(inode
)->index_cnt
= 2;
4395 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4397 btrfs_free_path(path
);
4402 * helper to find a free sequence number in a given directory. This current
4403 * code is very simple, later versions will do smarter things in the btree
4405 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4409 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4410 ret
= btrfs_inode_delayed_dir_index_count(dir
);
4412 ret
= btrfs_set_inode_index_count(dir
);
4418 *index
= BTRFS_I(dir
)->index_cnt
;
4419 BTRFS_I(dir
)->index_cnt
++;
4424 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4425 struct btrfs_root
*root
,
4427 const char *name
, int name_len
,
4428 u64 ref_objectid
, u64 objectid
,
4429 umode_t mode
, u64
*index
)
4431 struct inode
*inode
;
4432 struct btrfs_inode_item
*inode_item
;
4433 struct btrfs_key
*location
;
4434 struct btrfs_path
*path
;
4435 struct btrfs_inode_ref
*ref
;
4436 struct btrfs_key key
[2];
4442 path
= btrfs_alloc_path();
4444 return ERR_PTR(-ENOMEM
);
4446 inode
= new_inode(root
->fs_info
->sb
);
4448 btrfs_free_path(path
);
4449 return ERR_PTR(-ENOMEM
);
4453 * we have to initialize this early, so we can reclaim the inode
4454 * number if we fail afterwards in this function.
4456 inode
->i_ino
= objectid
;
4459 trace_btrfs_inode_request(dir
);
4461 ret
= btrfs_set_inode_index(dir
, index
);
4463 btrfs_free_path(path
);
4465 return ERR_PTR(ret
);
4469 * index_cnt is ignored for everything but a dir,
4470 * btrfs_get_inode_index_count has an explanation for the magic
4473 BTRFS_I(inode
)->index_cnt
= 2;
4474 BTRFS_I(inode
)->root
= root
;
4475 BTRFS_I(inode
)->generation
= trans
->transid
;
4476 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4477 btrfs_set_inode_space_info(root
, inode
);
4484 key
[0].objectid
= objectid
;
4485 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4488 key
[1].objectid
= objectid
;
4489 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4490 key
[1].offset
= ref_objectid
;
4492 sizes
[0] = sizeof(struct btrfs_inode_item
);
4493 sizes
[1] = name_len
+ sizeof(*ref
);
4495 path
->leave_spinning
= 1;
4496 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4500 inode_init_owner(inode
, dir
, mode
);
4501 inode_set_bytes(inode
, 0);
4502 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4503 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4504 struct btrfs_inode_item
);
4505 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4507 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4508 struct btrfs_inode_ref
);
4509 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4510 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4511 ptr
= (unsigned long)(ref
+ 1);
4512 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4514 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4515 btrfs_free_path(path
);
4517 location
= &BTRFS_I(inode
)->location
;
4518 location
->objectid
= objectid
;
4519 location
->offset
= 0;
4520 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4522 btrfs_inherit_iflags(inode
, dir
);
4524 if (S_ISREG(mode
)) {
4525 if (btrfs_test_opt(root
, NODATASUM
))
4526 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4527 if (btrfs_test_opt(root
, NODATACOW
) ||
4528 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4529 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4532 insert_inode_hash(inode
);
4533 inode_tree_add(inode
);
4535 trace_btrfs_inode_new(inode
);
4536 btrfs_set_inode_last_trans(trans
, inode
);
4541 BTRFS_I(dir
)->index_cnt
--;
4542 btrfs_free_path(path
);
4544 return ERR_PTR(ret
);
4547 static inline u8
btrfs_inode_type(struct inode
*inode
)
4549 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4553 * utility function to add 'inode' into 'parent_inode' with
4554 * a give name and a given sequence number.
4555 * if 'add_backref' is true, also insert a backref from the
4556 * inode to the parent directory.
4558 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4559 struct inode
*parent_inode
, struct inode
*inode
,
4560 const char *name
, int name_len
, int add_backref
, u64 index
)
4563 struct btrfs_key key
;
4564 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4565 u64 ino
= btrfs_ino(inode
);
4566 u64 parent_ino
= btrfs_ino(parent_inode
);
4568 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4569 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4572 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4576 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4577 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4578 key
.objectid
, root
->root_key
.objectid
,
4579 parent_ino
, index
, name
, name_len
);
4580 } else if (add_backref
) {
4581 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
4586 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4588 btrfs_inode_type(inode
), index
);
4592 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4594 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4595 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4600 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4603 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4604 key
.objectid
, root
->root_key
.objectid
,
4605 parent_ino
, &local_index
, name
, name_len
);
4607 } else if (add_backref
) {
4611 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
4612 ino
, parent_ino
, &local_index
);
4617 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4618 struct inode
*dir
, struct dentry
*dentry
,
4619 struct inode
*inode
, int backref
, u64 index
)
4621 int err
= btrfs_add_link(trans
, dir
, inode
,
4622 dentry
->d_name
.name
, dentry
->d_name
.len
,
4629 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4630 umode_t mode
, dev_t rdev
)
4632 struct btrfs_trans_handle
*trans
;
4633 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4634 struct inode
*inode
= NULL
;
4638 unsigned long nr
= 0;
4641 if (!new_valid_dev(rdev
))
4645 * 2 for inode item and ref
4647 * 1 for xattr if selinux is on
4649 trans
= btrfs_start_transaction(root
, 5);
4651 return PTR_ERR(trans
);
4653 err
= btrfs_find_free_ino(root
, &objectid
);
4657 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4658 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4660 if (IS_ERR(inode
)) {
4661 err
= PTR_ERR(inode
);
4665 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4672 * If the active LSM wants to access the inode during
4673 * d_instantiate it needs these. Smack checks to see
4674 * if the filesystem supports xattrs by looking at the
4678 inode
->i_op
= &btrfs_special_inode_operations
;
4679 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4683 init_special_inode(inode
, inode
->i_mode
, rdev
);
4684 btrfs_update_inode(trans
, root
, inode
);
4685 d_instantiate(dentry
, inode
);
4688 nr
= trans
->blocks_used
;
4689 btrfs_end_transaction(trans
, root
);
4690 btrfs_btree_balance_dirty(root
, nr
);
4692 inode_dec_link_count(inode
);
4698 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4699 umode_t mode
, struct nameidata
*nd
)
4701 struct btrfs_trans_handle
*trans
;
4702 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4703 struct inode
*inode
= NULL
;
4706 unsigned long nr
= 0;
4711 * 2 for inode item and ref
4713 * 1 for xattr if selinux is on
4715 trans
= btrfs_start_transaction(root
, 5);
4717 return PTR_ERR(trans
);
4719 err
= btrfs_find_free_ino(root
, &objectid
);
4723 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4724 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4726 if (IS_ERR(inode
)) {
4727 err
= PTR_ERR(inode
);
4731 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4738 * If the active LSM wants to access the inode during
4739 * d_instantiate it needs these. Smack checks to see
4740 * if the filesystem supports xattrs by looking at the
4743 inode
->i_fop
= &btrfs_file_operations
;
4744 inode
->i_op
= &btrfs_file_inode_operations
;
4746 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4750 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4751 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4752 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4753 d_instantiate(dentry
, inode
);
4756 nr
= trans
->blocks_used
;
4757 btrfs_end_transaction(trans
, root
);
4759 inode_dec_link_count(inode
);
4762 btrfs_btree_balance_dirty(root
, nr
);
4766 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4767 struct dentry
*dentry
)
4769 struct btrfs_trans_handle
*trans
;
4770 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4771 struct inode
*inode
= old_dentry
->d_inode
;
4773 unsigned long nr
= 0;
4777 /* do not allow sys_link's with other subvols of the same device */
4778 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4781 if (inode
->i_nlink
== ~0U)
4784 err
= btrfs_set_inode_index(dir
, &index
);
4789 * 2 items for inode and inode ref
4790 * 2 items for dir items
4791 * 1 item for parent inode
4793 trans
= btrfs_start_transaction(root
, 5);
4794 if (IS_ERR(trans
)) {
4795 err
= PTR_ERR(trans
);
4799 btrfs_inc_nlink(inode
);
4800 inode
->i_ctime
= CURRENT_TIME
;
4803 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4808 struct dentry
*parent
= dentry
->d_parent
;
4809 err
= btrfs_update_inode(trans
, root
, inode
);
4811 d_instantiate(dentry
, inode
);
4812 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4815 nr
= trans
->blocks_used
;
4816 btrfs_end_transaction(trans
, root
);
4819 inode_dec_link_count(inode
);
4822 btrfs_btree_balance_dirty(root
, nr
);
4826 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
4828 struct inode
*inode
= NULL
;
4829 struct btrfs_trans_handle
*trans
;
4830 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4832 int drop_on_err
= 0;
4835 unsigned long nr
= 1;
4838 * 2 items for inode and ref
4839 * 2 items for dir items
4840 * 1 for xattr if selinux is on
4842 trans
= btrfs_start_transaction(root
, 5);
4844 return PTR_ERR(trans
);
4846 err
= btrfs_find_free_ino(root
, &objectid
);
4850 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4851 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4852 S_IFDIR
| mode
, &index
);
4853 if (IS_ERR(inode
)) {
4854 err
= PTR_ERR(inode
);
4860 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4864 inode
->i_op
= &btrfs_dir_inode_operations
;
4865 inode
->i_fop
= &btrfs_dir_file_operations
;
4867 btrfs_i_size_write(inode
, 0);
4868 err
= btrfs_update_inode(trans
, root
, inode
);
4872 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4873 dentry
->d_name
.len
, 0, index
);
4877 d_instantiate(dentry
, inode
);
4881 nr
= trans
->blocks_used
;
4882 btrfs_end_transaction(trans
, root
);
4885 btrfs_btree_balance_dirty(root
, nr
);
4889 /* helper for btfs_get_extent. Given an existing extent in the tree,
4890 * and an extent that you want to insert, deal with overlap and insert
4891 * the new extent into the tree.
4893 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4894 struct extent_map
*existing
,
4895 struct extent_map
*em
,
4896 u64 map_start
, u64 map_len
)
4900 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4901 start_diff
= map_start
- em
->start
;
4902 em
->start
= map_start
;
4904 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4905 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4906 em
->block_start
+= start_diff
;
4907 em
->block_len
-= start_diff
;
4909 return add_extent_mapping(em_tree
, em
);
4912 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4913 struct inode
*inode
, struct page
*page
,
4914 size_t pg_offset
, u64 extent_offset
,
4915 struct btrfs_file_extent_item
*item
)
4918 struct extent_buffer
*leaf
= path
->nodes
[0];
4921 unsigned long inline_size
;
4925 WARN_ON(pg_offset
!= 0);
4926 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4927 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4928 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4929 btrfs_item_nr(leaf
, path
->slots
[0]));
4930 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4933 ptr
= btrfs_file_extent_inline_start(item
);
4935 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
4937 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
4938 ret
= btrfs_decompress(compress_type
, tmp
, page
,
4939 extent_offset
, inline_size
, max_size
);
4941 char *kaddr
= kmap_atomic(page
, KM_USER0
);
4942 unsigned long copy_size
= min_t(u64
,
4943 PAGE_CACHE_SIZE
- pg_offset
,
4944 max_size
- extent_offset
);
4945 memset(kaddr
+ pg_offset
, 0, copy_size
);
4946 kunmap_atomic(kaddr
, KM_USER0
);
4953 * a bit scary, this does extent mapping from logical file offset to the disk.
4954 * the ugly parts come from merging extents from the disk with the in-ram
4955 * representation. This gets more complex because of the data=ordered code,
4956 * where the in-ram extents might be locked pending data=ordered completion.
4958 * This also copies inline extents directly into the page.
4961 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
4962 size_t pg_offset
, u64 start
, u64 len
,
4968 u64 extent_start
= 0;
4970 u64 objectid
= btrfs_ino(inode
);
4972 struct btrfs_path
*path
= NULL
;
4973 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4974 struct btrfs_file_extent_item
*item
;
4975 struct extent_buffer
*leaf
;
4976 struct btrfs_key found_key
;
4977 struct extent_map
*em
= NULL
;
4978 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4979 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4980 struct btrfs_trans_handle
*trans
= NULL
;
4984 read_lock(&em_tree
->lock
);
4985 em
= lookup_extent_mapping(em_tree
, start
, len
);
4987 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4988 read_unlock(&em_tree
->lock
);
4991 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
4992 free_extent_map(em
);
4993 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
4994 free_extent_map(em
);
4998 em
= alloc_extent_map();
5003 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5004 em
->start
= EXTENT_MAP_HOLE
;
5005 em
->orig_start
= EXTENT_MAP_HOLE
;
5007 em
->block_len
= (u64
)-1;
5010 path
= btrfs_alloc_path();
5016 * Chances are we'll be called again, so go ahead and do
5022 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5023 objectid
, start
, trans
!= NULL
);
5030 if (path
->slots
[0] == 0)
5035 leaf
= path
->nodes
[0];
5036 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5037 struct btrfs_file_extent_item
);
5038 /* are we inside the extent that was found? */
5039 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5040 found_type
= btrfs_key_type(&found_key
);
5041 if (found_key
.objectid
!= objectid
||
5042 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5046 found_type
= btrfs_file_extent_type(leaf
, item
);
5047 extent_start
= found_key
.offset
;
5048 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5049 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5050 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5051 extent_end
= extent_start
+
5052 btrfs_file_extent_num_bytes(leaf
, item
);
5053 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5055 size
= btrfs_file_extent_inline_len(leaf
, item
);
5056 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5057 ~((u64
)root
->sectorsize
- 1);
5060 if (start
>= extent_end
) {
5062 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5063 ret
= btrfs_next_leaf(root
, path
);
5070 leaf
= path
->nodes
[0];
5072 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5073 if (found_key
.objectid
!= objectid
||
5074 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5076 if (start
+ len
<= found_key
.offset
)
5079 em
->len
= found_key
.offset
- start
;
5083 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5084 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5085 em
->start
= extent_start
;
5086 em
->len
= extent_end
- extent_start
;
5087 em
->orig_start
= extent_start
-
5088 btrfs_file_extent_offset(leaf
, item
);
5089 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5091 em
->block_start
= EXTENT_MAP_HOLE
;
5094 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5095 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5096 em
->compress_type
= compress_type
;
5097 em
->block_start
= bytenr
;
5098 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5101 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5102 em
->block_start
= bytenr
;
5103 em
->block_len
= em
->len
;
5104 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5105 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5108 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5112 size_t extent_offset
;
5115 em
->block_start
= EXTENT_MAP_INLINE
;
5116 if (!page
|| create
) {
5117 em
->start
= extent_start
;
5118 em
->len
= extent_end
- extent_start
;
5122 size
= btrfs_file_extent_inline_len(leaf
, item
);
5123 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5124 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5125 size
- extent_offset
);
5126 em
->start
= extent_start
+ extent_offset
;
5127 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5128 ~((u64
)root
->sectorsize
- 1);
5129 em
->orig_start
= EXTENT_MAP_INLINE
;
5130 if (compress_type
) {
5131 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5132 em
->compress_type
= compress_type
;
5134 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5135 if (create
== 0 && !PageUptodate(page
)) {
5136 if (btrfs_file_extent_compression(leaf
, item
) !=
5137 BTRFS_COMPRESS_NONE
) {
5138 ret
= uncompress_inline(path
, inode
, page
,
5140 extent_offset
, item
);
5144 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5146 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5147 memset(map
+ pg_offset
+ copy_size
, 0,
5148 PAGE_CACHE_SIZE
- pg_offset
-
5153 flush_dcache_page(page
);
5154 } else if (create
&& PageUptodate(page
)) {
5158 free_extent_map(em
);
5161 btrfs_release_path(path
);
5162 trans
= btrfs_join_transaction(root
);
5165 return ERR_CAST(trans
);
5169 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5172 btrfs_mark_buffer_dirty(leaf
);
5174 set_extent_uptodate(io_tree
, em
->start
,
5175 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5178 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5185 em
->block_start
= EXTENT_MAP_HOLE
;
5186 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5188 btrfs_release_path(path
);
5189 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5190 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5191 "[%llu %llu]\n", (unsigned long long)em
->start
,
5192 (unsigned long long)em
->len
,
5193 (unsigned long long)start
,
5194 (unsigned long long)len
);
5200 write_lock(&em_tree
->lock
);
5201 ret
= add_extent_mapping(em_tree
, em
);
5202 /* it is possible that someone inserted the extent into the tree
5203 * while we had the lock dropped. It is also possible that
5204 * an overlapping map exists in the tree
5206 if (ret
== -EEXIST
) {
5207 struct extent_map
*existing
;
5211 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5212 if (existing
&& (existing
->start
> start
||
5213 existing
->start
+ existing
->len
<= start
)) {
5214 free_extent_map(existing
);
5218 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5221 err
= merge_extent_mapping(em_tree
, existing
,
5224 free_extent_map(existing
);
5226 free_extent_map(em
);
5231 free_extent_map(em
);
5235 free_extent_map(em
);
5240 write_unlock(&em_tree
->lock
);
5243 trace_btrfs_get_extent(root
, em
);
5246 btrfs_free_path(path
);
5248 ret
= btrfs_end_transaction(trans
, root
);
5253 free_extent_map(em
);
5254 return ERR_PTR(err
);
5259 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5260 size_t pg_offset
, u64 start
, u64 len
,
5263 struct extent_map
*em
;
5264 struct extent_map
*hole_em
= NULL
;
5265 u64 range_start
= start
;
5271 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5276 * if our em maps to a hole, there might
5277 * actually be delalloc bytes behind it
5279 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5285 /* check to see if we've wrapped (len == -1 or similar) */
5294 /* ok, we didn't find anything, lets look for delalloc */
5295 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5296 end
, len
, EXTENT_DELALLOC
, 1);
5297 found_end
= range_start
+ found
;
5298 if (found_end
< range_start
)
5299 found_end
= (u64
)-1;
5302 * we didn't find anything useful, return
5303 * the original results from get_extent()
5305 if (range_start
> end
|| found_end
<= start
) {
5311 /* adjust the range_start to make sure it doesn't
5312 * go backwards from the start they passed in
5314 range_start
= max(start
,range_start
);
5315 found
= found_end
- range_start
;
5318 u64 hole_start
= start
;
5321 em
= alloc_extent_map();
5327 * when btrfs_get_extent can't find anything it
5328 * returns one huge hole
5330 * make sure what it found really fits our range, and
5331 * adjust to make sure it is based on the start from
5335 u64 calc_end
= extent_map_end(hole_em
);
5337 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5338 free_extent_map(hole_em
);
5341 hole_start
= max(hole_em
->start
, start
);
5342 hole_len
= calc_end
- hole_start
;
5346 if (hole_em
&& range_start
> hole_start
) {
5347 /* our hole starts before our delalloc, so we
5348 * have to return just the parts of the hole
5349 * that go until the delalloc starts
5351 em
->len
= min(hole_len
,
5352 range_start
- hole_start
);
5353 em
->start
= hole_start
;
5354 em
->orig_start
= hole_start
;
5356 * don't adjust block start at all,
5357 * it is fixed at EXTENT_MAP_HOLE
5359 em
->block_start
= hole_em
->block_start
;
5360 em
->block_len
= hole_len
;
5362 em
->start
= range_start
;
5364 em
->orig_start
= range_start
;
5365 em
->block_start
= EXTENT_MAP_DELALLOC
;
5366 em
->block_len
= found
;
5368 } else if (hole_em
) {
5373 free_extent_map(hole_em
);
5375 free_extent_map(em
);
5376 return ERR_PTR(err
);
5381 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5382 struct extent_map
*em
,
5385 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5386 struct btrfs_trans_handle
*trans
;
5387 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5388 struct btrfs_key ins
;
5391 bool insert
= false;
5394 * Ok if the extent map we looked up is a hole and is for the exact
5395 * range we want, there is no reason to allocate a new one, however if
5396 * it is not right then we need to free this one and drop the cache for
5399 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5401 free_extent_map(em
);
5404 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5407 trans
= btrfs_join_transaction(root
);
5409 return ERR_CAST(trans
);
5411 if (start
<= BTRFS_I(inode
)->disk_i_size
&& len
< 64 * 1024)
5412 btrfs_add_inode_defrag(trans
, inode
);
5414 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5416 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5417 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5418 alloc_hint
, (u64
)-1, &ins
, 1);
5425 em
= alloc_extent_map();
5427 em
= ERR_PTR(-ENOMEM
);
5433 em
->orig_start
= em
->start
;
5434 em
->len
= ins
.offset
;
5436 em
->block_start
= ins
.objectid
;
5437 em
->block_len
= ins
.offset
;
5438 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5441 * We need to do this because if we're using the original em we searched
5442 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5445 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5448 write_lock(&em_tree
->lock
);
5449 ret
= add_extent_mapping(em_tree
, em
);
5450 write_unlock(&em_tree
->lock
);
5453 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5456 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5457 ins
.offset
, ins
.offset
, 0);
5459 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5463 btrfs_end_transaction(trans
, root
);
5468 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5469 * block must be cow'd
5471 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5472 struct inode
*inode
, u64 offset
, u64 len
)
5474 struct btrfs_path
*path
;
5476 struct extent_buffer
*leaf
;
5477 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5478 struct btrfs_file_extent_item
*fi
;
5479 struct btrfs_key key
;
5487 path
= btrfs_alloc_path();
5491 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
5496 slot
= path
->slots
[0];
5499 /* can't find the item, must cow */
5506 leaf
= path
->nodes
[0];
5507 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5508 if (key
.objectid
!= btrfs_ino(inode
) ||
5509 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5510 /* not our file or wrong item type, must cow */
5514 if (key
.offset
> offset
) {
5515 /* Wrong offset, must cow */
5519 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5520 found_type
= btrfs_file_extent_type(leaf
, fi
);
5521 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5522 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5523 /* not a regular extent, must cow */
5526 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5527 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5529 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5530 if (extent_end
< offset
+ len
) {
5531 /* extent doesn't include our full range, must cow */
5535 if (btrfs_extent_readonly(root
, disk_bytenr
))
5539 * look for other files referencing this extent, if we
5540 * find any we must cow
5542 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
5543 key
.offset
- backref_offset
, disk_bytenr
))
5547 * adjust disk_bytenr and num_bytes to cover just the bytes
5548 * in this extent we are about to write. If there
5549 * are any csums in that range we have to cow in order
5550 * to keep the csums correct
5552 disk_bytenr
+= backref_offset
;
5553 disk_bytenr
+= offset
- key
.offset
;
5554 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5555 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5558 * all of the above have passed, it is safe to overwrite this extent
5563 btrfs_free_path(path
);
5567 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5568 struct buffer_head
*bh_result
, int create
)
5570 struct extent_map
*em
;
5571 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5572 u64 start
= iblock
<< inode
->i_blkbits
;
5573 u64 len
= bh_result
->b_size
;
5574 struct btrfs_trans_handle
*trans
;
5576 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5581 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5582 * io. INLINE is special, and we could probably kludge it in here, but
5583 * it's still buffered so for safety lets just fall back to the generic
5586 * For COMPRESSED we _have_ to read the entire extent in so we can
5587 * decompress it, so there will be buffering required no matter what we
5588 * do, so go ahead and fallback to buffered.
5590 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5591 * to buffered IO. Don't blame me, this is the price we pay for using
5594 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5595 em
->block_start
== EXTENT_MAP_INLINE
) {
5596 free_extent_map(em
);
5600 /* Just a good old fashioned hole, return */
5601 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5602 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5603 free_extent_map(em
);
5604 /* DIO will do one hole at a time, so just unlock a sector */
5605 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5606 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5611 * We don't allocate a new extent in the following cases
5613 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5615 * 2) The extent is marked as PREALLOC. We're good to go here and can
5616 * just use the extent.
5620 len
= em
->len
- (start
- em
->start
);
5624 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5625 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5626 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5631 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5632 type
= BTRFS_ORDERED_PREALLOC
;
5634 type
= BTRFS_ORDERED_NOCOW
;
5635 len
= min(len
, em
->len
- (start
- em
->start
));
5636 block_start
= em
->block_start
+ (start
- em
->start
);
5639 * we're not going to log anything, but we do need
5640 * to make sure the current transaction stays open
5641 * while we look for nocow cross refs
5643 trans
= btrfs_join_transaction(root
);
5647 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5648 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5649 block_start
, len
, len
, type
);
5650 btrfs_end_transaction(trans
, root
);
5652 free_extent_map(em
);
5657 btrfs_end_transaction(trans
, root
);
5661 * this will cow the extent, reset the len in case we changed
5664 len
= bh_result
->b_size
;
5665 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
5668 len
= min(len
, em
->len
- (start
- em
->start
));
5670 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5671 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5674 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5676 bh_result
->b_size
= len
;
5677 bh_result
->b_bdev
= em
->bdev
;
5678 set_buffer_mapped(bh_result
);
5679 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5680 set_buffer_new(bh_result
);
5682 free_extent_map(em
);
5687 struct btrfs_dio_private
{
5688 struct inode
*inode
;
5695 /* number of bios pending for this dio */
5696 atomic_t pending_bios
;
5701 struct bio
*orig_bio
;
5704 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5706 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5707 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5708 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5709 struct inode
*inode
= dip
->inode
;
5710 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5712 u32
*private = dip
->csums
;
5714 start
= dip
->logical_offset
;
5716 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5717 struct page
*page
= bvec
->bv_page
;
5720 unsigned long flags
;
5722 local_irq_save(flags
);
5723 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5724 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5725 csum
, bvec
->bv_len
);
5726 btrfs_csum_final(csum
, (char *)&csum
);
5727 kunmap_atomic(kaddr
, KM_IRQ0
);
5728 local_irq_restore(flags
);
5730 flush_dcache_page(bvec
->bv_page
);
5731 if (csum
!= *private) {
5732 printk(KERN_ERR
"btrfs csum failed ino %llu off"
5733 " %llu csum %u private %u\n",
5734 (unsigned long long)btrfs_ino(inode
),
5735 (unsigned long long)start
,
5741 start
+= bvec
->bv_len
;
5744 } while (bvec
<= bvec_end
);
5746 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5747 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5748 bio
->bi_private
= dip
->private;
5753 /* If we had a csum failure make sure to clear the uptodate flag */
5755 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5756 dio_end_io(bio
, err
);
5759 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5761 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5762 struct inode
*inode
= dip
->inode
;
5763 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5764 struct btrfs_trans_handle
*trans
;
5765 struct btrfs_ordered_extent
*ordered
= NULL
;
5766 struct extent_state
*cached_state
= NULL
;
5767 u64 ordered_offset
= dip
->logical_offset
;
5768 u64 ordered_bytes
= dip
->bytes
;
5774 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5782 trans
= btrfs_join_transaction(root
);
5783 if (IS_ERR(trans
)) {
5787 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5789 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5790 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5792 err
= btrfs_update_inode_fallback(trans
, root
, inode
);
5796 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5797 ordered
->file_offset
+ ordered
->len
- 1, 0,
5798 &cached_state
, GFP_NOFS
);
5800 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5801 ret
= btrfs_mark_extent_written(trans
, inode
,
5802 ordered
->file_offset
,
5803 ordered
->file_offset
+
5810 ret
= insert_reserved_file_extent(trans
, inode
,
5811 ordered
->file_offset
,
5817 BTRFS_FILE_EXTENT_REG
);
5818 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5819 ordered
->file_offset
, ordered
->len
);
5827 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5828 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5829 if (!ret
|| !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
))
5830 btrfs_update_inode_fallback(trans
, root
, inode
);
5833 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5834 ordered
->file_offset
+ ordered
->len
- 1,
5835 &cached_state
, GFP_NOFS
);
5837 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5838 btrfs_end_transaction(trans
, root
);
5839 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5840 btrfs_put_ordered_extent(ordered
);
5841 btrfs_put_ordered_extent(ordered
);
5845 * our bio might span multiple ordered extents. If we haven't
5846 * completed the accounting for the whole dio, go back and try again
5848 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5849 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5854 bio
->bi_private
= dip
->private;
5859 /* If we had an error make sure to clear the uptodate flag */
5861 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5862 dio_end_io(bio
, err
);
5865 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5866 struct bio
*bio
, int mirror_num
,
5867 unsigned long bio_flags
, u64 offset
)
5870 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5871 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5876 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5878 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5881 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
5882 "sector %#Lx len %u err no %d\n",
5883 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
5884 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5888 * before atomic variable goto zero, we must make sure
5889 * dip->errors is perceived to be set.
5891 smp_mb__before_atomic_dec();
5894 /* if there are more bios still pending for this dio, just exit */
5895 if (!atomic_dec_and_test(&dip
->pending_bios
))
5899 bio_io_error(dip
->orig_bio
);
5901 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5902 bio_endio(dip
->orig_bio
, 0);
5908 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5909 u64 first_sector
, gfp_t gfp_flags
)
5911 int nr_vecs
= bio_get_nr_vecs(bdev
);
5912 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5915 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5916 int rw
, u64 file_offset
, int skip_sum
,
5917 u32
*csums
, int async_submit
)
5919 int write
= rw
& REQ_WRITE
;
5920 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5924 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5931 if (write
&& async_submit
) {
5932 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5933 inode
, rw
, bio
, 0, 0,
5935 __btrfs_submit_bio_start_direct_io
,
5936 __btrfs_submit_bio_done
);
5940 * If we aren't doing async submit, calculate the csum of the
5943 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
5946 } else if (!skip_sum
) {
5947 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5948 file_offset
, csums
);
5954 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
5960 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5963 struct inode
*inode
= dip
->inode
;
5964 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5965 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5967 struct bio
*orig_bio
= dip
->orig_bio
;
5968 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5969 u64 start_sector
= orig_bio
->bi_sector
;
5970 u64 file_offset
= dip
->logical_offset
;
5974 u32
*csums
= dip
->csums
;
5976 int async_submit
= 0;
5977 int write
= rw
& REQ_WRITE
;
5979 map_length
= orig_bio
->bi_size
;
5980 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
5981 &map_length
, NULL
, 0);
5987 if (map_length
>= orig_bio
->bi_size
) {
5993 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
5996 bio
->bi_private
= dip
;
5997 bio
->bi_end_io
= btrfs_end_dio_bio
;
5998 atomic_inc(&dip
->pending_bios
);
6000 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
6001 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
6002 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
6003 bvec
->bv_offset
) < bvec
->bv_len
)) {
6005 * inc the count before we submit the bio so
6006 * we know the end IO handler won't happen before
6007 * we inc the count. Otherwise, the dip might get freed
6008 * before we're done setting it up
6010 atomic_inc(&dip
->pending_bios
);
6011 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
6012 file_offset
, skip_sum
,
6013 csums
, async_submit
);
6016 atomic_dec(&dip
->pending_bios
);
6020 /* Write's use the ordered csums */
6021 if (!write
&& !skip_sum
)
6022 csums
= csums
+ nr_pages
;
6023 start_sector
+= submit_len
>> 9;
6024 file_offset
+= submit_len
;
6029 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6030 start_sector
, GFP_NOFS
);
6033 bio
->bi_private
= dip
;
6034 bio
->bi_end_io
= btrfs_end_dio_bio
;
6036 map_length
= orig_bio
->bi_size
;
6037 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6038 &map_length
, NULL
, 0);
6044 submit_len
+= bvec
->bv_len
;
6051 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6052 csums
, async_submit
);
6060 * before atomic variable goto zero, we must
6061 * make sure dip->errors is perceived to be set.
6063 smp_mb__before_atomic_dec();
6064 if (atomic_dec_and_test(&dip
->pending_bios
))
6065 bio_io_error(dip
->orig_bio
);
6067 /* bio_end_io() will handle error, so we needn't return it */
6071 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6074 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6075 struct btrfs_dio_private
*dip
;
6076 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6078 int write
= rw
& REQ_WRITE
;
6081 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6083 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6090 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6091 if (!write
&& !skip_sum
) {
6092 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6100 dip
->private = bio
->bi_private
;
6102 dip
->logical_offset
= file_offset
;
6106 dip
->bytes
+= bvec
->bv_len
;
6108 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6110 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6111 bio
->bi_private
= dip
;
6113 dip
->orig_bio
= bio
;
6114 atomic_set(&dip
->pending_bios
, 0);
6117 bio
->bi_end_io
= btrfs_endio_direct_write
;
6119 bio
->bi_end_io
= btrfs_endio_direct_read
;
6121 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6126 * If this is a write, we need to clean up the reserved space and kill
6127 * the ordered extent.
6130 struct btrfs_ordered_extent
*ordered
;
6131 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6132 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6133 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6134 btrfs_free_reserved_extent(root
, ordered
->start
,
6136 btrfs_put_ordered_extent(ordered
);
6137 btrfs_put_ordered_extent(ordered
);
6139 bio_endio(bio
, ret
);
6142 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6143 const struct iovec
*iov
, loff_t offset
,
6144 unsigned long nr_segs
)
6150 unsigned blocksize_mask
= root
->sectorsize
- 1;
6151 ssize_t retval
= -EINVAL
;
6152 loff_t end
= offset
;
6154 if (offset
& blocksize_mask
)
6157 /* Check the memory alignment. Blocks cannot straddle pages */
6158 for (seg
= 0; seg
< nr_segs
; seg
++) {
6159 addr
= (unsigned long)iov
[seg
].iov_base
;
6160 size
= iov
[seg
].iov_len
;
6162 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6165 /* If this is a write we don't need to check anymore */
6170 * Check to make sure we don't have duplicate iov_base's in this
6171 * iovec, if so return EINVAL, otherwise we'll get csum errors
6172 * when reading back.
6174 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6175 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6183 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6184 const struct iovec
*iov
, loff_t offset
,
6185 unsigned long nr_segs
)
6187 struct file
*file
= iocb
->ki_filp
;
6188 struct inode
*inode
= file
->f_mapping
->host
;
6189 struct btrfs_ordered_extent
*ordered
;
6190 struct extent_state
*cached_state
= NULL
;
6191 u64 lockstart
, lockend
;
6193 int writing
= rw
& WRITE
;
6195 size_t count
= iov_length(iov
, nr_segs
);
6197 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6203 lockend
= offset
+ count
- 1;
6206 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6212 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6213 0, &cached_state
, GFP_NOFS
);
6215 * We're concerned with the entire range that we're going to be
6216 * doing DIO to, so we need to make sure theres no ordered
6217 * extents in this range.
6219 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6220 lockend
- lockstart
+ 1);
6223 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6224 &cached_state
, GFP_NOFS
);
6225 btrfs_start_ordered_extent(inode
, ordered
, 1);
6226 btrfs_put_ordered_extent(ordered
);
6231 * we don't use btrfs_set_extent_delalloc because we don't want
6232 * the dirty or uptodate bits
6235 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6236 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6237 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6240 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6241 lockend
, EXTENT_LOCKED
| write_bits
,
6242 1, 0, &cached_state
, GFP_NOFS
);
6247 free_extent_state(cached_state
);
6248 cached_state
= NULL
;
6250 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6251 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6252 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6253 btrfs_submit_direct
, 0);
6255 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6256 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6257 offset
+ iov_length(iov
, nr_segs
) - 1,
6258 EXTENT_LOCKED
| write_bits
, 1, 0,
6259 &cached_state
, GFP_NOFS
);
6260 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6262 * We're falling back to buffered, unlock the section we didn't
6265 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6266 offset
+ iov_length(iov
, nr_segs
) - 1,
6267 EXTENT_LOCKED
| write_bits
, 1, 0,
6268 &cached_state
, GFP_NOFS
);
6271 free_extent_state(cached_state
);
6275 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6276 __u64 start
, __u64 len
)
6278 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6281 int btrfs_readpage(struct file
*file
, struct page
*page
)
6283 struct extent_io_tree
*tree
;
6284 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6285 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
6288 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6290 struct extent_io_tree
*tree
;
6293 if (current
->flags
& PF_MEMALLOC
) {
6294 redirty_page_for_writepage(wbc
, page
);
6298 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6299 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6302 int btrfs_writepages(struct address_space
*mapping
,
6303 struct writeback_control
*wbc
)
6305 struct extent_io_tree
*tree
;
6307 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6308 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6312 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6313 struct list_head
*pages
, unsigned nr_pages
)
6315 struct extent_io_tree
*tree
;
6316 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6317 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6320 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6322 struct extent_io_tree
*tree
;
6323 struct extent_map_tree
*map
;
6326 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6327 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6328 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6330 ClearPagePrivate(page
);
6331 set_page_private(page
, 0);
6332 page_cache_release(page
);
6337 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6339 if (PageWriteback(page
) || PageDirty(page
))
6341 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6344 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6346 struct extent_io_tree
*tree
;
6347 struct btrfs_ordered_extent
*ordered
;
6348 struct extent_state
*cached_state
= NULL
;
6349 u64 page_start
= page_offset(page
);
6350 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6354 * we have the page locked, so new writeback can't start,
6355 * and the dirty bit won't be cleared while we are here.
6357 * Wait for IO on this page so that we can safely clear
6358 * the PagePrivate2 bit and do ordered accounting
6360 wait_on_page_writeback(page
);
6362 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6364 btrfs_releasepage(page
, GFP_NOFS
);
6367 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6369 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6373 * IO on this page will never be started, so we need
6374 * to account for any ordered extents now
6376 clear_extent_bit(tree
, page_start
, page_end
,
6377 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6378 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6379 &cached_state
, GFP_NOFS
);
6381 * whoever cleared the private bit is responsible
6382 * for the finish_ordered_io
6384 if (TestClearPagePrivate2(page
)) {
6385 btrfs_finish_ordered_io(page
->mapping
->host
,
6386 page_start
, page_end
);
6388 btrfs_put_ordered_extent(ordered
);
6389 cached_state
= NULL
;
6390 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6393 clear_extent_bit(tree
, page_start
, page_end
,
6394 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6395 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6396 __btrfs_releasepage(page
, GFP_NOFS
);
6398 ClearPageChecked(page
);
6399 if (PagePrivate(page
)) {
6400 ClearPagePrivate(page
);
6401 set_page_private(page
, 0);
6402 page_cache_release(page
);
6407 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6408 * called from a page fault handler when a page is first dirtied. Hence we must
6409 * be careful to check for EOF conditions here. We set the page up correctly
6410 * for a written page which means we get ENOSPC checking when writing into
6411 * holes and correct delalloc and unwritten extent mapping on filesystems that
6412 * support these features.
6414 * We are not allowed to take the i_mutex here so we have to play games to
6415 * protect against truncate races as the page could now be beyond EOF. Because
6416 * vmtruncate() writes the inode size before removing pages, once we have the
6417 * page lock we can determine safely if the page is beyond EOF. If it is not
6418 * beyond EOF, then the page is guaranteed safe against truncation until we
6421 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6423 struct page
*page
= vmf
->page
;
6424 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6425 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6426 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6427 struct btrfs_ordered_extent
*ordered
;
6428 struct extent_state
*cached_state
= NULL
;
6430 unsigned long zero_start
;
6437 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6439 ret
= btrfs_update_time(vma
->vm_file
);
6445 else /* -ENOSPC, -EIO, etc */
6446 ret
= VM_FAULT_SIGBUS
;
6452 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6455 size
= i_size_read(inode
);
6456 page_start
= page_offset(page
);
6457 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6459 if ((page
->mapping
!= inode
->i_mapping
) ||
6460 (page_start
>= size
)) {
6461 /* page got truncated out from underneath us */
6464 wait_on_page_writeback(page
);
6466 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6468 set_page_extent_mapped(page
);
6471 * we can't set the delalloc bits if there are pending ordered
6472 * extents. Drop our locks and wait for them to finish
6474 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6476 unlock_extent_cached(io_tree
, page_start
, page_end
,
6477 &cached_state
, GFP_NOFS
);
6479 btrfs_start_ordered_extent(inode
, ordered
, 1);
6480 btrfs_put_ordered_extent(ordered
);
6485 * XXX - page_mkwrite gets called every time the page is dirtied, even
6486 * if it was already dirty, so for space accounting reasons we need to
6487 * clear any delalloc bits for the range we are fixing to save. There
6488 * is probably a better way to do this, but for now keep consistent with
6489 * prepare_pages in the normal write path.
6491 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6492 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6493 0, 0, &cached_state
, GFP_NOFS
);
6495 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6498 unlock_extent_cached(io_tree
, page_start
, page_end
,
6499 &cached_state
, GFP_NOFS
);
6500 ret
= VM_FAULT_SIGBUS
;
6505 /* page is wholly or partially inside EOF */
6506 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6507 zero_start
= size
& ~PAGE_CACHE_MASK
;
6509 zero_start
= PAGE_CACHE_SIZE
;
6511 if (zero_start
!= PAGE_CACHE_SIZE
) {
6513 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6514 flush_dcache_page(page
);
6517 ClearPageChecked(page
);
6518 set_page_dirty(page
);
6519 SetPageUptodate(page
);
6521 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6522 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6524 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6528 return VM_FAULT_LOCKED
;
6531 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6536 static int btrfs_truncate(struct inode
*inode
)
6538 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6539 struct btrfs_block_rsv
*rsv
;
6542 struct btrfs_trans_handle
*trans
;
6544 u64 mask
= root
->sectorsize
- 1;
6545 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
6547 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6551 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6552 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6555 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6556 * 3 things going on here
6558 * 1) We need to reserve space for our orphan item and the space to
6559 * delete our orphan item. Lord knows we don't want to have a dangling
6560 * orphan item because we didn't reserve space to remove it.
6562 * 2) We need to reserve space to update our inode.
6564 * 3) We need to have something to cache all the space that is going to
6565 * be free'd up by the truncate operation, but also have some slack
6566 * space reserved in case it uses space during the truncate (thank you
6567 * very much snapshotting).
6569 * And we need these to all be seperate. The fact is we can use alot of
6570 * space doing the truncate, and we have no earthly idea how much space
6571 * we will use, so we need the truncate reservation to be seperate so it
6572 * doesn't end up using space reserved for updating the inode or
6573 * removing the orphan item. We also need to be able to stop the
6574 * transaction and start a new one, which means we need to be able to
6575 * update the inode several times, and we have no idea of knowing how
6576 * many times that will be, so we can't just reserve 1 item for the
6577 * entirety of the opration, so that has to be done seperately as well.
6578 * Then there is the orphan item, which does indeed need to be held on
6579 * to for the whole operation, and we need nobody to touch this reserved
6580 * space except the orphan code.
6582 * So that leaves us with
6584 * 1) root->orphan_block_rsv - for the orphan deletion.
6585 * 2) rsv - for the truncate reservation, which we will steal from the
6586 * transaction reservation.
6587 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6588 * updating the inode.
6590 rsv
= btrfs_alloc_block_rsv(root
);
6593 rsv
->size
= min_size
;
6596 * 1 for the truncate slack space
6597 * 1 for the orphan item we're going to add
6598 * 1 for the orphan item deletion
6599 * 1 for updating the inode.
6601 trans
= btrfs_start_transaction(root
, 4);
6602 if (IS_ERR(trans
)) {
6603 err
= PTR_ERR(trans
);
6607 /* Migrate the slack space for the truncate to our reserve */
6608 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
6612 ret
= btrfs_orphan_add(trans
, inode
);
6614 btrfs_end_transaction(trans
, root
);
6619 * setattr is responsible for setting the ordered_data_close flag,
6620 * but that is only tested during the last file release. That
6621 * could happen well after the next commit, leaving a great big
6622 * window where new writes may get lost if someone chooses to write
6623 * to this file after truncating to zero
6625 * The inode doesn't have any dirty data here, and so if we commit
6626 * this is a noop. If someone immediately starts writing to the inode
6627 * it is very likely we'll catch some of their writes in this
6628 * transaction, and the commit will find this file on the ordered
6629 * data list with good things to send down.
6631 * This is a best effort solution, there is still a window where
6632 * using truncate to replace the contents of the file will
6633 * end up with a zero length file after a crash.
6635 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6636 btrfs_add_ordered_operation(trans
, root
, inode
);
6639 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
);
6642 * This can only happen with the original transaction we
6643 * started above, every other time we shouldn't have a
6644 * transaction started yet.
6653 /* Just need the 1 for updating the inode */
6654 trans
= btrfs_start_transaction(root
, 1);
6655 if (IS_ERR(trans
)) {
6656 ret
= err
= PTR_ERR(trans
);
6662 trans
->block_rsv
= rsv
;
6664 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6666 BTRFS_EXTENT_DATA_KEY
);
6667 if (ret
!= -EAGAIN
) {
6672 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6673 ret
= btrfs_update_inode(trans
, root
, inode
);
6679 nr
= trans
->blocks_used
;
6680 btrfs_end_transaction(trans
, root
);
6682 btrfs_btree_balance_dirty(root
, nr
);
6685 if (ret
== 0 && inode
->i_nlink
> 0) {
6686 trans
->block_rsv
= root
->orphan_block_rsv
;
6687 ret
= btrfs_orphan_del(trans
, inode
);
6690 } else if (ret
&& inode
->i_nlink
> 0) {
6692 * Failed to do the truncate, remove us from the in memory
6695 ret
= btrfs_orphan_del(NULL
, inode
);
6699 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6700 ret
= btrfs_update_inode(trans
, root
, inode
);
6704 nr
= trans
->blocks_used
;
6705 ret
= btrfs_end_transaction(trans
, root
);
6706 btrfs_btree_balance_dirty(root
, nr
);
6710 btrfs_free_block_rsv(root
, rsv
);
6719 * create a new subvolume directory/inode (helper for the ioctl).
6721 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6722 struct btrfs_root
*new_root
, u64 new_dirid
)
6724 struct inode
*inode
;
6728 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
6729 new_dirid
, new_dirid
,
6730 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
6733 return PTR_ERR(inode
);
6734 inode
->i_op
= &btrfs_dir_inode_operations
;
6735 inode
->i_fop
= &btrfs_dir_file_operations
;
6737 set_nlink(inode
, 1);
6738 btrfs_i_size_write(inode
, 0);
6740 err
= btrfs_update_inode(trans
, new_root
, inode
);
6747 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6749 struct btrfs_inode
*ei
;
6750 struct inode
*inode
;
6752 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6757 ei
->space_info
= NULL
;
6761 ei
->last_sub_trans
= 0;
6762 ei
->logged_trans
= 0;
6763 ei
->delalloc_bytes
= 0;
6764 ei
->disk_i_size
= 0;
6767 ei
->index_cnt
= (u64
)-1;
6768 ei
->last_unlink_trans
= 0;
6770 spin_lock_init(&ei
->lock
);
6771 ei
->outstanding_extents
= 0;
6772 ei
->reserved_extents
= 0;
6774 ei
->ordered_data_close
= 0;
6775 ei
->orphan_meta_reserved
= 0;
6776 ei
->dummy_inode
= 0;
6778 ei
->delalloc_meta_reserved
= 0;
6779 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6781 ei
->delayed_node
= NULL
;
6783 inode
= &ei
->vfs_inode
;
6784 extent_map_tree_init(&ei
->extent_tree
);
6785 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
6786 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
6787 mutex_init(&ei
->log_mutex
);
6788 mutex_init(&ei
->delalloc_mutex
);
6789 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6790 INIT_LIST_HEAD(&ei
->i_orphan
);
6791 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6792 INIT_LIST_HEAD(&ei
->ordered_operations
);
6793 RB_CLEAR_NODE(&ei
->rb_node
);
6798 static void btrfs_i_callback(struct rcu_head
*head
)
6800 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6801 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6804 void btrfs_destroy_inode(struct inode
*inode
)
6806 struct btrfs_ordered_extent
*ordered
;
6807 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6809 WARN_ON(!list_empty(&inode
->i_dentry
));
6810 WARN_ON(inode
->i_data
.nrpages
);
6811 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
6812 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
6813 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
6814 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
6817 * This can happen where we create an inode, but somebody else also
6818 * created the same inode and we need to destroy the one we already
6825 * Make sure we're properly removed from the ordered operation
6829 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6830 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6831 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6832 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6835 spin_lock(&root
->orphan_lock
);
6836 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6837 printk(KERN_INFO
"BTRFS: inode %llu still on the orphan list\n",
6838 (unsigned long long)btrfs_ino(inode
));
6839 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6841 spin_unlock(&root
->orphan_lock
);
6844 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6848 printk(KERN_ERR
"btrfs found ordered "
6849 "extent %llu %llu on inode cleanup\n",
6850 (unsigned long long)ordered
->file_offset
,
6851 (unsigned long long)ordered
->len
);
6852 btrfs_remove_ordered_extent(inode
, ordered
);
6853 btrfs_put_ordered_extent(ordered
);
6854 btrfs_put_ordered_extent(ordered
);
6857 inode_tree_del(inode
);
6858 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6860 btrfs_remove_delayed_node(inode
);
6861 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6864 int btrfs_drop_inode(struct inode
*inode
)
6866 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6868 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6869 !btrfs_is_free_space_inode(root
, inode
))
6872 return generic_drop_inode(inode
);
6875 static void init_once(void *foo
)
6877 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6879 inode_init_once(&ei
->vfs_inode
);
6882 void btrfs_destroy_cachep(void)
6884 if (btrfs_inode_cachep
)
6885 kmem_cache_destroy(btrfs_inode_cachep
);
6886 if (btrfs_trans_handle_cachep
)
6887 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6888 if (btrfs_transaction_cachep
)
6889 kmem_cache_destroy(btrfs_transaction_cachep
);
6890 if (btrfs_path_cachep
)
6891 kmem_cache_destroy(btrfs_path_cachep
);
6892 if (btrfs_free_space_cachep
)
6893 kmem_cache_destroy(btrfs_free_space_cachep
);
6896 int btrfs_init_cachep(void)
6898 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6899 sizeof(struct btrfs_inode
), 0,
6900 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6901 if (!btrfs_inode_cachep
)
6904 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6905 sizeof(struct btrfs_trans_handle
), 0,
6906 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6907 if (!btrfs_trans_handle_cachep
)
6910 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6911 sizeof(struct btrfs_transaction
), 0,
6912 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6913 if (!btrfs_transaction_cachep
)
6916 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6917 sizeof(struct btrfs_path
), 0,
6918 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6919 if (!btrfs_path_cachep
)
6922 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6923 sizeof(struct btrfs_free_space
), 0,
6924 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6925 if (!btrfs_free_space_cachep
)
6930 btrfs_destroy_cachep();
6934 static int btrfs_getattr(struct vfsmount
*mnt
,
6935 struct dentry
*dentry
, struct kstat
*stat
)
6937 struct inode
*inode
= dentry
->d_inode
;
6938 u32 blocksize
= inode
->i_sb
->s_blocksize
;
6940 generic_fillattr(inode
, stat
);
6941 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
6942 stat
->blksize
= PAGE_CACHE_SIZE
;
6943 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
6944 ALIGN(BTRFS_I(inode
)->delalloc_bytes
, blocksize
)) >> 9;
6949 * If a file is moved, it will inherit the cow and compression flags of the new
6952 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
6954 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
6955 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
6957 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
6958 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
6960 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
6962 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
)
6963 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
6965 b_inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6968 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6969 struct inode
*new_dir
, struct dentry
*new_dentry
)
6971 struct btrfs_trans_handle
*trans
;
6972 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6973 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6974 struct inode
*new_inode
= new_dentry
->d_inode
;
6975 struct inode
*old_inode
= old_dentry
->d_inode
;
6976 struct timespec ctime
= CURRENT_TIME
;
6980 u64 old_ino
= btrfs_ino(old_inode
);
6982 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
6985 /* we only allow rename subvolume link between subvolumes */
6986 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
6989 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
6990 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
6993 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
6994 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
6997 * we're using rename to replace one file with another.
6998 * and the replacement file is large. Start IO on it now so
6999 * we don't add too much work to the end of the transaction
7001 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
7002 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
7003 filemap_flush(old_inode
->i_mapping
);
7005 /* close the racy window with snapshot create/destroy ioctl */
7006 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7007 down_read(&root
->fs_info
->subvol_sem
);
7009 * We want to reserve the absolute worst case amount of items. So if
7010 * both inodes are subvols and we need to unlink them then that would
7011 * require 4 item modifications, but if they are both normal inodes it
7012 * would require 5 item modifications, so we'll assume their normal
7013 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7014 * should cover the worst case number of items we'll modify.
7016 trans
= btrfs_start_transaction(root
, 20);
7017 if (IS_ERR(trans
)) {
7018 ret
= PTR_ERR(trans
);
7023 btrfs_record_root_in_trans(trans
, dest
);
7025 ret
= btrfs_set_inode_index(new_dir
, &index
);
7029 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7030 /* force full log commit if subvolume involved. */
7031 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7033 ret
= btrfs_insert_inode_ref(trans
, dest
,
7034 new_dentry
->d_name
.name
,
7035 new_dentry
->d_name
.len
,
7037 btrfs_ino(new_dir
), index
);
7041 * this is an ugly little race, but the rename is required
7042 * to make sure that if we crash, the inode is either at the
7043 * old name or the new one. pinning the log transaction lets
7044 * us make sure we don't allow a log commit to come in after
7045 * we unlink the name but before we add the new name back in.
7047 btrfs_pin_log_trans(root
);
7050 * make sure the inode gets flushed if it is replacing
7053 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
7054 btrfs_add_ordered_operation(trans
, root
, old_inode
);
7056 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
7057 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
7058 old_inode
->i_ctime
= ctime
;
7060 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
7061 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
7063 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7064 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
7065 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
7066 old_dentry
->d_name
.name
,
7067 old_dentry
->d_name
.len
);
7069 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7070 old_dentry
->d_inode
,
7071 old_dentry
->d_name
.name
,
7072 old_dentry
->d_name
.len
);
7074 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7079 new_inode
->i_ctime
= CURRENT_TIME
;
7080 if (unlikely(btrfs_ino(new_inode
) ==
7081 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7082 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7083 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7085 new_dentry
->d_name
.name
,
7086 new_dentry
->d_name
.len
);
7087 BUG_ON(new_inode
->i_nlink
== 0);
7089 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7090 new_dentry
->d_inode
,
7091 new_dentry
->d_name
.name
,
7092 new_dentry
->d_name
.len
);
7095 if (new_inode
->i_nlink
== 0) {
7096 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7101 fixup_inode_flags(new_dir
, old_inode
);
7103 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7104 new_dentry
->d_name
.name
,
7105 new_dentry
->d_name
.len
, 0, index
);
7108 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7109 struct dentry
*parent
= new_dentry
->d_parent
;
7110 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7111 btrfs_end_log_trans(root
);
7114 btrfs_end_transaction(trans
, root
);
7116 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7117 up_read(&root
->fs_info
->subvol_sem
);
7123 * some fairly slow code that needs optimization. This walks the list
7124 * of all the inodes with pending delalloc and forces them to disk.
7126 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7128 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7129 struct btrfs_inode
*binode
;
7130 struct inode
*inode
;
7132 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7135 spin_lock(&root
->fs_info
->delalloc_lock
);
7136 while (!list_empty(head
)) {
7137 binode
= list_entry(head
->next
, struct btrfs_inode
,
7139 inode
= igrab(&binode
->vfs_inode
);
7141 list_del_init(&binode
->delalloc_inodes
);
7142 spin_unlock(&root
->fs_info
->delalloc_lock
);
7144 filemap_flush(inode
->i_mapping
);
7146 btrfs_add_delayed_iput(inode
);
7151 spin_lock(&root
->fs_info
->delalloc_lock
);
7153 spin_unlock(&root
->fs_info
->delalloc_lock
);
7155 /* the filemap_flush will queue IO into the worker threads, but
7156 * we have to make sure the IO is actually started and that
7157 * ordered extents get created before we return
7159 atomic_inc(&root
->fs_info
->async_submit_draining
);
7160 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7161 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7162 wait_event(root
->fs_info
->async_submit_wait
,
7163 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7164 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7166 atomic_dec(&root
->fs_info
->async_submit_draining
);
7170 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7171 const char *symname
)
7173 struct btrfs_trans_handle
*trans
;
7174 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7175 struct btrfs_path
*path
;
7176 struct btrfs_key key
;
7177 struct inode
*inode
= NULL
;
7185 struct btrfs_file_extent_item
*ei
;
7186 struct extent_buffer
*leaf
;
7187 unsigned long nr
= 0;
7189 name_len
= strlen(symname
) + 1;
7190 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7191 return -ENAMETOOLONG
;
7194 * 2 items for inode item and ref
7195 * 2 items for dir items
7196 * 1 item for xattr if selinux is on
7198 trans
= btrfs_start_transaction(root
, 5);
7200 return PTR_ERR(trans
);
7202 err
= btrfs_find_free_ino(root
, &objectid
);
7206 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7207 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
7208 S_IFLNK
|S_IRWXUGO
, &index
);
7209 if (IS_ERR(inode
)) {
7210 err
= PTR_ERR(inode
);
7214 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7221 * If the active LSM wants to access the inode during
7222 * d_instantiate it needs these. Smack checks to see
7223 * if the filesystem supports xattrs by looking at the
7226 inode
->i_fop
= &btrfs_file_operations
;
7227 inode
->i_op
= &btrfs_file_inode_operations
;
7229 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7233 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7234 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7235 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7240 path
= btrfs_alloc_path();
7246 key
.objectid
= btrfs_ino(inode
);
7248 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7249 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7250 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7254 btrfs_free_path(path
);
7257 leaf
= path
->nodes
[0];
7258 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7259 struct btrfs_file_extent_item
);
7260 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7261 btrfs_set_file_extent_type(leaf
, ei
,
7262 BTRFS_FILE_EXTENT_INLINE
);
7263 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7264 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7265 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7266 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7268 ptr
= btrfs_file_extent_inline_start(ei
);
7269 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7270 btrfs_mark_buffer_dirty(leaf
);
7271 btrfs_free_path(path
);
7273 inode
->i_op
= &btrfs_symlink_inode_operations
;
7274 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7275 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7276 inode_set_bytes(inode
, name_len
);
7277 btrfs_i_size_write(inode
, name_len
- 1);
7278 err
= btrfs_update_inode(trans
, root
, inode
);
7284 d_instantiate(dentry
, inode
);
7285 nr
= trans
->blocks_used
;
7286 btrfs_end_transaction(trans
, root
);
7288 inode_dec_link_count(inode
);
7291 btrfs_btree_balance_dirty(root
, nr
);
7295 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7296 u64 start
, u64 num_bytes
, u64 min_size
,
7297 loff_t actual_len
, u64
*alloc_hint
,
7298 struct btrfs_trans_handle
*trans
)
7300 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7301 struct btrfs_key ins
;
7302 u64 cur_offset
= start
;
7305 bool own_trans
= true;
7309 while (num_bytes
> 0) {
7311 trans
= btrfs_start_transaction(root
, 3);
7312 if (IS_ERR(trans
)) {
7313 ret
= PTR_ERR(trans
);
7318 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7319 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7322 btrfs_end_transaction(trans
, root
);
7326 ret
= insert_reserved_file_extent(trans
, inode
,
7327 cur_offset
, ins
.objectid
,
7328 ins
.offset
, ins
.offset
,
7329 ins
.offset
, 0, 0, 0,
7330 BTRFS_FILE_EXTENT_PREALLOC
);
7332 btrfs_drop_extent_cache(inode
, cur_offset
,
7333 cur_offset
+ ins
.offset
-1, 0);
7335 num_bytes
-= ins
.offset
;
7336 cur_offset
+= ins
.offset
;
7337 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7339 inode
->i_ctime
= CURRENT_TIME
;
7340 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7341 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7342 (actual_len
> inode
->i_size
) &&
7343 (cur_offset
> inode
->i_size
)) {
7344 if (cur_offset
> actual_len
)
7345 i_size
= actual_len
;
7347 i_size
= cur_offset
;
7348 i_size_write(inode
, i_size
);
7349 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7352 ret
= btrfs_update_inode(trans
, root
, inode
);
7356 btrfs_end_transaction(trans
, root
);
7361 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7362 u64 start
, u64 num_bytes
, u64 min_size
,
7363 loff_t actual_len
, u64
*alloc_hint
)
7365 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7366 min_size
, actual_len
, alloc_hint
,
7370 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7371 struct btrfs_trans_handle
*trans
, int mode
,
7372 u64 start
, u64 num_bytes
, u64 min_size
,
7373 loff_t actual_len
, u64
*alloc_hint
)
7375 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7376 min_size
, actual_len
, alloc_hint
, trans
);
7379 static int btrfs_set_page_dirty(struct page
*page
)
7381 return __set_page_dirty_nobuffers(page
);
7384 static int btrfs_permission(struct inode
*inode
, int mask
)
7386 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7387 umode_t mode
= inode
->i_mode
;
7389 if (mask
& MAY_WRITE
&&
7390 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
7391 if (btrfs_root_readonly(root
))
7393 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
7396 return generic_permission(inode
, mask
);
7399 static const struct inode_operations btrfs_dir_inode_operations
= {
7400 .getattr
= btrfs_getattr
,
7401 .lookup
= btrfs_lookup
,
7402 .create
= btrfs_create
,
7403 .unlink
= btrfs_unlink
,
7405 .mkdir
= btrfs_mkdir
,
7406 .rmdir
= btrfs_rmdir
,
7407 .rename
= btrfs_rename
,
7408 .symlink
= btrfs_symlink
,
7409 .setattr
= btrfs_setattr
,
7410 .mknod
= btrfs_mknod
,
7411 .setxattr
= btrfs_setxattr
,
7412 .getxattr
= btrfs_getxattr
,
7413 .listxattr
= btrfs_listxattr
,
7414 .removexattr
= btrfs_removexattr
,
7415 .permission
= btrfs_permission
,
7416 .get_acl
= btrfs_get_acl
,
7418 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7419 .lookup
= btrfs_lookup
,
7420 .permission
= btrfs_permission
,
7421 .get_acl
= btrfs_get_acl
,
7424 static const struct file_operations btrfs_dir_file_operations
= {
7425 .llseek
= generic_file_llseek
,
7426 .read
= generic_read_dir
,
7427 .readdir
= btrfs_real_readdir
,
7428 .unlocked_ioctl
= btrfs_ioctl
,
7429 #ifdef CONFIG_COMPAT
7430 .compat_ioctl
= btrfs_ioctl
,
7432 .release
= btrfs_release_file
,
7433 .fsync
= btrfs_sync_file
,
7436 static struct extent_io_ops btrfs_extent_io_ops
= {
7437 .fill_delalloc
= run_delalloc_range
,
7438 .submit_bio_hook
= btrfs_submit_bio_hook
,
7439 .merge_bio_hook
= btrfs_merge_bio_hook
,
7440 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7441 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7442 .writepage_start_hook
= btrfs_writepage_start_hook
,
7443 .set_bit_hook
= btrfs_set_bit_hook
,
7444 .clear_bit_hook
= btrfs_clear_bit_hook
,
7445 .merge_extent_hook
= btrfs_merge_extent_hook
,
7446 .split_extent_hook
= btrfs_split_extent_hook
,
7450 * btrfs doesn't support the bmap operation because swapfiles
7451 * use bmap to make a mapping of extents in the file. They assume
7452 * these extents won't change over the life of the file and they
7453 * use the bmap result to do IO directly to the drive.
7455 * the btrfs bmap call would return logical addresses that aren't
7456 * suitable for IO and they also will change frequently as COW
7457 * operations happen. So, swapfile + btrfs == corruption.
7459 * For now we're avoiding this by dropping bmap.
7461 static const struct address_space_operations btrfs_aops
= {
7462 .readpage
= btrfs_readpage
,
7463 .writepage
= btrfs_writepage
,
7464 .writepages
= btrfs_writepages
,
7465 .readpages
= btrfs_readpages
,
7466 .direct_IO
= btrfs_direct_IO
,
7467 .invalidatepage
= btrfs_invalidatepage
,
7468 .releasepage
= btrfs_releasepage
,
7469 .set_page_dirty
= btrfs_set_page_dirty
,
7470 .error_remove_page
= generic_error_remove_page
,
7473 static const struct address_space_operations btrfs_symlink_aops
= {
7474 .readpage
= btrfs_readpage
,
7475 .writepage
= btrfs_writepage
,
7476 .invalidatepage
= btrfs_invalidatepage
,
7477 .releasepage
= btrfs_releasepage
,
7480 static const struct inode_operations btrfs_file_inode_operations
= {
7481 .getattr
= btrfs_getattr
,
7482 .setattr
= btrfs_setattr
,
7483 .setxattr
= btrfs_setxattr
,
7484 .getxattr
= btrfs_getxattr
,
7485 .listxattr
= btrfs_listxattr
,
7486 .removexattr
= btrfs_removexattr
,
7487 .permission
= btrfs_permission
,
7488 .fiemap
= btrfs_fiemap
,
7489 .get_acl
= btrfs_get_acl
,
7491 static const struct inode_operations btrfs_special_inode_operations
= {
7492 .getattr
= btrfs_getattr
,
7493 .setattr
= btrfs_setattr
,
7494 .permission
= btrfs_permission
,
7495 .setxattr
= btrfs_setxattr
,
7496 .getxattr
= btrfs_getxattr
,
7497 .listxattr
= btrfs_listxattr
,
7498 .removexattr
= btrfs_removexattr
,
7499 .get_acl
= btrfs_get_acl
,
7501 static const struct inode_operations btrfs_symlink_inode_operations
= {
7502 .readlink
= generic_readlink
,
7503 .follow_link
= page_follow_link_light
,
7504 .put_link
= page_put_link
,
7505 .getattr
= btrfs_getattr
,
7506 .setattr
= btrfs_setattr
,
7507 .permission
= btrfs_permission
,
7508 .setxattr
= btrfs_setxattr
,
7509 .getxattr
= btrfs_getxattr
,
7510 .listxattr
= btrfs_listxattr
,
7511 .removexattr
= btrfs_removexattr
,
7512 .get_acl
= btrfs_get_acl
,
7515 const struct dentry_operations btrfs_dentry_operations
= {
7516 .d_delete
= btrfs_dentry_delete
,
7517 .d_release
= btrfs_dentry_release
,