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 btrfs_ordered_extent
*ordered_extent
);
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
,
157 leaf
= path
->nodes
[0];
158 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
159 struct btrfs_file_extent_item
);
160 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
161 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
162 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
163 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
164 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
165 ptr
= btrfs_file_extent_inline_start(ei
);
167 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
170 while (compressed_size
> 0) {
171 cpage
= compressed_pages
[i
];
172 cur_size
= min_t(unsigned long, compressed_size
,
175 kaddr
= kmap_atomic(cpage
);
176 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
177 kunmap_atomic(kaddr
);
181 compressed_size
-= cur_size
;
183 btrfs_set_file_extent_compression(leaf
, ei
,
186 page
= find_get_page(inode
->i_mapping
,
187 start
>> PAGE_CACHE_SHIFT
);
188 btrfs_set_file_extent_compression(leaf
, ei
, 0);
189 kaddr
= kmap_atomic(page
);
190 offset
= start
& (PAGE_CACHE_SIZE
- 1);
191 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
192 kunmap_atomic(kaddr
);
193 page_cache_release(page
);
195 btrfs_mark_buffer_dirty(leaf
);
196 btrfs_free_path(path
);
199 * we're an inline extent, so nobody can
200 * extend the file past i_size without locking
201 * a page we already have locked.
203 * We must do any isize and inode updates
204 * before we unlock the pages. Otherwise we
205 * could end up racing with unlink.
207 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
208 ret
= btrfs_update_inode(trans
, root
, inode
);
212 btrfs_free_path(path
);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
223 struct btrfs_root
*root
,
224 struct inode
*inode
, u64 start
, u64 end
,
225 size_t compressed_size
, int compress_type
,
226 struct page
**compressed_pages
)
228 u64 isize
= i_size_read(inode
);
229 u64 actual_end
= min(end
+ 1, isize
);
230 u64 inline_len
= actual_end
- start
;
231 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
232 ~((u64
)root
->sectorsize
- 1);
233 u64 data_len
= inline_len
;
237 data_len
= compressed_size
;
240 actual_end
>= PAGE_CACHE_SIZE
||
241 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
243 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
245 data_len
> root
->fs_info
->max_inline
) {
249 ret
= btrfs_drop_extents(trans
, root
, inode
, start
, aligned_end
, 1);
253 if (isize
> actual_end
)
254 inline_len
= min_t(u64
, isize
, actual_end
);
255 ret
= insert_inline_extent(trans
, root
, inode
, start
,
256 inline_len
, compressed_size
,
257 compress_type
, compressed_pages
);
258 if (ret
&& ret
!= -ENOSPC
) {
259 btrfs_abort_transaction(trans
, root
, ret
);
261 } else if (ret
== -ENOSPC
) {
265 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
266 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
270 struct async_extent
{
275 unsigned long nr_pages
;
277 struct list_head list
;
282 struct btrfs_root
*root
;
283 struct page
*locked_page
;
286 struct list_head extents
;
287 struct btrfs_work work
;
290 static noinline
int add_async_extent(struct async_cow
*cow
,
291 u64 start
, u64 ram_size
,
294 unsigned long nr_pages
,
297 struct async_extent
*async_extent
;
299 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
300 BUG_ON(!async_extent
); /* -ENOMEM */
301 async_extent
->start
= start
;
302 async_extent
->ram_size
= ram_size
;
303 async_extent
->compressed_size
= compressed_size
;
304 async_extent
->pages
= pages
;
305 async_extent
->nr_pages
= nr_pages
;
306 async_extent
->compress_type
= compress_type
;
307 list_add_tail(&async_extent
->list
, &cow
->extents
);
312 * we create compressed extents in two phases. The first
313 * phase compresses a range of pages that have already been
314 * locked (both pages and state bits are locked).
316 * This is done inside an ordered work queue, and the compression
317 * is spread across many cpus. The actual IO submission is step
318 * two, and the ordered work queue takes care of making sure that
319 * happens in the same order things were put onto the queue by
320 * writepages and friends.
322 * If this code finds it can't get good compression, it puts an
323 * entry onto the work queue to write the uncompressed bytes. This
324 * makes sure that both compressed inodes and uncompressed inodes
325 * are written in the same order that the flusher thread sent them
328 static noinline
int compress_file_range(struct inode
*inode
,
329 struct page
*locked_page
,
331 struct async_cow
*async_cow
,
334 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
335 struct btrfs_trans_handle
*trans
;
337 u64 blocksize
= root
->sectorsize
;
339 u64 isize
= i_size_read(inode
);
341 struct page
**pages
= NULL
;
342 unsigned long nr_pages
;
343 unsigned long nr_pages_ret
= 0;
344 unsigned long total_compressed
= 0;
345 unsigned long total_in
= 0;
346 unsigned long max_compressed
= 128 * 1024;
347 unsigned long max_uncompressed
= 128 * 1024;
350 int compress_type
= root
->fs_info
->compress_type
;
352 /* if this is a small write inside eof, kick off a defrag */
353 if ((end
- start
+ 1) < 16 * 1024 &&
354 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
355 btrfs_add_inode_defrag(NULL
, inode
);
357 actual_end
= min_t(u64
, isize
, end
+ 1);
360 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
361 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
364 * we don't want to send crud past the end of i_size through
365 * compression, that's just a waste of CPU time. So, if the
366 * end of the file is before the start of our current
367 * requested range of bytes, we bail out to the uncompressed
368 * cleanup code that can deal with all of this.
370 * It isn't really the fastest way to fix things, but this is a
371 * very uncommon corner.
373 if (actual_end
<= start
)
374 goto cleanup_and_bail_uncompressed
;
376 total_compressed
= actual_end
- start
;
378 /* we want to make sure that amount of ram required to uncompress
379 * an extent is reasonable, so we limit the total size in ram
380 * of a compressed extent to 128k. This is a crucial number
381 * because it also controls how easily we can spread reads across
382 * cpus for decompression.
384 * We also want to make sure the amount of IO required to do
385 * a random read is reasonably small, so we limit the size of
386 * a compressed extent to 128k.
388 total_compressed
= min(total_compressed
, max_uncompressed
);
389 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
390 num_bytes
= max(blocksize
, num_bytes
);
395 * we do compression for mount -o compress and when the
396 * inode has not been flagged as nocompress. This flag can
397 * change at any time if we discover bad compression ratios.
399 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
400 (btrfs_test_opt(root
, COMPRESS
) ||
401 (BTRFS_I(inode
)->force_compress
) ||
402 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
404 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
406 /* just bail out to the uncompressed code */
410 if (BTRFS_I(inode
)->force_compress
)
411 compress_type
= BTRFS_I(inode
)->force_compress
;
413 ret
= btrfs_compress_pages(compress_type
,
414 inode
->i_mapping
, start
,
415 total_compressed
, pages
,
416 nr_pages
, &nr_pages_ret
,
422 unsigned long offset
= total_compressed
&
423 (PAGE_CACHE_SIZE
- 1);
424 struct page
*page
= pages
[nr_pages_ret
- 1];
427 /* zero the tail end of the last page, we might be
428 * sending it down to disk
431 kaddr
= kmap_atomic(page
);
432 memset(kaddr
+ offset
, 0,
433 PAGE_CACHE_SIZE
- offset
);
434 kunmap_atomic(kaddr
);
441 trans
= btrfs_join_transaction(root
);
443 ret
= PTR_ERR(trans
);
445 goto cleanup_and_out
;
447 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
449 /* lets try to make an inline extent */
450 if (ret
|| total_in
< (actual_end
- start
)) {
451 /* we didn't compress the entire range, try
452 * to make an uncompressed inline extent.
454 ret
= cow_file_range_inline(trans
, root
, inode
,
455 start
, end
, 0, 0, NULL
);
457 /* try making a compressed inline extent */
458 ret
= cow_file_range_inline(trans
, root
, inode
,
461 compress_type
, pages
);
465 * inline extent creation worked or returned error,
466 * we don't need to create any more async work items.
467 * Unlock and free up our temp pages.
469 extent_clear_unlock_delalloc(inode
,
470 &BTRFS_I(inode
)->io_tree
,
472 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
473 EXTENT_CLEAR_DELALLOC
|
474 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
476 btrfs_end_transaction(trans
, root
);
479 btrfs_end_transaction(trans
, root
);
484 * we aren't doing an inline extent round the compressed size
485 * up to a block size boundary so the allocator does sane
488 total_compressed
= (total_compressed
+ blocksize
- 1) &
492 * one last check to make sure the compression is really a
493 * win, compare the page count read with the blocks on disk
495 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
496 ~(PAGE_CACHE_SIZE
- 1);
497 if (total_compressed
>= total_in
) {
500 num_bytes
= total_in
;
503 if (!will_compress
&& pages
) {
505 * the compression code ran but failed to make things smaller,
506 * free any pages it allocated and our page pointer array
508 for (i
= 0; i
< nr_pages_ret
; i
++) {
509 WARN_ON(pages
[i
]->mapping
);
510 page_cache_release(pages
[i
]);
514 total_compressed
= 0;
517 /* flag the file so we don't compress in the future */
518 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
519 !(BTRFS_I(inode
)->force_compress
)) {
520 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
526 /* the async work queues will take care of doing actual
527 * allocation on disk for these compressed pages,
528 * and will submit them to the elevator.
530 add_async_extent(async_cow
, start
, num_bytes
,
531 total_compressed
, pages
, nr_pages_ret
,
534 if (start
+ num_bytes
< end
) {
541 cleanup_and_bail_uncompressed
:
543 * No compression, but we still need to write the pages in
544 * the file we've been given so far. redirty the locked
545 * page if it corresponds to our extent and set things up
546 * for the async work queue to run cow_file_range to do
547 * the normal delalloc dance
549 if (page_offset(locked_page
) >= start
&&
550 page_offset(locked_page
) <= end
) {
551 __set_page_dirty_nobuffers(locked_page
);
552 /* unlocked later on in the async handlers */
554 add_async_extent(async_cow
, start
, end
- start
+ 1,
555 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
563 for (i
= 0; i
< nr_pages_ret
; i
++) {
564 WARN_ON(pages
[i
]->mapping
);
565 page_cache_release(pages
[i
]);
572 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
574 EXTENT_CLEAR_UNLOCK_PAGE
|
576 EXTENT_CLEAR_DELALLOC
|
577 EXTENT_SET_WRITEBACK
|
578 EXTENT_END_WRITEBACK
);
579 if (!trans
|| IS_ERR(trans
))
580 btrfs_error(root
->fs_info
, ret
, "Failed to join transaction");
582 btrfs_abort_transaction(trans
, root
, ret
);
587 * phase two of compressed writeback. This is the ordered portion
588 * of the code, which only gets called in the order the work was
589 * queued. We walk all the async extents created by compress_file_range
590 * and send them down to the disk.
592 static noinline
int submit_compressed_extents(struct inode
*inode
,
593 struct async_cow
*async_cow
)
595 struct async_extent
*async_extent
;
597 struct btrfs_trans_handle
*trans
;
598 struct btrfs_key ins
;
599 struct extent_map
*em
;
600 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
601 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
602 struct extent_io_tree
*io_tree
;
605 if (list_empty(&async_cow
->extents
))
609 while (!list_empty(&async_cow
->extents
)) {
610 async_extent
= list_entry(async_cow
->extents
.next
,
611 struct async_extent
, list
);
612 list_del(&async_extent
->list
);
614 io_tree
= &BTRFS_I(inode
)->io_tree
;
617 /* did the compression code fall back to uncompressed IO? */
618 if (!async_extent
->pages
) {
619 int page_started
= 0;
620 unsigned long nr_written
= 0;
622 lock_extent(io_tree
, async_extent
->start
,
623 async_extent
->start
+
624 async_extent
->ram_size
- 1);
626 /* allocate blocks */
627 ret
= cow_file_range(inode
, async_cow
->locked_page
,
629 async_extent
->start
+
630 async_extent
->ram_size
- 1,
631 &page_started
, &nr_written
, 0);
636 * if page_started, cow_file_range inserted an
637 * inline extent and took care of all the unlocking
638 * and IO for us. Otherwise, we need to submit
639 * all those pages down to the drive.
641 if (!page_started
&& !ret
)
642 extent_write_locked_range(io_tree
,
643 inode
, async_extent
->start
,
644 async_extent
->start
+
645 async_extent
->ram_size
- 1,
653 lock_extent(io_tree
, async_extent
->start
,
654 async_extent
->start
+ async_extent
->ram_size
- 1);
656 trans
= btrfs_join_transaction(root
);
658 ret
= PTR_ERR(trans
);
660 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
661 ret
= btrfs_reserve_extent(trans
, root
,
662 async_extent
->compressed_size
,
663 async_extent
->compressed_size
,
664 0, alloc_hint
, &ins
, 1);
665 if (ret
&& ret
!= -ENOSPC
)
666 btrfs_abort_transaction(trans
, root
, ret
);
667 btrfs_end_transaction(trans
, root
);
672 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
673 WARN_ON(async_extent
->pages
[i
]->mapping
);
674 page_cache_release(async_extent
->pages
[i
]);
676 kfree(async_extent
->pages
);
677 async_extent
->nr_pages
= 0;
678 async_extent
->pages
= NULL
;
679 unlock_extent(io_tree
, async_extent
->start
,
680 async_extent
->start
+
681 async_extent
->ram_size
- 1);
684 goto out_free
; /* JDM: Requeue? */
688 * here we're doing allocation and writeback of the
691 btrfs_drop_extent_cache(inode
, async_extent
->start
,
692 async_extent
->start
+
693 async_extent
->ram_size
- 1, 0);
695 em
= alloc_extent_map();
696 BUG_ON(!em
); /* -ENOMEM */
697 em
->start
= async_extent
->start
;
698 em
->len
= async_extent
->ram_size
;
699 em
->orig_start
= em
->start
;
701 em
->block_start
= ins
.objectid
;
702 em
->block_len
= ins
.offset
;
703 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
704 em
->compress_type
= async_extent
->compress_type
;
705 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
706 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
709 write_lock(&em_tree
->lock
);
710 ret
= add_extent_mapping(em_tree
, em
);
711 write_unlock(&em_tree
->lock
);
712 if (ret
!= -EEXIST
) {
716 btrfs_drop_extent_cache(inode
, async_extent
->start
,
717 async_extent
->start
+
718 async_extent
->ram_size
- 1, 0);
721 ret
= btrfs_add_ordered_extent_compress(inode
,
724 async_extent
->ram_size
,
726 BTRFS_ORDERED_COMPRESSED
,
727 async_extent
->compress_type
);
728 BUG_ON(ret
); /* -ENOMEM */
731 * clear dirty, set writeback and unlock the pages.
733 extent_clear_unlock_delalloc(inode
,
734 &BTRFS_I(inode
)->io_tree
,
736 async_extent
->start
+
737 async_extent
->ram_size
- 1,
738 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
739 EXTENT_CLEAR_UNLOCK
|
740 EXTENT_CLEAR_DELALLOC
|
741 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
743 ret
= btrfs_submit_compressed_write(inode
,
745 async_extent
->ram_size
,
747 ins
.offset
, async_extent
->pages
,
748 async_extent
->nr_pages
);
750 BUG_ON(ret
); /* -ENOMEM */
751 alloc_hint
= ins
.objectid
+ ins
.offset
;
763 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
766 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
767 struct extent_map
*em
;
770 read_lock(&em_tree
->lock
);
771 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
774 * if block start isn't an actual block number then find the
775 * first block in this inode and use that as a hint. If that
776 * block is also bogus then just don't worry about it.
778 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
780 em
= search_extent_mapping(em_tree
, 0, 0);
781 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
782 alloc_hint
= em
->block_start
;
786 alloc_hint
= em
->block_start
;
790 read_unlock(&em_tree
->lock
);
796 * when extent_io.c finds a delayed allocation range in the file,
797 * the call backs end up in this code. The basic idea is to
798 * allocate extents on disk for the range, and create ordered data structs
799 * in ram to track those extents.
801 * locked_page is the page that writepage had locked already. We use
802 * it to make sure we don't do extra locks or unlocks.
804 * *page_started is set to one if we unlock locked_page and do everything
805 * required to start IO on it. It may be clean and already done with
808 static noinline
int cow_file_range(struct inode
*inode
,
809 struct page
*locked_page
,
810 u64 start
, u64 end
, int *page_started
,
811 unsigned long *nr_written
,
814 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
815 struct btrfs_trans_handle
*trans
;
818 unsigned long ram_size
;
821 u64 blocksize
= root
->sectorsize
;
822 struct btrfs_key ins
;
823 struct extent_map
*em
;
824 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
827 BUG_ON(btrfs_is_free_space_inode(inode
));
828 trans
= btrfs_join_transaction(root
);
830 extent_clear_unlock_delalloc(inode
,
831 &BTRFS_I(inode
)->io_tree
,
832 start
, end
, locked_page
,
833 EXTENT_CLEAR_UNLOCK_PAGE
|
834 EXTENT_CLEAR_UNLOCK
|
835 EXTENT_CLEAR_DELALLOC
|
837 EXTENT_SET_WRITEBACK
|
838 EXTENT_END_WRITEBACK
);
839 return PTR_ERR(trans
);
841 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
843 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
844 num_bytes
= max(blocksize
, num_bytes
);
845 disk_num_bytes
= num_bytes
;
848 /* if this is a small write inside eof, kick off defrag */
849 if (num_bytes
< 64 * 1024 &&
850 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
851 btrfs_add_inode_defrag(trans
, inode
);
854 /* lets try to make an inline extent */
855 ret
= cow_file_range_inline(trans
, root
, inode
,
856 start
, end
, 0, 0, NULL
);
858 extent_clear_unlock_delalloc(inode
,
859 &BTRFS_I(inode
)->io_tree
,
861 EXTENT_CLEAR_UNLOCK_PAGE
|
862 EXTENT_CLEAR_UNLOCK
|
863 EXTENT_CLEAR_DELALLOC
|
865 EXTENT_SET_WRITEBACK
|
866 EXTENT_END_WRITEBACK
);
868 *nr_written
= *nr_written
+
869 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
872 } else if (ret
< 0) {
873 btrfs_abort_transaction(trans
, root
, ret
);
878 BUG_ON(disk_num_bytes
>
879 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
881 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
882 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
884 while (disk_num_bytes
> 0) {
887 cur_alloc_size
= disk_num_bytes
;
888 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
889 root
->sectorsize
, 0, alloc_hint
,
892 btrfs_abort_transaction(trans
, root
, ret
);
896 em
= alloc_extent_map();
897 BUG_ON(!em
); /* -ENOMEM */
899 em
->orig_start
= em
->start
;
900 ram_size
= ins
.offset
;
901 em
->len
= ins
.offset
;
903 em
->block_start
= ins
.objectid
;
904 em
->block_len
= ins
.offset
;
905 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
906 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
909 write_lock(&em_tree
->lock
);
910 ret
= add_extent_mapping(em_tree
, em
);
911 write_unlock(&em_tree
->lock
);
912 if (ret
!= -EEXIST
) {
916 btrfs_drop_extent_cache(inode
, start
,
917 start
+ ram_size
- 1, 0);
920 cur_alloc_size
= ins
.offset
;
921 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
922 ram_size
, cur_alloc_size
, 0);
923 BUG_ON(ret
); /* -ENOMEM */
925 if (root
->root_key
.objectid
==
926 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
927 ret
= btrfs_reloc_clone_csums(inode
, start
,
930 btrfs_abort_transaction(trans
, root
, ret
);
935 if (disk_num_bytes
< cur_alloc_size
)
938 /* we're not doing compressed IO, don't unlock the first
939 * page (which the caller expects to stay locked), don't
940 * clear any dirty bits and don't set any writeback bits
942 * Do set the Private2 bit so we know this page was properly
943 * setup for writepage
945 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
946 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
949 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
950 start
, start
+ ram_size
- 1,
952 disk_num_bytes
-= cur_alloc_size
;
953 num_bytes
-= cur_alloc_size
;
954 alloc_hint
= ins
.objectid
+ ins
.offset
;
955 start
+= cur_alloc_size
;
959 btrfs_end_transaction(trans
, root
);
963 extent_clear_unlock_delalloc(inode
,
964 &BTRFS_I(inode
)->io_tree
,
965 start
, end
, locked_page
,
966 EXTENT_CLEAR_UNLOCK_PAGE
|
967 EXTENT_CLEAR_UNLOCK
|
968 EXTENT_CLEAR_DELALLOC
|
970 EXTENT_SET_WRITEBACK
|
971 EXTENT_END_WRITEBACK
);
977 * work queue call back to started compression on a file and pages
979 static noinline
void async_cow_start(struct btrfs_work
*work
)
981 struct async_cow
*async_cow
;
983 async_cow
= container_of(work
, struct async_cow
, work
);
985 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
986 async_cow
->start
, async_cow
->end
, async_cow
,
988 if (num_added
== 0) {
989 btrfs_add_delayed_iput(async_cow
->inode
);
990 async_cow
->inode
= NULL
;
995 * work queue call back to submit previously compressed pages
997 static noinline
void async_cow_submit(struct btrfs_work
*work
)
999 struct async_cow
*async_cow
;
1000 struct btrfs_root
*root
;
1001 unsigned long nr_pages
;
1003 async_cow
= container_of(work
, struct async_cow
, work
);
1005 root
= async_cow
->root
;
1006 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1009 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1011 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1012 wake_up(&root
->fs_info
->async_submit_wait
);
1014 if (async_cow
->inode
)
1015 submit_compressed_extents(async_cow
->inode
, async_cow
);
1018 static noinline
void async_cow_free(struct btrfs_work
*work
)
1020 struct async_cow
*async_cow
;
1021 async_cow
= container_of(work
, struct async_cow
, work
);
1022 if (async_cow
->inode
)
1023 btrfs_add_delayed_iput(async_cow
->inode
);
1027 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1028 u64 start
, u64 end
, int *page_started
,
1029 unsigned long *nr_written
)
1031 struct async_cow
*async_cow
;
1032 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1033 unsigned long nr_pages
;
1035 int limit
= 10 * 1024 * 1024;
1037 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1038 1, 0, NULL
, GFP_NOFS
);
1039 while (start
< end
) {
1040 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1041 BUG_ON(!async_cow
); /* -ENOMEM */
1042 async_cow
->inode
= igrab(inode
);
1043 async_cow
->root
= root
;
1044 async_cow
->locked_page
= locked_page
;
1045 async_cow
->start
= start
;
1047 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
1050 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1052 async_cow
->end
= cur_end
;
1053 INIT_LIST_HEAD(&async_cow
->extents
);
1055 async_cow
->work
.func
= async_cow_start
;
1056 async_cow
->work
.ordered_func
= async_cow_submit
;
1057 async_cow
->work
.ordered_free
= async_cow_free
;
1058 async_cow
->work
.flags
= 0;
1060 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1062 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1064 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
1067 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1068 wait_event(root
->fs_info
->async_submit_wait
,
1069 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1073 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1074 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1075 wait_event(root
->fs_info
->async_submit_wait
,
1076 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1080 *nr_written
+= nr_pages
;
1081 start
= cur_end
+ 1;
1087 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1088 u64 bytenr
, u64 num_bytes
)
1091 struct btrfs_ordered_sum
*sums
;
1094 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1095 bytenr
+ num_bytes
- 1, &list
, 0);
1096 if (ret
== 0 && list_empty(&list
))
1099 while (!list_empty(&list
)) {
1100 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1101 list_del(&sums
->list
);
1108 * when nowcow writeback call back. This checks for snapshots or COW copies
1109 * of the extents that exist in the file, and COWs the file as required.
1111 * If no cow copies or snapshots exist, we write directly to the existing
1114 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1115 struct page
*locked_page
,
1116 u64 start
, u64 end
, int *page_started
, int force
,
1117 unsigned long *nr_written
)
1119 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1120 struct btrfs_trans_handle
*trans
;
1121 struct extent_buffer
*leaf
;
1122 struct btrfs_path
*path
;
1123 struct btrfs_file_extent_item
*fi
;
1124 struct btrfs_key found_key
;
1137 u64 ino
= btrfs_ino(inode
);
1139 path
= btrfs_alloc_path();
1141 extent_clear_unlock_delalloc(inode
,
1142 &BTRFS_I(inode
)->io_tree
,
1143 start
, end
, locked_page
,
1144 EXTENT_CLEAR_UNLOCK_PAGE
|
1145 EXTENT_CLEAR_UNLOCK
|
1146 EXTENT_CLEAR_DELALLOC
|
1147 EXTENT_CLEAR_DIRTY
|
1148 EXTENT_SET_WRITEBACK
|
1149 EXTENT_END_WRITEBACK
);
1153 nolock
= btrfs_is_free_space_inode(inode
);
1156 trans
= btrfs_join_transaction_nolock(root
);
1158 trans
= btrfs_join_transaction(root
);
1160 if (IS_ERR(trans
)) {
1161 extent_clear_unlock_delalloc(inode
,
1162 &BTRFS_I(inode
)->io_tree
,
1163 start
, end
, locked_page
,
1164 EXTENT_CLEAR_UNLOCK_PAGE
|
1165 EXTENT_CLEAR_UNLOCK
|
1166 EXTENT_CLEAR_DELALLOC
|
1167 EXTENT_CLEAR_DIRTY
|
1168 EXTENT_SET_WRITEBACK
|
1169 EXTENT_END_WRITEBACK
);
1170 btrfs_free_path(path
);
1171 return PTR_ERR(trans
);
1174 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1176 cow_start
= (u64
)-1;
1179 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1182 btrfs_abort_transaction(trans
, root
, ret
);
1185 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1186 leaf
= path
->nodes
[0];
1187 btrfs_item_key_to_cpu(leaf
, &found_key
,
1188 path
->slots
[0] - 1);
1189 if (found_key
.objectid
== ino
&&
1190 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1195 leaf
= path
->nodes
[0];
1196 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1197 ret
= btrfs_next_leaf(root
, path
);
1199 btrfs_abort_transaction(trans
, root
, ret
);
1204 leaf
= path
->nodes
[0];
1210 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1212 if (found_key
.objectid
> ino
||
1213 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1214 found_key
.offset
> end
)
1217 if (found_key
.offset
> cur_offset
) {
1218 extent_end
= found_key
.offset
;
1223 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1224 struct btrfs_file_extent_item
);
1225 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1227 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1228 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1229 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1230 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1231 extent_end
= found_key
.offset
+
1232 btrfs_file_extent_num_bytes(leaf
, fi
);
1233 if (extent_end
<= start
) {
1237 if (disk_bytenr
== 0)
1239 if (btrfs_file_extent_compression(leaf
, fi
) ||
1240 btrfs_file_extent_encryption(leaf
, fi
) ||
1241 btrfs_file_extent_other_encoding(leaf
, fi
))
1243 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1245 if (btrfs_extent_readonly(root
, disk_bytenr
))
1247 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1249 extent_offset
, disk_bytenr
))
1251 disk_bytenr
+= extent_offset
;
1252 disk_bytenr
+= cur_offset
- found_key
.offset
;
1253 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1255 * force cow if csum exists in the range.
1256 * this ensure that csum for a given extent are
1257 * either valid or do not exist.
1259 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1262 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1263 extent_end
= found_key
.offset
+
1264 btrfs_file_extent_inline_len(leaf
, fi
);
1265 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1270 if (extent_end
<= start
) {
1275 if (cow_start
== (u64
)-1)
1276 cow_start
= cur_offset
;
1277 cur_offset
= extent_end
;
1278 if (cur_offset
> end
)
1284 btrfs_release_path(path
);
1285 if (cow_start
!= (u64
)-1) {
1286 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1287 found_key
.offset
- 1, page_started
,
1290 btrfs_abort_transaction(trans
, root
, ret
);
1293 cow_start
= (u64
)-1;
1296 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1297 struct extent_map
*em
;
1298 struct extent_map_tree
*em_tree
;
1299 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1300 em
= alloc_extent_map();
1301 BUG_ON(!em
); /* -ENOMEM */
1302 em
->start
= cur_offset
;
1303 em
->orig_start
= em
->start
;
1304 em
->len
= num_bytes
;
1305 em
->block_len
= num_bytes
;
1306 em
->block_start
= disk_bytenr
;
1307 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1308 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1309 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
1311 write_lock(&em_tree
->lock
);
1312 ret
= add_extent_mapping(em_tree
, em
);
1313 write_unlock(&em_tree
->lock
);
1314 if (ret
!= -EEXIST
) {
1315 free_extent_map(em
);
1318 btrfs_drop_extent_cache(inode
, em
->start
,
1319 em
->start
+ em
->len
- 1, 0);
1321 type
= BTRFS_ORDERED_PREALLOC
;
1323 type
= BTRFS_ORDERED_NOCOW
;
1326 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1327 num_bytes
, num_bytes
, type
);
1328 BUG_ON(ret
); /* -ENOMEM */
1330 if (root
->root_key
.objectid
==
1331 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1332 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1335 btrfs_abort_transaction(trans
, root
, ret
);
1340 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1341 cur_offset
, cur_offset
+ num_bytes
- 1,
1342 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1343 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1344 EXTENT_SET_PRIVATE2
);
1345 cur_offset
= extent_end
;
1346 if (cur_offset
> end
)
1349 btrfs_release_path(path
);
1351 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1352 cow_start
= cur_offset
;
1356 if (cow_start
!= (u64
)-1) {
1357 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1358 page_started
, nr_written
, 1);
1360 btrfs_abort_transaction(trans
, root
, ret
);
1366 err
= btrfs_end_transaction(trans
, root
);
1370 if (ret
&& cur_offset
< end
)
1371 extent_clear_unlock_delalloc(inode
,
1372 &BTRFS_I(inode
)->io_tree
,
1373 cur_offset
, end
, locked_page
,
1374 EXTENT_CLEAR_UNLOCK_PAGE
|
1375 EXTENT_CLEAR_UNLOCK
|
1376 EXTENT_CLEAR_DELALLOC
|
1377 EXTENT_CLEAR_DIRTY
|
1378 EXTENT_SET_WRITEBACK
|
1379 EXTENT_END_WRITEBACK
);
1381 btrfs_free_path(path
);
1386 * extent_io.c call back to do delayed allocation processing
1388 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1389 u64 start
, u64 end
, int *page_started
,
1390 unsigned long *nr_written
)
1393 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1395 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) {
1396 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1397 page_started
, 1, nr_written
);
1398 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
) {
1399 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1400 page_started
, 0, nr_written
);
1401 } else if (!btrfs_test_opt(root
, COMPRESS
) &&
1402 !(BTRFS_I(inode
)->force_compress
) &&
1403 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
)) {
1404 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1405 page_started
, nr_written
, 1);
1407 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1408 &BTRFS_I(inode
)->runtime_flags
);
1409 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1410 page_started
, nr_written
);
1415 static void btrfs_split_extent_hook(struct inode
*inode
,
1416 struct extent_state
*orig
, u64 split
)
1418 /* not delalloc, ignore it */
1419 if (!(orig
->state
& EXTENT_DELALLOC
))
1422 spin_lock(&BTRFS_I(inode
)->lock
);
1423 BTRFS_I(inode
)->outstanding_extents
++;
1424 spin_unlock(&BTRFS_I(inode
)->lock
);
1428 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1429 * extents so we can keep track of new extents that are just merged onto old
1430 * extents, such as when we are doing sequential writes, so we can properly
1431 * account for the metadata space we'll need.
1433 static void btrfs_merge_extent_hook(struct inode
*inode
,
1434 struct extent_state
*new,
1435 struct extent_state
*other
)
1437 /* not delalloc, ignore it */
1438 if (!(other
->state
& EXTENT_DELALLOC
))
1441 spin_lock(&BTRFS_I(inode
)->lock
);
1442 BTRFS_I(inode
)->outstanding_extents
--;
1443 spin_unlock(&BTRFS_I(inode
)->lock
);
1447 * extent_io.c set_bit_hook, used to track delayed allocation
1448 * bytes in this file, and to maintain the list of inodes that
1449 * have pending delalloc work to be done.
1451 static void btrfs_set_bit_hook(struct inode
*inode
,
1452 struct extent_state
*state
, int *bits
)
1456 * set_bit and clear bit hooks normally require _irqsave/restore
1457 * but in this case, we are only testing for the DELALLOC
1458 * bit, which is only set or cleared with irqs on
1460 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1461 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1462 u64 len
= state
->end
+ 1 - state
->start
;
1463 bool do_list
= !btrfs_is_free_space_inode(inode
);
1465 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1466 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1468 spin_lock(&BTRFS_I(inode
)->lock
);
1469 BTRFS_I(inode
)->outstanding_extents
++;
1470 spin_unlock(&BTRFS_I(inode
)->lock
);
1473 spin_lock(&root
->fs_info
->delalloc_lock
);
1474 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1475 root
->fs_info
->delalloc_bytes
+= len
;
1476 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1477 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1478 &root
->fs_info
->delalloc_inodes
);
1480 spin_unlock(&root
->fs_info
->delalloc_lock
);
1485 * extent_io.c clear_bit_hook, see set_bit_hook for why
1487 static void btrfs_clear_bit_hook(struct inode
*inode
,
1488 struct extent_state
*state
, int *bits
)
1491 * set_bit and clear bit hooks normally require _irqsave/restore
1492 * but in this case, we are only testing for the DELALLOC
1493 * bit, which is only set or cleared with irqs on
1495 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1496 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1497 u64 len
= state
->end
+ 1 - state
->start
;
1498 bool do_list
= !btrfs_is_free_space_inode(inode
);
1500 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1501 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1502 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1503 spin_lock(&BTRFS_I(inode
)->lock
);
1504 BTRFS_I(inode
)->outstanding_extents
--;
1505 spin_unlock(&BTRFS_I(inode
)->lock
);
1508 if (*bits
& EXTENT_DO_ACCOUNTING
)
1509 btrfs_delalloc_release_metadata(inode
, len
);
1511 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1513 btrfs_free_reserved_data_space(inode
, len
);
1515 spin_lock(&root
->fs_info
->delalloc_lock
);
1516 root
->fs_info
->delalloc_bytes
-= len
;
1517 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1519 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1520 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1521 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1523 spin_unlock(&root
->fs_info
->delalloc_lock
);
1528 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1529 * we don't create bios that span stripes or chunks
1531 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1532 size_t size
, struct bio
*bio
,
1533 unsigned long bio_flags
)
1535 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1536 struct btrfs_mapping_tree
*map_tree
;
1537 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1542 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1545 length
= bio
->bi_size
;
1546 map_tree
= &root
->fs_info
->mapping_tree
;
1547 map_length
= length
;
1548 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1549 &map_length
, NULL
, 0);
1550 /* Will always return 0 or 1 with map_multi == NULL */
1552 if (map_length
< length
+ size
)
1558 * in order to insert checksums into the metadata in large chunks,
1559 * we wait until bio submission time. All the pages in the bio are
1560 * checksummed and sums are attached onto the ordered extent record.
1562 * At IO completion time the cums attached on the ordered extent record
1563 * are inserted into the btree
1565 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1566 struct bio
*bio
, int mirror_num
,
1567 unsigned long bio_flags
,
1570 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1573 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1574 BUG_ON(ret
); /* -ENOMEM */
1579 * in order to insert checksums into the metadata in large chunks,
1580 * we wait until bio submission time. All the pages in the bio are
1581 * checksummed and sums are attached onto the ordered extent record.
1583 * At IO completion time the cums attached on the ordered extent record
1584 * are inserted into the btree
1586 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1587 int mirror_num
, unsigned long bio_flags
,
1590 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1591 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1595 * extent_io.c submission hook. This does the right thing for csum calculation
1596 * on write, or reading the csums from the tree before a read
1598 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1599 int mirror_num
, unsigned long bio_flags
,
1602 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1607 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1609 if (btrfs_is_free_space_inode(inode
))
1612 if (!(rw
& REQ_WRITE
)) {
1613 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1617 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1618 return btrfs_submit_compressed_read(inode
, bio
,
1619 mirror_num
, bio_flags
);
1620 } else if (!skip_sum
) {
1621 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1626 } else if (!skip_sum
) {
1627 /* csum items have already been cloned */
1628 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1630 /* we're doing a write, do the async checksumming */
1631 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1632 inode
, rw
, bio
, mirror_num
,
1633 bio_flags
, bio_offset
,
1634 __btrfs_submit_bio_start
,
1635 __btrfs_submit_bio_done
);
1639 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1643 * given a list of ordered sums record them in the inode. This happens
1644 * at IO completion time based on sums calculated at bio submission time.
1646 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1647 struct inode
*inode
, u64 file_offset
,
1648 struct list_head
*list
)
1650 struct btrfs_ordered_sum
*sum
;
1652 list_for_each_entry(sum
, list
, list
) {
1653 btrfs_csum_file_blocks(trans
,
1654 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1659 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1660 struct extent_state
**cached_state
)
1662 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1664 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1665 cached_state
, GFP_NOFS
);
1668 /* see btrfs_writepage_start_hook for details on why this is required */
1669 struct btrfs_writepage_fixup
{
1671 struct btrfs_work work
;
1674 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1676 struct btrfs_writepage_fixup
*fixup
;
1677 struct btrfs_ordered_extent
*ordered
;
1678 struct extent_state
*cached_state
= NULL
;
1680 struct inode
*inode
;
1685 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1689 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1690 ClearPageChecked(page
);
1694 inode
= page
->mapping
->host
;
1695 page_start
= page_offset(page
);
1696 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1698 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1701 /* already ordered? We're done */
1702 if (PagePrivate2(page
))
1705 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1707 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1708 page_end
, &cached_state
, GFP_NOFS
);
1710 btrfs_start_ordered_extent(inode
, ordered
, 1);
1711 btrfs_put_ordered_extent(ordered
);
1715 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1717 mapping_set_error(page
->mapping
, ret
);
1718 end_extent_writepage(page
, ret
, page_start
, page_end
);
1719 ClearPageChecked(page
);
1723 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1724 ClearPageChecked(page
);
1725 set_page_dirty(page
);
1727 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1728 &cached_state
, GFP_NOFS
);
1731 page_cache_release(page
);
1736 * There are a few paths in the higher layers of the kernel that directly
1737 * set the page dirty bit without asking the filesystem if it is a
1738 * good idea. This causes problems because we want to make sure COW
1739 * properly happens and the data=ordered rules are followed.
1741 * In our case any range that doesn't have the ORDERED bit set
1742 * hasn't been properly setup for IO. We kick off an async process
1743 * to fix it up. The async helper will wait for ordered extents, set
1744 * the delalloc bit and make it safe to write the page.
1746 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1748 struct inode
*inode
= page
->mapping
->host
;
1749 struct btrfs_writepage_fixup
*fixup
;
1750 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1752 /* this page is properly in the ordered list */
1753 if (TestClearPagePrivate2(page
))
1756 if (PageChecked(page
))
1759 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1763 SetPageChecked(page
);
1764 page_cache_get(page
);
1765 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1767 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1771 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1772 struct inode
*inode
, u64 file_pos
,
1773 u64 disk_bytenr
, u64 disk_num_bytes
,
1774 u64 num_bytes
, u64 ram_bytes
,
1775 u8 compression
, u8 encryption
,
1776 u16 other_encoding
, int extent_type
)
1778 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1779 struct btrfs_file_extent_item
*fi
;
1780 struct btrfs_path
*path
;
1781 struct extent_buffer
*leaf
;
1782 struct btrfs_key ins
;
1785 path
= btrfs_alloc_path();
1789 path
->leave_spinning
= 1;
1792 * we may be replacing one extent in the tree with another.
1793 * The new extent is pinned in the extent map, and we don't want
1794 * to drop it from the cache until it is completely in the btree.
1796 * So, tell btrfs_drop_extents to leave this extent in the cache.
1797 * the caller is expected to unpin it and allow it to be merged
1800 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1801 file_pos
+ num_bytes
, 0);
1805 ins
.objectid
= btrfs_ino(inode
);
1806 ins
.offset
= file_pos
;
1807 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1808 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1811 leaf
= path
->nodes
[0];
1812 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1813 struct btrfs_file_extent_item
);
1814 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1815 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1816 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1817 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1818 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1819 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1820 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1821 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1822 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1823 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1825 btrfs_mark_buffer_dirty(leaf
);
1826 btrfs_release_path(path
);
1828 inode_add_bytes(inode
, num_bytes
);
1830 ins
.objectid
= disk_bytenr
;
1831 ins
.offset
= disk_num_bytes
;
1832 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1833 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1834 root
->root_key
.objectid
,
1835 btrfs_ino(inode
), file_pos
, &ins
);
1837 btrfs_free_path(path
);
1843 * helper function for btrfs_finish_ordered_io, this
1844 * just reads in some of the csum leaves to prime them into ram
1845 * before we start the transaction. It limits the amount of btree
1846 * reads required while inside the transaction.
1848 /* as ordered data IO finishes, this gets called so we can finish
1849 * an ordered extent if the range of bytes in the file it covers are
1852 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
1854 struct inode
*inode
= ordered_extent
->inode
;
1855 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1856 struct btrfs_trans_handle
*trans
= NULL
;
1857 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1858 struct extent_state
*cached_state
= NULL
;
1859 int compress_type
= 0;
1863 nolock
= btrfs_is_free_space_inode(inode
);
1865 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
1870 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1871 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
1872 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1875 trans
= btrfs_join_transaction_nolock(root
);
1877 trans
= btrfs_join_transaction(root
);
1878 if (IS_ERR(trans
)) {
1879 ret
= PTR_ERR(trans
);
1883 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1884 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
1885 if (ret
) /* -ENOMEM or corruption */
1886 btrfs_abort_transaction(trans
, root
, ret
);
1891 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1892 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1896 trans
= btrfs_join_transaction_nolock(root
);
1898 trans
= btrfs_join_transaction(root
);
1899 if (IS_ERR(trans
)) {
1900 ret
= PTR_ERR(trans
);
1904 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1906 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1907 compress_type
= ordered_extent
->compress_type
;
1908 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1909 BUG_ON(compress_type
);
1910 ret
= btrfs_mark_extent_written(trans
, inode
,
1911 ordered_extent
->file_offset
,
1912 ordered_extent
->file_offset
+
1913 ordered_extent
->len
);
1915 BUG_ON(root
== root
->fs_info
->tree_root
);
1916 ret
= insert_reserved_file_extent(trans
, inode
,
1917 ordered_extent
->file_offset
,
1918 ordered_extent
->start
,
1919 ordered_extent
->disk_len
,
1920 ordered_extent
->len
,
1921 ordered_extent
->len
,
1922 compress_type
, 0, 0,
1923 BTRFS_FILE_EXTENT_REG
);
1925 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1926 ordered_extent
->file_offset
, ordered_extent
->len
,
1929 btrfs_abort_transaction(trans
, root
, ret
);
1933 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1934 &ordered_extent
->list
);
1936 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1937 if (!ret
|| !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1938 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
1939 if (ret
) { /* -ENOMEM or corruption */
1940 btrfs_abort_transaction(trans
, root
, ret
);
1944 btrfs_set_inode_last_trans(trans
, inode
);
1948 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1949 ordered_extent
->file_offset
+
1950 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1952 if (root
!= root
->fs_info
->tree_root
)
1953 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1955 btrfs_end_transaction(trans
, root
);
1958 clear_extent_uptodate(io_tree
, ordered_extent
->file_offset
,
1959 ordered_extent
->file_offset
+
1960 ordered_extent
->len
- 1, NULL
, GFP_NOFS
);
1963 * This needs to be done to make sure anybody waiting knows we are done
1964 * updating everything for this ordered extent.
1966 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1969 btrfs_put_ordered_extent(ordered_extent
);
1970 /* once for the tree */
1971 btrfs_put_ordered_extent(ordered_extent
);
1976 static void finish_ordered_fn(struct btrfs_work
*work
)
1978 struct btrfs_ordered_extent
*ordered_extent
;
1979 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
1980 btrfs_finish_ordered_io(ordered_extent
);
1983 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1984 struct extent_state
*state
, int uptodate
)
1986 struct inode
*inode
= page
->mapping
->host
;
1987 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1988 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1989 struct btrfs_workers
*workers
;
1991 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1993 ClearPagePrivate2(page
);
1994 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1995 end
- start
+ 1, uptodate
))
1998 ordered_extent
->work
.func
= finish_ordered_fn
;
1999 ordered_extent
->work
.flags
= 0;
2001 if (btrfs_is_free_space_inode(inode
))
2002 workers
= &root
->fs_info
->endio_freespace_worker
;
2004 workers
= &root
->fs_info
->endio_write_workers
;
2005 btrfs_queue_worker(workers
, &ordered_extent
->work
);
2011 * when reads are done, we need to check csums to verify the data is correct
2012 * if there's a match, we allow the bio to finish. If not, the code in
2013 * extent_io.c will try to find good copies for us.
2015 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2016 struct extent_state
*state
, int mirror
)
2018 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2019 struct inode
*inode
= page
->mapping
->host
;
2020 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2022 u64
private = ~(u32
)0;
2024 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2027 if (PageChecked(page
)) {
2028 ClearPageChecked(page
);
2032 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
2035 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
2036 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
2037 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
2042 if (state
&& state
->start
== start
) {
2043 private = state
->private;
2046 ret
= get_state_private(io_tree
, start
, &private);
2048 kaddr
= kmap_atomic(page
);
2052 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
2053 btrfs_csum_final(csum
, (char *)&csum
);
2054 if (csum
!= private)
2057 kunmap_atomic(kaddr
);
2062 printk_ratelimited(KERN_INFO
"btrfs csum failed ino %llu off %llu csum %u "
2064 (unsigned long long)btrfs_ino(page
->mapping
->host
),
2065 (unsigned long long)start
, csum
,
2066 (unsigned long long)private);
2067 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2068 flush_dcache_page(page
);
2069 kunmap_atomic(kaddr
);
2075 struct delayed_iput
{
2076 struct list_head list
;
2077 struct inode
*inode
;
2080 /* JDM: If this is fs-wide, why can't we add a pointer to
2081 * btrfs_inode instead and avoid the allocation? */
2082 void btrfs_add_delayed_iput(struct inode
*inode
)
2084 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2085 struct delayed_iput
*delayed
;
2087 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2090 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2091 delayed
->inode
= inode
;
2093 spin_lock(&fs_info
->delayed_iput_lock
);
2094 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2095 spin_unlock(&fs_info
->delayed_iput_lock
);
2098 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2101 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2102 struct delayed_iput
*delayed
;
2105 spin_lock(&fs_info
->delayed_iput_lock
);
2106 empty
= list_empty(&fs_info
->delayed_iputs
);
2107 spin_unlock(&fs_info
->delayed_iput_lock
);
2111 spin_lock(&fs_info
->delayed_iput_lock
);
2112 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2113 spin_unlock(&fs_info
->delayed_iput_lock
);
2115 while (!list_empty(&list
)) {
2116 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2117 list_del(&delayed
->list
);
2118 iput(delayed
->inode
);
2123 enum btrfs_orphan_cleanup_state
{
2124 ORPHAN_CLEANUP_STARTED
= 1,
2125 ORPHAN_CLEANUP_DONE
= 2,
2129 * This is called in transaction commit time. If there are no orphan
2130 * files in the subvolume, it removes orphan item and frees block_rsv
2133 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2134 struct btrfs_root
*root
)
2136 struct btrfs_block_rsv
*block_rsv
;
2139 if (atomic_read(&root
->orphan_inodes
) ||
2140 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2143 spin_lock(&root
->orphan_lock
);
2144 if (atomic_read(&root
->orphan_inodes
)) {
2145 spin_unlock(&root
->orphan_lock
);
2149 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
2150 spin_unlock(&root
->orphan_lock
);
2154 block_rsv
= root
->orphan_block_rsv
;
2155 root
->orphan_block_rsv
= NULL
;
2156 spin_unlock(&root
->orphan_lock
);
2158 if (root
->orphan_item_inserted
&&
2159 btrfs_root_refs(&root
->root_item
) > 0) {
2160 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2161 root
->root_key
.objectid
);
2163 root
->orphan_item_inserted
= 0;
2167 WARN_ON(block_rsv
->size
> 0);
2168 btrfs_free_block_rsv(root
, block_rsv
);
2173 * This creates an orphan entry for the given inode in case something goes
2174 * wrong in the middle of an unlink/truncate.
2176 * NOTE: caller of this function should reserve 5 units of metadata for
2179 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2181 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2182 struct btrfs_block_rsv
*block_rsv
= NULL
;
2187 if (!root
->orphan_block_rsv
) {
2188 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
2193 spin_lock(&root
->orphan_lock
);
2194 if (!root
->orphan_block_rsv
) {
2195 root
->orphan_block_rsv
= block_rsv
;
2196 } else if (block_rsv
) {
2197 btrfs_free_block_rsv(root
, block_rsv
);
2201 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2202 &BTRFS_I(inode
)->runtime_flags
)) {
2205 * For proper ENOSPC handling, we should do orphan
2206 * cleanup when mounting. But this introduces backward
2207 * compatibility issue.
2209 if (!xchg(&root
->orphan_item_inserted
, 1))
2215 atomic_inc(&root
->orphan_inodes
);
2218 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
2219 &BTRFS_I(inode
)->runtime_flags
))
2221 spin_unlock(&root
->orphan_lock
);
2223 /* grab metadata reservation from transaction handle */
2225 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2226 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
2229 /* insert an orphan item to track this unlinked/truncated file */
2231 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
2232 if (ret
&& ret
!= -EEXIST
) {
2233 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2234 &BTRFS_I(inode
)->runtime_flags
);
2235 btrfs_abort_transaction(trans
, root
, ret
);
2241 /* insert an orphan item to track subvolume contains orphan files */
2243 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2244 root
->root_key
.objectid
);
2245 if (ret
&& ret
!= -EEXIST
) {
2246 btrfs_abort_transaction(trans
, root
, ret
);
2254 * We have done the truncate/delete so we can go ahead and remove the orphan
2255 * item for this particular inode.
2257 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2259 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2260 int delete_item
= 0;
2261 int release_rsv
= 0;
2264 spin_lock(&root
->orphan_lock
);
2265 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2266 &BTRFS_I(inode
)->runtime_flags
))
2269 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
2270 &BTRFS_I(inode
)->runtime_flags
))
2272 spin_unlock(&root
->orphan_lock
);
2274 if (trans
&& delete_item
) {
2275 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
2276 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
2280 btrfs_orphan_release_metadata(inode
);
2281 atomic_dec(&root
->orphan_inodes
);
2288 * this cleans up any orphans that may be left on the list from the last use
2291 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2293 struct btrfs_path
*path
;
2294 struct extent_buffer
*leaf
;
2295 struct btrfs_key key
, found_key
;
2296 struct btrfs_trans_handle
*trans
;
2297 struct inode
*inode
;
2298 u64 last_objectid
= 0;
2299 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2301 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2304 path
= btrfs_alloc_path();
2311 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2312 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2313 key
.offset
= (u64
)-1;
2316 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2321 * if ret == 0 means we found what we were searching for, which
2322 * is weird, but possible, so only screw with path if we didn't
2323 * find the key and see if we have stuff that matches
2327 if (path
->slots
[0] == 0)
2332 /* pull out the item */
2333 leaf
= path
->nodes
[0];
2334 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2336 /* make sure the item matches what we want */
2337 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2339 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2342 /* release the path since we're done with it */
2343 btrfs_release_path(path
);
2346 * this is where we are basically btrfs_lookup, without the
2347 * crossing root thing. we store the inode number in the
2348 * offset of the orphan item.
2351 if (found_key
.offset
== last_objectid
) {
2352 printk(KERN_ERR
"btrfs: Error removing orphan entry, "
2353 "stopping orphan cleanup\n");
2358 last_objectid
= found_key
.offset
;
2360 found_key
.objectid
= found_key
.offset
;
2361 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2362 found_key
.offset
= 0;
2363 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2364 ret
= PTR_RET(inode
);
2365 if (ret
&& ret
!= -ESTALE
)
2368 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
2369 struct btrfs_root
*dead_root
;
2370 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2371 int is_dead_root
= 0;
2374 * this is an orphan in the tree root. Currently these
2375 * could come from 2 sources:
2376 * a) a snapshot deletion in progress
2377 * b) a free space cache inode
2378 * We need to distinguish those two, as the snapshot
2379 * orphan must not get deleted.
2380 * find_dead_roots already ran before us, so if this
2381 * is a snapshot deletion, we should find the root
2382 * in the dead_roots list
2384 spin_lock(&fs_info
->trans_lock
);
2385 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
2387 if (dead_root
->root_key
.objectid
==
2388 found_key
.objectid
) {
2393 spin_unlock(&fs_info
->trans_lock
);
2395 /* prevent this orphan from being found again */
2396 key
.offset
= found_key
.objectid
- 1;
2401 * Inode is already gone but the orphan item is still there,
2402 * kill the orphan item.
2404 if (ret
== -ESTALE
) {
2405 trans
= btrfs_start_transaction(root
, 1);
2406 if (IS_ERR(trans
)) {
2407 ret
= PTR_ERR(trans
);
2410 printk(KERN_ERR
"auto deleting %Lu\n",
2411 found_key
.objectid
);
2412 ret
= btrfs_del_orphan_item(trans
, root
,
2413 found_key
.objectid
);
2414 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
2415 btrfs_end_transaction(trans
, root
);
2420 * add this inode to the orphan list so btrfs_orphan_del does
2421 * the proper thing when we hit it
2423 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2424 &BTRFS_I(inode
)->runtime_flags
);
2426 /* if we have links, this was a truncate, lets do that */
2427 if (inode
->i_nlink
) {
2428 if (!S_ISREG(inode
->i_mode
)) {
2434 ret
= btrfs_truncate(inode
);
2439 /* this will do delete_inode and everything for us */
2444 /* release the path since we're done with it */
2445 btrfs_release_path(path
);
2447 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2449 if (root
->orphan_block_rsv
)
2450 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2453 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2454 trans
= btrfs_join_transaction(root
);
2456 btrfs_end_transaction(trans
, root
);
2460 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2462 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2466 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2467 btrfs_free_path(path
);
2472 * very simple check to peek ahead in the leaf looking for xattrs. If we
2473 * don't find any xattrs, we know there can't be any acls.
2475 * slot is the slot the inode is in, objectid is the objectid of the inode
2477 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2478 int slot
, u64 objectid
)
2480 u32 nritems
= btrfs_header_nritems(leaf
);
2481 struct btrfs_key found_key
;
2485 while (slot
< nritems
) {
2486 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2488 /* we found a different objectid, there must not be acls */
2489 if (found_key
.objectid
!= objectid
)
2492 /* we found an xattr, assume we've got an acl */
2493 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2497 * we found a key greater than an xattr key, there can't
2498 * be any acls later on
2500 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2507 * it goes inode, inode backrefs, xattrs, extents,
2508 * so if there are a ton of hard links to an inode there can
2509 * be a lot of backrefs. Don't waste time searching too hard,
2510 * this is just an optimization
2515 /* we hit the end of the leaf before we found an xattr or
2516 * something larger than an xattr. We have to assume the inode
2523 * read an inode from the btree into the in-memory inode
2525 static void btrfs_read_locked_inode(struct inode
*inode
)
2527 struct btrfs_path
*path
;
2528 struct extent_buffer
*leaf
;
2529 struct btrfs_inode_item
*inode_item
;
2530 struct btrfs_timespec
*tspec
;
2531 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2532 struct btrfs_key location
;
2536 bool filled
= false;
2538 ret
= btrfs_fill_inode(inode
, &rdev
);
2542 path
= btrfs_alloc_path();
2546 path
->leave_spinning
= 1;
2547 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2549 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2553 leaf
= path
->nodes
[0];
2558 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2559 struct btrfs_inode_item
);
2560 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2561 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
2562 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
2563 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
2564 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2566 tspec
= btrfs_inode_atime(inode_item
);
2567 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2568 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2570 tspec
= btrfs_inode_mtime(inode_item
);
2571 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2572 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2574 tspec
= btrfs_inode_ctime(inode_item
);
2575 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2576 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2578 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2579 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2580 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
2583 * If we were modified in the current generation and evicted from memory
2584 * and then re-read we need to do a full sync since we don't have any
2585 * idea about which extents were modified before we were evicted from
2588 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
2589 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
2590 &BTRFS_I(inode
)->runtime_flags
);
2592 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
2593 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2595 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2597 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2598 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2601 * try to precache a NULL acl entry for files that don't have
2602 * any xattrs or acls
2604 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
2607 cache_no_acl(inode
);
2609 btrfs_free_path(path
);
2611 switch (inode
->i_mode
& S_IFMT
) {
2613 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2614 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2615 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2616 inode
->i_fop
= &btrfs_file_operations
;
2617 inode
->i_op
= &btrfs_file_inode_operations
;
2620 inode
->i_fop
= &btrfs_dir_file_operations
;
2621 if (root
== root
->fs_info
->tree_root
)
2622 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2624 inode
->i_op
= &btrfs_dir_inode_operations
;
2627 inode
->i_op
= &btrfs_symlink_inode_operations
;
2628 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2629 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2632 inode
->i_op
= &btrfs_special_inode_operations
;
2633 init_special_inode(inode
, inode
->i_mode
, rdev
);
2637 btrfs_update_iflags(inode
);
2641 btrfs_free_path(path
);
2642 make_bad_inode(inode
);
2646 * given a leaf and an inode, copy the inode fields into the leaf
2648 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2649 struct extent_buffer
*leaf
,
2650 struct btrfs_inode_item
*item
,
2651 struct inode
*inode
)
2653 btrfs_set_inode_uid(leaf
, item
, i_uid_read(inode
));
2654 btrfs_set_inode_gid(leaf
, item
, i_gid_read(inode
));
2655 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2656 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2657 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2659 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2660 inode
->i_atime
.tv_sec
);
2661 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2662 inode
->i_atime
.tv_nsec
);
2664 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2665 inode
->i_mtime
.tv_sec
);
2666 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2667 inode
->i_mtime
.tv_nsec
);
2669 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2670 inode
->i_ctime
.tv_sec
);
2671 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2672 inode
->i_ctime
.tv_nsec
);
2674 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2675 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2676 btrfs_set_inode_sequence(leaf
, item
, inode
->i_version
);
2677 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2678 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2679 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2680 btrfs_set_inode_block_group(leaf
, item
, 0);
2684 * copy everything in the in-memory inode into the btree.
2686 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
2687 struct btrfs_root
*root
, struct inode
*inode
)
2689 struct btrfs_inode_item
*inode_item
;
2690 struct btrfs_path
*path
;
2691 struct extent_buffer
*leaf
;
2694 path
= btrfs_alloc_path();
2698 path
->leave_spinning
= 1;
2699 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
2707 btrfs_unlock_up_safe(path
, 1);
2708 leaf
= path
->nodes
[0];
2709 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2710 struct btrfs_inode_item
);
2712 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2713 btrfs_mark_buffer_dirty(leaf
);
2714 btrfs_set_inode_last_trans(trans
, inode
);
2717 btrfs_free_path(path
);
2722 * copy everything in the in-memory inode into the btree.
2724 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2725 struct btrfs_root
*root
, struct inode
*inode
)
2730 * If the inode is a free space inode, we can deadlock during commit
2731 * if we put it into the delayed code.
2733 * The data relocation inode should also be directly updated
2736 if (!btrfs_is_free_space_inode(inode
)
2737 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
2738 btrfs_update_root_times(trans
, root
);
2740 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
2742 btrfs_set_inode_last_trans(trans
, inode
);
2746 return btrfs_update_inode_item(trans
, root
, inode
);
2749 static noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
2750 struct btrfs_root
*root
, struct inode
*inode
)
2754 ret
= btrfs_update_inode(trans
, root
, inode
);
2756 return btrfs_update_inode_item(trans
, root
, inode
);
2761 * unlink helper that gets used here in inode.c and in the tree logging
2762 * recovery code. It remove a link in a directory with a given name, and
2763 * also drops the back refs in the inode to the directory
2765 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2766 struct btrfs_root
*root
,
2767 struct inode
*dir
, struct inode
*inode
,
2768 const char *name
, int name_len
)
2770 struct btrfs_path
*path
;
2772 struct extent_buffer
*leaf
;
2773 struct btrfs_dir_item
*di
;
2774 struct btrfs_key key
;
2776 u64 ino
= btrfs_ino(inode
);
2777 u64 dir_ino
= btrfs_ino(dir
);
2779 path
= btrfs_alloc_path();
2785 path
->leave_spinning
= 1;
2786 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2787 name
, name_len
, -1);
2796 leaf
= path
->nodes
[0];
2797 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2798 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2801 btrfs_release_path(path
);
2803 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
2806 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2807 "inode %llu parent %llu\n", name_len
, name
,
2808 (unsigned long long)ino
, (unsigned long long)dir_ino
);
2809 btrfs_abort_transaction(trans
, root
, ret
);
2813 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2815 btrfs_abort_transaction(trans
, root
, ret
);
2819 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2821 if (ret
!= 0 && ret
!= -ENOENT
) {
2822 btrfs_abort_transaction(trans
, root
, ret
);
2826 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2831 btrfs_free_path(path
);
2835 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2836 inode_inc_iversion(inode
);
2837 inode_inc_iversion(dir
);
2838 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2839 ret
= btrfs_update_inode(trans
, root
, dir
);
2844 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2845 struct btrfs_root
*root
,
2846 struct inode
*dir
, struct inode
*inode
,
2847 const char *name
, int name_len
)
2850 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2852 btrfs_drop_nlink(inode
);
2853 ret
= btrfs_update_inode(trans
, root
, inode
);
2859 /* helper to check if there is any shared block in the path */
2860 static int check_path_shared(struct btrfs_root
*root
,
2861 struct btrfs_path
*path
)
2863 struct extent_buffer
*eb
;
2867 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2870 if (!path
->nodes
[level
])
2872 eb
= path
->nodes
[level
];
2873 if (!btrfs_block_can_be_shared(root
, eb
))
2875 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2884 * helper to start transaction for unlink and rmdir.
2886 * unlink and rmdir are special in btrfs, they do not always free space.
2887 * so in enospc case, we should make sure they will free space before
2888 * allowing them to use the global metadata reservation.
2890 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2891 struct dentry
*dentry
)
2893 struct btrfs_trans_handle
*trans
;
2894 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2895 struct btrfs_path
*path
;
2896 struct btrfs_dir_item
*di
;
2897 struct inode
*inode
= dentry
->d_inode
;
2902 u64 ino
= btrfs_ino(inode
);
2903 u64 dir_ino
= btrfs_ino(dir
);
2906 * 1 for the possible orphan item
2907 * 1 for the dir item
2908 * 1 for the dir index
2909 * 1 for the inode ref
2910 * 1 for the inode ref in the tree log
2911 * 2 for the dir entries in the log
2914 trans
= btrfs_start_transaction(root
, 8);
2915 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2918 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2919 return ERR_PTR(-ENOSPC
);
2921 /* check if there is someone else holds reference */
2922 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2923 return ERR_PTR(-ENOSPC
);
2925 if (atomic_read(&inode
->i_count
) > 2)
2926 return ERR_PTR(-ENOSPC
);
2928 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2929 return ERR_PTR(-ENOSPC
);
2931 path
= btrfs_alloc_path();
2933 root
->fs_info
->enospc_unlink
= 0;
2934 return ERR_PTR(-ENOMEM
);
2937 /* 1 for the orphan item */
2938 trans
= btrfs_start_transaction(root
, 1);
2939 if (IS_ERR(trans
)) {
2940 btrfs_free_path(path
);
2941 root
->fs_info
->enospc_unlink
= 0;
2945 path
->skip_locking
= 1;
2946 path
->search_commit_root
= 1;
2948 ret
= btrfs_lookup_inode(trans
, root
, path
,
2949 &BTRFS_I(dir
)->location
, 0);
2955 if (check_path_shared(root
, path
))
2960 btrfs_release_path(path
);
2962 ret
= btrfs_lookup_inode(trans
, root
, path
,
2963 &BTRFS_I(inode
)->location
, 0);
2969 if (check_path_shared(root
, path
))
2974 btrfs_release_path(path
);
2976 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2977 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2983 BUG_ON(ret
== 0); /* Corruption */
2984 if (check_path_shared(root
, path
))
2986 btrfs_release_path(path
);
2994 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2995 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
3001 if (check_path_shared(root
, path
))
3007 btrfs_release_path(path
);
3009 ret
= btrfs_get_inode_ref_index(trans
, root
, path
, dentry
->d_name
.name
,
3010 dentry
->d_name
.len
, ino
, dir_ino
, 0,
3017 if (check_path_shared(root
, path
))
3020 btrfs_release_path(path
);
3023 * This is a commit root search, if we can lookup inode item and other
3024 * relative items in the commit root, it means the transaction of
3025 * dir/file creation has been committed, and the dir index item that we
3026 * delay to insert has also been inserted into the commit root. So
3027 * we needn't worry about the delayed insertion of the dir index item
3030 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
3031 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
3036 BUG_ON(ret
== -ENOENT
);
3037 if (check_path_shared(root
, path
))
3042 btrfs_free_path(path
);
3043 /* Migrate the orphan reservation over */
3045 err
= btrfs_block_rsv_migrate(trans
->block_rsv
,
3046 &root
->fs_info
->global_block_rsv
,
3047 trans
->bytes_reserved
);
3050 btrfs_end_transaction(trans
, root
);
3051 root
->fs_info
->enospc_unlink
= 0;
3052 return ERR_PTR(err
);
3055 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
3059 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
3060 struct btrfs_root
*root
)
3062 if (trans
->block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
) {
3063 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3064 trans
->bytes_reserved
);
3065 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3066 BUG_ON(!root
->fs_info
->enospc_unlink
);
3067 root
->fs_info
->enospc_unlink
= 0;
3069 btrfs_end_transaction(trans
, root
);
3072 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3074 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3075 struct btrfs_trans_handle
*trans
;
3076 struct inode
*inode
= dentry
->d_inode
;
3078 unsigned long nr
= 0;
3080 trans
= __unlink_start_trans(dir
, dentry
);
3082 return PTR_ERR(trans
);
3084 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
3086 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3087 dentry
->d_name
.name
, dentry
->d_name
.len
);
3091 if (inode
->i_nlink
== 0) {
3092 ret
= btrfs_orphan_add(trans
, inode
);
3098 nr
= trans
->blocks_used
;
3099 __unlink_end_trans(trans
, root
);
3100 btrfs_btree_balance_dirty(root
, nr
);
3104 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
3105 struct btrfs_root
*root
,
3106 struct inode
*dir
, u64 objectid
,
3107 const char *name
, int name_len
)
3109 struct btrfs_path
*path
;
3110 struct extent_buffer
*leaf
;
3111 struct btrfs_dir_item
*di
;
3112 struct btrfs_key key
;
3115 u64 dir_ino
= btrfs_ino(dir
);
3117 path
= btrfs_alloc_path();
3121 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3122 name
, name_len
, -1);
3123 if (IS_ERR_OR_NULL(di
)) {
3131 leaf
= path
->nodes
[0];
3132 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3133 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3134 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3136 btrfs_abort_transaction(trans
, root
, ret
);
3139 btrfs_release_path(path
);
3141 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3142 objectid
, root
->root_key
.objectid
,
3143 dir_ino
, &index
, name
, name_len
);
3145 if (ret
!= -ENOENT
) {
3146 btrfs_abort_transaction(trans
, root
, ret
);
3149 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
3151 if (IS_ERR_OR_NULL(di
)) {
3156 btrfs_abort_transaction(trans
, root
, ret
);
3160 leaf
= path
->nodes
[0];
3161 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3162 btrfs_release_path(path
);
3165 btrfs_release_path(path
);
3167 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3169 btrfs_abort_transaction(trans
, root
, ret
);
3173 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3174 inode_inc_iversion(dir
);
3175 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3176 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
3178 btrfs_abort_transaction(trans
, root
, ret
);
3180 btrfs_free_path(path
);
3184 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3186 struct inode
*inode
= dentry
->d_inode
;
3188 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3189 struct btrfs_trans_handle
*trans
;
3190 unsigned long nr
= 0;
3192 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
3194 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3197 trans
= __unlink_start_trans(dir
, dentry
);
3199 return PTR_ERR(trans
);
3201 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3202 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3203 BTRFS_I(inode
)->location
.objectid
,
3204 dentry
->d_name
.name
,
3205 dentry
->d_name
.len
);
3209 err
= btrfs_orphan_add(trans
, inode
);
3213 /* now the directory is empty */
3214 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3215 dentry
->d_name
.name
, dentry
->d_name
.len
);
3217 btrfs_i_size_write(inode
, 0);
3219 nr
= trans
->blocks_used
;
3220 __unlink_end_trans(trans
, root
);
3221 btrfs_btree_balance_dirty(root
, nr
);
3227 * this can truncate away extent items, csum items and directory items.
3228 * It starts at a high offset and removes keys until it can't find
3229 * any higher than new_size
3231 * csum items that cross the new i_size are truncated to the new size
3234 * min_type is the minimum key type to truncate down to. If set to 0, this
3235 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3237 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3238 struct btrfs_root
*root
,
3239 struct inode
*inode
,
3240 u64 new_size
, u32 min_type
)
3242 struct btrfs_path
*path
;
3243 struct extent_buffer
*leaf
;
3244 struct btrfs_file_extent_item
*fi
;
3245 struct btrfs_key key
;
3246 struct btrfs_key found_key
;
3247 u64 extent_start
= 0;
3248 u64 extent_num_bytes
= 0;
3249 u64 extent_offset
= 0;
3251 u64 mask
= root
->sectorsize
- 1;
3252 u32 found_type
= (u8
)-1;
3255 int pending_del_nr
= 0;
3256 int pending_del_slot
= 0;
3257 int extent_type
= -1;
3260 u64 ino
= btrfs_ino(inode
);
3262 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3264 path
= btrfs_alloc_path();
3270 * We want to drop from the next block forward in case this new size is
3271 * not block aligned since we will be keeping the last block of the
3272 * extent just the way it is.
3274 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3275 btrfs_drop_extent_cache(inode
, (new_size
+ mask
) & (~mask
), (u64
)-1, 0);
3278 * This function is also used to drop the items in the log tree before
3279 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3280 * it is used to drop the loged items. So we shouldn't kill the delayed
3283 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3284 btrfs_kill_delayed_inode_items(inode
);
3287 key
.offset
= (u64
)-1;
3291 path
->leave_spinning
= 1;
3292 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3299 /* there are no items in the tree for us to truncate, we're
3302 if (path
->slots
[0] == 0)
3309 leaf
= path
->nodes
[0];
3310 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3311 found_type
= btrfs_key_type(&found_key
);
3313 if (found_key
.objectid
!= ino
)
3316 if (found_type
< min_type
)
3319 item_end
= found_key
.offset
;
3320 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3321 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3322 struct btrfs_file_extent_item
);
3323 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3324 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3326 btrfs_file_extent_num_bytes(leaf
, fi
);
3327 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3328 item_end
+= btrfs_file_extent_inline_len(leaf
,
3333 if (found_type
> min_type
) {
3336 if (item_end
< new_size
)
3338 if (found_key
.offset
>= new_size
)
3344 /* FIXME, shrink the extent if the ref count is only 1 */
3345 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3348 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3350 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3352 u64 orig_num_bytes
=
3353 btrfs_file_extent_num_bytes(leaf
, fi
);
3354 extent_num_bytes
= new_size
-
3355 found_key
.offset
+ root
->sectorsize
- 1;
3356 extent_num_bytes
= extent_num_bytes
&
3357 ~((u64
)root
->sectorsize
- 1);
3358 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3360 num_dec
= (orig_num_bytes
-
3362 if (root
->ref_cows
&& extent_start
!= 0)
3363 inode_sub_bytes(inode
, num_dec
);
3364 btrfs_mark_buffer_dirty(leaf
);
3367 btrfs_file_extent_disk_num_bytes(leaf
,
3369 extent_offset
= found_key
.offset
-
3370 btrfs_file_extent_offset(leaf
, fi
);
3372 /* FIXME blocksize != 4096 */
3373 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3374 if (extent_start
!= 0) {
3377 inode_sub_bytes(inode
, num_dec
);
3380 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3382 * we can't truncate inline items that have had
3386 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3387 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3388 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3389 u32 size
= new_size
- found_key
.offset
;
3391 if (root
->ref_cows
) {
3392 inode_sub_bytes(inode
, item_end
+ 1 -
3396 btrfs_file_extent_calc_inline_size(size
);
3397 btrfs_truncate_item(trans
, root
, path
,
3399 } else if (root
->ref_cows
) {
3400 inode_sub_bytes(inode
, item_end
+ 1 -
3406 if (!pending_del_nr
) {
3407 /* no pending yet, add ourselves */
3408 pending_del_slot
= path
->slots
[0];
3410 } else if (pending_del_nr
&&
3411 path
->slots
[0] + 1 == pending_del_slot
) {
3412 /* hop on the pending chunk */
3414 pending_del_slot
= path
->slots
[0];
3421 if (found_extent
&& (root
->ref_cows
||
3422 root
== root
->fs_info
->tree_root
)) {
3423 btrfs_set_path_blocking(path
);
3424 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3425 extent_num_bytes
, 0,
3426 btrfs_header_owner(leaf
),
3427 ino
, extent_offset
, 0);
3431 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3434 if (path
->slots
[0] == 0 ||
3435 path
->slots
[0] != pending_del_slot
) {
3436 if (pending_del_nr
) {
3437 ret
= btrfs_del_items(trans
, root
, path
,
3441 btrfs_abort_transaction(trans
,
3447 btrfs_release_path(path
);
3454 if (pending_del_nr
) {
3455 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3458 btrfs_abort_transaction(trans
, root
, ret
);
3461 btrfs_free_path(path
);
3466 * btrfs_truncate_page - read, zero a chunk and write a page
3467 * @inode - inode that we're zeroing
3468 * @from - the offset to start zeroing
3469 * @len - the length to zero, 0 to zero the entire range respective to the
3471 * @front - zero up to the offset instead of from the offset on
3473 * This will find the page for the "from" offset and cow the page and zero the
3474 * part we want to zero. This is used with truncate and hole punching.
3476 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
3479 struct address_space
*mapping
= inode
->i_mapping
;
3480 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3481 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3482 struct btrfs_ordered_extent
*ordered
;
3483 struct extent_state
*cached_state
= NULL
;
3485 u32 blocksize
= root
->sectorsize
;
3486 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3487 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3489 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
3494 if ((offset
& (blocksize
- 1)) == 0 &&
3495 (!len
|| ((len
& (blocksize
- 1)) == 0)))
3497 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3503 page
= find_or_create_page(mapping
, index
, mask
);
3505 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3509 page_start
= page_offset(page
);
3510 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3512 if (!PageUptodate(page
)) {
3513 ret
= btrfs_readpage(NULL
, page
);
3515 if (page
->mapping
!= mapping
) {
3517 page_cache_release(page
);
3520 if (!PageUptodate(page
)) {
3525 wait_on_page_writeback(page
);
3527 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
3528 set_page_extent_mapped(page
);
3530 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3532 unlock_extent_cached(io_tree
, page_start
, page_end
,
3533 &cached_state
, GFP_NOFS
);
3535 page_cache_release(page
);
3536 btrfs_start_ordered_extent(inode
, ordered
, 1);
3537 btrfs_put_ordered_extent(ordered
);
3541 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3542 EXTENT_DIRTY
| EXTENT_DELALLOC
|
3543 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
3544 0, 0, &cached_state
, GFP_NOFS
);
3546 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3549 unlock_extent_cached(io_tree
, page_start
, page_end
,
3550 &cached_state
, GFP_NOFS
);
3555 if (offset
!= PAGE_CACHE_SIZE
) {
3557 len
= PAGE_CACHE_SIZE
- offset
;
3560 memset(kaddr
, 0, offset
);
3562 memset(kaddr
+ offset
, 0, len
);
3563 flush_dcache_page(page
);
3566 ClearPageChecked(page
);
3567 set_page_dirty(page
);
3568 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3573 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3575 page_cache_release(page
);
3581 * This function puts in dummy file extents for the area we're creating a hole
3582 * for. So if we are truncating this file to a larger size we need to insert
3583 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3584 * the range between oldsize and size
3586 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3588 struct btrfs_trans_handle
*trans
;
3589 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3590 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3591 struct extent_map
*em
= NULL
;
3592 struct extent_state
*cached_state
= NULL
;
3593 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3594 u64 mask
= root
->sectorsize
- 1;
3595 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3596 u64 block_end
= (size
+ mask
) & ~mask
;
3602 if (size
<= hole_start
)
3606 struct btrfs_ordered_extent
*ordered
;
3607 btrfs_wait_ordered_range(inode
, hole_start
,
3608 block_end
- hole_start
);
3609 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3611 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3614 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3615 &cached_state
, GFP_NOFS
);
3616 btrfs_put_ordered_extent(ordered
);
3619 cur_offset
= hole_start
;
3621 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3622 block_end
- cur_offset
, 0);
3627 last_byte
= min(extent_map_end(em
), block_end
);
3628 last_byte
= (last_byte
+ mask
) & ~mask
;
3629 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3630 struct extent_map
*hole_em
;
3631 hole_size
= last_byte
- cur_offset
;
3633 trans
= btrfs_start_transaction(root
, 3);
3634 if (IS_ERR(trans
)) {
3635 err
= PTR_ERR(trans
);
3639 err
= btrfs_drop_extents(trans
, root
, inode
,
3641 cur_offset
+ hole_size
, 1);
3643 btrfs_abort_transaction(trans
, root
, err
);
3644 btrfs_end_transaction(trans
, root
);
3648 err
= btrfs_insert_file_extent(trans
, root
,
3649 btrfs_ino(inode
), cur_offset
, 0,
3650 0, hole_size
, 0, hole_size
,
3653 btrfs_abort_transaction(trans
, root
, err
);
3654 btrfs_end_transaction(trans
, root
);
3658 btrfs_drop_extent_cache(inode
, cur_offset
,
3659 cur_offset
+ hole_size
- 1, 0);
3660 hole_em
= alloc_extent_map();
3662 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3663 &BTRFS_I(inode
)->runtime_flags
);
3666 hole_em
->start
= cur_offset
;
3667 hole_em
->len
= hole_size
;
3668 hole_em
->orig_start
= cur_offset
;
3670 hole_em
->block_start
= EXTENT_MAP_HOLE
;
3671 hole_em
->block_len
= 0;
3672 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3673 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
3674 hole_em
->generation
= trans
->transid
;
3677 write_lock(&em_tree
->lock
);
3678 err
= add_extent_mapping(em_tree
, hole_em
);
3680 list_move(&hole_em
->list
,
3681 &em_tree
->modified_extents
);
3682 write_unlock(&em_tree
->lock
);
3685 btrfs_drop_extent_cache(inode
, cur_offset
,
3689 free_extent_map(hole_em
);
3691 btrfs_update_inode(trans
, root
, inode
);
3692 btrfs_end_transaction(trans
, root
);
3694 free_extent_map(em
);
3696 cur_offset
= last_byte
;
3697 if (cur_offset
>= block_end
)
3701 free_extent_map(em
);
3702 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3707 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3709 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3710 struct btrfs_trans_handle
*trans
;
3711 loff_t oldsize
= i_size_read(inode
);
3714 if (newsize
== oldsize
)
3717 if (newsize
> oldsize
) {
3718 truncate_pagecache(inode
, oldsize
, newsize
);
3719 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3723 trans
= btrfs_start_transaction(root
, 1);
3725 return PTR_ERR(trans
);
3727 i_size_write(inode
, newsize
);
3728 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3729 ret
= btrfs_update_inode(trans
, root
, inode
);
3730 btrfs_end_transaction(trans
, root
);
3734 * We're truncating a file that used to have good data down to
3735 * zero. Make sure it gets into the ordered flush list so that
3736 * any new writes get down to disk quickly.
3739 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
3740 &BTRFS_I(inode
)->runtime_flags
);
3742 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3743 truncate_setsize(inode
, newsize
);
3744 ret
= btrfs_truncate(inode
);
3750 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3752 struct inode
*inode
= dentry
->d_inode
;
3753 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3756 if (btrfs_root_readonly(root
))
3759 err
= inode_change_ok(inode
, attr
);
3763 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3764 err
= btrfs_setsize(inode
, attr
->ia_size
);
3769 if (attr
->ia_valid
) {
3770 setattr_copy(inode
, attr
);
3771 inode_inc_iversion(inode
);
3772 err
= btrfs_dirty_inode(inode
);
3774 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
3775 err
= btrfs_acl_chmod(inode
);
3781 void btrfs_evict_inode(struct inode
*inode
)
3783 struct btrfs_trans_handle
*trans
;
3784 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3785 struct btrfs_block_rsv
*rsv
, *global_rsv
;
3786 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
3790 trace_btrfs_inode_evict(inode
);
3792 truncate_inode_pages(&inode
->i_data
, 0);
3793 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3794 btrfs_is_free_space_inode(inode
)))
3797 if (is_bad_inode(inode
)) {
3798 btrfs_orphan_del(NULL
, inode
);
3801 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3802 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3804 if (root
->fs_info
->log_root_recovering
) {
3805 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3806 &BTRFS_I(inode
)->runtime_flags
));
3810 if (inode
->i_nlink
> 0) {
3811 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3815 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3817 btrfs_orphan_del(NULL
, inode
);
3820 rsv
->size
= min_size
;
3822 global_rsv
= &root
->fs_info
->global_block_rsv
;
3824 btrfs_i_size_write(inode
, 0);
3827 * This is a bit simpler than btrfs_truncate since we've already
3828 * reserved our space for our orphan item in the unlink, so we just
3829 * need to reserve some slack space in case we add bytes and update
3830 * inode item when doing the truncate.
3833 ret
= btrfs_block_rsv_refill_noflush(root
, rsv
, min_size
);
3836 * Try and steal from the global reserve since we will
3837 * likely not use this space anyway, we want to try as
3838 * hard as possible to get this to work.
3841 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, min_size
);
3844 printk(KERN_WARNING
"Could not get space for a "
3845 "delete, will truncate on mount %d\n", ret
);
3846 btrfs_orphan_del(NULL
, inode
);
3847 btrfs_free_block_rsv(root
, rsv
);
3851 trans
= btrfs_start_transaction_noflush(root
, 1);
3852 if (IS_ERR(trans
)) {
3853 btrfs_orphan_del(NULL
, inode
);
3854 btrfs_free_block_rsv(root
, rsv
);
3858 trans
->block_rsv
= rsv
;
3860 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3864 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3865 ret
= btrfs_update_inode(trans
, root
, inode
);
3868 nr
= trans
->blocks_used
;
3869 btrfs_end_transaction(trans
, root
);
3871 btrfs_btree_balance_dirty(root
, nr
);
3874 btrfs_free_block_rsv(root
, rsv
);
3877 trans
->block_rsv
= root
->orphan_block_rsv
;
3878 ret
= btrfs_orphan_del(trans
, inode
);
3882 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3883 if (!(root
== root
->fs_info
->tree_root
||
3884 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
3885 btrfs_return_ino(root
, btrfs_ino(inode
));
3887 nr
= trans
->blocks_used
;
3888 btrfs_end_transaction(trans
, root
);
3889 btrfs_btree_balance_dirty(root
, nr
);
3896 * this returns the key found in the dir entry in the location pointer.
3897 * If no dir entries were found, location->objectid is 0.
3899 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3900 struct btrfs_key
*location
)
3902 const char *name
= dentry
->d_name
.name
;
3903 int namelen
= dentry
->d_name
.len
;
3904 struct btrfs_dir_item
*di
;
3905 struct btrfs_path
*path
;
3906 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3909 path
= btrfs_alloc_path();
3913 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
3918 if (IS_ERR_OR_NULL(di
))
3921 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3923 btrfs_free_path(path
);
3926 location
->objectid
= 0;
3931 * when we hit a tree root in a directory, the btrfs part of the inode
3932 * needs to be changed to reflect the root directory of the tree root. This
3933 * is kind of like crossing a mount point.
3935 static int fixup_tree_root_location(struct btrfs_root
*root
,
3937 struct dentry
*dentry
,
3938 struct btrfs_key
*location
,
3939 struct btrfs_root
**sub_root
)
3941 struct btrfs_path
*path
;
3942 struct btrfs_root
*new_root
;
3943 struct btrfs_root_ref
*ref
;
3944 struct extent_buffer
*leaf
;
3948 path
= btrfs_alloc_path();
3955 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3956 BTRFS_I(dir
)->root
->root_key
.objectid
,
3957 location
->objectid
);
3964 leaf
= path
->nodes
[0];
3965 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3966 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
3967 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3970 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3971 (unsigned long)(ref
+ 1),
3972 dentry
->d_name
.len
);
3976 btrfs_release_path(path
);
3978 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3979 if (IS_ERR(new_root
)) {
3980 err
= PTR_ERR(new_root
);
3984 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3989 *sub_root
= new_root
;
3990 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3991 location
->type
= BTRFS_INODE_ITEM_KEY
;
3992 location
->offset
= 0;
3995 btrfs_free_path(path
);
3999 static void inode_tree_add(struct inode
*inode
)
4001 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4002 struct btrfs_inode
*entry
;
4004 struct rb_node
*parent
;
4005 u64 ino
= btrfs_ino(inode
);
4007 p
= &root
->inode_tree
.rb_node
;
4010 if (inode_unhashed(inode
))
4013 spin_lock(&root
->inode_lock
);
4016 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
4018 if (ino
< btrfs_ino(&entry
->vfs_inode
))
4019 p
= &parent
->rb_left
;
4020 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
4021 p
= &parent
->rb_right
;
4023 WARN_ON(!(entry
->vfs_inode
.i_state
&
4024 (I_WILL_FREE
| I_FREEING
)));
4025 rb_erase(parent
, &root
->inode_tree
);
4026 RB_CLEAR_NODE(parent
);
4027 spin_unlock(&root
->inode_lock
);
4031 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
4032 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4033 spin_unlock(&root
->inode_lock
);
4036 static void inode_tree_del(struct inode
*inode
)
4038 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4041 spin_lock(&root
->inode_lock
);
4042 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
4043 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4044 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
4045 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4047 spin_unlock(&root
->inode_lock
);
4050 * Free space cache has inodes in the tree root, but the tree root has a
4051 * root_refs of 0, so this could end up dropping the tree root as a
4052 * snapshot, so we need the extra !root->fs_info->tree_root check to
4053 * make sure we don't drop it.
4055 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
4056 root
!= root
->fs_info
->tree_root
) {
4057 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
4058 spin_lock(&root
->inode_lock
);
4059 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4060 spin_unlock(&root
->inode_lock
);
4062 btrfs_add_dead_root(root
);
4066 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
4068 struct rb_node
*node
;
4069 struct rb_node
*prev
;
4070 struct btrfs_inode
*entry
;
4071 struct inode
*inode
;
4074 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4076 spin_lock(&root
->inode_lock
);
4078 node
= root
->inode_tree
.rb_node
;
4082 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4084 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
4085 node
= node
->rb_left
;
4086 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
4087 node
= node
->rb_right
;
4093 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4094 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
4098 prev
= rb_next(prev
);
4102 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4103 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
4104 inode
= igrab(&entry
->vfs_inode
);
4106 spin_unlock(&root
->inode_lock
);
4107 if (atomic_read(&inode
->i_count
) > 1)
4108 d_prune_aliases(inode
);
4110 * btrfs_drop_inode will have it removed from
4111 * the inode cache when its usage count
4116 spin_lock(&root
->inode_lock
);
4120 if (cond_resched_lock(&root
->inode_lock
))
4123 node
= rb_next(node
);
4125 spin_unlock(&root
->inode_lock
);
4128 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4130 struct btrfs_iget_args
*args
= p
;
4131 inode
->i_ino
= args
->ino
;
4132 BTRFS_I(inode
)->root
= args
->root
;
4136 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4138 struct btrfs_iget_args
*args
= opaque
;
4139 return args
->ino
== btrfs_ino(inode
) &&
4140 args
->root
== BTRFS_I(inode
)->root
;
4143 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4145 struct btrfs_root
*root
)
4147 struct inode
*inode
;
4148 struct btrfs_iget_args args
;
4149 args
.ino
= objectid
;
4152 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4153 btrfs_init_locked_inode
,
4158 /* Get an inode object given its location and corresponding root.
4159 * Returns in *is_new if the inode was read from disk
4161 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4162 struct btrfs_root
*root
, int *new)
4164 struct inode
*inode
;
4166 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4168 return ERR_PTR(-ENOMEM
);
4170 if (inode
->i_state
& I_NEW
) {
4171 BTRFS_I(inode
)->root
= root
;
4172 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4173 btrfs_read_locked_inode(inode
);
4174 if (!is_bad_inode(inode
)) {
4175 inode_tree_add(inode
);
4176 unlock_new_inode(inode
);
4180 unlock_new_inode(inode
);
4182 inode
= ERR_PTR(-ESTALE
);
4189 static struct inode
*new_simple_dir(struct super_block
*s
,
4190 struct btrfs_key
*key
,
4191 struct btrfs_root
*root
)
4193 struct inode
*inode
= new_inode(s
);
4196 return ERR_PTR(-ENOMEM
);
4198 BTRFS_I(inode
)->root
= root
;
4199 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4200 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
4202 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4203 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
4204 inode
->i_fop
= &simple_dir_operations
;
4205 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4206 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4211 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4213 struct inode
*inode
;
4214 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4215 struct btrfs_root
*sub_root
= root
;
4216 struct btrfs_key location
;
4220 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4221 return ERR_PTR(-ENAMETOOLONG
);
4223 if (unlikely(d_need_lookup(dentry
))) {
4224 memcpy(&location
, dentry
->d_fsdata
, sizeof(struct btrfs_key
));
4225 kfree(dentry
->d_fsdata
);
4226 dentry
->d_fsdata
= NULL
;
4227 /* This thing is hashed, drop it for now */
4230 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4234 return ERR_PTR(ret
);
4236 if (location
.objectid
== 0)
4239 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4240 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4244 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4246 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4247 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4248 &location
, &sub_root
);
4251 inode
= ERR_PTR(ret
);
4253 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4255 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4257 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4259 if (!IS_ERR(inode
) && root
!= sub_root
) {
4260 down_read(&root
->fs_info
->cleanup_work_sem
);
4261 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4262 ret
= btrfs_orphan_cleanup(sub_root
);
4263 up_read(&root
->fs_info
->cleanup_work_sem
);
4265 inode
= ERR_PTR(ret
);
4271 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4273 struct btrfs_root
*root
;
4274 struct inode
*inode
= dentry
->d_inode
;
4276 if (!inode
&& !IS_ROOT(dentry
))
4277 inode
= dentry
->d_parent
->d_inode
;
4280 root
= BTRFS_I(inode
)->root
;
4281 if (btrfs_root_refs(&root
->root_item
) == 0)
4284 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
4290 static void btrfs_dentry_release(struct dentry
*dentry
)
4292 if (dentry
->d_fsdata
)
4293 kfree(dentry
->d_fsdata
);
4296 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4301 ret
= d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
4302 if (unlikely(d_need_lookup(dentry
))) {
4303 spin_lock(&dentry
->d_lock
);
4304 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
4305 spin_unlock(&dentry
->d_lock
);
4310 unsigned char btrfs_filetype_table
[] = {
4311 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4314 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4317 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4318 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4319 struct btrfs_item
*item
;
4320 struct btrfs_dir_item
*di
;
4321 struct btrfs_key key
;
4322 struct btrfs_key found_key
;
4323 struct btrfs_path
*path
;
4324 struct list_head ins_list
;
4325 struct list_head del_list
;
4327 struct extent_buffer
*leaf
;
4329 unsigned char d_type
;
4334 int key_type
= BTRFS_DIR_INDEX_KEY
;
4338 int is_curr
= 0; /* filp->f_pos points to the current index? */
4340 /* FIXME, use a real flag for deciding about the key type */
4341 if (root
->fs_info
->tree_root
== root
)
4342 key_type
= BTRFS_DIR_ITEM_KEY
;
4344 /* special case for "." */
4345 if (filp
->f_pos
== 0) {
4346 over
= filldir(dirent
, ".", 1,
4347 filp
->f_pos
, btrfs_ino(inode
), DT_DIR
);
4352 /* special case for .., just use the back ref */
4353 if (filp
->f_pos
== 1) {
4354 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4355 over
= filldir(dirent
, "..", 2,
4356 filp
->f_pos
, pino
, DT_DIR
);
4361 path
= btrfs_alloc_path();
4367 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4368 INIT_LIST_HEAD(&ins_list
);
4369 INIT_LIST_HEAD(&del_list
);
4370 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
4373 btrfs_set_key_type(&key
, key_type
);
4374 key
.offset
= filp
->f_pos
;
4375 key
.objectid
= btrfs_ino(inode
);
4377 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4382 leaf
= path
->nodes
[0];
4383 slot
= path
->slots
[0];
4384 if (slot
>= btrfs_header_nritems(leaf
)) {
4385 ret
= btrfs_next_leaf(root
, path
);
4393 item
= btrfs_item_nr(leaf
, slot
);
4394 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4396 if (found_key
.objectid
!= key
.objectid
)
4398 if (btrfs_key_type(&found_key
) != key_type
)
4400 if (found_key
.offset
< filp
->f_pos
)
4402 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
4403 btrfs_should_delete_dir_index(&del_list
,
4407 filp
->f_pos
= found_key
.offset
;
4410 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4412 di_total
= btrfs_item_size(leaf
, item
);
4414 while (di_cur
< di_total
) {
4415 struct btrfs_key location
;
4417 if (verify_dir_item(root
, leaf
, di
))
4420 name_len
= btrfs_dir_name_len(leaf
, di
);
4421 if (name_len
<= sizeof(tmp_name
)) {
4422 name_ptr
= tmp_name
;
4424 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4430 read_extent_buffer(leaf
, name_ptr
,
4431 (unsigned long)(di
+ 1), name_len
);
4433 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4434 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4437 /* is this a reference to our own snapshot? If so
4440 * In contrast to old kernels, we insert the snapshot's
4441 * dir item and dir index after it has been created, so
4442 * we won't find a reference to our own snapshot. We
4443 * still keep the following code for backward
4446 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4447 location
.objectid
== root
->root_key
.objectid
) {
4451 over
= filldir(dirent
, name_ptr
, name_len
,
4452 found_key
.offset
, location
.objectid
,
4456 if (name_ptr
!= tmp_name
)
4461 di_len
= btrfs_dir_name_len(leaf
, di
) +
4462 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4464 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4470 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4473 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
4479 /* Reached end of directory/root. Bump pos past the last item. */
4480 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4482 * 32-bit glibc will use getdents64, but then strtol -
4483 * so the last number we can serve is this.
4485 filp
->f_pos
= 0x7fffffff;
4491 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4492 btrfs_put_delayed_items(&ins_list
, &del_list
);
4493 btrfs_free_path(path
);
4497 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4499 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4500 struct btrfs_trans_handle
*trans
;
4502 bool nolock
= false;
4504 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
4507 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
4510 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4512 trans
= btrfs_join_transaction_nolock(root
);
4514 trans
= btrfs_join_transaction(root
);
4516 return PTR_ERR(trans
);
4517 ret
= btrfs_commit_transaction(trans
, root
);
4523 * This is somewhat expensive, updating the tree every time the
4524 * inode changes. But, it is most likely to find the inode in cache.
4525 * FIXME, needs more benchmarking...there are no reasons other than performance
4526 * to keep or drop this code.
4528 int btrfs_dirty_inode(struct inode
*inode
)
4530 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4531 struct btrfs_trans_handle
*trans
;
4534 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
4537 trans
= btrfs_join_transaction(root
);
4539 return PTR_ERR(trans
);
4541 ret
= btrfs_update_inode(trans
, root
, inode
);
4542 if (ret
&& ret
== -ENOSPC
) {
4543 /* whoops, lets try again with the full transaction */
4544 btrfs_end_transaction(trans
, root
);
4545 trans
= btrfs_start_transaction(root
, 1);
4547 return PTR_ERR(trans
);
4549 ret
= btrfs_update_inode(trans
, root
, inode
);
4551 btrfs_end_transaction(trans
, root
);
4552 if (BTRFS_I(inode
)->delayed_node
)
4553 btrfs_balance_delayed_items(root
);
4559 * This is a copy of file_update_time. We need this so we can return error on
4560 * ENOSPC for updating the inode in the case of file write and mmap writes.
4562 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
4565 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4567 if (btrfs_root_readonly(root
))
4570 if (flags
& S_VERSION
)
4571 inode_inc_iversion(inode
);
4572 if (flags
& S_CTIME
)
4573 inode
->i_ctime
= *now
;
4574 if (flags
& S_MTIME
)
4575 inode
->i_mtime
= *now
;
4576 if (flags
& S_ATIME
)
4577 inode
->i_atime
= *now
;
4578 return btrfs_dirty_inode(inode
);
4582 * find the highest existing sequence number in a directory
4583 * and then set the in-memory index_cnt variable to reflect
4584 * free sequence numbers
4586 static int btrfs_set_inode_index_count(struct inode
*inode
)
4588 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4589 struct btrfs_key key
, found_key
;
4590 struct btrfs_path
*path
;
4591 struct extent_buffer
*leaf
;
4594 key
.objectid
= btrfs_ino(inode
);
4595 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4596 key
.offset
= (u64
)-1;
4598 path
= btrfs_alloc_path();
4602 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4605 /* FIXME: we should be able to handle this */
4611 * MAGIC NUMBER EXPLANATION:
4612 * since we search a directory based on f_pos we have to start at 2
4613 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4614 * else has to start at 2
4616 if (path
->slots
[0] == 0) {
4617 BTRFS_I(inode
)->index_cnt
= 2;
4623 leaf
= path
->nodes
[0];
4624 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4626 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4627 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4628 BTRFS_I(inode
)->index_cnt
= 2;
4632 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4634 btrfs_free_path(path
);
4639 * helper to find a free sequence number in a given directory. This current
4640 * code is very simple, later versions will do smarter things in the btree
4642 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4646 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4647 ret
= btrfs_inode_delayed_dir_index_count(dir
);
4649 ret
= btrfs_set_inode_index_count(dir
);
4655 *index
= BTRFS_I(dir
)->index_cnt
;
4656 BTRFS_I(dir
)->index_cnt
++;
4661 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4662 struct btrfs_root
*root
,
4664 const char *name
, int name_len
,
4665 u64 ref_objectid
, u64 objectid
,
4666 umode_t mode
, u64
*index
)
4668 struct inode
*inode
;
4669 struct btrfs_inode_item
*inode_item
;
4670 struct btrfs_key
*location
;
4671 struct btrfs_path
*path
;
4672 struct btrfs_inode_ref
*ref
;
4673 struct btrfs_key key
[2];
4679 path
= btrfs_alloc_path();
4681 return ERR_PTR(-ENOMEM
);
4683 inode
= new_inode(root
->fs_info
->sb
);
4685 btrfs_free_path(path
);
4686 return ERR_PTR(-ENOMEM
);
4690 * we have to initialize this early, so we can reclaim the inode
4691 * number if we fail afterwards in this function.
4693 inode
->i_ino
= objectid
;
4696 trace_btrfs_inode_request(dir
);
4698 ret
= btrfs_set_inode_index(dir
, index
);
4700 btrfs_free_path(path
);
4702 return ERR_PTR(ret
);
4706 * index_cnt is ignored for everything but a dir,
4707 * btrfs_get_inode_index_count has an explanation for the magic
4710 BTRFS_I(inode
)->index_cnt
= 2;
4711 BTRFS_I(inode
)->root
= root
;
4712 BTRFS_I(inode
)->generation
= trans
->transid
;
4713 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4716 * We could have gotten an inode number from somebody who was fsynced
4717 * and then removed in this same transaction, so let's just set full
4718 * sync since it will be a full sync anyway and this will blow away the
4719 * old info in the log.
4721 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
4728 key
[0].objectid
= objectid
;
4729 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4733 * Start new inodes with an inode_ref. This is slightly more
4734 * efficient for small numbers of hard links since they will
4735 * be packed into one item. Extended refs will kick in if we
4736 * add more hard links than can fit in the ref item.
4738 key
[1].objectid
= objectid
;
4739 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4740 key
[1].offset
= ref_objectid
;
4742 sizes
[0] = sizeof(struct btrfs_inode_item
);
4743 sizes
[1] = name_len
+ sizeof(*ref
);
4745 path
->leave_spinning
= 1;
4746 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4750 inode_init_owner(inode
, dir
, mode
);
4751 inode_set_bytes(inode
, 0);
4752 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4753 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4754 struct btrfs_inode_item
);
4755 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
4756 sizeof(*inode_item
));
4757 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4759 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4760 struct btrfs_inode_ref
);
4761 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4762 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4763 ptr
= (unsigned long)(ref
+ 1);
4764 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4766 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4767 btrfs_free_path(path
);
4769 location
= &BTRFS_I(inode
)->location
;
4770 location
->objectid
= objectid
;
4771 location
->offset
= 0;
4772 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4774 btrfs_inherit_iflags(inode
, dir
);
4776 if (S_ISREG(mode
)) {
4777 if (btrfs_test_opt(root
, NODATASUM
))
4778 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4779 if (btrfs_test_opt(root
, NODATACOW
) ||
4780 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4781 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4784 insert_inode_hash(inode
);
4785 inode_tree_add(inode
);
4787 trace_btrfs_inode_new(inode
);
4788 btrfs_set_inode_last_trans(trans
, inode
);
4790 btrfs_update_root_times(trans
, root
);
4795 BTRFS_I(dir
)->index_cnt
--;
4796 btrfs_free_path(path
);
4798 return ERR_PTR(ret
);
4801 static inline u8
btrfs_inode_type(struct inode
*inode
)
4803 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4807 * utility function to add 'inode' into 'parent_inode' with
4808 * a give name and a given sequence number.
4809 * if 'add_backref' is true, also insert a backref from the
4810 * inode to the parent directory.
4812 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4813 struct inode
*parent_inode
, struct inode
*inode
,
4814 const char *name
, int name_len
, int add_backref
, u64 index
)
4817 struct btrfs_key key
;
4818 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4819 u64 ino
= btrfs_ino(inode
);
4820 u64 parent_ino
= btrfs_ino(parent_inode
);
4822 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4823 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4826 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4830 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4831 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4832 key
.objectid
, root
->root_key
.objectid
,
4833 parent_ino
, index
, name
, name_len
);
4834 } else if (add_backref
) {
4835 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
4839 /* Nothing to clean up yet */
4843 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4845 btrfs_inode_type(inode
), index
);
4849 btrfs_abort_transaction(trans
, root
, ret
);
4853 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4855 inode_inc_iversion(parent_inode
);
4856 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4857 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4859 btrfs_abort_transaction(trans
, root
, ret
);
4863 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4866 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4867 key
.objectid
, root
->root_key
.objectid
,
4868 parent_ino
, &local_index
, name
, name_len
);
4870 } else if (add_backref
) {
4874 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
4875 ino
, parent_ino
, &local_index
);
4880 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4881 struct inode
*dir
, struct dentry
*dentry
,
4882 struct inode
*inode
, int backref
, u64 index
)
4884 int err
= btrfs_add_link(trans
, dir
, inode
,
4885 dentry
->d_name
.name
, dentry
->d_name
.len
,
4892 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4893 umode_t mode
, dev_t rdev
)
4895 struct btrfs_trans_handle
*trans
;
4896 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4897 struct inode
*inode
= NULL
;
4901 unsigned long nr
= 0;
4904 if (!new_valid_dev(rdev
))
4908 * 2 for inode item and ref
4910 * 1 for xattr if selinux is on
4912 trans
= btrfs_start_transaction(root
, 5);
4914 return PTR_ERR(trans
);
4916 err
= btrfs_find_free_ino(root
, &objectid
);
4920 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4921 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4923 if (IS_ERR(inode
)) {
4924 err
= PTR_ERR(inode
);
4928 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4935 * If the active LSM wants to access the inode during
4936 * d_instantiate it needs these. Smack checks to see
4937 * if the filesystem supports xattrs by looking at the
4941 inode
->i_op
= &btrfs_special_inode_operations
;
4942 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4946 init_special_inode(inode
, inode
->i_mode
, rdev
);
4947 btrfs_update_inode(trans
, root
, inode
);
4948 d_instantiate(dentry
, inode
);
4951 nr
= trans
->blocks_used
;
4952 btrfs_end_transaction(trans
, root
);
4953 btrfs_btree_balance_dirty(root
, nr
);
4955 inode_dec_link_count(inode
);
4961 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4962 umode_t mode
, bool excl
)
4964 struct btrfs_trans_handle
*trans
;
4965 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4966 struct inode
*inode
= NULL
;
4969 unsigned long nr
= 0;
4974 * 2 for inode item and ref
4976 * 1 for xattr if selinux is on
4978 trans
= btrfs_start_transaction(root
, 5);
4980 return PTR_ERR(trans
);
4982 err
= btrfs_find_free_ino(root
, &objectid
);
4986 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4987 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4989 if (IS_ERR(inode
)) {
4990 err
= PTR_ERR(inode
);
4994 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5001 * If the active LSM wants to access the inode during
5002 * d_instantiate it needs these. Smack checks to see
5003 * if the filesystem supports xattrs by looking at the
5006 inode
->i_fop
= &btrfs_file_operations
;
5007 inode
->i_op
= &btrfs_file_inode_operations
;
5009 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
5013 inode
->i_mapping
->a_ops
= &btrfs_aops
;
5014 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
5015 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
5016 d_instantiate(dentry
, inode
);
5019 nr
= trans
->blocks_used
;
5020 btrfs_end_transaction(trans
, root
);
5022 inode_dec_link_count(inode
);
5025 btrfs_btree_balance_dirty(root
, nr
);
5029 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
5030 struct dentry
*dentry
)
5032 struct btrfs_trans_handle
*trans
;
5033 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5034 struct inode
*inode
= old_dentry
->d_inode
;
5036 unsigned long nr
= 0;
5040 /* do not allow sys_link's with other subvols of the same device */
5041 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
5044 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
5047 err
= btrfs_set_inode_index(dir
, &index
);
5052 * 2 items for inode and inode ref
5053 * 2 items for dir items
5054 * 1 item for parent inode
5056 trans
= btrfs_start_transaction(root
, 5);
5057 if (IS_ERR(trans
)) {
5058 err
= PTR_ERR(trans
);
5062 btrfs_inc_nlink(inode
);
5063 inode_inc_iversion(inode
);
5064 inode
->i_ctime
= CURRENT_TIME
;
5067 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
5072 struct dentry
*parent
= dentry
->d_parent
;
5073 err
= btrfs_update_inode(trans
, root
, inode
);
5076 d_instantiate(dentry
, inode
);
5077 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
5080 nr
= trans
->blocks_used
;
5081 btrfs_end_transaction(trans
, root
);
5084 inode_dec_link_count(inode
);
5087 btrfs_btree_balance_dirty(root
, nr
);
5091 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
5093 struct inode
*inode
= NULL
;
5094 struct btrfs_trans_handle
*trans
;
5095 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5097 int drop_on_err
= 0;
5100 unsigned long nr
= 1;
5103 * 2 items for inode and ref
5104 * 2 items for dir items
5105 * 1 for xattr if selinux is on
5107 trans
= btrfs_start_transaction(root
, 5);
5109 return PTR_ERR(trans
);
5111 err
= btrfs_find_free_ino(root
, &objectid
);
5115 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5116 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5117 S_IFDIR
| mode
, &index
);
5118 if (IS_ERR(inode
)) {
5119 err
= PTR_ERR(inode
);
5125 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5129 inode
->i_op
= &btrfs_dir_inode_operations
;
5130 inode
->i_fop
= &btrfs_dir_file_operations
;
5132 btrfs_i_size_write(inode
, 0);
5133 err
= btrfs_update_inode(trans
, root
, inode
);
5137 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
5138 dentry
->d_name
.len
, 0, index
);
5142 d_instantiate(dentry
, inode
);
5146 nr
= trans
->blocks_used
;
5147 btrfs_end_transaction(trans
, root
);
5150 btrfs_btree_balance_dirty(root
, nr
);
5154 /* helper for btfs_get_extent. Given an existing extent in the tree,
5155 * and an extent that you want to insert, deal with overlap and insert
5156 * the new extent into the tree.
5158 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
5159 struct extent_map
*existing
,
5160 struct extent_map
*em
,
5161 u64 map_start
, u64 map_len
)
5165 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
5166 start_diff
= map_start
- em
->start
;
5167 em
->start
= map_start
;
5169 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
5170 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
5171 em
->block_start
+= start_diff
;
5172 em
->block_len
-= start_diff
;
5174 return add_extent_mapping(em_tree
, em
);
5177 static noinline
int uncompress_inline(struct btrfs_path
*path
,
5178 struct inode
*inode
, struct page
*page
,
5179 size_t pg_offset
, u64 extent_offset
,
5180 struct btrfs_file_extent_item
*item
)
5183 struct extent_buffer
*leaf
= path
->nodes
[0];
5186 unsigned long inline_size
;
5190 WARN_ON(pg_offset
!= 0);
5191 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5192 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
5193 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
5194 btrfs_item_nr(leaf
, path
->slots
[0]));
5195 tmp
= kmalloc(inline_size
, GFP_NOFS
);
5198 ptr
= btrfs_file_extent_inline_start(item
);
5200 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
5202 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
5203 ret
= btrfs_decompress(compress_type
, tmp
, page
,
5204 extent_offset
, inline_size
, max_size
);
5206 char *kaddr
= kmap_atomic(page
);
5207 unsigned long copy_size
= min_t(u64
,
5208 PAGE_CACHE_SIZE
- pg_offset
,
5209 max_size
- extent_offset
);
5210 memset(kaddr
+ pg_offset
, 0, copy_size
);
5211 kunmap_atomic(kaddr
);
5218 * a bit scary, this does extent mapping from logical file offset to the disk.
5219 * the ugly parts come from merging extents from the disk with the in-ram
5220 * representation. This gets more complex because of the data=ordered code,
5221 * where the in-ram extents might be locked pending data=ordered completion.
5223 * This also copies inline extents directly into the page.
5226 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
5227 size_t pg_offset
, u64 start
, u64 len
,
5233 u64 extent_start
= 0;
5235 u64 objectid
= btrfs_ino(inode
);
5237 struct btrfs_path
*path
= NULL
;
5238 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5239 struct btrfs_file_extent_item
*item
;
5240 struct extent_buffer
*leaf
;
5241 struct btrfs_key found_key
;
5242 struct extent_map
*em
= NULL
;
5243 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5244 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5245 struct btrfs_trans_handle
*trans
= NULL
;
5249 read_lock(&em_tree
->lock
);
5250 em
= lookup_extent_mapping(em_tree
, start
, len
);
5252 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5253 read_unlock(&em_tree
->lock
);
5256 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5257 free_extent_map(em
);
5258 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5259 free_extent_map(em
);
5263 em
= alloc_extent_map();
5268 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5269 em
->start
= EXTENT_MAP_HOLE
;
5270 em
->orig_start
= EXTENT_MAP_HOLE
;
5272 em
->block_len
= (u64
)-1;
5275 path
= btrfs_alloc_path();
5281 * Chances are we'll be called again, so go ahead and do
5287 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5288 objectid
, start
, trans
!= NULL
);
5295 if (path
->slots
[0] == 0)
5300 leaf
= path
->nodes
[0];
5301 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5302 struct btrfs_file_extent_item
);
5303 /* are we inside the extent that was found? */
5304 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5305 found_type
= btrfs_key_type(&found_key
);
5306 if (found_key
.objectid
!= objectid
||
5307 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5311 found_type
= btrfs_file_extent_type(leaf
, item
);
5312 extent_start
= found_key
.offset
;
5313 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5314 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5315 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5316 extent_end
= extent_start
+
5317 btrfs_file_extent_num_bytes(leaf
, item
);
5318 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5320 size
= btrfs_file_extent_inline_len(leaf
, item
);
5321 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5322 ~((u64
)root
->sectorsize
- 1);
5325 if (start
>= extent_end
) {
5327 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5328 ret
= btrfs_next_leaf(root
, path
);
5335 leaf
= path
->nodes
[0];
5337 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5338 if (found_key
.objectid
!= objectid
||
5339 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5341 if (start
+ len
<= found_key
.offset
)
5344 em
->len
= found_key
.offset
- start
;
5348 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5349 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5350 em
->start
= extent_start
;
5351 em
->len
= extent_end
- extent_start
;
5352 em
->orig_start
= extent_start
-
5353 btrfs_file_extent_offset(leaf
, item
);
5354 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5356 em
->block_start
= EXTENT_MAP_HOLE
;
5359 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5360 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5361 em
->compress_type
= compress_type
;
5362 em
->block_start
= bytenr
;
5363 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5366 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5367 em
->block_start
= bytenr
;
5368 em
->block_len
= em
->len
;
5369 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5370 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5373 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5377 size_t extent_offset
;
5380 em
->block_start
= EXTENT_MAP_INLINE
;
5381 if (!page
|| create
) {
5382 em
->start
= extent_start
;
5383 em
->len
= extent_end
- extent_start
;
5387 size
= btrfs_file_extent_inline_len(leaf
, item
);
5388 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5389 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5390 size
- extent_offset
);
5391 em
->start
= extent_start
+ extent_offset
;
5392 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5393 ~((u64
)root
->sectorsize
- 1);
5394 em
->orig_start
= EXTENT_MAP_INLINE
;
5395 if (compress_type
) {
5396 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5397 em
->compress_type
= compress_type
;
5399 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5400 if (create
== 0 && !PageUptodate(page
)) {
5401 if (btrfs_file_extent_compression(leaf
, item
) !=
5402 BTRFS_COMPRESS_NONE
) {
5403 ret
= uncompress_inline(path
, inode
, page
,
5405 extent_offset
, item
);
5406 BUG_ON(ret
); /* -ENOMEM */
5409 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5411 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5412 memset(map
+ pg_offset
+ copy_size
, 0,
5413 PAGE_CACHE_SIZE
- pg_offset
-
5418 flush_dcache_page(page
);
5419 } else if (create
&& PageUptodate(page
)) {
5423 free_extent_map(em
);
5426 btrfs_release_path(path
);
5427 trans
= btrfs_join_transaction(root
);
5430 return ERR_CAST(trans
);
5434 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5437 btrfs_mark_buffer_dirty(leaf
);
5439 set_extent_uptodate(io_tree
, em
->start
,
5440 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5443 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5450 em
->block_start
= EXTENT_MAP_HOLE
;
5451 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5453 btrfs_release_path(path
);
5454 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5455 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5456 "[%llu %llu]\n", (unsigned long long)em
->start
,
5457 (unsigned long long)em
->len
,
5458 (unsigned long long)start
,
5459 (unsigned long long)len
);
5465 write_lock(&em_tree
->lock
);
5466 ret
= add_extent_mapping(em_tree
, em
);
5467 /* it is possible that someone inserted the extent into the tree
5468 * while we had the lock dropped. It is also possible that
5469 * an overlapping map exists in the tree
5471 if (ret
== -EEXIST
) {
5472 struct extent_map
*existing
;
5476 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5477 if (existing
&& (existing
->start
> start
||
5478 existing
->start
+ existing
->len
<= start
)) {
5479 free_extent_map(existing
);
5483 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5486 err
= merge_extent_mapping(em_tree
, existing
,
5489 free_extent_map(existing
);
5491 free_extent_map(em
);
5496 free_extent_map(em
);
5500 free_extent_map(em
);
5505 write_unlock(&em_tree
->lock
);
5509 trace_btrfs_get_extent(root
, em
);
5512 btrfs_free_path(path
);
5514 ret
= btrfs_end_transaction(trans
, root
);
5519 free_extent_map(em
);
5520 return ERR_PTR(err
);
5522 BUG_ON(!em
); /* Error is always set */
5526 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5527 size_t pg_offset
, u64 start
, u64 len
,
5530 struct extent_map
*em
;
5531 struct extent_map
*hole_em
= NULL
;
5532 u64 range_start
= start
;
5538 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5543 * if our em maps to a hole, there might
5544 * actually be delalloc bytes behind it
5546 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5552 /* check to see if we've wrapped (len == -1 or similar) */
5561 /* ok, we didn't find anything, lets look for delalloc */
5562 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5563 end
, len
, EXTENT_DELALLOC
, 1);
5564 found_end
= range_start
+ found
;
5565 if (found_end
< range_start
)
5566 found_end
= (u64
)-1;
5569 * we didn't find anything useful, return
5570 * the original results from get_extent()
5572 if (range_start
> end
|| found_end
<= start
) {
5578 /* adjust the range_start to make sure it doesn't
5579 * go backwards from the start they passed in
5581 range_start
= max(start
,range_start
);
5582 found
= found_end
- range_start
;
5585 u64 hole_start
= start
;
5588 em
= alloc_extent_map();
5594 * when btrfs_get_extent can't find anything it
5595 * returns one huge hole
5597 * make sure what it found really fits our range, and
5598 * adjust to make sure it is based on the start from
5602 u64 calc_end
= extent_map_end(hole_em
);
5604 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5605 free_extent_map(hole_em
);
5608 hole_start
= max(hole_em
->start
, start
);
5609 hole_len
= calc_end
- hole_start
;
5613 if (hole_em
&& range_start
> hole_start
) {
5614 /* our hole starts before our delalloc, so we
5615 * have to return just the parts of the hole
5616 * that go until the delalloc starts
5618 em
->len
= min(hole_len
,
5619 range_start
- hole_start
);
5620 em
->start
= hole_start
;
5621 em
->orig_start
= hole_start
;
5623 * don't adjust block start at all,
5624 * it is fixed at EXTENT_MAP_HOLE
5626 em
->block_start
= hole_em
->block_start
;
5627 em
->block_len
= hole_len
;
5629 em
->start
= range_start
;
5631 em
->orig_start
= range_start
;
5632 em
->block_start
= EXTENT_MAP_DELALLOC
;
5633 em
->block_len
= found
;
5635 } else if (hole_em
) {
5640 free_extent_map(hole_em
);
5642 free_extent_map(em
);
5643 return ERR_PTR(err
);
5648 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5649 struct extent_map
*em
,
5652 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5653 struct btrfs_trans_handle
*trans
;
5654 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5655 struct btrfs_key ins
;
5658 bool insert
= false;
5661 * Ok if the extent map we looked up is a hole and is for the exact
5662 * range we want, there is no reason to allocate a new one, however if
5663 * it is not right then we need to free this one and drop the cache for
5666 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5668 free_extent_map(em
);
5671 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5674 trans
= btrfs_join_transaction(root
);
5676 return ERR_CAST(trans
);
5678 if (start
<= BTRFS_I(inode
)->disk_i_size
&& len
< 64 * 1024)
5679 btrfs_add_inode_defrag(trans
, inode
);
5681 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5683 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5684 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5685 alloc_hint
, &ins
, 1);
5692 em
= alloc_extent_map();
5694 em
= ERR_PTR(-ENOMEM
);
5700 em
->orig_start
= em
->start
;
5701 em
->len
= ins
.offset
;
5703 em
->block_start
= ins
.objectid
;
5704 em
->block_len
= ins
.offset
;
5705 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5708 * We need to do this because if we're using the original em we searched
5709 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5712 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5715 write_lock(&em_tree
->lock
);
5716 ret
= add_extent_mapping(em_tree
, em
);
5717 write_unlock(&em_tree
->lock
);
5720 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5723 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5724 ins
.offset
, ins
.offset
, 0);
5726 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5730 btrfs_end_transaction(trans
, root
);
5735 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5736 * block must be cow'd
5738 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5739 struct inode
*inode
, u64 offset
, u64 len
)
5741 struct btrfs_path
*path
;
5743 struct extent_buffer
*leaf
;
5744 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5745 struct btrfs_file_extent_item
*fi
;
5746 struct btrfs_key key
;
5754 path
= btrfs_alloc_path();
5758 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
5763 slot
= path
->slots
[0];
5766 /* can't find the item, must cow */
5773 leaf
= path
->nodes
[0];
5774 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5775 if (key
.objectid
!= btrfs_ino(inode
) ||
5776 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5777 /* not our file or wrong item type, must cow */
5781 if (key
.offset
> offset
) {
5782 /* Wrong offset, must cow */
5786 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5787 found_type
= btrfs_file_extent_type(leaf
, fi
);
5788 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5789 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5790 /* not a regular extent, must cow */
5793 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5794 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5796 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5797 if (extent_end
< offset
+ len
) {
5798 /* extent doesn't include our full range, must cow */
5802 if (btrfs_extent_readonly(root
, disk_bytenr
))
5806 * look for other files referencing this extent, if we
5807 * find any we must cow
5809 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
5810 key
.offset
- backref_offset
, disk_bytenr
))
5814 * adjust disk_bytenr and num_bytes to cover just the bytes
5815 * in this extent we are about to write. If there
5816 * are any csums in that range we have to cow in order
5817 * to keep the csums correct
5819 disk_bytenr
+= backref_offset
;
5820 disk_bytenr
+= offset
- key
.offset
;
5821 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5822 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5825 * all of the above have passed, it is safe to overwrite this extent
5830 btrfs_free_path(path
);
5834 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
5835 struct extent_state
**cached_state
, int writing
)
5837 struct btrfs_ordered_extent
*ordered
;
5841 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
5844 * We're concerned with the entire range that we're going to be
5845 * doing DIO to, so we need to make sure theres no ordered
5846 * extents in this range.
5848 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
5849 lockend
- lockstart
+ 1);
5852 * We need to make sure there are no buffered pages in this
5853 * range either, we could have raced between the invalidate in
5854 * generic_file_direct_write and locking the extent. The
5855 * invalidate needs to happen so that reads after a write do not
5858 if (!ordered
&& (!writing
||
5859 !test_range_bit(&BTRFS_I(inode
)->io_tree
,
5860 lockstart
, lockend
, EXTENT_UPTODATE
, 0,
5864 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
5865 cached_state
, GFP_NOFS
);
5868 btrfs_start_ordered_extent(inode
, ordered
, 1);
5869 btrfs_put_ordered_extent(ordered
);
5871 /* Screw you mmap */
5872 ret
= filemap_write_and_wait_range(inode
->i_mapping
,
5879 * If we found a page that couldn't be invalidated just
5880 * fall back to buffered.
5882 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
5883 lockstart
>> PAGE_CACHE_SHIFT
,
5884 lockend
>> PAGE_CACHE_SHIFT
);
5895 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
5896 u64 len
, u64 orig_start
,
5897 u64 block_start
, u64 block_len
,
5900 struct extent_map_tree
*em_tree
;
5901 struct extent_map
*em
;
5902 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5905 em_tree
= &BTRFS_I(inode
)->extent_tree
;
5906 em
= alloc_extent_map();
5908 return ERR_PTR(-ENOMEM
);
5911 em
->orig_start
= orig_start
;
5913 em
->block_len
= block_len
;
5914 em
->block_start
= block_start
;
5915 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5916 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5917 if (type
== BTRFS_ORDERED_PREALLOC
)
5918 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5921 btrfs_drop_extent_cache(inode
, em
->start
,
5922 em
->start
+ em
->len
- 1, 0);
5923 write_lock(&em_tree
->lock
);
5924 ret
= add_extent_mapping(em_tree
, em
);
5925 write_unlock(&em_tree
->lock
);
5926 } while (ret
== -EEXIST
);
5929 free_extent_map(em
);
5930 return ERR_PTR(ret
);
5937 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5938 struct buffer_head
*bh_result
, int create
)
5940 struct extent_map
*em
;
5941 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5942 struct extent_state
*cached_state
= NULL
;
5943 u64 start
= iblock
<< inode
->i_blkbits
;
5944 u64 lockstart
, lockend
;
5945 u64 len
= bh_result
->b_size
;
5946 struct btrfs_trans_handle
*trans
;
5947 int unlock_bits
= EXTENT_LOCKED
;
5951 ret
= btrfs_delalloc_reserve_space(inode
, len
);
5954 unlock_bits
|= EXTENT_DELALLOC
| EXTENT_DIRTY
;
5956 len
= min_t(u64
, len
, root
->sectorsize
);
5960 lockend
= start
+ len
- 1;
5963 * If this errors out it's because we couldn't invalidate pagecache for
5964 * this range and we need to fallback to buffered.
5966 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
, create
))
5970 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
5971 lockend
, EXTENT_DELALLOC
, NULL
,
5972 &cached_state
, GFP_NOFS
);
5977 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5984 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5985 * io. INLINE is special, and we could probably kludge it in here, but
5986 * it's still buffered so for safety lets just fall back to the generic
5989 * For COMPRESSED we _have_ to read the entire extent in so we can
5990 * decompress it, so there will be buffering required no matter what we
5991 * do, so go ahead and fallback to buffered.
5993 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5994 * to buffered IO. Don't blame me, this is the price we pay for using
5997 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5998 em
->block_start
== EXTENT_MAP_INLINE
) {
5999 free_extent_map(em
);
6004 /* Just a good old fashioned hole, return */
6005 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
6006 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
6007 free_extent_map(em
);
6013 * We don't allocate a new extent in the following cases
6015 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6017 * 2) The extent is marked as PREALLOC. We're good to go here and can
6018 * just use the extent.
6022 len
= min(len
, em
->len
- (start
- em
->start
));
6023 lockstart
= start
+ len
;
6027 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
6028 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
6029 em
->block_start
!= EXTENT_MAP_HOLE
)) {
6034 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6035 type
= BTRFS_ORDERED_PREALLOC
;
6037 type
= BTRFS_ORDERED_NOCOW
;
6038 len
= min(len
, em
->len
- (start
- em
->start
));
6039 block_start
= em
->block_start
+ (start
- em
->start
);
6042 * we're not going to log anything, but we do need
6043 * to make sure the current transaction stays open
6044 * while we look for nocow cross refs
6046 trans
= btrfs_join_transaction(root
);
6050 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
6051 u64 orig_start
= em
->start
;
6053 if (type
== BTRFS_ORDERED_PREALLOC
) {
6054 free_extent_map(em
);
6055 em
= create_pinned_em(inode
, start
, len
,
6057 block_start
, len
, type
);
6059 btrfs_end_transaction(trans
, root
);
6064 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
6065 block_start
, len
, len
, type
);
6066 btrfs_end_transaction(trans
, root
);
6068 free_extent_map(em
);
6073 btrfs_end_transaction(trans
, root
);
6077 * this will cow the extent, reset the len in case we changed
6080 len
= bh_result
->b_size
;
6081 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
6086 len
= min(len
, em
->len
- (start
- em
->start
));
6088 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
6090 bh_result
->b_size
= len
;
6091 bh_result
->b_bdev
= em
->bdev
;
6092 set_buffer_mapped(bh_result
);
6094 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6095 set_buffer_new(bh_result
);
6098 * Need to update the i_size under the extent lock so buffered
6099 * readers will get the updated i_size when we unlock.
6101 if (start
+ len
> i_size_read(inode
))
6102 i_size_write(inode
, start
+ len
);
6106 * In the case of write we need to clear and unlock the entire range,
6107 * in the case of read we need to unlock only the end area that we
6108 * aren't using if there is any left over space.
6110 if (lockstart
< lockend
) {
6111 if (create
&& len
< lockend
- lockstart
) {
6112 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6113 lockstart
+ len
- 1,
6114 unlock_bits
| EXTENT_DEFRAG
, 1, 0,
6115 &cached_state
, GFP_NOFS
);
6117 * Beside unlock, we also need to cleanup reserved space
6118 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6120 clear_extent_bit(&BTRFS_I(inode
)->io_tree
,
6121 lockstart
+ len
, lockend
,
6122 unlock_bits
| EXTENT_DO_ACCOUNTING
|
6123 EXTENT_DEFRAG
, 1, 0, NULL
, GFP_NOFS
);
6125 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6126 lockend
, unlock_bits
, 1, 0,
6127 &cached_state
, GFP_NOFS
);
6130 free_extent_state(cached_state
);
6133 free_extent_map(em
);
6139 unlock_bits
|= EXTENT_DO_ACCOUNTING
;
6141 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6142 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
6146 struct btrfs_dio_private
{
6147 struct inode
*inode
;
6153 /* number of bios pending for this dio */
6154 atomic_t pending_bios
;
6159 struct bio
*orig_bio
;
6162 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
6164 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6165 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
6166 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6167 struct inode
*inode
= dip
->inode
;
6168 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6171 start
= dip
->logical_offset
;
6173 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
6174 struct page
*page
= bvec
->bv_page
;
6177 u64
private = ~(u32
)0;
6178 unsigned long flags
;
6180 if (get_state_private(&BTRFS_I(inode
)->io_tree
,
6183 local_irq_save(flags
);
6184 kaddr
= kmap_atomic(page
);
6185 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
6186 csum
, bvec
->bv_len
);
6187 btrfs_csum_final(csum
, (char *)&csum
);
6188 kunmap_atomic(kaddr
);
6189 local_irq_restore(flags
);
6191 flush_dcache_page(bvec
->bv_page
);
6192 if (csum
!= private) {
6194 printk(KERN_ERR
"btrfs csum failed ino %llu off"
6195 " %llu csum %u private %u\n",
6196 (unsigned long long)btrfs_ino(inode
),
6197 (unsigned long long)start
,
6198 csum
, (unsigned)private);
6203 start
+= bvec
->bv_len
;
6205 } while (bvec
<= bvec_end
);
6207 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
6208 dip
->logical_offset
+ dip
->bytes
- 1);
6209 bio
->bi_private
= dip
->private;
6213 /* If we had a csum failure make sure to clear the uptodate flag */
6215 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
6216 dio_end_io(bio
, err
);
6219 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
6221 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6222 struct inode
*inode
= dip
->inode
;
6223 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6224 struct btrfs_ordered_extent
*ordered
= NULL
;
6225 u64 ordered_offset
= dip
->logical_offset
;
6226 u64 ordered_bytes
= dip
->bytes
;
6232 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
6234 ordered_bytes
, !err
);
6238 ordered
->work
.func
= finish_ordered_fn
;
6239 ordered
->work
.flags
= 0;
6240 btrfs_queue_worker(&root
->fs_info
->endio_write_workers
,
6244 * our bio might span multiple ordered extents. If we haven't
6245 * completed the accounting for the whole dio, go back and try again
6247 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
6248 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
6254 bio
->bi_private
= dip
->private;
6258 /* If we had an error make sure to clear the uptodate flag */
6260 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
6261 dio_end_io(bio
, err
);
6264 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
6265 struct bio
*bio
, int mirror_num
,
6266 unsigned long bio_flags
, u64 offset
)
6269 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6270 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
6271 BUG_ON(ret
); /* -ENOMEM */
6275 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
6277 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6280 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
6281 "sector %#Lx len %u err no %d\n",
6282 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
6283 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
6287 * before atomic variable goto zero, we must make sure
6288 * dip->errors is perceived to be set.
6290 smp_mb__before_atomic_dec();
6293 /* if there are more bios still pending for this dio, just exit */
6294 if (!atomic_dec_and_test(&dip
->pending_bios
))
6298 bio_io_error(dip
->orig_bio
);
6300 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
6301 bio_endio(dip
->orig_bio
, 0);
6307 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
6308 u64 first_sector
, gfp_t gfp_flags
)
6310 int nr_vecs
= bio_get_nr_vecs(bdev
);
6311 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
6314 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
6315 int rw
, u64 file_offset
, int skip_sum
,
6318 int write
= rw
& REQ_WRITE
;
6319 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6325 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
6333 if (write
&& async_submit
) {
6334 ret
= btrfs_wq_submit_bio(root
->fs_info
,
6335 inode
, rw
, bio
, 0, 0,
6337 __btrfs_submit_bio_start_direct_io
,
6338 __btrfs_submit_bio_done
);
6342 * If we aren't doing async submit, calculate the csum of the
6345 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
6348 } else if (!skip_sum
) {
6349 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
, file_offset
);
6355 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
6361 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
6364 struct inode
*inode
= dip
->inode
;
6365 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6366 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
6368 struct bio
*orig_bio
= dip
->orig_bio
;
6369 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
6370 u64 start_sector
= orig_bio
->bi_sector
;
6371 u64 file_offset
= dip
->logical_offset
;
6376 int async_submit
= 0;
6378 map_length
= orig_bio
->bi_size
;
6379 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6380 &map_length
, NULL
, 0);
6386 if (map_length
>= orig_bio
->bi_size
) {
6392 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
6395 bio
->bi_private
= dip
;
6396 bio
->bi_end_io
= btrfs_end_dio_bio
;
6397 atomic_inc(&dip
->pending_bios
);
6399 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
6400 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
6401 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
6402 bvec
->bv_offset
) < bvec
->bv_len
)) {
6404 * inc the count before we submit the bio so
6405 * we know the end IO handler won't happen before
6406 * we inc the count. Otherwise, the dip might get freed
6407 * before we're done setting it up
6409 atomic_inc(&dip
->pending_bios
);
6410 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
6411 file_offset
, skip_sum
,
6415 atomic_dec(&dip
->pending_bios
);
6419 start_sector
+= submit_len
>> 9;
6420 file_offset
+= submit_len
;
6425 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6426 start_sector
, GFP_NOFS
);
6429 bio
->bi_private
= dip
;
6430 bio
->bi_end_io
= btrfs_end_dio_bio
;
6432 map_length
= orig_bio
->bi_size
;
6433 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6434 &map_length
, NULL
, 0);
6440 submit_len
+= bvec
->bv_len
;
6447 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6456 * before atomic variable goto zero, we must
6457 * make sure dip->errors is perceived to be set.
6459 smp_mb__before_atomic_dec();
6460 if (atomic_dec_and_test(&dip
->pending_bios
))
6461 bio_io_error(dip
->orig_bio
);
6463 /* bio_end_io() will handle error, so we needn't return it */
6467 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6470 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6471 struct btrfs_dio_private
*dip
;
6472 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6474 int write
= rw
& REQ_WRITE
;
6477 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6479 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6485 dip
->private = bio
->bi_private
;
6487 dip
->logical_offset
= file_offset
;
6491 dip
->bytes
+= bvec
->bv_len
;
6493 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6495 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6496 bio
->bi_private
= dip
;
6498 dip
->orig_bio
= bio
;
6499 atomic_set(&dip
->pending_bios
, 0);
6502 bio
->bi_end_io
= btrfs_endio_direct_write
;
6504 bio
->bi_end_io
= btrfs_endio_direct_read
;
6506 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6511 * If this is a write, we need to clean up the reserved space and kill
6512 * the ordered extent.
6515 struct btrfs_ordered_extent
*ordered
;
6516 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6517 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6518 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6519 btrfs_free_reserved_extent(root
, ordered
->start
,
6521 btrfs_put_ordered_extent(ordered
);
6522 btrfs_put_ordered_extent(ordered
);
6524 bio_endio(bio
, ret
);
6527 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6528 const struct iovec
*iov
, loff_t offset
,
6529 unsigned long nr_segs
)
6535 unsigned blocksize_mask
= root
->sectorsize
- 1;
6536 ssize_t retval
= -EINVAL
;
6537 loff_t end
= offset
;
6539 if (offset
& blocksize_mask
)
6542 /* Check the memory alignment. Blocks cannot straddle pages */
6543 for (seg
= 0; seg
< nr_segs
; seg
++) {
6544 addr
= (unsigned long)iov
[seg
].iov_base
;
6545 size
= iov
[seg
].iov_len
;
6547 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6550 /* If this is a write we don't need to check anymore */
6555 * Check to make sure we don't have duplicate iov_base's in this
6556 * iovec, if so return EINVAL, otherwise we'll get csum errors
6557 * when reading back.
6559 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6560 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6569 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6570 const struct iovec
*iov
, loff_t offset
,
6571 unsigned long nr_segs
)
6573 struct file
*file
= iocb
->ki_filp
;
6574 struct inode
*inode
= file
->f_mapping
->host
;
6576 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6580 return __blockdev_direct_IO(rw
, iocb
, inode
,
6581 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6582 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6583 btrfs_submit_direct
, 0);
6586 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6587 __u64 start
, __u64 len
)
6589 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6592 int btrfs_readpage(struct file
*file
, struct page
*page
)
6594 struct extent_io_tree
*tree
;
6595 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6596 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
6599 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6601 struct extent_io_tree
*tree
;
6604 if (current
->flags
& PF_MEMALLOC
) {
6605 redirty_page_for_writepage(wbc
, page
);
6609 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6610 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6613 int btrfs_writepages(struct address_space
*mapping
,
6614 struct writeback_control
*wbc
)
6616 struct extent_io_tree
*tree
;
6618 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6619 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6623 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6624 struct list_head
*pages
, unsigned nr_pages
)
6626 struct extent_io_tree
*tree
;
6627 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6628 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6631 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6633 struct extent_io_tree
*tree
;
6634 struct extent_map_tree
*map
;
6637 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6638 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6639 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6641 ClearPagePrivate(page
);
6642 set_page_private(page
, 0);
6643 page_cache_release(page
);
6648 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6650 if (PageWriteback(page
) || PageDirty(page
))
6652 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6655 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6657 struct inode
*inode
= page
->mapping
->host
;
6658 struct extent_io_tree
*tree
;
6659 struct btrfs_ordered_extent
*ordered
;
6660 struct extent_state
*cached_state
= NULL
;
6661 u64 page_start
= page_offset(page
);
6662 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6665 * we have the page locked, so new writeback can't start,
6666 * and the dirty bit won't be cleared while we are here.
6668 * Wait for IO on this page so that we can safely clear
6669 * the PagePrivate2 bit and do ordered accounting
6671 wait_on_page_writeback(page
);
6673 tree
= &BTRFS_I(inode
)->io_tree
;
6675 btrfs_releasepage(page
, GFP_NOFS
);
6678 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
6679 ordered
= btrfs_lookup_ordered_extent(inode
,
6683 * IO on this page will never be started, so we need
6684 * to account for any ordered extents now
6686 clear_extent_bit(tree
, page_start
, page_end
,
6687 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6688 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
6689 EXTENT_DEFRAG
, 1, 0, &cached_state
, GFP_NOFS
);
6691 * whoever cleared the private bit is responsible
6692 * for the finish_ordered_io
6694 if (TestClearPagePrivate2(page
) &&
6695 btrfs_dec_test_ordered_pending(inode
, &ordered
, page_start
,
6696 PAGE_CACHE_SIZE
, 1)) {
6697 btrfs_finish_ordered_io(ordered
);
6699 btrfs_put_ordered_extent(ordered
);
6700 cached_state
= NULL
;
6701 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
6703 clear_extent_bit(tree
, page_start
, page_end
,
6704 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6705 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
6706 &cached_state
, GFP_NOFS
);
6707 __btrfs_releasepage(page
, GFP_NOFS
);
6709 ClearPageChecked(page
);
6710 if (PagePrivate(page
)) {
6711 ClearPagePrivate(page
);
6712 set_page_private(page
, 0);
6713 page_cache_release(page
);
6718 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6719 * called from a page fault handler when a page is first dirtied. Hence we must
6720 * be careful to check for EOF conditions here. We set the page up correctly
6721 * for a written page which means we get ENOSPC checking when writing into
6722 * holes and correct delalloc and unwritten extent mapping on filesystems that
6723 * support these features.
6725 * We are not allowed to take the i_mutex here so we have to play games to
6726 * protect against truncate races as the page could now be beyond EOF. Because
6727 * vmtruncate() writes the inode size before removing pages, once we have the
6728 * page lock we can determine safely if the page is beyond EOF. If it is not
6729 * beyond EOF, then the page is guaranteed safe against truncation until we
6732 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6734 struct page
*page
= vmf
->page
;
6735 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6736 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6737 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6738 struct btrfs_ordered_extent
*ordered
;
6739 struct extent_state
*cached_state
= NULL
;
6741 unsigned long zero_start
;
6748 sb_start_pagefault(inode
->i_sb
);
6749 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6751 ret
= file_update_time(vma
->vm_file
);
6757 else /* -ENOSPC, -EIO, etc */
6758 ret
= VM_FAULT_SIGBUS
;
6764 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6767 size
= i_size_read(inode
);
6768 page_start
= page_offset(page
);
6769 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6771 if ((page
->mapping
!= inode
->i_mapping
) ||
6772 (page_start
>= size
)) {
6773 /* page got truncated out from underneath us */
6776 wait_on_page_writeback(page
);
6778 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
6779 set_page_extent_mapped(page
);
6782 * we can't set the delalloc bits if there are pending ordered
6783 * extents. Drop our locks and wait for them to finish
6785 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6787 unlock_extent_cached(io_tree
, page_start
, page_end
,
6788 &cached_state
, GFP_NOFS
);
6790 btrfs_start_ordered_extent(inode
, ordered
, 1);
6791 btrfs_put_ordered_extent(ordered
);
6796 * XXX - page_mkwrite gets called every time the page is dirtied, even
6797 * if it was already dirty, so for space accounting reasons we need to
6798 * clear any delalloc bits for the range we are fixing to save. There
6799 * is probably a better way to do this, but for now keep consistent with
6800 * prepare_pages in the normal write path.
6802 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6803 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6804 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
6805 0, 0, &cached_state
, GFP_NOFS
);
6807 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6810 unlock_extent_cached(io_tree
, page_start
, page_end
,
6811 &cached_state
, GFP_NOFS
);
6812 ret
= VM_FAULT_SIGBUS
;
6817 /* page is wholly or partially inside EOF */
6818 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6819 zero_start
= size
& ~PAGE_CACHE_MASK
;
6821 zero_start
= PAGE_CACHE_SIZE
;
6823 if (zero_start
!= PAGE_CACHE_SIZE
) {
6825 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6826 flush_dcache_page(page
);
6829 ClearPageChecked(page
);
6830 set_page_dirty(page
);
6831 SetPageUptodate(page
);
6833 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6834 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6835 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
6837 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6841 sb_end_pagefault(inode
->i_sb
);
6842 return VM_FAULT_LOCKED
;
6846 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6848 sb_end_pagefault(inode
->i_sb
);
6852 static int btrfs_truncate(struct inode
*inode
)
6854 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6855 struct btrfs_block_rsv
*rsv
;
6858 struct btrfs_trans_handle
*trans
;
6860 u64 mask
= root
->sectorsize
- 1;
6861 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
6863 ret
= btrfs_truncate_page(inode
, inode
->i_size
, 0, 0);
6867 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6868 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6871 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6872 * 3 things going on here
6874 * 1) We need to reserve space for our orphan item and the space to
6875 * delete our orphan item. Lord knows we don't want to have a dangling
6876 * orphan item because we didn't reserve space to remove it.
6878 * 2) We need to reserve space to update our inode.
6880 * 3) We need to have something to cache all the space that is going to
6881 * be free'd up by the truncate operation, but also have some slack
6882 * space reserved in case it uses space during the truncate (thank you
6883 * very much snapshotting).
6885 * And we need these to all be seperate. The fact is we can use alot of
6886 * space doing the truncate, and we have no earthly idea how much space
6887 * we will use, so we need the truncate reservation to be seperate so it
6888 * doesn't end up using space reserved for updating the inode or
6889 * removing the orphan item. We also need to be able to stop the
6890 * transaction and start a new one, which means we need to be able to
6891 * update the inode several times, and we have no idea of knowing how
6892 * many times that will be, so we can't just reserve 1 item for the
6893 * entirety of the opration, so that has to be done seperately as well.
6894 * Then there is the orphan item, which does indeed need to be held on
6895 * to for the whole operation, and we need nobody to touch this reserved
6896 * space except the orphan code.
6898 * So that leaves us with
6900 * 1) root->orphan_block_rsv - for the orphan deletion.
6901 * 2) rsv - for the truncate reservation, which we will steal from the
6902 * transaction reservation.
6903 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6904 * updating the inode.
6906 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
6909 rsv
->size
= min_size
;
6913 * 1 for the truncate slack space
6914 * 1 for the orphan item we're going to add
6915 * 1 for the orphan item deletion
6916 * 1 for updating the inode.
6918 trans
= btrfs_start_transaction(root
, 4);
6919 if (IS_ERR(trans
)) {
6920 err
= PTR_ERR(trans
);
6924 /* Migrate the slack space for the truncate to our reserve */
6925 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
6929 ret
= btrfs_orphan_add(trans
, inode
);
6931 btrfs_end_transaction(trans
, root
);
6936 * setattr is responsible for setting the ordered_data_close flag,
6937 * but that is only tested during the last file release. That
6938 * could happen well after the next commit, leaving a great big
6939 * window where new writes may get lost if someone chooses to write
6940 * to this file after truncating to zero
6942 * The inode doesn't have any dirty data here, and so if we commit
6943 * this is a noop. If someone immediately starts writing to the inode
6944 * it is very likely we'll catch some of their writes in this
6945 * transaction, and the commit will find this file on the ordered
6946 * data list with good things to send down.
6948 * This is a best effort solution, there is still a window where
6949 * using truncate to replace the contents of the file will
6950 * end up with a zero length file after a crash.
6952 if (inode
->i_size
== 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
6953 &BTRFS_I(inode
)->runtime_flags
))
6954 btrfs_add_ordered_operation(trans
, root
, inode
);
6957 * So if we truncate and then write and fsync we normally would just
6958 * write the extents that changed, which is a problem if we need to
6959 * first truncate that entire inode. So set this flag so we write out
6960 * all of the extents in the inode to the sync log so we're completely
6963 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6964 trans
->block_rsv
= rsv
;
6967 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6969 BTRFS_EXTENT_DATA_KEY
);
6970 if (ret
!= -ENOSPC
) {
6975 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6976 ret
= btrfs_update_inode(trans
, root
, inode
);
6982 nr
= trans
->blocks_used
;
6983 btrfs_end_transaction(trans
, root
);
6984 btrfs_btree_balance_dirty(root
, nr
);
6986 trans
= btrfs_start_transaction(root
, 2);
6987 if (IS_ERR(trans
)) {
6988 ret
= err
= PTR_ERR(trans
);
6993 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
6995 BUG_ON(ret
); /* shouldn't happen */
6996 trans
->block_rsv
= rsv
;
6999 if (ret
== 0 && inode
->i_nlink
> 0) {
7000 trans
->block_rsv
= root
->orphan_block_rsv
;
7001 ret
= btrfs_orphan_del(trans
, inode
);
7004 } else if (ret
&& inode
->i_nlink
> 0) {
7006 * Failed to do the truncate, remove us from the in memory
7009 ret
= btrfs_orphan_del(NULL
, inode
);
7013 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
7014 ret
= btrfs_update_inode(trans
, root
, inode
);
7018 nr
= trans
->blocks_used
;
7019 ret
= btrfs_end_transaction(trans
, root
);
7020 btrfs_btree_balance_dirty(root
, nr
);
7024 btrfs_free_block_rsv(root
, rsv
);
7033 * create a new subvolume directory/inode (helper for the ioctl).
7035 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
7036 struct btrfs_root
*new_root
, u64 new_dirid
)
7038 struct inode
*inode
;
7042 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
7043 new_dirid
, new_dirid
,
7044 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
7047 return PTR_ERR(inode
);
7048 inode
->i_op
= &btrfs_dir_inode_operations
;
7049 inode
->i_fop
= &btrfs_dir_file_operations
;
7051 set_nlink(inode
, 1);
7052 btrfs_i_size_write(inode
, 0);
7054 err
= btrfs_update_inode(trans
, new_root
, inode
);
7060 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
7062 struct btrfs_inode
*ei
;
7063 struct inode
*inode
;
7065 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
7072 ei
->last_sub_trans
= 0;
7073 ei
->logged_trans
= 0;
7074 ei
->delalloc_bytes
= 0;
7075 ei
->disk_i_size
= 0;
7078 ei
->index_cnt
= (u64
)-1;
7079 ei
->last_unlink_trans
= 0;
7080 ei
->last_log_commit
= 0;
7082 spin_lock_init(&ei
->lock
);
7083 ei
->outstanding_extents
= 0;
7084 ei
->reserved_extents
= 0;
7086 ei
->runtime_flags
= 0;
7087 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
7089 ei
->delayed_node
= NULL
;
7091 inode
= &ei
->vfs_inode
;
7092 extent_map_tree_init(&ei
->extent_tree
);
7093 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
7094 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
7095 ei
->io_tree
.track_uptodate
= 1;
7096 ei
->io_failure_tree
.track_uptodate
= 1;
7097 mutex_init(&ei
->log_mutex
);
7098 mutex_init(&ei
->delalloc_mutex
);
7099 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
7100 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
7101 INIT_LIST_HEAD(&ei
->ordered_operations
);
7102 RB_CLEAR_NODE(&ei
->rb_node
);
7107 static void btrfs_i_callback(struct rcu_head
*head
)
7109 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
7110 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
7113 void btrfs_destroy_inode(struct inode
*inode
)
7115 struct btrfs_ordered_extent
*ordered
;
7116 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7118 WARN_ON(!hlist_empty(&inode
->i_dentry
));
7119 WARN_ON(inode
->i_data
.nrpages
);
7120 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
7121 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
7122 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
7123 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
7126 * This can happen where we create an inode, but somebody else also
7127 * created the same inode and we need to destroy the one we already
7134 * Make sure we're properly removed from the ordered operation
7138 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
7139 spin_lock(&root
->fs_info
->ordered_extent_lock
);
7140 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
7141 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
7144 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
7145 &BTRFS_I(inode
)->runtime_flags
)) {
7146 printk(KERN_INFO
"BTRFS: inode %llu still on the orphan list\n",
7147 (unsigned long long)btrfs_ino(inode
));
7148 atomic_dec(&root
->orphan_inodes
);
7152 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
7156 printk(KERN_ERR
"btrfs found ordered "
7157 "extent %llu %llu on inode cleanup\n",
7158 (unsigned long long)ordered
->file_offset
,
7159 (unsigned long long)ordered
->len
);
7160 btrfs_remove_ordered_extent(inode
, ordered
);
7161 btrfs_put_ordered_extent(ordered
);
7162 btrfs_put_ordered_extent(ordered
);
7165 inode_tree_del(inode
);
7166 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
7168 btrfs_remove_delayed_node(inode
);
7169 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
7172 int btrfs_drop_inode(struct inode
*inode
)
7174 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7176 if (btrfs_root_refs(&root
->root_item
) == 0 &&
7177 !btrfs_is_free_space_inode(inode
))
7180 return generic_drop_inode(inode
);
7183 static void init_once(void *foo
)
7185 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
7187 inode_init_once(&ei
->vfs_inode
);
7190 void btrfs_destroy_cachep(void)
7193 * Make sure all delayed rcu free inodes are flushed before we
7197 if (btrfs_inode_cachep
)
7198 kmem_cache_destroy(btrfs_inode_cachep
);
7199 if (btrfs_trans_handle_cachep
)
7200 kmem_cache_destroy(btrfs_trans_handle_cachep
);
7201 if (btrfs_transaction_cachep
)
7202 kmem_cache_destroy(btrfs_transaction_cachep
);
7203 if (btrfs_path_cachep
)
7204 kmem_cache_destroy(btrfs_path_cachep
);
7205 if (btrfs_free_space_cachep
)
7206 kmem_cache_destroy(btrfs_free_space_cachep
);
7209 int btrfs_init_cachep(void)
7211 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
7212 sizeof(struct btrfs_inode
), 0,
7213 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
7214 if (!btrfs_inode_cachep
)
7217 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
7218 sizeof(struct btrfs_trans_handle
), 0,
7219 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7220 if (!btrfs_trans_handle_cachep
)
7223 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
7224 sizeof(struct btrfs_transaction
), 0,
7225 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7226 if (!btrfs_transaction_cachep
)
7229 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
7230 sizeof(struct btrfs_path
), 0,
7231 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7232 if (!btrfs_path_cachep
)
7235 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
7236 sizeof(struct btrfs_free_space
), 0,
7237 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7238 if (!btrfs_free_space_cachep
)
7243 btrfs_destroy_cachep();
7247 static int btrfs_getattr(struct vfsmount
*mnt
,
7248 struct dentry
*dentry
, struct kstat
*stat
)
7250 struct inode
*inode
= dentry
->d_inode
;
7251 u32 blocksize
= inode
->i_sb
->s_blocksize
;
7253 generic_fillattr(inode
, stat
);
7254 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
7255 stat
->blksize
= PAGE_CACHE_SIZE
;
7256 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
7257 ALIGN(BTRFS_I(inode
)->delalloc_bytes
, blocksize
)) >> 9;
7262 * If a file is moved, it will inherit the cow and compression flags of the new
7265 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
7267 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
7268 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
7270 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
7271 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
7273 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
7275 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
) {
7276 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
7277 b_inode
->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
7279 b_inode
->flags
&= ~(BTRFS_INODE_COMPRESS
|
7280 BTRFS_INODE_NOCOMPRESS
);
7284 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
7285 struct inode
*new_dir
, struct dentry
*new_dentry
)
7287 struct btrfs_trans_handle
*trans
;
7288 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
7289 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
7290 struct inode
*new_inode
= new_dentry
->d_inode
;
7291 struct inode
*old_inode
= old_dentry
->d_inode
;
7292 struct timespec ctime
= CURRENT_TIME
;
7296 u64 old_ino
= btrfs_ino(old_inode
);
7298 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
7301 /* we only allow rename subvolume link between subvolumes */
7302 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
7305 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
7306 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
7309 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
7310 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
7313 * we're using rename to replace one file with another.
7314 * and the replacement file is large. Start IO on it now so
7315 * we don't add too much work to the end of the transaction
7317 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
7318 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
7319 filemap_flush(old_inode
->i_mapping
);
7321 /* close the racy window with snapshot create/destroy ioctl */
7322 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7323 down_read(&root
->fs_info
->subvol_sem
);
7325 * We want to reserve the absolute worst case amount of items. So if
7326 * both inodes are subvols and we need to unlink them then that would
7327 * require 4 item modifications, but if they are both normal inodes it
7328 * would require 5 item modifications, so we'll assume their normal
7329 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7330 * should cover the worst case number of items we'll modify.
7332 trans
= btrfs_start_transaction(root
, 20);
7333 if (IS_ERR(trans
)) {
7334 ret
= PTR_ERR(trans
);
7339 btrfs_record_root_in_trans(trans
, dest
);
7341 ret
= btrfs_set_inode_index(new_dir
, &index
);
7345 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7346 /* force full log commit if subvolume involved. */
7347 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7349 ret
= btrfs_insert_inode_ref(trans
, dest
,
7350 new_dentry
->d_name
.name
,
7351 new_dentry
->d_name
.len
,
7353 btrfs_ino(new_dir
), index
);
7357 * this is an ugly little race, but the rename is required
7358 * to make sure that if we crash, the inode is either at the
7359 * old name or the new one. pinning the log transaction lets
7360 * us make sure we don't allow a log commit to come in after
7361 * we unlink the name but before we add the new name back in.
7363 btrfs_pin_log_trans(root
);
7366 * make sure the inode gets flushed if it is replacing
7369 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
7370 btrfs_add_ordered_operation(trans
, root
, old_inode
);
7372 inode_inc_iversion(old_dir
);
7373 inode_inc_iversion(new_dir
);
7374 inode_inc_iversion(old_inode
);
7375 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
7376 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
7377 old_inode
->i_ctime
= ctime
;
7379 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
7380 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
7382 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7383 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
7384 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
7385 old_dentry
->d_name
.name
,
7386 old_dentry
->d_name
.len
);
7388 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7389 old_dentry
->d_inode
,
7390 old_dentry
->d_name
.name
,
7391 old_dentry
->d_name
.len
);
7393 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7396 btrfs_abort_transaction(trans
, root
, ret
);
7401 inode_inc_iversion(new_inode
);
7402 new_inode
->i_ctime
= CURRENT_TIME
;
7403 if (unlikely(btrfs_ino(new_inode
) ==
7404 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7405 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7406 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7408 new_dentry
->d_name
.name
,
7409 new_dentry
->d_name
.len
);
7410 BUG_ON(new_inode
->i_nlink
== 0);
7412 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7413 new_dentry
->d_inode
,
7414 new_dentry
->d_name
.name
,
7415 new_dentry
->d_name
.len
);
7417 if (!ret
&& new_inode
->i_nlink
== 0) {
7418 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7422 btrfs_abort_transaction(trans
, root
, ret
);
7427 fixup_inode_flags(new_dir
, old_inode
);
7429 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7430 new_dentry
->d_name
.name
,
7431 new_dentry
->d_name
.len
, 0, index
);
7433 btrfs_abort_transaction(trans
, root
, ret
);
7437 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7438 struct dentry
*parent
= new_dentry
->d_parent
;
7439 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7440 btrfs_end_log_trans(root
);
7443 btrfs_end_transaction(trans
, root
);
7445 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7446 up_read(&root
->fs_info
->subvol_sem
);
7452 * some fairly slow code that needs optimization. This walks the list
7453 * of all the inodes with pending delalloc and forces them to disk.
7455 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7457 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7458 struct btrfs_inode
*binode
;
7459 struct inode
*inode
;
7461 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7464 spin_lock(&root
->fs_info
->delalloc_lock
);
7465 while (!list_empty(head
)) {
7466 binode
= list_entry(head
->next
, struct btrfs_inode
,
7468 inode
= igrab(&binode
->vfs_inode
);
7470 list_del_init(&binode
->delalloc_inodes
);
7471 spin_unlock(&root
->fs_info
->delalloc_lock
);
7473 filemap_flush(inode
->i_mapping
);
7475 btrfs_add_delayed_iput(inode
);
7480 spin_lock(&root
->fs_info
->delalloc_lock
);
7482 spin_unlock(&root
->fs_info
->delalloc_lock
);
7484 /* the filemap_flush will queue IO into the worker threads, but
7485 * we have to make sure the IO is actually started and that
7486 * ordered extents get created before we return
7488 atomic_inc(&root
->fs_info
->async_submit_draining
);
7489 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7490 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7491 wait_event(root
->fs_info
->async_submit_wait
,
7492 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7493 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7495 atomic_dec(&root
->fs_info
->async_submit_draining
);
7499 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7500 const char *symname
)
7502 struct btrfs_trans_handle
*trans
;
7503 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7504 struct btrfs_path
*path
;
7505 struct btrfs_key key
;
7506 struct inode
*inode
= NULL
;
7514 struct btrfs_file_extent_item
*ei
;
7515 struct extent_buffer
*leaf
;
7516 unsigned long nr
= 0;
7518 name_len
= strlen(symname
) + 1;
7519 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7520 return -ENAMETOOLONG
;
7523 * 2 items for inode item and ref
7524 * 2 items for dir items
7525 * 1 item for xattr if selinux is on
7527 trans
= btrfs_start_transaction(root
, 5);
7529 return PTR_ERR(trans
);
7531 err
= btrfs_find_free_ino(root
, &objectid
);
7535 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7536 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
7537 S_IFLNK
|S_IRWXUGO
, &index
);
7538 if (IS_ERR(inode
)) {
7539 err
= PTR_ERR(inode
);
7543 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7550 * If the active LSM wants to access the inode during
7551 * d_instantiate it needs these. Smack checks to see
7552 * if the filesystem supports xattrs by looking at the
7555 inode
->i_fop
= &btrfs_file_operations
;
7556 inode
->i_op
= &btrfs_file_inode_operations
;
7558 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7562 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7563 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7564 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7569 path
= btrfs_alloc_path();
7575 key
.objectid
= btrfs_ino(inode
);
7577 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7578 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7579 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7583 btrfs_free_path(path
);
7586 leaf
= path
->nodes
[0];
7587 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7588 struct btrfs_file_extent_item
);
7589 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7590 btrfs_set_file_extent_type(leaf
, ei
,
7591 BTRFS_FILE_EXTENT_INLINE
);
7592 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7593 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7594 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7595 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7597 ptr
= btrfs_file_extent_inline_start(ei
);
7598 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7599 btrfs_mark_buffer_dirty(leaf
);
7600 btrfs_free_path(path
);
7602 inode
->i_op
= &btrfs_symlink_inode_operations
;
7603 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7604 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7605 inode_set_bytes(inode
, name_len
);
7606 btrfs_i_size_write(inode
, name_len
- 1);
7607 err
= btrfs_update_inode(trans
, root
, inode
);
7613 d_instantiate(dentry
, inode
);
7614 nr
= trans
->blocks_used
;
7615 btrfs_end_transaction(trans
, root
);
7617 inode_dec_link_count(inode
);
7620 btrfs_btree_balance_dirty(root
, nr
);
7624 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7625 u64 start
, u64 num_bytes
, u64 min_size
,
7626 loff_t actual_len
, u64
*alloc_hint
,
7627 struct btrfs_trans_handle
*trans
)
7629 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
7630 struct extent_map
*em
;
7631 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7632 struct btrfs_key ins
;
7633 u64 cur_offset
= start
;
7636 bool own_trans
= true;
7640 while (num_bytes
> 0) {
7642 trans
= btrfs_start_transaction(root
, 3);
7643 if (IS_ERR(trans
)) {
7644 ret
= PTR_ERR(trans
);
7649 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7650 0, *alloc_hint
, &ins
, 1);
7653 btrfs_end_transaction(trans
, root
);
7657 ret
= insert_reserved_file_extent(trans
, inode
,
7658 cur_offset
, ins
.objectid
,
7659 ins
.offset
, ins
.offset
,
7660 ins
.offset
, 0, 0, 0,
7661 BTRFS_FILE_EXTENT_PREALLOC
);
7663 btrfs_abort_transaction(trans
, root
, ret
);
7665 btrfs_end_transaction(trans
, root
);
7668 btrfs_drop_extent_cache(inode
, cur_offset
,
7669 cur_offset
+ ins
.offset
-1, 0);
7671 em
= alloc_extent_map();
7673 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
7674 &BTRFS_I(inode
)->runtime_flags
);
7678 em
->start
= cur_offset
;
7679 em
->orig_start
= cur_offset
;
7680 em
->len
= ins
.offset
;
7681 em
->block_start
= ins
.objectid
;
7682 em
->block_len
= ins
.offset
;
7683 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7684 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7685 em
->generation
= trans
->transid
;
7688 write_lock(&em_tree
->lock
);
7689 ret
= add_extent_mapping(em_tree
, em
);
7691 list_move(&em
->list
,
7692 &em_tree
->modified_extents
);
7693 write_unlock(&em_tree
->lock
);
7696 btrfs_drop_extent_cache(inode
, cur_offset
,
7697 cur_offset
+ ins
.offset
- 1,
7700 free_extent_map(em
);
7702 num_bytes
-= ins
.offset
;
7703 cur_offset
+= ins
.offset
;
7704 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7706 inode_inc_iversion(inode
);
7707 inode
->i_ctime
= CURRENT_TIME
;
7708 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7709 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7710 (actual_len
> inode
->i_size
) &&
7711 (cur_offset
> inode
->i_size
)) {
7712 if (cur_offset
> actual_len
)
7713 i_size
= actual_len
;
7715 i_size
= cur_offset
;
7716 i_size_write(inode
, i_size
);
7717 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7720 ret
= btrfs_update_inode(trans
, root
, inode
);
7723 btrfs_abort_transaction(trans
, root
, ret
);
7725 btrfs_end_transaction(trans
, root
);
7730 btrfs_end_transaction(trans
, root
);
7735 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7736 u64 start
, u64 num_bytes
, u64 min_size
,
7737 loff_t actual_len
, u64
*alloc_hint
)
7739 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7740 min_size
, actual_len
, alloc_hint
,
7744 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7745 struct btrfs_trans_handle
*trans
, int mode
,
7746 u64 start
, u64 num_bytes
, u64 min_size
,
7747 loff_t actual_len
, u64
*alloc_hint
)
7749 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7750 min_size
, actual_len
, alloc_hint
, trans
);
7753 static int btrfs_set_page_dirty(struct page
*page
)
7755 return __set_page_dirty_nobuffers(page
);
7758 static int btrfs_permission(struct inode
*inode
, int mask
)
7760 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7761 umode_t mode
= inode
->i_mode
;
7763 if (mask
& MAY_WRITE
&&
7764 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
7765 if (btrfs_root_readonly(root
))
7767 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
7770 return generic_permission(inode
, mask
);
7773 static const struct inode_operations btrfs_dir_inode_operations
= {
7774 .getattr
= btrfs_getattr
,
7775 .lookup
= btrfs_lookup
,
7776 .create
= btrfs_create
,
7777 .unlink
= btrfs_unlink
,
7779 .mkdir
= btrfs_mkdir
,
7780 .rmdir
= btrfs_rmdir
,
7781 .rename
= btrfs_rename
,
7782 .symlink
= btrfs_symlink
,
7783 .setattr
= btrfs_setattr
,
7784 .mknod
= btrfs_mknod
,
7785 .setxattr
= btrfs_setxattr
,
7786 .getxattr
= btrfs_getxattr
,
7787 .listxattr
= btrfs_listxattr
,
7788 .removexattr
= btrfs_removexattr
,
7789 .permission
= btrfs_permission
,
7790 .get_acl
= btrfs_get_acl
,
7792 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7793 .lookup
= btrfs_lookup
,
7794 .permission
= btrfs_permission
,
7795 .get_acl
= btrfs_get_acl
,
7798 static const struct file_operations btrfs_dir_file_operations
= {
7799 .llseek
= generic_file_llseek
,
7800 .read
= generic_read_dir
,
7801 .readdir
= btrfs_real_readdir
,
7802 .unlocked_ioctl
= btrfs_ioctl
,
7803 #ifdef CONFIG_COMPAT
7804 .compat_ioctl
= btrfs_ioctl
,
7806 .release
= btrfs_release_file
,
7807 .fsync
= btrfs_sync_file
,
7810 static struct extent_io_ops btrfs_extent_io_ops
= {
7811 .fill_delalloc
= run_delalloc_range
,
7812 .submit_bio_hook
= btrfs_submit_bio_hook
,
7813 .merge_bio_hook
= btrfs_merge_bio_hook
,
7814 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7815 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7816 .writepage_start_hook
= btrfs_writepage_start_hook
,
7817 .set_bit_hook
= btrfs_set_bit_hook
,
7818 .clear_bit_hook
= btrfs_clear_bit_hook
,
7819 .merge_extent_hook
= btrfs_merge_extent_hook
,
7820 .split_extent_hook
= btrfs_split_extent_hook
,
7824 * btrfs doesn't support the bmap operation because swapfiles
7825 * use bmap to make a mapping of extents in the file. They assume
7826 * these extents won't change over the life of the file and they
7827 * use the bmap result to do IO directly to the drive.
7829 * the btrfs bmap call would return logical addresses that aren't
7830 * suitable for IO and they also will change frequently as COW
7831 * operations happen. So, swapfile + btrfs == corruption.
7833 * For now we're avoiding this by dropping bmap.
7835 static const struct address_space_operations btrfs_aops
= {
7836 .readpage
= btrfs_readpage
,
7837 .writepage
= btrfs_writepage
,
7838 .writepages
= btrfs_writepages
,
7839 .readpages
= btrfs_readpages
,
7840 .direct_IO
= btrfs_direct_IO
,
7841 .invalidatepage
= btrfs_invalidatepage
,
7842 .releasepage
= btrfs_releasepage
,
7843 .set_page_dirty
= btrfs_set_page_dirty
,
7844 .error_remove_page
= generic_error_remove_page
,
7847 static const struct address_space_operations btrfs_symlink_aops
= {
7848 .readpage
= btrfs_readpage
,
7849 .writepage
= btrfs_writepage
,
7850 .invalidatepage
= btrfs_invalidatepage
,
7851 .releasepage
= btrfs_releasepage
,
7854 static const struct inode_operations btrfs_file_inode_operations
= {
7855 .getattr
= btrfs_getattr
,
7856 .setattr
= btrfs_setattr
,
7857 .setxattr
= btrfs_setxattr
,
7858 .getxattr
= btrfs_getxattr
,
7859 .listxattr
= btrfs_listxattr
,
7860 .removexattr
= btrfs_removexattr
,
7861 .permission
= btrfs_permission
,
7862 .fiemap
= btrfs_fiemap
,
7863 .get_acl
= btrfs_get_acl
,
7864 .update_time
= btrfs_update_time
,
7866 static const struct inode_operations btrfs_special_inode_operations
= {
7867 .getattr
= btrfs_getattr
,
7868 .setattr
= btrfs_setattr
,
7869 .permission
= btrfs_permission
,
7870 .setxattr
= btrfs_setxattr
,
7871 .getxattr
= btrfs_getxattr
,
7872 .listxattr
= btrfs_listxattr
,
7873 .removexattr
= btrfs_removexattr
,
7874 .get_acl
= btrfs_get_acl
,
7875 .update_time
= btrfs_update_time
,
7877 static const struct inode_operations btrfs_symlink_inode_operations
= {
7878 .readlink
= generic_readlink
,
7879 .follow_link
= page_follow_link_light
,
7880 .put_link
= page_put_link
,
7881 .getattr
= btrfs_getattr
,
7882 .setattr
= btrfs_setattr
,
7883 .permission
= btrfs_permission
,
7884 .setxattr
= btrfs_setxattr
,
7885 .getxattr
= btrfs_getxattr
,
7886 .listxattr
= btrfs_listxattr
,
7887 .removexattr
= btrfs_removexattr
,
7888 .get_acl
= btrfs_get_acl
,
7889 .update_time
= btrfs_update_time
,
7892 const struct dentry_operations btrfs_dentry_operations
= {
7893 .d_delete
= btrfs_dentry_delete
,
7894 .d_release
= btrfs_dentry_release
,