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>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args
{
57 struct btrfs_root
*root
;
60 static const struct inode_operations btrfs_dir_inode_operations
;
61 static const struct inode_operations btrfs_symlink_inode_operations
;
62 static const struct inode_operations btrfs_dir_ro_inode_operations
;
63 static const struct inode_operations btrfs_special_inode_operations
;
64 static const struct inode_operations btrfs_file_inode_operations
;
65 static const struct address_space_operations btrfs_aops
;
66 static const struct address_space_operations btrfs_symlink_aops
;
67 static const struct file_operations btrfs_dir_file_operations
;
68 static struct extent_io_ops btrfs_extent_io_ops
;
70 static struct kmem_cache
*btrfs_inode_cachep
;
71 struct kmem_cache
*btrfs_trans_handle_cachep
;
72 struct kmem_cache
*btrfs_transaction_cachep
;
73 struct kmem_cache
*btrfs_path_cachep
;
74 struct kmem_cache
*btrfs_free_space_cachep
;
77 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
78 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
79 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
80 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
81 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
82 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
83 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
84 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
87 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
);
88 static int btrfs_truncate(struct inode
*inode
);
89 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
);
90 static noinline
int cow_file_range(struct inode
*inode
,
91 struct page
*locked_page
,
92 u64 start
, u64 end
, int *page_started
,
93 unsigned long *nr_written
, int unlock
);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
96 struct inode
*inode
, struct inode
*dir
)
100 err
= btrfs_init_acl(trans
, inode
, dir
);
102 err
= btrfs_xattr_security_init(trans
, inode
, dir
);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
112 struct btrfs_root
*root
, struct inode
*inode
,
113 u64 start
, size_t size
, size_t compressed_size
,
115 struct page
**compressed_pages
)
117 struct btrfs_key key
;
118 struct btrfs_path
*path
;
119 struct extent_buffer
*leaf
;
120 struct page
*page
= NULL
;
123 struct btrfs_file_extent_item
*ei
;
126 size_t cur_size
= size
;
128 unsigned long offset
;
130 if (compressed_size
&& compressed_pages
)
131 cur_size
= compressed_size
;
133 path
= btrfs_alloc_path();
137 path
->leave_spinning
= 1;
138 btrfs_set_trans_block_group(trans
, inode
);
140 key
.objectid
= inode
->i_ino
;
142 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
143 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
145 inode_add_bytes(inode
, size
);
146 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
153 leaf
= path
->nodes
[0];
154 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
155 struct btrfs_file_extent_item
);
156 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
157 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
158 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
159 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
160 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
161 ptr
= btrfs_file_extent_inline_start(ei
);
163 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
166 while (compressed_size
> 0) {
167 cpage
= compressed_pages
[i
];
168 cur_size
= min_t(unsigned long, compressed_size
,
171 kaddr
= kmap_atomic(cpage
, KM_USER0
);
172 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
173 kunmap_atomic(kaddr
, KM_USER0
);
177 compressed_size
-= cur_size
;
179 btrfs_set_file_extent_compression(leaf
, ei
,
182 page
= find_get_page(inode
->i_mapping
,
183 start
>> PAGE_CACHE_SHIFT
);
184 btrfs_set_file_extent_compression(leaf
, ei
, 0);
185 kaddr
= kmap_atomic(page
, KM_USER0
);
186 offset
= start
& (PAGE_CACHE_SIZE
- 1);
187 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
188 kunmap_atomic(kaddr
, KM_USER0
);
189 page_cache_release(page
);
191 btrfs_mark_buffer_dirty(leaf
);
192 btrfs_free_path(path
);
195 * we're an inline extent, so nobody can
196 * extend the file past i_size without locking
197 * a page we already have locked.
199 * We must do any isize and inode updates
200 * before we unlock the pages. Otherwise we
201 * could end up racing with unlink.
203 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
204 btrfs_update_inode(trans
, root
, inode
);
208 btrfs_free_path(path
);
214 * conditionally insert an inline extent into the file. This
215 * does the checks required to make sure the data is small enough
216 * to fit as an inline extent.
218 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
219 struct btrfs_root
*root
,
220 struct inode
*inode
, u64 start
, u64 end
,
221 size_t compressed_size
, int compress_type
,
222 struct page
**compressed_pages
)
224 u64 isize
= i_size_read(inode
);
225 u64 actual_end
= min(end
+ 1, isize
);
226 u64 inline_len
= actual_end
- start
;
227 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
228 ~((u64
)root
->sectorsize
- 1);
230 u64 data_len
= inline_len
;
234 data_len
= compressed_size
;
237 actual_end
>= PAGE_CACHE_SIZE
||
238 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
240 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
242 data_len
> root
->fs_info
->max_inline
) {
246 ret
= btrfs_drop_extents(trans
, inode
, start
, aligned_end
,
250 if (isize
> actual_end
)
251 inline_len
= min_t(u64
, isize
, actual_end
);
252 ret
= insert_inline_extent(trans
, root
, inode
, start
,
253 inline_len
, compressed_size
,
254 compress_type
, compressed_pages
);
256 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
257 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
261 struct async_extent
{
266 unsigned long nr_pages
;
268 struct list_head list
;
273 struct btrfs_root
*root
;
274 struct page
*locked_page
;
277 struct list_head extents
;
278 struct btrfs_work work
;
281 static noinline
int add_async_extent(struct async_cow
*cow
,
282 u64 start
, u64 ram_size
,
285 unsigned long nr_pages
,
288 struct async_extent
*async_extent
;
290 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
291 BUG_ON(!async_extent
);
292 async_extent
->start
= start
;
293 async_extent
->ram_size
= ram_size
;
294 async_extent
->compressed_size
= compressed_size
;
295 async_extent
->pages
= pages
;
296 async_extent
->nr_pages
= nr_pages
;
297 async_extent
->compress_type
= compress_type
;
298 list_add_tail(&async_extent
->list
, &cow
->extents
);
303 * we create compressed extents in two phases. The first
304 * phase compresses a range of pages that have already been
305 * locked (both pages and state bits are locked).
307 * This is done inside an ordered work queue, and the compression
308 * is spread across many cpus. The actual IO submission is step
309 * two, and the ordered work queue takes care of making sure that
310 * happens in the same order things were put onto the queue by
311 * writepages and friends.
313 * If this code finds it can't get good compression, it puts an
314 * entry onto the work queue to write the uncompressed bytes. This
315 * makes sure that both compressed inodes and uncompressed inodes
316 * are written in the same order that pdflush sent them down.
318 static noinline
int compress_file_range(struct inode
*inode
,
319 struct page
*locked_page
,
321 struct async_cow
*async_cow
,
324 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
325 struct btrfs_trans_handle
*trans
;
327 u64 blocksize
= root
->sectorsize
;
329 u64 isize
= i_size_read(inode
);
331 struct page
**pages
= NULL
;
332 unsigned long nr_pages
;
333 unsigned long nr_pages_ret
= 0;
334 unsigned long total_compressed
= 0;
335 unsigned long total_in
= 0;
336 unsigned long max_compressed
= 128 * 1024;
337 unsigned long max_uncompressed
= 128 * 1024;
340 int compress_type
= root
->fs_info
->compress_type
;
342 actual_end
= min_t(u64
, isize
, end
+ 1);
345 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
346 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
349 * we don't want to send crud past the end of i_size through
350 * compression, that's just a waste of CPU time. So, if the
351 * end of the file is before the start of our current
352 * requested range of bytes, we bail out to the uncompressed
353 * cleanup code that can deal with all of this.
355 * It isn't really the fastest way to fix things, but this is a
356 * very uncommon corner.
358 if (actual_end
<= start
)
359 goto cleanup_and_bail_uncompressed
;
361 total_compressed
= actual_end
- start
;
363 /* we want to make sure that amount of ram required to uncompress
364 * an extent is reasonable, so we limit the total size in ram
365 * of a compressed extent to 128k. This is a crucial number
366 * because it also controls how easily we can spread reads across
367 * cpus for decompression.
369 * We also want to make sure the amount of IO required to do
370 * a random read is reasonably small, so we limit the size of
371 * a compressed extent to 128k.
373 total_compressed
= min(total_compressed
, max_uncompressed
);
374 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
375 num_bytes
= max(blocksize
, num_bytes
);
380 * we do compression for mount -o compress and when the
381 * inode has not been flagged as nocompress. This flag can
382 * change at any time if we discover bad compression ratios.
384 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
385 (btrfs_test_opt(root
, COMPRESS
) ||
386 (BTRFS_I(inode
)->force_compress
) ||
387 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
389 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
392 if (BTRFS_I(inode
)->force_compress
)
393 compress_type
= BTRFS_I(inode
)->force_compress
;
395 ret
= btrfs_compress_pages(compress_type
,
396 inode
->i_mapping
, start
,
397 total_compressed
, pages
,
398 nr_pages
, &nr_pages_ret
,
404 unsigned long offset
= total_compressed
&
405 (PAGE_CACHE_SIZE
- 1);
406 struct page
*page
= pages
[nr_pages_ret
- 1];
409 /* zero the tail end of the last page, we might be
410 * sending it down to disk
413 kaddr
= kmap_atomic(page
, KM_USER0
);
414 memset(kaddr
+ offset
, 0,
415 PAGE_CACHE_SIZE
- offset
);
416 kunmap_atomic(kaddr
, KM_USER0
);
422 trans
= btrfs_join_transaction(root
, 1);
423 BUG_ON(IS_ERR(trans
));
424 btrfs_set_trans_block_group(trans
, inode
);
425 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
427 /* lets try to make an inline extent */
428 if (ret
|| total_in
< (actual_end
- start
)) {
429 /* we didn't compress the entire range, try
430 * to make an uncompressed inline extent.
432 ret
= cow_file_range_inline(trans
, root
, inode
,
433 start
, end
, 0, 0, NULL
);
435 /* try making a compressed inline extent */
436 ret
= cow_file_range_inline(trans
, root
, inode
,
439 compress_type
, pages
);
443 * inline extent creation worked, we don't need
444 * to create any more async work items. Unlock
445 * and free up our temp pages.
447 extent_clear_unlock_delalloc(inode
,
448 &BTRFS_I(inode
)->io_tree
,
450 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
451 EXTENT_CLEAR_DELALLOC
|
452 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
454 btrfs_end_transaction(trans
, root
);
457 btrfs_end_transaction(trans
, root
);
462 * we aren't doing an inline extent round the compressed size
463 * up to a block size boundary so the allocator does sane
466 total_compressed
= (total_compressed
+ blocksize
- 1) &
470 * one last check to make sure the compression is really a
471 * win, compare the page count read with the blocks on disk
473 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
474 ~(PAGE_CACHE_SIZE
- 1);
475 if (total_compressed
>= total_in
) {
478 num_bytes
= total_in
;
481 if (!will_compress
&& pages
) {
483 * the compression code ran but failed to make things smaller,
484 * free any pages it allocated and our page pointer array
486 for (i
= 0; i
< nr_pages_ret
; i
++) {
487 WARN_ON(pages
[i
]->mapping
);
488 page_cache_release(pages
[i
]);
492 total_compressed
= 0;
495 /* flag the file so we don't compress in the future */
496 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
497 !(BTRFS_I(inode
)->force_compress
)) {
498 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
504 /* the async work queues will take care of doing actual
505 * allocation on disk for these compressed pages,
506 * and will submit them to the elevator.
508 add_async_extent(async_cow
, start
, num_bytes
,
509 total_compressed
, pages
, nr_pages_ret
,
512 if (start
+ num_bytes
< end
) {
519 cleanup_and_bail_uncompressed
:
521 * No compression, but we still need to write the pages in
522 * the file we've been given so far. redirty the locked
523 * page if it corresponds to our extent and set things up
524 * for the async work queue to run cow_file_range to do
525 * the normal delalloc dance
527 if (page_offset(locked_page
) >= start
&&
528 page_offset(locked_page
) <= end
) {
529 __set_page_dirty_nobuffers(locked_page
);
530 /* unlocked later on in the async handlers */
532 add_async_extent(async_cow
, start
, end
- start
+ 1,
533 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
541 for (i
= 0; i
< nr_pages_ret
; i
++) {
542 WARN_ON(pages
[i
]->mapping
);
543 page_cache_release(pages
[i
]);
551 * phase two of compressed writeback. This is the ordered portion
552 * of the code, which only gets called in the order the work was
553 * queued. We walk all the async extents created by compress_file_range
554 * and send them down to the disk.
556 static noinline
int submit_compressed_extents(struct inode
*inode
,
557 struct async_cow
*async_cow
)
559 struct async_extent
*async_extent
;
561 struct btrfs_trans_handle
*trans
;
562 struct btrfs_key ins
;
563 struct extent_map
*em
;
564 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
565 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
566 struct extent_io_tree
*io_tree
;
569 if (list_empty(&async_cow
->extents
))
573 while (!list_empty(&async_cow
->extents
)) {
574 async_extent
= list_entry(async_cow
->extents
.next
,
575 struct async_extent
, list
);
576 list_del(&async_extent
->list
);
578 io_tree
= &BTRFS_I(inode
)->io_tree
;
581 /* did the compression code fall back to uncompressed IO? */
582 if (!async_extent
->pages
) {
583 int page_started
= 0;
584 unsigned long nr_written
= 0;
586 lock_extent(io_tree
, async_extent
->start
,
587 async_extent
->start
+
588 async_extent
->ram_size
- 1, GFP_NOFS
);
590 /* allocate blocks */
591 ret
= cow_file_range(inode
, async_cow
->locked_page
,
593 async_extent
->start
+
594 async_extent
->ram_size
- 1,
595 &page_started
, &nr_written
, 0);
598 * if page_started, cow_file_range inserted an
599 * inline extent and took care of all the unlocking
600 * and IO for us. Otherwise, we need to submit
601 * all those pages down to the drive.
603 if (!page_started
&& !ret
)
604 extent_write_locked_range(io_tree
,
605 inode
, async_extent
->start
,
606 async_extent
->start
+
607 async_extent
->ram_size
- 1,
615 lock_extent(io_tree
, async_extent
->start
,
616 async_extent
->start
+ async_extent
->ram_size
- 1,
619 trans
= btrfs_join_transaction(root
, 1);
620 BUG_ON(IS_ERR(trans
));
621 ret
= btrfs_reserve_extent(trans
, root
,
622 async_extent
->compressed_size
,
623 async_extent
->compressed_size
,
626 btrfs_end_transaction(trans
, root
);
630 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
631 WARN_ON(async_extent
->pages
[i
]->mapping
);
632 page_cache_release(async_extent
->pages
[i
]);
634 kfree(async_extent
->pages
);
635 async_extent
->nr_pages
= 0;
636 async_extent
->pages
= NULL
;
637 unlock_extent(io_tree
, async_extent
->start
,
638 async_extent
->start
+
639 async_extent
->ram_size
- 1, GFP_NOFS
);
644 * here we're doing allocation and writeback of the
647 btrfs_drop_extent_cache(inode
, async_extent
->start
,
648 async_extent
->start
+
649 async_extent
->ram_size
- 1, 0);
651 em
= alloc_extent_map(GFP_NOFS
);
653 em
->start
= async_extent
->start
;
654 em
->len
= async_extent
->ram_size
;
655 em
->orig_start
= em
->start
;
657 em
->block_start
= ins
.objectid
;
658 em
->block_len
= ins
.offset
;
659 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
660 em
->compress_type
= async_extent
->compress_type
;
661 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
662 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
665 write_lock(&em_tree
->lock
);
666 ret
= add_extent_mapping(em_tree
, em
);
667 write_unlock(&em_tree
->lock
);
668 if (ret
!= -EEXIST
) {
672 btrfs_drop_extent_cache(inode
, async_extent
->start
,
673 async_extent
->start
+
674 async_extent
->ram_size
- 1, 0);
677 ret
= btrfs_add_ordered_extent_compress(inode
,
680 async_extent
->ram_size
,
682 BTRFS_ORDERED_COMPRESSED
,
683 async_extent
->compress_type
);
687 * clear dirty, set writeback and unlock the pages.
689 extent_clear_unlock_delalloc(inode
,
690 &BTRFS_I(inode
)->io_tree
,
692 async_extent
->start
+
693 async_extent
->ram_size
- 1,
694 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
695 EXTENT_CLEAR_UNLOCK
|
696 EXTENT_CLEAR_DELALLOC
|
697 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
699 ret
= btrfs_submit_compressed_write(inode
,
701 async_extent
->ram_size
,
703 ins
.offset
, async_extent
->pages
,
704 async_extent
->nr_pages
);
707 alloc_hint
= ins
.objectid
+ ins
.offset
;
715 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
718 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
719 struct extent_map
*em
;
722 read_lock(&em_tree
->lock
);
723 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
726 * if block start isn't an actual block number then find the
727 * first block in this inode and use that as a hint. If that
728 * block is also bogus then just don't worry about it.
730 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
732 em
= search_extent_mapping(em_tree
, 0, 0);
733 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
734 alloc_hint
= em
->block_start
;
738 alloc_hint
= em
->block_start
;
742 read_unlock(&em_tree
->lock
);
748 * when extent_io.c finds a delayed allocation range in the file,
749 * the call backs end up in this code. The basic idea is to
750 * allocate extents on disk for the range, and create ordered data structs
751 * in ram to track those extents.
753 * locked_page is the page that writepage had locked already. We use
754 * it to make sure we don't do extra locks or unlocks.
756 * *page_started is set to one if we unlock locked_page and do everything
757 * required to start IO on it. It may be clean and already done with
760 static noinline
int cow_file_range(struct inode
*inode
,
761 struct page
*locked_page
,
762 u64 start
, u64 end
, int *page_started
,
763 unsigned long *nr_written
,
766 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
767 struct btrfs_trans_handle
*trans
;
770 unsigned long ram_size
;
773 u64 blocksize
= root
->sectorsize
;
774 struct btrfs_key ins
;
775 struct extent_map
*em
;
776 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
779 BUG_ON(root
== root
->fs_info
->tree_root
);
780 trans
= btrfs_join_transaction(root
, 1);
781 BUG_ON(IS_ERR(trans
));
782 btrfs_set_trans_block_group(trans
, inode
);
783 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
785 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
786 num_bytes
= max(blocksize
, num_bytes
);
787 disk_num_bytes
= num_bytes
;
791 /* lets try to make an inline extent */
792 ret
= cow_file_range_inline(trans
, root
, inode
,
793 start
, end
, 0, 0, NULL
);
795 extent_clear_unlock_delalloc(inode
,
796 &BTRFS_I(inode
)->io_tree
,
798 EXTENT_CLEAR_UNLOCK_PAGE
|
799 EXTENT_CLEAR_UNLOCK
|
800 EXTENT_CLEAR_DELALLOC
|
802 EXTENT_SET_WRITEBACK
|
803 EXTENT_END_WRITEBACK
);
805 *nr_written
= *nr_written
+
806 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
813 BUG_ON(disk_num_bytes
>
814 btrfs_super_total_bytes(&root
->fs_info
->super_copy
));
816 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
817 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
819 while (disk_num_bytes
> 0) {
822 cur_alloc_size
= disk_num_bytes
;
823 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
824 root
->sectorsize
, 0, alloc_hint
,
828 em
= alloc_extent_map(GFP_NOFS
);
831 em
->orig_start
= em
->start
;
832 ram_size
= ins
.offset
;
833 em
->len
= ins
.offset
;
835 em
->block_start
= ins
.objectid
;
836 em
->block_len
= ins
.offset
;
837 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
838 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
841 write_lock(&em_tree
->lock
);
842 ret
= add_extent_mapping(em_tree
, em
);
843 write_unlock(&em_tree
->lock
);
844 if (ret
!= -EEXIST
) {
848 btrfs_drop_extent_cache(inode
, start
,
849 start
+ ram_size
- 1, 0);
852 cur_alloc_size
= ins
.offset
;
853 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
854 ram_size
, cur_alloc_size
, 0);
857 if (root
->root_key
.objectid
==
858 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
859 ret
= btrfs_reloc_clone_csums(inode
, start
,
864 if (disk_num_bytes
< cur_alloc_size
)
867 /* we're not doing compressed IO, don't unlock the first
868 * page (which the caller expects to stay locked), don't
869 * clear any dirty bits and don't set any writeback bits
871 * Do set the Private2 bit so we know this page was properly
872 * setup for writepage
874 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
875 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
878 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
879 start
, start
+ ram_size
- 1,
881 disk_num_bytes
-= cur_alloc_size
;
882 num_bytes
-= cur_alloc_size
;
883 alloc_hint
= ins
.objectid
+ ins
.offset
;
884 start
+= cur_alloc_size
;
888 btrfs_end_transaction(trans
, root
);
894 * work queue call back to started compression on a file and pages
896 static noinline
void async_cow_start(struct btrfs_work
*work
)
898 struct async_cow
*async_cow
;
900 async_cow
= container_of(work
, struct async_cow
, work
);
902 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
903 async_cow
->start
, async_cow
->end
, async_cow
,
906 async_cow
->inode
= NULL
;
910 * work queue call back to submit previously compressed pages
912 static noinline
void async_cow_submit(struct btrfs_work
*work
)
914 struct async_cow
*async_cow
;
915 struct btrfs_root
*root
;
916 unsigned long nr_pages
;
918 async_cow
= container_of(work
, struct async_cow
, work
);
920 root
= async_cow
->root
;
921 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
924 atomic_sub(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
926 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
928 waitqueue_active(&root
->fs_info
->async_submit_wait
))
929 wake_up(&root
->fs_info
->async_submit_wait
);
931 if (async_cow
->inode
)
932 submit_compressed_extents(async_cow
->inode
, async_cow
);
935 static noinline
void async_cow_free(struct btrfs_work
*work
)
937 struct async_cow
*async_cow
;
938 async_cow
= container_of(work
, struct async_cow
, work
);
942 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
943 u64 start
, u64 end
, int *page_started
,
944 unsigned long *nr_written
)
946 struct async_cow
*async_cow
;
947 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
948 unsigned long nr_pages
;
950 int limit
= 10 * 1024 * 1042;
952 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
953 1, 0, NULL
, GFP_NOFS
);
954 while (start
< end
) {
955 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
956 async_cow
->inode
= inode
;
957 async_cow
->root
= root
;
958 async_cow
->locked_page
= locked_page
;
959 async_cow
->start
= start
;
961 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
964 cur_end
= min(end
, start
+ 512 * 1024 - 1);
966 async_cow
->end
= cur_end
;
967 INIT_LIST_HEAD(&async_cow
->extents
);
969 async_cow
->work
.func
= async_cow_start
;
970 async_cow
->work
.ordered_func
= async_cow_submit
;
971 async_cow
->work
.ordered_free
= async_cow_free
;
972 async_cow
->work
.flags
= 0;
974 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
976 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
978 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
981 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
982 wait_event(root
->fs_info
->async_submit_wait
,
983 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
987 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
988 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
989 wait_event(root
->fs_info
->async_submit_wait
,
990 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
994 *nr_written
+= nr_pages
;
1001 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1002 u64 bytenr
, u64 num_bytes
)
1005 struct btrfs_ordered_sum
*sums
;
1008 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1009 bytenr
+ num_bytes
- 1, &list
);
1010 if (ret
== 0 && list_empty(&list
))
1013 while (!list_empty(&list
)) {
1014 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1015 list_del(&sums
->list
);
1022 * when nowcow writeback call back. This checks for snapshots or COW copies
1023 * of the extents that exist in the file, and COWs the file as required.
1025 * If no cow copies or snapshots exist, we write directly to the existing
1028 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1029 struct page
*locked_page
,
1030 u64 start
, u64 end
, int *page_started
, int force
,
1031 unsigned long *nr_written
)
1033 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1034 struct btrfs_trans_handle
*trans
;
1035 struct extent_buffer
*leaf
;
1036 struct btrfs_path
*path
;
1037 struct btrfs_file_extent_item
*fi
;
1038 struct btrfs_key found_key
;
1050 bool nolock
= false;
1052 path
= btrfs_alloc_path();
1054 if (root
== root
->fs_info
->tree_root
) {
1056 trans
= btrfs_join_transaction_nolock(root
, 1);
1058 trans
= btrfs_join_transaction(root
, 1);
1060 BUG_ON(IS_ERR(trans
));
1062 cow_start
= (u64
)-1;
1065 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
1068 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1069 leaf
= path
->nodes
[0];
1070 btrfs_item_key_to_cpu(leaf
, &found_key
,
1071 path
->slots
[0] - 1);
1072 if (found_key
.objectid
== inode
->i_ino
&&
1073 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1078 leaf
= path
->nodes
[0];
1079 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1080 ret
= btrfs_next_leaf(root
, path
);
1085 leaf
= path
->nodes
[0];
1091 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1093 if (found_key
.objectid
> inode
->i_ino
||
1094 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1095 found_key
.offset
> end
)
1098 if (found_key
.offset
> cur_offset
) {
1099 extent_end
= found_key
.offset
;
1104 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1105 struct btrfs_file_extent_item
);
1106 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1108 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1109 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1110 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1111 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1112 extent_end
= found_key
.offset
+
1113 btrfs_file_extent_num_bytes(leaf
, fi
);
1114 if (extent_end
<= start
) {
1118 if (disk_bytenr
== 0)
1120 if (btrfs_file_extent_compression(leaf
, fi
) ||
1121 btrfs_file_extent_encryption(leaf
, fi
) ||
1122 btrfs_file_extent_other_encoding(leaf
, fi
))
1124 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1126 if (btrfs_extent_readonly(root
, disk_bytenr
))
1128 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
1130 extent_offset
, disk_bytenr
))
1132 disk_bytenr
+= extent_offset
;
1133 disk_bytenr
+= cur_offset
- found_key
.offset
;
1134 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1136 * force cow if csum exists in the range.
1137 * this ensure that csum for a given extent are
1138 * either valid or do not exist.
1140 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1143 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1144 extent_end
= found_key
.offset
+
1145 btrfs_file_extent_inline_len(leaf
, fi
);
1146 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1151 if (extent_end
<= start
) {
1156 if (cow_start
== (u64
)-1)
1157 cow_start
= cur_offset
;
1158 cur_offset
= extent_end
;
1159 if (cur_offset
> end
)
1165 btrfs_release_path(root
, path
);
1166 if (cow_start
!= (u64
)-1) {
1167 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1168 found_key
.offset
- 1, page_started
,
1171 cow_start
= (u64
)-1;
1174 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1175 struct extent_map
*em
;
1176 struct extent_map_tree
*em_tree
;
1177 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1178 em
= alloc_extent_map(GFP_NOFS
);
1180 em
->start
= cur_offset
;
1181 em
->orig_start
= em
->start
;
1182 em
->len
= num_bytes
;
1183 em
->block_len
= num_bytes
;
1184 em
->block_start
= disk_bytenr
;
1185 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1186 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1188 write_lock(&em_tree
->lock
);
1189 ret
= add_extent_mapping(em_tree
, em
);
1190 write_unlock(&em_tree
->lock
);
1191 if (ret
!= -EEXIST
) {
1192 free_extent_map(em
);
1195 btrfs_drop_extent_cache(inode
, em
->start
,
1196 em
->start
+ em
->len
- 1, 0);
1198 type
= BTRFS_ORDERED_PREALLOC
;
1200 type
= BTRFS_ORDERED_NOCOW
;
1203 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1204 num_bytes
, num_bytes
, type
);
1207 if (root
->root_key
.objectid
==
1208 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1209 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1214 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1215 cur_offset
, cur_offset
+ num_bytes
- 1,
1216 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1217 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1218 EXTENT_SET_PRIVATE2
);
1219 cur_offset
= extent_end
;
1220 if (cur_offset
> end
)
1223 btrfs_release_path(root
, path
);
1225 if (cur_offset
<= end
&& cow_start
== (u64
)-1)
1226 cow_start
= cur_offset
;
1227 if (cow_start
!= (u64
)-1) {
1228 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1229 page_started
, nr_written
, 1);
1234 ret
= btrfs_end_transaction_nolock(trans
, root
);
1237 ret
= btrfs_end_transaction(trans
, root
);
1240 btrfs_free_path(path
);
1245 * extent_io.c call back to do delayed allocation processing
1247 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1248 u64 start
, u64 end
, int *page_started
,
1249 unsigned long *nr_written
)
1252 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1254 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
)
1255 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1256 page_started
, 1, nr_written
);
1257 else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
)
1258 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1259 page_started
, 0, nr_written
);
1260 else if (!btrfs_test_opt(root
, COMPRESS
) &&
1261 !(BTRFS_I(inode
)->force_compress
) &&
1262 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))
1263 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1264 page_started
, nr_written
, 1);
1266 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1267 page_started
, nr_written
);
1271 static int btrfs_split_extent_hook(struct inode
*inode
,
1272 struct extent_state
*orig
, u64 split
)
1274 /* not delalloc, ignore it */
1275 if (!(orig
->state
& EXTENT_DELALLOC
))
1278 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1283 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1284 * extents so we can keep track of new extents that are just merged onto old
1285 * extents, such as when we are doing sequential writes, so we can properly
1286 * account for the metadata space we'll need.
1288 static int btrfs_merge_extent_hook(struct inode
*inode
,
1289 struct extent_state
*new,
1290 struct extent_state
*other
)
1292 /* not delalloc, ignore it */
1293 if (!(other
->state
& EXTENT_DELALLOC
))
1296 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1301 * extent_io.c set_bit_hook, used to track delayed allocation
1302 * bytes in this file, and to maintain the list of inodes that
1303 * have pending delalloc work to be done.
1305 static int btrfs_set_bit_hook(struct inode
*inode
,
1306 struct extent_state
*state
, int *bits
)
1310 * set_bit and clear bit hooks normally require _irqsave/restore
1311 * but in this case, we are only testeing for the DELALLOC
1312 * bit, which is only set or cleared with irqs on
1314 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1315 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1316 u64 len
= state
->end
+ 1 - state
->start
;
1317 int do_list
= (root
->root_key
.objectid
!=
1318 BTRFS_ROOT_TREE_OBJECTID
);
1320 if (*bits
& EXTENT_FIRST_DELALLOC
)
1321 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1323 atomic_inc(&BTRFS_I(inode
)->outstanding_extents
);
1325 spin_lock(&root
->fs_info
->delalloc_lock
);
1326 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1327 root
->fs_info
->delalloc_bytes
+= len
;
1328 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1329 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1330 &root
->fs_info
->delalloc_inodes
);
1332 spin_unlock(&root
->fs_info
->delalloc_lock
);
1338 * extent_io.c clear_bit_hook, see set_bit_hook for why
1340 static int btrfs_clear_bit_hook(struct inode
*inode
,
1341 struct extent_state
*state
, int *bits
)
1344 * set_bit and clear bit hooks normally require _irqsave/restore
1345 * but in this case, we are only testeing for the DELALLOC
1346 * bit, which is only set or cleared with irqs on
1348 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1349 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1350 u64 len
= state
->end
+ 1 - state
->start
;
1351 int do_list
= (root
->root_key
.objectid
!=
1352 BTRFS_ROOT_TREE_OBJECTID
);
1354 if (*bits
& EXTENT_FIRST_DELALLOC
)
1355 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1356 else if (!(*bits
& EXTENT_DO_ACCOUNTING
))
1357 atomic_dec(&BTRFS_I(inode
)->outstanding_extents
);
1359 if (*bits
& EXTENT_DO_ACCOUNTING
)
1360 btrfs_delalloc_release_metadata(inode
, len
);
1362 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1364 btrfs_free_reserved_data_space(inode
, len
);
1366 spin_lock(&root
->fs_info
->delalloc_lock
);
1367 root
->fs_info
->delalloc_bytes
-= len
;
1368 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1370 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1371 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1372 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1374 spin_unlock(&root
->fs_info
->delalloc_lock
);
1380 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1381 * we don't create bios that span stripes or chunks
1383 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1384 size_t size
, struct bio
*bio
,
1385 unsigned long bio_flags
)
1387 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1388 struct btrfs_mapping_tree
*map_tree
;
1389 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1394 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1397 length
= bio
->bi_size
;
1398 map_tree
= &root
->fs_info
->mapping_tree
;
1399 map_length
= length
;
1400 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1401 &map_length
, NULL
, 0);
1403 if (map_length
< length
+ size
)
1409 * in order to insert checksums into the metadata in large chunks,
1410 * we wait until bio submission time. All the pages in the bio are
1411 * checksummed and sums are attached onto the ordered extent record.
1413 * At IO completion time the cums attached on the ordered extent record
1414 * are inserted into the btree
1416 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1417 struct bio
*bio
, int mirror_num
,
1418 unsigned long bio_flags
,
1421 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1424 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1430 * in order to insert checksums into the metadata in large chunks,
1431 * we wait until bio submission time. All the pages in the bio are
1432 * checksummed and sums are attached onto the ordered extent record.
1434 * At IO completion time the cums attached on the ordered extent record
1435 * are inserted into the btree
1437 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1438 int mirror_num
, unsigned long bio_flags
,
1441 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1442 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1446 * extent_io.c submission hook. This does the right thing for csum calculation
1447 * on write, or reading the csums from the tree before a read
1449 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1450 int mirror_num
, unsigned long bio_flags
,
1453 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1457 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1459 if (root
== root
->fs_info
->tree_root
)
1460 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 2);
1462 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
1465 if (!(rw
& REQ_WRITE
)) {
1466 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1467 return btrfs_submit_compressed_read(inode
, bio
,
1468 mirror_num
, bio_flags
);
1469 } else if (!skip_sum
) {
1470 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1475 } else if (!skip_sum
) {
1476 /* csum items have already been cloned */
1477 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1479 /* we're doing a write, do the async checksumming */
1480 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1481 inode
, rw
, bio
, mirror_num
,
1482 bio_flags
, bio_offset
,
1483 __btrfs_submit_bio_start
,
1484 __btrfs_submit_bio_done
);
1488 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1492 * given a list of ordered sums record them in the inode. This happens
1493 * at IO completion time based on sums calculated at bio submission time.
1495 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1496 struct inode
*inode
, u64 file_offset
,
1497 struct list_head
*list
)
1499 struct btrfs_ordered_sum
*sum
;
1501 btrfs_set_trans_block_group(trans
, inode
);
1503 list_for_each_entry(sum
, list
, list
) {
1504 btrfs_csum_file_blocks(trans
,
1505 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1510 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1511 struct extent_state
**cached_state
)
1513 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1515 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1516 cached_state
, GFP_NOFS
);
1519 /* see btrfs_writepage_start_hook for details on why this is required */
1520 struct btrfs_writepage_fixup
{
1522 struct btrfs_work work
;
1525 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1527 struct btrfs_writepage_fixup
*fixup
;
1528 struct btrfs_ordered_extent
*ordered
;
1529 struct extent_state
*cached_state
= NULL
;
1531 struct inode
*inode
;
1535 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1539 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1540 ClearPageChecked(page
);
1544 inode
= page
->mapping
->host
;
1545 page_start
= page_offset(page
);
1546 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1548 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1549 &cached_state
, GFP_NOFS
);
1551 /* already ordered? We're done */
1552 if (PagePrivate2(page
))
1555 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1557 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1558 page_end
, &cached_state
, GFP_NOFS
);
1560 btrfs_start_ordered_extent(inode
, ordered
, 1);
1565 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1566 ClearPageChecked(page
);
1568 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1569 &cached_state
, GFP_NOFS
);
1572 page_cache_release(page
);
1577 * There are a few paths in the higher layers of the kernel that directly
1578 * set the page dirty bit without asking the filesystem if it is a
1579 * good idea. This causes problems because we want to make sure COW
1580 * properly happens and the data=ordered rules are followed.
1582 * In our case any range that doesn't have the ORDERED bit set
1583 * hasn't been properly setup for IO. We kick off an async process
1584 * to fix it up. The async helper will wait for ordered extents, set
1585 * the delalloc bit and make it safe to write the page.
1587 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1589 struct inode
*inode
= page
->mapping
->host
;
1590 struct btrfs_writepage_fixup
*fixup
;
1591 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1593 /* this page is properly in the ordered list */
1594 if (TestClearPagePrivate2(page
))
1597 if (PageChecked(page
))
1600 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1604 SetPageChecked(page
);
1605 page_cache_get(page
);
1606 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1608 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1612 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1613 struct inode
*inode
, u64 file_pos
,
1614 u64 disk_bytenr
, u64 disk_num_bytes
,
1615 u64 num_bytes
, u64 ram_bytes
,
1616 u8 compression
, u8 encryption
,
1617 u16 other_encoding
, int extent_type
)
1619 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1620 struct btrfs_file_extent_item
*fi
;
1621 struct btrfs_path
*path
;
1622 struct extent_buffer
*leaf
;
1623 struct btrfs_key ins
;
1627 path
= btrfs_alloc_path();
1630 path
->leave_spinning
= 1;
1633 * we may be replacing one extent in the tree with another.
1634 * The new extent is pinned in the extent map, and we don't want
1635 * to drop it from the cache until it is completely in the btree.
1637 * So, tell btrfs_drop_extents to leave this extent in the cache.
1638 * the caller is expected to unpin it and allow it to be merged
1641 ret
= btrfs_drop_extents(trans
, inode
, file_pos
, file_pos
+ num_bytes
,
1645 ins
.objectid
= inode
->i_ino
;
1646 ins
.offset
= file_pos
;
1647 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1648 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1650 leaf
= path
->nodes
[0];
1651 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1652 struct btrfs_file_extent_item
);
1653 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1654 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1655 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1656 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1657 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1658 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1659 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1660 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1661 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1662 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1664 btrfs_unlock_up_safe(path
, 1);
1665 btrfs_set_lock_blocking(leaf
);
1667 btrfs_mark_buffer_dirty(leaf
);
1669 inode_add_bytes(inode
, num_bytes
);
1671 ins
.objectid
= disk_bytenr
;
1672 ins
.offset
= disk_num_bytes
;
1673 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1674 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1675 root
->root_key
.objectid
,
1676 inode
->i_ino
, file_pos
, &ins
);
1678 btrfs_free_path(path
);
1684 * helper function for btrfs_finish_ordered_io, this
1685 * just reads in some of the csum leaves to prime them into ram
1686 * before we start the transaction. It limits the amount of btree
1687 * reads required while inside the transaction.
1689 /* as ordered data IO finishes, this gets called so we can finish
1690 * an ordered extent if the range of bytes in the file it covers are
1693 static int btrfs_finish_ordered_io(struct inode
*inode
, u64 start
, u64 end
)
1695 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1696 struct btrfs_trans_handle
*trans
= NULL
;
1697 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1698 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1699 struct extent_state
*cached_state
= NULL
;
1700 int compress_type
= 0;
1702 bool nolock
= false;
1704 ret
= btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1708 BUG_ON(!ordered_extent
);
1710 nolock
= (root
== root
->fs_info
->tree_root
);
1712 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1713 BUG_ON(!list_empty(&ordered_extent
->list
));
1714 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1717 trans
= btrfs_join_transaction_nolock(root
, 1);
1719 trans
= btrfs_join_transaction(root
, 1);
1720 BUG_ON(IS_ERR(trans
));
1721 btrfs_set_trans_block_group(trans
, inode
);
1722 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1723 ret
= btrfs_update_inode(trans
, root
, inode
);
1729 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1730 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1731 0, &cached_state
, GFP_NOFS
);
1734 trans
= btrfs_join_transaction_nolock(root
, 1);
1736 trans
= btrfs_join_transaction(root
, 1);
1737 BUG_ON(IS_ERR(trans
));
1738 btrfs_set_trans_block_group(trans
, inode
);
1739 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1741 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1742 compress_type
= ordered_extent
->compress_type
;
1743 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1744 BUG_ON(compress_type
);
1745 ret
= btrfs_mark_extent_written(trans
, inode
,
1746 ordered_extent
->file_offset
,
1747 ordered_extent
->file_offset
+
1748 ordered_extent
->len
);
1751 BUG_ON(root
== root
->fs_info
->tree_root
);
1752 ret
= insert_reserved_file_extent(trans
, inode
,
1753 ordered_extent
->file_offset
,
1754 ordered_extent
->start
,
1755 ordered_extent
->disk_len
,
1756 ordered_extent
->len
,
1757 ordered_extent
->len
,
1758 compress_type
, 0, 0,
1759 BTRFS_FILE_EXTENT_REG
);
1760 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1761 ordered_extent
->file_offset
,
1762 ordered_extent
->len
);
1765 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1766 ordered_extent
->file_offset
+
1767 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1769 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1770 &ordered_extent
->list
);
1772 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1773 ret
= btrfs_update_inode(trans
, root
, inode
);
1778 btrfs_end_transaction_nolock(trans
, root
);
1780 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1782 btrfs_end_transaction(trans
, root
);
1786 btrfs_put_ordered_extent(ordered_extent
);
1787 /* once for the tree */
1788 btrfs_put_ordered_extent(ordered_extent
);
1793 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1794 struct extent_state
*state
, int uptodate
)
1796 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1798 ClearPagePrivate2(page
);
1799 return btrfs_finish_ordered_io(page
->mapping
->host
, start
, end
);
1803 * When IO fails, either with EIO or csum verification fails, we
1804 * try other mirrors that might have a good copy of the data. This
1805 * io_failure_record is used to record state as we go through all the
1806 * mirrors. If another mirror has good data, the page is set up to date
1807 * and things continue. If a good mirror can't be found, the original
1808 * bio end_io callback is called to indicate things have failed.
1810 struct io_failure_record
{
1815 unsigned long bio_flags
;
1819 static int btrfs_io_failed_hook(struct bio
*failed_bio
,
1820 struct page
*page
, u64 start
, u64 end
,
1821 struct extent_state
*state
)
1823 struct io_failure_record
*failrec
= NULL
;
1825 struct extent_map
*em
;
1826 struct inode
*inode
= page
->mapping
->host
;
1827 struct extent_io_tree
*failure_tree
= &BTRFS_I(inode
)->io_failure_tree
;
1828 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
1835 ret
= get_state_private(failure_tree
, start
, &private);
1837 failrec
= kmalloc(sizeof(*failrec
), GFP_NOFS
);
1840 failrec
->start
= start
;
1841 failrec
->len
= end
- start
+ 1;
1842 failrec
->last_mirror
= 0;
1843 failrec
->bio_flags
= 0;
1845 read_lock(&em_tree
->lock
);
1846 em
= lookup_extent_mapping(em_tree
, start
, failrec
->len
);
1847 if (em
->start
> start
|| em
->start
+ em
->len
< start
) {
1848 free_extent_map(em
);
1851 read_unlock(&em_tree
->lock
);
1853 if (!em
|| IS_ERR(em
)) {
1857 logical
= start
- em
->start
;
1858 logical
= em
->block_start
+ logical
;
1859 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
1860 logical
= em
->block_start
;
1861 failrec
->bio_flags
= EXTENT_BIO_COMPRESSED
;
1862 extent_set_compress_type(&failrec
->bio_flags
,
1865 failrec
->logical
= logical
;
1866 free_extent_map(em
);
1867 set_extent_bits(failure_tree
, start
, end
, EXTENT_LOCKED
|
1868 EXTENT_DIRTY
, GFP_NOFS
);
1869 set_state_private(failure_tree
, start
,
1870 (u64
)(unsigned long)failrec
);
1872 failrec
= (struct io_failure_record
*)(unsigned long)private;
1874 num_copies
= btrfs_num_copies(
1875 &BTRFS_I(inode
)->root
->fs_info
->mapping_tree
,
1876 failrec
->logical
, failrec
->len
);
1877 failrec
->last_mirror
++;
1879 spin_lock(&BTRFS_I(inode
)->io_tree
.lock
);
1880 state
= find_first_extent_bit_state(&BTRFS_I(inode
)->io_tree
,
1883 if (state
&& state
->start
!= failrec
->start
)
1885 spin_unlock(&BTRFS_I(inode
)->io_tree
.lock
);
1887 if (!state
|| failrec
->last_mirror
> num_copies
) {
1888 set_state_private(failure_tree
, failrec
->start
, 0);
1889 clear_extent_bits(failure_tree
, failrec
->start
,
1890 failrec
->start
+ failrec
->len
- 1,
1891 EXTENT_LOCKED
| EXTENT_DIRTY
, GFP_NOFS
);
1895 bio
= bio_alloc(GFP_NOFS
, 1);
1896 bio
->bi_private
= state
;
1897 bio
->bi_end_io
= failed_bio
->bi_end_io
;
1898 bio
->bi_sector
= failrec
->logical
>> 9;
1899 bio
->bi_bdev
= failed_bio
->bi_bdev
;
1902 bio_add_page(bio
, page
, failrec
->len
, start
- page_offset(page
));
1903 if (failed_bio
->bi_rw
& REQ_WRITE
)
1908 ret
= BTRFS_I(inode
)->io_tree
.ops
->submit_bio_hook(inode
, rw
, bio
,
1909 failrec
->last_mirror
,
1910 failrec
->bio_flags
, 0);
1915 * each time an IO finishes, we do a fast check in the IO failure tree
1916 * to see if we need to process or clean up an io_failure_record
1918 static int btrfs_clean_io_failures(struct inode
*inode
, u64 start
)
1921 u64 private_failure
;
1922 struct io_failure_record
*failure
;
1926 if (count_range_bits(&BTRFS_I(inode
)->io_failure_tree
, &private,
1927 (u64
)-1, 1, EXTENT_DIRTY
, 0)) {
1928 ret
= get_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1929 start
, &private_failure
);
1931 failure
= (struct io_failure_record
*)(unsigned long)
1933 set_state_private(&BTRFS_I(inode
)->io_failure_tree
,
1935 clear_extent_bits(&BTRFS_I(inode
)->io_failure_tree
,
1937 failure
->start
+ failure
->len
- 1,
1938 EXTENT_DIRTY
| EXTENT_LOCKED
,
1947 * when reads are done, we need to check csums to verify the data is correct
1948 * if there's a match, we allow the bio to finish. If not, we go through
1949 * the io_failure_record routines to find good copies
1951 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1952 struct extent_state
*state
)
1954 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
1955 struct inode
*inode
= page
->mapping
->host
;
1956 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1958 u64
private = ~(u32
)0;
1960 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1963 if (PageChecked(page
)) {
1964 ClearPageChecked(page
);
1968 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
1971 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
1972 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
1973 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
1978 if (state
&& state
->start
== start
) {
1979 private = state
->private;
1982 ret
= get_state_private(io_tree
, start
, &private);
1984 kaddr
= kmap_atomic(page
, KM_USER0
);
1988 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
1989 btrfs_csum_final(csum
, (char *)&csum
);
1990 if (csum
!= private)
1993 kunmap_atomic(kaddr
, KM_USER0
);
1995 /* if the io failure tree for this inode is non-empty,
1996 * check to see if we've recovered from a failed IO
1998 btrfs_clean_io_failures(inode
, start
);
2002 if (printk_ratelimit()) {
2003 printk(KERN_INFO
"btrfs csum failed ino %lu off %llu csum %u "
2004 "private %llu\n", page
->mapping
->host
->i_ino
,
2005 (unsigned long long)start
, csum
,
2006 (unsigned long long)private);
2008 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2009 flush_dcache_page(page
);
2010 kunmap_atomic(kaddr
, KM_USER0
);
2016 struct delayed_iput
{
2017 struct list_head list
;
2018 struct inode
*inode
;
2021 void btrfs_add_delayed_iput(struct inode
*inode
)
2023 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2024 struct delayed_iput
*delayed
;
2026 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2029 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2030 delayed
->inode
= inode
;
2032 spin_lock(&fs_info
->delayed_iput_lock
);
2033 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2034 spin_unlock(&fs_info
->delayed_iput_lock
);
2037 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2040 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2041 struct delayed_iput
*delayed
;
2044 spin_lock(&fs_info
->delayed_iput_lock
);
2045 empty
= list_empty(&fs_info
->delayed_iputs
);
2046 spin_unlock(&fs_info
->delayed_iput_lock
);
2050 down_read(&root
->fs_info
->cleanup_work_sem
);
2051 spin_lock(&fs_info
->delayed_iput_lock
);
2052 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2053 spin_unlock(&fs_info
->delayed_iput_lock
);
2055 while (!list_empty(&list
)) {
2056 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2057 list_del(&delayed
->list
);
2058 iput(delayed
->inode
);
2061 up_read(&root
->fs_info
->cleanup_work_sem
);
2065 * calculate extra metadata reservation when snapshotting a subvolume
2066 * contains orphan files.
2068 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle
*trans
,
2069 struct btrfs_pending_snapshot
*pending
,
2070 u64
*bytes_to_reserve
)
2072 struct btrfs_root
*root
;
2073 struct btrfs_block_rsv
*block_rsv
;
2077 root
= pending
->root
;
2078 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2081 block_rsv
= root
->orphan_block_rsv
;
2083 /* orphan block reservation for the snapshot */
2084 num_bytes
= block_rsv
->size
;
2087 * after the snapshot is created, COWing tree blocks may use more
2088 * space than it frees. So we should make sure there is enough
2091 index
= trans
->transid
& 0x1;
2092 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2093 num_bytes
+= block_rsv
->size
-
2094 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2097 *bytes_to_reserve
+= num_bytes
;
2100 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle
*trans
,
2101 struct btrfs_pending_snapshot
*pending
)
2103 struct btrfs_root
*root
= pending
->root
;
2104 struct btrfs_root
*snap
= pending
->snap
;
2105 struct btrfs_block_rsv
*block_rsv
;
2110 if (!root
->orphan_block_rsv
|| list_empty(&root
->orphan_list
))
2113 /* refill source subvolume's orphan block reservation */
2114 block_rsv
= root
->orphan_block_rsv
;
2115 index
= trans
->transid
& 0x1;
2116 if (block_rsv
->reserved
+ block_rsv
->freed
[index
] < block_rsv
->size
) {
2117 num_bytes
= block_rsv
->size
-
2118 (block_rsv
->reserved
+ block_rsv
->freed
[index
]);
2119 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2120 root
->orphan_block_rsv
,
2125 /* setup orphan block reservation for the snapshot */
2126 block_rsv
= btrfs_alloc_block_rsv(snap
);
2129 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2130 snap
->orphan_block_rsv
= block_rsv
;
2132 num_bytes
= root
->orphan_block_rsv
->size
;
2133 ret
= btrfs_block_rsv_migrate(&pending
->block_rsv
,
2134 block_rsv
, num_bytes
);
2138 /* insert orphan item for the snapshot */
2139 WARN_ON(!root
->orphan_item_inserted
);
2140 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2141 snap
->root_key
.objectid
);
2143 snap
->orphan_item_inserted
= 1;
2147 enum btrfs_orphan_cleanup_state
{
2148 ORPHAN_CLEANUP_STARTED
= 1,
2149 ORPHAN_CLEANUP_DONE
= 2,
2153 * This is called in transaction commmit time. If there are no orphan
2154 * files in the subvolume, it removes orphan item and frees block_rsv
2157 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2158 struct btrfs_root
*root
)
2162 if (!list_empty(&root
->orphan_list
) ||
2163 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2166 if (root
->orphan_item_inserted
&&
2167 btrfs_root_refs(&root
->root_item
) > 0) {
2168 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2169 root
->root_key
.objectid
);
2171 root
->orphan_item_inserted
= 0;
2174 if (root
->orphan_block_rsv
) {
2175 WARN_ON(root
->orphan_block_rsv
->size
> 0);
2176 btrfs_free_block_rsv(root
, root
->orphan_block_rsv
);
2177 root
->orphan_block_rsv
= NULL
;
2182 * This creates an orphan entry for the given inode in case something goes
2183 * wrong in the middle of an unlink/truncate.
2185 * NOTE: caller of this function should reserve 5 units of metadata for
2188 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2190 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2191 struct btrfs_block_rsv
*block_rsv
= NULL
;
2196 if (!root
->orphan_block_rsv
) {
2197 block_rsv
= btrfs_alloc_block_rsv(root
);
2201 spin_lock(&root
->orphan_lock
);
2202 if (!root
->orphan_block_rsv
) {
2203 root
->orphan_block_rsv
= block_rsv
;
2204 } else if (block_rsv
) {
2205 btrfs_free_block_rsv(root
, block_rsv
);
2209 if (list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2210 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2213 * For proper ENOSPC handling, we should do orphan
2214 * cleanup when mounting. But this introduces backward
2215 * compatibility issue.
2217 if (!xchg(&root
->orphan_item_inserted
, 1))
2225 if (!BTRFS_I(inode
)->orphan_meta_reserved
) {
2226 BTRFS_I(inode
)->orphan_meta_reserved
= 1;
2229 spin_unlock(&root
->orphan_lock
);
2232 btrfs_add_durable_block_rsv(root
->fs_info
, block_rsv
);
2234 /* grab metadata reservation from transaction handle */
2236 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2240 /* insert an orphan item to track this unlinked/truncated file */
2242 ret
= btrfs_insert_orphan_item(trans
, root
, inode
->i_ino
);
2246 /* insert an orphan item to track subvolume contains orphan files */
2248 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2249 root
->root_key
.objectid
);
2256 * We have done the truncate/delete so we can go ahead and remove the orphan
2257 * item for this particular inode.
2259 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2261 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2262 int delete_item
= 0;
2263 int release_rsv
= 0;
2266 spin_lock(&root
->orphan_lock
);
2267 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
2268 list_del_init(&BTRFS_I(inode
)->i_orphan
);
2272 if (BTRFS_I(inode
)->orphan_meta_reserved
) {
2273 BTRFS_I(inode
)->orphan_meta_reserved
= 0;
2276 spin_unlock(&root
->orphan_lock
);
2278 if (trans
&& delete_item
) {
2279 ret
= btrfs_del_orphan_item(trans
, root
, inode
->i_ino
);
2284 btrfs_orphan_release_metadata(inode
);
2290 * this cleans up any orphans that may be left on the list from the last use
2293 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2295 struct btrfs_path
*path
;
2296 struct extent_buffer
*leaf
;
2297 struct btrfs_key key
, found_key
;
2298 struct btrfs_trans_handle
*trans
;
2299 struct inode
*inode
;
2300 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2302 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2305 path
= btrfs_alloc_path();
2312 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2313 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2314 key
.offset
= (u64
)-1;
2317 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2322 * if ret == 0 means we found what we were searching for, which
2323 * is weird, but possible, so only screw with path if we didnt
2324 * find the key and see if we have stuff that matches
2328 if (path
->slots
[0] == 0)
2333 /* pull out the item */
2334 leaf
= path
->nodes
[0];
2335 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2337 /* make sure the item matches what we want */
2338 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2340 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2343 /* release the path since we're done with it */
2344 btrfs_release_path(root
, path
);
2347 * this is where we are basically btrfs_lookup, without the
2348 * crossing root thing. we store the inode number in the
2349 * offset of the orphan item.
2351 found_key
.objectid
= found_key
.offset
;
2352 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2353 found_key
.offset
= 0;
2354 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2355 if (IS_ERR(inode
)) {
2356 ret
= PTR_ERR(inode
);
2361 * add this inode to the orphan list so btrfs_orphan_del does
2362 * the proper thing when we hit it
2364 spin_lock(&root
->orphan_lock
);
2365 list_add(&BTRFS_I(inode
)->i_orphan
, &root
->orphan_list
);
2366 spin_unlock(&root
->orphan_lock
);
2369 * if this is a bad inode, means we actually succeeded in
2370 * removing the inode, but not the orphan record, which means
2371 * we need to manually delete the orphan since iput will just
2372 * do a destroy_inode
2374 if (is_bad_inode(inode
)) {
2375 trans
= btrfs_start_transaction(root
, 0);
2376 if (IS_ERR(trans
)) {
2377 ret
= PTR_ERR(trans
);
2380 btrfs_orphan_del(trans
, inode
);
2381 btrfs_end_transaction(trans
, root
);
2386 /* if we have links, this was a truncate, lets do that */
2387 if (inode
->i_nlink
) {
2388 if (!S_ISREG(inode
->i_mode
)) {
2394 ret
= btrfs_truncate(inode
);
2399 /* this will do delete_inode and everything for us */
2404 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2406 if (root
->orphan_block_rsv
)
2407 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2410 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2411 trans
= btrfs_join_transaction(root
, 1);
2413 btrfs_end_transaction(trans
, root
);
2417 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2419 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2423 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2424 btrfs_free_path(path
);
2429 * very simple check to peek ahead in the leaf looking for xattrs. If we
2430 * don't find any xattrs, we know there can't be any acls.
2432 * slot is the slot the inode is in, objectid is the objectid of the inode
2434 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2435 int slot
, u64 objectid
)
2437 u32 nritems
= btrfs_header_nritems(leaf
);
2438 struct btrfs_key found_key
;
2442 while (slot
< nritems
) {
2443 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2445 /* we found a different objectid, there must not be acls */
2446 if (found_key
.objectid
!= objectid
)
2449 /* we found an xattr, assume we've got an acl */
2450 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2454 * we found a key greater than an xattr key, there can't
2455 * be any acls later on
2457 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2464 * it goes inode, inode backrefs, xattrs, extents,
2465 * so if there are a ton of hard links to an inode there can
2466 * be a lot of backrefs. Don't waste time searching too hard,
2467 * this is just an optimization
2472 /* we hit the end of the leaf before we found an xattr or
2473 * something larger than an xattr. We have to assume the inode
2480 * read an inode from the btree into the in-memory inode
2482 static void btrfs_read_locked_inode(struct inode
*inode
)
2484 struct btrfs_path
*path
;
2485 struct extent_buffer
*leaf
;
2486 struct btrfs_inode_item
*inode_item
;
2487 struct btrfs_timespec
*tspec
;
2488 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2489 struct btrfs_key location
;
2491 u64 alloc_group_block
;
2495 path
= btrfs_alloc_path();
2497 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2499 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2503 leaf
= path
->nodes
[0];
2504 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2505 struct btrfs_inode_item
);
2507 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2508 inode
->i_nlink
= btrfs_inode_nlink(leaf
, inode_item
);
2509 inode
->i_uid
= btrfs_inode_uid(leaf
, inode_item
);
2510 inode
->i_gid
= btrfs_inode_gid(leaf
, inode_item
);
2511 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2513 tspec
= btrfs_inode_atime(inode_item
);
2514 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2515 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2517 tspec
= btrfs_inode_mtime(inode_item
);
2518 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2519 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2521 tspec
= btrfs_inode_ctime(inode_item
);
2522 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2523 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2525 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2526 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2527 BTRFS_I(inode
)->sequence
= btrfs_inode_sequence(leaf
, inode_item
);
2528 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2530 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2532 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2533 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2535 alloc_group_block
= btrfs_inode_block_group(leaf
, inode_item
);
2538 * try to precache a NULL acl entry for files that don't have
2539 * any xattrs or acls
2541 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0], inode
->i_ino
);
2543 cache_no_acl(inode
);
2545 BTRFS_I(inode
)->block_group
= btrfs_find_block_group(root
, 0,
2546 alloc_group_block
, 0);
2547 btrfs_free_path(path
);
2550 switch (inode
->i_mode
& S_IFMT
) {
2552 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2553 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2554 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2555 inode
->i_fop
= &btrfs_file_operations
;
2556 inode
->i_op
= &btrfs_file_inode_operations
;
2559 inode
->i_fop
= &btrfs_dir_file_operations
;
2560 if (root
== root
->fs_info
->tree_root
)
2561 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2563 inode
->i_op
= &btrfs_dir_inode_operations
;
2566 inode
->i_op
= &btrfs_symlink_inode_operations
;
2567 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2568 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2571 inode
->i_op
= &btrfs_special_inode_operations
;
2572 init_special_inode(inode
, inode
->i_mode
, rdev
);
2576 btrfs_update_iflags(inode
);
2580 btrfs_free_path(path
);
2581 make_bad_inode(inode
);
2585 * given a leaf and an inode, copy the inode fields into the leaf
2587 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2588 struct extent_buffer
*leaf
,
2589 struct btrfs_inode_item
*item
,
2590 struct inode
*inode
)
2592 btrfs_set_inode_uid(leaf
, item
, inode
->i_uid
);
2593 btrfs_set_inode_gid(leaf
, item
, inode
->i_gid
);
2594 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2595 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2596 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2598 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2599 inode
->i_atime
.tv_sec
);
2600 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2601 inode
->i_atime
.tv_nsec
);
2603 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2604 inode
->i_mtime
.tv_sec
);
2605 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2606 inode
->i_mtime
.tv_nsec
);
2608 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2609 inode
->i_ctime
.tv_sec
);
2610 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2611 inode
->i_ctime
.tv_nsec
);
2613 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2614 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2615 btrfs_set_inode_sequence(leaf
, item
, BTRFS_I(inode
)->sequence
);
2616 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2617 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2618 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2619 btrfs_set_inode_block_group(leaf
, item
, BTRFS_I(inode
)->block_group
);
2623 * copy everything in the in-memory inode into the btree.
2625 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2626 struct btrfs_root
*root
, struct inode
*inode
)
2628 struct btrfs_inode_item
*inode_item
;
2629 struct btrfs_path
*path
;
2630 struct extent_buffer
*leaf
;
2633 path
= btrfs_alloc_path();
2635 path
->leave_spinning
= 1;
2636 ret
= btrfs_lookup_inode(trans
, root
, path
,
2637 &BTRFS_I(inode
)->location
, 1);
2644 btrfs_unlock_up_safe(path
, 1);
2645 leaf
= path
->nodes
[0];
2646 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2647 struct btrfs_inode_item
);
2649 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2650 btrfs_mark_buffer_dirty(leaf
);
2651 btrfs_set_inode_last_trans(trans
, inode
);
2654 btrfs_free_path(path
);
2660 * unlink helper that gets used here in inode.c and in the tree logging
2661 * recovery code. It remove a link in a directory with a given name, and
2662 * also drops the back refs in the inode to the directory
2664 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2665 struct btrfs_root
*root
,
2666 struct inode
*dir
, struct inode
*inode
,
2667 const char *name
, int name_len
)
2669 struct btrfs_path
*path
;
2671 struct extent_buffer
*leaf
;
2672 struct btrfs_dir_item
*di
;
2673 struct btrfs_key key
;
2676 path
= btrfs_alloc_path();
2682 path
->leave_spinning
= 1;
2683 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2684 name
, name_len
, -1);
2693 leaf
= path
->nodes
[0];
2694 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2695 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2698 btrfs_release_path(root
, path
);
2700 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
2702 dir
->i_ino
, &index
);
2704 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2705 "inode %lu parent %lu\n", name_len
, name
,
2706 inode
->i_ino
, dir
->i_ino
);
2710 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
2711 index
, name
, name_len
, -1);
2720 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2721 btrfs_release_path(root
, path
);
2723 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2725 BUG_ON(ret
!= 0 && ret
!= -ENOENT
);
2727 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2732 btrfs_free_path(path
);
2736 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2737 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2738 btrfs_update_inode(trans
, root
, dir
);
2743 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2744 struct btrfs_root
*root
,
2745 struct inode
*dir
, struct inode
*inode
,
2746 const char *name
, int name_len
)
2749 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2751 btrfs_drop_nlink(inode
);
2752 ret
= btrfs_update_inode(trans
, root
, inode
);
2758 /* helper to check if there is any shared block in the path */
2759 static int check_path_shared(struct btrfs_root
*root
,
2760 struct btrfs_path
*path
)
2762 struct extent_buffer
*eb
;
2766 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2769 if (!path
->nodes
[level
])
2771 eb
= path
->nodes
[level
];
2772 if (!btrfs_block_can_be_shared(root
, eb
))
2774 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2783 * helper to start transaction for unlink and rmdir.
2785 * unlink and rmdir are special in btrfs, they do not always free space.
2786 * so in enospc case, we should make sure they will free space before
2787 * allowing them to use the global metadata reservation.
2789 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2790 struct dentry
*dentry
)
2792 struct btrfs_trans_handle
*trans
;
2793 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2794 struct btrfs_path
*path
;
2795 struct btrfs_inode_ref
*ref
;
2796 struct btrfs_dir_item
*di
;
2797 struct inode
*inode
= dentry
->d_inode
;
2803 trans
= btrfs_start_transaction(root
, 10);
2804 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2807 if (inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2808 return ERR_PTR(-ENOSPC
);
2810 /* check if there is someone else holds reference */
2811 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2812 return ERR_PTR(-ENOSPC
);
2814 if (atomic_read(&inode
->i_count
) > 2)
2815 return ERR_PTR(-ENOSPC
);
2817 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2818 return ERR_PTR(-ENOSPC
);
2820 path
= btrfs_alloc_path();
2822 root
->fs_info
->enospc_unlink
= 0;
2823 return ERR_PTR(-ENOMEM
);
2826 trans
= btrfs_start_transaction(root
, 0);
2827 if (IS_ERR(trans
)) {
2828 btrfs_free_path(path
);
2829 root
->fs_info
->enospc_unlink
= 0;
2833 path
->skip_locking
= 1;
2834 path
->search_commit_root
= 1;
2836 ret
= btrfs_lookup_inode(trans
, root
, path
,
2837 &BTRFS_I(dir
)->location
, 0);
2843 if (check_path_shared(root
, path
))
2848 btrfs_release_path(root
, path
);
2850 ret
= btrfs_lookup_inode(trans
, root
, path
,
2851 &BTRFS_I(inode
)->location
, 0);
2857 if (check_path_shared(root
, path
))
2862 btrfs_release_path(root
, path
);
2864 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2865 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2866 inode
->i_ino
, (u64
)-1, 0);
2872 if (check_path_shared(root
, path
))
2874 btrfs_release_path(root
, path
);
2882 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2883 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2889 if (check_path_shared(root
, path
))
2895 btrfs_release_path(root
, path
);
2897 ref
= btrfs_lookup_inode_ref(trans
, root
, path
,
2898 dentry
->d_name
.name
, dentry
->d_name
.len
,
2899 inode
->i_ino
, dir
->i_ino
, 0);
2905 if (check_path_shared(root
, path
))
2907 index
= btrfs_inode_ref_index(path
->nodes
[0], ref
);
2908 btrfs_release_path(root
, path
);
2910 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
, index
,
2911 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
2916 BUG_ON(ret
== -ENOENT
);
2917 if (check_path_shared(root
, path
))
2922 btrfs_free_path(path
);
2924 btrfs_end_transaction(trans
, root
);
2925 root
->fs_info
->enospc_unlink
= 0;
2926 return ERR_PTR(err
);
2929 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
2933 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
2934 struct btrfs_root
*root
)
2936 if (trans
->block_rsv
== &root
->fs_info
->global_block_rsv
) {
2937 BUG_ON(!root
->fs_info
->enospc_unlink
);
2938 root
->fs_info
->enospc_unlink
= 0;
2940 btrfs_end_transaction_throttle(trans
, root
);
2943 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2945 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2946 struct btrfs_trans_handle
*trans
;
2947 struct inode
*inode
= dentry
->d_inode
;
2949 unsigned long nr
= 0;
2951 trans
= __unlink_start_trans(dir
, dentry
);
2953 return PTR_ERR(trans
);
2955 btrfs_set_trans_block_group(trans
, dir
);
2957 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
2959 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
2960 dentry
->d_name
.name
, dentry
->d_name
.len
);
2963 if (inode
->i_nlink
== 0) {
2964 ret
= btrfs_orphan_add(trans
, inode
);
2968 nr
= trans
->blocks_used
;
2969 __unlink_end_trans(trans
, root
);
2970 btrfs_btree_balance_dirty(root
, nr
);
2974 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
2975 struct btrfs_root
*root
,
2976 struct inode
*dir
, u64 objectid
,
2977 const char *name
, int name_len
)
2979 struct btrfs_path
*path
;
2980 struct extent_buffer
*leaf
;
2981 struct btrfs_dir_item
*di
;
2982 struct btrfs_key key
;
2986 path
= btrfs_alloc_path();
2990 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir
->i_ino
,
2991 name
, name_len
, -1);
2992 BUG_ON(!di
|| IS_ERR(di
));
2994 leaf
= path
->nodes
[0];
2995 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2996 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
2997 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2999 btrfs_release_path(root
, path
);
3001 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3002 objectid
, root
->root_key
.objectid
,
3003 dir
->i_ino
, &index
, name
, name_len
);
3005 BUG_ON(ret
!= -ENOENT
);
3006 di
= btrfs_search_dir_index_item(root
, path
, dir
->i_ino
,
3008 BUG_ON(!di
|| IS_ERR(di
));
3010 leaf
= path
->nodes
[0];
3011 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3012 btrfs_release_path(root
, path
);
3016 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir
->i_ino
,
3017 index
, name
, name_len
, -1);
3018 BUG_ON(!di
|| IS_ERR(di
));
3020 leaf
= path
->nodes
[0];
3021 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3022 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3023 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3025 btrfs_release_path(root
, path
);
3027 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3028 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3029 ret
= btrfs_update_inode(trans
, root
, dir
);
3032 btrfs_free_path(path
);
3036 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3038 struct inode
*inode
= dentry
->d_inode
;
3040 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3041 struct btrfs_trans_handle
*trans
;
3042 unsigned long nr
= 0;
3044 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
||
3045 inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
3048 trans
= __unlink_start_trans(dir
, dentry
);
3050 return PTR_ERR(trans
);
3052 btrfs_set_trans_block_group(trans
, dir
);
3054 if (unlikely(inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3055 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3056 BTRFS_I(inode
)->location
.objectid
,
3057 dentry
->d_name
.name
,
3058 dentry
->d_name
.len
);
3062 err
= btrfs_orphan_add(trans
, inode
);
3066 /* now the directory is empty */
3067 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3068 dentry
->d_name
.name
, dentry
->d_name
.len
);
3070 btrfs_i_size_write(inode
, 0);
3072 nr
= trans
->blocks_used
;
3073 __unlink_end_trans(trans
, root
);
3074 btrfs_btree_balance_dirty(root
, nr
);
3081 * when truncating bytes in a file, it is possible to avoid reading
3082 * the leaves that contain only checksum items. This can be the
3083 * majority of the IO required to delete a large file, but it must
3084 * be done carefully.
3086 * The keys in the level just above the leaves are checked to make sure
3087 * the lowest key in a given leaf is a csum key, and starts at an offset
3088 * after the new size.
3090 * Then the key for the next leaf is checked to make sure it also has
3091 * a checksum item for the same file. If it does, we know our target leaf
3092 * contains only checksum items, and it can be safely freed without reading
3095 * This is just an optimization targeted at large files. It may do
3096 * nothing. It will return 0 unless things went badly.
3098 static noinline
int drop_csum_leaves(struct btrfs_trans_handle
*trans
,
3099 struct btrfs_root
*root
,
3100 struct btrfs_path
*path
,
3101 struct inode
*inode
, u64 new_size
)
3103 struct btrfs_key key
;
3106 struct btrfs_key found_key
;
3107 struct btrfs_key other_key
;
3108 struct btrfs_leaf_ref
*ref
;
3112 path
->lowest_level
= 1;
3113 key
.objectid
= inode
->i_ino
;
3114 key
.type
= BTRFS_CSUM_ITEM_KEY
;
3115 key
.offset
= new_size
;
3117 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3121 if (path
->nodes
[1] == NULL
) {
3126 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, path
->slots
[1]);
3127 nritems
= btrfs_header_nritems(path
->nodes
[1]);
3132 if (path
->slots
[1] >= nritems
)
3135 /* did we find a key greater than anything we want to delete? */
3136 if (found_key
.objectid
> inode
->i_ino
||
3137 (found_key
.objectid
== inode
->i_ino
&& found_key
.type
> key
.type
))
3140 /* we check the next key in the node to make sure the leave contains
3141 * only checksum items. This comparison doesn't work if our
3142 * leaf is the last one in the node
3144 if (path
->slots
[1] + 1 >= nritems
) {
3146 /* search forward from the last key in the node, this
3147 * will bring us into the next node in the tree
3149 btrfs_node_key_to_cpu(path
->nodes
[1], &found_key
, nritems
- 1);
3151 /* unlikely, but we inc below, so check to be safe */
3152 if (found_key
.offset
== (u64
)-1)
3155 /* search_forward needs a path with locks held, do the
3156 * search again for the original key. It is possible
3157 * this will race with a balance and return a path that
3158 * we could modify, but this drop is just an optimization
3159 * and is allowed to miss some leaves.
3161 btrfs_release_path(root
, path
);
3164 /* setup a max key for search_forward */
3165 other_key
.offset
= (u64
)-1;
3166 other_key
.type
= key
.type
;
3167 other_key
.objectid
= key
.objectid
;
3169 path
->keep_locks
= 1;
3170 ret
= btrfs_search_forward(root
, &found_key
, &other_key
,
3172 path
->keep_locks
= 0;
3173 if (ret
|| found_key
.objectid
!= key
.objectid
||
3174 found_key
.type
!= key
.type
) {
3179 key
.offset
= found_key
.offset
;
3180 btrfs_release_path(root
, path
);
3185 /* we know there's one more slot after us in the tree,
3186 * read that key so we can verify it is also a checksum item
3188 btrfs_node_key_to_cpu(path
->nodes
[1], &other_key
, path
->slots
[1] + 1);
3190 if (found_key
.objectid
< inode
->i_ino
)
3193 if (found_key
.type
!= key
.type
|| found_key
.offset
< new_size
)
3197 * if the key for the next leaf isn't a csum key from this objectid,
3198 * we can't be sure there aren't good items inside this leaf.
3201 if (other_key
.objectid
!= inode
->i_ino
|| other_key
.type
!= key
.type
)
3204 leaf_start
= btrfs_node_blockptr(path
->nodes
[1], path
->slots
[1]);
3205 leaf_gen
= btrfs_node_ptr_generation(path
->nodes
[1], path
->slots
[1]);
3207 * it is safe to delete this leaf, it contains only
3208 * csum items from this inode at an offset >= new_size
3210 ret
= btrfs_del_leaf(trans
, root
, path
, leaf_start
);
3213 if (root
->ref_cows
&& leaf_gen
< trans
->transid
) {
3214 ref
= btrfs_alloc_leaf_ref(root
, 0);
3216 ref
->root_gen
= root
->root_key
.offset
;
3217 ref
->bytenr
= leaf_start
;
3219 ref
->generation
= leaf_gen
;
3222 btrfs_sort_leaf_ref(ref
);
3224 ret
= btrfs_add_leaf_ref(root
, ref
, 0);
3226 btrfs_free_leaf_ref(root
, ref
);
3232 btrfs_release_path(root
, path
);
3234 if (other_key
.objectid
== inode
->i_ino
&&
3235 other_key
.type
== key
.type
&& other_key
.offset
> key
.offset
) {
3236 key
.offset
= other_key
.offset
;
3242 /* fixup any changes we've made to the path */
3243 path
->lowest_level
= 0;
3244 path
->keep_locks
= 0;
3245 btrfs_release_path(root
, path
);
3252 * this can truncate away extent items, csum items and directory items.
3253 * It starts at a high offset and removes keys until it can't find
3254 * any higher than new_size
3256 * csum items that cross the new i_size are truncated to the new size
3259 * min_type is the minimum key type to truncate down to. If set to 0, this
3260 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3262 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3263 struct btrfs_root
*root
,
3264 struct inode
*inode
,
3265 u64 new_size
, u32 min_type
)
3267 struct btrfs_path
*path
;
3268 struct extent_buffer
*leaf
;
3269 struct btrfs_file_extent_item
*fi
;
3270 struct btrfs_key key
;
3271 struct btrfs_key found_key
;
3272 u64 extent_start
= 0;
3273 u64 extent_num_bytes
= 0;
3274 u64 extent_offset
= 0;
3276 u64 mask
= root
->sectorsize
- 1;
3277 u32 found_type
= (u8
)-1;
3280 int pending_del_nr
= 0;
3281 int pending_del_slot
= 0;
3282 int extent_type
= -1;
3287 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3289 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3290 btrfs_drop_extent_cache(inode
, new_size
& (~mask
), (u64
)-1, 0);
3292 path
= btrfs_alloc_path();
3296 key
.objectid
= inode
->i_ino
;
3297 key
.offset
= (u64
)-1;
3301 path
->leave_spinning
= 1;
3302 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3309 /* there are no items in the tree for us to truncate, we're
3312 if (path
->slots
[0] == 0)
3319 leaf
= path
->nodes
[0];
3320 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3321 found_type
= btrfs_key_type(&found_key
);
3324 if (found_key
.objectid
!= inode
->i_ino
)
3327 if (found_type
< min_type
)
3330 item_end
= found_key
.offset
;
3331 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3332 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3333 struct btrfs_file_extent_item
);
3334 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3335 encoding
= btrfs_file_extent_compression(leaf
, fi
);
3336 encoding
|= btrfs_file_extent_encryption(leaf
, fi
);
3337 encoding
|= btrfs_file_extent_other_encoding(leaf
, fi
);
3339 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3341 btrfs_file_extent_num_bytes(leaf
, fi
);
3342 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3343 item_end
+= btrfs_file_extent_inline_len(leaf
,
3348 if (found_type
> min_type
) {
3351 if (item_end
< new_size
)
3353 if (found_key
.offset
>= new_size
)
3359 /* FIXME, shrink the extent if the ref count is only 1 */
3360 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3363 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3365 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3366 if (!del_item
&& !encoding
) {
3367 u64 orig_num_bytes
=
3368 btrfs_file_extent_num_bytes(leaf
, fi
);
3369 extent_num_bytes
= new_size
-
3370 found_key
.offset
+ root
->sectorsize
- 1;
3371 extent_num_bytes
= extent_num_bytes
&
3372 ~((u64
)root
->sectorsize
- 1);
3373 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3375 num_dec
= (orig_num_bytes
-
3377 if (root
->ref_cows
&& extent_start
!= 0)
3378 inode_sub_bytes(inode
, num_dec
);
3379 btrfs_mark_buffer_dirty(leaf
);
3382 btrfs_file_extent_disk_num_bytes(leaf
,
3384 extent_offset
= found_key
.offset
-
3385 btrfs_file_extent_offset(leaf
, fi
);
3387 /* FIXME blocksize != 4096 */
3388 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3389 if (extent_start
!= 0) {
3392 inode_sub_bytes(inode
, num_dec
);
3395 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3397 * we can't truncate inline items that have had
3401 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3402 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3403 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3404 u32 size
= new_size
- found_key
.offset
;
3406 if (root
->ref_cows
) {
3407 inode_sub_bytes(inode
, item_end
+ 1 -
3411 btrfs_file_extent_calc_inline_size(size
);
3412 ret
= btrfs_truncate_item(trans
, root
, path
,
3415 } else if (root
->ref_cows
) {
3416 inode_sub_bytes(inode
, item_end
+ 1 -
3422 if (!pending_del_nr
) {
3423 /* no pending yet, add ourselves */
3424 pending_del_slot
= path
->slots
[0];
3426 } else if (pending_del_nr
&&
3427 path
->slots
[0] + 1 == pending_del_slot
) {
3428 /* hop on the pending chunk */
3430 pending_del_slot
= path
->slots
[0];
3437 if (found_extent
&& (root
->ref_cows
||
3438 root
== root
->fs_info
->tree_root
)) {
3439 btrfs_set_path_blocking(path
);
3440 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3441 extent_num_bytes
, 0,
3442 btrfs_header_owner(leaf
),
3443 inode
->i_ino
, extent_offset
);
3447 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3450 if (path
->slots
[0] == 0 ||
3451 path
->slots
[0] != pending_del_slot
) {
3452 if (root
->ref_cows
) {
3456 if (pending_del_nr
) {
3457 ret
= btrfs_del_items(trans
, root
, path
,
3463 btrfs_release_path(root
, path
);
3470 if (pending_del_nr
) {
3471 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3475 btrfs_free_path(path
);
3480 * taken from block_truncate_page, but does cow as it zeros out
3481 * any bytes left in the last page in the file.
3483 static int btrfs_truncate_page(struct address_space
*mapping
, loff_t from
)
3485 struct inode
*inode
= mapping
->host
;
3486 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3487 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3488 struct btrfs_ordered_extent
*ordered
;
3489 struct extent_state
*cached_state
= NULL
;
3491 u32 blocksize
= root
->sectorsize
;
3492 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3493 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3499 if ((offset
& (blocksize
- 1)) == 0)
3501 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3507 page
= grab_cache_page(mapping
, index
);
3509 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3513 page_start
= page_offset(page
);
3514 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3516 if (!PageUptodate(page
)) {
3517 ret
= btrfs_readpage(NULL
, page
);
3519 if (page
->mapping
!= mapping
) {
3521 page_cache_release(page
);
3524 if (!PageUptodate(page
)) {
3529 wait_on_page_writeback(page
);
3531 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
3533 set_page_extent_mapped(page
);
3535 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3537 unlock_extent_cached(io_tree
, page_start
, page_end
,
3538 &cached_state
, GFP_NOFS
);
3540 page_cache_release(page
);
3541 btrfs_start_ordered_extent(inode
, ordered
, 1);
3542 btrfs_put_ordered_extent(ordered
);
3546 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3547 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
3548 0, 0, &cached_state
, GFP_NOFS
);
3550 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3553 unlock_extent_cached(io_tree
, page_start
, page_end
,
3554 &cached_state
, GFP_NOFS
);
3559 if (offset
!= PAGE_CACHE_SIZE
) {
3561 memset(kaddr
+ offset
, 0, PAGE_CACHE_SIZE
- offset
);
3562 flush_dcache_page(page
);
3565 ClearPageChecked(page
);
3566 set_page_dirty(page
);
3567 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3572 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3574 page_cache_release(page
);
3580 * This function puts in dummy file extents for the area we're creating a hole
3581 * for. So if we are truncating this file to a larger size we need to insert
3582 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3583 * the range between oldsize and size
3585 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3587 struct btrfs_trans_handle
*trans
;
3588 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3589 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3590 struct extent_map
*em
= NULL
;
3591 struct extent_state
*cached_state
= NULL
;
3592 u64 mask
= root
->sectorsize
- 1;
3593 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3594 u64 block_end
= (size
+ mask
) & ~mask
;
3600 if (size
<= hole_start
)
3604 struct btrfs_ordered_extent
*ordered
;
3605 btrfs_wait_ordered_range(inode
, hole_start
,
3606 block_end
- hole_start
);
3607 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3608 &cached_state
, GFP_NOFS
);
3609 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3612 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3613 &cached_state
, GFP_NOFS
);
3614 btrfs_put_ordered_extent(ordered
);
3617 cur_offset
= hole_start
;
3619 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3620 block_end
- cur_offset
, 0);
3621 BUG_ON(IS_ERR(em
) || !em
);
3622 last_byte
= min(extent_map_end(em
), block_end
);
3623 last_byte
= (last_byte
+ mask
) & ~mask
;
3624 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3626 hole_size
= last_byte
- cur_offset
;
3628 trans
= btrfs_start_transaction(root
, 2);
3629 if (IS_ERR(trans
)) {
3630 err
= PTR_ERR(trans
);
3633 btrfs_set_trans_block_group(trans
, inode
);
3635 err
= btrfs_drop_extents(trans
, inode
, cur_offset
,
3636 cur_offset
+ hole_size
,
3641 err
= btrfs_insert_file_extent(trans
, root
,
3642 inode
->i_ino
, cur_offset
, 0,
3643 0, hole_size
, 0, hole_size
,
3648 btrfs_drop_extent_cache(inode
, hole_start
,
3651 btrfs_end_transaction(trans
, root
);
3653 free_extent_map(em
);
3655 cur_offset
= last_byte
;
3656 if (cur_offset
>= block_end
)
3660 free_extent_map(em
);
3661 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3666 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3668 loff_t oldsize
= i_size_read(inode
);
3671 if (newsize
== oldsize
)
3674 if (newsize
> oldsize
) {
3675 i_size_write(inode
, newsize
);
3676 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3677 truncate_pagecache(inode
, oldsize
, newsize
);
3678 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3680 btrfs_setsize(inode
, oldsize
);
3684 mark_inode_dirty(inode
);
3688 * We're truncating a file that used to have good data down to
3689 * zero. Make sure it gets into the ordered flush list so that
3690 * any new writes get down to disk quickly.
3693 BTRFS_I(inode
)->ordered_data_close
= 1;
3695 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3696 truncate_setsize(inode
, newsize
);
3697 ret
= btrfs_truncate(inode
);
3703 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3705 struct inode
*inode
= dentry
->d_inode
;
3706 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3709 if (btrfs_root_readonly(root
))
3712 err
= inode_change_ok(inode
, attr
);
3716 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3717 err
= btrfs_setsize(inode
, attr
->ia_size
);
3722 if (attr
->ia_valid
) {
3723 setattr_copy(inode
, attr
);
3724 mark_inode_dirty(inode
);
3726 if (attr
->ia_valid
& ATTR_MODE
)
3727 err
= btrfs_acl_chmod(inode
);
3733 void btrfs_evict_inode(struct inode
*inode
)
3735 struct btrfs_trans_handle
*trans
;
3736 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3740 trace_btrfs_inode_evict(inode
);
3742 truncate_inode_pages(&inode
->i_data
, 0);
3743 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3744 root
== root
->fs_info
->tree_root
))
3747 if (is_bad_inode(inode
)) {
3748 btrfs_orphan_del(NULL
, inode
);
3751 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3752 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3754 if (root
->fs_info
->log_root_recovering
) {
3755 BUG_ON(!list_empty(&BTRFS_I(inode
)->i_orphan
));
3759 if (inode
->i_nlink
> 0) {
3760 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3764 btrfs_i_size_write(inode
, 0);
3767 trans
= btrfs_start_transaction(root
, 0);
3768 BUG_ON(IS_ERR(trans
));
3769 btrfs_set_trans_block_group(trans
, inode
);
3770 trans
->block_rsv
= root
->orphan_block_rsv
;
3772 ret
= btrfs_block_rsv_check(trans
, root
,
3773 root
->orphan_block_rsv
, 0, 5);
3775 BUG_ON(ret
!= -EAGAIN
);
3776 ret
= btrfs_commit_transaction(trans
, root
);
3781 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3785 nr
= trans
->blocks_used
;
3786 btrfs_end_transaction(trans
, root
);
3788 btrfs_btree_balance_dirty(root
, nr
);
3793 ret
= btrfs_orphan_del(trans
, inode
);
3797 nr
= trans
->blocks_used
;
3798 btrfs_end_transaction(trans
, root
);
3799 btrfs_btree_balance_dirty(root
, nr
);
3801 end_writeback(inode
);
3806 * this returns the key found in the dir entry in the location pointer.
3807 * If no dir entries were found, location->objectid is 0.
3809 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3810 struct btrfs_key
*location
)
3812 const char *name
= dentry
->d_name
.name
;
3813 int namelen
= dentry
->d_name
.len
;
3814 struct btrfs_dir_item
*di
;
3815 struct btrfs_path
*path
;
3816 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3819 path
= btrfs_alloc_path();
3822 di
= btrfs_lookup_dir_item(NULL
, root
, path
, dir
->i_ino
, name
,
3827 if (!di
|| IS_ERR(di
))
3830 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3832 btrfs_free_path(path
);
3835 location
->objectid
= 0;
3840 * when we hit a tree root in a directory, the btrfs part of the inode
3841 * needs to be changed to reflect the root directory of the tree root. This
3842 * is kind of like crossing a mount point.
3844 static int fixup_tree_root_location(struct btrfs_root
*root
,
3846 struct dentry
*dentry
,
3847 struct btrfs_key
*location
,
3848 struct btrfs_root
**sub_root
)
3850 struct btrfs_path
*path
;
3851 struct btrfs_root
*new_root
;
3852 struct btrfs_root_ref
*ref
;
3853 struct extent_buffer
*leaf
;
3857 path
= btrfs_alloc_path();
3864 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3865 BTRFS_I(dir
)->root
->root_key
.objectid
,
3866 location
->objectid
);
3873 leaf
= path
->nodes
[0];
3874 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3875 if (btrfs_root_ref_dirid(leaf
, ref
) != dir
->i_ino
||
3876 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3879 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3880 (unsigned long)(ref
+ 1),
3881 dentry
->d_name
.len
);
3885 btrfs_release_path(root
->fs_info
->tree_root
, path
);
3887 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3888 if (IS_ERR(new_root
)) {
3889 err
= PTR_ERR(new_root
);
3893 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3898 *sub_root
= new_root
;
3899 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3900 location
->type
= BTRFS_INODE_ITEM_KEY
;
3901 location
->offset
= 0;
3904 btrfs_free_path(path
);
3908 static void inode_tree_add(struct inode
*inode
)
3910 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3911 struct btrfs_inode
*entry
;
3913 struct rb_node
*parent
;
3915 p
= &root
->inode_tree
.rb_node
;
3918 if (inode_unhashed(inode
))
3921 spin_lock(&root
->inode_lock
);
3924 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
3926 if (inode
->i_ino
< entry
->vfs_inode
.i_ino
)
3927 p
= &parent
->rb_left
;
3928 else if (inode
->i_ino
> entry
->vfs_inode
.i_ino
)
3929 p
= &parent
->rb_right
;
3931 WARN_ON(!(entry
->vfs_inode
.i_state
&
3932 (I_WILL_FREE
| I_FREEING
)));
3933 rb_erase(parent
, &root
->inode_tree
);
3934 RB_CLEAR_NODE(parent
);
3935 spin_unlock(&root
->inode_lock
);
3939 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
3940 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3941 spin_unlock(&root
->inode_lock
);
3944 static void inode_tree_del(struct inode
*inode
)
3946 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3949 spin_lock(&root
->inode_lock
);
3950 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
3951 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
3952 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
3953 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3955 spin_unlock(&root
->inode_lock
);
3958 * Free space cache has inodes in the tree root, but the tree root has a
3959 * root_refs of 0, so this could end up dropping the tree root as a
3960 * snapshot, so we need the extra !root->fs_info->tree_root check to
3961 * make sure we don't drop it.
3963 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
3964 root
!= root
->fs_info
->tree_root
) {
3965 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
3966 spin_lock(&root
->inode_lock
);
3967 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
3968 spin_unlock(&root
->inode_lock
);
3970 btrfs_add_dead_root(root
);
3974 int btrfs_invalidate_inodes(struct btrfs_root
*root
)
3976 struct rb_node
*node
;
3977 struct rb_node
*prev
;
3978 struct btrfs_inode
*entry
;
3979 struct inode
*inode
;
3982 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
3984 spin_lock(&root
->inode_lock
);
3986 node
= root
->inode_tree
.rb_node
;
3990 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
3992 if (objectid
< entry
->vfs_inode
.i_ino
)
3993 node
= node
->rb_left
;
3994 else if (objectid
> entry
->vfs_inode
.i_ino
)
3995 node
= node
->rb_right
;
4001 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4002 if (objectid
<= entry
->vfs_inode
.i_ino
) {
4006 prev
= rb_next(prev
);
4010 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4011 objectid
= entry
->vfs_inode
.i_ino
+ 1;
4012 inode
= igrab(&entry
->vfs_inode
);
4014 spin_unlock(&root
->inode_lock
);
4015 if (atomic_read(&inode
->i_count
) > 1)
4016 d_prune_aliases(inode
);
4018 * btrfs_drop_inode will have it removed from
4019 * the inode cache when its usage count
4024 spin_lock(&root
->inode_lock
);
4028 if (cond_resched_lock(&root
->inode_lock
))
4031 node
= rb_next(node
);
4033 spin_unlock(&root
->inode_lock
);
4037 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4039 struct btrfs_iget_args
*args
= p
;
4040 inode
->i_ino
= args
->ino
;
4041 BTRFS_I(inode
)->root
= args
->root
;
4042 btrfs_set_inode_space_info(args
->root
, inode
);
4046 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4048 struct btrfs_iget_args
*args
= opaque
;
4049 return args
->ino
== inode
->i_ino
&&
4050 args
->root
== BTRFS_I(inode
)->root
;
4053 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4055 struct btrfs_root
*root
)
4057 struct inode
*inode
;
4058 struct btrfs_iget_args args
;
4059 args
.ino
= objectid
;
4062 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4063 btrfs_init_locked_inode
,
4068 /* Get an inode object given its location and corresponding root.
4069 * Returns in *is_new if the inode was read from disk
4071 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4072 struct btrfs_root
*root
, int *new)
4074 struct inode
*inode
;
4076 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4078 return ERR_PTR(-ENOMEM
);
4080 if (inode
->i_state
& I_NEW
) {
4081 BTRFS_I(inode
)->root
= root
;
4082 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4083 btrfs_read_locked_inode(inode
);
4084 inode_tree_add(inode
);
4085 unlock_new_inode(inode
);
4093 static struct inode
*new_simple_dir(struct super_block
*s
,
4094 struct btrfs_key
*key
,
4095 struct btrfs_root
*root
)
4097 struct inode
*inode
= new_inode(s
);
4100 return ERR_PTR(-ENOMEM
);
4102 BTRFS_I(inode
)->root
= root
;
4103 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4104 BTRFS_I(inode
)->dummy_inode
= 1;
4106 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4107 inode
->i_op
= &simple_dir_inode_operations
;
4108 inode
->i_fop
= &simple_dir_operations
;
4109 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4110 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4115 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4117 struct inode
*inode
;
4118 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4119 struct btrfs_root
*sub_root
= root
;
4120 struct btrfs_key location
;
4124 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4125 return ERR_PTR(-ENAMETOOLONG
);
4127 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4130 return ERR_PTR(ret
);
4132 if (location
.objectid
== 0)
4135 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4136 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4140 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4142 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4143 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4144 &location
, &sub_root
);
4147 inode
= ERR_PTR(ret
);
4149 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4151 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4153 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4155 if (!IS_ERR(inode
) && root
!= sub_root
) {
4156 down_read(&root
->fs_info
->cleanup_work_sem
);
4157 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4158 ret
= btrfs_orphan_cleanup(sub_root
);
4159 up_read(&root
->fs_info
->cleanup_work_sem
);
4161 inode
= ERR_PTR(ret
);
4167 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4169 struct btrfs_root
*root
;
4171 if (!dentry
->d_inode
&& !IS_ROOT(dentry
))
4172 dentry
= dentry
->d_parent
;
4174 if (dentry
->d_inode
) {
4175 root
= BTRFS_I(dentry
->d_inode
)->root
;
4176 if (btrfs_root_refs(&root
->root_item
) == 0)
4182 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4183 struct nameidata
*nd
)
4185 struct inode
*inode
;
4187 inode
= btrfs_lookup_dentry(dir
, dentry
);
4189 return ERR_CAST(inode
);
4191 return d_splice_alias(inode
, dentry
);
4194 static unsigned char btrfs_filetype_table
[] = {
4195 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4198 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4201 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4202 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4203 struct btrfs_item
*item
;
4204 struct btrfs_dir_item
*di
;
4205 struct btrfs_key key
;
4206 struct btrfs_key found_key
;
4207 struct btrfs_path
*path
;
4210 struct extent_buffer
*leaf
;
4213 unsigned char d_type
;
4218 int key_type
= BTRFS_DIR_INDEX_KEY
;
4223 /* FIXME, use a real flag for deciding about the key type */
4224 if (root
->fs_info
->tree_root
== root
)
4225 key_type
= BTRFS_DIR_ITEM_KEY
;
4227 /* special case for "." */
4228 if (filp
->f_pos
== 0) {
4229 over
= filldir(dirent
, ".", 1,
4236 /* special case for .., just use the back ref */
4237 if (filp
->f_pos
== 1) {
4238 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4239 over
= filldir(dirent
, "..", 2,
4245 path
= btrfs_alloc_path();
4248 btrfs_set_key_type(&key
, key_type
);
4249 key
.offset
= filp
->f_pos
;
4250 key
.objectid
= inode
->i_ino
;
4252 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4258 leaf
= path
->nodes
[0];
4259 nritems
= btrfs_header_nritems(leaf
);
4260 slot
= path
->slots
[0];
4261 if (advance
|| slot
>= nritems
) {
4262 if (slot
>= nritems
- 1) {
4263 ret
= btrfs_next_leaf(root
, path
);
4266 leaf
= path
->nodes
[0];
4267 nritems
= btrfs_header_nritems(leaf
);
4268 slot
= path
->slots
[0];
4276 item
= btrfs_item_nr(leaf
, slot
);
4277 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4279 if (found_key
.objectid
!= key
.objectid
)
4281 if (btrfs_key_type(&found_key
) != key_type
)
4283 if (found_key
.offset
< filp
->f_pos
)
4286 filp
->f_pos
= found_key
.offset
;
4288 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4290 di_total
= btrfs_item_size(leaf
, item
);
4292 while (di_cur
< di_total
) {
4293 struct btrfs_key location
;
4295 if (verify_dir_item(root
, leaf
, di
))
4298 name_len
= btrfs_dir_name_len(leaf
, di
);
4299 if (name_len
<= sizeof(tmp_name
)) {
4300 name_ptr
= tmp_name
;
4302 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4308 read_extent_buffer(leaf
, name_ptr
,
4309 (unsigned long)(di
+ 1), name_len
);
4311 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4312 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4314 /* is this a reference to our own snapshot? If so
4317 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4318 location
.objectid
== root
->root_key
.objectid
) {
4322 over
= filldir(dirent
, name_ptr
, name_len
,
4323 found_key
.offset
, location
.objectid
,
4327 if (name_ptr
!= tmp_name
)
4332 di_len
= btrfs_dir_name_len(leaf
, di
) +
4333 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4335 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4339 /* Reached end of directory/root. Bump pos past the last item. */
4340 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4342 * 32-bit glibc will use getdents64, but then strtol -
4343 * so the last number we can serve is this.
4345 filp
->f_pos
= 0x7fffffff;
4351 btrfs_free_path(path
);
4355 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4357 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4358 struct btrfs_trans_handle
*trans
;
4360 bool nolock
= false;
4362 if (BTRFS_I(inode
)->dummy_inode
)
4366 nolock
= (root
->fs_info
->closing
&& root
== root
->fs_info
->tree_root
);
4368 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4370 trans
= btrfs_join_transaction_nolock(root
, 1);
4372 trans
= btrfs_join_transaction(root
, 1);
4374 return PTR_ERR(trans
);
4375 btrfs_set_trans_block_group(trans
, inode
);
4377 ret
= btrfs_end_transaction_nolock(trans
, root
);
4379 ret
= btrfs_commit_transaction(trans
, root
);
4385 * This is somewhat expensive, updating the tree every time the
4386 * inode changes. But, it is most likely to find the inode in cache.
4387 * FIXME, needs more benchmarking...there are no reasons other than performance
4388 * to keep or drop this code.
4390 void btrfs_dirty_inode(struct inode
*inode
)
4392 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4393 struct btrfs_trans_handle
*trans
;
4396 if (BTRFS_I(inode
)->dummy_inode
)
4399 trans
= btrfs_join_transaction(root
, 1);
4400 BUG_ON(IS_ERR(trans
));
4401 btrfs_set_trans_block_group(trans
, inode
);
4403 ret
= btrfs_update_inode(trans
, root
, inode
);
4404 if (ret
&& ret
== -ENOSPC
) {
4405 /* whoops, lets try again with the full transaction */
4406 btrfs_end_transaction(trans
, root
);
4407 trans
= btrfs_start_transaction(root
, 1);
4408 if (IS_ERR(trans
)) {
4409 if (printk_ratelimit()) {
4410 printk(KERN_ERR
"btrfs: fail to "
4411 "dirty inode %lu error %ld\n",
4412 inode
->i_ino
, PTR_ERR(trans
));
4416 btrfs_set_trans_block_group(trans
, inode
);
4418 ret
= btrfs_update_inode(trans
, root
, inode
);
4420 if (printk_ratelimit()) {
4421 printk(KERN_ERR
"btrfs: fail to "
4422 "dirty inode %lu error %d\n",
4427 btrfs_end_transaction(trans
, root
);
4431 * find the highest existing sequence number in a directory
4432 * and then set the in-memory index_cnt variable to reflect
4433 * free sequence numbers
4435 static int btrfs_set_inode_index_count(struct inode
*inode
)
4437 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4438 struct btrfs_key key
, found_key
;
4439 struct btrfs_path
*path
;
4440 struct extent_buffer
*leaf
;
4443 key
.objectid
= inode
->i_ino
;
4444 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4445 key
.offset
= (u64
)-1;
4447 path
= btrfs_alloc_path();
4451 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4454 /* FIXME: we should be able to handle this */
4460 * MAGIC NUMBER EXPLANATION:
4461 * since we search a directory based on f_pos we have to start at 2
4462 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4463 * else has to start at 2
4465 if (path
->slots
[0] == 0) {
4466 BTRFS_I(inode
)->index_cnt
= 2;
4472 leaf
= path
->nodes
[0];
4473 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4475 if (found_key
.objectid
!= inode
->i_ino
||
4476 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4477 BTRFS_I(inode
)->index_cnt
= 2;
4481 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4483 btrfs_free_path(path
);
4488 * helper to find a free sequence number in a given directory. This current
4489 * code is very simple, later versions will do smarter things in the btree
4491 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4495 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4496 ret
= btrfs_set_inode_index_count(dir
);
4501 *index
= BTRFS_I(dir
)->index_cnt
;
4502 BTRFS_I(dir
)->index_cnt
++;
4507 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4508 struct btrfs_root
*root
,
4510 const char *name
, int name_len
,
4511 u64 ref_objectid
, u64 objectid
,
4512 u64 alloc_hint
, int mode
, u64
*index
)
4514 struct inode
*inode
;
4515 struct btrfs_inode_item
*inode_item
;
4516 struct btrfs_key
*location
;
4517 struct btrfs_path
*path
;
4518 struct btrfs_inode_ref
*ref
;
4519 struct btrfs_key key
[2];
4525 path
= btrfs_alloc_path();
4528 inode
= new_inode(root
->fs_info
->sb
);
4530 btrfs_free_path(path
);
4531 return ERR_PTR(-ENOMEM
);
4535 trace_btrfs_inode_request(dir
);
4537 ret
= btrfs_set_inode_index(dir
, index
);
4539 btrfs_free_path(path
);
4541 return ERR_PTR(ret
);
4545 * index_cnt is ignored for everything but a dir,
4546 * btrfs_get_inode_index_count has an explanation for the magic
4549 BTRFS_I(inode
)->index_cnt
= 2;
4550 BTRFS_I(inode
)->root
= root
;
4551 BTRFS_I(inode
)->generation
= trans
->transid
;
4552 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4553 btrfs_set_inode_space_info(root
, inode
);
4559 BTRFS_I(inode
)->block_group
=
4560 btrfs_find_block_group(root
, 0, alloc_hint
, owner
);
4562 key
[0].objectid
= objectid
;
4563 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4566 key
[1].objectid
= objectid
;
4567 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4568 key
[1].offset
= ref_objectid
;
4570 sizes
[0] = sizeof(struct btrfs_inode_item
);
4571 sizes
[1] = name_len
+ sizeof(*ref
);
4573 path
->leave_spinning
= 1;
4574 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4578 inode_init_owner(inode
, dir
, mode
);
4579 inode
->i_ino
= objectid
;
4580 inode_set_bytes(inode
, 0);
4581 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4582 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4583 struct btrfs_inode_item
);
4584 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4586 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4587 struct btrfs_inode_ref
);
4588 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4589 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4590 ptr
= (unsigned long)(ref
+ 1);
4591 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4593 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4594 btrfs_free_path(path
);
4596 location
= &BTRFS_I(inode
)->location
;
4597 location
->objectid
= objectid
;
4598 location
->offset
= 0;
4599 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4601 btrfs_inherit_iflags(inode
, dir
);
4603 if ((mode
& S_IFREG
)) {
4604 if (btrfs_test_opt(root
, NODATASUM
))
4605 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4606 if (btrfs_test_opt(root
, NODATACOW
) ||
4607 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4608 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4611 insert_inode_hash(inode
);
4612 inode_tree_add(inode
);
4614 trace_btrfs_inode_new(inode
);
4619 BTRFS_I(dir
)->index_cnt
--;
4620 btrfs_free_path(path
);
4622 return ERR_PTR(ret
);
4625 static inline u8
btrfs_inode_type(struct inode
*inode
)
4627 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4631 * utility function to add 'inode' into 'parent_inode' with
4632 * a give name and a given sequence number.
4633 * if 'add_backref' is true, also insert a backref from the
4634 * inode to the parent directory.
4636 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4637 struct inode
*parent_inode
, struct inode
*inode
,
4638 const char *name
, int name_len
, int add_backref
, u64 index
)
4641 struct btrfs_key key
;
4642 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4644 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4645 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4647 key
.objectid
= inode
->i_ino
;
4648 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4652 if (unlikely(inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4653 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4654 key
.objectid
, root
->root_key
.objectid
,
4655 parent_inode
->i_ino
,
4656 index
, name
, name_len
);
4657 } else if (add_backref
) {
4658 ret
= btrfs_insert_inode_ref(trans
, root
,
4659 name
, name_len
, inode
->i_ino
,
4660 parent_inode
->i_ino
, index
);
4664 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4665 parent_inode
->i_ino
, &key
,
4666 btrfs_inode_type(inode
), index
);
4669 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4671 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4672 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4677 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4678 struct inode
*dir
, struct dentry
*dentry
,
4679 struct inode
*inode
, int backref
, u64 index
)
4681 int err
= btrfs_add_link(trans
, dir
, inode
,
4682 dentry
->d_name
.name
, dentry
->d_name
.len
,
4685 d_instantiate(dentry
, inode
);
4693 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4694 int mode
, dev_t rdev
)
4696 struct btrfs_trans_handle
*trans
;
4697 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4698 struct inode
*inode
= NULL
;
4702 unsigned long nr
= 0;
4705 if (!new_valid_dev(rdev
))
4708 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4713 * 2 for inode item and ref
4715 * 1 for xattr if selinux is on
4717 trans
= btrfs_start_transaction(root
, 5);
4719 return PTR_ERR(trans
);
4721 btrfs_set_trans_block_group(trans
, dir
);
4723 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4724 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4725 BTRFS_I(dir
)->block_group
, mode
, &index
);
4726 err
= PTR_ERR(inode
);
4730 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4736 btrfs_set_trans_block_group(trans
, inode
);
4737 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4741 inode
->i_op
= &btrfs_special_inode_operations
;
4742 init_special_inode(inode
, inode
->i_mode
, rdev
);
4743 btrfs_update_inode(trans
, root
, inode
);
4745 btrfs_update_inode_block_group(trans
, inode
);
4746 btrfs_update_inode_block_group(trans
, dir
);
4748 nr
= trans
->blocks_used
;
4749 btrfs_end_transaction_throttle(trans
, root
);
4750 btrfs_btree_balance_dirty(root
, nr
);
4752 inode_dec_link_count(inode
);
4758 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4759 int mode
, struct nameidata
*nd
)
4761 struct btrfs_trans_handle
*trans
;
4762 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4763 struct inode
*inode
= NULL
;
4766 unsigned long nr
= 0;
4770 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4774 * 2 for inode item and ref
4776 * 1 for xattr if selinux is on
4778 trans
= btrfs_start_transaction(root
, 5);
4780 return PTR_ERR(trans
);
4782 btrfs_set_trans_block_group(trans
, dir
);
4784 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4785 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4786 BTRFS_I(dir
)->block_group
, mode
, &index
);
4787 err
= PTR_ERR(inode
);
4791 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4797 btrfs_set_trans_block_group(trans
, inode
);
4798 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4802 inode
->i_mapping
->a_ops
= &btrfs_aops
;
4803 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
4804 inode
->i_fop
= &btrfs_file_operations
;
4805 inode
->i_op
= &btrfs_file_inode_operations
;
4806 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
4808 btrfs_update_inode_block_group(trans
, inode
);
4809 btrfs_update_inode_block_group(trans
, dir
);
4811 nr
= trans
->blocks_used
;
4812 btrfs_end_transaction_throttle(trans
, root
);
4814 inode_dec_link_count(inode
);
4817 btrfs_btree_balance_dirty(root
, nr
);
4821 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
4822 struct dentry
*dentry
)
4824 struct btrfs_trans_handle
*trans
;
4825 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4826 struct inode
*inode
= old_dentry
->d_inode
;
4828 unsigned long nr
= 0;
4832 if (inode
->i_nlink
== 0)
4835 /* do not allow sys_link's with other subvols of the same device */
4836 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
4839 if (inode
->i_nlink
== ~0U)
4842 btrfs_inc_nlink(inode
);
4843 inode
->i_ctime
= CURRENT_TIME
;
4845 err
= btrfs_set_inode_index(dir
, &index
);
4850 * 2 items for inode and inode ref
4851 * 2 items for dir items
4852 * 1 item for parent inode
4854 trans
= btrfs_start_transaction(root
, 5);
4855 if (IS_ERR(trans
)) {
4856 err
= PTR_ERR(trans
);
4860 btrfs_set_trans_block_group(trans
, dir
);
4863 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
4868 struct dentry
*parent
= dget_parent(dentry
);
4869 btrfs_update_inode_block_group(trans
, dir
);
4870 err
= btrfs_update_inode(trans
, root
, inode
);
4872 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
4876 nr
= trans
->blocks_used
;
4877 btrfs_end_transaction_throttle(trans
, root
);
4880 inode_dec_link_count(inode
);
4883 btrfs_btree_balance_dirty(root
, nr
);
4887 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
4889 struct inode
*inode
= NULL
;
4890 struct btrfs_trans_handle
*trans
;
4891 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4893 int drop_on_err
= 0;
4896 unsigned long nr
= 1;
4898 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
4903 * 2 items for inode and ref
4904 * 2 items for dir items
4905 * 1 for xattr if selinux is on
4907 trans
= btrfs_start_transaction(root
, 5);
4909 return PTR_ERR(trans
);
4910 btrfs_set_trans_block_group(trans
, dir
);
4912 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4913 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
4914 BTRFS_I(dir
)->block_group
, S_IFDIR
| mode
,
4916 if (IS_ERR(inode
)) {
4917 err
= PTR_ERR(inode
);
4923 err
= btrfs_init_inode_security(trans
, inode
, dir
);
4927 inode
->i_op
= &btrfs_dir_inode_operations
;
4928 inode
->i_fop
= &btrfs_dir_file_operations
;
4929 btrfs_set_trans_block_group(trans
, inode
);
4931 btrfs_i_size_write(inode
, 0);
4932 err
= btrfs_update_inode(trans
, root
, inode
);
4936 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
4937 dentry
->d_name
.len
, 0, index
);
4941 d_instantiate(dentry
, inode
);
4943 btrfs_update_inode_block_group(trans
, inode
);
4944 btrfs_update_inode_block_group(trans
, dir
);
4947 nr
= trans
->blocks_used
;
4948 btrfs_end_transaction_throttle(trans
, root
);
4951 btrfs_btree_balance_dirty(root
, nr
);
4955 /* helper for btfs_get_extent. Given an existing extent in the tree,
4956 * and an extent that you want to insert, deal with overlap and insert
4957 * the new extent into the tree.
4959 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
4960 struct extent_map
*existing
,
4961 struct extent_map
*em
,
4962 u64 map_start
, u64 map_len
)
4966 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
4967 start_diff
= map_start
- em
->start
;
4968 em
->start
= map_start
;
4970 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
4971 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
4972 em
->block_start
+= start_diff
;
4973 em
->block_len
-= start_diff
;
4975 return add_extent_mapping(em_tree
, em
);
4978 static noinline
int uncompress_inline(struct btrfs_path
*path
,
4979 struct inode
*inode
, struct page
*page
,
4980 size_t pg_offset
, u64 extent_offset
,
4981 struct btrfs_file_extent_item
*item
)
4984 struct extent_buffer
*leaf
= path
->nodes
[0];
4987 unsigned long inline_size
;
4991 WARN_ON(pg_offset
!= 0);
4992 compress_type
= btrfs_file_extent_compression(leaf
, item
);
4993 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
4994 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
4995 btrfs_item_nr(leaf
, path
->slots
[0]));
4996 tmp
= kmalloc(inline_size
, GFP_NOFS
);
4997 ptr
= btrfs_file_extent_inline_start(item
);
4999 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
5001 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
5002 ret
= btrfs_decompress(compress_type
, tmp
, page
,
5003 extent_offset
, inline_size
, max_size
);
5005 char *kaddr
= kmap_atomic(page
, KM_USER0
);
5006 unsigned long copy_size
= min_t(u64
,
5007 PAGE_CACHE_SIZE
- pg_offset
,
5008 max_size
- extent_offset
);
5009 memset(kaddr
+ pg_offset
, 0, copy_size
);
5010 kunmap_atomic(kaddr
, KM_USER0
);
5017 * a bit scary, this does extent mapping from logical file offset to the disk.
5018 * the ugly parts come from merging extents from the disk with the in-ram
5019 * representation. This gets more complex because of the data=ordered code,
5020 * where the in-ram extents might be locked pending data=ordered completion.
5022 * This also copies inline extents directly into the page.
5025 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
5026 size_t pg_offset
, u64 start
, u64 len
,
5032 u64 extent_start
= 0;
5034 u64 objectid
= inode
->i_ino
;
5036 struct btrfs_path
*path
= NULL
;
5037 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5038 struct btrfs_file_extent_item
*item
;
5039 struct extent_buffer
*leaf
;
5040 struct btrfs_key found_key
;
5041 struct extent_map
*em
= NULL
;
5042 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5043 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5044 struct btrfs_trans_handle
*trans
= NULL
;
5048 read_lock(&em_tree
->lock
);
5049 em
= lookup_extent_mapping(em_tree
, start
, len
);
5051 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5052 read_unlock(&em_tree
->lock
);
5055 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5056 free_extent_map(em
);
5057 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5058 free_extent_map(em
);
5062 em
= alloc_extent_map(GFP_NOFS
);
5067 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5068 em
->start
= EXTENT_MAP_HOLE
;
5069 em
->orig_start
= EXTENT_MAP_HOLE
;
5071 em
->block_len
= (u64
)-1;
5074 path
= btrfs_alloc_path();
5078 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5079 objectid
, start
, trans
!= NULL
);
5086 if (path
->slots
[0] == 0)
5091 leaf
= path
->nodes
[0];
5092 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5093 struct btrfs_file_extent_item
);
5094 /* are we inside the extent that was found? */
5095 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5096 found_type
= btrfs_key_type(&found_key
);
5097 if (found_key
.objectid
!= objectid
||
5098 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5102 found_type
= btrfs_file_extent_type(leaf
, item
);
5103 extent_start
= found_key
.offset
;
5104 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5105 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5106 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5107 extent_end
= extent_start
+
5108 btrfs_file_extent_num_bytes(leaf
, item
);
5109 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5111 size
= btrfs_file_extent_inline_len(leaf
, item
);
5112 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5113 ~((u64
)root
->sectorsize
- 1);
5116 if (start
>= extent_end
) {
5118 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5119 ret
= btrfs_next_leaf(root
, path
);
5126 leaf
= path
->nodes
[0];
5128 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5129 if (found_key
.objectid
!= objectid
||
5130 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5132 if (start
+ len
<= found_key
.offset
)
5135 em
->len
= found_key
.offset
- start
;
5139 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5140 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5141 em
->start
= extent_start
;
5142 em
->len
= extent_end
- extent_start
;
5143 em
->orig_start
= extent_start
-
5144 btrfs_file_extent_offset(leaf
, item
);
5145 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5147 em
->block_start
= EXTENT_MAP_HOLE
;
5150 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5151 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5152 em
->compress_type
= compress_type
;
5153 em
->block_start
= bytenr
;
5154 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5157 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5158 em
->block_start
= bytenr
;
5159 em
->block_len
= em
->len
;
5160 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5161 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5164 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5168 size_t extent_offset
;
5171 em
->block_start
= EXTENT_MAP_INLINE
;
5172 if (!page
|| create
) {
5173 em
->start
= extent_start
;
5174 em
->len
= extent_end
- extent_start
;
5178 size
= btrfs_file_extent_inline_len(leaf
, item
);
5179 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5180 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5181 size
- extent_offset
);
5182 em
->start
= extent_start
+ extent_offset
;
5183 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5184 ~((u64
)root
->sectorsize
- 1);
5185 em
->orig_start
= EXTENT_MAP_INLINE
;
5186 if (compress_type
) {
5187 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5188 em
->compress_type
= compress_type
;
5190 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5191 if (create
== 0 && !PageUptodate(page
)) {
5192 if (btrfs_file_extent_compression(leaf
, item
) !=
5193 BTRFS_COMPRESS_NONE
) {
5194 ret
= uncompress_inline(path
, inode
, page
,
5196 extent_offset
, item
);
5200 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5202 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5203 memset(map
+ pg_offset
+ copy_size
, 0,
5204 PAGE_CACHE_SIZE
- pg_offset
-
5209 flush_dcache_page(page
);
5210 } else if (create
&& PageUptodate(page
)) {
5214 free_extent_map(em
);
5216 btrfs_release_path(root
, path
);
5217 trans
= btrfs_join_transaction(root
, 1);
5219 return ERR_CAST(trans
);
5223 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5226 btrfs_mark_buffer_dirty(leaf
);
5228 set_extent_uptodate(io_tree
, em
->start
,
5229 extent_map_end(em
) - 1, GFP_NOFS
);
5232 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5239 em
->block_start
= EXTENT_MAP_HOLE
;
5240 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5242 btrfs_release_path(root
, path
);
5243 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5244 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5245 "[%llu %llu]\n", (unsigned long long)em
->start
,
5246 (unsigned long long)em
->len
,
5247 (unsigned long long)start
,
5248 (unsigned long long)len
);
5254 write_lock(&em_tree
->lock
);
5255 ret
= add_extent_mapping(em_tree
, em
);
5256 /* it is possible that someone inserted the extent into the tree
5257 * while we had the lock dropped. It is also possible that
5258 * an overlapping map exists in the tree
5260 if (ret
== -EEXIST
) {
5261 struct extent_map
*existing
;
5265 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5266 if (existing
&& (existing
->start
> start
||
5267 existing
->start
+ existing
->len
<= start
)) {
5268 free_extent_map(existing
);
5272 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5275 err
= merge_extent_mapping(em_tree
, existing
,
5278 free_extent_map(existing
);
5280 free_extent_map(em
);
5285 free_extent_map(em
);
5289 free_extent_map(em
);
5294 write_unlock(&em_tree
->lock
);
5297 trace_btrfs_get_extent(root
, em
);
5300 btrfs_free_path(path
);
5302 ret
= btrfs_end_transaction(trans
, root
);
5307 free_extent_map(em
);
5308 return ERR_PTR(err
);
5313 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5314 size_t pg_offset
, u64 start
, u64 len
,
5317 struct extent_map
*em
;
5318 struct extent_map
*hole_em
= NULL
;
5319 u64 range_start
= start
;
5325 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5330 * if our em maps to a hole, there might
5331 * actually be delalloc bytes behind it
5333 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5339 /* check to see if we've wrapped (len == -1 or similar) */
5348 /* ok, we didn't find anything, lets look for delalloc */
5349 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5350 end
, len
, EXTENT_DELALLOC
, 1);
5351 found_end
= range_start
+ found
;
5352 if (found_end
< range_start
)
5353 found_end
= (u64
)-1;
5356 * we didn't find anything useful, return
5357 * the original results from get_extent()
5359 if (range_start
> end
|| found_end
<= start
) {
5365 /* adjust the range_start to make sure it doesn't
5366 * go backwards from the start they passed in
5368 range_start
= max(start
,range_start
);
5369 found
= found_end
- range_start
;
5372 u64 hole_start
= start
;
5375 em
= alloc_extent_map(GFP_NOFS
);
5381 * when btrfs_get_extent can't find anything it
5382 * returns one huge hole
5384 * make sure what it found really fits our range, and
5385 * adjust to make sure it is based on the start from
5389 u64 calc_end
= extent_map_end(hole_em
);
5391 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5392 free_extent_map(hole_em
);
5395 hole_start
= max(hole_em
->start
, start
);
5396 hole_len
= calc_end
- hole_start
;
5400 if (hole_em
&& range_start
> hole_start
) {
5401 /* our hole starts before our delalloc, so we
5402 * have to return just the parts of the hole
5403 * that go until the delalloc starts
5405 em
->len
= min(hole_len
,
5406 range_start
- hole_start
);
5407 em
->start
= hole_start
;
5408 em
->orig_start
= hole_start
;
5410 * don't adjust block start at all,
5411 * it is fixed at EXTENT_MAP_HOLE
5413 em
->block_start
= hole_em
->block_start
;
5414 em
->block_len
= hole_len
;
5416 em
->start
= range_start
;
5418 em
->orig_start
= range_start
;
5419 em
->block_start
= EXTENT_MAP_DELALLOC
;
5420 em
->block_len
= found
;
5422 } else if (hole_em
) {
5427 free_extent_map(hole_em
);
5429 free_extent_map(em
);
5430 return ERR_PTR(err
);
5435 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5438 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5439 struct btrfs_trans_handle
*trans
;
5440 struct extent_map
*em
;
5441 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5442 struct btrfs_key ins
;
5446 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5448 trans
= btrfs_join_transaction(root
, 0);
5450 return ERR_CAST(trans
);
5452 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5454 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5455 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5456 alloc_hint
, (u64
)-1, &ins
, 1);
5462 em
= alloc_extent_map(GFP_NOFS
);
5464 em
= ERR_PTR(-ENOMEM
);
5469 em
->orig_start
= em
->start
;
5470 em
->len
= ins
.offset
;
5472 em
->block_start
= ins
.objectid
;
5473 em
->block_len
= ins
.offset
;
5474 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5475 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5478 write_lock(&em_tree
->lock
);
5479 ret
= add_extent_mapping(em_tree
, em
);
5480 write_unlock(&em_tree
->lock
);
5483 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5486 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5487 ins
.offset
, ins
.offset
, 0);
5489 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5493 btrfs_end_transaction(trans
, root
);
5498 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5499 * block must be cow'd
5501 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5502 struct inode
*inode
, u64 offset
, u64 len
)
5504 struct btrfs_path
*path
;
5506 struct extent_buffer
*leaf
;
5507 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5508 struct btrfs_file_extent_item
*fi
;
5509 struct btrfs_key key
;
5517 path
= btrfs_alloc_path();
5521 ret
= btrfs_lookup_file_extent(trans
, root
, path
, inode
->i_ino
,
5526 slot
= path
->slots
[0];
5529 /* can't find the item, must cow */
5536 leaf
= path
->nodes
[0];
5537 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5538 if (key
.objectid
!= inode
->i_ino
||
5539 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5540 /* not our file or wrong item type, must cow */
5544 if (key
.offset
> offset
) {
5545 /* Wrong offset, must cow */
5549 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5550 found_type
= btrfs_file_extent_type(leaf
, fi
);
5551 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5552 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5553 /* not a regular extent, must cow */
5556 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5557 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5559 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5560 if (extent_end
< offset
+ len
) {
5561 /* extent doesn't include our full range, must cow */
5565 if (btrfs_extent_readonly(root
, disk_bytenr
))
5569 * look for other files referencing this extent, if we
5570 * find any we must cow
5572 if (btrfs_cross_ref_exist(trans
, root
, inode
->i_ino
,
5573 key
.offset
- backref_offset
, disk_bytenr
))
5577 * adjust disk_bytenr and num_bytes to cover just the bytes
5578 * in this extent we are about to write. If there
5579 * are any csums in that range we have to cow in order
5580 * to keep the csums correct
5582 disk_bytenr
+= backref_offset
;
5583 disk_bytenr
+= offset
- key
.offset
;
5584 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5585 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5588 * all of the above have passed, it is safe to overwrite this extent
5593 btrfs_free_path(path
);
5597 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5598 struct buffer_head
*bh_result
, int create
)
5600 struct extent_map
*em
;
5601 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5602 u64 start
= iblock
<< inode
->i_blkbits
;
5603 u64 len
= bh_result
->b_size
;
5604 struct btrfs_trans_handle
*trans
;
5606 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5611 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5612 * io. INLINE is special, and we could probably kludge it in here, but
5613 * it's still buffered so for safety lets just fall back to the generic
5616 * For COMPRESSED we _have_ to read the entire extent in so we can
5617 * decompress it, so there will be buffering required no matter what we
5618 * do, so go ahead and fallback to buffered.
5620 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5621 * to buffered IO. Don't blame me, this is the price we pay for using
5624 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5625 em
->block_start
== EXTENT_MAP_INLINE
) {
5626 free_extent_map(em
);
5630 /* Just a good old fashioned hole, return */
5631 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
5632 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
5633 free_extent_map(em
);
5634 /* DIO will do one hole at a time, so just unlock a sector */
5635 unlock_extent(&BTRFS_I(inode
)->io_tree
, start
,
5636 start
+ root
->sectorsize
- 1, GFP_NOFS
);
5641 * We don't allocate a new extent in the following cases
5643 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5645 * 2) The extent is marked as PREALLOC. We're good to go here and can
5646 * just use the extent.
5650 len
= em
->len
- (start
- em
->start
);
5654 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
5655 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
5656 em
->block_start
!= EXTENT_MAP_HOLE
)) {
5661 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5662 type
= BTRFS_ORDERED_PREALLOC
;
5664 type
= BTRFS_ORDERED_NOCOW
;
5665 len
= min(len
, em
->len
- (start
- em
->start
));
5666 block_start
= em
->block_start
+ (start
- em
->start
);
5669 * we're not going to log anything, but we do need
5670 * to make sure the current transaction stays open
5671 * while we look for nocow cross refs
5673 trans
= btrfs_join_transaction(root
, 0);
5677 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
5678 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
5679 block_start
, len
, len
, type
);
5680 btrfs_end_transaction(trans
, root
);
5682 free_extent_map(em
);
5687 btrfs_end_transaction(trans
, root
);
5691 * this will cow the extent, reset the len in case we changed
5694 len
= bh_result
->b_size
;
5695 free_extent_map(em
);
5696 em
= btrfs_new_extent_direct(inode
, start
, len
);
5699 len
= min(len
, em
->len
- (start
- em
->start
));
5701 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, start
+ len
- 1,
5702 EXTENT_LOCKED
| EXTENT_DELALLOC
| EXTENT_DIRTY
, 1,
5705 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
5707 bh_result
->b_size
= len
;
5708 bh_result
->b_bdev
= em
->bdev
;
5709 set_buffer_mapped(bh_result
);
5710 if (create
&& !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
5711 set_buffer_new(bh_result
);
5713 free_extent_map(em
);
5718 struct btrfs_dio_private
{
5719 struct inode
*inode
;
5726 /* number of bios pending for this dio */
5727 atomic_t pending_bios
;
5732 struct bio
*orig_bio
;
5735 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
5737 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5738 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
5739 struct bio_vec
*bvec
= bio
->bi_io_vec
;
5740 struct inode
*inode
= dip
->inode
;
5741 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5743 u32
*private = dip
->csums
;
5745 start
= dip
->logical_offset
;
5747 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
5748 struct page
*page
= bvec
->bv_page
;
5751 unsigned long flags
;
5753 local_irq_save(flags
);
5754 kaddr
= kmap_atomic(page
, KM_IRQ0
);
5755 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
5756 csum
, bvec
->bv_len
);
5757 btrfs_csum_final(csum
, (char *)&csum
);
5758 kunmap_atomic(kaddr
, KM_IRQ0
);
5759 local_irq_restore(flags
);
5761 flush_dcache_page(bvec
->bv_page
);
5762 if (csum
!= *private) {
5763 printk(KERN_ERR
"btrfs csum failed ino %lu off"
5764 " %llu csum %u private %u\n",
5765 inode
->i_ino
, (unsigned long long)start
,
5771 start
+= bvec
->bv_len
;
5774 } while (bvec
<= bvec_end
);
5776 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
5777 dip
->logical_offset
+ dip
->bytes
- 1, GFP_NOFS
);
5778 bio
->bi_private
= dip
->private;
5783 /* If we had a csum failure make sure to clear the uptodate flag */
5785 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5786 dio_end_io(bio
, err
);
5789 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
5791 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5792 struct inode
*inode
= dip
->inode
;
5793 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5794 struct btrfs_trans_handle
*trans
;
5795 struct btrfs_ordered_extent
*ordered
= NULL
;
5796 struct extent_state
*cached_state
= NULL
;
5797 u64 ordered_offset
= dip
->logical_offset
;
5798 u64 ordered_bytes
= dip
->bytes
;
5804 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
5812 trans
= btrfs_join_transaction(root
, 1);
5813 if (IS_ERR(trans
)) {
5817 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5819 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
)) {
5820 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered
);
5822 ret
= btrfs_update_inode(trans
, root
, inode
);
5827 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5828 ordered
->file_offset
+ ordered
->len
- 1, 0,
5829 &cached_state
, GFP_NOFS
);
5831 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
)) {
5832 ret
= btrfs_mark_extent_written(trans
, inode
,
5833 ordered
->file_offset
,
5834 ordered
->file_offset
+
5841 ret
= insert_reserved_file_extent(trans
, inode
,
5842 ordered
->file_offset
,
5848 BTRFS_FILE_EXTENT_REG
);
5849 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
5850 ordered
->file_offset
, ordered
->len
);
5858 add_pending_csums(trans
, inode
, ordered
->file_offset
, &ordered
->list
);
5859 btrfs_ordered_update_i_size(inode
, 0, ordered
);
5860 btrfs_update_inode(trans
, root
, inode
);
5862 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, ordered
->file_offset
,
5863 ordered
->file_offset
+ ordered
->len
- 1,
5864 &cached_state
, GFP_NOFS
);
5866 btrfs_delalloc_release_metadata(inode
, ordered
->len
);
5867 btrfs_end_transaction(trans
, root
);
5868 ordered_offset
= ordered
->file_offset
+ ordered
->len
;
5869 btrfs_put_ordered_extent(ordered
);
5870 btrfs_put_ordered_extent(ordered
);
5874 * our bio might span multiple ordered extents. If we haven't
5875 * completed the accounting for the whole dio, go back and try again
5877 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
5878 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
5883 bio
->bi_private
= dip
->private;
5888 /* If we had an error make sure to clear the uptodate flag */
5890 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
5891 dio_end_io(bio
, err
);
5894 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
5895 struct bio
*bio
, int mirror_num
,
5896 unsigned long bio_flags
, u64 offset
)
5899 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5900 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
5905 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
5907 struct btrfs_dio_private
*dip
= bio
->bi_private
;
5910 printk(KERN_ERR
"btrfs direct IO failed ino %lu rw %lu "
5911 "sector %#Lx len %u err no %d\n",
5912 dip
->inode
->i_ino
, bio
->bi_rw
,
5913 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
5917 * before atomic variable goto zero, we must make sure
5918 * dip->errors is perceived to be set.
5920 smp_mb__before_atomic_dec();
5923 /* if there are more bios still pending for this dio, just exit */
5924 if (!atomic_dec_and_test(&dip
->pending_bios
))
5928 bio_io_error(dip
->orig_bio
);
5930 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
5931 bio_endio(dip
->orig_bio
, 0);
5937 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
5938 u64 first_sector
, gfp_t gfp_flags
)
5940 int nr_vecs
= bio_get_nr_vecs(bdev
);
5941 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
5944 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
5945 int rw
, u64 file_offset
, int skip_sum
,
5948 int write
= rw
& REQ_WRITE
;
5949 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5953 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
5957 if (write
&& !skip_sum
) {
5958 ret
= btrfs_wq_submit_bio(root
->fs_info
,
5959 inode
, rw
, bio
, 0, 0,
5961 __btrfs_submit_bio_start_direct_io
,
5962 __btrfs_submit_bio_done
);
5964 } else if (!skip_sum
) {
5965 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
,
5966 file_offset
, csums
);
5971 ret
= btrfs_map_bio(root
, rw
, bio
, 0, 1);
5977 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
5980 struct inode
*inode
= dip
->inode
;
5981 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5982 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
5984 struct bio
*orig_bio
= dip
->orig_bio
;
5985 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
5986 u64 start_sector
= orig_bio
->bi_sector
;
5987 u64 file_offset
= dip
->logical_offset
;
5991 u32
*csums
= dip
->csums
;
5993 int write
= rw
& REQ_WRITE
;
5995 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
5998 bio
->bi_private
= dip
;
5999 bio
->bi_end_io
= btrfs_end_dio_bio
;
6000 atomic_inc(&dip
->pending_bios
);
6002 map_length
= orig_bio
->bi_size
;
6003 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6004 &map_length
, NULL
, 0);
6010 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
6011 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
6012 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
6013 bvec
->bv_offset
) < bvec
->bv_len
)) {
6015 * inc the count before we submit the bio so
6016 * we know the end IO handler won't happen before
6017 * we inc the count. Otherwise, the dip might get freed
6018 * before we're done setting it up
6020 atomic_inc(&dip
->pending_bios
);
6021 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
6022 file_offset
, skip_sum
,
6026 atomic_dec(&dip
->pending_bios
);
6030 /* Write's use the ordered csums */
6031 if (!write
&& !skip_sum
)
6032 csums
= csums
+ nr_pages
;
6033 start_sector
+= submit_len
>> 9;
6034 file_offset
+= submit_len
;
6039 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6040 start_sector
, GFP_NOFS
);
6043 bio
->bi_private
= dip
;
6044 bio
->bi_end_io
= btrfs_end_dio_bio
;
6046 map_length
= orig_bio
->bi_size
;
6047 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6048 &map_length
, NULL
, 0);
6054 submit_len
+= bvec
->bv_len
;
6060 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6069 * before atomic variable goto zero, we must
6070 * make sure dip->errors is perceived to be set.
6072 smp_mb__before_atomic_dec();
6073 if (atomic_dec_and_test(&dip
->pending_bios
))
6074 bio_io_error(dip
->orig_bio
);
6076 /* bio_end_io() will handle error, so we needn't return it */
6080 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6083 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6084 struct btrfs_dio_private
*dip
;
6085 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6087 int write
= rw
& REQ_WRITE
;
6090 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6092 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6099 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6100 if (!write
&& !skip_sum
) {
6101 dip
->csums
= kmalloc(sizeof(u32
) * bio
->bi_vcnt
, GFP_NOFS
);
6109 dip
->private = bio
->bi_private
;
6111 dip
->logical_offset
= file_offset
;
6115 dip
->bytes
+= bvec
->bv_len
;
6117 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6119 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6120 bio
->bi_private
= dip
;
6122 dip
->orig_bio
= bio
;
6123 atomic_set(&dip
->pending_bios
, 0);
6126 bio
->bi_end_io
= btrfs_endio_direct_write
;
6128 bio
->bi_end_io
= btrfs_endio_direct_read
;
6130 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6135 * If this is a write, we need to clean up the reserved space and kill
6136 * the ordered extent.
6139 struct btrfs_ordered_extent
*ordered
;
6140 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6141 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6142 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6143 btrfs_free_reserved_extent(root
, ordered
->start
,
6145 btrfs_put_ordered_extent(ordered
);
6146 btrfs_put_ordered_extent(ordered
);
6148 bio_endio(bio
, ret
);
6151 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6152 const struct iovec
*iov
, loff_t offset
,
6153 unsigned long nr_segs
)
6159 unsigned blocksize_mask
= root
->sectorsize
- 1;
6160 ssize_t retval
= -EINVAL
;
6161 loff_t end
= offset
;
6163 if (offset
& blocksize_mask
)
6166 /* Check the memory alignment. Blocks cannot straddle pages */
6167 for (seg
= 0; seg
< nr_segs
; seg
++) {
6168 addr
= (unsigned long)iov
[seg
].iov_base
;
6169 size
= iov
[seg
].iov_len
;
6171 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6174 /* If this is a write we don't need to check anymore */
6179 * Check to make sure we don't have duplicate iov_base's in this
6180 * iovec, if so return EINVAL, otherwise we'll get csum errors
6181 * when reading back.
6183 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6184 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6192 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6193 const struct iovec
*iov
, loff_t offset
,
6194 unsigned long nr_segs
)
6196 struct file
*file
= iocb
->ki_filp
;
6197 struct inode
*inode
= file
->f_mapping
->host
;
6198 struct btrfs_ordered_extent
*ordered
;
6199 struct extent_state
*cached_state
= NULL
;
6200 u64 lockstart
, lockend
;
6202 int writing
= rw
& WRITE
;
6204 size_t count
= iov_length(iov
, nr_segs
);
6206 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6212 lockend
= offset
+ count
- 1;
6215 ret
= btrfs_delalloc_reserve_space(inode
, count
);
6221 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6222 0, &cached_state
, GFP_NOFS
);
6224 * We're concerned with the entire range that we're going to be
6225 * doing DIO to, so we need to make sure theres no ordered
6226 * extents in this range.
6228 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6229 lockend
- lockstart
+ 1);
6232 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6233 &cached_state
, GFP_NOFS
);
6234 btrfs_start_ordered_extent(inode
, ordered
, 1);
6235 btrfs_put_ordered_extent(ordered
);
6240 * we don't use btrfs_set_extent_delalloc because we don't want
6241 * the dirty or uptodate bits
6244 write_bits
= EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
;
6245 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6246 EXTENT_DELALLOC
, 0, NULL
, &cached_state
,
6249 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6250 lockend
, EXTENT_LOCKED
| write_bits
,
6251 1, 0, &cached_state
, GFP_NOFS
);
6256 free_extent_state(cached_state
);
6257 cached_state
= NULL
;
6259 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
6260 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6261 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6262 btrfs_submit_direct
, 0);
6264 if (ret
< 0 && ret
!= -EIOCBQUEUED
) {
6265 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
,
6266 offset
+ iov_length(iov
, nr_segs
) - 1,
6267 EXTENT_LOCKED
| write_bits
, 1, 0,
6268 &cached_state
, GFP_NOFS
);
6269 } else if (ret
>= 0 && ret
< iov_length(iov
, nr_segs
)) {
6271 * We're falling back to buffered, unlock the section we didn't
6274 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, offset
+ ret
,
6275 offset
+ iov_length(iov
, nr_segs
) - 1,
6276 EXTENT_LOCKED
| write_bits
, 1, 0,
6277 &cached_state
, GFP_NOFS
);
6280 free_extent_state(cached_state
);
6284 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6285 __u64 start
, __u64 len
)
6287 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6290 int btrfs_readpage(struct file
*file
, struct page
*page
)
6292 struct extent_io_tree
*tree
;
6293 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6294 return extent_read_full_page(tree
, page
, btrfs_get_extent
);
6297 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6299 struct extent_io_tree
*tree
;
6302 if (current
->flags
& PF_MEMALLOC
) {
6303 redirty_page_for_writepage(wbc
, page
);
6307 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6308 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6311 int btrfs_writepages(struct address_space
*mapping
,
6312 struct writeback_control
*wbc
)
6314 struct extent_io_tree
*tree
;
6316 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6317 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6321 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6322 struct list_head
*pages
, unsigned nr_pages
)
6324 struct extent_io_tree
*tree
;
6325 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6326 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6329 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6331 struct extent_io_tree
*tree
;
6332 struct extent_map_tree
*map
;
6335 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6336 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6337 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6339 ClearPagePrivate(page
);
6340 set_page_private(page
, 0);
6341 page_cache_release(page
);
6346 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6348 if (PageWriteback(page
) || PageDirty(page
))
6350 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6353 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6355 struct extent_io_tree
*tree
;
6356 struct btrfs_ordered_extent
*ordered
;
6357 struct extent_state
*cached_state
= NULL
;
6358 u64 page_start
= page_offset(page
);
6359 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6363 * we have the page locked, so new writeback can't start,
6364 * and the dirty bit won't be cleared while we are here.
6366 * Wait for IO on this page so that we can safely clear
6367 * the PagePrivate2 bit and do ordered accounting
6369 wait_on_page_writeback(page
);
6371 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6373 btrfs_releasepage(page
, GFP_NOFS
);
6376 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6378 ordered
= btrfs_lookup_ordered_extent(page
->mapping
->host
,
6382 * IO on this page will never be started, so we need
6383 * to account for any ordered extents now
6385 clear_extent_bit(tree
, page_start
, page_end
,
6386 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6387 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
, 1, 0,
6388 &cached_state
, GFP_NOFS
);
6390 * whoever cleared the private bit is responsible
6391 * for the finish_ordered_io
6393 if (TestClearPagePrivate2(page
)) {
6394 btrfs_finish_ordered_io(page
->mapping
->host
,
6395 page_start
, page_end
);
6397 btrfs_put_ordered_extent(ordered
);
6398 cached_state
= NULL
;
6399 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
,
6402 clear_extent_bit(tree
, page_start
, page_end
,
6403 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6404 EXTENT_DO_ACCOUNTING
, 1, 1, &cached_state
, GFP_NOFS
);
6405 __btrfs_releasepage(page
, GFP_NOFS
);
6407 ClearPageChecked(page
);
6408 if (PagePrivate(page
)) {
6409 ClearPagePrivate(page
);
6410 set_page_private(page
, 0);
6411 page_cache_release(page
);
6416 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6417 * called from a page fault handler when a page is first dirtied. Hence we must
6418 * be careful to check for EOF conditions here. We set the page up correctly
6419 * for a written page which means we get ENOSPC checking when writing into
6420 * holes and correct delalloc and unwritten extent mapping on filesystems that
6421 * support these features.
6423 * We are not allowed to take the i_mutex here so we have to play games to
6424 * protect against truncate races as the page could now be beyond EOF. Because
6425 * vmtruncate() writes the inode size before removing pages, once we have the
6426 * page lock we can determine safely if the page is beyond EOF. If it is not
6427 * beyond EOF, then the page is guaranteed safe against truncation until we
6430 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6432 struct page
*page
= vmf
->page
;
6433 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6434 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6435 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6436 struct btrfs_ordered_extent
*ordered
;
6437 struct extent_state
*cached_state
= NULL
;
6439 unsigned long zero_start
;
6445 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6449 else /* -ENOSPC, -EIO, etc */
6450 ret
= VM_FAULT_SIGBUS
;
6454 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6457 size
= i_size_read(inode
);
6458 page_start
= page_offset(page
);
6459 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6461 if ((page
->mapping
!= inode
->i_mapping
) ||
6462 (page_start
>= size
)) {
6463 /* page got truncated out from underneath us */
6466 wait_on_page_writeback(page
);
6468 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
,
6470 set_page_extent_mapped(page
);
6473 * we can't set the delalloc bits if there are pending ordered
6474 * extents. Drop our locks and wait for them to finish
6476 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6478 unlock_extent_cached(io_tree
, page_start
, page_end
,
6479 &cached_state
, GFP_NOFS
);
6481 btrfs_start_ordered_extent(inode
, ordered
, 1);
6482 btrfs_put_ordered_extent(ordered
);
6487 * XXX - page_mkwrite gets called every time the page is dirtied, even
6488 * if it was already dirty, so for space accounting reasons we need to
6489 * clear any delalloc bits for the range we are fixing to save. There
6490 * is probably a better way to do this, but for now keep consistent with
6491 * prepare_pages in the normal write path.
6493 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6494 EXTENT_DIRTY
| EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
,
6495 0, 0, &cached_state
, GFP_NOFS
);
6497 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6500 unlock_extent_cached(io_tree
, page_start
, page_end
,
6501 &cached_state
, GFP_NOFS
);
6502 ret
= VM_FAULT_SIGBUS
;
6507 /* page is wholly or partially inside EOF */
6508 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6509 zero_start
= size
& ~PAGE_CACHE_MASK
;
6511 zero_start
= PAGE_CACHE_SIZE
;
6513 if (zero_start
!= PAGE_CACHE_SIZE
) {
6515 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6516 flush_dcache_page(page
);
6519 ClearPageChecked(page
);
6520 set_page_dirty(page
);
6521 SetPageUptodate(page
);
6523 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6524 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6526 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6530 return VM_FAULT_LOCKED
;
6532 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6537 static int btrfs_truncate(struct inode
*inode
)
6539 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6542 struct btrfs_trans_handle
*trans
;
6544 u64 mask
= root
->sectorsize
- 1;
6546 ret
= btrfs_truncate_page(inode
->i_mapping
, inode
->i_size
);
6550 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6551 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6553 trans
= btrfs_start_transaction(root
, 5);
6555 return PTR_ERR(trans
);
6557 btrfs_set_trans_block_group(trans
, inode
);
6559 ret
= btrfs_orphan_add(trans
, inode
);
6561 btrfs_end_transaction(trans
, root
);
6565 nr
= trans
->blocks_used
;
6566 btrfs_end_transaction(trans
, root
);
6567 btrfs_btree_balance_dirty(root
, nr
);
6569 /* Now start a transaction for the truncate */
6570 trans
= btrfs_start_transaction(root
, 0);
6572 return PTR_ERR(trans
);
6573 btrfs_set_trans_block_group(trans
, inode
);
6574 trans
->block_rsv
= root
->orphan_block_rsv
;
6577 * setattr is responsible for setting the ordered_data_close flag,
6578 * but that is only tested during the last file release. That
6579 * could happen well after the next commit, leaving a great big
6580 * window where new writes may get lost if someone chooses to write
6581 * to this file after truncating to zero
6583 * The inode doesn't have any dirty data here, and so if we commit
6584 * this is a noop. If someone immediately starts writing to the inode
6585 * it is very likely we'll catch some of their writes in this
6586 * transaction, and the commit will find this file on the ordered
6587 * data list with good things to send down.
6589 * This is a best effort solution, there is still a window where
6590 * using truncate to replace the contents of the file will
6591 * end up with a zero length file after a crash.
6593 if (inode
->i_size
== 0 && BTRFS_I(inode
)->ordered_data_close
)
6594 btrfs_add_ordered_operation(trans
, root
, inode
);
6598 trans
= btrfs_start_transaction(root
, 0);
6600 return PTR_ERR(trans
);
6601 btrfs_set_trans_block_group(trans
, inode
);
6602 trans
->block_rsv
= root
->orphan_block_rsv
;
6605 ret
= btrfs_block_rsv_check(trans
, root
,
6606 root
->orphan_block_rsv
, 0, 5);
6607 if (ret
== -EAGAIN
) {
6608 ret
= btrfs_commit_transaction(trans
, root
);
6618 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6620 BTRFS_EXTENT_DATA_KEY
);
6621 if (ret
!= -EAGAIN
) {
6626 ret
= btrfs_update_inode(trans
, root
, inode
);
6632 nr
= trans
->blocks_used
;
6633 btrfs_end_transaction(trans
, root
);
6635 btrfs_btree_balance_dirty(root
, nr
);
6638 if (ret
== 0 && inode
->i_nlink
> 0) {
6639 ret
= btrfs_orphan_del(trans
, inode
);
6642 } else if (ret
&& inode
->i_nlink
> 0) {
6644 * Failed to do the truncate, remove us from the in memory
6647 ret
= btrfs_orphan_del(NULL
, inode
);
6650 ret
= btrfs_update_inode(trans
, root
, inode
);
6654 nr
= trans
->blocks_used
;
6655 ret
= btrfs_end_transaction_throttle(trans
, root
);
6658 btrfs_btree_balance_dirty(root
, nr
);
6664 * create a new subvolume directory/inode (helper for the ioctl).
6666 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
6667 struct btrfs_root
*new_root
,
6668 u64 new_dirid
, u64 alloc_hint
)
6670 struct inode
*inode
;
6674 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2, new_dirid
,
6675 new_dirid
, alloc_hint
, S_IFDIR
| 0700, &index
);
6677 return PTR_ERR(inode
);
6678 inode
->i_op
= &btrfs_dir_inode_operations
;
6679 inode
->i_fop
= &btrfs_dir_file_operations
;
6682 btrfs_i_size_write(inode
, 0);
6684 err
= btrfs_update_inode(trans
, new_root
, inode
);
6691 /* helper function for file defrag and space balancing. This
6692 * forces readahead on a given range of bytes in an inode
6694 unsigned long btrfs_force_ra(struct address_space
*mapping
,
6695 struct file_ra_state
*ra
, struct file
*file
,
6696 pgoff_t offset
, pgoff_t last_index
)
6698 pgoff_t req_size
= last_index
- offset
+ 1;
6700 page_cache_sync_readahead(mapping
, ra
, file
, offset
, req_size
);
6701 return offset
+ req_size
;
6704 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
6706 struct btrfs_inode
*ei
;
6707 struct inode
*inode
;
6709 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
6714 ei
->space_info
= NULL
;
6718 ei
->last_sub_trans
= 0;
6719 ei
->logged_trans
= 0;
6720 ei
->delalloc_bytes
= 0;
6721 ei
->reserved_bytes
= 0;
6722 ei
->disk_i_size
= 0;
6724 ei
->index_cnt
= (u64
)-1;
6725 ei
->last_unlink_trans
= 0;
6727 atomic_set(&ei
->outstanding_extents
, 0);
6728 atomic_set(&ei
->reserved_extents
, 0);
6730 ei
->ordered_data_close
= 0;
6731 ei
->orphan_meta_reserved
= 0;
6732 ei
->dummy_inode
= 0;
6733 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
6735 inode
= &ei
->vfs_inode
;
6736 extent_map_tree_init(&ei
->extent_tree
, GFP_NOFS
);
6737 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
, GFP_NOFS
);
6738 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
, GFP_NOFS
);
6739 mutex_init(&ei
->log_mutex
);
6740 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
6741 INIT_LIST_HEAD(&ei
->i_orphan
);
6742 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
6743 INIT_LIST_HEAD(&ei
->ordered_operations
);
6744 RB_CLEAR_NODE(&ei
->rb_node
);
6749 static void btrfs_i_callback(struct rcu_head
*head
)
6751 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
6752 INIT_LIST_HEAD(&inode
->i_dentry
);
6753 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
6756 void btrfs_destroy_inode(struct inode
*inode
)
6758 struct btrfs_ordered_extent
*ordered
;
6759 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6761 WARN_ON(!list_empty(&inode
->i_dentry
));
6762 WARN_ON(inode
->i_data
.nrpages
);
6763 WARN_ON(atomic_read(&BTRFS_I(inode
)->outstanding_extents
));
6764 WARN_ON(atomic_read(&BTRFS_I(inode
)->reserved_extents
));
6767 * This can happen where we create an inode, but somebody else also
6768 * created the same inode and we need to destroy the one we already
6775 * Make sure we're properly removed from the ordered operation
6779 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
6780 spin_lock(&root
->fs_info
->ordered_extent_lock
);
6781 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
6782 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
6785 if (root
== root
->fs_info
->tree_root
) {
6786 struct btrfs_block_group_cache
*block_group
;
6788 block_group
= btrfs_lookup_block_group(root
->fs_info
,
6789 BTRFS_I(inode
)->block_group
);
6790 if (block_group
&& block_group
->inode
== inode
) {
6791 spin_lock(&block_group
->lock
);
6792 block_group
->inode
= NULL
;
6793 spin_unlock(&block_group
->lock
);
6794 btrfs_put_block_group(block_group
);
6795 } else if (block_group
) {
6796 btrfs_put_block_group(block_group
);
6800 spin_lock(&root
->orphan_lock
);
6801 if (!list_empty(&BTRFS_I(inode
)->i_orphan
)) {
6802 printk(KERN_INFO
"BTRFS: inode %lu still on the orphan list\n",
6804 list_del_init(&BTRFS_I(inode
)->i_orphan
);
6806 spin_unlock(&root
->orphan_lock
);
6809 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
6813 printk(KERN_ERR
"btrfs found ordered "
6814 "extent %llu %llu on inode cleanup\n",
6815 (unsigned long long)ordered
->file_offset
,
6816 (unsigned long long)ordered
->len
);
6817 btrfs_remove_ordered_extent(inode
, ordered
);
6818 btrfs_put_ordered_extent(ordered
);
6819 btrfs_put_ordered_extent(ordered
);
6822 inode_tree_del(inode
);
6823 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
6825 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
6828 int btrfs_drop_inode(struct inode
*inode
)
6830 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6832 if (btrfs_root_refs(&root
->root_item
) == 0 &&
6833 root
!= root
->fs_info
->tree_root
)
6836 return generic_drop_inode(inode
);
6839 static void init_once(void *foo
)
6841 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
6843 inode_init_once(&ei
->vfs_inode
);
6846 void btrfs_destroy_cachep(void)
6848 if (btrfs_inode_cachep
)
6849 kmem_cache_destroy(btrfs_inode_cachep
);
6850 if (btrfs_trans_handle_cachep
)
6851 kmem_cache_destroy(btrfs_trans_handle_cachep
);
6852 if (btrfs_transaction_cachep
)
6853 kmem_cache_destroy(btrfs_transaction_cachep
);
6854 if (btrfs_path_cachep
)
6855 kmem_cache_destroy(btrfs_path_cachep
);
6856 if (btrfs_free_space_cachep
)
6857 kmem_cache_destroy(btrfs_free_space_cachep
);
6860 int btrfs_init_cachep(void)
6862 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode_cache",
6863 sizeof(struct btrfs_inode
), 0,
6864 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
6865 if (!btrfs_inode_cachep
)
6868 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle_cache",
6869 sizeof(struct btrfs_trans_handle
), 0,
6870 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6871 if (!btrfs_trans_handle_cachep
)
6874 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction_cache",
6875 sizeof(struct btrfs_transaction
), 0,
6876 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6877 if (!btrfs_transaction_cachep
)
6880 btrfs_path_cachep
= kmem_cache_create("btrfs_path_cache",
6881 sizeof(struct btrfs_path
), 0,
6882 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6883 if (!btrfs_path_cachep
)
6886 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space_cache",
6887 sizeof(struct btrfs_free_space
), 0,
6888 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
6889 if (!btrfs_free_space_cachep
)
6894 btrfs_destroy_cachep();
6898 static int btrfs_getattr(struct vfsmount
*mnt
,
6899 struct dentry
*dentry
, struct kstat
*stat
)
6901 struct inode
*inode
= dentry
->d_inode
;
6902 generic_fillattr(inode
, stat
);
6903 stat
->dev
= BTRFS_I(inode
)->root
->anon_super
.s_dev
;
6904 stat
->blksize
= PAGE_CACHE_SIZE
;
6905 stat
->blocks
= (inode_get_bytes(inode
) +
6906 BTRFS_I(inode
)->delalloc_bytes
) >> 9;
6911 * If a file is moved, it will inherit the cow and compression flags of the new
6914 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
6916 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
6917 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
6919 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
6920 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
6922 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
6924 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
)
6925 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
6927 b_inode
->flags
&= ~BTRFS_INODE_COMPRESS
;
6930 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
6931 struct inode
*new_dir
, struct dentry
*new_dentry
)
6933 struct btrfs_trans_handle
*trans
;
6934 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
6935 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
6936 struct inode
*new_inode
= new_dentry
->d_inode
;
6937 struct inode
*old_inode
= old_dentry
->d_inode
;
6938 struct timespec ctime
= CURRENT_TIME
;
6943 if (new_dir
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
6946 /* we only allow rename subvolume link between subvolumes */
6947 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
6950 if (old_inode
->i_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
6951 (new_inode
&& new_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
))
6954 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
6955 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
6958 * we're using rename to replace one file with another.
6959 * and the replacement file is large. Start IO on it now so
6960 * we don't add too much work to the end of the transaction
6962 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
6963 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
6964 filemap_flush(old_inode
->i_mapping
);
6966 /* close the racy window with snapshot create/destroy ioctl */
6967 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
6968 down_read(&root
->fs_info
->subvol_sem
);
6970 * We want to reserve the absolute worst case amount of items. So if
6971 * both inodes are subvols and we need to unlink them then that would
6972 * require 4 item modifications, but if they are both normal inodes it
6973 * would require 5 item modifications, so we'll assume their normal
6974 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6975 * should cover the worst case number of items we'll modify.
6977 trans
= btrfs_start_transaction(root
, 20);
6978 if (IS_ERR(trans
)) {
6979 ret
= PTR_ERR(trans
);
6983 btrfs_set_trans_block_group(trans
, new_dir
);
6986 btrfs_record_root_in_trans(trans
, dest
);
6988 ret
= btrfs_set_inode_index(new_dir
, &index
);
6992 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
6993 /* force full log commit if subvolume involved. */
6994 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
6996 ret
= btrfs_insert_inode_ref(trans
, dest
,
6997 new_dentry
->d_name
.name
,
6998 new_dentry
->d_name
.len
,
7000 new_dir
->i_ino
, index
);
7004 * this is an ugly little race, but the rename is required
7005 * to make sure that if we crash, the inode is either at the
7006 * old name or the new one. pinning the log transaction lets
7007 * us make sure we don't allow a log commit to come in after
7008 * we unlink the name but before we add the new name back in.
7010 btrfs_pin_log_trans(root
);
7013 * make sure the inode gets flushed if it is replacing
7016 if (new_inode
&& new_inode
->i_size
&&
7017 old_inode
&& S_ISREG(old_inode
->i_mode
)) {
7018 btrfs_add_ordered_operation(trans
, root
, old_inode
);
7021 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
7022 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
7023 old_inode
->i_ctime
= ctime
;
7025 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
7026 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
7028 if (unlikely(old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7029 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
7030 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
7031 old_dentry
->d_name
.name
,
7032 old_dentry
->d_name
.len
);
7034 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7035 old_dentry
->d_inode
,
7036 old_dentry
->d_name
.name
,
7037 old_dentry
->d_name
.len
);
7039 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7044 new_inode
->i_ctime
= CURRENT_TIME
;
7045 if (unlikely(new_inode
->i_ino
==
7046 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7047 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7048 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7050 new_dentry
->d_name
.name
,
7051 new_dentry
->d_name
.len
);
7052 BUG_ON(new_inode
->i_nlink
== 0);
7054 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7055 new_dentry
->d_inode
,
7056 new_dentry
->d_name
.name
,
7057 new_dentry
->d_name
.len
);
7060 if (new_inode
->i_nlink
== 0) {
7061 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7066 fixup_inode_flags(new_dir
, old_inode
);
7068 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7069 new_dentry
->d_name
.name
,
7070 new_dentry
->d_name
.len
, 0, index
);
7073 if (old_inode
->i_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7074 struct dentry
*parent
= dget_parent(new_dentry
);
7075 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7077 btrfs_end_log_trans(root
);
7080 btrfs_end_transaction_throttle(trans
, root
);
7082 if (old_inode
->i_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7083 up_read(&root
->fs_info
->subvol_sem
);
7089 * some fairly slow code that needs optimization. This walks the list
7090 * of all the inodes with pending delalloc and forces them to disk.
7092 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7094 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7095 struct btrfs_inode
*binode
;
7096 struct inode
*inode
;
7098 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7101 spin_lock(&root
->fs_info
->delalloc_lock
);
7102 while (!list_empty(head
)) {
7103 binode
= list_entry(head
->next
, struct btrfs_inode
,
7105 inode
= igrab(&binode
->vfs_inode
);
7107 list_del_init(&binode
->delalloc_inodes
);
7108 spin_unlock(&root
->fs_info
->delalloc_lock
);
7110 filemap_flush(inode
->i_mapping
);
7112 btrfs_add_delayed_iput(inode
);
7117 spin_lock(&root
->fs_info
->delalloc_lock
);
7119 spin_unlock(&root
->fs_info
->delalloc_lock
);
7121 /* the filemap_flush will queue IO into the worker threads, but
7122 * we have to make sure the IO is actually started and that
7123 * ordered extents get created before we return
7125 atomic_inc(&root
->fs_info
->async_submit_draining
);
7126 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7127 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7128 wait_event(root
->fs_info
->async_submit_wait
,
7129 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7130 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7132 atomic_dec(&root
->fs_info
->async_submit_draining
);
7136 int btrfs_start_one_delalloc_inode(struct btrfs_root
*root
, int delay_iput
,
7139 struct btrfs_inode
*binode
;
7140 struct inode
*inode
= NULL
;
7142 spin_lock(&root
->fs_info
->delalloc_lock
);
7143 while (!list_empty(&root
->fs_info
->delalloc_inodes
)) {
7144 binode
= list_entry(root
->fs_info
->delalloc_inodes
.next
,
7145 struct btrfs_inode
, delalloc_inodes
);
7146 inode
= igrab(&binode
->vfs_inode
);
7148 list_move_tail(&binode
->delalloc_inodes
,
7149 &root
->fs_info
->delalloc_inodes
);
7153 list_del_init(&binode
->delalloc_inodes
);
7154 cond_resched_lock(&root
->fs_info
->delalloc_lock
);
7156 spin_unlock(&root
->fs_info
->delalloc_lock
);
7160 filemap_write_and_wait(inode
->i_mapping
);
7162 * We have to do this because compression doesn't
7163 * actually set PG_writeback until it submits the pages
7164 * for IO, which happens in an async thread, so we could
7165 * race and not actually wait for any writeback pages
7166 * because they've not been submitted yet. Technically
7167 * this could still be the case for the ordered stuff
7168 * since the async thread may not have started to do its
7169 * work yet. If this becomes the case then we need to
7170 * figure out a way to make sure that in writepage we
7171 * wait for any async pages to be submitted before
7172 * returning so that fdatawait does what its supposed to
7175 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
7177 filemap_flush(inode
->i_mapping
);
7180 btrfs_add_delayed_iput(inode
);
7188 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7189 const char *symname
)
7191 struct btrfs_trans_handle
*trans
;
7192 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7193 struct btrfs_path
*path
;
7194 struct btrfs_key key
;
7195 struct inode
*inode
= NULL
;
7203 struct btrfs_file_extent_item
*ei
;
7204 struct extent_buffer
*leaf
;
7205 unsigned long nr
= 0;
7207 name_len
= strlen(symname
) + 1;
7208 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7209 return -ENAMETOOLONG
;
7211 err
= btrfs_find_free_objectid(NULL
, root
, dir
->i_ino
, &objectid
);
7215 * 2 items for inode item and ref
7216 * 2 items for dir items
7217 * 1 item for xattr if selinux is on
7219 trans
= btrfs_start_transaction(root
, 5);
7221 return PTR_ERR(trans
);
7223 btrfs_set_trans_block_group(trans
, dir
);
7225 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7226 dentry
->d_name
.len
, dir
->i_ino
, objectid
,
7227 BTRFS_I(dir
)->block_group
, S_IFLNK
|S_IRWXUGO
,
7229 err
= PTR_ERR(inode
);
7233 err
= btrfs_init_inode_security(trans
, inode
, dir
);
7239 btrfs_set_trans_block_group(trans
, inode
);
7240 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7244 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7245 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7246 inode
->i_fop
= &btrfs_file_operations
;
7247 inode
->i_op
= &btrfs_file_inode_operations
;
7248 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7250 btrfs_update_inode_block_group(trans
, inode
);
7251 btrfs_update_inode_block_group(trans
, dir
);
7255 path
= btrfs_alloc_path();
7257 key
.objectid
= inode
->i_ino
;
7259 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7260 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7261 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7267 leaf
= path
->nodes
[0];
7268 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7269 struct btrfs_file_extent_item
);
7270 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7271 btrfs_set_file_extent_type(leaf
, ei
,
7272 BTRFS_FILE_EXTENT_INLINE
);
7273 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7274 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7275 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7276 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7278 ptr
= btrfs_file_extent_inline_start(ei
);
7279 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7280 btrfs_mark_buffer_dirty(leaf
);
7281 btrfs_free_path(path
);
7283 inode
->i_op
= &btrfs_symlink_inode_operations
;
7284 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7285 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7286 inode_set_bytes(inode
, name_len
);
7287 btrfs_i_size_write(inode
, name_len
- 1);
7288 err
= btrfs_update_inode(trans
, root
, inode
);
7293 nr
= trans
->blocks_used
;
7294 btrfs_end_transaction_throttle(trans
, root
);
7296 inode_dec_link_count(inode
);
7299 btrfs_btree_balance_dirty(root
, nr
);
7303 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7304 u64 start
, u64 num_bytes
, u64 min_size
,
7305 loff_t actual_len
, u64
*alloc_hint
,
7306 struct btrfs_trans_handle
*trans
)
7308 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7309 struct btrfs_key ins
;
7310 u64 cur_offset
= start
;
7313 bool own_trans
= true;
7317 while (num_bytes
> 0) {
7319 trans
= btrfs_start_transaction(root
, 3);
7320 if (IS_ERR(trans
)) {
7321 ret
= PTR_ERR(trans
);
7326 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7327 0, *alloc_hint
, (u64
)-1, &ins
, 1);
7330 btrfs_end_transaction(trans
, root
);
7334 ret
= insert_reserved_file_extent(trans
, inode
,
7335 cur_offset
, ins
.objectid
,
7336 ins
.offset
, ins
.offset
,
7337 ins
.offset
, 0, 0, 0,
7338 BTRFS_FILE_EXTENT_PREALLOC
);
7340 btrfs_drop_extent_cache(inode
, cur_offset
,
7341 cur_offset
+ ins
.offset
-1, 0);
7343 num_bytes
-= ins
.offset
;
7344 cur_offset
+= ins
.offset
;
7345 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7347 inode
->i_ctime
= CURRENT_TIME
;
7348 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7349 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7350 (actual_len
> inode
->i_size
) &&
7351 (cur_offset
> inode
->i_size
)) {
7352 if (cur_offset
> actual_len
)
7353 i_size
= actual_len
;
7355 i_size
= cur_offset
;
7356 i_size_write(inode
, i_size
);
7357 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7360 ret
= btrfs_update_inode(trans
, root
, inode
);
7364 btrfs_end_transaction(trans
, root
);
7369 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7370 u64 start
, u64 num_bytes
, u64 min_size
,
7371 loff_t actual_len
, u64
*alloc_hint
)
7373 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7374 min_size
, actual_len
, alloc_hint
,
7378 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7379 struct btrfs_trans_handle
*trans
, int mode
,
7380 u64 start
, u64 num_bytes
, u64 min_size
,
7381 loff_t actual_len
, u64
*alloc_hint
)
7383 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7384 min_size
, actual_len
, alloc_hint
, trans
);
7387 static int btrfs_set_page_dirty(struct page
*page
)
7389 return __set_page_dirty_nobuffers(page
);
7392 static int btrfs_permission(struct inode
*inode
, int mask
, unsigned int flags
)
7394 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7396 if (btrfs_root_readonly(root
) && (mask
& MAY_WRITE
))
7398 if ((BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
) && (mask
& MAY_WRITE
))
7400 return generic_permission(inode
, mask
, flags
, btrfs_check_acl
);
7403 static const struct inode_operations btrfs_dir_inode_operations
= {
7404 .getattr
= btrfs_getattr
,
7405 .lookup
= btrfs_lookup
,
7406 .create
= btrfs_create
,
7407 .unlink
= btrfs_unlink
,
7409 .mkdir
= btrfs_mkdir
,
7410 .rmdir
= btrfs_rmdir
,
7411 .rename
= btrfs_rename
,
7412 .symlink
= btrfs_symlink
,
7413 .setattr
= btrfs_setattr
,
7414 .mknod
= btrfs_mknod
,
7415 .setxattr
= btrfs_setxattr
,
7416 .getxattr
= btrfs_getxattr
,
7417 .listxattr
= btrfs_listxattr
,
7418 .removexattr
= btrfs_removexattr
,
7419 .permission
= btrfs_permission
,
7421 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7422 .lookup
= btrfs_lookup
,
7423 .permission
= btrfs_permission
,
7426 static const struct file_operations btrfs_dir_file_operations
= {
7427 .llseek
= generic_file_llseek
,
7428 .read
= generic_read_dir
,
7429 .readdir
= btrfs_real_readdir
,
7430 .unlocked_ioctl
= btrfs_ioctl
,
7431 #ifdef CONFIG_COMPAT
7432 .compat_ioctl
= btrfs_ioctl
,
7434 .release
= btrfs_release_file
,
7435 .fsync
= btrfs_sync_file
,
7438 static struct extent_io_ops btrfs_extent_io_ops
= {
7439 .fill_delalloc
= run_delalloc_range
,
7440 .submit_bio_hook
= btrfs_submit_bio_hook
,
7441 .merge_bio_hook
= btrfs_merge_bio_hook
,
7442 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7443 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7444 .writepage_start_hook
= btrfs_writepage_start_hook
,
7445 .readpage_io_failed_hook
= btrfs_io_failed_hook
,
7446 .set_bit_hook
= btrfs_set_bit_hook
,
7447 .clear_bit_hook
= btrfs_clear_bit_hook
,
7448 .merge_extent_hook
= btrfs_merge_extent_hook
,
7449 .split_extent_hook
= btrfs_split_extent_hook
,
7453 * btrfs doesn't support the bmap operation because swapfiles
7454 * use bmap to make a mapping of extents in the file. They assume
7455 * these extents won't change over the life of the file and they
7456 * use the bmap result to do IO directly to the drive.
7458 * the btrfs bmap call would return logical addresses that aren't
7459 * suitable for IO and they also will change frequently as COW
7460 * operations happen. So, swapfile + btrfs == corruption.
7462 * For now we're avoiding this by dropping bmap.
7464 static const struct address_space_operations btrfs_aops
= {
7465 .readpage
= btrfs_readpage
,
7466 .writepage
= btrfs_writepage
,
7467 .writepages
= btrfs_writepages
,
7468 .readpages
= btrfs_readpages
,
7469 .sync_page
= block_sync_page
,
7470 .direct_IO
= btrfs_direct_IO
,
7471 .invalidatepage
= btrfs_invalidatepage
,
7472 .releasepage
= btrfs_releasepage
,
7473 .set_page_dirty
= btrfs_set_page_dirty
,
7474 .error_remove_page
= generic_error_remove_page
,
7477 static const struct address_space_operations btrfs_symlink_aops
= {
7478 .readpage
= btrfs_readpage
,
7479 .writepage
= btrfs_writepage
,
7480 .invalidatepage
= btrfs_invalidatepage
,
7481 .releasepage
= btrfs_releasepage
,
7484 static const struct inode_operations btrfs_file_inode_operations
= {
7485 .getattr
= btrfs_getattr
,
7486 .setattr
= btrfs_setattr
,
7487 .setxattr
= btrfs_setxattr
,
7488 .getxattr
= btrfs_getxattr
,
7489 .listxattr
= btrfs_listxattr
,
7490 .removexattr
= btrfs_removexattr
,
7491 .permission
= btrfs_permission
,
7492 .fiemap
= btrfs_fiemap
,
7494 static const struct inode_operations btrfs_special_inode_operations
= {
7495 .getattr
= btrfs_getattr
,
7496 .setattr
= btrfs_setattr
,
7497 .permission
= btrfs_permission
,
7498 .setxattr
= btrfs_setxattr
,
7499 .getxattr
= btrfs_getxattr
,
7500 .listxattr
= btrfs_listxattr
,
7501 .removexattr
= btrfs_removexattr
,
7503 static const struct inode_operations btrfs_symlink_inode_operations
= {
7504 .readlink
= generic_readlink
,
7505 .follow_link
= page_follow_link_light
,
7506 .put_link
= page_put_link
,
7507 .getattr
= btrfs_getattr
,
7508 .permission
= btrfs_permission
,
7509 .setxattr
= btrfs_setxattr
,
7510 .getxattr
= btrfs_getxattr
,
7511 .listxattr
= btrfs_listxattr
,
7512 .removexattr
= btrfs_removexattr
,
7515 const struct dentry_operations btrfs_dentry_operations
= {
7516 .d_delete
= btrfs_dentry_delete
,